Journal of Pathology Informatics Journal of Pathology Informatics
Contact us | Home | Login   |  Users Online: 153  Print this pageEmail this pageSmall font sizeDefault font sizeIncrease font size 




 
Table of Contents    
ABSTRACT
J Pathol Inform 2015,  6:24

Abstracts: Pathology Informatics 2015


Date of Web Publication02-Jun-2015

Correspondence Address:
Login to access the Email id

Source of Support: None, Conflict of Interest: None


Rights and PermissionsRights and Permissions

How to cite this article:
. Abstracts: Pathology Informatics 2015. J Pathol Inform 2015;6:24

How to cite this URL:
. Abstracts: Pathology Informatics 2015. J Pathol Inform [serial online] 2015 [cited 2017 Nov 23];6:24. Available from: http://www.jpathinformatics.org/text.asp?2015/6/1/24/157992

Short Abstracts Presentations

Locations: Wyndham Grand Pittsburgh Hotel, May 5 - 8 2015, Ballroom 1,
King's Garden 1, King's Garden 2/3, Brigade Room



VarBase: A Platform for the Storage and Clinical Interpretation of Next Generation Sequencing Data

Wade L. Schul 1 , John G. Howe 1 , Karl Hager 1 , Henry M. Rinder 1

1 Department of Laboratory Medicine, Yale University, New Haven, CT, USA. E-mail: wade.schulz@yale.edu

Content

Next generation sequencing (NGS) has rapidly expanded in the clinical laboratory. However, efficient clinical interpretation of NGS data is difficult due to the lack of support for genetic information within laboratory information systems. We present a framework developed within our institution named VarBase that provides access to NGS data for clinical interpretation and research studies.

Technology

The VarBase platform is based on two database technologies: a relational database (SQL Server, Microsoft, Redmond, WA, USA) and a document-oriented search platform (Elasticsearch, Elasticsearch, Los Altos, CA, USA). These are hosted on Windows- and Linux-based virtual machines, respectively. Information parsed from publically accessible databases such as ClinVar, COSMIC, and the 1000 Genomes project are housed within a publically accessible server to provide variant-specific annotations for clinical interpretation.

Design

Identifiable patient information is stored within the SQL database, since it is easily encrypted and provides all traditional security models. Deidentified variant information is also submitted to the VarBase research engine based on Elasticsearch, which can be used to create dynamic, real-time queries against the entire dataset. A graphical user interface provides access for variant interpretation by pathologists and review by oncology teams. Finally, a set of technology-agnostic web services allow authorized users to access sequencing data for research projects and clinical trials.

Results

The VarBase system has been in use since October 2014 for the clinical interpretation of a custom NGS panel for acute myeloid leukemia and myelodysplastic syndrome. The use of this system has decreased the turnaround time for clinical interpretation from approximately 1 month to 1 week from the date of sample collection. In addition, the web service interface has been integrated into multiple clinical research projects in different departments.

Conclusion

The flexibility and scalability of the VarBase platform have allowed our institution to provide a rapid turnaround time for NGS interpretation. Unlike many commercial applications, the open platform with a common set of web service interfaces has allowed several investigators access to sequencing data for ongoing clinical studies. This approach to NGS data management will significantly increase the availability of relevant information on patient outcomes to pathologists and primary clinical teams.

Novel Label-free Chemical Imaging for the Identification of Biomarkers of recurrent Diabetic Nephropathy

Vishal K. Varma 1,2 , Andre Kajdacsy-Balla 1 , Sanjeev Akkina 3 , Suman Setty 1 , Michael Walsh 1,2

Departments of 1 Pathology and 2 Bioengineering, College of Medicine, University of Illinois at Chicago, 3 Department of Medicine, Division of Nephrology, College of Medicine, University of Illinois at Chicago, USA. E-mail: vvarma4@uic.edu

Content

Kidney transplantation is the main treatment for end-stage renal disease, however, close monitoring of posttransplant biopsies is required to monitor and identify subclinical complications. In high-risk patients, surveillance biopsies are acquired approximately every 6-12 months posttransplantation to examine tissue histology to find markers associated with complications. Recent technological advances in developing high resolution imaging approaches have allowed for the interrogation of the biochemical status of glomerular and tubular structures to identify markers of disease processes.

Technology

We focused on identifying biochemical markers associated with recurrent diabetic nephropathy using the chemical imaging approach, Fourier transform infrared (FT-IR) spectroscopic imaging. FT-IR imaging is an emerging approach to obtain label-free images of the biochemical composition of tissue biopsies (including proteins, lipids, collagen, DNA, and glycation).

Design

A study was performed on native kidney tissues in order to identify biochemical markers associated with diabetic nephropathy. Serial sections were acquired and stained with PAS or imaged using chemical imaging. The histological parameters used for diagnosis were compared to the biochemical markers. Finally, a follow-up studied was performed on renal transplant patients to replicate the first study.

Results

Biomarkers were identified that were changed in renal structures associated with the progression of diabetic nephropathy, including increased levels of glycation. Using these two biomarkers alone, it is possible to diagnose the disease with specificity of 91% and sensitivity of 88%. In comparison, using the histological parameter, mesangial fractional surface area alone gives a specificity of 82% and sensitivity of 89%. When all the biochemical information is used to diagnose the cases, we obtained a 100% specificity and 100% sensitivity. These biomarkers were found to be increasing in the cohort of transplant patients that underwent rapid diabetic nephropathy recurrence. In addition, the early biopsies from the patients who underwent later diabetic nephropathy progression were biochemically different from the nonprogressive patients, suggesting that chemical imaging may identify prehistological biomarkers that will predict outcome.

Conclusion

We have demonstrated that a number of biomarkers are associated with the advancement of diabetic nephropathy and that we can track the early recurrence of diabetic nephropathy in surveillance biopsies.

Google Glass Imaging Modality for Integrated Gross Pathology Workflow

Jeffrey Taylor 1 , Brian Kolowitz 1 , Ishtiaque Ahmed 2 , Anil V. Parwani 2 , Liron Pantanowitz 1

1 Technology Development Center, University of Pittsburge Medical Centre, 2 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: pantanowitzl@upmc.edu

Content

Google Glasses (Glass™) have great potential in healthcare. Several proofs of concept have been tested such as enhanced patient interaction, telemedicine, and education. Glass has also been used for telepathology. Current imaging in gross pathology requires a hands-on approach, cameras capable of macrophotography, and computers to import captured images into the laboratory information system. Our aim was to use Glass to develop a hands-free imaging modality integrated into gross pathology workflow.

Technology

Google Glass (explorer prototype) device. Glass web application program interface (HTML5, CSS, JavaScript, Google Glass Mirror API). Glass application server (Microsoft.Net 4.5, Microsoft ASP.Net MVC 3, Google Glass Mirror API, Object Store API, Patient and Exam Query API).

Design

Using Google Glass connected to Google cloud, a user activates our Glass Modality App inside the glass display (Figure). Using a Glass App Server, the user can query a patient by searching the master patient index on our local institution server. After saying "ok glass" to take a snapshot, the static image with associated metadata is then stored in the laboratory information system.



Results

Through the use of our Glass Modality App, a user is able to capture and upload pathology images acquired with the wearable Glass device into a patient's electronic record by means of voice commands.

Conclusions

The Glass Modality App is useful in situations like grossing where the user cannot use their hands, mobility and point-of-view are beneficial, and static images of gross pathology are needed. These custom interfaces allow abstract query of patient and metadata, storage of captured images using the Glass device, and integration with the laboratory information system.

Comparison of Companion Virtuoso and Stand-alone Visiopharm Breast Marker Image Analysis Software

Marais Combrinck 1,2 , Jeffrey L. Fine 3 , Liron Pantanowitz 3

1 Departments of Anatomical Pathology, PathWest, QEII Medical Centre, Perth, 2 School of Pathology and Laboratory Medicine, University of Western Australia, Australia, 3 Department of Pathology, UPMC, Pittsburgh, PA, USA. E-mail: pantanowitzl@upmc.edu

Content

Image analysis to quantify immunohistochemical markers is recommended to improve accuracy, consistency, and reproducibility. To date, several commercial image analysis algorithms have been developed to quantify the breast immunohistochemical markers estrogen receptor (ER), progesterone receptor (PR), Her2/neu, and Ki-67. The aim of this study was to compare the performance of two image analysis software packages at quantifying these breast markers.

Technology

Whole slide scanners used to digitize slides were iScan Coreo Au (Ventana) and Aperio XT (Leica). Image analysis was performed using companion software with the iScan device (Virtuoso version 5.2.2, Ventana) and the stand-alone software TissueMorphDP (Visiopharm Integrator System, version 4.6.2.811).

Design

Twenty consecutive breast invasive carcinoma cases were selected. ER, PR, Her2/neu, and Ki-67 immunohistochemical stains from these cases were scanned. Due to file format compatibility issues slides had to be scanned twice (Aperio and Ventana). Image analysis using Virtuoso software involved a pathologist manually selecting representative regions of interest to be scored. The Visiopharm unaltered basic nuclear and HER-2 algorithms (HER2-IHC, Connectivity) were used to score the entire digital slide. The results from both software platforms were compared.

Results

All but one case could be analyzed (one ER case repeatedly failed with Virtuoso). For ER analysis, four cases had discrepancies below 10%, five differed by 10-20%, and 10 cases had a score discrepancy over 20%. Virtuoso software gave higher ER scores. For PR analysis, eight cases showed a discrepancy below 10%, five differed by 10-20%, and seven had discrepancies over 20%. For Her2/neu analysis, 15 cases showed concordant results. In three cases, Visopharm scored Her2/neu as 3+ compared to Virtuoso where two of these cases were 1+ and the other was 0. Over-scoring in these 3 cases was due to artifacts from scoring of air bubbles and inked tissue edges. For Ki-67 analysis, six cases showed a difference of less than 10%, two cases differed by 10-20%, and 12 cases differed by more than 20%.

Conclusion

Image analysis of breast markers using different software platforms produced varying results that could impact patient care. Better optimization of algorithms instead of using default settings and human intervention to eliminate artifacts could improve performance. Image analysis results in this study could be skewed due to differences in image file format or from scoring regions of interest versus whole digital slides. Without a better balance of automation versus need for manual supervision, image analysis may not yet represent a prudent expenditure of limited laboratory resources. Practical guidelines are needed and standardization is recommended to improve the performance of digital image analysis in current pathology practice.

Utilizing Structured Data in Pathology Reporting for Clinical Flow Cytometry

Aaron C. Shaver 1

1 Department of Pathology, Microbiology and Immunology, Vanderbilt University Medical Center, Nashville, TN, USA. E-mail: aaron.shaver@vanderbilt.edu

Content

Pathology reporting is well-suited to templating, due to the structured nature of the information being conveyed. For complicated reports like clinical flow cytometry, an underlying data structure is of great benefit compared to "fill-in-the-blanks" style templating, both in terms of ease of use for the operator as well as the ability to extend the system as new capabilities are added.

Technology

A web-based interactive form was designed using HTML 5, JavaScript, and the publicly available jquery library with templates stored in custom XML files.

Design

The form was designed with underlying data structures that included classes for flow cytometry markers, cell types, and marker panels. Interaction between the classes was modeled so that, for example, a description of B lymphocytes would only include the subset of markers performed that were relevant for that cell type. The class structures also included attributes that automatically generate diagnostic lines and interpretive comments.

Results

The software tool was designed for use by pathologists reviewing the flow cytometry data. Emphasis was placed on the ability to generate reports with complex diagnostic lines and comments that were easily readable while minimizing the amount of effort required for report generation. Commonly encountered diagnostic scenarios (for example, a new diagnosis of acute leukemia) were included in a set of templates which would prepopulate fields in a manner depending on the other variables selected (such as the panels of tubes used). Extensibility of the system (for example, adding a new marker to a panel) was ensured via design of the data structures to allow addition or subtraction of new members without the need for major reformatting. The tool was placed into general clinical practice in July 2014. Users reported increased efficiency and reproducibility with no major drawbacks.

Conclusion

As the information being conveyed in pathology reports grows increasingly more complicated, traditional synoptic-style reporting becomes less and less efficient as a method for generating reports in a templated manner. Although it requires a greater initial investment, designing custom data structures for a reporting tool greatly increase its utility for the user.

Developing a Pseudo-bidirectional Order Entry Interface Using Laboratory Information System

Mehrvash Haghighi 1 , Mable Rosario 1

1 Department of Pathology, Columbia Medical Center, New York Presbyterian Hospital, New York, NY, USA. E-mail: mh3373@cumc.columbia.edu

Content

CPOEs ensure proper collection and exchange of patient's health data between laboratories and clinics and promote continuity of care. However, there is ongoing challenge of adoption by clinicians due to interruption to workflow. Customizing the order entry interface based on specific needs of practice, routine process will improve the usability and increase adoption.

Technology

LIS: Cerner Copath v2013.01.1.136 (Kansas City, MO, USA); Outreach vendor: 4Medica version 15.10 (Culver City, CA, USA); Practice EMR: Athena v1.2 (Marina Del Rey, CA, USA).

Design

The main objective of this work was to streamline the ordering process and result retrieval. We created a complete set of combined tests for pap smears and gynecology biopsies. A new set of part types was also built in LIS corresponding to test orders. The new part types contain both the test names and test codes. Our outreach vendor did customized programming to strip out the test names and codes and place them in target segments of OBR designated by EMR.

Results

Clinician places the orders in EMR and prints the requisition. The laboratory accessions, the received specimen in LIS using new part types. The result output includes part type name containing test name and code which will be parsed out. The returned test name and code values will link the result back to the original order and change the status of ordered test from "order" to "review result." When clinician checks the lab information, the result displays with exact label of original order with new status. Without this customization, the result would create a separate value with a general label such as "surgical pathology" or "cytology" which requires clinician to review the previous progress notes to find the original orders and open multiple reports to find the corresponding result.

Conclusion

Flexible, customized order entry is the key to high adoption and successful implementation. We found the followings as the most important factors for the success of this project:

  • Focus on automating physician's ordering process
  • Customizing the design to improve efficiency
  • Minimize the impacts to ancillary workflow
  • Allowing for a flexible rollout strategy.


Microimaging: Seeing the Unseen in Living Patients

Christopher Garcia 1 , Guillermo Tearney 2

1 Department of Pathology, Massachusetts General Hospital, 2 Massachusetts General Hospital, Wellman Center for Photomedicine, Boston, MA, USA. E-mail: cgarcia14@partners.org

Content

Today's gold standard for medical diagnosis is histology of excised biopsies or surgical specimens where tissue is taken out of the body, processed, sectioned, stained an looked at under a light microscope by a pathologist. There are many limitations of this technique, including the fact that it is inherently invasive, time-consuming, costly, and dangerous for some organs. Furthermore, oftentimes, the diseased tissue is not readily seen by visual inspection and as a result, the tissue is sampled at a random location, which can be highly inaccurate. If we could instead conduct microscopy inside the body, then we could provide tools for screening, targeting biopsies, making primary disease diagnosis, and guiding intervention on the cellular basis. This promise has motivated the development of a new field, termed in vivo microscopy (IVM), the goal of which is to obtain microscopic images from living human patients.

Technology

Two IVM technologies, confocal microscopy (CM) and optical coherence tomography, are currently available and in clinical use. Upcoming developments including whole organ microscopy, swallowable microscopy capsules, molecular imaging, and very high resolution microscopic devices are under development.

Design

This is an explanatory overview of IVM technologies as they currently exist in the market and some of the new technologies and imaging modalities that are under development.

Results

The clinical application of IVM is relatively new with little pathologist involvement at this time. The College of American Pathologists is involved in several initiatives to increase pathologist awareness of the technologies, increase competency in pathologist diagnosis of IVM images, and foster interest and involvement in IVM employment.

Conclusion

CM and optical coherence tomography, in addition to the other IVM technologies and modalities that are currently under development, will likely revolutionize how disease is diagnosed and how medicine is practiced in the future.

Constellation: A Web 2.0 Adult Autopsy Reporting and Analytics Platform

Matthew D. Cain 1 , Joseph Drwiega 1 , Zheng Ping 1 , Seung Park 1

1 Department of Pathology, University of Alabama at Birmingham, Birmingham, AL, USA. E-mail: cainmd@uab.edu

Content

Autopsies are an integral part of patient care in the modern academic medical center, and the potential usefulness of the immense amounts of data that these autopsies generate is widely accepted. However, due in large part to antiquated anatomic pathology laboratory information systems that are: (a) Not specifically designed around the autopsy pathology workflow and (b) extremely difficult to query for data analytics purposes, these data generally remain locked away. In order to solve these problems at our institution, we have developed, tested, and deployed a modern, Web 2.0 adult autopsy reporting and analytics platform codenamed "Constellation."

Technology

Server Hardware: Dell Precision T3600; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 14.04 LTS 64-bit; Web Server: nginx 1.7; Database Management System: MariaDB 10.0; Programming Language: PHP-FPM 5.5; User Interface Framework: Twitter Bootstrap 3.3.

Design

Constellation consists of an autopsy report generator with a searchable database. Its frontend is written in PHP-FPM, its relational database layer is MariaDB, and Twitter Bootstrap is used for the user interface and form validation. Constellation's user interface is streamlined for our institution's autopsy pathology workflow, and gives the user one-click customization options for autopsy reports based on our institution's standard template. The backend database stores relevant report data - including demographics and diagnoses - as discrete data elements which can be arbitrarily searched and queried via SQL.

Results

Constellation has been implemented, tested, and deployed for our autopsy pathology service. In preliminary testing, constellation reduces the amount of time an experienced resident takes to generate a full autopsy report by 17%; we expect the time savings for inexperienced residents to be even larger.

Conclusion

Constellation provides meaningful speed increases in resident generation of autopsy reports, drives down error, and allows for seamless collection of data for analytics purposes. As constellation continues to be used, its database will allow for increasingly sophisticated data analytics. Efforts to use it to track interesting cases for morbidity and mortality interdepartmental conferences, teaching, and infection control tracking are being actively developed.

Re-imagining Pathology Consultations Using the Scalable Adaptive Graphics Environment: A Live Demonstration

Bruce Levy 1 , Tushar Patel 1 , Victor Mateevitsi 2

1 Department of Pathology, 2 Electronic Visualization Laboratory, University of Illinois, IL, USA. E-mail: bplevy@uic.edu

Content

Whole-slide imaging (WSI), while technologically mature, has not become mainstream. One reason is that current methods of visualizing and using WSIs follow long-existing workflows for glass slides. If we take advantage of the digital nature of WSIs and use them in novel ways that were not possible with physical slides, we will make significant progress to convince pathologists to embrace this technology and advance medical care, research, and education.

Technology

The Scalable Adaptive Graphics Environment (SAGE) was developed at the University of Illinois at Chicago's Electronic Visualization Laboratory to facilitate collaborative efforts that require the sharing of data-intensive information for analysis. SAGE is an open-source visualization and collaboration windowing environment that runs in a web browser, enabling users to access, display, and share a variety of data-intensive information, in a variety of resolutions and formats from multiple sources taking advantage of HTML5 and the high-performance graphics and networking capabilities contained in modern web browsers. A WSI-viewer for SAGE was developed.

Design

We will demonstrate this technology through the performance of a live simulated pathology consultation between two pathologists. The pathologists will communicate through a videoconference window in SAGE while both macroscopic images and WSIs are displayed, manipulated, and discussed in real-time. Given sufficient time, we hope to invite members of the audience to connect with the consultation through the creation of individual pointers on their personal devices.

Results

In a previous demonstration, we successfully performed a consultation between two pathologists. In addition, this technology has been successfully used for multidisciplinary conferences and the education of medical students and pathology residents. We have shown in these scenarios that any number of users can connect to the specific instance of SAGE simultaneously via their computers, tablets or smart phones to view and manipulate WSI and other visual and nonvisual information.

Conclusion

SAGE is capable of using WSIs in ways that are not possible with glass slides or the current generation of WSI-viewers or web sharing technology. It can change the playing field in favor of WSIs.

Applied Telemicroscopy for Microbiology: Comparing the Accuracy of Whole Slide Images to Static Photomicrographs

Daniel D. Rhoads 1 , Nadia F. Habib-Bein 1,2 , Rahman Hariri 1 , Douglas J. Hartman 1 , Sara E. Monaco 1 , Liron Pantanowitz 1

1 Departments of Pathology, University of Pittsburgh Medical Center, 2 Veterans Affairs Pittsburgh Healthcare System, Pittsburgh, USA. E-mail: danrhoads@gmail.com

Content

Telemicroscopy has the potential to improve the rapidity of microbiology-related consultations, but the optimal imaging modalities have not been thoroughly investigated for this application, particularly in regard to interpretive accuracy. This study compares ×40 whole slide imaging (WSI) and static photomicroscopy for telemicrobiology.

Technology

The Aperio XT system (Leica Biosystems, Nussloch, Germany) was used for ×40 WSI acquisition. An Olympus (Center Valley, Pennsylvania, USA) system (BX46 microscope, DP73 camera, cellSense Standard 1.11 software) was used to obtain digital photomicrographs.

Design

Four evaluators (pathologists and clinical microbiologists) each evaluated 30 microbiology-relevant slides using 3 modalities: ×40 WSI, representative ×10 and ×100 static digital photomicrographs, and glass slides. A minimum 2-week wash-out period was employed between evaluations. Various specimens (cytology, histology, laboratory preparations) containing viruses, bacteria, mycobacteria, fungi, parasites, and negative controls using different stains were studied. Evaluators provided a diagnosis and assessed the imaging technology for all cases. Statistical analyses were performed using two-tailed Student's t-test in Microsoft Excel 2010 (Redmond, Washington, USA).

Results

Diagnoses using glass slides and static images were more accurate than ×40 WSI (P = 0.02 and 0.03, respectively). Diagnostic accuracy using glass slides and static images was equivalent (P = 0.83). Evaluators perceived image resolution and focus to be limiting factors in photomicrographs and WSI. False negative interpretations occurred less frequently when using static images (3%) than glass slides (8%) or ×40 WSI (7%).

Conclusion

Image resolution and suitable focus appear to be limitations of digital imaging for telemicrobiology. If photographed appropriately, the accuracy of static telemicrobiology is comparable with glass slide analysis. For telemicrobiology, the diagnostic accuracy of ×40 WSI is inferior to using representative static images. Better imaging technology is required to facilitate telemicrobiology.

Toward Accurate Reporting of Next-generation Sequencing Results: Deriving Measurement Precision in Limited Clinical Specimens

Thomas Blomquist 1 , Erin Crawford 1 , James C. Willey 1

1 Department of Pathology and Medicine, University of Toledo Medical Center, Toledo, OH, USA. E-mail: thomas.blomquist@utoledo.edu

Content

Small or degraded clinical specimens pose unique challenges to accurate reporting in next-generation sequencing (NGS)-based diagnostics. These challenges, in part, stem from stochastic sampling during both specimen preparation and sequencing steps. For these specimens, appropriate design of NGS assay standards and how their analysis is integrated in the informatics pipeline may enable more accurate reporting of measurement precision. For small or degraded specimens, we hypothesized that NGS-based assay coefficient of variation (CV) would be predicted by: (a) Sequencing coverage (i.e., read counts) and (b) input molarity (i.e., number of intact molecules).

Technology

Monte Carlo simulations of Poisson sampling distributions were used to derive three equations to predict expected NGS assay CV from: (1) Sequencing coverage, (2) input molarity, or (3) both sequencing coverage and input molarity. Recently described targeted amplicon specimen preparation for NGS with competitive internal standards, and its accompanying informatics pipeline, was used to provide measurement of input molarity, sequencing counts, and observed CV across replicates.

Design

Cell lines with known allelic composition and ratio were mixed and prepared for NGS such that a broad range of limiting allelic molar proportion and/or sequence read counts were observed (46 sets of quadruplicate measurements). Observed CV was then compared to expected CV based on derived equations.

Results

For the equation derived from both sequencing coverage and input molarity (#3), expected CV was very close to observed (average [observed CV/expected CV] = 1.01) and explained 74% of observed assay variance [Figure 1]. In contrast, observed CV was on average 13-fold, or 1.5-fold, higher than expected CV based on sequencing coverage (#1), or input molarity (#2), prediction models alone.



Conclusions

For limited or degraded specimens, knowledge of both input molarity as well as sequencing coverage provides accurate reporting of precision for single replicate measurements in targeted NGS-based assays. Reliance on sequencing coverage information alone may grossly underestimate NGS precision for single measurements. Demonstration of the basic math and various molecular standardization approaches required to create accurate reporting of NGS measurement precision will be provided during this presentation, and afterward via E-mail.

Independent, Student-led Software Engineering at a Free Medical Student-run Clinic: Pathology Informatics Leads the Way

Timothy Kennell Jr. 1 , Omar Ramadan 1 , L. Nicholas Herrera 2 , Nathan Haywood 2 , Jae Sung 2 , Robin Lorenz 1 , Seung Park, MD 3

1 NIH Medical Scientist Training Program, University of Alabama at Birmingham School of Medicine, 2 University of Alabama at Birmingham School of Medicine, 3 Department of Pathology, Division of Informatics, University of Alabama at Birmingham, Birmingham, AL, USA. E-mail: tikenn@uab.edu

Content

In this era of Big Data in medicine, independent software engineering capabilities have emerged as key drivers of health care innovation. While many student-run clinics have informatics committees, few have the infrastructure and support that would allow for meaningful clinical software engineering. Creation of such infrastructure, as well as training of students to serve in software engineering roles, therefore became a key focus for our clinic in academic year 2014-2015. In this abstract, we present the 1-year impact that pathology informatics has had at our institution's student-run clinic.

Technology

Server Hardware: Dell PowerEdge 2590; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 14.04 LTS 64-bit; Web Server: nginx 1.7; Database Management System: MariaDB 10.0; Server-Side Programming Language: PHP-FPM 5.5; User Interface Library: Twitter Bootstrap 3.3.

Design

We designed and led a 1-week-long boot camp in clinical informatics and software engineering. The students performed a survey of the informatics landscape at our clinic, identifying unmet needs and engaging with stakeholders. We chose to design and implement a custom, Web 2.0, real-time volunteer signup, tracking, and training management platform as the first critical subsystem of a future laboratory information system and electronic medical record.

Results

Within 1 year of the boot camp, under pathology informatics faculty guidance, the students designed, built, and deployed the platform. 100% of users have converted over, and greater than 30% of users take advantage of the smartphone-optimized layout. It has reduced clinic coordinator burden from 20+ h/week to under 1 h/week, and has driven down human error in scheduling. Furthermore, our underlying databases have been increasingly useful for advanced analytics.

Conclusion

Pathology Informatics has set the vision and led the way for an informatics and Big Data-driven future at our institution's student-led clinic. In 1 year, we have gone from using off-the shelf, ill-fitting software systems to generating our own custom solutions. Our finalized platform is an integral part of our clinic's patient care. This effort shows the utility of informaticist-supervised student software engineering both as a learning opportunity and as a driver of cultural change.

The Diagnostic Accuracy of Digital Microscopy: A Systematic Review

Edward Goacher 1,2 , Rebecca Randell 3 , Darren Treanor 1,2

1 Department of Pathology, Leeds Institute of Cancer and Pathology, 3 School of Health Care, Leeds University, 2 Department of Pathology, Bexley Wing, St. James' University Hospital, Leeds, UK. E-mail: um11eg@leeds.ac.uk

Content

Digital microscopy in pathology involves the generation of digital images from glass slides. They can be viewed, stored, and shared virtually. At present, digital images are used routinely in both education and research. They are not currently used on a large scale in routine primary diagnosis, and are not approved for clinical use in all jurisdictions. No systematic review of their diagnostic accuracy has been performed. This systematic review aims to assess the effect of digital microscopy on diagnostic accuracy when compared to glass slides. Secondary outcome measures, including the impact of the technology systems used, will also be assessed.

Technology

A systematic review was performed. Studies examining the use of whole slide imaging, static telepathology, and dynamic telepathology in comparison to glass slides were included in the review. Studies comparing the different digital imaging technologies were also included.

Design

The review was designed according to the Cochrane systematic review guidelines. The primary reviewer received specialist training in conducting systematic reviewers prior to undertaking the review. Multiple (n = 5) electronic databases were searched, reference searching of included papers, and citation tracking were performed to retrieve studies. A search of the grey literature was performed to minimize bias. Two independent reviewers, one of whom an expert in the field, subjected all studies to a predefined screening algorithm. A third independent reviewer was consulted in cases where there was a disagreement. Data from all included papers were extracted using the EPOC data collection template and its quality subjected to the QUADAS tool by two independent reviewers.

Results

A literature search retrieved 1256 studies that are currently being subjected to the screening algorithm. Univariable and multivariable logistic regression models, adjusted for study risk of bias, will be used to investigate associations between outcomes of interest and study-specific covariates. If justified, a meta-analysis of the data will be performed.

Conclusions

Data analysis is ongoing and the results will be presented.

Optical Coherence Tomography for Evaluation of Pelvic Wash Cytology Specimens

Jeffrey L. Fine 1 , Muhammad A. Syed 1 , Liron Pantanowitz 1

1 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: finejl@upmc.edu

Content

Optical coherence tomography (OCT) is an advanced microscopic imaging technique that has advantages over traditional microscopy. Most OCT work to date has dealt with tissues. We are not aware of any OCT studies involving cytopathology. The aim of this study was to determine the feasibility of OCT for imaging cytology samples.

Technology

Archival pelvic wash specimens fixed in Cytolyt were imaged by OCT (LightCT, LLTech SAS, Paris France) at 1.6 micron transverse (x,y) resolution by 1 micron axial (z) resolution. Image stacks comprised of two-dimensional (2D) slices were acquired at increasing depth into the specimen.

Design

Fixed cytology specimens were centrifuged. Drops of the specimen enriched with cells were placed on a coverslip and inserted within the OCT imaging chamber. OCT images were compared to ThinPrep glass slides to locate and characterize cells.

Results

OCT image files of 2D slices were small (<500 KB each) compared with tissue images. Imaging artifacts hampered good quality images. These were related to improvisations required to acquire cytology images as the OCT system was designed for tissue imaging. Cells were clearly visible and size determinations were possible. While nuclear detail was lacking there were recognizable differences between benign and malignant specimens [Figure 1].



Conclusion

This feasibility study demonstrates that isolated cells in cytology samples can be imaged using OCT. Compared to traditional microscopy, OCT is quick, nondestructive, and requires no slide preparation or scanning. It is plausible that in the near future portable OCT devices could be used to image body fluids (e.g., effusions, ascites) in vivo for immediate diagnostic purposes.

Digital Pathology Integrative Platform

Ilker Ersoy 1 , Mikhail Kovalenko 1 , Chi-Ren Shyu 1,2,3 , William Krause 1 , Donald Doll 1 , Richard Hammer 1 , Dmitriy Shin 4-6

Department of 1 Pathology, 2 MU Informatics Institute, 3 Computer Science, 4 Electrical and Computer Engineering and 5Department of Medicine, 6 Ellis Fischel Cancer Center, University of Missouri, Columbia, MO 65212, USA. E-mail: Shindm@health.missouri.edu

Content

We present Digital Pathology Integrative Platform (DPIP) that unifies Whole slide imaging (WSI) technology with web-based interactive WSI viewing, annotation, workflow/logic processing framework, gaze tracking and bio-objects detection and identification module to support educational and research tools in pathology informatics. Collected Human Computer Interaction data facilitate best practices in learning/training experience and allows capturing visual diagnostic heuristics from WSI. We demonstrate utility of DPIP using a WSI web-based interactive hematopathology atlas with game-like training workflow and WSI web-based interactive histology atlas, in which learning of histological structures is facilitated by their functional annotations.

Technology

Aperio ScanScope CS was used to digitize glass slides. DPIP platform was developed in-house using JavaScript and Java technologies. Open-source based software such as "Openseadragon" viewer, "VIPS" library, "IIPIMage" server was incorporated into DPIP using custom-built software adapters.

Design

DPIP consists of several software modules that are depicted in [Figure 1]. The first volume of hematopathology atlas was developed using DPIP based on 60 patient cases from Ellis Fischel Cancer Center, Columbia, MO and includes complete set of diagnostic materials such as Hematoxylin and Eosin/IHC slides, flow cytometry, radiographs, and patient history. Penalty-based diagnostic training scheme was developed for each of these cases. The proof-of-concept version of the interactive histology atlas includes tissue images from male and female reproductive organs. Biological objects (cells, follicles, tissues, etc.) have been delineated and functionally annotated for training and research purposes.



Results

DPIP-based educational suites were evaluated by hematopathology, oncology, and histology experts and educators. There has been high inter-expert agreement that DPIP-based tools provide significant improvement over traditional WSI-based educational and research resources. DPIP-based tools create a synergetic effect that enables conducting trainee behavior analysis, expert versus novice, specialist versus generalist pathologist studies as well as WSI image analytics for pathology diagnosis and quality assurance purposes.

Conclusions

We conclude that DPIP provides a means to build new generation of WSI tools where HCI data are exploited to design more effective applications for educational, research, and clinical purposes.

Implementation of the EPIC 2014 Beaker Clinical Pathology Module at Stanford Health Care

Brent Tan 1

1 Department of Pathology, Stanford School of Medicine, Stanford, CA, USA. E-mail: btan@stanford.edu

Content

An account of the Beaker Clinical Pathology implementation at Stanford Health Care, which went live on February 21, 2015.

Technology

EPIC Beaker Clinical Pathology 2014 (Verona, WI, USA), 2 instances of EPIC EMR 2014 (Stanford Health Care instance with Beaker and Stanford Children's Hospital instance), EPIC-to-EPIC interface, EPIC Blood Administration Module version 0.9, CITRIX Web Interface XenApp 6.5, Data Innovations (South Burlington, Vermont, USA) Instrument Manager v8.13.00.10, SafeTrace version 3.9 (Braintree, MA, USA), Rhodes Group (Vernon, CT, USA) Clinical Laboratory Repository, 716 Intermec (Fort Mill, SC, USA) PB50 mobile printers, 154 Intermec PC43D printers, 722 Datalogic (Bologna, Italy) Gryphon GD 4400-B 2D barcode readers, 35 EPIC Rover/Apple iPODs (Cupertino, CA, USA) with Honeywell (Fort Mill, SC, USA) Captuvo SL22h enterprise sleds, 3774 Windows 7 workstations.

Design

The governance and information gathering process leading to selection of a new laboratory information system including the rationale for selecting Beaker will be shared. The project timeline and staffing requirements will be highlighted, including incremental deliverables from both information technology and laboratory operations. Hardware and software components and key aspects of system design will be reviewed, including system architecture, archival data strategy, and interfaces. The software build and design, testing process, user training, and sign-off for the system will be examined and compared to the vendor's recommendations. The cut-over and immediate support plan will be reviewed.

Results

Key performance indicators including turn-around-times for tests and critical test value calls, and number of mislabeled specimens will be presented. Major issues encountered included: (1) Underestimation of the testing and iterative redesign process led to delays in the timeline and operational staffing shortages. (2) Training gaps related to overly generic training, lacking laboratory section specific processes and moderately complex workflows. (3) Insufficient time to test auto verification programmed within Beaker. (4) Technical challenges with the EPIC-to-EPIC interface.

Conclusions

EPIC Beaker version 2014 is a viable, fully functional laboratory information system. Strengths include vertical integration with the EPIC EMR for easier information exchange and integrated positive patient identification within nurse and clinic collections. Shortcomings include QC, aliquoting, lack of component level result verification, and critical call workflow.

Lessons Learned: Results after 1 Year Operation of Enterprise-wide Educational Whole Slide Image Resource

Christopher Garcia 1 , Mohamed Salama 2

1 Department of Pathology, Massachusetts General Hospital, Boston, MA, 2 Department of Pathology, University of Utah, UT, USA. E-mail: cgarcia14@mgh.harvard.edu

Content

Whole slide imaging (WSI) has allowed for the creation of online educational materials that can both reach a wide audience, as well as address the educational needs of different audiences. The Department of Pathology at the University of Utah created a "Digital Pathology Project" that is open to the public, but is curated by faculty members and trainees of the school of medicine. After a year of going live, we have usage statistics to present, as well as lessons learned concerning governance and curation of information within a site that is fully open for public access.

Technology

The Digital Pathology Project at the University of Utah is based on OpenDigital Pathology, an open-source, web-based application created with MySQL, Django, Python, jQuery, HTML5, and CSS3. It is housed on a webserver running Ubuntu 13.04. The WSI are converted by OpenSlide Python and the Google Maps API is used to display, navigate, and annotate the images.

Design

At the start of the project (February 2014), there were 10 study sets and 718 slides in total. Each volunteer "curator" was in charge of collecting, organizing, and uploading slides to the site. Each was given basic instructions, and video instructions are available on YouTube.com. The number of study sets, slides, and slide views was tracked over a period of 12 months.

Results

At 12 months of operation (February 2015), there are 11 study sets consisting of 1144 slides with a total of 12,756 views. Two study sets were deleted from the system, while 3 were added. 426 slides have been added. Surgical pathology showed the most frequent views; however, head and neck specialty showed was highest views relative to the number of slides. The tally details of slides present, number of views of slides, and slides added over 12 months are available to be viewed in [Table 1].



Conclusions

After 1 year of operations, the Digital Pathology Project at the University of Utah has grown and is gaining public attention and traffic. Due to the loose governance/curation model employed, different study sets have grown (and gained audiences) at different rates. Furthermore, we recognize that a public repository for deidentified cases cannot meet all the educational needs of the enterprise. However, it does create the flexibility to reach a wide range of trainees in many content areas.

Fair Use of Copyrighted Material in Medical Education: Opportunities and Risks

Philip J. Boyer 1 , Patrick R. Mann 2

1 Department of Pathology, East Carolina University, Brody School of Medicine, Greenville, NC, 2 Department of Pathology, University of Colorado Denver, Aurora, CO, USA. E-mail: boyerp14@ecu.edu

Content

Given the ever-expanding breadth and depth of the medical knowledge base and the ever-increasing expectations of medical students and residents, it is challenging to construct comprehensive and high-quality medical education teaching materials and interactions based on one's own collection of images, diagrams, and tables. Increasingly, easy electronic access to a wealth of outstanding material from journal articles, textbooks, and internet content allows for the gathering of content to illustrate a specific concept. All of these potential resources are copyrighted, either explicitly or implicitly. There is virtually no peer-reviewed, published guidance regarding the use of such material in the medical education setting.

Technology

This study incorporated electronic Internet searches for peer-reviewed publications (PubMed) about and copyrighted resources available for use in medical and graduate medical education.

Design

In collaboration with a University of Texas Austin Library lawyer who specializes in copyright law, this study sought to: (1) Evaluate existing United States copyright law, including the four "fair use" criteria and the Technology, Education, and Copyright Harmonization (TEACH) Act, with respect medical education and (2) define the opportunities and limitations in the use of copyrighted material by evaluating a wide variety of use-cases.

Results
"Fair use" criteria are incorporated into the United States copyright law and should both guide and facilitate decisions about the utilization of copyrighted materials in the medical education setting. Fair use strictures explicitly allow for the use of copyrighted content (e.g., a diagram or image from a textbook, a medical journal, or an on-line resource) to which students and trainees often already have "fair use" access rights. Appropriate attribution of the work must be undertaken to acknowledge the source of such material. While the TEACH Act primarily addressed "distance education," it also introduced limitations, requirements, and nuances (e.g., regarding the storage and serving of files from within learning management systems) that are relevant to traditional "face-to-face" courses.

Conclusions

Educators need to be aware of the legal implications and limitations of both: (1) The primary use and (2) the electronic storage and distribution of such content.

Optical Coherence Tomography Imaging of Pathology Specimens; Freedom from the Tyranny of Glass Slides

Jeffrey L. Fine 1

1 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: finejl@upmc.edu

Content

Optical coherence tomography (OCT) images often resemble whole slide images (WSIs), which may make OCT a natural choice for advanced imaging applications in pathology. This report details our early experience with OCT as a pathology imaging modality, including a pathologist-friendly introduction. We will present images from several tissue types, including breast and endometrium, and we will discuss our strategies for propelling current research into future clinical applications.

Technology

Numerous formalin fixed, tissue block size tissue pieces were imaged by OCT (LightCT, LLTech SAS, Paris France), at 1.6 micron transverse (x,y) resolution by 1 micron axial (z) resolution. A variety of acquisitions were used, including wide-field two-dimensional (2D) slices (mimics a histologic section), narrow-field stacks of 2D images (three-dimensional data set), and video files. Live images were viewed but were not acquired. OCT data were exported in multiple formats, including JPEG and 16-bit gray scale, and including DICOM formats. In general, WSIs were then created from Hematoxylin and Eosin slides made from the OCT-imaged tissue (Aperio ScanScope XT, Leica, Vista, CA, USA). Image data manipulation was attempted using Adobe Photoshop and ImageJ (NIH) software.

Design

OCT images were compared with WSIs and also were viewed under varying conditions, within the included viewer and also after export into other software. For OCT/WSI comparisons, full-size JPEG or TIFF files were compared side-by-side at similar zoom/magnification. For interpretation studies, collections of cropped images were placed into Microsoft PowerPoint presentations easier review by subject pathologists. Image manipulations included strategies: (1) Look-up table manipulation of 16-bit gray scale images, and (2) application of image filters followed by recombination of postprocessed image with original image.

Results

Some tissues are easily seen "as is," such as endometrium; it is likely that intra-operative and/or in vivo endometrial assessment will future OCT applications based upon this early system's current image quality [Figure 1]. Other tissues like breast will likely require a combination of rigorous histopathology study and more sophisticated image processing in order to yield reproducible clinical applications [Figure 2].





Conclusions

OCT is microscopy; therefore, pathologists are a natural choice for providing OCT-based diagnosis. OCT is very promising provided that it is not seen as a direct WSI replacement; pathologists should seek to adapt OCT according to its strengths for specific clinical applications. OCT is rapid, it does not require histology equipment or chemicals, and it may be possible to project OCT diagnostics into situations where slides cannot go (e.g., in vivo microscopy).

Perceptions of Pathology Informatics by Noninformatician Pathologists and Trainees

Addie Walker 1 , Christopher Garcia 2 , Jason Baron 2 , Thomas Gudewicz 2 , John Gilbertson 2 , Walter Henricks 1 , Roy Lee 1

1 Cleveland Clinic, Robert J. Tomsich Pathology and Laboratory Medicine Institute, Cleveland, OH, 2 Department of Pathology, Massachusetts General Hospital, Boston, MA, USA. E-mail: walkera5@ccf.org

Content

Although pathology informatics (PI) is essential to modern pathology practice, it remains misunderstood by noninformatics pathologists. To understand their current perceptions of PI better, we surveyed pathology trainees and staff at Cleveland Clinic (CC) and Massachusetts General Hospital (MGH): two, large academic centers with different PI cultures. CC has an IT help desk group embedded within PI, separate from central IT. MGH has a ubiquitous, continuous informatics presence, and a fellowship program.

Technology

A Likert survey was created using surveymonkey.com. An anonymous URL was generated and emailed to survey participants.

Design

Our 17-question survey assessed three general areas regarding PI: definition, interest, and utility. One question asked if the participant understood what PI encompasses, followed by a checklist of possible informatics activities, with some being pure IT or help desk activities. Practicing informaticians and informatics fellows were excluded.

Results

Totally, 107 individuals took the survey, of which 59 individuals agreed they understood what PI encompasses. Although more MGH versus CC respondents felt their institution placed a higher value on informatics training, this did not correlate with correct understanding of PI. Over 50% of these 59 participants (except for MGH trainees - 25%) associated help desk activities with informatics [Table]. Most agreed that pathologists need to be familiar with and knowledgeable in PI (n = 86), but "agree" responses decreased with "I plan to use PI in my career" (n = 71). Most respondents (n = 79) disagreed with "I am not interested in using informatics," while a small number (n = 13) agreed.



Conclusions

Correct understanding of PI is needed for noninformatics trained pathologists to effectively utilize informatics methodologies. PI is viewed positively by the majority of them, and there is strong interest in learning it. However, the definition of informatics varies greatly-incorrect beliefs exist and confuse PI with IT and help desk services, even to those who believe they understand PI. The findings also illustrate how differences in departmental culture can have on individual perception of PI. Further efforts by the PI subspecialty should continue to correct these misperceptions.

Biomarkers for Pancreatic Cancer - Identification Through Meta-analysis and Validation on Tissue Microarrays Utilizing Digital Pathology for Potential Clinical Application

Asif Ali 1 , Victoria Brown 2 , Simon Denley 3 , Nigel B. Jamieson 3 , Jennifer P. Morton 4 , Colin Nixon 4 , Janet S. Graham 5 , Zia Ul-Haq 1,6 , Daniel Francis MacKay 6 , Owen J. Sansom 4 , C. Ross Carter 3 , Colin J. McKay 3 , Fraser R. Duthie 7 , Karin A. Oien 8

1 Khyber Medical University, Peshawar, Pakistan, 2 Pathology Laboratory, Forth Valley Royal Hospital, Larbert FK5 4WR, 3 West of Scotland Pancreatic Unit and Glasgow Royal Infirmary, Alexandra Parade, Glasgow G31 2ER, 4 Beatson Institute for Cancer Research, Glasgow G61 1BD, 5 Departemnt of Medical Oncology, Beatson West of Scotland Cancer Centre, Glasgow G12 0YN, 6 Public Health Institute for Health and Wellbeing, University of Glasgow, Glasgow, 7 Department of Pathology, Southern General Hospital, Greater Glasgow and Clyde NHS, Glasgow G51 4TF, 8 Wolfson Wohl Cancer Research Centre, Institute of Cancer Sciences, College of Medical Veterinary and Life Sciences, University of Glasgow, Garscube Estate, Bearsden G61 1QH, UK. E-mail: draliasif7@gmail.com

Content

Pancreatico-biliary adenocarcinomas (PBA) have a poor prognosis. Diagnosis is usually achieved by imaging and/or endoscopy with confirmatory cytology. Cytological interpretation can be difficult especially in the setting of chronic pancreatitis/cholangitis. Immunohistochemistry biomarkers could act as an adjunct to cytology to improve the diagnosis. Thus, we performed a meta-analysis and selected KOC, S100P, mesothelin, and MUC1 for further validation in PBA resection specimens.

Technology

Hamamatsu (Japan):

  • NanoZoomer Digital Pathology Scanner Leica Biosystems (UK) Ltd., and SlidePath (Dublin, Ireland)
  • Digital Image Hub
  • Distiller (Slidepath Digital Slide Solutions).


Design

Tissue microarrays containing tumor and normal cores in a ratio of 3:2, from 99 surgically resected PBA patients, were used for immunohistochemistry. Immunohistochemistry was performed on an automated platform using antibodies against KOC, S100P, mesothelin, and MUC1. Stained slides for each antibody were scanned in NanoZoomer Digital Pathology Scanner and uploaded as dynamic digital images in Digital Image Hub and Distiller. Tissue cores were then digitally scored for staining intensity and proportion of tissue stained using Histoscore method (range, 0-300). Sensitivity and specificity for individual biomarkers, as well as biomarker panels, were determined with different cut-off for positivity.

Results

The expression of all four biomarkers was high in PBA versus normal ducts, with a mean Histoscore of 150 versus 0.4 for KOC, 165 versus 0.3 for S100P, 115 versus 0.5 for mesothelin and 200 versus 14 for MUC1 (P < 0.0001 for all comparisons). Five cut-offs were identified for sensitivity/specificity analyses, namely 5%, 10%, or 20% positive cells, Histoscore 20 and moderate-strong staining intensity using receiver operating characteristics curves. Using 20% positive cells as a cut-off achieved higher sensitivity/specificity values: KOC 84%/100%; S100P 83%/100%; mesothelin 88%/92%; and MUC1 89%/63%. Analysis of a panel of KOC, S100P, and mesothelin achieved 100% sensitivity and 99% specificity if at least 2 biomarkers were positive for 10% cut-off; and 100% sensitivity and specificity for 20% cut-off. The panel approach was facilitated in Distiller.

Conclusion

A biomarker panel of KOC, S100P, and mesothelin with at least 2 biomarkers positive was found to be an optimum panel with both 10% and 20% cut-offs in resection specimens from patients with PBA.

Optical Coherence Tomography Visualization of Breast Tissue

Matthew G. Hanna 1 , Alexis Bruhat 2 , Jeffrey L. Fine 2

1 Department of Pathology, Icahn School of Medicine at Mount Sinai, The Mount Sinai Hospital, New York, NY, 2Department of Pathology, University of Pittsburgh School of Medicine, Pittsburgh, PA, USA. E-mail: matthew.hanna@mountsinai.org

Content

Optical coherence tomography (OCT) permits rapid, nondestructive tissue microscopy without the need for microscope slides. OCT images can strongly resemble medium-magnification whole slide images (WSIs), thereby making it an attractive alternative or adjunct to traditional microscopy in selected clinical applications such as intra-operative consultation. This study presents OCT breast histopathology, correlated with traditional Hematoxylin and Eosin (H&E) histopathology.

Technology

Forty-five formalin-fixed tissue specimens from nine patients were imaged by OCT (LightCT, LLTech SAS, Paris France), at 1.6 micron transverse (x,y) resolution by 1 micron axial (z) resolution. Wide-field two-dimensional (2D) slices were acquired, at multiple tissue depths. OCT image data high-resolution JPEG and 16-bit gray scale images. Following OCT, tissue was submitted for H&E histology, then WSIs were made at 0.5 microns per pixel (Aperio ScanScope XT, Leica, Vista CA, USA).

Design

Signed-out cases were examined with an emphasis on finding interesting pathological findings; tissue block sized pieces were collected. After imaging OCT and WSIs were compared in order to find correlating, recognizable landmarks that were then used to identify subtle structures. Tissue findings were retrospectively categorized. Specimen inking was also assessed.

Results

OCT acquisition required 10-20 min per 2D slice, each about 200 megapixels in size. OCT images were immediately viewable prior to export. Large landmarks included fat/vessel patterns and tissue shape. Invasive carcinoma showed low signal, versus benign stroma, which was bright and shiny with organized collagen bundles. Benign ducts appear as low signal areas. These features were visible with practice but lacked contrast. Ink was bright and easily seen [Figure 1].



Conclusions

OCT is microscopy and pathologists can recognize many histopathologic features in these images. Tissue contrast is lacking and needs to improve; presence of subtle, shadow-like features may be amenable to image processing or image analysis to facilitate diagnosis. While OCT may not replace WSIs as-is, it may be useful in selected applications where rapid, nondestructive imaging is required. Further work is required, but these results are promising.

Evaluation of Panoramic Digital Images Using Panoptiq for Frozen Section Diagnosis


Dinesh Pradhan 1 , Sara E. Monaco 1 , Anil V. Parwani 1 , Ishtiaque Ahmed 1 , Liron Pantanowitz 1

1 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: pradhand@upmc.edu

Content

Whole slide imaging (WSI) permits intraoperative consultations (frozen sections) to be performed remotely. However, WSI files are large and can be problematic if there are tissue artifacts (e.g., folds) or when slides are scanned without multiplanes (z-stacks) to permit focusing. The Panoptiq imaging system allows users to create their own digital files that combine low power panoramic images with regions of interest that are imaged using high power z-stacks. The aim of this study was to determine the utility of Panoptiq for frozen section telepathology

Technology

Glass slides were digitized using a Prosilica GT camera (model GT1920C, Allied Vision Technologies) attached to an Olympus BX45 microscope and Dell Precision Tower 810 computer (Dell). Panotiq 3 version 3.1.2 software was used for image acquisition and Panoptiq View version 3.1.2 to view images (ViewsIQ, Canada). These digital slides were viewed using ImageScope software (Aperio ePathology, Leica, US) on HP ZR24w (1920 × 1200) monitors.

Design

Twenty randomly selected genitourinary pathology frozen section cases were evaluated using conventional light microscopy and digital slides created using Panoptiq and Aperio. Image acquisition using Panoptiq software involved a pathology resident manually creating digital maps (×4 objective) and then selecting representative regions of interest to generate z-stacks at higher magnification (×40 objective). Three pathologists were asked to render diagnoses and rate the image quality (1-10) and their diagnostic confidence (1-10) for each modality.

Results

The table compares glass slide, Panoptiq images, and Aperio WSI evaluations. Complaints regarding WSI included poor focus near tissue folds and air bubbles. Panoptiq permitted fine focusing on tissue folds [Figure] and bubbles. Issues with Panoptiq images included low resolution of ×4 image maps and tiling artifacts. In some cases, the z-stacked areas scanned were considered by pathologists to be limited or nonrepresentative regions of the sample.



Conclusions

The Panoptiq imaging system is a novel tool that can be used for frozen section telepathology. Panoptiq digital images are easy to generate and navigate, of relatively small file size, and offer a mechanism to overcome focusing problems commonly encountered with whole slide images of frozen sections. However, the acquisition of representative Panoptiq images is dependent upon a trained individual creating these files.

Concordance of Digital Imaging in Neuropathology Intraoperative Consultation


Matthew Hanna 1 , Cheng-Hsuan Chiang 2 , Clayton Wiley 2 , Laura Drogowski 3 , Simran Parwani 3 , Liron Pantanowitz 2 , Anil V. Parwani 2

1 Department of Pathology, Icahn School of Medicine at Mount Sinai, The Mount Sinai Hospital, New York, NY, 2 Department of Pathology, University of Pittsburgh Medical Center, 3 Omnyx™, Pittsburgh, PA, USA. E-mail: matthew.hanna@mountsinai.org

Content

Neuropathology intraoperative consultation (IOC) is being performed by telepathology in several medical centers. These cases are unique, because submitted tissue is generally scant and involves a combination of cytology preparations (smears, touch preparations) and frozen sections. The aim of this study was to evaluate the performance of the Omnyx digital imaging system for neuropathology IOC.

Technology

Glass slides were scanned using Omnyx™ VL4 at ×20. Cases were created and viewed using a dual monitor Omnyx™ Integrated Digital Pathology solution (version 1.2) (Omnyx, Pittsburgh, PA, USA).

Design

Retrospective analysis of 104 slides collected from 76 parts of 65 sequentially accessioned cases was selected from IOC between 2013 and 2014. Cases included inflammatory and neoplastic lesions [Figure 1a]. Slides included 43 touch preparations, 40 smears, and 21 frozen sections. Each case was presented to the pathologist with the clinical history and radiology available at the time of the original IOC, to simulate real life diagnostic conditions. Two neuropathologists reviewed all digitized slides and identified the assessability of the image, categorical and descriptive diagnosis [Figure 1b], and quality of the image. Slide quality was rated from 1 (unacceptable) to 7 (excellent). Times to diagnosis and rescan rates were also noted.



Results

Concordance rates between glass and digital slides for the categorical and descriptive diagnoses were 93%, 83% and 92%, 72% for both pathologists, respectively. Interobserver concordance for all 76 parts was 91% and 89% for categorical and descriptive diagnoses. Accuracy was 89.5% and 84%. Both pathologists had a mean time to diagnosis of 96 and 97 s (range 25 s - 4 min 41 s). The quality assessment of the scanned slides had a mean score of 6 (out of 7) for the digitized slides by both pathologists. Four slides (3.8%) needed rescans because of air bubbles. Touch preparations had the highest deferral rate and most discordant cases. Frozen sections had the least deferrals and did not show an association with concordance.

Conclusions

The Omnyx system performed well with neuropathology IOC. Using this imaging system, the data show high concordance rates with glass slide interpretation. Scanned touch preparations are more challenging than frozen sections because they were associated with higher deferral rates and discordant cases.

IT-tools to Overcome Integration Problems in Digital Microscopy within Collaborative Research Networks

Lika Svanadze 1 , Thomas Franke 1 , Karoline Buckow 1 , Erik Bahn 2 , Otto Rienhoff 1

Departments of 1 Medical Informatics and 2 Neuropathology, University Medical Center Göttingen, Germany. E-mail: lika.svanadze@med.uni-goettingen.de

Content

Laboratory Information Management Systems (LIMS) are a comprehensive solution for a wide range of laboratory processes and provide valuable information regarding specimen. Despite the powerful features of modern LIMS, there is a growing need for the integration of the new data sources - e.g., from new pathology methods. Within the scope of an infrastructure research project for the Competence Network Multiple Sclerosis, high resolution digital microscopic images should be made easily accessible while viewing the histology-specimen from which they originate. This requires integration of digital microscopy features into the LIMS. Here, we present automated and manual tools by which LIMS can be interfaced with Digital Pathology System.

Technology

An Olympus dotSlide virtual microscopy system was used for digitization process. Digital slides are stored on a Net Image Server and laboratory information regarding the specimen is registered into the STARLIMS software.

Design

Microscopic images are handled by a Digital Pathology System. A web-based image database was set up and configured. Automated and manual tools were defined for the linkage of digital images to the corresponding specimen into the LIMS.

Results

Digital images in conjunction with metadata are stored in the Digital Pathology System and they are freely accessible via the web-based viewer from within our institution's secure network. Most LIMS allow customers to rapidly configure the system in a way that gives us the ability to easily customize a new tab for scanned images in order to manually link microscopic images to the corresponding specimen and facilitate viewing of specimen in conjunction with the digital images. Furthermore, our STARLIMS supports connectivity for multiple external applications via SOAP-based web services. Using this web services interface, multiple images could be automatically linked to the corresponding specimen using unique identifiers.

Conclusions

The proposed methods deliver a layout for an effective prototype for the integration of the digital microscopic images into the LIMS for the Competence Network Multiple Sclerosis IT infrastructure. These tools can significantly reduce workload and eliminate handling errors of specimen and this increase the quality of the collaborative research project.

Acknowledgments

This work was supported by the Competence Network Multiples Sclerosis (01GI1304B), funded by the German Federal Ministry of Education and Research.

Novel Computational Image Features For Prostate Cancer Grading

Keluo Yao 1 , M. Khalid Khan Niazi 2 , Debra L. Zynger 1 , James Chen 2 , Steven Clinton 3 , Mehmet Koyutürk 4 , Tom LaFramboise 5 , Metin Gurcan 2

Departments of 1 Pathology and 2 Biomedical Informatics, 3 Department of Internal Medicine, Comprehensive Cancer Center, The Ohio State University, Departments of 4 Electrical Engineering and Computer Science and 5 Genetics and Genome Sciences, Case Western Reserve University, Cleveland, OH, USA. E-mail: keluo.yao@osumc.edu

Content

The Gleason score is a histopathologic system for manually grading prostate cancer which is utilized for clinical management. Attempts to automate grading have failed to gain acceptance due to lack of visually meaningful and clinically interpretable features. The aim of this study was to develop, calibrate, and validate novel, and histopathologically meaningful features to grade prostate cancer.

Technology

Digitalized slides from The Cancer Genome Atlas (TCGA) and The Ohio State University (OSU) were obtained. Digital slides were annotated and tumor regions were selected by an expert pathologist and extracted with an Aperio Imagescope. Image analysis software was developed at The Clinical Image Analysis Laboratory at OSU.

Design

Totally, 14 Gleason ≤6 and 29 Gleason ≥8 images were extracted from TCGA slides and 49 Gleason ≤6 and 39 Gleason ≥8 images were extracted from OSU slides. Computer image analysis consisted of: segmentation, feature extraction, and classification. In segmentation, the software divided the tumor region into stroma and luminal spaces. Glandular cells were extracted by computing the shortest path from the nuclei to their closest luminal spaces. From these data, we further extracted: ratio of glandular nuclei to the total number of nuclei and average shortest distance of all nuclei to their closest luminal spaces. Architectural changes were quantized using directional filter banks. TCGA images were used as a training set to build the classifier using the subspace learning method and OSU images were used to test the classifier.

Results

The classifier achieved 93.0% and 97.6% accuracy in differentiating Gleason ≤6 from Gleason ≥8 in the TCGA training and OSU test set, respectively. In both TCGA and OSU image sets, the ratio of glandular nuclei to the total number of nuclei was the most significant contributor to the classification.

Conclusions

We have developed a novel set of image features that can distinguish high and low risk Gleason scores using objective criteria. In the future, our investigation will focus on elucidating image features to differentiate the Gleason score of 7 as well as automated cancer region detection.

Primary Diagnosis of Lung Biopsies in Uncut Specimens Visualized Using Stackstreamer

Richard Torres 1 , Michael Levene 2 , Eben Olson 1

1 Department of Laboratory Medicine, Yale School of Medicine, New Haven, 2 Applikate Technologies, LLC, East Haven, CT, USA. E-mail: richard.torres@yale.edu

Content


Small lung biopsy samples present a significant challenge for pathologists in balancing the need for sufficient sections for visual diagnosis against tissue preservation for ancillary studies. We have been developing new methods for ex-vivo tissue imaging of uncut, un-embedded samples using multiphoton microscopy (MPM). In this study, we sought to evaluate our method as applied to rendering primary lung biopsy diagnoses. As histology z-stacks are not easily reviewable from a remote server using readily available software, we employed a web-based interface that we have developed for remote navigation of three-dimensional histology image sets of arbitrary size (Stackstreamer™).

Technology

Tissue processing was performed as previously described, consisting of sample dehydration in methanol prior to clearing with benzyl alcohol/benzyl benzoate. Staining was done with fluorescent dyes and images were pseudo-colored using inversion of Hematoxylin and Eosin color convolution matrices. Imaging was performed on a custom multiphoton microscope controlled with Scanimage software and specialized MATLAB functions. Postprocessing was done in FIJI and with Python scripts. Stacks were subdivided into layers, each composed of "pyramidal" structures of variable resolution tiles, analogous to DeepZoom, but with addition of z-levels. The Stackstreamer viewer was written in JavaScript using WebSockets.

Design

Paired core biopsies of fresh and formalin-fixed human lung nodule tissue samples were obtained from discarded pathologic tissue at resection. Image stacks were evaluated by two anatomic pathologists who independently assessed primary diagnostic quality and compared them subsequently to traditionally processed specimens. Visualization was done both with locally loaded images using Windows PhotoViewer and remotely using Stackstreamer.

Results

All cleared samples analyzed with MPM yielded images comparable to traditional Hematoxylin and Eosin sections [Figure 1] at depths up to several hundred microns, and were available within same day of tissue receipt. Primary diagnosis from image stacks was possible with remote viewing using Stackstreamer with only small lag-effects.



Conclusions

On the basis of this qualitative analysis, clearing/MPM appears to be a practical method for lung biopsy primary diagnosis that does not consume tissue, increases reviewable volume, requires less manual skilled labor, avoids cutting artifacts, and can be adequately visualized remotely using Stackstreamer.

Location and Proximity Sensing Using Mobile Devices in the Hospital Environment: A Technical Survey to Establish Performance

Christopher Metts 1 , Jeff Szymanski 1 , Ronald Jackups 1

1 Department of Pathology and Immunology, Washington University School of Medicine, St. Louis, MO, USA. E-mail: cmetts@path.wustl.edu

Content

Mobile devices contain sensors that provide data to add context to the user's interaction with an application. In particular, their location and proximity to other users, events, or structures can be used to predict information needs. Location sensing indoors without the benefit of GPS is challenging. Triangulation of radio signals is used, but with diminished accuracy. Methods to improve accuracy include audits of radio signals and installation of strategically placed Bluetooth radios. The purpose of this study was to survey the accuracy of indoor location sensing on our campus. By determining baseline performance, results of efforts to improve accuracy can be quantified.

Technology

Map overlay production: MapTiler (Klokan Technologies, Unterager, Switzerland). Web application: Python on Google App Engine iPhone application: Objective-C.

Methods

A list of forty-three points-of-interest (POI) on campus was compiled in a web application, with coordinates determined using Google Maps. A map of the campus was geo-referenced and tiled for overlay. The list and tiles were served to an iPhone app, in which POI were presented in tabular format sorted by distance from the user. Apple's Maps SDK was used to render the tiles as image overlays, POI as push-pins, and the predicted location. A survey of 100 locations distributed throughout main corridors was conducted. At each location, the surveyor indicated their location on the map in the iPhone application. The coordinates for the predicted and actual locations were uploaded to the web application for analysis.

Results

The mean discrepancy between actual and predicted coordinates for all survey locations is 7.8 m (SD 5.3 m), the median is 7.3 m, and the range is 0-25.6 m. Accuracy is highest in the windowed walkways connecting the main buildings, and lowest in internal hallways.

Conclusions

Mobile applications using location and proximity sensing can be used for orientation of new staff with the performance available on our campus. Efforts to improve accuracy would likely increase the usefulness of applications for patients and visitors. Installation of Bluetooth radios is required to elevate accuracy to a level appropriate for clinical and operations support, including patient, provider, and specimen location.



Hepatitis C Virus Genie: A Web-based Hepatitis C Virus Interpretation Platform for the Versant Hepatitis C Virus Genotype 2.0 Line Probe Assay

Abha Soni 1 , Seung L. Park 1

1 Department of Pathology and Informatics, University of Alabama at Birmingham, Birmingham, AL, USA. E-mail: abhasoni@uabmc.edu

Content

Hepatitis C virus (HCV) genotyping at our institution is performed using the Versant HCV genotype 2.0 Line Probe Assay (LiPA). The last step of this procedure is a manual, time-consuming, and error prone process that involves the comparison of bands on a test strip to a physical reference table. The aim of this project was to develop a web-based HCV genotype interpretation platform that would (a) minimize interpretation time, (b) reduce error, and therefore (c) increase the quality of patient care delivered through this test methodology.

Technology

Server Hardware: Dell Precision T3600; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 14.04 LTS 64-bit; Web Server: nginx 1.7; Database Management System: MariaDB 10.0; Programming Language: PHP-FPM 5.5; User Interface Framework: Twitter Bootstrap 3.3.

Design

Using the virtual machine, the resident (a) Designed, built, and deployed and automated HCV genotype interpretation program called "HCV Genie," (b) populated the reference database of that program utilizing a comma-separated value derived from Versant's physical reference table, (c) built a user interface for band pattern query, and (d) clinically validated the final program against the current manual interpretation methodology.

Results

Our program was written, deployed, clinically validated, and proven to be identical to human expert interpretation (n = 200) over the course of 2 weeks. It decreases the time needed to interpret results by 53% in residents, but results among experienced laboratory technicians are more equivocal.

Conclusion

Our program provides results that are identical to the manual workflow, but (a) With reduced manual steps and (b) in a timeframe similar to that of the most well-trained manual interpreter, regardless of the program user's experience level. Future iterations of this program will focus on minimizing user input even further, and as user experience with our program grows, we expect further decreases in interpretation time.

Utility of the Lytro Light-field Camera as a Microscope Image Capture Platform and Ancillary Tool for Pathology Specimen Interpretation that benefits from Examination of Multiple Focal Stacks

Grzegorz T. Gurda 1 , Sara E. Monaco 1 , Somak Roy 1 , Jeffrey L. Fine 1

1 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: gurdag@upmc.edu

Content

Photography of microscopic specimens in anatomic and clinical pathology is a requisite to sharing an image obtained from a glass slide for diagnostic consultation, documentation of clinical reports, education and publication. Traditional top-mounted microscope cameras are expensive, difficult to upgrade, and often require proprietary/nonstandardized image processing; alternatively, eyepiece-mounted smart phone cameras are intuitive and inexpensive, but may be difficult to operate due to unstable adapters or pose risk to patient privacy with loss/theft of the device. Here, we developed a stable and inexpensive platform using a Lytro camera to obtain static photographs with image quality on par with smart phone cameras. In addition, the light-field technology creates multiple focal stacks, a potential diagnostic benefit.

Technology

Lytro™ camera and mount adapter were purchased from Lytro, Inc (Mountain View, CA, USA). MicroMount 38 mm MMT1 anodized aluminum microscope ocular adapter was purchased from CNC Parts Supply, Inc. (Cape Coral, FL, USA). Optional 37 mm magnetic lens adaptor is available from Viewpoint Labs, LLC (New Haven, CT, USA). For comparison, additional images were obtained via Nikon Digital Sight DS-Fi2 top-mount camera (Toyko, Japan) and LG G3 smart phone with a Skylight adapter.

Design

Ocular mounted Lytro camera was used to obtain images, which were retrieved as static image files (JPEG), z-stacks (TIFF) and focal stack movie files (MP4). Setup, image acquisition time, image processing, image size, and image quality were assessed and compared to top-mounted professional camera and a smart phone camera. Utility of multiple focal stacks was also evaluated.

Results

Lytro camera setup is fast (30 s or less) and easily adaptable to any personal microscope. Images can be acquired rapidly. Image quality is on par with a smart phone camera and only somewhat inferior to a professional top-mounted camera. Postimage processing requires a free proprietary Lytro software suite, but creates standard and relatively small (500 KB-2 MB static, 1-20 MB movie) files amendable to easy sharing, storage and further processing or analysis. Image capture of multiple focal stacks is useful to evaluate specimens that benefit from a review of multiple focal planes. For surgical pathology, we evaluated: breast microcalcifications, mitosis versus debris artifact and intracellular inclusions (i.e., hyaline bodies). For cytopathology, we evaluated: urine crystals, degeneration versus viral effect, viral effect versus high grade dysplasia and mucin-containing adenocarcinoma versus mesothelioma.

Conclusion

Lytro camera platform is inexpensive and easy to use. Capture of multi-focal stacks (particularly as movies) may be useful for pathology specimens whose interpretation benefits from examination of multiple focal planes. Lytro camera platform is a reasonable, dedicated alternative to smart phones and a potentially useful ancillary image acquisition tool to traditional, static images, including whose slide image technology.

Strategies to Optimize Data Storage for Clinical Next Generation Sequencing

S. Roy 1 , J. McHugh 1 , R. Mitchell 1 , G. Burdelski 1 , A. V. Parwani 1 , A. I. Wald 1 , K. Callenberg 1 , L. Santos 1 , S. Zhong 1 , M. N. Nikiforova 1 , Y. E. Nikiforov 1 , L. Pantanowitz 1

Departments of Pathology, Information Services Division, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: roys@upmc.edu

Content

Data storage for next generation sequencing (NGS)-based assays constitutes a significant component of clinical laboratory operations. Scalable, redundant, and secure infrastructure solutions are essential to manage analyzed NGS data. With continuously evolving sequencing technology, laboratory workflow and cumulative experience with data analytics, strategies for optimal utilization of storage infrastructure resources is surfacing as an important aspect of data management. The aim of this study was to present our institution's experience with data storage for NGS.

Technology

The molecular and genomic pathology laboratory houses six NGS sequencers with or without accompanying high-performance workstation servers for data analytics and local storage (7-10 terabytes per server). Resource intensive data analytics (e.g., whole transcriptome) is performed on a high performance compute (HPC) cluster, provisioned by our institution's HIPAA-compliant data center with onboard "hot" storage. Scale out network attached storage (SONAS) and high-performance enterprise storage (HPES) at our data center were used for archiving clinical NGS test data (cold storage).

Design

Using high-bandwidth network connections, raw signal files were moved to the SONAS archives whereas intermediate data (FASTQ-BAM-VCF) were stored in the HPES pool. The latter provided low-latency access to FASTQ and other files for alternate bioinformatics analyses that were run on virtual machines and the HPC cluster. Downstream test results, including annotated genomic alterations and knowledgebase, generated by custom developed web applications were hosted and stored in virtualized database servers in our data center.

Results

With initial NGS based testing, raw data files (100-250 GB per run) and intermediate data (100-500 MB) were archived to allow re-analysis and alternative analysis, respectively. However, increasing test volumes and introduction of new NGS assays (new gene panels, gene fusion, and copy number variation analysis) and novel workflows required optimization of allocated storage. Such strategies included creation of data retention policies to limit storage of relevant raw data, data compression, and optimal use of our low-latency storage pool. Institution wide disaster recovery (DR) plans enhanced storage reliability and security, but also helped define data types based on relevance to patient care. All NGS data were stored off-site in our data center as part of this DR plan.

Conclusion

The steady plummet of storage cost has greatly facilitated large-scale NGS data storage. However, strategies to optimally manage stored data are becoming increasingly important because of the increasing demand for NGS testing, competing capital needs, complexity of Big Data, and need to re-analyze archival data.

Convolutional Neural Networks for Segmentation of Multiphoton Microscopy Images

Eben Olsen 1 , Richard Torres 1

1 Department of Laboratory Medicine, Yale University, New Haven, CT, USA. E-mail: eben.olson@yale.edu

Content

The combination of optical tissue clearing with multiphoton microscopy allows imaging of tissue sections greater than 1 mm in thickness, while maintaining subcellular resolution. Three-dimensional imaging increases observation coverage and can enable novel methods of quantitative morphometric analysis. Tools for automated processing provide vital aid in handling the greatly increased data volume. Convolutional neural networks (CNNs) are powerful tools for image analysis. They consist of multiple layers of convolutional filters followed by nonlinear transformations, and are capable of learning hierarchical feature representations directly from pixel data. Due to increasing computational power and numerous practical innovations, in recent years, CNNs have been established as a preeminent technique for image classification, object detection and image segmentation. We sought to apply these techniques to segmentation of glomeruli in image stacks of human kidney.

Technology

Images were acquired using previously described methods of optical tissue clearing and a custom multiphoton microscope. Networks were implemented in Python using the Theano symbolic computation framework, which enables seamless GPU acceleration. A fully-convolutional architecture was used to allow efficient training and prediction on large images.

Design

Portions of formalin-fixed human kidney were obtained from discarded tissue at autopsy or resection. Blocks approximately 5 mm × 5 mm × 1 mm in size were processed and imaged to a depth of 500 microns with a step size of 5 microns. For training and evaluation, segmentation maps highlighting glomeruli were manually drawn using an ImageJ plugin. Trained networks were applied to classify the full data sets, producing three-dimensional segmentation volumes. Results were compared to manual segmentation.

Results

We find that CNNs can be trained to effectively segment glomeruli in multiphoton images, enabling automated counting and estimation of glomerular volume and shape with minimal human effort. We also find that semi-supervised learning can leverage the large amounts of unlabeled image data made available by volumetric microscopy to aid in training.

Conclusions

CNNs are an effective tool for segmentation of tissue images, reducing the human labor required to analyze large volumetric data sets. Use of these techniques may enable novel methods of analysis which would otherwise be infeasible due to time and effort required.

Asterism: A Web 2.0 Molecular Diagnostics Templating, Reporting, and Analytics System

Rance C. Siniard 1 , Shuko Harada 2 , Seung Park 1

1 Department of Pathology, Division of Informatics, University of Alabama at Birmingham, 2 Department of Pathology, Section of Molecular Diagnostics, Division of Anatomic Pathology, University of Alabama at Birmingham, Birmingham, AL, USA. E-mail: chad.siniard@gmail.com

Content

Molecular diagnostics is an integral part of patient care in the modern academic medical center, and the usefulness of the immense amount of data that these tests generate is undisputed. However, these data are not easily integrated into sign-out reports, resulting in manual entry of data. This (a) Presents a persistent source of human error and (b) impedes efforts in data analytics, often forcing the use of natural language processing. In order to solve these problems at our institution, we have developed, tested, and deployed a modern, Web 2.0 molecular genetics templating, reporting, and analytics system codenamed "Asterism."

Technology

Server Hardware: Dell Precision T3600; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 14.04 LTS 64-bit; Web Server: nginx 1.7; Database Management System: MariaDB 10.0; Programming Language: PHP-FPM 5.5; User Interface Framework: Twitter Bootstrap 3.3.

Design

Asterism distills narrative molecular diagnostics reports into discrete data elements before the report ever hits the LIS. We do this by offering a report templating and generation system that intelligently offers its users limited numbers of choices based on the information being provided. This system generates a complete report to be inserted into the upstream LIS, while storing all relevant information as discrete data elements in its own database.

Results

Asterism has been implemented, tested, and deployed for our molecular genetics pathology service. In preliminary testing, asterism has been found to (a) reduce the paperwork phase time of sign-out by up to 50% and (b) significantly reduce human error in transcription (e.g., from elements in a Microsoft Word template that were left in the final report by accident). Since all clinically relevant information is stored in asterism's databases [Figure 1], it will become a powerful data warehouse for molecular diagnostics as usage increases.



Conclusions

Asterism provides increased efficiency in the generation of complete sign-out reports for the molecular genetics service. In addition, the data are now efficiently stored for future research, patient care, or educational use. In future releases, search capabilities will be added, as well as a user-management system to more easily assign cases to pathologists on the service.

Critical Value Notification in Downstream Systems

Douglas J. Hartman 1 , Darlene Sutara 2 , James Blumer 2 , Shangcong Zeng 2 , Anthony Piccoli 2 , Richard Ambrosino 3

1 Department of Pathology, 2 Information Services Division, University of Pittsburgh Medical Center, 3 Department of Biomedical Informatics, University of Pittsburgh, Pittsburgh, PA, USA. E-mail: hartmandj@upmc.edu

Content

Criteria for values within anatomic pathology that constitute critical value results have been described. In the increasingly demanding field of medical delivery communication of critical values can be lost within the day-to-day demands of clinical service.

Technology

We have identified a mechanism to notify the clinical teams when a critical value in anatomic pathology has been identified. The critical value is assigned within the Pathology Laboratory Information System (Cerner CoPath) by "TC66." We expanded a preexisting structure for notifying clinicians of postdischarge laboratory values for notification of critical values in anatomic pathology.

Design

We generated a signal within the headers of our outgoing anatomic pathology reports that could be parsed. Using the outbound HL7 report feed, the Message Router monitors the feed for the critical result indicator, TC66 using a string search. When encountered, Message Router directs these data to a separate flow. Key fields in the message are used to determine MedTrak logic: Attending Doctor Last^First Send as is; Referring Doctor Last^First Send as is; Consulting Doctor Last^First Send as is; Admitting Doctor #####^Last^First Send as is; Results Copies To ######^Last^First~ Send as is.

Results

When the critical value string is identified, two things occur: (a) The report content is formatted into a PDF based on a predefined template and (b) the HL7 message is transformed into the agreed format for the notification system (MedTrak). A diagram of the process is below. The reformatted message is directed to MedTrak for delivery thru the Automated Notification process.

Following is a visual of the process:



Conclusion

This notification system represents an improved mechanism to notify clinicians when a critical anatomic value is identified. The solution for this problem required an interdisciplinary team of both medical directors and directors both within the originating system, the downstream systems and the notification systems. Future changes in healthcare are going to necessitate more projects across interdisciplinary teams such as this project to provide a safer healthcare environment, effective/efficient communication between clinical team members and optimize clinical care delivery.

Interfaces Between Digital Pathology Image Management Systems and Anatomic Pathology Laboratory Information Systems


Nicholas C. Jones 1 , Christopher Garcia 1 , Luigi K. Rao 2 , James Floyd 1 , John Gilbertson 1

1 Department of Pathology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, 2 Department of Pathology and Laboratory Services, Water Reed National Military Medical Center, Bethesda, MD, USA. E-mail: ncjones@partners.org

Content

Scalable clinical application of whole slide images (WSIs) will require system communications between Anatomic Pathology Laboratory Information Systems and Digital Pathology Image Management Systems for scan ordering, image use, and workflow purposes. Analysis of the implementation and management of a Health Level 7 interface has yielded insight into the benefits and current limitations of such an interface.

Technology

Philips Digital Pathology Image Management System (up to version 2.3.1.1) and Philips Ultra Fast Scanner (Philips Digital Pathology, Best, The Netherlands), CoPath Plus (up to version 6.1.1006, Tucson, Arizona, USA).

Design

A bidirectional Health Level 7 interface was implemented between the two systems. Orders on cases in the Anatomic Pathology Laboratory Information System would send select information about the case and slide to the Image Management System. Upon successful scanning, the Image Management System reads the barcode and attributes the case and slide information to the scan data set. It then passes a link for the WSI back to the Anatomic Pathology Laboratory Information System. The system was has been employed for use over the past 14 months in scanning for clinical conferences and education.

Results

The interface was effective in passing basic information between both systems including the slide unique identifier, the stain, stain type, and a hyperlink for the WSI. This allowed for the two systems to achieve the fundamental requirements for clinical management; in 2014, over 2000 slides were scanned through the interface.

Conclusion

The interface implementation has been successful, but further development of the interface could be beneficial. Interfacing information like the case structure (including specimen and block information) could allow for better and more automated organization within the Image Management System. As content, features, and workflow tools are created and evolve in both types of systems to handle WSI workflow, the information shared by the interface will need to grow and evolve as well.

Harnessing Web Application Programming Interface for Optimizing Clinical Next Generation Sequencing Workflow


S. Roy 1 , S. Zhong 1 , R. Mitchell 2 , G. Burdelski 2 , J. McHugh 2 , D. Maglicco 2 , A. V. Parwani 1 , L. Pantanowitz 1 , Y. E. Nikiforov 1 , M. N. Nikiforova 1

1 Department of Pathology, 2 Information Services Division, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: roys@upmc.edu

Content

Application programming interface (API) is a set of protocols that facilitate software development by exposing useful functions and content for consumption by other applications either at a system level or across a network (Web API). The latter is a popular component of modern data-driven web applications that has recently gained popularity for sharing large-scale genomic datasets. The aim of this project was to use Web API based messaging for transfer of genomic data to improve our laboratory's next generation sequencing (NGS) workflow.

Technology

Our pipeline consisted of servers performing primary bioinformatics analysis on semiconductor-based raw sequence data and custom developed web applications, VariantExplorer and SeqReporter, for downstream data analytics. VariantExplorer also used novel algorithms for processing copy number variation (CNV) and gene fusion data. The sequencing servers and VariantExplorer were hosted in an Ubuntu Linux operating system. SeqReporter was deployed in Microsoft Windows Server operating system. Django framework v1.5 and ASP.NET framework v4.5 were used to develop VariantExplorer and SeqReporter, respectively. The web API was hosted by VariantExplorer and used JavaScript Object Notation format for messaging. All data communications occurred over high-bandwidth secure network (1 GB/s).

Design

Upon completion of primary bioinformatics analysis, technologists submitted analysis jobs to VariantExplorer, which extracted the genomic data (BAM and VCF files) from the sequencing servers. Subsequently, VariantExplorer enabled asynchronous consumption of the annotated DNA variants as well as CNV and gene fusion analyses results by SeqReporter using a set of web APIs. The received data were then further annotated with our in-house knowledge base and converted into a LIS-compatible clinical report.

Results

A total of 344 cases across different NGS gene panels (thyroid, advanced solid tumor, and brain tumor panels) were analyzed using the new framework. In contrast to the older method that relied on manual uploads of genomic data, the web API-based method substantially streamlined the data analysis workflow in our molecular laboratory. There was a significant decrease in user interaction and input during the entire process enabling more hands-off time, which also significantly decreased probability of human errors, especially with constantly increasing test volumes. A real time monitoring interface enabled easy tracking of background data analytic processes in both applications.

Conclusion

High-complexity and high-volume clinical testing with NGS assays require close interactions between bioinformatics and clinical informatics systems to optimize laboratory workflow. Web API employed in our laboratory's NGS workflow successfully bridged this gap to enable seamless communications in an operating system agnostic manner with promising preliminary results.

Clinical Reporting of Tumor DNA Sequencing results in Pathology

Peter Gershkovich 1 , Neil Mutnick 1 , John Sinard 1

1 Department of Pathology, Yale Medical School, New Haven CT, USA. E-mail: peter.gershkovich@yale.edu

Context

Clear and accurate reporting of tumor DNA Sequencing results is essential to fulfill the promise of personalized medicine. The information that goes into molecular pathology reports is complex and diverse. It varies in significance and comes from sources that are frequently updated. Our knowledge of genomic variations often changes and new treatment targets routinely become available. Nationally and internationally funded genomic projects with manually curated resources are essential in providing timely information to molecular pathologists. New types of technology are required to assist molecular pathologists in integrating information from all available resources.

Technology

Our "Downstream Reporting" software was written using the Java programming language and Open Source frameworks. Google Web Toolkit framework was used to build a web-based user interface that brings together external resources such as PubMed, Enembl, OMIM, GeneCards, Cosmic, KEGG pathways and Uniprot via RESTful API. MongoDb - a modern no-SQL database was used to store data.

Design

Downstream Reporting is designed to assist bioinformaticians who upload results of two parallel pipelines and Variant Effect Predictor files to produce a single Excel file with conveniently formatted raw data necessary to filter artifacts and select variants for further evaluation by molecular pathologists. It is also designed to assist molecular pathologists with access to discovered variants, patient information, previously reported variant archetypes and a range of external resources. Ultimately, this information is consolidated in a pathology report that is automatically uploaded into the Anatomic Pathology LIS.

Results

The system consolidates access to raw data and facilitates a higher level of abstraction that culminates in creation of a tumor DNA sequencing evaluation report. The system improves efficiency and accuracy of reporting through flexible integration with internal and external resources.

Conclusions

Accurate and clear reporting of tumor DNA sequencing results by molecular pathologists requires new types of technology solutions that can integrate already existing departmental resources with analytic pipelines and vast publicly available repositories of cancer genomic data. The Downstream Reporting application demonstrates a range of technical approaches that reduce the complexity of clinical reporting of tumor DNA sequencing results.

Internal Quality Assessment for Multicolor Flow Cytometry

Daniel S. Herman 1 , David Wu 1 , Brent L. Wood1, David Ng 1

1 Department of Laboratory Medicine, University of Washington, Seattle, WA, USA, E-mail: hermands@uw.edu

Content

Multicolor flow cytometry is used in the evaluation for leukemia and lymphoma to identify cell population with abnormal antigen distributions. This approach relies upon conscientious quality management. Our laboratory's quality program includes external controls, but depends highly upon manual review of individual cases by experienced technologists and pathologists. We are developing approaches to internally assess flow cytometry data variability with the goal of improving quality and potentially enabling automation, auto-verification, and better computer-aided interpretation.

Technology

Specimens were assessed in clinical practice using 10-color BD LSRII flow cytometers and two distinct reagent cocktails, which included different antibody-fluorophore conjugates (PE-Cy7 and PE-CF594) for the B-lymphocyte antigen CD19. We analyzed the fluorescence intensity data using python, sqlite3, and R.

Design

Retrospective analyses were performed on clinical samples processed in 2014. Fluorescent intensities were compensated, biexponential-transformed, rescaled (0-100), and gated roughly for lymphocytes (forward-scatter area >0.1 and side-scatter height <0.7). Focusing on CD19, 1D population peaks were identified, aligned between samples (Munkres algorithm), and compared by orthogonal linear regression. In addition, pairs of 25-bin, sample 1D distributions were compared using Earth Mover's Distance and rescaled (0-100).

Results

The peak fluorescence intensity of the CD19-positive population analyzed by a single instrument (n = 1602) had a mean of 75 and standard deviation of 3.3 in our myeloid blast assay and a mean of 73 and standard deviation of 3.0 in our B-lymphocyte assay. Directly comparing tests of the same specimen revealed a fixed bias of 6.7 and a proportional bias of 0.88 between assays (R2 x = 0.3, R2 y = 0.4). Pairwise comparisons of binned 1D distributions showed a mean distance of 1.2% between tests of the same specimen (n = 1677) and 2.4% between tests of different specimens (n = 5,621,304; P = 4 × 10 232). Linear regression suggests that assay- and sample-specific effects contributed to 0.05% and 0.02% of total between-test variation.

Conclusions

Current quality control practices for clinical flow cytometry permit considerable analytic variability. The between-assay bias identified for the CD19-positive population suggests data normalization could facilitate cross-assay analyses. Ongoing studies will better characterize this variation, address its impact on clinical interpretation, and develop methods to improve quality.

From Two to Three Dimensions in Digital Pathology: Three-Dimensional Analysis and Volume Rendering Using Routine Histology Sections


Beverly Faulkner-Jones 1 , Charles Law 2,3 , Stephen Turney 3 , Seymour Rosen 1

1 Department of Pathology, Beth Israel Deaconess Medical Center, Boston, 3 Center for Brain Science, Harvard University, Cambridge, MA, 2 Kitware Inc., Clifton Park, New York, USA. E-mail: bfaulkne@bidmc.harvard.edu

Content

Pathologists render a diagnosis from thin tissue sections mounted on glass slides. The ability to then infer three-dimensional structure from these two-dimensional tissue sections is crucial for the diagnosis and full characterization of many disease processes. This is dependent on the skill and experience of the individual pathologist. With advances in digital methods, it is now feasible to have a new workflow in which pathologists are presented with aligned whole slide images (WSIs) of serial sections. We are developing web-based tools for aligning and viewing WSI datasets generated from renal needle core biopsies as our test system. The expectation is for faster evaluation, greater objectivity, and improved diagnostic accuracy.

Technology

Serial sections from renal needle core biopsies were scanned at ×40 using a Philips Ultra Fast Whole Slide Scanner. The WSIs were viewed using our high-performance web-based client server system (http://slide-atlas.org). Tools were developed for local adaptive alignment of the WSIs and incorporated into Slide Atlas. Aligned WSIs are displayed side by side, and can be quickly manipulated and stepped through to find a region of interest that can be compared simultaneously in adjacent sections. Alignment information is stored with the images and applied interactively for display. Precise alignment is computed and maintained at different zoom levels. Structures of interest (e.g., glomeruli, tubular deposits) can be segmented and volume rendered.

Design

Serially sectioned renal needle core biopsies received in the routine diagnostic workflow were stained with our standard panel of histochemical stains and then digitized. Serial sections (up to ~70 per biopsy) were aligned, inspected and structures of interest segmented, extracted and volume rendered.

Results

The serial WSIs from renal biopsies can be aligned, and the algorithms can compensate for artifacts intrinsic to paraffin-processed tissue. Alignments can be made from a few or many serial WSIs, creating "short" and "long" image stacks. Large regular structures such as glomeruli as well as variably sized and irregular tubular crystalline deposits can be aligned, segmented, and rendered into three-dimensional structures.

Conclusion

Serial WSIs from routine specimens can be can be aligned and viewed, facilitating three-dimensional analyses.

Webpage Graphical Animations for Pathology Education

Edward C. Klatt 1

1 Department of Biomedical Sciences, Mercer University School of Medicine, Macon, GA, USA. E-mail: klatt_ec@mercer.edu

Content

Animations to enhance health science education were added to webpages in general and organ systems pathology (http://library.med.utah.edu/WebPath/webpath.html). Selected webpages emphasizing educational constructs that could be enhanced with a simple graphical animation were modified. Constructs included pathophysiologic mechanisms of disease, diagnostic characteristics, and complications of disease.

Technology

Graphical animations were developed with World Wide Web 3 (WWW3) guidelines for the HyperText markup Language version 5 (HTML5), Cascading Style Sheets version 3 (CSS3), and Scalable Vector Graphics (SVG) standards, and coded into.html files with TexEdit 4.10 (http://www.tex-edit.com). The svg-edit-2.7 program (http://code.google.com/p/svg-edit/) was used with the Safari 7.1.3 web browser to develop the coding for SVG animations.

Design

Standards to create the HTML5, CSS3, and SVG coding were taken from http://www.w3schools.com. A total of 150 animations were developed. Their display size ranged from 25 × 50 pixels to 200 × 250 pixels. Each of these animations was placed onto a single webpage adjacent to an existing gross or microscopic pathologic image of a disease condition. There were 50 animations utilizing either mouse-over or touch screen on a mobile device to initiate the animation. The animations were set to a play length between 1 s and 9 s, and they could repeat indefinitely.

Results

The animations required from 10 min to 1 h each to develop. Webpage.html markup with the animation and uploading to the website server took less than 5 min per webpage. Learners using these webpages stayed on a webpage with animation an average of 20 s longer than pages with equivalent size of text and images without animation.

Conclusion

CSS3 animations let a webpage element gradually change from one style to another. SVG can display complex, irregular shapes. These animations use coding created and edited with any text editor and presented via a common open source interface displayed via a web browser without the need for additional software or plugins. Animations can be added to webpages with minimal time and resources. Animations encourage active learning via interaction to lengthen attention span and hold learner interest on the page. Animations support conceptualization of visual models for learning.

Implementation and Optimization of Electronic Synoptic Reporting at a Large Academic Hospital

Veronica E. Klepeis 1 , Thomas M. Gudewicz 1 , John R. Gilbertson 1

1 Department of Pathology, Massachusetts General Hospital, Boston, MA, USA. E-mail: vklepeis@partners.org

Content

Synoptic reporting is considered superior to traditional narrative reporting, as it standardizes, simplifies, and prioritizes recording of information. Beyond these benefits, electronic synoptic reporting has the potential to collect discrete data elements, resulting in advanced data-querying capabilities, automated analysis, and decision support. This study presents key factors that influenced implementation of electronic synoptic reporting at a large academic hospital in three categories: workflow, synoptic design, and synoptic content.

Technology

Electronic synoptic reporting was implemented using mTuitive software (Centerville, MA, USA). Specifically, synoptic reports were designed using Agile Author (v2.0.1.6) and synoptic data were entered using xPert (v3.0.0.27).

Design

Our anatomic pathology service is completely subspecialized. Until this implementation, paper synoptics were used, with the vast majority of pathologists relying heavily on transcriptionists. For this project, two paper synoptics per subspecialty were translated into electronic format, including both cancer and noncancer synoptics. In addition, new paper versions of the electronic synoptics were also made available. Pathologists and transcriptionists were individually trained on the software.

Results

A major obstacle to incorporating electronic synoptic reporting into our workflow included inadequate and outdated sign-out rooms, making computers and monitors largely inaccessible to the pathologist. Factors having most significant impact on workflow included the resulting use of paper versions of electronic synoptics and, therefore, involvement of transcriptionists in data entry. Furthermore, to optimize synoptic design, we applied functionality that made data capture more efficient (decreasing number of clicks by defaulting answers, staging automatically, incorporating list-in-list features and branching logic), provided ample opportunity for free text entry to motivate compliance, and introduced elements that improved report readability (clearly defining parts of case included, presenting tumor stage summary up-front, removing extraneous information relevant only during data entry). Finally, it was necessary to address cancer synoptic content-related issues, which, while not specific to electronic reporting, were a deviation from our prior practice. Most important were incorporation of results from margins and lymph nodes separately submitted from the main specimen into cancer synoptics and integrating results from prior specimens.

Conclusion

By paying attention to workflow, synoptic design and content, we identified and managed factors critical to optimal implementation of electronic synoptic reporting.

An Open-source Database Application for Clinical Pathology Consults

Patrick C. Mathias 1 , Cigdem H. Ussakli 1 , Daniel S. Herman 1 , Sinan Ussakli 2 , Noah G. Hoffman 1

1 Department of Laboratory Medicine, University of Washington, Seattle, 2 Microsoft Corporation, Redmond, WA, USA. E-mail: pcm10@uw.edu

Content

Clinical pathology consults elicit valuable information about patients and laboratory tests that are not routinely captured in the electronic health record. At our institution, clinical pathology consults are a significant service component of residency, so in 2004, we developed an online call database to capture resident effort and knowledge and serve as a resource when residents are consulted. Because of technical problems, from its growth and outdated software components, we recently updated our database to achieve better performance and add new features.

Technology

The revised database was built using open-source components. The application was built with a Python web framework called Flask (http://flask.pocoo.org) and used an Apache web server (http://httpd.apache.org) and a department-hosted PostgreSQL database (http://www.postgresql.org). We used our institution's pubCookie web authentication services combined with account management within the application to ensure appropriate levels of access for users. The application is available to download at http://github.com/compumaster/oncalldb and a demo is available at http://oncalldb.azurewebsites.net.

Design

The primary functions of the database are to capture both structured and narrative on-call data, and to track progress of consultations. Based on resident input, multiple features were added: frequent auto-saving, robust full-text search, alerts indicating a patient have been previously consulted on, views of records without structured protected health information, commenting capabilities for faculty and other residents, tagging calls for classification purposes, and multiple user privilege levels.

Results

More than 6000 resident consults have been logged since the new database went live in July 2013. Throughout the more than 1.5 years of use, there have been no unscheduled downtimes. More than 30% of calls were for resident approval of laboratory tests, and another 23% of calls were for clinical consultations. Other frequent categories of calls included transfusion reactions (13%), requisition clarifications (8%), and critical values (8%).

Conclusions

We have built and deployed a freely available application to log resident consults that includes case tracking capabilities, auto-saving, full-text search, the ability to comment on cases, and multiple user privilege levels.

Use of No-SQL Data Model to Address Complexity of Tumor DNA Sequencing Analysis

Peter Gershkovich 1

1 Department of Pathology, Yale Medical School, New Haven, CT, USA. E-mail: peter.gershkovich@yale.edu

Context

Tumor DNA Sequencing generates complex datasets that require even more complex phenotype annotations. Adequate data models are critical for efficient storage, retrieval, exchange, and rapid understanding of interrelationships between various data elements that come from a range of sources used in bioinformatics pipelines and in pathological evaluation of variants. Relational models that have been dominating systems design for the last 30 years have been criticized for being too rigid, too fragmented, and incongruent to the needs of service oriented information exchange. No-SQL databases and document-based principles of data modeling may be more suitable to support the needs of molecular pathology.

Technology

MongoDB - a modern no-SQL database was used to store data for Downstream Reporting software - a Java-based reporting tool that integrates Anatomic Pathology LIS, bioinformatics pipelines, and external resources such as PubMed, Enembl, OMIM, KEGG, GeneCards, and Uniprot via RESTful API to facilitate creation of Pathology Reports for Tumor DNA Sequencing tests.

Design

MongoDB provides storage for "Downstream Reporting" - a web application deployed on Apache Tomcat servlet container. Users of the system upload results of two parallel bioinformatics pipelines to filter artifacts, select and annotate driver mutations, and generate molecular pathology reports. Four core tables and a handful of cached dictionaries were created to support all data needs of the application. Classic one-to-many and many-to-many relationships were replaced with document-based embedded data model.

Results

Most relationships between entities in DNA sequencing data that a relational model would consider as one-to-many appear to be rather one-to-few (e.g., a few genes in copy number variation region, or a few literature references for predicted effects of a variant). Keeping them together by means of embedded data model helped to reduce systems complexity and speed up development time. This data model worked well with JSON-based information exchange and required less code to create desired functionality.

Conclusions

The complexity and flexibility of data in cancer genomics require new approaches to simplify data models, facilitate understanding of data, and speed up the development process. No-SQL databases and MongoDB in particular appear to be well-suited to address these needs.

Segmenting Salient Objects in Images of Hematoxylin and Eosin Stained Tissues

S. Chakra Chennubhotla 1 , Luong Nguyen 1 , Jeffrey Fine 1 , Bruce Campbell 1 , Adrian Lee 1 , Lans Taylor 1

1 Department of Computational and Systems Biology, University of Pittsburgh, Pittsburgh, PA, USA. E-mail: chakracs@pitt.edu

Content

Pathologists examine Hematoxylin and Eosin (H&E) stained tumor slides to decide on the progression of disease and to inform oncologists on the appropriate course of cancer treatments. A key component of this analysis is to identify architectural elements of a tissue, such as glands/ducts, stroma, infiltrating lymphocytes, blood vessels, and associate diagnostic features (spatial and morphological) to these salient structures.

Technology

Despite the significance of segmenting a tissue into its architectural components, there are a limited number of available methods, most of which are supervised and/or utilize highly engineered features. Motivated by the recent work in computer vision on segmenting natural image boundaries, we propose to use internal image statistics in combination with a novel color presentation to segment H&E images of breast tissues into salient objects.

Design

First, we construct an opponent color representation, termed phi-psi space, to account for stain variability in H&E images. The signals in the opponent color space can be separated into pink, purple, and white (the three main colors of H&E images), by a mixture of univariate von Mises distributions. Second, to account for the spatial dependencies inside salient structures, we model the joint distributions between neighboring pixels with bivariate von Mises mixture models. Third, from the joint distribution, we calculate the point-wise mutual information (phi-psi) and further use it as an affinity measure between pixels in a graph-based spectral segmentation approach to finding the salient objects. Finally, salient object annotation data are being collected from highly experienced pathologists with a plan to validate our method against the ground truth dataset. Further, the annotation dataset will be released in public domain.

Results

[Figure 1] shows salient tissue object detection results with various strategies proposed in this work. Overall, integrating color opponent information with internal statistics (last column) gives the most accurate segmentation results.



Conclusion

Our method is the first to make the assumption that the statistical association between pixels within a tissue object is higher than the association between pixels on different objects in histopathology images. We test this assumption and showed that it can improve segmentation of salient tissue objects in H&E images.

A New Theory of Reference Intervals Based on Clinical Risks

Alan B. Solinger 1 , Steven I. Rothman 1 , G. Duncan Finlay 1

1 Department of Research, F.A.R. Institute, Sarasota, FL, USA, E-mail: alan.solinger@farinstitute.org

Content

Reference intervals are usually determined by statistical methodology unrelated to clinical outcomes: a "healthy" cohort's central 95% of test values are the "Reference Interval;" others are "Low" or "High." Methodological problems include defining "healthy" and illogic of flagging outer values when entire cohort is defined as healthy. These problems arise from methodology established before modern electronic medical records (EMR). We seek replacement methodology from perspectives of informatics.

Technology

Data extracted from AllScripts SCM version 5.0 (Sunrise Clinical Manager); statistical analysis performed with JMP version 11.2.1 (SAS Corporation, Cary, NC, USA).

Design

We extracted discharge dispositions; laboratory test results and demographics for 375,747 adult patient admissions at Sarasota Memorial Hospital (Florida). Similar smaller extracts were obtained from an academic medical center (Northeast USA) and a regional hospital (Southwest USA). For each analyte, we calculated an Outcome Risk function: OR(x) = (ONOwith in x)/(ONOwithout x)

where ONOwith in x odds of Negative Outcome for test results within x,

ONOwithout x = odds of Negative Outcome for results not with in x,

x = mean value of test results within an interval x,

Logistic regression, adjusted for confounding variables, determined final OR(x) for each analyte.

Results

Risks of both mortality and unfavorable discharge are below average within intervals reported in [Table 1]. Further, this approach provides clinical risks (e.g., mortality odds ratios) for values outside cut-points. Concurrence was found with other Negative Outcomes (e.g., 1-year postdischarge mortality), and among various medical centers.



Conclusions

We devised novel method to associate risks of negative patient outcomes, independent of diagnosis, with analyte test values. Clinical risk analyses determine reference interval cut-points and quantitative risks beyond them. In first approximation, outcome risk functions and cut-points can be calculated by spreadsheet in the laboratory from readily available EMR data.

High Normal Potassium is Dangerous for All Patients, Not Just Acute Myocardial Infarction

Alan B. Solinger 1 , Steven I. Rothman 1 , G. Duncan Finlay 1

1 Department of Research, F.A.R. Institute, Sarasota, FL, USA. E-mail: alan.solinger@farinstitute.org

Content

Recent studies show increased mortality in acute myocardial infarction (AMI) patients with serum potassium levels of 4.5-5.0 mEq/L, which is within the reference interval used by most laboratories. These findings have created an unresolved controversy challenging established potassium repletion therapeutic targets. We hypothesize this higher risk is applicable generally, not just to AMI patients .

Technology

Data extracted from AllScripts SCM version 5.0 (Sunrise Clinical Manager); statistical analysis performed with JMP version 11.2.1 (SAS Corporation, Cary, NC, USA).

Design

Retrospective study of 375,747 admissions at Sarasota Memorial Hospital (Florida); 54,916 at an academic medical center (Northeast); 37,419 a regional hospital (Southwest). Outcomes were in-hospital and 1-year mortality. Mortality models were fit by logistic regression.

Results

Utilizing logistic regression with adjustment for possible confounding factors, our analysis for all patients, independent of diagnosis, yields lowest odds of mortality at potassium values from 3.5 to 4.5 mEq/L, with significantly higher risks beyond 4.5 mEq/L. For both the AMI cohort and the non-AMI cohort, in-hospital all-cause mortality odds ratios were above 1.8 (P < 0.001) for potassium between 4.5 and 5.0 mEq/L (within usual reference interval); and were above 3 (P < 0.001) for potassium between 5.0 and 5.5 mEq/L. Adjusting for serum creatinine levels >2.0 mg/dL produced the same high "normal" mortality risks. Our findings hold for 1-year postdischarge, as well as in-hospital mortality odds ratios. While the risk functions differ in detail between AMI and other patients, we find that both show minimum risk within the same cut-points, with substantial increased risk above 4.5 mEq/L.

Conclusions

Our analysis extends the AMI finding: All patients have an increased mortality risk for serum potassium levels above 4.5 mEq/L. The etiology of death associated with mild hyperkalemia remains unclear. Presence of renal insufficiency appears not to account for this increased mortality. Without prospective studies, our findings cannot establish safety or danger of potassium repletion therapeutic targets. Finally, we point out that standard reference intervals are not based on patient risk, but are defined as the central 95% of test results for a "healthy" cohort. Reference interval cut-points would be more meaningful with a risk-based methodology.

Telepathology Network for the New York-New Jersey VA Integrated Health System

Charles Ladoulis 1,2 , Rosemary Wieczorek 1,2 , Nicholas Cassai 2 , Shahida Ahmed 2 , Rakhee Saxena 2 , Gary Clarke 2 , Paul Endres 2 , Matthew Pincus 1,2 , Cathy Cruise 2 , Telepathology Working Group of VA NY/NJ Integrated Health System

1 Department of Pathology and Laboratory Medicine, State University of New York at Brooklyn, NY, 2 Department of Veterans Affairs, Veterans Integrated System Network 3, VA, USA. E-mail: charles.ladoulis@gmail.com

Content

A telepathology network has been implemented in the VA New York/New Jersey (NY/NJ) Veterans Healthcare Network in 2014. The pathology departments of six NY/NJ VA Hospital units include Manhattan, Brooklyn, Bronx, NJ, Northport and Hudson Valley Veterans Administration hospitals.

Technology

All six facilities have been networked with identical instrumentation including robotic microscopes, whole slide imaging, and gross specimen photographic equipment. Robotic microscopes, local dedicated server, and remote regional server were all provided by ApolloPACS (ApolloPACS, Falls Church, VA, USA) for remote storage, access, and retrieval by all six departments. Whole slide imaging systems, MikroScan D2 (Mikron Instruments, Carlsbad, CA, USA) are installed for imaging, storage, and transmission to the ApolloPACS remote enterprise image management system. Gross specimen photographic imaging using the MacroPath (Milestone Medical Technologies, Kalamazoo, MI, USA) will provide for surgical and autopsy gross specimen imaging, annotation, and distribution.

Design

ApolloPACS enterprise image management system enables all six facilities to store, retrieve, and integrate images from a regional server at the VA facility in Richmond, Virginia. Teleconferences of pathologists, laboratory technologists and VA network and telemedicine technical staff meet weekly to install equipment, to create technical policies and procedures, to develop standards for telepathology consultations, and for compliance with HIPAA and federal regulations. Validation of whole slide imaging for diagnosis is in progress to comply with CAP recommendations developed by experts in pathology informatics and endorsed by the College of American Pathologists.

Results

Robotic microscopy consultation at all six VA hospitals using robotic microscopy is implemented for clinical review of consultation cases or integrated into conference case review and discussion, for internal departmental consultation or for interdepartmental specialty conferences. Whole slide imaging validation for all six facilities is in progress to meet CAP guidelines in 2015.

Conclusions

Weekly teleconferences for pathologists, laboratory technologists, and telehealth network coordinators ongoing for 18 months were critical for implementation and integration of hardware, software, and professional staff. VA Telehealth Support personnel is critical for successful integration, for compliance with federal regulatory requirements, and for consistent application of pathology practice guidelines.

One Year of Fried Brains: Informatics as a Driver of Cultural Change at an NIH Medical Scientists Training Program


Seung Park 1 , Timothy Kennell Jr. 2 , Robin Lorenz 2

1 Department of Pathology, NIH Medical Scientist Training Program, University of Alabama at Birmingham School of Medicine, 2 Department of Pathology, Division of Informatics, University of Alabama at Birmingham, Birmingham, AL, USA. E-mail: seungp@uab.edu

Content


One year ago, we instituted a unique 1- to 2-week long flipped-classroom, total-immersion data sciences experience that was proven to have a significant impact on participants' abilities in both informatics and Big Data analytics. This course has proven so popular that a majority of first- and second-year students at our institution's NIH Medical Scientists Training Program (MSTP) have taken it. This has provided us with the ability to measure the cultural effects of such a course on an entire training program.

Technology

E-mail correspondence.

Design

We maintained long-term contact with the students who had taken our course, allowing us to gauge impact of the course on the students. We provided both specific assistance on projects and general informatics mentorship. In addition, throughout this time, we assessed the impact of the course, projects, and mentorship on the attitude of the students toward the use of informatics and the decisions that they made in order to incorporate it into their careers.

Results

Projects chosen by the MSTP students at the end of the class are considerably larger in scope and are far more likely (82% vs. 64%) to be pursued to completion than the M.D.-only or Ph.D.-only student projects. Of the 17 MSTP projects, 14 have been completed and presented at national research conferences. After the completion of their projects, over 70% of the students have chosen to make informatics a large component of their research. Furthermore, four students thus far have made informatics their primary professional focus, two of whom have indicated that they would not have made this choice without the course.

Conclusion

MSTP students have a unique insight into the use of informatics due to their dual career focus. With Pathology Informatics leading the way in our institution's informatics education efforts, our MSTP has experienced an unprecedented cultural shift toward Big Data analytics and Clinical Informatics. These future M.D./Ph.D.'s are poised to become pillars of the critical bridge between clinical care, bench research, and informatics that will be all-important as we blaze a trail into the future of Personalized Medicine.

Three-dimensional Quantitative Morphological Characterization of Pancreatic Carcinoma Nuclei

Yi Gao 1 , Elizabeth Vanner 1 , Luisa Escobar Hoyos 2 , Christopher Metter 2 , Tahsin Kurc 1 , Kenneth Shroyer 2 , Joel Saltz 1

Departments of 1 Biomedical Informatics and 2 Pathology, Stony Brook University, NY, USA. E-mail: yi.Gao@Stonybrookmedicine.edu

Content

Pancreatic cancer is the fourth most common cause of deaths due to cancer in the USA. 1 The pathological evaluation of the fine-needle aspiration is currently adopted as the diagnosis references. However, up to 30% of the patients lack strong evidences in the traditional two-dimensional (2D) Papanicolaou staining slides. This study proposes a quantitative analysis on the three-dimensional (3D) nuclei morphological property extracted from 3D confocal/super resolution imaging of the immunofluorecence labeled tissue sample.

Technology

Pancreatic cancer tissue is stained with the 4',6-diamidino-2-phenylindole and the Zeiss LSM 510 META NLO two-photon laser scanning confocal microscope with Chameleon XR tunable 705-980 nm laser system. The 3D volume is acquired at 0.5 μm for each slice with a total 20 μm thickness.

Design

Institutional guidelines regarding animal experimentation were followed. The nuclei surfaces are extracted using an adaptive geodesic path approach. 2 The 3D surfaces are constructed and the closest distance to the enclosing convex hull is computed to measure the concavity of the nuclei surface.

Results

In [Figure 1], the panels a-c show the three orthogonal views of one nucleus from a healthy cell. The red contour depicts the automatically generated surface around the nucleus. d shows its 3D surface view and e overlay the concavity color-map over the surface. A region with more red-oriented color indicates more significant concaveness. Same for f-j for a cancer cell nucleus.



Conclusions

The volumetric analysis based on the 3D confocal imaging of the immunofluorescent tissue sample provides detailed and quantitative characterizations of the cell nuclei that are not available in the current 2D image based diagnosing approach. This would enable an objective and quantitative classification to facilitate pathologists' decision especially in the atypical cases

References

  1. American Cancer Society. Lifetime Risk of Developing or Dying from Cancer; 2014.
  2. Zhu L, Kolesov I, Gao Y, Kikinis R, Tannenbaum A. An effective interactive medical image segmentation method using fast growcut. In MICCAI Workshop on Interactive Medical Image Computing; 2014.


LOCHI: A Suite of R Programs to Automate Chimerism Analysis

Huazhang Guo 1 , Shiquan Jiang 1 , William Lam 1 , Junaid Ibrahim 1 , David Zhang 1 , Janina Longtine 1 , Fei Ye 1

1 Department of Pathology, Mount Sinai Hospital, New York, NY, USA. E-mail: huazhang.guo@gmail.com

Content

Quantitative chimerism is an indicator of the cellular dynamics involving engraftment of donor hematopoietic stem cells (HSCs) or relapse of recipient malignant cells in HSC transplant (HSCT) patients. Currently, the most common method for monitoring chimerism following HSCT is by polymerase chain reaction (PCR) amplification of short tandem repeat (STR) loci followed by capillary gel electrophoresis. However, this method can be challenging and tedious, including selection of informative loci and the repetition of quantifying chimerism for multiple loci from multiple cell types and multiple patients. Manual computation is not only time-consuming but also has the potential of human errors. Currently, there is no free software to fully automate the chimerism analysis.

Technology

PCR amplification of STR was done using Promega Power Plex 16 System. Capillary electrophoresis was done on ABI 3130x Genetic Analyzer. LOCHI, a suite of programs to automate chimerism analysis, was developed using statistical language R (version 3.1.1). R was chosen for programming because it is free and has strong matrix arithmetic capability.

Design

The algorithm for single donor (SD) chimerism calculation is well-established. LOCHI, a suite of R programs, was developed to automatically pick informative loci and calculate chimerism. In short, before transplantation, the donor and recipient alleles at each locus are compared to identify informative loci; after transplantation, chimerism is calculated from the peak areas of the informative loci. Alleles, not shared between donor and recipient, are used to calculate chimerism. In the case of informative locus with one shared and one unshared alleles, the value of unshared allele is used to estimate the shared allele assuming balanced amplification. The average chimerism value of all informative loci is reported. Similarly, we developed an algorithm to calculate chimerism in double cord blood transplantation and implemented it in LOCHI.

Results

LOCHI starts reading data from files generated by ABI GeneMapper. It filters the input data to get rid of stutter peaks and other noise. In addition to selecting informative loci and calculating chimerism, it also generates the checksum data to help verify the data integrity and pinpoint the outliers. The accuracy of the programs was compared with manual calculation on 20 patient samples. It is 90% concordant with manual calculation. As expected, the nonconcordance was due to human errors from the manual calculation.

Conclusion

By automating chimerism analysis, LOCHI significantly saves time and avoids human errors. Even though it was designed for our specific laboratory procedure, it can be easily adapted to other STR platforms with minor modification. The programs are freely available upon request.

Impact of Monitor Color Calibration on Digital Pathology Interpretation

Matthew G. Hanna 1 , Sara E. Monaco 1 , Ishtiaque Ahmed 1 , Anil V. Parwani 1 , Liron Pantanowitz 1

1 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: matthew.hanna@mountsinai.org

Content

Color is emerging as an important parameter that may impact digital imaging in pathology. Color in digital images is impacted by tissue stains, image acquisition devices, and computer displays used to view images. Color calibration of stained slides, whole slide scanners and/or monitors is alleged to be important in standardizing digital pathology. This study aimed to determine if color calibration of monitors produced a noticeable difference when interpreting digital slides.

Technology

Digital slides scanned at ×40 magnification using an Aperio ScanScope XT (Leica Biosystems, USA) were viewed on 24 inch widescreen Hewlett-Packard displays (ZR24w monitor, HP, USA) with 1920 × 1200 resolution. Display color calibration was performed using Spyder4 Pro (Datacolor, Lawrenceville, NJ, USA) software and a color sensor attached via USB to each monitor workstation.

Design

Two pathologists each interpreted 12 whole slide images using side-by-side HP monitors with equivalent display brightness, contrast ratio, pixel pitch, and color gamut. Digital slides of various stains were selected (Hematoxylin and Eosin, Diff-Quik, Papanicolaou, trichrome, Gram, acid fast, Grocott's methenamine silver, India ink, immunohistochemistry). One monitor had baseline factor settings, while the other was color calibrated using Spyder4 Pro adjusting for ambient light. Pathologists were blinded as to which monitor was calibrated. Their preference for monitor, color predilection, and diagnostic accuracy was recorded.

Results

Pathologists overall preferred (92% and 75% of cases) viewing digital slides on the uncalibrated monitor. They favored image color and brightness on the uncalibrated display. The calibrated monitor appeared to have a blue undertone compared to the affable yellow uncalibrated monitor [Figure 1]. Calibration or lack thereof did not affect diagnosis.



Conclusions

This study unexpectedly revealed that when viewing whole slide images from a desktop workstation, in the majority of cases, pathologists prefer an uncalibrated computer monitor to one that is color calibrated. Additional studies comparing different color calibration techniques and display settings are needed to better appreciate how adjustment of color affects digital pathology

Systematized Nomenclature of Medicine Encoding in Pathology Report

John Hon Man Mok 1 , Kent Tsui 1

1 Department of Health Information, The Hospital Authority, Hong Kong, China. E-mail: mokhm@ha.org.hk

Content

There is an auto-encoding function for Systematized Nomenclature of Medicine III (SNOMED III) in our eleven anatomical pathology laboratory information systems in the Hong Kong Hospital Authority. For reporting pathology result, apart from printing report, pathologists are using SNOMED III to encode the "Diagnosis result" in their local LISs. An operation list of SNOMED III codes including locally created terms and synonyms has been created in the Hong Kong Hospital Authority. The creation of the operation list serves two purposes; to enhance the SNOMED III encoding process in LIS, and to create a mapping to a standard list of SNOMED Clinical Terms (SNOMED CT). The data retrieval and analysis of pathology report can be made either by SNOMED III or SNOMED CT concepts.

Technology

In local LIS, the SNOMED III encoding process is made by a matching of typed keywords in the Diagnosis field to the SNOMED III operation list in a local LIS dictionary. The encoded SNOMED III codes are stored along with the each pathology report.

Design

A working group has been formed with the hospital pathologists, IT and health informatician. The objective of the working group is to create a corporate-wise operation list of SNOMED III codes and a standard list of SNOMED CT.

Steps to create the operation list and standard list:

  • Associate with the eleven local SNOMED III tables
  • Cleanse the locally created terms
  • Map the SNOMED III codes to SNOMED CT concepts
  • Submit new concepts to IHTSDO if needed
  • Deploy the corporate-wise SNOMED III operation list to local LIS
  • Create a mapping table of SNOMED III operational list to the SNOMED CT standard list.


Results

  • The corporate-wise SNOMED III operation list contains over 13,000 terms and their synonyms
  • The SNOMED CT standard list has 6000 distinct SNOMED CT concepts [Appendix 1].




Conclusions

The benefits of the creation of corporate-wise SNOMED III operation list and SNOMED CT standard list are:

  • To enhance the automatic encoding in LIS
  • To facilitate data retrieval and analysis of pathology data
  • To be a groundwork for clinical information exchange in ePR and secondary use of pathology data.


HemaVue: A Web-based Portal for Digital Peripheral Blood Smears

Liron Pantanowitz 1 , Thomas Harper 2 , Anil V. Parwani 1 , Lydia Contis 1

1 Department of Pathology, 2 Information Services Division, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: pantanowitzl@upmc.edu

Content

Morphological evaluation of a peripheral blood smear by clinicians is occasionally required for patient management. Traditionally, this involves a hematologist requesting a patient's blood smear from the hematology laboratory for microscopic review. This is time-consuming for clinicians, manually intensive for laboratory staff, and only available for a limited time period before the glass slide gets discarded. In several hematology labs at our institution, we use CellaVision instruments that employ automated digital techniques to quantify and classify cells on peripheral blood smears. Our aim was to make these "digital blood smears" remotely available to clinicians.

Technology

CellaVision DM96 (AB, Lund, Sweden) digital hematology analyzers. Hardware included a Microsoft Windows Server (2008) and Microsoft SQL Server (2012). Microsoft Internet Information Services 7 web server software and ColdFusion 9 middleware.

Design

We created a website called "HemaVue" that runs on a Microsoft Windows Server. The website provides a portal to a database of tables on a Microsoft SQL Server. The tables comprise data for CellaVision runs from three instruments located in separate labs. The patient databases include images (jpeg files of blood cells) and associated metadata (patient details and blood counts). Data are exported to the database daily by laboratory personnel from standalone workstations. An automated process checks for exports, archives a copy of the export file, and notifies the database administrator and network manager if the export was not run. Using this secure web portal, users can remotely look up a patient's digital blood smear.

Results

The database currently has over 3.5 million images for around 23,000 patients. Over 130 users have access to the HemaVue site, including hematology/oncology faculty and laboratory technicians. Maintaining this service is challenging since the CellaVision system does not typically support exporting its data. With three laboratories supplying digital data from separate systems, occasionally run identifiers may overlap. Adding a new database column rectified some incidents, and manual intervention is sometimes required.

Conclusions

The HemaVue web-based portal permits clinicians at our institution to remotely view their patient's digital peripheral blood smears anywhere and at any time. These digital blood smears are permanently retained. The next phase of our project is to incorporate patient digital blood smears into the electronic medical record.

Impact of Social Media on Scholarly Informatics Articles

Liron Pantanowitz 1 , Navid Farahani 2 , Anil V. Parwani 1

1 Department of Pathology, University of Pittsburgh School of Medicine, Pittsburgh, PA, 2 Department of Pathology and Laboratory Medicine, Cedars-Sinai Medical Center, Los Angeles, CA, USA. E-mail: pantanowitzl@upmc.com

Content

Citations in peer-reviewed journals and impact factors are currently considered the gold-standard of gauging the scientific influence of published articles. However, formal citations in the literature take time to accumulate. Traditional journal metrics like citation index have also been influenced by the launch of Open Access journals. The role of alternative metrics (aka altmetrics) such as mentions in blogs and social media platforms for measuring the impact of scholarly publications is uncertain. Our aim was to explore the impact of social media websites on articles published in the Journal of Pathology Informatics (JPI).

Technology

JPI (published by Medknow, part of Wolters Kluwer) is an Open Access peer-reviewed journal. Facebook and Twitter accounts were established to publicize JPI articles.

Design

Two articles both about digital imaging in pathology that were published close together in JPI were chosen for this study. Article #1 was posted as a hyperlink on Facebook and Twitter 2 days following publication. Article #2 was not mentioned on any of the JPI social media websites. We evaluated the statistics (number of views, prints, emails, and downloads) for these two articles from the JPI website at 3 and 6 months following publication.

Results

The JPI Facebook homepage has 5541 likes and Twitter account 496 followers. The table compares the statistics for these two articles.

Conclusion

These data do not support the notion that publicizing scholarly articles with social media greatly influences their popularity. Perhaps following more and different types articles over longer time frames and using other measures (e.g., rate of re-posting articles by followers) may be more informative. Additional investigation into cyberpsychology and the role of social media on scientific research and articles is warranted.

Retrospective Blinded Study of Thyroid Fine Needle Aspirates Using Whole Slide Imaging Compared to Glass Slides for Primary Diagnosis

Simpal K. Gill 1 , Vijayalakshmi Padmanabhan 1 , Jonathan Marotti 1 , Kristen Muller 1 , Gerald Jackman Jr. 1 , Laura Gordon 1

1 Dartmouth Hitchcock Medical Center, Lebanon, NH, USA. E-mail: simpal.k.gill@hitchcock.org

Content

The utility of whole-slide imaging (WSI) for primary diagnosis of thyroid fine needle aspirates (FNAs) remains unclear. The goal of this study was to investigate the diagnostic accuracy, interobserver agreement, and feasibility of WSI for a series of thyroid FNAs compared to traditional light microscopy.

Technology

Leica SCN 400 digital scanner.

Design

Twenty-five consecutive previously diagnosed thyroid FNAs were retrieved and two cytopathologists (#1 and #2) independently reviewed all glass slides (smears, LBP and cell block). These were then scanned using the Leica SCN 400 digital scanner and reviewed in the digital image hub (Leica Biosystems) after a washout period of at least 1 week. Diagnoses for glass slides and WSI were based on The Bethesda System. The original diagnosis was considered correct; concordance was based on degree of agreement with the original diagnosis and analyzed between cytopathologists.

Results

Case distribution and interobserver agreement for the two modalities are shown in the table below. The total percent of correct diagnoses was 74% with similar findings between glass slides and WSI (74% vs. 70%, P = 0.8). However, WSI had an overall lower interobserver agreement (κ = 0.52) compared to glass slides (κ = 0.74), with the lowest level in atypia of undetermined significance cases (0.52). Benign cases had the greatest level of WSI interobserver agreement (κ = 0.90). There was one major (two step) discrepancy for WSI (benign classified as suspicious for malignancy) and no major discrepancies for glass slides. Specific limitations of WSI reported by cytopathologists included increased effort required to screen/maneuver digital images, increased length of time required to review digital images (particularly LBPs), and perceived decrease in nuclear cytologic detail.



Conclusion

WSI of thyroid aspirates showed poorer interobserver agreement compared to review of glass slides. Further studies of WSI are warranted to minimize limitations and enhance utilization in thyroid cytology.

An Alternative to SOAP: Formatting Electronic Medical Records for Pathologists

Chancy Christenson 1 , John Scott 1

1 Department of Pathology and Laboratory Medicine, Tulane University Medical Center, New Orleans, LA, USA. E-mail: cchrist@tulane.edu

Content

Widespread adoption of electronic medical records (EMR) was designated a national goal in 2009 by the American Recovery and Reinvestment Act. Although many of the core objectives set by the act are outside the scope of a normal pathology practice, the College of American Pathologists is encouraging pathologists to coordinate with clinicians in adopting the new technology. EMRs are written in the format of Subjective, Objective, Assessment and Plan, which is not used in pathology. Pathology notes describe the final diagnosis, micrscopic and gross description. Although there are a few EMRs such as CoPath and Epic which accommodate the unique format of pathology, they are hospital-based systems and are not available for pathology departments with independent consultation services.

Technology

The Praxis IV EMR system from Infor-Med was installed on a PC with Windows XP operating system and a 1 terabyte hard drive with remote access capability through the Tulane Internal network.

Design

A basic template was written in the format of final diagnosis, microscopic description, and gross description under patient instructions; subsequent consults for that disease were created under that template. Cataloging the data this way allowed for the creation of a "tree of documents" where multiple presentations of a disease were stored under a single disease term. This led to the creation of a diagnostic library of templates for each unique disease entity which could be modified as necessary.

Results

Standard EMRs use templates written in a Subjective, Objective, Assessment and Plan format and are static with preprogrammed checklists. The Praxis EMR uses dynamic templates instead which can be developed and modified by the user. The authors repurposed the Assessments and the Instructions sections to develop a diagnostic library of searchable, customizable templates catalogued by each disease and referenced by frequency of use. The templates allowed users to rapidly generate and modify a consult note from a specific disease template and populate it with an individual patient's unique information.

Conclusion

Current EMRs use an unsuitable format for pathology. However, Praxis is flexible enough to give independent pathology departments a viable alternative to generate consult notes and interface with other EMRs.

Live Robotic Telepathology: A Comparative Evaluation of VisionTek ® M6 and VisionTek ® Digital Microscopes

Ehab A. ElGabry 1 , Liron Pantanowitz 1 , Ishtiaque Ahmed 1 , Tsetan Dolkar 2 , Anil V. Parwani 1

1 Departments of Pathology, UPMC Shadyside hospital, Pittsburgh, PA, 2 Danbury Hospital, Danbury, Connecticut, USA. E-mail: drgabry2013@gmail.com

Content

VisionTek digital microscopes offer hybrid live robotic microscopy and slide scanning capability (whole slide image, partial slide image). The new VisionTek M6 has faster scan speeds at 0.275 μm/pixels (1.5 min vs. 3 min for 15 mm × 15 mm), higher magnification objectives and digital magnifications, illumination control and gamma correction for enhanced subcellular detail examination, and multiple z-plane capture at single field of view. We compared the performance of the VisionTek M6 to the VisionTek for telepathology.

Technology

VisionTek M6 (Sakura Finetek, USA) with ×5, ×20 and ×40 objectives and digital magnifications of ×2.5, ×10 and ×63. VisionTek (Sakura Finetek, USA) with ×2.5, ×10, ×20 objectives and ×40 digital magnification. TeamViewer (version 8.0.18051) for digital microscope remote control and live image review.

Design

A total of 60 prostate biopsy and 20 fine-needle aspiration slides were examined via telepathology using VisionTek M6 and a conventional microscope with a 2 week washout period. Findings were compared to prior validation results for VisionTek.

Results

VisionTek M6 showed better concordance (97% vs. 84%) for digital versus glass diagnostic performance. VisionTek M6 showed slightly better concordance than VisionTek (98% vs. 95%) for digital versus glass satisfactory evaluations. [Figure 1] shows that VisionTek M6 resulted in fewer diagnostic discrepancies and less technical errors.



Conclusion

Live robotic telepathology of surgical pathology biopsies and cytopathology cases using the VisionTek M6 resulted in improved diagnostic performance and fewer technical errors. The higher magnifications together with increased precision when remotely changing objectives greatly improved remote live slide review. Incorporating technology into digital microscopes that enhances resolution for live review and z-scanning capabilities contributes to improved image quality and thereby clinical performance, especially for remote review of cytology slides.

A Clinical-Grade Variant Template Designed To Support Genomic Data Integration Into Clinical Applications

Edward R. Lockhart 1 , Ira M. Lubin 1 , FACMG on behalf of the Clinical-Grade Variant File Workgroup

1 Division of Laboratory Programs, Standards, and Services, Centers for Disease Control and Prevention, Atlanta, GA, USA. E-mail: elockhart@cdc.gov

Content

The absence of adopted standards for representing human genomic sequence variants detected by clinical next-generation sequencing is a challenge to developing interoperable systems for clinical and public health applications. We present a clinical-grade variant template that provides a model for genomic data representation that addresses these challenges.

Technology

Health information systems and messaging formats are evolving to manage genomic data. The Clinical-grade Variant Template was developed to support clinical genomic sequencing applications by providing a format and constrained data fields amenable to the messaging of a patient's genomic data generated from next-generation sequencing technologies.

Design

A national workgroup was convened and facilitated by the Centers for Disease Control and Prevention to develop a clinical-grade variant file template. To assure synergy with existing initiatives, the workgroup collaborated with a number of federal partners (US Food and Drug Administration, National Institute of Standards, National Center for Biotechnology Information), and others in public and private settings.

Results

The workgroup developed a clinical-grade variant template. The template contains three sections: (1) General data about the patient and test methods, (2) the clinically relevant findings, and (3) the sequence dataset generated by genomic sequencing prior to clinical assessment (The VCF specification is recommended for this latter element). Certain laboratory methods require standardization to permit the generation of constrained data. For example, the designation of consistent position assignments requires alignment against the human genome reference assembly. The workgroup also recommended that variant callers be set to output reference variant, no call, and local phasing data, at least in clinically important regions of the genome to reduce ambiguity in variant descriptions. Other constrained data were recommended as relevant to data exchange among clinical entities and consistent with other initiatives (e.g., the HL7 Clinical Genomics Workgroup). However, the description of quality metrics remains challenging because calibration protocols vary by method and laboratory setting.

Conclusion

The clinical-grade variant template is designed to inform processes for genomic data representation amenable to clinical applications and to support of systems interoperability. Currently, the workgroup is designing a use case to evaluate the utility of the template.

The Road to "Personalized Medicine": How Can Patient-specific Forecasts of Survival be Developed?

Navid Farahani 1 , Amanda Lo1, Alberto Marchevsky 1

1 Department of Pathology and Laboratory Medicine, Cedars-Sinai Medical Center, Los Angeles, CA, USA. E-mail: nfarahan@gmail.org

Content

Cancer survival is generally estimated with univariate Kaplan-Meier and multivariate Cox hazard models, which do not provide survival probabilities or odds for individual patients. There is a need to develop patient-specific predictive forecasting models based on clinical, pathologic, molecular, and other data. Bayesian statistics are increasingly been employed to develop forecasting models based on multivariate conditional probabilities, but they have been seldom applied for the estimation of survival for individual cancer patients.

Technology

A simple Bayesian model was created using Excel (Microsoft, Redmond, WA, USA) and the Bayes' theorem formula, to demonstrate the effect of prior probabilities of 5-year survival by age and gender on the conditional probabilities of survival estimated for patients with stage II nonsmall cell lung carcinoma (NSCLC), and/or chronic obstructive pulmonary disease (COPD).

Design

Data from age and gender specific life tables (Centers for Disease Control 2010), were used to obtain prior probabilities of 5-year survival. Conditional probabilities of survival for patients with severe COPD (forced expiratory volume in 1 s <37%) and/or Stage II NSCLC were used to explore the effect of a multivariate Bayesian model on individualized posterior probabilities of survival.

Results

Multiple models by age and gender were calculated. [Table 1] shows an example of 5-year posterior probabilities of survival for 60-year-old Caucasians.



Conclusion

The Bayesian models provide posterior probabilities of survival that are quite different from averages published in literature for severe COPD and Stage II NSCLC patients. Bayesian derived posterior probabilities are heavily influenced by prior probability of survival. Further studies based on specific cohorts are needed to evaluate the utility of this methodology for the development of patient-specific forecasting models of survival for cancer patients.

The Use of Apple's Siri Virtual Assistant and Google's Mobile Application for Rapid Literature Searches

Navid Farahani 1 , Alberto Marchevsky 1

1 Department of Pathology and Laboratory Medicine, Cedars-Sinai Medical Center, Los Angeles, CA, USA. E-mail: nfarahan@gmail.org

Content

Multiple mobile applications are available but pathologists are probably underutilizing them as there is little awareness about their potential to facilitate daily practice. We tested the applicability of a smart phone and two widely available mobile applications from leading information technology corporations to perform rapid literature searches.

Technology

An iPhone 6 (Apple, Cupertino, CA, USA) smart phone equipped with Siri (Apple iOS v. 8.1.3) and Google's (Google, Mountain View, CA, USA) mobile application (v. 5.2.0) was used to query for literature information. Both applications use voice activated natural language user interfaces. They function as virtual personal assistants that delegate user requests to various web services and easily enable quick access to large, networked information databases.

Design

Ten questions pertaining to lung cancer were formulated and dictated to each of the mobile apps. For each query, the total number of retrieved links and the clinical relevance of the first ten links retrieved by each application were recorded. The latter were classified as either relevant or nonrelevant, based on our clinical expertise. Results were compared with the F-test.

Results

Both applications retrieved results surprisingly rapidly, with as many as 24,700,000 links per question in <1 s. There was no significant difference in the number of total of links retrieved by either software. On average, 6.5 (±1.4) and 7.4 (±1.3) of retrievals by Siri and Google's mobile application, respectively, were relevant. Differences are not statistically significant (P = 0.72).

Conclusion

Smart phones and their intrinsic applications represent vast, untapped resources that can serve multiple purposes because of their portability, ubiquitous Internet access, and ease of use. Both of the tested mobile applications from Apple and Google were easily able to process medical queries using natural language input with reasonable accuracy. Optimization of these natural language-based applications may potentially reduce turn-around-time associated with surgical pathology sign-out by enabling rapid access to pertinent differential diagnostic information.

Whole Slide Images of Prostate Core Needle Biopsies: How Do Compression Levels Impact Pathologist's Interpretation?

Ehab A. ElGabry 1 , Gabriela Quiroga-Garza 1 , Dinesh Pradhan 1 , Jon Duboy 1 , Liron Pantanowitz 1 , Anil V. Parwani 1

1 Department of Pathology, UPMC Shadyside Hospital, Pittsburgh, PA, USA. E-mail: drgabry2013@gmail.com

Content

Compression of whole slide images may improve their transmission rates and need for storage. However, image compression may negatively impact diagnostic performance. The objective of this study was to determine the impact of different image compression levels when interpreting prostate biopsies.

Technology

Glass slides were scanned using an Aperio ScanScope XT (Leica Microsystems Inc., USA). Computer monitors (HP ZR24w 1920 × 1200) were used to view digital slides with ImageScope software (Aperio ePathology, Leica, USA).

Design

Fifteen randomly selected prostate core needle biopsies (97 slides) were diagnosed using conventional light microscopy and digital slides. Slides were scanned at 3 different compression settings: 70 (low quality default setting, low compression), 50 (mid-level compression), and 30 (high compression). Reviewers were blinded to these compression levels and asked to rate image quality and their diagnostic confidence.

Results

[Table 1] shows the differences when evaluating digital slides with different compressions.



Conclusion

These data show that compression of whole slide images do not impact the diagnostic confidence and perceived image quality by pathologists. Diagnostic discrepancy between glass and digital slides was unrelated to image compression levels.

A Quick Prototyping Methodology for Segmenting Histology Images

Nipun Patel 1 , Benjamin A. Lakin 2,3 , Joshua S. Shelofsky 2,3 , Brian D. Snyder 2,4 , Mark W. Grinstaff 3,5

1 Department of Imaging, Gnnovation Inc., Beth Israel Deaconess Medical Center, Boston, Massachusetts, 2 Center for Advanced Orthopaedic Studies, Beth Israel Deaconess Medical Center, Harvard Medical School, 4 Department of Orthopaedic Surgery, Children's Hospital, Departments of 3 Biomedical Engineering and 5 Chemistry, Boston University, Boston, MA, USA. E-mail: npatel@gnnovationinc.com

Content

There are tremendous inefficiencies in how histology/pathology slides are evaluated in the clinic. Use of computer aided diagnostics to evaluate these slides has the potential to improve the efficiencies in the pathology practice: By providing quantitative assessments, consistent analyses, fast throughput, and better usage of the time of trained pathologists. Our idea is to design and develop these systems for specific disease areas.

In addition, the images have a color component which gives even better option to extract features from images. However, variety of data requires a quick way to do prototype testing of segmentation method. Our expertise with MATLAB (The MathWorks Inc., Natick, MA, USA) and understanding of image processing helped develop the Graphic User Interface based prototype for testing out the results. This results in scalable model that eventually can be tested modified and finally plugged into the workflow in a relatively short amount of time.

Technology

High resolution microscopic digital images of Safranin-O stained histological sections were obtained and saved as TIFF images for further analysis. The Graphic User Interface for image analysis was developed using several toolboxes within MATLAB (The MathWorks Inc., Natick, MA, USA)

Design

The histology images (Saff-O stained sections of bovine plugs) were segmented into background and foreground 1,2,… (sections of same osteochondral plug). The following algorithm steps were involved:

  • Graphics User Interface (allow user to select relevant image(s) for processing)
  • Select background image to get background intensity values for normalization
  • Apply normalization to the foreground image
  • Extract the region of interest and quantify the results.


Results

The goal of this study was to develop a software model to compare the quantification of the stained slides to the GAG content of DMMB Assay and find a correlation between the two. The correlation for n = 42 was strong (R2 = 0.93) and statistically significant (P < 0.05), indicating that quantifying the red content from Safranin-O stained histology sections reflects the GAG content of the cartilage tissue.

Conclusion

Developing a scalable and flexible GUI based "testing platform" can allow easy modification of the parameters to achieve the desired performance to be close to human judgment ("gold standard") in efficient manner.

Implementation of Whole Slide Imaging in a Histology Laboratory for Image Archiving of High Volume Pathology Practice

Michael Suriawinata 1 , Laura Gordon 1 , Gerald Jackman 1 , Matthew Stone 2 , Tamara Fairbank 1

1 Department of Pathology, 2 Ancillary Information Systems, Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire, USA. E-mail: michaelsuriawinata@gmail.com

Content

Although whole slide imaging has been widely accepted in education and research settings, its implementation in a clinical pathology practice is not without challenges. Workflow, data storage and personnel support are common challenges encountered in clinical laboratories. The Department of Pathology at Dartmouth-Hitchcock Medical Center (DHMC) has successfully implemented routine whole slide scanning for archiving in the histology laboratory of a high volume pathology practice.

Technology

Whole slide imaging.

Design

The Department of Pathology at DHMC has been performing whole slide scanning for slide archiving for 2 years, using two Leica SCN400 scanners (4- and 384-slide) in histology laboratory. This service is supported by two histotechnologists and a high school student. One to four pertinent positive cytopathology, dermatopathology, surgical pathology, and consultation slides, are selected and documented in pathology report. The slides are then submitted for routine or priority scanning after the cases are signed out throughout the day, maintaining a continuous throughput. Priority slides are scanned in a 4-slide scanner, while routine slides are scanned in a 384-slide scanner. Images are reviewed for their quality, including sharpness, clarity, and areas of interest inclusion; then uploaded to a server at the hospital data center. Scanned slides are marked and filed in the slide storage room.

Results

In 2014, the total number of clinical slides scanned was 19,066 representing 19% of accessioned cases. The average turnaround time for routine and priority scanning were 24 h and 0.5 h, respectively. The scanning process requires 0.75 of a FTE of either a histotechnologist or support staff and adds a step between sign out and slide filing. Scanner operation and quality control of images are best-performed by histotechnologists. All pathologists now rely on whole slide images for image capture, conferences, tumor boards, and teaching; and have abandoned the use of digital microscope camera. Storage of image files at hospital data center enables image viewing throughout intranet and ensures data security and reliability.

Conclusion

We have shown that whole slide image scanning could be integrated to a busy pathology service, with a well-designed workflow, histotechnologist support, and a supportive staff to assist in system integration.

Utilization of Whole Slide Image Digital Pathology Communication System for Surgical Pathology Quality Assurance Program

Michael Suriawinata 1 , Laura Gordon 1 , Gerald Jackman 1 , Matthew Stone 2 , Robin Weisburger 3 , Tamara Fairbank 1

1 Department of Pathology, 2 Ancillary Information Systems, Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire, 3 Corista Digital Pathology, Concord, MA, USA. E-mail: michaelsuriawinata@gmail.com

Content

Whole slide imaging (WSI) has been around for more than a decade, but pathologists' comfort and familiarity with using such a system varies. Quality assurance program (QA) is an important component of a surgical pathology service and generally employs a multitude of case review methods. The Department of Pathology at Dartmouth-Hitchcock Medical Center (DHMC) validated and established a postsign out QA program using WSIs, supported by a WSI digital pathology communication system.

Technology

WSI.

Design

The Department of Pathology at DHMC has been performing routine WSI archiving of pertinent slides for the past 2 years. To establish a postsign out QA, positive or malignant biopsy case is accessioned to Corista DP3™ digital pathology communication system. Archived WSIs of the case are then imported from the departmental image database to and merged with patient demographics, clinical information, gross description and pathologic diagnosis from the laboratory information system in the Corista system through its interface. Histotechnologists and trained support staff provide quality control of the imported WSIs. The Corista system then routes the case to another pathologist who did not sign the case out for QA. The QA pathologist reviews the case on personal desktop computer and provides agreement or disagreement to the diagnosis of the original pathologist. If disagreement arises, the pathologist may communicate to other pathologists for resolution of the diagnosis and follow the departmental QA protocol. The system captures all steps and communications, and provides a timely QA report.

Results

The benefits of QA using WSI digital pathology communication system: (1) Improved QA workflow, (2) reduction in QA steps, (3) elimination of glass slide circulation between pathologists, (4) improvement of pathologists' familiarity and comfort to a diagnostic WSI system, (5) pathologists' preparation for WSI telepathology in the future.

Conclusion

We have established a case review QA program utilizing a WSI digital pathology communication system, which significantly improves our QA and its workflow, and indirectly prepares pathologists for diagnostic telepathology in the future.

Applied Informatics for Clinical Sequencing: A Tailored Data Sciences Experience for Molecular Diagnostics


David Figge 1 , David Standaert 1 , Shuko Harada 2 , Seung Park 3

1 Department of Neurology, University of Alabama at Birmingham, 2 Department of Pathology, Section of Molecular Diagnostics, Division of Anatomic Pathology, University of Alabama at Birmingham, 3 Department of Pathology, Division of Informatics, University of Alabama at Birmingham, Birmingham, AL, USA.

E-mail: figgeda@uab.edu

Content

In an age of decreasing nucleotide sequencing costs, clinical exome sequencing has become a major diagnostic modality in personalized medicine. Significant challenges include: (a) Issues arising from management of high-volume data, (b) clinician understanding of the uses and limitations of sequencing, and (c) lack of a standardized data-driven culture in personalized medicine. At our institution, we have chosen to tackle these problems by designing and implementing a novel training paradigm in molecular informatics.

Design

The experience begins with an axiomatic introduction to computer science, including systems architecture, procedural and object-oriented programming, and software engineering principles. Following this, participants are immediately tasked with creating an exome analysis pipeline utilizing the Broad Institute's Genome Analysis Toolkit (GATK) Best Practices. GATK has the advantage of being the most popular exome analytics toolkit for research and increasingly for clinical care; as such, it has extensive documentation on pipeline assembly and validation. Throughout the experience, participants are offered mentorship and guidance at all times, with the expectation of independence by the end. Once participants have successfully assembled and tested their pipelines, they are given a curated 50-genome set against which their pipelines can be formally validated.

Technology

Virtualization Software: Oracle VirtualBox 4.3; Guest Operating System: Ubuntu Linux Server 14.04 LTS 64-bit; Exome Sequencing Software: BWA 7.5, Picard 1.94, SamTools 1.19, GATK 3.3.

Results

Five participants thus far have taken this experience, ranging from a graduate student from our institution's Medical Scientist Training Program to an attending pathologist in molecular diagnostics. All 5 have successfully assembled and validated their pipelines. In doing so, participants have gained deep experience in genomic informatics - a subject that is often treated as a blackbox at many institutions.

Conclusions

This experience shows the utility of informatics training as directly tailored to a pathology subspecialty. By enhancing the participants' understanding of sequencing analysis, we have helped create a bridge between clinical care, research, and informatics. This bridge has already begun to form the foundations of a Big-Data and analytics-driven multidisciplinary culture at our institution, which we believe will lead both directly and indirectly to more effective diagnostics and improved patient outcomes.

BBQ'D: An Extensible, Open-standards, Web 2.0 Pathology Education Crowdsourcing and Analytics Platform


Mina S. Mousa 1 , Silvio Litovsky 2 , Seung Park 3

1 Departments of Pathology, University of Alabama at Birmingham School of Medicine, 2 Division of Anatomic Pathology, University of Alabama at Birmingham, 3 Department of Pathology, Division of Informatics, University of Alabama at Birmingham, Birmingham, AL, USA. E-mail: msm3w3@uab.edu

Content

Medical education has undergone a profound paradigmatic shift. Older lecture-based didactic methods have given way to systems-and problem-based small group learning. With the growing realization that Big Data analytics and informatics are core competencies of the future of medicine, we stand at the brink of another profound transition in medical education. At our institution, we have designed, implemented, tested, and deployed an open-standards, Web 2.0 pathology education crowdsourcing, testing, and analytics platform codenamed "BBQ'D" as an experiment in flipped-classroom methodologies.

Technology

Server Hardware: Dell Precision T3600; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 14.04 LTS 64-bit; Web Server: nginx 1.7; Database Management System: MariaDB 10.0; Programming Language: PHP-FPM 5.5; User Interface Framework: Twitter Bootstrap 3.3.

Design

BBQ'D consists of the following components: a question submission module, a peer review module, a quiz/test module, and an analytics platform. Students are actively encouraged to both review the study material and gain an understanding of proper test question writing by writing boards-type review questions in the question submission module. Instructors can edit and approve submitted questions in the peer review module. The quiz/test module can be used both for self-study and formal testing. Student performance metrics - from time spent on questions to a breakdown of student performance on subject areas - are made available by the analytics platform.

Results

BBQ'D has been designed, implemented, tested, and deployed for both undergraduate and graduate medical education. It is currently under formal consideration as a teaching modality by our institution's Undergraduate Medical Education committee, and will be made available for beta testing shortly. The analytics platform in particular is recognized to be of immense potential for future medical education.

Conclusions

Specially-tailored informatics systems have the potential to revolutionize the state of the art in education. Flipped-classroom and practicum-based methodologies especially stand to benefit from such systems. We will monitor BBQ'D use patterns and statistics going forward. When a critical mass of validated review questions has been established, we will also load BBQ'D onto single-board Linux computers to be sent to medical schools in Zambia and other resource-poor areas.

Using Adaptive eLearning in Cervical Cytopathology Resident Education

Laura Taylor 1 , Chelsea Mehr 1 , Roseann Wu 1 , Michael Feldman 1

1 Department of Pathology and Laboratory Medicine, Hospital of the University of Pennsylvania, PA, USA. E-mail: laura.taylor@uphs.upenn.edu

Content

The integration of technology into physician training has the potential to substantially improve the breadth and quality of medical education. This use of technology is particularly relevant in the field of pathology, where critical visual information can be presented in innovative ways. Smart Sparrow (Smart Sparrow Pty Ltd., Sydney, Australia) is an online system that is designed to facilitate the integration of technology and adaptive learning into medical education.

Technology

Smart Sparrow is an adaptive, web-based E-learning platform that can be used to present information and provide real-time feedback. The technology uses student responses to individualize the educational curriculum based on current research and learning theory. In contrast to systems that rely on expensive servers to house files and host programs that require firewall protection, Smart Sparrow is cloud-based and can be accessed outside of a protected system.

Design

Smart Sparrow technology was used to create an adaptive E-learning module covering basic cervical cytopathology principles and Bethesda System terminology, intended for pathology residents at various levels of training. Static images and whole scanned slides from our institution's cytopathology archives were integrated into the software. Accompanying explanatory text and relevant feedback were also provided. An identical assessment of knowledge was given to learners before and after engaging with the module. Basic demographics were also collected. Institutional guidelines regarding research on human subjects were followed.

Results

The cervical cytopathology E-learning module has been created and well received by cytopathology faculty and residents with varying cytopathology experience. A broader evaluation is pending to formally evaluate the utility of this technology. The assessment given before and after using the module will allow for a quantitative metric of learning.

Conclusion

In addition to the positive feedback from residents and faculty to the Smart Sparrow technology, we anticipate that learners will show improvement in their cervical cytopathology knowledge following completion of the module. These data will help establish the utility of Smart Sparrow and adaptive E-learning in the education of not only pathology residents but medical trainees at all levels and specialties. In the future, this tool could also be used by residency programs for competency-based assessment.

Web Application Design as an Adjunct to Resident Teaching in Pathology Informatics

Chris L. Williams 1 , Katherine VandenHeuvel 1

1 Department of Pathology, University of Oklahoma Health Sciences Center, OK, USA.

E-mail: christopher-williams@ouhsc.edu

Content

In order to meet the demands placed on today's pathologist, increasing emphasis is being placed on resident education in informatics. Giving residents a meaningful, standardized experience with a live informatics project, however, can be difficult for numerous reasons. We wanted to see if a small-scale, hands-on project could be used to supplement the didactics of an informatics rotation for pathology residents without prior programming experience.

Design

A web application was chosen for several reasons, including the abundance of development tools, cross-platform compatibility, and increasing popularity. The resident was encouraged to create a project meeting three criteria: immediate utility, personal interest, and feasibility of completion within 2 weeks. The project chosen was a template to generate placental pathology reports which could be easily transferred into LIS.

Technology

A web-based development environment, Cloud9 (www.c9.io ; San Francisco, CA, USA), was selected for ease of setup and configuration. The application uses the basic triad of languages common to most web applications: HTML5 (www.w3.org , W3C) to describe content, cascading style sheets (W3C) to style content for presentation, and JavaScript (Mozilla, Mountain View, CA, USA) to add functionality. Two additional open source frameworks were used to abstract low level mechanics in order to focus on broad design concepts: Bootstrap (www.getbootstrap.com, Twitter, San Francisco, CA, USA) for style and layout, and AngularJS (www.angularjs.org , Google, Mountain View, CA, USA), a JavaScript framework utilized for manipulation of HTML content.

Results

With guidance from a faculty member, the resident was able to learn basic syntax and use the previously mentioned tools to complete a fully functional prototype within 1 week. The project was successful in significantly reducing the time needed to transcribe a paper checklist into a formatted report. In addition, the resident was very engaged and had positive feedback about the experience.

Conclusion

The overall objective was achieved and the resident gained familiarity with some fundamental informatics concepts in a relatively short period of time. Small project-based experiences such as this may be useful in supplementing resident education, especially for those with an interest in informatics.

Pathology Informatics Essentials for Residents: A Novel Curriculum for Educating Pathology Residents


Walter Henricks 1 , Raymond D. Aller 2 , Philip J. Boyer 3 , Victor B. Brodsky 4 , Alexis B. Carter 5 , Rajesh C. Dash 6 , Michael D. Feldman 7 , Rebecca Fulcer 8 , John R. Gilbertson 9 , Trish Glover 8 , James H. Harrison Jr. 10 , Kristen A. Johnson 8 , Donald S. Karcher 11 , Priscilla Markwood 12 , Ann Neumann 8 , Anil V. Parwani 13 , Sue Plath 8 , Suzanne Z. Powell1 4 , Michael W. Riben 15 , Rodney A. Schmidt 16 , John H. Sinard 17 , Enrique Terrazas 18 , J. Allan Tucker 19 , J. Mark Tuthill 20 , Myra L. Wilkerson 21 , Liron Pantanowitz 13

Department of 1 Pathology, Cleveland Clinic, 2 Pathology, University of Southern California, 3 Pathology and Laboratory Medicine, East Carolina University, 4 Pathology and Laboratory Medicine, Weill Cornell Medical College, 5 Department of Pathology and Laboratory Medicine, Emory University, 6 Pathology, Duke University School of Medicine, 7 Pathology and Laboratory Medicine, University of Pennsylvania, 8 College of American Pathologists, 9 Pathology, Massachusetts General Hospital, 10 Pathology, University of Virginia School of Medicine, 11 Pathology, The George Washington University, 12 Association of Pathology Chairs, 13 Pathology, University of Pittsburgh, 14 Pathology, Houston Methodist, 15 Pathology and Laboratory Medicine, MD Anderson, 16 Pathology, University of Washington, 17 Pathology, Yale University, 18 Laboratory Medicine, University of California San Francisco, 19 Pathology, University of South Alabama, 20 Pathology, Henry Ford Health System, 21 Pathology, Geisinger Medical Laboratories, E-mail: henricw@ccf.org

Content

Multiple factors are driving an increased need for pathologists to be proficient in informatics. However, adequate informatics training is currently lacking in most pathology residency programs. We report a cross-organizational initiative to develop a pathology informatics curriculum for residents aimed at providing training in core informatics knowledge and skills that are important for current and future pathology practice.

Technology

Interactive Portable Document Files (Acrobat, Adobe, San Jose, CA, USA). Microsoft applications (Word and PowerPoint, Microsoft, Redmond, WA, USA).

Design

The College of American Pathologists, Association of Pathology Chairs, and Association for Pathology Informatics convened a joint workgroup consisting of about nineteen pathologists, expert in informatics. A core team of three pathologists and education specialists provided oversight and project management. The workgroup met by conference calls over a period of about 7 months. A subset of the workgroup met in person to finalize the first release. Pathology Informatics Essentials for Residents (PIER) is aimed at training all pathologists, not informatics specialists.

Results

The group developed PIER. Release 1 of PIER launched in September 2014. PIER contains 4 Essentials, each with a set of peer-reviewed informatics knowledge and skill statements, designed with increasing complexity. PIER content is mapped to recent American Council on Graduate Medical Education informatics milestones for residents. PIER materials consist of an instructional resource guide and interactive toolkit (http://www.apcprods.org/pier/). The toolkit enables residents and program directors to track and to document individuals' progress through PIER. Alpha testing has begun at 14 training programs of diverse size and settings.

Conclusions

PIER is a novel curriculum developed by experts that pathology residency program directors may use to address their informatics training needs. This flexible curriculum is designed to easily align with American Council on Graduate Medical Education milestones. We anticipate widespread adoption and implementation to ensure that all pathology residents acquire the informatics training needed for modern pathology practice. © 2015 APC/API/CAP. All rights reserved.

Applications of Clinical Looking Glass Software for Effectiveness and Clinical Significance of Endometrial Cells Presence in PAP Smears

Gloria Ramos-Rivera 1 , Antonio Cajigas 1 , Mark Suhrland 1 , Ljiljana Vasovic 2

1 Department of Pathology, Montefiore Medical Center, Albert Einstein College of Medicine, Bronx, 2 New York-Presbyterian Hospital, Weill Cornell Medical College, New York, NY, USA. E-mail: gramos@montefiore.org

Content

Endometrial cancer is the most common malignancy of female genital tract in the United States. Currently, there is no accepted screening test. Suspicion for endometrial cancer is based on symptoms and presence of endometrial cells (EMC) in pap smears. Bethesda 2001 recommends reporting normal EMC in women over 40 years. We intend to use the Clinical Looking Glass (CLG) software to extract this information and assess the effectiveness of the EMC diagnosis with respect to capturing significant pathology.

Technology

CLG version 4.3.1, a user-friendly interactive software application developed at Montefiore Medical Center, Bronx, NY, USA to evaluate health care quality, effectiveness, and efficiency.

Design

A cohort was created with the CLG tool to extract the cytology report of female patients over 40 years old diagnosed with EMC present in pap smears in 2010 along with subsequent surgical pathology reports. The data were exported in an excel file and analyzed.

Results

The cohort consisted of 117 female patients with a diagnosis of EMC in pap smears on 2010. The average age was 55.6 years old. After manual verification of the cytopathology reports, we classified these patients in two categories: 103 patients with EMC present and 14 patients with atypical EMC present. Then, we manually selected surgical pathology reports related to gynecological procedures as follow-up of PAP smear diagnosis. From patients with diagnosis of EMC, only 36 patients (35%) were worked-up with a benign outcome. From the patients diagnosed with atypical EMC, four were followed up and one had a positive diagnosis of endometrial adenocarcinoma.

Conclusions

The CLG software is very useful creating cohorts of specific demographics, time periods, and diagnosis search. Nevertheless, the searching criteria used were not enough to eliminate the manual work of going through every report to extract the information needed for the study. From the information obtained, we can be tempted to conclude that the EMC diagnosis and follow-up is not cost effective, but a bigger study is needed requiring more manual work or a better approach.

Feasibility of Using the Panoptiq Imaging System for Telemicrobiology

Daniel D. Rhoads 1 , Ishtiaque Ahmed 1 , Liron Pantanowitz 1

1 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. E-mail: danrhoads@gmail.com

Content

Telemicrobiology is challenging because of the high magnification and fine resolution required to identify microorganisms. The Panoptiq imaging system combines low power panoramic image acquisition with simultaneous high power recording for regions of interest. The aim of this study was to evaluate the feasibility of using Panoptiq for telemicrobiology.

Technology

Panoptiq v.3.1.2 (ViewsIQ, Richmond, British Columbia, Canada) software was used for image acquisition, and Panoptiq View 3.1.2 (ViewsIQ) software was used for image viewing. Hardware included a Dell Precision Tower 5810, Olympus BX45TF microscope, and Prosilica GT (Allied Vision Technology) digital camera.

Design

Twenty challenging microbiology cases were selected. Glass slides included blood smears, tissue biopsies, cytology samples, and Gram-stained specimens and positive blood cultures. The slides contained bacteria, mycobacteria, fungi, and parasites. Panoptiq was used to acquire multi-objective digital images in two stages: slides were first digitally mapped at 10x objective magnification and then select Z-stacks of areas of interest were recorded at ×100 (oil immersion).

Results

Image acquisition was easy and took approximately 3 min/slide. Registration maps were built by manually previewing the slide at low power (×10), and these maps (.svs format) ranged in size from 1.1 to 45.5 Mb. Annotated z-stack videos of regions of interest were captured with 100x oil immersion magnification, and these z-stack files ranged in size from 15.5 Mb to 175 Mb (approximately 1.0 Mb per z-frame). The software highlighted each location on the map that had an accompanying z-stack file, and the user could open each z-stack in a pop-up window for closer inspection (see Figure of synovial fluid containing Gram-positive cocci). The main challenge was correctly registering ×100 fields to the ×10 digital map.



Conclusion

Panoptiq is a promising digital imaging system with potential for telemicrobiology use. Benefits of this system include its small footprint, ease of use, speed of image acquisition, low and high magnification capabilities, and extensive z-stack capacity. Further software enhancement is needed to accurately register ×100 regions of interest to low magnification digital maps.

Standardization of Histopathological Diagnoses from the Dog and Cat Using Systematized Nomenclature of Medicine-based Information Model

Abdullah M. Awaysheh 1 , Jeffrey Wilcke 1 , François Elvinger 2 , Loren Rees 3 , Weiguo Fan 4 , Kurt Zimmerman 1

Departments of 1 Biomedical Sciences and Pathobiology, 2 Population Health Sciences, 3 Business Information Technology and 4 Accounting and Information Systems, Virginia Tech University, Blacksburg, VA, USA. E-mail: abod442@vt.edu

Content

The World Small Animal Veterinary Association Gastrointestinal Standardization Group proposed standards for reporting the microscopic findings of endoscopic biopsies for the dog and cat. However, these standardization efforts do not include recommendations on terminology and syntax for the morphologic diagnosis based upon the microscopic findings. In this work, an information model was created to provide a standard representation of histopathological diagnoses. These standardized diagnoses can serve as classification categories to facilitate the application of data mining and machine learning methods.

Technology

Retrospective gastrointestinal biopsy reports were examined from dogs (n = 320) and cats (n = 125) with clinical gastrointestinal disease. Animals were client owned and presented to the Veterinary Teaching Hospital, Virginia - Maryland Collage of Veterinary Medicine between November 1, 2006 and April 29, 2013. The Systematized Nomenclature of Medicine-clinical Terms (SNOMED CT) was used as a terminology for the semantic content in the information model for representing the histopathological diagnoses.

Design

Unstructured textual biopsy diagnoses were cataloged. Unique unstructured diagnoses were mapped to SNOMED. Unique unstructured diagnoses were then expressed as triplets. These triplets were comprised of SNOMED morphological abnormality concepts, SNOMED finding site relationship, and SNOMED body structure concepts.

Results

The biopsy reports contained 99 unique unstructured diagnoses. These diagnoses were associated with the following anatomic locations: n = 41 (27.3 %) stomach, n = 50 (33.3 %) small intestine and n = 59 (39.3 %) large intestine. SNOMED was only able to directly represent 59 (59.6%) of the unstructured original unique diagnoses. However, using the triplet information model, 100% of the original report diagnoses could be expressed using 33 SNOMED morphological abnormality concepts and 9 SNOMED body structure concepts.

Conclusions

In this study, the information model (morphological abnormality concept + finding site relationship + body structure concept) was able to represent the gastrointestinal histopathological diagnoses, thus, can be used to extend the World Small Animal Veterinary Association Gastrointestinal Standardization Group's efforts to standardize gastrointestinal biopsy reporting.

A Web 2.0, Extensible Clinical Decision Support System for Perioperative Management of Antiplatelet Therapy in Patients with Coronary Stents

Briana Gibson 1 , Marisa Marques 2 , Seung Park 3

1 Department of Pathology, University of Alabama at Birmingham, 2 Department of Pathology, Division of Blood Banking and Transfusion Medicine, University of Alabama at Birmingham, 3 Department of Pathology, Division of Informatics, University of Alabama at Birmingham, Birmingham, AL, USA. E-mail: brgibson@uabmc.edu

Content

Consensus-based standardized management protocols improve care and increase clinician compliance with recommended practice guidelines. Ensuring institution-wide accessibility and ease of use is critical for adoption in clinical practice. Institutions often develop flowcharts for standardization of clinical care, but these flowcharts are often so complex that they are not readily usable in the clinical milieu. In order to solve these problems, we designed, created, and deployed a database driven online decision tree for a protocol for preoperative antiplatelet therapy in patients with coronary stents developed by our institution's multidisciplinary Anticoagulation Task Force.

Technology

Server Hardware: Toshiba Portege M750 (Intel Core 2 Duo CPU P8600 @ 2.4 GHz x2, 2GB DDR2 RAM, 160GB HDD); Operating System: Ubuntu Linux Server 14.04 LTS 64-bit; Web Server: nginx 1.7.7; Database Management System: MariaDB 10.0.14; Programming Language: PHP-FPM 5.5; User Interface Framework: Twitter Bootstrap 3.2.

Design

The flowcharts developed by the Anticoagulation Task Force were translated into an application that separates each decision point and displays the next designated point based on the user's input. All possible paths and results are collected in tables in a central database. The user clicks through multiple decision points until arriving at a clinical recommendation. The design is responsive, providing an appropriate user experience for different form factors (e.g., desktop, tablet, smartphone).

Results

The Protocol for Preoperative Antiplatelet Therapy with Coronary Stents (PATC) is accessible on institutional computers, as well as personal tablets and mobile devices. In preliminary testing, it has greatly simplified and clarified perioperative management of antiplatelet therapy in patients with coronary stents.

Conclusions

We created an interactive, online decision tree that provides clinical recommendations in accordance with our institution's consensus guidelines. We are actively gathering statistics on workflow improvement and error reduction. Future upgrades to our clinical decision support system include data persistence for analytics. This project underscores the utility of clinical informatics in providing workflow-and culture-appropriate decision support for complex clinical situations.

A Comprehensive Web 2.0 Neuropathology Whole Slide Imaging Repository and Teaching Platform

J. Robinson Hackney 1 , Peter Anderson 2 , Alex Feldman 3 , Seung Park 1

1 Department of Pathology, Division of Informatics, University of Alabama at Birmingham, 2 Department of Pathology, Division of Molecular and Cellular Pathology, University of Alabama at Birmingham, 3 Department of Pathology, University of Alabama at Birmingham, Birmingham, AL, USA. E-mail: jhackney@uab.edu

Content

Last year, we created an instructional neuropathology wiki for the education of residents and fellows at our institution's Pathology residency and fellowship programs. This wiki is host to a curated collection of digital neuropathology images, presented within their clinical context. With the creation and deployment of a Web 2.0 universal whole slide imaging (WSI) system at our institution, we have chosen to create the largest publicly available neuropathology WSI repository and teaching platform on the web.

Technology

Server Hardware: Dell Precision T3600; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 14.04 LTS 64-bit; Web Server: nginx 1.7; Database Management System: MariaDB 10.0; Programming Language: PHP-FPM 5.5; User Interface Framework: Twitter Bootstrap 3.3; Rapid Publication Environment: MediaWiki 1.22; WSI Conversion Software: libVIPS 4.32 and OpenSlide 3.4; WSI Server System: PEIR-VM 0.2.

Design

We gathered all of our glass-slide neuropathology teaching sets, cataloged them, made fresh recuts where necessary, and most critically compiled detailed metadata on each slide into a CSV file. This CSV file was then fed into our existing PEIR-VM system, and all slides were scanned. The scanned slides were converted into DZI pyramids via libVIPS and OpenSlide; these DZI pyramids were then transferred to PEIR-VM for final display. The WSIs (complete with relevant metadata) can be directly viewed from PEIR-VM, or they can be embedded in their clinical context at our neuropathology teaching wiki.

Results

Over the course of several months, over 150 slides were vetted, scanned, compiled with metadata, and inserted into PEIR-VM. We now have a full library of neuropathology WSIs against which both teaching and image analytics research can be done. Our WSI library is fully integrated into our existing neuropathology teaching wiki, allowing residents and fellows to read full slides instead of having to be guided by static images alone.

Conclusions

The addition of our neuropathology WSI library, while time-consuming and system resource-intensive, has been worthwhile. We will continue to maintain this library as time goes on, adding new cases culled from our clinical service as appropriate. We are considering a parallel version of this system for clinical case archiving.

An Accurate Nuclei Detection and Segmentation Method based on Multi-scale Edge Selection in Polar Space


Chi Liu, Gustavo K. Rohde

Department of Biomedical Engineering, Center for Bioimage Informatics, Carnegie Mellon University, Pittsburgh, PA, USA. E-mail: chiliu@andrew.cmu.edu

Content

Cell nuclei in two-dimensional (2D) pathology images can yield quantitative information about the presence or the absence of disease processes. Segmenting nuclei correctly with minimum human effort is important for problems involving large numbers of cells, and patients. We describe an alternative nuclei segmentation method, which we call multi-scale edge selection in polar space. It takes as input 2D pathology images and outputs segmentation results with little effort for parameter tuning or human intervention.

Technology

  • Nuclei detection:
    • We construct a filter bank composed with rings of different sizes. Then the input image is convolved with the filters to generate the response map via normalized cross correlation. The peaks in the response map are thresholded and the seeds are finally obtained by calculating the mass center of each connected component.
  • Nuclei segmentation:
    • Edge maps are generated at different levels of image blurring followed by edge selection performed at each level in polar space. With an edge continuity constraint, the algorithm takes selected edges at coarser scales as guidance to select edges at finer scales. The algorithm outputs final contour with the highest likelihood of being correct, as measured through a well-defined energy term.


Design

We demonstrated the performance of our method on liver histopathology and mesothelioma cytopathology datasets. With hand segmented "ground truth data," the segmentation results are quantified using area error rate and normalized sum of distances. Moreover, we compared our method with several state of the art methods level set, template matching, the ovuscule.

Results

The quantified segmentation results are shown in [Table 1].



Conclusion

The proposed method showed satisfying performance by quantitative evaluation on real datasets. Most importantly, the algorithm is automatic and robust. Edge iteration in our method can make the contour tightly attached to the borders of nuclei, which can reduce the noisy information induced in segmentation step and thus potentially improve the classification accuracy.

Cost-effective Approach to Analyzing Operational Data in the Anatomic Pathology Laboratory

Caesar A. Llanos, Joseph Zeitouni

Department of Pathology, University of Miami, Miller School of Medicine, FL, USA. E-mail: cllanos2@med.miami.edu

Content

The modern anatomic pathology laboratory operating with an LIS is expected to provide a high level of data availability. However, out of the box reports included with most LIS at installation rarely suffice to satisfy the varied needs of the laboratory. Many laboratories, lack the expertise to write reports using sophisticated high-end reporting tools, leaving them to expensive consultant fees, or even compiling data manually. Here, we report our experience with custom report writing and data analysis that currently supports most divisions in our laboratory, using low-cost, readily available office software.

Technology

Queries are written using Microsoft Access (Redmond, WA, USA) with a read-only ODBC connection to the Sunquest Copath LIS database (Tucson, AZ, USA) and analyzed/displayed in Microsoft Excel.

Design

The LIS acquires multidimensional data with every step of processing clinical samples. With an intimate understanding of the processes in the laboratory and knowledge of the LIS database structure, we developed custom queries to obtain and analyze sets of data that are supporting the decision-making processes in the laboratory.

Results

Requests for multiple reports were reviewed from pathologists, staff from Quality Assurance, Billing and Laboratory Administration, as well as clinicians.

We summarized all the requests and developed a functional set of timestamps in the laboratory workflow that could be used to satisfy these requests as well as drive process improvements related to cases as well as to procedures/addenda. For billing and productivity, we created several queries that can be combined to provide a meaningful summary of each pathologist's workload.

Conclusions

The use of an ODBC driver combined with inexpensive desktop office software provided us with a cost effective way to quickly create and modify flexible reports for multiple and varied purposes in the clinical laboratory. As long as HIPAA compliant access restrictions are observed, we feel that pathologists should insist on the inclusion of an ODBC driver with every LIS installation.

Epidermal Growth Factor Receptor-sure Gold Nanorods Precisely Quantify Epidermal Growth Factor Receptor Expression

Ilker Ersoy 1,2 , Chuck Caldwell 4 , Chi-Ren Shyu 2,3,5 , Raghuraman Kannan 4,6 , Gerald Arthur 1,2 , Dmitriy Shin 1,2,3

Departments of 1 Pathology, 3 Computer Science, 4 Bioengineering, 5 Electrical and Computer Engineering and 6 Radiology, 2 Pathology, MU Informatics Institute, University of Missouri, Columbia, MO, USA. E-mail: shindm@health.missouri.edu

Content

Epidermal growth factor receptor (EGFR) is a cell surface receptor that is overexpressed in a variety of epithelial tumors such as colon carcinoma and nonsmall cell lung carcinoma. Precise determination of the EGFR expression levels is important in the selection of therapeutic approaches such as Cetuximab therapy. The only Food and Drug Administration approved test for EGFR is the immunohistochemical (IHC) method called EGFR PharmDx Kit developed by Dako. Several studies have confirmed that IHC methods are not sufficiently accurate in determining EGFR expression. Here, we report an automated image analysis method to show that our previously developed nanotechnology based EGFR detection kit (EGFR-Sure) is able to consistently quantify EGFR expression.

Technology

Gold nanorods (GNR) were synthesized at University of Missouri and linked with peptides engineered to bond with EGFR. Images of lung and colon carcinomas stained with an IHC method (EGFR PharmDx Kit) by Dako and EGFR-Sure nanorods were taken with a Leica DM5500 microscope. In-house developed image analysis software written in MATLAB by Mathworks was used to analyze images.

Design

Several fields of lung and colon adenocarcinomas stained with EGFR-Sure and PharmDx Kit were imaged to compare the robustness of quantification. Specimens were also stained with DAPI nuclear stain. In-house developed software first finds the nuclei in the images by running an active contour algorithm. Then, cytoplasm regions were computed by a watershed algorithm for each nucleus. The amount of stain was computed as a percentage per cell area by thresholding the stain channel and counting pixels within the cytoplasm. Several different thresholds were used to analyze robustness of the quantification.

Results

Quantification of the developed GNR is consistent among a variety of thresholds in comparison to conventional IHC stain as shown in the Figure. Variance of IHC stain quantity is about three fold greater than that of EGFR-Sure, and is dependent on the selection of the threshold.



Conclusions

Developed GNR is robust to the choice of parameters and produces consistent quantitative values. It provides precise quantification of EGFR expression and has also great potential to be suitable for precision medicine applications.

Laboratory Med Wiki: A Resource for Institution-Specific Knowledge Developed by and for Residents

Jeffrey Szymanski, Christopher Metts, Ronald Jackups

Department of Pathology and Immunology, Washington University, St. Louis, MO 63110, USA. E-mail: jszymanski@path.wustl.edu

Content

Well-established mechanisms to store, organize, and disseminate medical and scientific knowledge are readily available to residents training in laboratory medicine. However, residents also require a great deal of institution-specific knowledge to excel in their training programs. Often working in a new and unfamiliar health care system, residents begin each rotation unable to find contact information, schedules, patient data, instrument manuals, and other resources essential to their decision making process. Unfortunately, this institution-specific knowledge is often fragmented and difficult to access and can even go undocumented entirely. To address this problem, we created Laboratory Med Wiki, a wiki application to store, organize, and disseminate institution-specific knowledge for laboratory medicine residents.

Technology

Server Hardware: Dell PowerEdge R720, 32x2.3Ghz CPU cores, 192GB memory; Server Host Virtualization Hypervisor: VMWare ESXi 5.5; VM operating system: Ubuntu 14.04.2 LTS. Programming Language: PHP-FPM 5.3; Rapid Publication Environment: MediaWiki 1.24.1.

Design

A standard LAMP (Linux, Apache, MySQL, and PHP) stack was installed, and MediaWiki was installed and configured by residents with the assistance of the pathology department's information technology staff. Initial wiki pages were constructed based on extant department-reviewed service manuals ("survival guides") for each laboratory medicine service. Access to the wiki was provided to all laboratory medicine residents and fellows, and a brief introduction to the Wikimedia platform was provided at a department-wide meeting and through an instructional email.

Results

Laboratory Med Wiki was used to consolidate and organize institution-specific knowledge in the laboratory medicine environment. Several examples of residents updating and utilizing the wiki will be shown. Up to date usage statistics including unique users and total uploaded content will also be presented.

Conclusion

Laboratory Med Wiki has provided our residents with an organized platform to search and create institution-specific knowledge that has traditionally been difficult to access. By using MediaWiki, an open source platform, we will allow other departments and institutions to adopt the wiki format as a model to organize institution-specific knowledge.

Software Tools to Aid Blood Utilization Review

Brian Dangott, Philip Howard

Department of Pathology, East Carolina University, Greenville, NC, USA.

E-mail: dangottb@ecu.edu

Content

Blood management initiatives often involve retrospective analysis of product utilization. Current techniques usually include an automatically generated report which shows unit type, patient location, etc. However, relevant clinical data are often lacking. In this project, custom software was written to correlate issued blood bank products with clinically relevant data including the preceding hemoglobin and platelet levels for red cell units (red blood cell [RBC]) and platelet units respectively.

Technology

Custom programming using a mix of MUMPS, korn shell scripts, and perl was written for the Sunquest LIS (Tucson, AZ, USA) to parse the data. By default, the scripts run daily to identify transfusion time stamps from the previous calendar day. The time stamps are used to find the most recent hemoglobin and/or platelet values for a transfused patient. If no relevant values are found in a 5 day look back, the script will abort and move to the next transfusion. The default settings can be overridden by entering a date range to perform retrospective analysis. For this project, anonymized data from calendar year 2014 was exported and subsequently analyzed in Microsoft Excel.

Design

Extracted variables included: Age range, gender, pretransfusion hemoglobin level, pretransfusion platelet count and issued product.

Results

For all patients, the average pretransfusion hemoglobin level was 7.7 and the average pretransfusion platelet level was 61. A total of 20,393 red cell products and 2597 platelet products were issued during the calendar year. Approximately, 75% of both red cell and platelet transfusions took place after age 45. Comparing issued RBCs with age range shows a bimodal pattern with peaks at 15-25 years and 60-70 years. The platelet versus age comparison shows a similar bimodal pattern with peaks at 10-15 years and 55-60 years [Table 1].



Conclusions

By using software to incorporate relevant clinical values, blood utilization data can be stratified in ways that are very difficult using manual methods. Custom software allowed a retrospective review to illustrate how blood products are used in our facility. Going forward, the implementation of this software will help with blood utilization management.

CubiePEIR: A Next-generation Linux/armhf Single Board Computer with an Integrated Pathology Learning Environment and a Telepathology Platform

Elizabeth Staley 1 , Peter Anderson 2 , Jeremie Lever 3 , Matthew Anderson 2 , Rance C. Siniard 4 , Seung Park 4

1 Department of Pathology, 2 Division of Molecular and Cellular Pathology, 3 NIH Medical Scientists Training Program, University of Alabama at Birmingham, 4 Department of Pathology, Division of Informatics, University of Alabama at Birmingham, AL, USA. E-mail: estaley@uabmc.edu

Content

One year ago, we sent a Raspberry Pi Linux/armv6 single board computer loaded with educational content to medical schools in Zambia. This computer (codenamed "RaspberryPEIR"), while successful in its mission, had significant I/O and processing bottlenecks. As a result, as demand for RaspberryPEIR and its resources grew, perceived and actual systems performance was dramatically reduced. Furthermore, there arose a need for telepathology that RaspberryPEIR could not meet. We therefore designed and implemented a next-generation single board pathology informatics support environment codenamed "CubiePEIR."

Design

CubiePEIR utilizes a CubieTruck, which integrates an AllWinner A20 SoC. This SoC integrates two ARM Cortex-A7 CPU cores (instruction set architecture: ARMv7 with hardware floating-point unit), 2GB DDR3 SDRAM, an integrated 802.11n chip, and Serial ATA connectivity. The vastly increased performance of this unit allowed us to (a) Directly connect a Serial ATA hard drive as a boot and storage device, (b) compile software natively on the system to maximize performance, and (c) introduce a stripped-down, commodity telepathology system based on our previous work in commodity telepathology.

Technology

Hardware: CubieBoard CubieTruck (AllWinner A20 SoC, 2GB DDR3 SDRAM, Broadcom BCM4392 802.11n) and Western Digital 500GB SATA HDD; Operating System: Cubian Linux; Web Server: nginx 1.6; Database: MySQL 5.5; Programming Language: PHP-FPM 5.5; Web-Based WSI: OpenSlide 3.4.0, OpenSeadragon 1.0.0; Wireless Routing: ISC-DHCP-Server 4.1 and Bind9 9.8; Telepathology: OpenTelePath 0.2.

Results

CubiePEIR exhibits I/O performance that is an order of magnitude higher than that of RaspberryPEIR. It also integrates a much more advanced learning environment, including twice the number of whole slide images that were shipped with RaspberryPEIR, a full online course in Histology, and the ability to stream histology teaching sessions from a microscope via OpenTelePath.

Conclusions

The use of cheap commodity computing as a means to provide medical education in underserved countries is both cost-effective and sustainable. CubiePEIR will allow greater numbers of trainees to have access to PEIR content, furthering medical education in regions of the world where the limitation of resources serves as a formidable barrier to education.

Identifying Prognostic Immunophenotypic Profiles of Germinal Center Type Diffuse Large B-Cell Lymphoma through Hospital Information Systems

Evan Himchak, Etan Marks, Yanhua Wang

1 Department of Pathology, Montefiore Medical Center, Bronx, NY, USA. E-mail: ehimchak@montefiore.org

Content

Diffuse large B-cell lymphoma (DLBCL) is the most common type of aggressive non-Hodgkin's lymphoma. There are currently two broad categories of DLBCL based on the immunophenotypical profile, germinal center origin or nongerminal center origin. Within the germinal center DLBCL, it is unclear as to which immunophenotypical profile has a worse outcome. Using Clinical Looking Glass (CLG), a Montefiore Medical Center created program, we retrospectively looked at DLBCLs and their immunophenotype with regards to CD10, BCL-2, BCL-6, MUM1, and their Ki67 index.

Technology

A combination of CLG and Microsoft Excel was used to identify and sort patients with DLBCL and their immunohistochemical profile at diagnosis. CLG centralizes and anonymizes all of our institutional information systems into one searchable database. This provides current retrospective information on patient data. The output of CLG is an Excel table, which allows for further analysis.

Design

CLG was used to search for surgical pathology reports during a 10-year span of patients with a diagnosis of DLBCL. Reports containing the initial diagnosis of DLBCL and CD10, Ki67, Bcl2, Bcl6, or MUM1 were extracted to Excel with clinical information including age, gender, and survival data. Several Excel functions were used to categorize each patient's immunoprofile from their pathology report. This information, combined with survival data, was used to analyze prognostic value of the immunohistochemical markers.

Results

Totally, 166 patients were identified with germinal center DLBCL. Cox Proportional-Hazards Regression showed significantly better prognosis in patients with a Ki67 of 40% or less versus greater than 40% (P = 0.01). Kaplan-Meier curves suggest that Bcl6-may have a poorer prognosis than Bcl6+. It is not clear if CD10, Bcl2, or MUM1 status individually have prognostic value at this time.

Conclusion

This study demonstrated how powerful tools like CLG and Excel can analyze large amounts of retrospective data. By doing so, prognostic indicators can be identified and better indicate which patient populations require more aggressive treatments. Further study is necessary to identify which immunophenotypical profiles of DLBCL have predictive value.

Integration of Third Party Genetic Analysis Software into a Clinical Next Generation Sequencing Data Platform

Thomas J. S. Durant 1 , Wade L. Schulz 2

1 University of Connecticut, School of Medicine, Farmington, 2 Department of Clinical Pathology, Yale University, New Haven, CT, USA. E-mail: sakrisondurant@uchc.edu

Content

Advancements in next-generation sequencing (NGS) technology have allowed researchers and clinicians to generate genome-wide data sets. However, current information systems, which predate the arrival of NGS technology, lack adequate methods for NGS analysis and remain isolated from the systems which generate the data. In order to add meaningful annotations to clinical NGS data, we developed a system that allows us to request and dynamically format sequencing data for a variety of software applications.

Technology

The data management system is hosted on Windows Server 2012 (Microsoft, Redmond, WA, USA). Research software packages were run on Linux-based virtual machines running CentOS. Web service responses were consumed and parsed using Python 2.7 then visualized in R (version 3.1.2) using the SciClone R-package.

Design

In an effort to implement third-party data analysis software, we obtained variant call format files from technology agnostic web service endpoints and parsed the data with a python script. Data were formatted with the same Python script for analysis with a locally installed research application, called SciClone, which can be used to assess tumor heterogeneity from NGS results.

Results

Easily consumable web services support NGS utilization in clinical and research scenarios, without adversely impacting system performance. Results can be sent back to the data management system for use in ongoing clinical trials or for clinical interpretation as appropriate. Researchers or clinicians accessing the results need no additional training, and data can be provided based on available user permissions. Connecting clinicians and researchers to NGS results with a user-friendly environment, provided with visually interpretable graphs, is currently serving to improve turnaround time and closely integrate ongoing research studies with clinical sign-out in our department.

Conclusion

The ability to generate genome-wide sequences and analyze the resulting data in a clinically meaningful way are two distinct entities, which have been discordant in their points of evolution. With NGS technology evolving at a rapid pace, the technology used to interpret it struggles to keep up. To increase efficiency and provide increased data access, applications that easily and securely allow access to third-party research applications, such as web services, are important for furthering our understanding of sequencing data.

Are Coverslips Required Before Glass Slides Can Be Digitized?

Ehab A. ElGabry, Jon Duboy, Andrew Lesniak, Anil V. Parwani, Liron Pantanowitz

Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA.

E-mail: pantanowitzl@upmc.edu

Content

Whole slide imaging devices are typically calibrated to scan specific glass slides. A glass "slide" in pathology used for routine work (e.g., surgical pathology, cytopathology) includes a microscope slide (1.0 mm average thickness) and a mounted coverslip (0.15 mm average thickness). Some manufacturers recommend scanning with a glass coverslip. However, for some practices (e.g., rapid on-site evaluation of cytology specimens), glass slides may be prepared without coverslips. The effect of scanning glass slides without coverslips has not been well studied. The aim of this study was to determine the impact of uncovered glass slides on the digitization process of various commercially available scanners.

Technology

Whole slide scanners used: ScanScope XT (Aperio, Leica Microsystems, USA), NanoZoomer HT 2.0 (Hamamatsu, Japan), Mirax midi (3DHISTECH, Hungary), and VL4 (Omnyx, USA).

Design

Air-dried cytopathology glass slide smears (Diff-Quik stained) were prepared without coverslips. Attempts were made to digitize these uncovered slides using different scanners. The slides were subsequently mounted with glass coverslips and rescanned. The ability to scan slides and the image quality of these digital slides were evaluated.

Results

The table shows that all tested devices were able to successfully scan uncoverslipped glass slides. While image quality of both uncovered and covered digitized slides were satisfactory for diagnosis, for most scanners the covered digital slides were subjectively sharper in detail.



Conclusion

While some manufacturers claim that uncoverslipped glass slides may fail to scan with their whole slide imaging devices because they require a glass slide-coverslip interface for image capture, this was not observed in the scanners tested. If the presence of a coverslip is required to digitize a glass slide, this may warrant modification of workflow, especially during cytology on-site evaluation.

Creating a Wiki for the Adaptation, Management, and Execution of Pathology Informatics Essentials for Residents

Emilio Madrigal 1 , Matthew Hanna 2 , Shyam Prajapati 1 , Mark T. Friedman 1

1 Departments of Pathology, Mount Sinai Health System, St. Luke's, Roosevelt, and Beth Israel Hospitals, 2 Icahn School of Medicine at Mount Sinai, The Mount Sinai Hospital, New York, NY, USA. E-mail: emadrigal@chpnet.org

Content

A lack of structured educational resources regarding informatics is a common occurrence in pathology residency programs. This often leads to deficiencies in the core knowledge and skill sets required by graduating pathologists. At our institution, laboratory management and informatics is a required 1-month rotation, and a review of its syllabus revealed unstructured objectives without clear expectations, necessitating improvements based on a newly released research-based instructional resource guide: Pathology Informatics Essentials for Residents (PIER).

Technology

In order to actively implement PIER, the need for a system that would allow for the creation and modification of content was identified, and the dynamic framework of a wiki (content management system) was most suitable for this function. MediaWiki, a free, server-based, scalable, and feature-rich wiki application was installed and configured in an Apache Linux web server, which uses PHP to process and present information stored in a MySQL database.

Design

A wiki (http://slrbimcpathology.com/wiki) with a minimalistic user interface was created as a platform to manage the educational objectives formulated by PIER. In order to create and edit content, residents acquire a user account, whose access and ability permissions need to be approved by the wiki administrator, restricting content modification to authorized users.

Results

By utilizing MediaWiki's revision history tracker and a mobile timekeeping application, it was determined that the manual installation of the wiki, database configuration, and user interface design, comprised a total of 6.5 h. As an initial assessment of the project, three residents with an interest in pathology informatics were assigned a PIER Essentials 1 topic and asked to populate a wiki page based on the "Rationale," "PIER Outcomes," and "Content" described in the toolkit. Following a short instructional tutorial on a MediaWiki's wikitext format, the average time to completion of a topic page by a contributing resident was 67 min.

Conclusion

By using a wiki as an educational tool for pathology informatics training, we have been able to restructure our laboratory management and informatics rotation. Through the creation and editing of content, it is our impression that residents will be actively engaged in the learning process allowing them to better retain the knowledge and skills obtained. Future aims include the continuous maintenance of up-to-date wiki topics, evaluation of residents' pre and post pathology informatics competency, as well as an assessment of PIER outcome achievements.

Evaluation of Inflammatory Dermatoses using Whole Slide Images: Digital Images versus Glass Slides in the Analysis and Diagnosis of Neutrophilic Dermatoses

Kristin Burns 1 , Toni Peters 1 , Charles Law 2 , Beverly E. Faulkner-Jones 1

1 Department of Pathology, Beth Israel Deaconess Medical Center, Boston, MA, 2 Kitware Inc., Clifton Park, New York, USA. E-mail: kcburns@bidmc.harvard.edu

Content

The neutrophilic dermatoses are a clinically diverse group of diseases defined by a dense dermal infiltrate of mature neutrophils with leukocytoclasis. However, histologic variants exist and may pose diagnostic difficulty. Whole slide images (WSIs) have been used to assess of pathologic processes in several organ systems but there are few data on their use in dermatopathology. As part of a study to explore how neutrophil count impacts the development of histologic features in neutrophilic dermatoses, particularly Sweet's syndrome, we assessed the utility of high resolution WSIs for assessment and diagnosis. Sweet's syndrome can occur in the setting of leukemia and lymphoma, and timely diagnosis is desirable in this patient population. The use of WSIs would facilitate on-line collaborative subspecialty dermatopathology review and aid in the diagnosis of histologic variants of Sweet's Syndrome.

Technology

Glass slides were digitized using a Philips UFS slide scanner and the WSIs were viewed on http://slide-atlas.org , our high performance web-based viewing platform. Where indicated, local adaptive alignments of sequential WSIs were performed in Slide Atlas, and the aligned WSIs displayed side by side.

Design

Totally, 48 cases were selected by searching the laboratory information system for "neutrophilic dermatoses/infiltrates." These were digitized and uploaded to Slide Atlas. The glass slides and then the corresponding WSIs were evaluated with a 2-week "wash out" period. Features evaluated included the density and distribution of the neutrophilic infiltrate, the presence of papillary dermal edema, vasculopathy, leukocytoclasis, or "variant" features (immature neutrophils), and the presence of an associated nonneutrophilic inflammatory infiltrate.

Results

Neutrophilic infiltrates could be identified on WSIs in all cases. All 16 classic neutrophilic dermatoses were identified on the WSIs. Appropriate differential diagnoses were made for the 3 variant cases (superficial dermal infiltrate) on the WSIs, and equivalent to glass slide exam. WSIs had the advantage over glass slides of improved low zoom architectural assessment, the ability to computer-align sequential sections for limited three-dimensional assessment and collaborative review.

Conclusion

High resolution WSIs displayed using a high performance viewer can be used in the assessment and diagnosis of neutrophilic dermatoses and inflammatory dermatoses with neutrophilic infiltrates.




 

 
Top
  

    

 
  Search
 
   Browse articles
  
    Similar in PUBMED
    Access Statistics
    Email Alert *
    Add to My List *
* Registration required (free)  

 
  In this article

 Article Access Statistics
    Viewed3800    
    Printed30    
    Emailed0    
    PDF Downloaded327    
    Comments [Add]    

Recommend this journal