Journal of Pathology Informatics Journal of Pathology Informatics
Contact us | Home | Login   |  Users Online: 657  Print this pageEmail this pageSmall font sizeDefault font sizeIncrease font size 




 
Table of Contents    
ABSTRACT
J Pathol Inform 2014,  5:20

Abstracts: Pathology Informatics 2014


Date of Web Publication25-Jul-2014

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 2014. J Pathol Inform 2014;5:20

How to cite this URL:
. Abstracts: Pathology Informatics 2014. J Pathol Inform [serial online] 2014 [cited 2017 Sep 22];5:20. Available from: http://www.jpathinformatics.org/text.asp?2014/5/1/20/137080


   Short Abstract Presentations Top


Wednesday, May 14, 2014

Locations: Wyndham Grand Pittsburgh Hotel, Grand Ballroom 1, Kings Garden South/LeBateau, Kings Garden North

Advanced >Pathology Informatics


Grand Ballroom 1

Mongo on FHIR - A Governable Linked-Data Approach to HL7-Based Interoperability for Web Computing

Jonas S. Almeida, Alexander Grüneberg

Department of Pathology, Division of Informatics University of Alabama at Birmingham, . E-mail: jalmeida@uab.edu

Content

HL7 has recently developed the Fast Health Interoperable Resources specification (FHIR www.hl7.org/fhir) of the application programming interface (API) for interoperability with modern web computing (HTTP REST interface for CRUD operations). This development creates a framework to approach HL7 data streams as open linked data, using the JSON-LD formalism recently established by W3C (www.w3.org/TR/json-ld). This new level of syntactic interoperability proposed by HL7 is unprecedented for patient data. However, by itself, this would lead to an absence of governance that would also be unprecedented for this type of data. We have approached this implicit requirement for a portable (data-embedded) governance by extending the JSON-LD formalism to include user operators.

Technology

JSON-LD stands for Javascript Object Notation (JSON) for the Semantic Web's Linked Data (LD) framework. The critical advantage is that it allows the use of the rich toolbox of Web 3.0 technologies developed around W3C's Resource Description Framework (RDF), which underlies the Web's evolution towards effectively being a global Data Space.

Design

Web technologies have now reached the point where the regular web browser has become a full-stack software development platform. The development of FHIR by HL7 was the missing component to enable web computing approaches to patient derived data streams in clinical environments.

Results

The goal of embedding the governance of CRUD operations in JSON-LD formatted HL7 data streams was pursued, and achieved, without any redesign of the data itself, or the FIHR itself. We have validated this web computing for Health Informatics proposition with the development of a web service using only open source libraries. Specifically, the REST middle layer was developed in NodeJS, articulating a Big Data backend (NoSQL + MapReduce distribution) using MongoDB. A client-side JavaScript library to facilitate the development of web application was also developed. The open source library and a demonstration deployment with data about the 7,000+ patients described by The Cancer Genome Atlas (TCGA) are maintained at http://github.com/ibl/fhir.

Conclusion

HL7 Fast Health Interoperable Resources specification (FHIR) was found to deliver exactly what it was set out to do: a Web friendly programmatic interface for CRUD operations. This narrow configuration of FHIR API caused governance of those data streams to be defined as an entirely orthogonal pursuit. This created the opportunity, explored in the data modeling work described here, to embed governance within the data in a manner that does not intrude into the representation itself. The tools to achieve this goal have only emerged recently, and are found within the third generation of Web Technologies (Web 3.0 - Semantic Web). Specifically, the Linked Data approach to W3C's Resource Description Framework (RDF) now has a native representation as JSON-LD. The novel, non-intrusive, governance modeling described in this report was validated with an open source prototype and a public web service accompanying this report (see Results for address).

Automating Anatomic Pathology Stain Orders in Histology

Robert Stapp, Michael Czechowski, Kathy Roszka, J. Mark Tuthill

Department of Pathology, Henry Ford Hospital, Detroit, MI.

E-mail: rstapp1@hfhs.org

Content

Immunoperoxidase and special tissue stains are important routine aspects of pathology practice, with automated instrument platforms (AIPs) available to perform these assays. Orders in the anatomic pathology laboratory information system (AP-LIS), should be able to interface to these AIPs. Our goal was to eliminate dual order entry and reduce the tissue misidentification due to relabeling slides. We partnered with our vendors to implement bi-directional Health Level 7 (HL7) interfaces between our AP-LIS and AIP's for immunohistochemistry (IHC) and special staining platforms.

Technology

CoPath Plus v6.0 was used as the AP-LIS (Sunquest Information Systems, Tuscon, AZ). Three AutoLink 48 automatic IHC stainers, two Artisan Link Pro automated special stains platforms, and one DakoLink interface server, were utilized (Dako, Denmark).

Design

Initially, a unidirectional HL7 interface was implemented between CoPath and DakoLink. This allowed IHC orders placed in the AP-LIS to be received into the DakoLink instrument control software which sends electronic orders the IHC platforms.

Secondly, Copath was upgraded to a new version which provided the capability to assign pathology assets a unique identifier. This allowed our AP-LIS to send unique slide identifiers to the IHC platforms. Once implemented, the IHC AIPs could utilize native CoPath labels, eliminating slide relabeling.

Finally, a bi-directional HL7 interface was implemented which allowed status messages to be communicated from the IHC AIPs back to our AP-LIS. We then implemented the same interface between our AP-LIS the automated special stains platforms.

Results

The first phase simplified the staining run's setup process, while the second phase eliminated the slide relabeling. Currently, the Dako AIPs operate by scanning CoPath labels directly. By automating these processes, time savings of approximately 56 minutes (IHC) and 7.75 minutes (special stains) per run have been achieved. Total annual labor savings are estimated to be 570 hours for IHC and 100.75 hours for special stains workflows.

Conclusion

Elimination of dual order entry and relabeling has markedly decreased our assay run times. The potential for patient misidentification has been significantly reduced. In addition, automating these processes has increased stain order accuracy, improved laboratory throughput, and improved turnaround times.

KRAS Mutations in Pancreatic Adenocarcinoma: Caveats for Data Mining in Cancer Genomics

Edward Stites, David H. Spencer, Eric J. Duncavage, Ian S. Hagemann

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

Content

An effort to determine the observed frequency of KRAS mutations within our clinical next-generation sequencing experience identified several potential issues in mining clinical cancer genomic data.

Technology

Targeted sequencing of all exons of KRAS was performed as part of a multi-gene assay using DNA extracted from formalin-fixed, paraffin-embedded tumor tissue. Agilent SureSelect hybrid capture was followed by Illumina HiSeq or MiSeq sequencing. Single-nucleotide variants and insertion/deletion events were called by a custom bioinformatics pipeline based in the Genome Analysis Toolkit, with a variant allele frequency greater than 10% required for automated mutation reporting.

Design

A database of clinical cancer genome sequencing results was queried to retrieve cases of pancreatic cancer and their bioinformatically called KRAS genotypes. Chart reviews were then performed on all patients that satisfied the initial query to confirm the diagnoses and reported KRAS mutation status.

Results

Among 857 total patients, a naïve search identified 65 patients with pancreatic cancer, 39 of whom (60%) had a KRAS mutation. Examination of the electronic medical record revealed 11 of these patients had a diagnosis other than pancreatic adenocarcinoma. Review of clinical genomics reports identified 6 cases where the pathologist identified and reported a KRAS mutation that did not meet thresholds for automated mutation calling. This revised search therefore identified 54 patients with pancreatic cancer in the database, 42 of whom (78%) had KRAS mutations.

Conclusions

Our analysis highlights that caution must be used when querying clinical cancer genomic data. False positives and false negatives complicate the ability for clinical summary statistics to contribute to quality control and quality assurance. The "big data" movement claims that statisticians and informaticians can make important insights without understanding the underlying biology or medicine. Our work suggests that content-area knowledge is needed to execute maximally informative queries and recognize the limitations of data mining.

Adaptive Learning Based Data Extraction for Patient Search from Pathology Reports

Shuai Zheng', James J. Lu, 1 Daniel J. Brat, Fusheng Wang

1
Department of Mathematics and Computer Science, 2 Department of Pathology and Laboratory Medicine, 3 Department of Biomedical Informatics, Emory University, Atlanta, GA. E-mail: szheng5@emory.edu

Content

Extracting information from pathology reports, which usually contains both free-text and semi structured data, can improve the utility of the information to both human researchers and computing systems. The extracted information, when appropriately restructured and organized, enables complex and interesting queries such as, find all patients that have certain diagnosis and been treated with a specific therapy within a given period. We have developed an interactive, adaptive learning based information extraction system that, compared to existing techniques, greatly simplifies the identification, extraction and structuring of information in pathology reports.

Technology

ASLForm supports convenient transformation of information in free-text reports. The workflow follows the conventional process of manual annotation and extraction. It addition, through user interaction monitoring, ASLForm transparently and incrementally learns key data element contexts to improve its ability to automatically identify pertinent information in subsequent reports. ASLForm employs OpenNLP for basic natural language processing tasks, and implements an adaptive learning model based on Hidden Markov model (HMM).

Design

ASLForm consists of three components. The preprocessing component parses and indexes input text. The Answer generating component locates target values to fill the output form. System generated answers along with user revisions are analyzed by the adaptive learning component, to incrementally (and often quickly) improve the precision of the answer generating component. As the answer identification accuracy improves, the user's primary role evolves from annotator to verifier.

Results

ASLForm supports extraction of the following information: personal information such as gender and age, diagnosis, therapy and genetic marker, which consist of embedded attribute values from both semi-structured and unstructured text. Experiments over 50 patient records demonstrate high reliability and accuracy of ASLForm for pathology reports. For most attributes, the extraction has yielded a precision rate of over 90%.

Conclusions

Information from pathology reports holds great research and clinical values. ASLForm provides an effective tool for making such information available. The usability, adaptability and accuracy of ASLForm make it a convenient and powerful tool for extracting information from diverse pathology reports.

Imaging Informatics

Wednesday, May 14, 2014


Kings Garden South/LeBateau

Automated Detection and Characterization of Prostate Cancer-Containing Tissue Regions Using Slides Stained with AMACR, HMW Cytokeratin, and P63

Benjamin Brasseur 1 , Andrew Johnson 2 , Jonathan Henriksen 3 , Joseph Koopmeiners 4 , Anthony Rizzardi 3 , Gregory Metzger 5 , Stephen Schmechel 3

1
University of Minnesota, Medical School, Minneapolis, MN, 2 Department of Laboratory Medicine and Pathology, University of Minnesota, Minneapolis, MN, 3 Department of Pathology, University of Washington, Seattle, WA, 4 School of Public Health, University of Minnesota, Minneapolis, MN, 5 Department of Radiology, University of Minnesota, Minneapolis, MN, USA.

E-mail: bras0072@umn.edu

Content

Our laboratories are engaged in improving multi-parametric magnetic resonance imaging (mpMRI) detection of prostate cancer (PCa). Central to these efforts is spatial co-registration of pre-operative mpMRI data with post-prostatectomy pathology data (the "gold standard" for PCa presence in tissue sections). Pathologist manual annotations is prohibitively time consuming. Further, manual annotations provide partial data: PCa shows infiltrative growth such that regions of interest (ROIs) annotated as "cancer" are composed of admixed malignant and benign epithelial and stromal elements and gland lumens. Automated methods are desired to identify PCa-containing ROIs and assess fractional composition within ROIs.

Technology

We previously reported innovative SigMap software that aligns serial whole slide images (WSI), and overlay grids onto WSI images, to develop spatially arrayed pathology data. We used SigMap and image analysis software (Aperio ePathology, Leica Biosystems, Vista, CA, USA) on slides stained with triple antibody cocktail to a) identify grid ROIs containing PCa and to b) assess fractional tissue composition within ROIs. Pathologist manual annotations were benchmarks for these analyses.

Design

PCa-containing blocks were randomly selected and randomly-ordered adjacent sections were stained with H and E or triple antibody cocktail (AMACR in FastRed,34βE12 and p63 in diaminobenzidine/brown, and hematoxylin counterstain). Slides were digitized (ScanScopeXT, Aperio) and tissue areas were annotated by pathology trainees (BMB/ADJ) and validated by a pathologist (SCS). SigMap aligned adjacent sections and generated grids of 0.5 × 0.5 mm for analysis.

Results

Grid ROIs were identified using Color Deconvolution (Aperio) as PCa-containing or nonPCa-containing with 85% sensitivity and 85% specificity, versus pathologist manual annotation. Regression modeling revealed good correlation between computer-assessed composition within ROIs (%red, %brown, %blue, and %clear) and manually annotated fractional composition (cancer, benign epithelium, stroma, and luminal spaces) with correlations of 0.749, 0.598, 0.603, and 0.511, respectively.

Conclusions

Software methods can automatically detect PCa-containing ROIs, and assess fractional composition within ROIs. These methods will be useful for identification and validation of mpMRI signals useful to assess PCa extent preoperatively.

Interface of Biosample Whole Slide Images with Daedalus' Biomaterial Tracking and Management Research Software (BTM)

Clara E Magyar 1 , Juan Sebastian 1 , Azita Sharif 2 , Sarah M Dry 1

1
Department of Pathology and Laboratory Medicine, UCLA School of Medicine, Los Angeles, CA, 2 Daedalus, Inc., Cambridge, MA.

E-mail: cmagyar@mednet.ucla.edu

Content

The UCLA CAP accredited Institutional Biorepository uses Daedalus' BTM for biosample management. Before releasing banked samples, pathologists review a quality assurance (QA) slide to verify diagnosis and tumor content. We created a bidirectional interface between BTM's database and Aperio Spectrum (whole slide image database), to streamline subsequent tissue release requests and sample selection. This solution makes digital pathology a rich annotation for an integrated content approach for UCLA Biobanking/biomedical research.

Technology

Whole slide scanner: Aperio AT (Leica Biosystems, Buffalo Grove, IL) Biobanking database: BTM (Daedalus Software Inc., Cambridge, MA).

Design

We wanted to add rich content (scalable to millions of large pathology images) without affecting response-time and storage requirements. We accomplished this by adding another architectural tier. Rather than moving numerous bits around the network on every click, we show thumbnails of large images for each sample. Only on specific requests from the users do we fetch the entire image which opens in Aperio's web browser, Webscope. Adding another architectural tier introduces an identity problem among related entities (Bank Samples and their Pathology Images). We addressed this by using an identity hash-code. This solution works on SOAP protocol for web-based services. For performance enhancement, we wrote proprietary user-authenticators and SOAP response parsers.

Results

Integration was finalized January 2013. QA slides are labeled with the Daedalus Matrix barcode and scanned into Spectrum. The barcode links the slide image to the BTM sample family. One can search BTM for sample families with linked images, or directly check a sample's image tab. If a linked image exists, an image thumbnail appears adjacent to the image URL; clicking on the thumbnail directs the user to Webscope for viewing. To date, over 400 slides of banked sample families have been digitized and linked.

Conclusion

We created a seamless, bidirectional interface of our biorepository sample tracking system (BTM) with our whole slide image database (Spectrum) creating a more efficient and rich annotation system when searching for and releasing specimens. This facilitates researcher selection of samples and improves efficiency of QA review as consulting pathologists view digital images on demand.

Automated Identification of Glomeruli in Volumetric Multiphoton Images Using Texture Features

Eben Olson, Richard Torres

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

Content

Optical tissue clearing allows large volumes of tissue to be imaged in three dimensions with subcellular resolution. While additional qualitative and quantitative morphometric information is made available by the use of these techniques, the size of the resulting data sets make manual analysis more difficult or, in some cases, infeasible. An example of some research and clinical importance is identification, counting, and volumetric measurement of glomeruli in kidney samples. We report the development of an automated method for glomerular detection in volumetric multiphoton images of mouse kidney employing frequency-based texture segmentation algorithms.

Technology

A custom built multiphoton microscope was used in conjunction with optical tissue clearing to image millimeter-thick sections of mouse kidney tissue stained with protein dye. Images were segmented using several texture feature analysis techniques including binary Gabor patterns with a random forest classifier. Additional gradient-based and morphometric analysis was used to refine the results of this segmentation. The software was built using open-source libraries and custom code written in Python.

Design

A dataset consisting of 71 stacks of 300 images covering a total volume of approximately 9 mm^3 was analyzed. A subset of 100 randomly selected images were hand labeled for classifier training and testing. An additional 15 randomly selected sub-volumes of 100 images were manually segmented and used as the basis for validation testing. The full dataset was then automatically segmented by the newly developed software. An overlap of greater than 50% of the glomerular volume between manually determined region and automated segmentation was counted as adequate detection.

Results

Binary Gabor pattern-based texture segmentation was successfully applied to histologic image analysis on large three-dimensional data sets. Automated analysis achieved high accuracy, correctly detecting 163/174 glomeruli (sensitivity = 94%), while maintaining a small false discovery rate (10/173).

Conclusions

Machine learning of texture features provides a valuable tool for segmentation and preliminary analysis of large volumetric data sets. Binary Gabor pattern-based analysis is a robust approach for segmentation that is not dependent on color thresholding. Our automated image processing system can effectively identify regions containing glomeruli, allowing the extraction of volumes of interest for further analysis.

A MapReduce Based High Performance Whole Slide Image Analysis Framework in the Cloud

Hoang Vo 1 , Dejun Teng 1 , Yanhui Liang 2 , Ablimit Aji 1 , Jun Kong 2 , Fusheng Wang 2 ,

1
Department of Mathematics and Computer Science, Emory University, Atlanta, GA, 2 Department of Biomedical Informatics, Emory University, Atlanta, GA.

E-mail: hoang.vo@emory.edu

Content

Systematic analysis of high resolution whole slide images has many potential applications to support biomedical research and disease diagnosis. Due to the enormous size and dimensions of whole slide images, traditional image analysis techniques cannot be performed directly due to memory limitation and extremely long execution time for each image. In-house computing infrastructures often have limited computing capacity to process large batches of such images. We develop a scalable and cost effective MapReduce based high performance image analysis framework for massive whole slide image processing in the cloud by integrating image analysis algorithms and spatial queries. The framework is general and can be adapted to different image analysis algorithms.

Technology

We develop a MapReduce-based framework by adapting image analysis algorithms into MapReduce based processing pipelines running on commodity clusters and Amazon Elastic MapReduce (EMR), where image tiles are stored in distributed file systems (Hadoop Distributed File System and Amazon S3 respectively). Tile based parallelization is performed where task management and load balancing is automated by MapReduce. We use a cloud based spatial querying engine Hadoop-GIS to amend the problem of boundary objects due to partitioning and integrate it into the image processing pipeline.

Design

Tiles are extracted from whole slide images using a grid based overlapping partitioning scheme. Each data file consisting of single tile is supplied to a "map" function, where they are processed producing segmented boundaries. Segmented boundaries are further processed in a "reduce" function which provides geometry correction and duplicate elimination. File results are aggregated and stored in the distributed file system.

Results

We have performed an experiment for nuclei segmentation from a dataset of 500 whole slide pathology images using Amazon Elastic MapReduce (EMR). The framework achieves an almost perfectly linear scalability. In addition, the duplicate elimination step handling objects spanning tile boundaries takes a very small portion of execution time, indicating an insignificant overhead for producing sound results.

Conclusions

Our experiments have demonstrated a very effective, scalable and cost effective framework for pathology image analysis. The framework can handle cases of objects spanning multiple boundaries, while incurring a small addition execution time overhead. Furthermore, the post-processing function can be effortlessly extended to perform addition image analysis and result validation. The framework is highly adaptive for different image segmentation and feature extraction algorithms.

Applied Pathology Informatics

Wednesday, May 14, 2014


Kings Garden North

Alchemy: A Web 2.0 Real-Time Quality Assurance Platform for Human Immunodeficiency Virus and Hepatitis C Virus Quantitation Tests

Emmanuel Agosto-Arroyo, Gina M. Coshatt, Seung Lyung Park

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

Content

Quality assurance (QA) is integral to a well-run molecular diagnostics laboratory. QA statistical calculations are usually manual, time consuming, and often error-prone. The aim of this project was to develop a web-based real-time QA platform that would (a) Aautomate quality control reporting at the molecular diagnostics laboratory of the University of Alabama at Birmingham, and (b) minimize the time expended in preparing these reports.

Technology

A Dell Precision T3600 (Intel Xeon E5-1603 @ 2.8 GHz; 16GB DDR3 SDRAM; 256GB SSD; 1TB HDD; Microsoft Windows 7 × 64) was used to host a standard LEMP (Ubuntu Linux Server 12.04 LTS; ngin×1.5.6; MariaDB 5.5.32; PHP 5.4) stack virtual machine on Oracle VM Virtualbox virtualization software.

Design

Using the LEMP stack, we designed, built, and deployed a QA platform code-named "Alchemy", targeting our Human Immunodeficiency Virus (HIV) and Hepatitis C Virus (HCV) quantitation tests. Alchemy was fed with data autogenerated monthly as comma-separated value (CSV) files from the clinical pathology laboratory information system. Alchemy calculates key QA statistics in real time, including average TAT in days, tests falling outside expected TAT ranges, and test result ranges.

Results

Before Alchemy, reporting QA for the HIV and HCV quantitation tests took 45-60 minutes of personnel time per test per month. With Alchemy, that has shrunk to 15 minutes total per month.

Conclusions

Alchemy has significantly decreased the time and the human error associated with QA report generation in our molecular diagnostics lab. Other tests will be added in future updates. This effort shows the utility of informaticist-supervised resident/fellow programming projects as learning opportunities and workflow improvements.

A Real-time Web 2.0 Antibiogram System for the Clinical Microbiology Laboratory

Matthew Cain 1 , Stephen Moser 2 , Seung Park 3

University of Alabama at Birmingham, 1 Department of Pathology, 2 Department of Pathology, Division of Microbiology, 3 Department of Pathology, Division of Informatics, Birmingham, AL. E-mail: mdcain@uabmc.edu

Content

Committees composed of microbiologist, pharmacists, and Infectious Disease clinicians, serve an important role in efficacious antimicrobial therapy. Currently, our institution releases annual antibiograms detailing summary antibiotic susceptibility data. Prior to this project, a tab-separated value (TSV) file was generated by the electronic medical record system, which was then imported into Microsoft Access for further manipulation and data harmonization. In this project, we sought to automate this process as well as provide an efficient means for clinicians to see real-time data regarding antibiotic susceptibilities.

Technology

Server Hardware: Dell Precision T3600; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 12.04 LTS 64-bit; Web Server: nginx 1.5; Database Management System: MariaDB 5.5; Programming Language: PHP-FPM 5.3; User Interface Framework: Twitter Bootstrap 2.3.

Design

The TSV files were structurally analyzed. Tables for patient information, antibiotics, bacteria, bacterial source, and observations were created in MariaDB. A complete web application was written in PHP-FPM, its relational database connector (mysqli), and Twitter Bootstrap with the following functionality: (1) One-click upload of new monthly data, complete with automated detection and removal of duplicate/extraneous data; (2) dynamic generation of monthly reports; (3) real-time database search [Figure 1].



Results

Development of an upload system that parses and prepares data reduces the microbiologist workload by approximately 16 hours annually. Furthermore, the database represents real-time susceptibilities with easy access. Previously, antibiograms were released annually and only contained the number of isolates and the percent susceptibility on a given bacteria-antibiotic pair. Our database not only supplies up-to-date information, but provides the ability to search by bacterial source and hospital location.

Conclusions

We successfully developed software that reduces the time to develop conventional antibiograms, provides real-time data, and easy accessibility to highly detailed susceptibility statistics. The system parses and processes text files exported in Health Level 7 (HL7) format. Hospitals commonly use this format; therefore, this program could be used in other institutions with minor adjustments. Future goals include the incorporation of advanced statistics, such as cluster analysis to identify resistances in specific hospital units, as well as panels focused to specific patient populations.

Development of an Informatics Tracking Solution to Enhance Flow Cytometric Specimen Processing

Raymond E Felgar 1 Christine G Roth 2 , Anthony L Piccoli 2 , Karen L Freilino 2 , Kevin J Nauman 2 , Ralph A Nauman 2 , Michael Fitzgerald 2 , Liron Pantanowitz 2 , Anil V Parwani 2 , Steven H Swerdlow 2

1
University of Pittsburgh Medical Center, Department of Pathology, Division of Hematopathology, Pittsburgh, PA 2 University of Pittsburgh Medical Center, Department of Pathology, Pittsburgh, PA. E-mail: felgarre@upmc.edu

Content

Tracking of specimen status in high volume flow cytometric laboratories can be problematic. Laboratory efficiency may be reduced if this hampers determination of panel choice, sample priority, and estimated time left to completion of specimens being processed prior to pathologist staff review and sign-out. Our aim was to design a tracking system for processing flow cytometry specimens in a large academic medical center, with use of pre-existing tools in our anatomic pathology laboratory information system (AP-LIS), provide realtime display of case status for technologists and assist with short and long-term laboratory management issues.

Technology

The anatomic pathology laboratory information system (AP-LIS; CoPath, Cerner) histology tracking function was customized to record accessioning, sample type, priority, staining, acquisition, and completion of analysis steps for flow cytometry specimens. A tracking report was designed to allow for real-time assessment of specimen status within the lab on large screen monitors in the laboratory and updated in real-time by employing dedicated computers and System Scheduler software (Splinterware Software Solutions).

Design

Flow panel choices, coupled to technical billing for markers, were captured using the LIS stain ordering function. Tracking log report(s) also allow end users to determine the current stage of processing of a particular specimen in question or for a range of specimens, including completed cases (based on accession date, for example).

Results

The tracking report successfully displayed the dynamic status of all flow cytometry procedures and markers being tested [Figure 1]a. The display included color- coding by status. The report was also available on demand in the LIS to pathologists and technical staff. Tracking log reports [Figure 1]b allow one to track both pending and completed work and can also serve as a data reservoir for functions such as turn-around-times for specific steps in specimen processing; this can also facilitate work-flow analysis.



Conclusion

Real-time tracking of flow cytometric specimen processing and analysis status can be performed using existing AP-LIS functions. This tracking tool allows technical and supervisory staff to readily monitor laboratory workflow. The tracking data collected also allows for subsequent workflow analyses.

The Ongoing Evolution of the Core Curriculum of a Clinical Pathology Informatics Fellowship

Andrew M. Quinn 1 , John R. Gilbertson 2

Departments of Pathology, 1 Brigham and Women's Hospital, Boston, MA, 2 Massachusetts General Hospital, Boston, MA.

E-mail: aquinn3@partners.org

Content

The Partners Healthcare System's Clinical Fellowship in Pathology Informatics (Boston, MA, USA) faces ongoing challenges to the delivery of its core curriculum in the form of a new class of fellows with new and varying educational needs and increasingly fractured, enterprise-wide commitments; taxing electronic health record and laboratory information system implementations; and increasing interest in the subspecialty at the academic medical centers in the network.

Technology

(There is no relevant technology to discuss.).

Design

In response, the fellowship has piloted a network-wide pathology informatics lecture series and regular learning laboratories, as well as a modification of the existing didactic sessions [Figure 1]. Lectures are given by the informatics faculty, current and former fellows and information systems members in the network, and are open to all professional members of the pathology departments at the academic medical centers. Learning laboratories consist of small-group exercises geared toward a variety of learning styles and are driven by both the fellows and a member of the informatics faculty. These learning laboratories have also created a forum for discussing real-time and real-world pathology informatics matters, and incorporating awareness of and timely discussions about the latest pathology informatics literature. Didactic sessions now focus on group discussions of fellows' ongoing projects as well as updates on the enterprise-wide electronic health record and laboratory information system implementations and directed questions about weekly readings, whereas they had previously included more formal discussions of the four objectives of the core curriculum: information fundamentals, information systems, workflow and process, and governance and management.



Results

These changes have diversified the delivery of the fellowship's core curriculum, increased exposure of faculty, fellows and trainees to one another, and better distributed teaching responsibilities among the entirety of the pathology informatics asset in the network.

Conclusions

Though the above approach has only been in place for seven months, its characterization herein allows for continued discussion of evolving educational opportunities in pathology informatics and medical informatics in general, and highlights the importance of having a flexible fellowship with active participation from its fellows.


   Platform Short Abstract Presentations Top


Wednesday,May 14, 2014 (9:00 am - 5:20 pm)

Thursday, May 15, 2014 (9:00 am - Noon)


Location: Wyndham Grand Pittsburgh Hotel, Kings Garden North

Informatics Infrastructure Requirements for Clinical Next Generation Sequencing: Challenges and Solutions

Somak Roy 1 , Ryan Mitchell 2 , Gary Burdelski 2 , Jeff McHugh 2 , Shan Zhong 1 , Kevin Nauman 2 , Marina N. Nikiforova 1 , Yuri E. Nikiforov 1 , Anil V. Parwani 1 , Liron Pantanowitz 1

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

Content

Introduction of Next Generation Sequencing (NGS) in clinical molecular laboratories has unveiled obtrusive informatics challenges. Successful hardware and software solutions to support NGS data analysis and storage for the laboratory need to be scalable, redundant, secure and support sharing in an enterprise environment of a large academic center. We present our institution's experience in developing an informatics framework to support clinical NGS testing.

Technology

The clinical molecular laboratory at our institution houses five NGS sequencers and accompanying vendor provided high-performance workstation servers with 7 to 10 terabytes of local storage per server. Scale out network attached storage (SONAS) and high-performance enterprise storage (HPES) at our institution's HIPAA-compliant data center were used for archiving clinical NGS test data. High-bandwidth network connections (up to 1GB/s) facilitated data transfer [Figure 1]. Dynamically scalable Windows-based Virtual Machines (VM) facilitated data processing virtually and new algorithm development for NGS molecular testing.



Design

Using high-bandwidth network connections, raw signal files were moved to the SONAS archives (45TB) whereas data files starting and downstream to FASTQ files were stored in the HPES pool (1TB/sequencer). The latter provided immediate access to FASTQ and other files for bioinformatics analysis on VMs. Multiple laboratory staff access VMs using Microsoft Windows native remote desktop solution and are able to run concurrent resource intensive NGS data analyses.

Results

This enterprise informatics infrastructure initially provided highly redundant and secure storage to prevent accidental loss of clinical test data. However, the off-site location of raw data slowed data analysis. As a result, VMs were created to provide direct access to archived data, avoiding the need for transfer of massive data files over the network. To optimize NGS analysis time on VMs, input data for bioinformatics analysis was stored using HPES.

Conclusion

Informatics infrastructure required significant customization and close collaboration with laboratory professionals to successfully optimize NGS data analysis and storage. This highly scalable infrastructure has facilitated the implementation of HIPAA-compliant NGS testing in our high volume clinical molecular laboratory. This framework may provide a model for other institutions planning to pursue clinical NGS testing.

A Free-Standing Clinical Laboratory Data Warehouse in the Era of Enterprise Clinical and Research Warehouses: 16 Years of Experience, Future Options, and Strategic Considerations

Philip J. Boyer 1 , Bradley B. Brimhall 3 , Connie Williamson 2 , Thurston Matsuura 2 , Chad Vanderbilt 1 , Joan Coleman 2 , Ronald Lepoff 1

1
Department of Pathology, University of Colorado School of Medicine and University of Colorado Hospital, 2 Clinical Laboratory, University of Colorado Hospital, Aurora, CO, 3 Department of Pathology, University of Mississippi, Jackson. E-mail: philip.boyer@ucdenver.edu

Content

A clinical laboratory data warehouse (CLDW) was established over 16 years ago at the University of Colorado Hospital (UCH). Laboratory Information system (LIS) data is downloaded to the warehouse daily. UCH will be transitioning from Cerner Classic to Epic Beaker in 2015 and the continued need for the CLDW has been questioned.

Technology

A SQL server database was established with several tables and over 70 fields (patient demographics, billing, and test information). As it contains protected health information, access is strictly limited. Laboratory personnel access data by either (1) Web page-based SQL queries or (2) request to an LIS technologist for a complex query.

Design

The cost and utilization of the CLDW is evaluated and the feasibility of using current and future enterprise database options for queries is considered.

Results

The initial cost to create the CLDW (hardware, software, professional time) was approximately $61,000; annual upkeep costs are less than $20,000. The CLDW is extensively utilized for both routine and ad hoc queries including assessment of turn-around time, test utilization, infectious disease monitoring, quality improvement, and workflow management. The UCH Epic database contains over 8 years of test result data; however, LIS test-specific information is not present. That database will contain Beaker LIS data beginning in 2015. An enterprise research data warehouse will go live in 2015; while detailed LIS data could be incorporated, historic data would likely not be sent.

Conclusions

The UCH CLDW contains over 1.5 decades of patient and test-related data. It facilitates routine and ad hoc queries to meet clinical, operational, financial, and research goals, justifying the modest development and maintenance costs. Maintaining the CLDW and interfacing Beaker data is justified, particularly given (1) the need to accumulate many months of data prior to meaningful data mining in one of the enterprise databases and (2) the inherent value of over 16 years of longitudinal data. Maintaining the CLDW also presents critical strategic advantages to the laboratory, particularly during the time of transition to a new LIS, by allowing laboratory autonomy, with direct data access, using familiar and time-tested query tools.

"WSI ZoomViewer": A Vendor Neutral HTML5 Whole Slide Image Viewer

Eugene Tseytlin 2 , Anil V. Parwani 1 , William Cable 3 , Liron Pantanowitz 1

1
Department of Pathology, University of Pittsburgh Medical Center, 2 Department of Biomedical Informatics, University of Pittsburgh, 3 University of Pittsburgh Medical Center, Information Services Division, Pittsburgh, PA.

E-mail: tseytlin@pitt.edu

Content

Digital whole slide images (WSI) are being increasingly used in Pathology for a variety of clinical, education and research purposes. Unfortunately, unlike other medical fields such as Radiology, viewing of digital images from various WSI vendors continues to be a challenge as there are no standard image formats. Most vendors offer their own proprietary WSI solutions for various use cases. Their software is often bundled with WSI scanners, is expensive and does not always inter-operate well with other vendors and image formats. There are currently limited viewer solutions that can open WSI files from multiple vendors, such as SlideViewer (http://slidetutor.upmc.edu/viewer) and OpenSlide (http://openslide.org/). SlideViewer uses Java technology which requires frequent updates, is no longer popular for the web front-end applications, and is not supported by Apple mobile devices. Therefore, our aim was to develop a new vendor neutral WSI viewer based on the latest HTML5 technology.

Technology

The WSI viewer was based on the OpenSeadragon (http://openseadragon.github.io/) HTML5 framework, which is an open source implementation of Microsoft DeepZoom technology.

Design

OpenSeadragon allows users to view very large image files by breaking underlying images into tiles and transmitting only tiles at an appropriate zoom level and view position to save on bandwidth. Instead of converting WSI files to an OpenSeadragon format, the new viewer permits the application programming interface to interact with a vendor's WSI server software natively. This allowed the new viewer to access remote WSI hosting sites without the need to install any additional software on servers.

Results

The new HTML5 viewer ("WSI ZoomViewer") worked on all modern web browsers and all mobile platforms, including Apple devices. The current version supports Leica and Hamamatsu vendors, with the ability to add additional formats in the future. Additional features that were developed specifically for our institution's telepathology consultation portal were multiple slide support, navigation panel, orientation specific layout for mobile devices, and slide snapshot uploading and viewing [Figure 1].



Conclusion

WSI ZoomViewer is a vendor neutral HTML5 digital slide viewer that fills an important niche in today's digital pathology market. The advantage of HTML5 is its ease of deployment. This novel WSI viewer will allow institutions to avoid vendor lock-in by working with digital files from different vendor's, and facilitate collaboration by allowing heterogeneous WSI repositories hosted by different institutions to be easily accessed.

Reader Studies for Digital Pathology: Software for Simulation, Analysis, and Sizing

Weijie Chen, Adam Wunderlich, Nicholas Petrick, Brandon D. Gallas

Food and Drug Administration, Center for Devices and Radiological Health, Office of Science and Engineering Laboratories, Silver Spring, MD.

E-mail: weijie.chen@fda.hhs.gov

Content

In certain types of clinical studies comparing Whole Slide Imaging with optical microscopy, multiple pathologists (i.e. readers) read/evaluate images of multiple cases from both modalities and the pathologist's diagnostic assessment for each case is converted to a binary score by comparing with a reference standard (1: Agree, 0: Disagree). The summary metric is the probability of agreement obtained by averaging the binary scores for each modality. The study hypothesis is that Whole Slide Imaging is non-inferior to optical microscopy where the conclusion should generalize to both the population of readers (pathologists) and the population of cases (patients). In order to achieve such a generalization, both readers and cases in the study should be treated as random samples representative of their respective populations. We present a framework for simulation, analysis, and sizing of such studies.

Technology

We introduce a computational model that simulates binary data with a correlation structure that allows for accounting for two sources of variation observed in real studies: variability due to the random reader sample and the random cases. We illustrate procedures for using our simulation model to validate a statistical analysis method and to estimate the sample size (both the number of readers and cases) needed to achieve a desired statistical power.

Design

We adopt two methods for the analysis of reader study data: (a) The method of Obuchowski-Rockette-Hillis that uses a correlated analysis of variance model; and (b) the method of Gallas that is based on the U statistics. We use our simulation model to validate these two methods in terms of coverage probability of 95% confidence intervals for percentage agreement.

Results

We found that the 95% confidence intervals from both methods have satisfactory coverage probability and therefore are appropriate for the analysis of binary reader study data. We also demonstrated the application of our simulation model to sizing a study.

Conclusions

The general methodological framework we present here and the associated freely-available software package (http://code.google.com/p/imrmc/wiki/iMRMC_Binary) are useful for simulation, analysis, and sizing of multireader multicase reader studies with binary assessments.

An Open Standards, Enterprise Whole Slide Imaging Warehouse and Layered Governance Model

Seung Park 2 , Sean Wilkinson 1

1
Department of Biomedical Engineering, University of Alabama at Birmingham, 2 Department of Pathology, Division of Informatics, University of Alabama at Birmingham.

E-mail: seungp@uab.edu

Content

Many whole slide imaging (WSI) solutions exist, most with a coupled client-server viewer model. These solutions (a) are hardware intensive, (b) assume unified metadata governance, (c) are Windows-only, and (d) provide zero meaningful interoperability. We have developed an open-standards, Web 2.0 WSI repository and layered governance model that allows for divergent metadata access depending on the use case (educational, research, clinical).

Technology

Hardware: Dell PowerEdge 2950; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 12.04 LTS 64-bit; Web Server: nginx 1.5; Database Management System: MariaDB 5.5; Programming Language: PHP-FPM 5.3; WSI Decode Library: OpenSlide 3.4; Image Processing Library: VIPS 4.38; Deep Zoom Image (DZI) Viewer: OpenSeadragon 1.0.

Design

Upon WSI file generation, these files are converted into DZI pyramids by an OpenSlide + VIPS powered application. DZI pyramids, written in a self-describing tiled JPEG file-based interchange format, allow for direct, high-performance access to arbitrary regions of interest. The deidentified DZIs are uploaded to a public-facing web server. The corresponding clinical metadata are uploaded to a hospital-access-only database server (code-named 'ASPIRE'). Public accession numbers (PANs) on the public server are unrelated to the true clinical accession numbers (CANs). An OpenSeadragon-based viewing app on ASPIRE translates CANs into PANs and embeds the appropriate WSI from the public server, allowing clinicians to utilize CANs only within the hospital intranet. Researchers and educational users do not have access to CANs, and simply use PANs instead.

Results

Our WSI governance system successfully provides interchange for disparate user groups who have hitherto utilized disparate WSI file formats. Prior to implementation of our system, there were 8 different WSI repositories maintained in individual silos; these have been unified. All WSIs are available for real-time viewing, and our system supports hundreds of concurrent users with no discernable performance degradation.

Conclusion

Our system is now the department-mandated solution for WSI interchange and data governance going forward. It has supplanted individual slide boxes at our medical school, and forms the basis of our electronic consult slide archiving repository. Multiple advanced image analytics pipelines for our system are also in development.

Integrating Robotic Analyzer with LIS to Reduce Human Errors in HPV Reporting

Peter Gershkovich, John Sinard

Yale Medical School, Department of Pathology, New Haven, CT.

E-mail: peter.gershkovich@yale.edu

Content

Human errors in reporting lab results for high-risk Human Papilloma Virus (HPV)- a sexually transmitted viral infection that causes cervical cancer - may lead to a wrong treatment and can damage personal relationships. Despite rigorous quality control, manual transfer of data from the analyzer to the LIS resulted in a 0.04% undetected error rate in our lab, or approximately 24 cases per year. In addition, we found 0.2% of cases where existing QA processes identified human errors leading to approximately 120 amended reports per year. An electronic link that sends results from the analyzer to our LIS, eliminating the need for manual entry of the results, can prevent these errors.

Technology

Software was written using the Java programming language and Open Source frameworks (Quartz, Apache Mina, Apache Commons, etc.). It consists of two modules: ASTM interface with a Cobas 4800 analyzer from Roche and the specimen-tracking and reporting module linked to the LIS.

Design

An HPV Result Interface Module was added to our in-house custom-built Histology Asset Tracking system. A new concept of "derived specimen" was introduced to track the HPV samples and to provide a link between the tests accessioned in the LIS and the tests tubes placed in the analyzer. The interface between the HPV analyzer and the LIS is a web-based application that communicates over Ethernet with the LIS and the analyzer using TCP/IP and ASTM protocols. The results are automatically incorporated as addenda into Final Reports of the core LIS.

Results

The system enabled tracking and automatic transmission of the results from the analyzer to the LIS eliminating the manual transfer of results for HPV screening. There were no errors in HPV reporting after the implementation of the project based on the evaluation of 2321 cases. There were no amendments related to incorrect reporting of results.

Conclusions

The logic of assembling HPV results and recommendations in a Cytology Report is algorithmic. The exceptions are rare and easy to detect. The integration between the LIS and Robotic Analyzers of HPV eliminates manual transfer of results and improves the accuracy of HPV reporting.

This is Your Brain on Informatics: A Total-Immersion Data Sciences Course for the Next Generation of Informaticists

Timothy Kennell Jr. 1 , Vincent Laufer 1 , Robinna Lorenz 1 , Seung Park 2

1
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. E-mail: tikenn@uab.edu

Content

The need to train informaticists is critical, yet such training is rarely built into standard curricula. Even at informatics-savvy institutions, informatics education in medicine is almost invariably an afterthought, usually focusing on imparting testable knowledge rather than on building technical prowess. We have developed and implemented a one week, 40 hour long, total-immersion experience in clinical and research informatics/data sciences that enables trainees at all levels (medical students to faculty; most with zero prior exposure to informatics) to independently create and administer complex informatics systems.

Design

On day 1, students are introduced to computer science in an axiomatic fashion, learning the fundamentals of systems architecture and algorithmic thinking while they assemble a functional LEMP (Linux, nginx, MariaDB, PHP-FPM) stack from the ground up, and write their first set of exploratory programs in both procedural and object-oriented programming languages. Day 2 covers relational database design, normalization, querying and maintenance. Day 3 covers Web 2.0 user interface elements. Days 4-5 comprise a supervised hackathon in which students form into groups, design real-world informatics projects, and begin implementation. Students continue working on their projects until completion (usually with light supervision).

Technology

Virtualization Software: Oracle VirtualBox4.3; Guest Operating System: Ubuntu Linux Server 12.04 LTS 64-bit; Web Server: Nginx1.5; Database Management System: MariaDB 5.5; Programming Languages: PHP-FPM 5.3, HTML5, CSS3, JavaScript, and SQL; User Interface Library: Twitter Bootstrap 3.3.

Results

Over two sessions of this class thus far, 21 students have undertaken 10 major informatics projects, with a current total of 14 abstracts submitted to national research conferences. 3 are medical education projects, all of which have been formally adopted by our institution. 4 are research-oriented analysis projects, in various stages of completion. 3 are clinical informatics projects, each of which (a) generates significant time and money savings in clinical operations and (b) has been formally adopted by our institution.

Conclusions

Engineering education-derived practices, including real-world group projects and technical skill building, reap major dividends when applied to informatics education. Future iterations of this class will see the completion and deployment of more major informatics systems.

Large-Scale Retrospective Data Analysis for Postmarketing Surveillance of Point-Of-Care Glucose Meter Accuracy

Lee Schroeder, May Louie, Nigam Shah

Department of Pathology, Stanford University, Stanford, CA.

E-mail: lschroed@stanford.edu

Content

Tight glycemic control protocols for critically ill patients routinely incorporate point-of-care (POC) glucose meters. This is despite FDA warnings against such use due to concerns with accuracy and a scarcity of postmarketing surveillance. We demonstrate the utility of mining retrospective laboratory information system data for ongoing surveillance of POC glucose meters.

Technology

Sunquest extracts into a SQL database allowing direct query of aggregated data.

Design

It is common practice in intensive care units to use arterial blood for POC glucose measurements, often with simultaneous draws for other laboratory testing. We queried our SQL database for all events in which both an intensive care unit POC and standard glucose measurement (automated hexokinase or whole-blood glucose oxidase) were collected within the same 5-minute period. For comparison, we used 144 data points from annual POC instrument correlation studies with hexokinase methodology.

Results

We found 20,537 pairs of POC and standard glucose measurements meeting criteria over 4 years. 32% of events exceeded recent FDA draft guidance (+/-7 mg/dL for values <70 mg/dL, +/-10% for values >=70 mg/dL). Systematic bias of the POC glucose meter was +5.0% (ci +4.8%/+5.2%) within the interval of 60 to 200 mg/dl. Bias increased to +18.4% (ci +4.3%/+32.6%) at lower standard glucose values and decreased to -20.8% (ci -25.4%/-16.1%) for glucose values above 350 mg/dl. A similar bias was found whether POC collection occurred before or after the standard (P < 0.59), or whether standard measurements were resulted 0-30 or 30-60 minutes after POC results (P < 0.34). Correlation study bias was not statistically different than that of retrospective analysis (P < 0.075). Of 364 hypoglycemic events, 117 were missed by POC testing. Of these, only 2 were misclassified such that an insulin dose would have been administered according to our tight glycemic control protocol. Multiple linear regression against POC bias was performed on concurrent testing from the comprehensive metabolic, cbc, and arterial blood gas panels. A robust interference by hemoglobin was found such that a shift of hemoglobin from 14.0 to 7.0 g/dl was associated with +13.7% (ci +12.7%/+14.8%) bias.

Conclusion

This analysis shows proof-of-concept for harnessing retrospective patient data in postmarketing surveillance of POC instruments.


   Electronic Poster Session Top


Wednesday, May 14, 2014

Presented in the Grand Foyer, 2 nd Floor, Wyndham Grand Pittsburgh Hotel

Archival System for Pathology Informatics, Research and Education: A Web-Based Image Platform for Molecular Pathology

Emmanuel Agosto-Arroyo, Mary P. Branscomb, Shuko Harada, Seung Lyung Park

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

Content

Image collection and retention are an imperative part of the Fluorescence In Situ Hybridization (FISH) analysis process. Photographic or digitized microscopic images are retained for documentation of all FISH assays. For neoplastic disorders, the minimum required time for retention is 10 years, which can translate to the requirement of a large storage volume, if the images are being saved in a local or shared drive.

Technology

A Dell Precision T3600 (Intel Xeon E5-1603 @ 2.8 GHz; 16GB DDR3 SDRAM; 256GB SSD; 1TB HDD; Microsoft Windows 7 × 64) was used to host a standard LEMP (Ubuntu Linux Server 12.04 LTS; nginx 1.5.6; MariaDB 5.5.34; PHP 5.3.10; Graphic library: ImageMagick 6.6.9-7).

Design

Piwigo 2.5.3 - an open-source PHP-based image gallery program - was customized in order to support image file upload from a departmental shared drive and then codenamed ASPIRE. This shared drive was slow and prone to network congestion-derived errors, yet was the place where all FISH images had been archived since 2012. These images were uploaded to ASPIRE, organized based on our test menu and placed in individual subgalleries by their case number. They were tagged according to the test type and result. All 2014 images have been directly saved to ASPIRE, without the need of saving into other drives.

Results

Before ASPIRE, saving a FISH image was a tedious process, including creating a folder, saving the images and copying them in a shared drive. These tasks used to take up to 7 minutes. After developing the ASPIRE web page, it has been an easier and direct process taking less than 2 minutes. ASPIRE has also allow for other capacities, like tagging the images with important information, facilitating the image search process, allowing quick searches based on case number, as well as, on results or test type. File search time has been decreased from 4-5 minutes to 3 seconds.

Conclusions

Aspire has proven to be an excellent tool for saving images from the different FISH cases performed in our laboratory. It has improved the workflow for saving these images and also accessing them for revision, decreasing the amount of time spent in these activities. Other image folders are being and will be added to the system.

ETOX: A Real-time Web 2.0 Toxicology Database

Matthew Cain 1 , Gregory G. Davis 2 , Seung Park 3

1
Department of Pathology, University of Alabama at Birmingham, 2 Department of Pathology, Division of Forensic Pathology, University of Alabama at Birmingham, 3 Department of Pathology, Division of Informatics, University of Alabama at Birmingham. E-mail: mdcain@uabmc.edu

Content

Forensic pathologists often rely on toxicology results to explain cause of death. Case reports and compilations of case reports are the mainstay of toxicology information when determining if a drug caused or contributed to an individual's death. We report our development of a real-time searchable toxicology database that catalogues not only all toxicology results, but demographics, body habitus, health conditions, and manner of death.

Technology

Server Hardware: Dell Precision T3600; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 12.04 LTS 64-bit; Web Server: nginx 1.5; Database Management System: MariaDB 5.5; Programming Language: PHP-FPM 5.3; User Interface Framework: Twitter Bootstrap 2.3.

Design

Our system consists of two parts: an extraction program and the Web 2.0 searchable database. First we created the extraction program, which harvests data from autopsy and toxicology reports and then compiles these sets of data into a common spreadsheet that can be divided appropriately for our online database. Then, tables for pathologist, institution, patient information, case, measurements, and observations were created in MariaDB [Figure 1]. A complete web application was written in PHP-FPM, its relational database connector (mysqli), and Twitter Bootstrap with the ability to form complex queries including multiple drugs, health conditions, and demographics.



Results

Manual extraction of autopsy and toxicology data averages 204 and 145 seconds per report, respectively (P = 0.0002 and P = 0.002). Our extraction program reduces this time to 1 second/report. Currently, even complex searches yield near instantaneous results. These search results provide averages and statistics on the following: demographics, organ masses, health conditions, cause and manner of death, drug results and an option to select individual autopsies that concluded drug intoxication as the cause of death.

Conclusions

Our system significantly reduces the time to extract autopsy and toxicology information. It provides real-time toxicology data as well as the ability to easily search for complex scenarios. While published case reports provide data on only a few deaths per report, our system provides detailed information on thousands of cases. Future goals include the incorporation of advanced statistics and a more sophisticated upload system so that other institutions can join our database.

PEIR-VM: A Universal, Open Standards, Web 2.0 Whole Slide Imaging Repository

Alex Feldman 1 , Jeremie Lever 2 , Timothy Awtrey 3 , Israel Ponce-Rodriguez 3 , Peter Anderson 4 , Matthew Anderson 4 , Seung Park 5

1
Department of Pathology, University of Alabama at Birmingham, 2 School of Medicine, NIH Medical Scholars Training Program, University of Alabama at Birmingham, 3 Department of Pathology, Division of Information Services, University of Alabama at Birmingham, 4 Department of Pathology, Division of Molecular and Cellular Pathology, University of Alabama at Birmingham, 5 Department of Pathology, Division of Informatics, University of Alabama at Birmingham. E-mail: feldm001@gmail.com

Content

Whole slide imaging (WSI) technology is increasingly used in medical education and research. However, significant barriers (e.g. closed file formats, lack of appropriate viewers for operating systems other than Windows) to universal adoption remain. We present an open source, open standards, Web 2.0 whole slide imaging repository system designed and implemented for ease of deployment, ease of use, and infinite extensibility.

Technology

Hardware: Dell PowerEdge 2950; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 12.04 LTS 64-bit; Web Server: nginx 1.5; Database Management System: MariaDB 5.5; Programming Language: PHP-FPM 5.3; WSI Decode Library: OpenSlide 3.4; Image Processing Library: VIPS 4.38; Deep Zoom Image Viewer: OpenSeadragon 1.0.

Design

OpenSlide and VIPS were linked and cross-compiled from Linux to Windows 32-bit binaries. A graphical user interface (code-named 'WholeBuddy') enabling single-click conversion of multiple WSI formats (notably.SVS and.BIF) was created. Previously-generated pathology and histology teaching WSIs were converted into DZI pyramids using WholeBuddy. A virtual machine running a LEMP (Linux, nginx, MariaDB, PHP-FPM) stack was generated, and a web app was created allowing for (a) Easy upload of WSIs with their associated metadata and (b) real-time display of WSIs via OpenSeadragon.

Results

PEIR-VM currently contains over 300 WSIs for teaching and research. It is viewable with any modern web browser; this is critical at an institution where 61% of students use Mac OS X instead of Windows. The tiling schema for DZI pyramids is into JPEG files with regularized filenames, making regions of interest trivial to access once identified. Conversion of WSIs into DZIs increases file size by a factor of 1.26-2.01 (average 1.52); we find this an acceptable tradeoff, as this removes the processing bottleneck that live WSI streaming platforms traditionally display. For instance, even once ported to an underpowered Raspberry Pi, PEIR-VM can support over 300 concurrent users with no appreciable drop in performance.

Conclusions

PEIR-VM has enjoyed rapid adoption among our students and researchers. Future revisions will include region of interest annotations, live tutorial sessions, and updated application programming interfaces for further extensibility.

Evaluation and Actualization of a Pathology Department/Residency Program Website: Developing an Enhanced Educational and Informational Tool

Emilio Madrigal, Mark T. Friedman

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

Content

The potential to be a learning resource as well as an up-to-date information hub creates marked variability in content and quality amongst pathology departments'/residency programs' websites. Although no clear guidelines have been designated to balance this material, we evaluated our previous website (continuumpathology.org) and identified enough deficiencies to warrant its reconstruction altogether (slrbimcpathology.com).

Technology

A well-architectured content management system and blogging tool was deemed necessary to maintain a dynamic website and for this, WordPress, an open source engine, was chosen. A suitable template was identified, and once installed the PHP, HTML and CSS codes were further manipulated. The rich plug-in architecture and multi-user capabilities enhanced customizability, allowing for the development of interactive activities.

Design

Evaluation of our previous website revealed that it was last updated in 2011, rendering its listed schedules, faculty, residents, and contact information outdated and incorrect. Further, the limited educational content was highly disorganized and without any interactivity. A completely indexed environment with a clean user interface was designed, consisting of: a security-encrypted page for current residents with monthly rotation schedules, contact sheets, and educational resources; interactive 'Case of the Month' presentations by residents; daily didactic lecture schedules. Multiple facets of the previous and current websites were evaluated using a newly developed validated tool that assesses the quality of medical education websites in pathology.

Results

The medical educational website quality evaluation tool (MEWQET) was used by two observers within our department. The previous domain yielded MEWQET scores of 36 and 37: 'not recommended' website. The current domain resulted in scores of 73 and 72: 'recommended' website [Table 1]; with complete concordance in these categories: aim, comprehensiveness, navigability, design, interactivity, and disclosures.



Conclusion

In its current form, our website is serving to improve workflow within our program by creating a centralized hub for grossing, frozen-section, and on-call schedules, allowing attendings and residents to know with whom they have been paired with on any given date. Residents have also gained an opportunity to advance their core competencies by creating interactive cases, with educational discussions. Future aims include the integration of: grossing dictation manual/atlas; secured laboratory protocols; and educational case presentations by faculty.

Challenges Developing a Model Digital Pathology Network in a Major Military Healthcare Environment

Anil V. Parwani 3 , Leslie Anthony 1 , Charles R Bond 1 , Jonhan Ho 2 , Liron Pantanowitz 3 , David Glinski 1 , Orly Aridor 1 , Daniel B. Smith 4

1 Office of Sponsored Programs and Research Support, University of Pittsburgh Medical Center, Pittsburgh, PA, 2 Department of Dermatopathology, University of Pittsburgh Medical Center, Pittsburgh, PA, 3 Department of Pathology, University of Pittsburgh School of Medicine, Pittsburgh, PA, 4 Department of Pathology, 81 st Medical Group, Keesler Air Force Base. E-mail: parwaniav@upmc.edu

Content

The United States Air Force Medical Service (AFMS) and University of Pittsburgh Medical Center (UPMC) collaborated on a congressionally-funded initiative focused on the tactical design and implementation of a model digital pathology (DP) network for the AFMS. This presentation highlights unique technology and operational requirements that had to be addressed to form a functional AFMS DP infrastructure.

Technology

Aperio ScanScope AT (400-slide capacity) WSI systems were installed at four regional centers and Aperio ScanScope CS (5-slide capacity) systems at two smaller centers. IT certifications prerequisite to connection of WSI systems to the AFMS network were required, known as Department of Defense (DoD) Information Assurance Certification and Accreditation Process (DIACAP).

Design

Close collaborations and shared decision making between UPMC researchers, AFMS pathologists, and military information technology (IT) leadership were conducted throughout project phases. AFMS pathology needs and workflow were assessed, using the contextual inquiry method, to identify clinical application targets. Commercial WSI systems were critically evaluated and a single system that met AFMS needs was selected for the model AFMS network. UPMC IT coordinated deployment of WSI scanners at various military bases. Research studies were designed for targeted clinical applications, including consultations, quality assurance, and education, to support DP adoption and demonstrate value. Unique technologies, processes, regulatory and people challenges associated with formation of this DP infrastructure were critically reviewed and documented.

Results

The recently established WSI network, is the largest and most geographically-dispersed WSI-based DP network within the DoD. DIACAP and other information assurance certifications proved to be a formidable, lengthy process; which resulted in waning pathology interest and suspension of planned research activities. Early use of technology by pathologists varied across sites. A legacy Laboratory Information System loomed as a challenge to integration with barcoding and workflow management software. Constant movement of personnel, inherent in the military, disrupted AFMS commitment to clinical process change. The autonomous nature of AFMS base operations further jeopardized AFMS-wide DP adoption. Grant limited time and funding placed post-project network sustainment and expansion to all AFMS pathology labs at high risk.

Conclusions

The emerging DP network was successfully established and is ready to be leveraged to optimize collaborative approaches within AFMS pathology and to serve as a model for other branches of the military. While DP implementation obstacles are not exclusive to a military healthcare environment, DoD IT requirements pose distinct challenges. Dedicated management and leadership commitment is paramount to the stability, sustainability, growth, and maturity of this model DP military network.

Pros and Cons of Using Digital Pathology for Genitourinary Tumor Board

Somak Roy, Jon Duboy, Ishtiaque Ahmed, Rajiv Dhir, Anil V. Parwani, Liron Pantanowitz

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

Content

Multidisciplinary clinicopathologic conferences are a key component in clinical practice, which allows physicians to determine consensus based treatment options for their patients. Pathology and radiology data contribute significantly to the discussion. Unlike radiology, pathology data has been traditionally presented using glass slides and/or static images in slideshow format. This requires substantial preparation time for the pathologist. We aimed to determine the benefits and difficulties using digital technology to present pathology at our institution's genitourinary tumor boards.

Technology

Whole slide images (WSI) were prepared using an Aperio ScanScope XT scanner (Leica Biosystems). Digitized slides stored on an image server were accessed via our institution's secure wireless network and viewed on an Apple iPad3 (5 th Gen, late 2013) using the Aperio ePathViewer. VGA adapters and switchers facilitated large screen projection.

Design

Weekly tumor board preparation involved glass slide retrieval and WSI scanning of selected slides. Digital slides were previewed for quality assurance.

Results

The new digital workflow has been used for 12 tumor boards so far involving 45 cases with 3 to 8 slides per case. Digitizing slides improved the efficiency of conference preparation for pathology. Total preparation time decreased from 2-3 hours/week to 1 hour/week. WSI presentations were received favorably by clinicians in the audience. The feature they most appreciated was shared real-time overview of entire slides. Margin status, tumor in relation to adjacent anatomical structures, and pattern of infiltration were better appreciated. WSI could also be later shared with the clinical team whenever desired. Drawbacks included the inability to annotate WSI with the ePathViewer, the need to toggle between the iPad and other devices (laptop computer) to display radiology images, text and gross pathology images, as well as difficulties accessing WSI during network outages.

Conclusion

The benefits of using digital pathology for genitourinary tumor boards outweighed the difficulties encountered employing this technology. Viewing WSI on an iPad was perceived to be cutting-edge, engaging and more informative for clinicians attending these conferences. Our aim is to further enhance this experience by more seamlessly integrating digitized slides with accompanying text and static gross images.

Utilizing Raspberry PI to Provide Educational Material for Medical Students in Zambia

Stephanie Simmons 1 , Peter Anderson 2 , Alex Feldman 1 , Jeremie Lever 3 , Seung Park 4

1
Department of Pathology, University of Alabama at Birmingham, 2 Division of Molecular and Cellular Pathology, University of Alabama at Birmingham, 3 School of Medicine, University of Alabama at Birmingham, 4 Department of Pathology, Division of Informatics, University of Alabama at Birmingham. E-mail: sdsimmons@uabmc.edu

Content

Sub-Saharan Africa is one of the most medically underserved regions in the world. Medical schools are in short supply of faculty, educational resources, and technology. Though the use of information technology has been proposed to mitigate these shortfalls, the main obstacle is lack of quality internet service, which makes online access to educational materials difficult or even impossible.

Design

The Pathology Education Instructional Resource (PEIR) (http://peir.path.uab.edu) was developed at the University of Alabama, Birmingham and been an instrumental component in first-year medical student pathology education. PEIR was recently updated to include a repository of web-viewable whole slide images (WSI) (http://peir-vm.path.uab.edu). Our goal was to take PEIR to Zambia in a way that could be easily accessible, even in regions with poor connectivity, by turning the Raspberry Pi, an inexpensive credit card-sized single-board computer, into a combination server and local Wi-Fi router that could provide access to PEIR within the medical school without the need of an internet connection.

Technology

Hardware: Raspberry Pi Model B, EdiMax USB Wi-Fi dongle and Western Digital 500GB USB HDD; Operating System: Raspbian Linux; Web Server: nginx 1.5; Database: MySQL 5.5; Programming Language: PHP-FPM 5.3; Web-Based WSI: OpenSlide 3.4.0, OpenSeadragon 1.0.0; Wireless Routing: DNSMasq 2.68

Results

The Raspberry Pi system project, code-named "Raspberry PEIR", was completed in less than one month with a cost of approximately $90 for the hardware. It includes a collection of over 30,000 histopathology images organized by topic with full keyword-based searching, over 300 whole slide images organized by topic, and an online pathology curriculum "in a box" that directly utilizes the images and WSIs. Raspberry PEIR also serves as an independent Wi-Fi router that directly serves PEIR content to connected wireless devices (e.g. smartphones and tablets), whether or not a stable connection to the Internet at large exists.

Conclusions

The use of cheap commodity computing in medical education in developing countries is a cost-effective and sustainable approach. To the best of our knowledge, this is the first usage of a Raspberry Pi in medical education in Sub-Saharan Africa. We will continue to add functionality with future versions.

Evaluation of the SPOT PathStation2 for Gross Telepathology

Muhammad Syed, Ishtiaque Ahmed, Anil V. Parwani, Ralph Anderson, Kara Balatincz, Douglas Hartman, Somak Roy, Liron Pantanowitz

Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA. E-mail: syedma2@upmc.edu

Content

Gross pathology evaluation is an important component of intraoperative pathology consultations and subsequent triaging of specimens by pathology staff to document findings, appropriately sample tissue for histological evaluation and ancillary testing or tissue banking, and to help render final diagnoses. Currently there are limited digital imaging systems that facilitate gross telepathology. Our aim was to evaluate the SPOT PathStation2 system for performing gross telepathology.

Technology

SPOT Imaging Solutions TM PathStation2 (SPOT Imaging Solutions, a division of Diagnostic Instruments, Inc. Sterling Heights, MI, USA). The imaging system includes a 1080p/30 frames per second compact, high definition digital camera with 16x optical zoom lens, 18" 1920 × 1080 resolution TFT touch screen All-In-One Computer and PathSuite2 software. TeamViewer (version 8) was employed to view live image feeds on remote desktops. The PathStation2 can be configured for mounting on a wall, stand, or inside a grossing hood. The configuration we evaluated was the stand version that included an adjustable height copy stand with high intensity 5000 K color balanced lighting.

Design

A PathStation2 desktop instrument was installed in the UPMC Shadyside hospital gross pathology laboratory (see figure) and configured on our institution's secure network. Various pathology specimens ranging from small needle core biopsies to large resections (e.g. kidney, prostate, and pancreas) were selected. Inked specimens were included to grossly evaluate margins. Three surgical pathologists remotely evaluated the live image quality of specimens by logging on to the PathStation2 workstation from their respective offices using TeamViewer. Remote acquisition of snapshots and the use of annotation tools to guide the prosector in the grossing room were also assessed.

Results

The PathStation2 provided high quality images of all specimens during live gross telepathology [Figure 1]. Specimen details and inked margins were easy to examine at low fields of view and when zooming in close to inspect specific areas. "Sharpness" and "depth of focus" setting tools facilitated viewing of specimens at low and high magnifications, respectively. The TeamViewer banner on remote workstations obscured some of the PathSuite2 icons available for annotation.



Conclusion

The SPOT PathStation2 gross imaging system incorporates excellent hardware components (high resolution digital camera with 16x optical zoom) and essential software features (annotation tools) that permit real-time gross telepathology for clinical use. Although TeamViewer provided adequate remote collaboration, future versions of the SPOT PathSuite software will improve the interactivity of the system for the remote pathologist.

VisionTek ® Live Robotic Digital Telecytology Validation at UPMC

Muhammad Syed, Anil V. Parwani, Walid Khalbuss, Payam Arya, Ishtiaque Ahmed, Ioan Cucoranu, Jackie Cuda, Liron Pantanowitz

Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA. E-mail: syedma2@upmc.edu

Content

Telecytology for rapid on-site evaluation (ROSE) is being increasingly used in the practice of pathology. To date, real-time telecytology has been accomplished largely by means of streaming live video images. Very few studies have employed robotic telemicroscopy for remote cytologic evaluation. New technology, such as the VisionTek ® digital microscope, has emerged on the market where whole slide image (WSI) scanners now incorporate robotic microscopy capabilities. These hybrid instruments can therefore be used for static whole slide imaging and/or live robotic telepathology by permitting remote control of their microscope. The aim of this study was to validate the utility of VisionTek ® for telecytology.

Technology

A VisionTek ® Digital Microscope (Sakura, Finetrek, USA) desktop instrument was installed at University of Pittsburgh Medical Center Shadyside hospital in Pittsburgh, PA, USA. This brightfield system permitted live robotic review of up to 4 glass slides. For remote viewing of digital images TeamViewer (version 8.0.18051) was used.

Design

Fifty (50) archival de-identified cytology glass slides of fine needle aspiration (FNA) direct smears (Diff-Quik stained) were selected. These cases represented a broad range of diagnoses (benign and malignant) and complexity (common and difficult diagnoses). Three board certified cytopathologists each evaluated these cases. Digital images using VisionTek® were remotely reviewed using similar monitors. Glass slides of all cases were examined 2 weeks later using a conventional light microscope. The participants were provided with a brief clinical history and site of the FNA for all cases. Pathologists recorded their diagnoses, time to reach a diagnosis, and any technical errors encountered. Discordant diagnoses were categorized into major (likely to impact patient care) or minor (unlikely to impact patient care) discrepancies.

Results

[Table 1] summarizes intraobserver study findings. Intraobserver concordance between digital and glass slides was high (average 95%) when cytopathologists had to determine if a cytology FNA contained adequate material. However, the intraobserver concordance when rendering a specific cytologic diagnosis was lower (average 84%). Cases in which there was diagnostic discordance were of major impact in 7% and minor clinical significance in 9% of cases. Technical errors were reported in 23 instances, and were related to failure of the microscope to autofocus or to interruptions with the TeamViewer application. Cases hampered by technical difficulty caused a delay in reaching a final diagnosis.



Conclusion

Employing the VisionTek ® Digital Microscope's real-time robotic feature for rapid telecytology evaluation was reliable for determining specimen adequacy in FNA specimens and for providing preliminary diagnoses of most cases in a timely manner. Discrepancies were attributed mainly to misinterpretation of digital images. Remote viewing of digital images was problematic in a few cases due to hardware or third party application software technical errors.

An Automated Assessment of Tumor Viability in Digitized Whole-Slide Sections of a Lung Cancer Mouse Model

Riku Turkki 1 , Nina Linder 1 , Yinhai Wang 1 , Tanja Holopainen 2 , Anne Grote 1 , Mikael Lundin 1 , Kari Alitalo 2,3 , Johan Lundin 1

1
Institute for Molecular Medicine (FIMM), University of Helsinki, Finland, 2 Translational Cancer Biology Program, Biomedicum Helsinki, University of Helsinki, Finland, 3 Wihuri Research Institute Biomedicum, University of Helsinki, Finland. E-mail: riku.turkki@fimm.fi

Content

In histological analysis tumor viability is measured as a fraction of viable tissue region within the whole tumor. Tumor viability is commonly assessed in pre-clinical research, especially in drug screening settings to measure and compare the treatment responses of antitumor therapies. Nonetheless, due to the challenging nature of visual examination, the quantification of tumor viability is problematic and methods suited for high-throughput analysis are needed. Visual examination is associated with intra- and inter observer variability and is not easily scaled up for analysis of larger samples sets.

In this study, we developed an automated segmentation for detecting the tissue subtypes required for the viability assessment. The approach is validated on sections of human non-small cell lung cancer xenograft tumors treated with antiangiogenic agents.

Technology

The tumor sections were digitally scanned with a whole-slide scanner and archived in a digital content management environment. A multi-class support vector machine classifier combined with rotation invariant texture descriptors and feature mapping, was used to construct a supervised segmentation tool to identify tissue subtypes. Tumor viability score was then derived from the ratio between recognized necrotic tissue area and whole tumor area.

Design

Altogether, 56 hematoxylin-eosin stained tumor sections were used in this study. Two experienced researchers annotated the digitized sections into 3 categories, i.e. necrotic tumor, viable tumor and other tissue types (e.g. including normal host tissue, stroma and muscle tissue). The annotations were used for training the algorithm (4 sections) and for testing (52 sections).

Results

The method achieved a median accuracy of F-score = 0.92 to assess the tumor viability when tested on 52 whole tumor sections. The correlation between the human experts and the automated method to measure the viability ratio was r = 0.79 (CI95% 0.65-0.87; P < 0.0001).

Conclusions

In this study, a novel automated tumor viability assessment was presented. Preliminary results indicated that the proposed method offer objective and reliable tumor viability assessment in xenograft samples when compared to human observers. Therefore, the method has the potential to further benefit pre-clinical xenograft studies and similar research settings where there is a need to quantify tumor viability.

Automated Grading of Renal Cell Carcinoma Using Whole Slide Imaging

Fang-Cheng Yeh 1 , Anil V. Parwani 2 , Liron Pantanowitz 2 , Chien Ho 3

1
Carnegie Mellon University, Biomedical Engineering, Pittsburgh, PA, 2 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, 3 Department of Biological Science, Pittsburgh NMR Center for Biomedical Research, Carnegie Mellon, University, Pittsburgh, PA. E-mail: frank.yeh@gmail.com

Content

Grading of clear cell renal cell carcinoma using Fuhrman nuclear grade for prognostic evaluation is a manual task traditionally performed by pathologists. Recent technology developments have demonstrated the benefit of applying image analysis tools to whole slide imaging (WSI). The aim of this study was to determine the feasibility of using automated image analysis to grade clear cell RCC.

Technology

Automatic analysis was applied to 39 H and E-stained digitized slides of clear cell RCC with varying grades that calculated nuclear size across the entire slide. The analysis results were correlated with manual Fuhrman nuclear grades.

Design

A total of 39 de-identified H and E-stained glass slides of clear cell RCC were selected from the archives at the University of Pittsburgh Medical Center following Institutional Review Board approval. The stain quality, diagnosis, and Fuhrman nuclear grade of RCC were determined by an experienced genitourinary surgical pathologist (AVP).A total of 3, 11, 12, and 13 cases were selected for grades I, II, III, and IV, respectively. We selected only 3 cases for grade I because of its rarity (5~10%) among clear cell RCC cases.

Results

The spatial distribution of nuclear size provided a panoramic view of the tissue sections that can facilitate locating high-grade tumor regions and areas with necrosis. The statistical analysis showed that maximum nuclear size was significantly different (P value < 0.001) between low-grade (grades I and II) and high-grade tumors (grades III and IV). The receiver operating characteristics (ROC) analysis showed that the maximum nuclear size distinguished high-grade and low-grade tumors with a false positive rate of 0.2 and a true positive rate of 1.0. The area under the curve is 0.97143.

Conclusion

Image analysis algorithms that employ nuclear size can be used to differentiate low-grade and high-grade clear cell RCC with good sensitivity and specificity. These data suggest that automatic WSI analysis of clear cell RCC may facilitate pathologic grading of renal tumors and reduce variability encountered with manual grading.


   Short Abstract Presentations Top


Locations: Wyndham Grand Pittsburgh Hotel, Kings Garden South/LeBateau, Kings Garden North

Imaging Informatics

Thursday, May 15, 2014


Grand Ballroom 1

HemaVue: A Web-Based Tool for Clinician Access to Digital Peripheral Blood Smears

William Cable 1 , Liron Pantanowitz 2 , Denine Maglicco 1 , Ryan Mitchell 1 , Jeff McHugh 1 , Anil V. Parwani 2 , Lydia Contis 2

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

Content

The peripheral blood smear (PBS) is an important component of a patient's clinical evaluation and provides a means to morphologically evaluate blood cells. The traditional approach has been to review blood smears in the hematology laboratory, which may not easily accessible. With advances in technology, the PBS can now be digitized and analyzed. The aim of our project was to develop a web-based tool to allow clinicians, residents and fellows to view PBS scans digitized by CellaVision ® . The tool facilitates the teaching of residents and fellows without the need for a microscope.

Technology

CellaVision DM96 analyzer, Windows Server 2008, MS Active Directory, Internet Information Services (IIS) Manager, Coldfusion 9, SQL Server 11.0.3128, IBM ® Scale Out Network Attached Storage (SONAS).

Design

The CellaVision ® analyzer scans blood smear glass slides and automatically identifies and classifies white blood cells. Images of these blood cells are stored in the CellaVision database (Microsoft Access) as binary data. A technologist creates an archive file with this data, which is moved to our SONAS. A scheduled job employs SQL Server Integration Services to convert the Access archive file and insert it into a SQL Server database. A Coldfusion web application was written to interface with the SQL database, query against the patient medical record number, generate the associated JPG image files of scanned blood cells, and display a grid of thumbnail images.

Results

CellaVision ® data including all digital images are securely archived. The "HemaVue" tool we developed permits CellaVision ® scan results to be viewed by authorized personnel on any terminal in the institution's network. Users are presented with an organized display of the digital PBS. Access to this tool is restricted to authenticated users via Active Directory and IIS. Clinician and trainee experience with this tool has been very positive and is widely and readily accepted as a teaching tool.

Conclusion

The HemaVue web-based tool has successfully allowed clinicians and trainees at our academic institution to remotely access digitized PBS generated by the CellaVision ® instrument. Future efforts are aimed at incorporating these digital blood smears into the electronic medical record for improved access to patient data.

Pilot Reader Studies Comparing Whole Slide Images with Different Gamma Settings

Brandon D. Gallas 1 , Marios Gavrielides 1 , Adam Wunderlich 1 , Jeffrey D. Seidman 2 , Jason Hipp 3 , Stephen M. Hewitt 3

1
US Food and Drug Administration, Division of Imaging and Applied Mathematics, OSEL/CDRH, 2 Division of Immunology and Hematology Devices, OIR/CDRH, Silver Spring, 3 National Institutes of Health, Laboratory of Pathology, Center for Cancer Research, National Cancer Institute, Bethesda. E-mail: brandon.gallas@fda.hhs.gov

Content

We report on reader studies comparing whole slide images with two different gamma settings in a quantitative assessment that removes sources of measurement uncertainty. We use an evaluation environment called eeDAP that allows cell-to-cell correspondence between digital and analog pathology. The pre-compiled, stand-alone, license-free, software component of eeDAP is described below and is freely available from http://code.google.com/p/eedap/.

Technology

We used a Hamamatsu Nanozoomer 2.0HT to scan H and E stained slides at 40x. eeDAP is a Matlab application that registers glass slides to corresponding whole slide images and collects reader study data. The registration uses a microscope-mounted camera, motorized stage, and software that is minimally guided by the user. The reader study collects quantitative evaluations of histopathological features; it runs in digital mode (showing ROIs from whole slide images) or microscope mode (stage automatically navigates to ROI locations on the slide). Using a reticle and corresponding digital overlay, eeDAP isolates the same individual cells for evaluation by pathologists in both modes. This registration allows the reduction or elimination of a large source of variability in comparing these modalities in the hands of the pathologist: The field of view (the tissue) being evaluated.

Design

We have executed two studies: (1) Readers had to decide if a cell was a mitotic figure or not (and their confidence) in H and E stained sarcoma tissue; (2) the readers had to decide if a cell was a plasma cell or not (and their confidence) in H and E stained colon tissue. For each study, five readers evaluated 50 cells in two digital modes (gamma = 1.0 and 1.8). The reference for digital-mode performance was a pathologist evaluating the same cells in microscope mode.

Results

The gamma setting did not appear to affect the digital-to-microscope agreement. Additionally, the intra-reader agreement across the digital modes was higher than the digital-to-microscope agreement, indicating that reader, not mode, is the dominant effect.

Conclusions

This was a pilot study for our evaluation environment (eeDAP), workflow, and analysis methodologies. We will use these results to design a pivotal study to test a specific hypothesis at a target statistical power.

TelePath: Commodity Telepathology for the Web 2.0 Age

Timothy Kennell Jr 1 , Kelly Taylor 2 , Samuel Borak 2 , Walter Bell 2 , Israel Ponce-Rodriguez 3 , Timothy Awtrey 3 , Seung Park 4

1
NIH Medical Scientist Training Program, University of Alabama at Birmingham School of Medicine, 2 Departments of Pathology, Community Practice Pathology Program, University of Alabama at Birmingham, 3 Division of Information Services, University of Alabama at Birmingham, 4 Division of Informatics, University of Alabama at Birmingham. E-mail: tikenn@uab.edu

Content

Telepathology remains underutilized despite its decades-long existence. Common roadblocks to adoption include high startup cost, unsatisfactory user interfaces, and unpredictable video latency. At our institution, we have created a telepathology system from commodity components that has proven to be cost-effective, user-friendly, and highly-utilized.

Technology

Microscope Camera: Lumenera Infinity HD; Digital Signal Processor: AVerMedia Live Gamer HD; Server Hardware: Dell PowerEdge 2590; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 12.04 LTS 64-bit; Web Server: nginx 1.5; Real Time Messaging Protocol Server: nginx-rtmp-module 1.1; Database Management System: MariaDB 5.5; Programming Languages: PHP-FPM 5.3 and Javascript; User Interface Library: Twitter Bootstrap 3.3; Client Video Playback: JWPlayer 6.8.

Design

Infinity HD cameras are installed on pathologist microscopes; these stream video via HDMI to the Live Gamer HD, which H.264 encodes the video feed in real-time. The feed is automatically routed to an nginx + nginx-rtmp-module server for user viewing [Figure 1]. The originating pathologist needs to only press a single hardware button to start the live stream. The consultant pathologist needs only to visit a website to view the stream. No additional hardware or software installation is required.



Results

TelePath is universally commended by its users for its ease of operation, and adds minimal cost ($150 for the Live Gamer HD; $3000 for the camera, but any existing microscope camera with video output can be used). 100% of consultant pathologists consider the video stream to be of adequate quality for diagnostic consultation. Latency ranges from 1-3 seconds, but is a constant (thereby predictable) latency per session that can easily be adjusted for. Originating pathologists utilize TelePath 1-5 times per week, each time avoiding the 24-48 hour turnaround time delay associated with delivering a case via courier to the consultants.

Conclusions

TelePath is an integral part of our pathology practice (most notably our community outreach program), and is available at every site we serve. By using commodity components and programming user-friendly interfaces, we have seen widespread adoption of telepathology as a consultation medium in our practice. Future versions will add robotic remote control capabilities and increased quality of service.

Framework for Data Management and Visualization of The National Lung Screening Trial Pathology Images

Ashish Sharma 1 , Ameen Kazerouni 1 , Yusuf Nadir Saghar 1 , Paul Commean 2 , Lawrence Tarbox 2 , Fred Prior 2

1
Department of Biomedical Informatics, Emory University, Atlanta, GA, 2 Mallinckrodt Institute of Radiology, Washington University, St. Louis, MO. E-mail: ashish.sharma@emory.edu

Content

This abstract reports on the use of caMicroscope to provide web-based access to digitized pathology images from the National Lung Screening Trial. caMicroscope is a web based data management and visualization system for digitized whole slide images.

Technology

caMicroscope follows the Software as a Service delivery model. It consists of four primary modules.

Data Loader: Utilizes the OpenSlide and VIPS open source libraries to perform data preparation. This includes, (1) Identification of images to process; (2) Conversion of the images from its vendor specific format to the open standard BigTIFF; (3) The extraction of image meta-data that is necessary for visualization of images and markups.

Image Viewer: A HTML5 based viewer that leverages OpenSeadragon and IIPServer for the visualization of the pathology images.

Markup Tools: Provide the user the ability to draw, store and retrieve annotations and markups.

Data Services: A collection of web data services that provide access to databases that manage images, metadata and markup data via secure REST API's.

Design

caMicroscope deployed for the National Lung Screening Trial, required a semi-automated data preparation process, that included the retrieval of anonymized identifiers from the image labels, and reconciliation of these labels with global identifiers. The Cancer Imaging Archive's Query Tool, allows investigators to access and query repositories of Clinical, Pathology and Radiology metadata. It is then used to launch the caMicroscope image viewer to visualize selected subjects. User generated markups are stored along with image identifiers, and can be restored on launch. The deployment facilitates, (1) Visualization of images; (2) Efficient sharing of images; (3) Rendering and sharing of Annotations and Markups.

Results

Performance tests have been performed to validate the scalability and reliability of caMicroscope. The system, shown in [Figure 1], has been live for 9 months. It provides access to 1253 pathology images to authorized users.

Conclusion

caMicroscope is a generic digital pathology data management and visualization system that was used to provide access to images in this project. It integrates with external metadata exploration clients such as the Query Tool and gives users the ability to visualize and annotate images.

Imaging Informatics

Thursday, May 15, 2014

Kings Garden South/LeBateau

Web-Based Pathology Practice Examination Usage


Edward C. Klatt

Mercer University School of Medicine, Biomedical Sciences, Savannah, GA. E-mail: klatt_ec@mercer.edu

Content

General, organ, and subject specific pathology practice examinations were placed onto a public access interent web site. Users had to make 4 clicks from the home web page at: (http://library.med.utah.edu/WebPath/EXAM/EXAMIDX.html).

Technology

Multiple choice questions were developed in United Stated Medical Licensing Examination style and coded into.html files with javascript functions that presented a common interface for web browser viewing in a timed format. A PERL programming script with common gateway interface for web page forms scored examinations and placed results for each question and total score per accession into a log file on an internet server.

Design

The 4 general review examinations of 30 questions each could be completed in up to 30 minutes. The 17 organ and subject specific examinations of 10 questions each, with accompanying gross or microscopic images, could be completed in up to 15 minutes each. The results of log files were compiled from June 2006 through January 2014.

Results

The 4 general review examinations had 31,639 accesses with completion of all questions, for an overall completion rate of 54% and average score of 75%. A score of 100% was achieved in 7% of accesses, 21% of users achieved a score ≥90%, and 95% of users achieved ≥50%. In top to bottom web page menu order, review examination usage was 44%, 24%, 17%, and 15% of all accessions. The 17 organ and subject specific examinations had 103,028 completions, with completion rate 73% and average score 74%. Mastery at 100% was 19.6% overall, 36.8% of users achieved a score ≥90%, and 90% achieved ≥50%. The first 3 menu items on the web page accounted for 12.6%, 10.0%, and 8.2% of all completions, and the bottom 3 accounted for no more than 2% each.

Conclusion

Completion rates were higher for shorter examinations of 10 questions in a single subject area. Usage was higher for examinations at the top of the web page menu. Though internet user attention span appeared limited based upon completion rates, scores achieved suggest that serious users completing the examinations had sufficient preparation to make use of the examinations to support their pathology education.

Defining the Next Generation of Digital Pathology in Swedish Clinical Routine

Claes Lundström

Linköping University and Sectra AB, Center for Medical Image Science and Visualization, Linköping, Sweden. E-mail: claes.lundstrom@liu.se

Content

In 2012 a research and clinical development project regarding Whole-Slide Imaging was launched, involving care providers covering almost half of Sweden as well as academic and industrial actors. We here present the results of the design phase, where the objective was to design optimal work practices for a fully digital environment and to establish all benefits of digital pathology in clinical routine. A notable foundation for this effort was the extensive experience of digital primary review practice in Kalmar and Linköping, where over 550,000 slides have been scanned to date.

Technology

Experiences from various digital pathology software and equipment have been one type of input to the design work.

Design

The analysis and design of work practices were carried out through a series of workshops. Input was gathered from experts in clinical pathology (pathologists, managers, technicians), process improvement and imaging informatics. The designs were documented and subsequently reviewed. Information with bearing on supportive IT tools was used as a requirement specification for prototype development including a pathologist's workstation.

Results

The designed workflows and prototypes point to many benefits of the digitization, some perhaps being less established. For example, fully digitizing the grossing documentation and deeply integrating it with the slide viewer is of significant diagnostic value. The enterprise scope is also underlined. Clinical value beyond MDT meetings arise when pathology is one part of an enterprise image management setup, and there are cost synergies with existing IT infrastructures for radiology imaging. Nevertheless, a major hurdle for adoption is the cost of long-term data storage and the need for smart information lifecycle management is highlighted. Finally, nation-wide collaboration in the form of a centralized image repository for primary review has been investigated and found to be technically feasible.

Conclusion

A strong knowledge base has been established to support continued adoption of digital pathology in Sweden. There is convincing rationale for this development, but a broad scope is necessary to meet all challenges and exploit all opportunities.

High Resolution Virtual Histology of Uncut, Unembedded Tissue

Richard Torres 1 , Eben Olson 1 , Michael J. Levene 2

Yale University School of Medicine, 1 Laboratory Medicine, 2 Biomedical Engineering New Haven, CT. E-mail: richard.torres@yale.edu

Content

Incomplete sample evaluation, artifacts of preparation, poorly quantitative measures, limited growth pattern information, and an extended manual preparative process are some of the aspects of traditional slide-based histologic analysis of human samples that limit further advancements in histology. With the increasing use of ancillary tests, these limitations are particularly relevant for small core biopsies. In an effort to expand the capabilities of tissue analysis, we report on the potential of a modified tissue processing method that can be used to image non-destructively through entire core biopsy specimens, eliminates the need for wax-embedding, and avoids cutting artifacts, while preserving tissue for ancillary studies.

Technology

The technique of multiphoton (MP) microscopy was combined with a benzyl alcohol benzyl benzoate (BABB) tissue-clearing protocol for deep tissue imaging. The microscope is based on a Ti: Sapphire pulsed laser, high numerical aperture objectives, and multichannel detection with Scanimage (Janelia Farm, Ashburn, VA) control software. Non-fluorescent second harmonic generation signal, corresponding to collagen fibers, was collected concurrently.

Design

Neoplastic and non-neoplastic samples of formalin-fixed human lung, breast, and kidney were obtained from discarded pathologic tissue specimens after resection. Core biopsy-sized fragments were dehydrated in alcohol before being cleared in BABB. Fluorescent dyes were used for staining. Specimens were imaged overnight. Fluorescence images were transformed into H and E-like images via a custom script that inverts the logarithmic matrix conversion of H and E color values. A subset of samples were subsequently processed using traditional techniques for comparison.

Results

Excellent image quality was obtained on all tissue types obtained using multiphoton microscopy and BABB clearing at depths greater than 1mm, sufficient for complete core biopsy evaluation. Faithful H and E-like reproduction was achieved. Second harmonic signals offered quantifiable fibrosis measures throughout the samples. There were no appreciable degradation effects or artifacts on subsequent wax-embedding and staining.

Conclusions

The combination of multiphoton imaging and tissue clearing shows great promise as a practical tool for histology that offers significant advantages over traditional techniques.

International Telepathology Consultation: Two Years of Experience Between the University of Pittsburgh Medical Center and Kingmed (China)

Chengquan Zhao 1 , Tao Wu 2 , Xiangdong Ding 2 , Anil V Parwani 1 , Hualin Chen 2 , Gonzalo Romero Lauro 1 , Xiaodong Feng 2 , Shangwei Wu 2 , Samuel Yousem 1 , Liron Pantanowitz 1,

1 Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, 2 Department of Surgical Pathology, KingMed Diagnostics, Guangzhou, Guangdong, China. E-mail: zhaoc@upmc.edu

Content

Telepathology is increasingly being employed to support diagnostic consultation services. Most telepathology publications have dealt more with the technology aspects than clinical experience of this practice. KingMed Diagnostics is a large network with 21 central laboratories serving 9,000 hospitals and clinics in China. Since 2012, the University of Pittsburgh Medical Center (UPMC) and KingMed have established an international telepathology consultation service. The aim of this study was to review the experience of this international telepathology partnership.

Technology

Whole slide images acquired using a NanoZoomer 2.0-HT were stored on a server in China using the NDP.serve database management system. Digital slides were accessed via a portal server in Pittsburgh, PA, USA. Using a customized portal and associated viewer software, UPMC pathologists were notified of incoming digital consultations, viewed digital slides and generated diagnostic reports.

Design

This is a retrospective study that summarizes telepathology consultation results between UPMC and KingMed over a 2 year period (January 2012 - December 2013).

Results

Most (59%) submitted cases were referred by Chinese pathologists, 38% per request of clinicians, and 3% by patients in China. There were 144 cases sent in 2012 and 614 in 2013 (P < 0.05). Case volume increased in the past 9 months, with an average of 61 cases per month. A summary of case distribution and turnaround time is listed in [Table 1]. Hematopathology received the most cases (25.3%), followed by bone/soft tissue (20.6%) and gynecologic/breast (18.6%) subspecialties. Average turnaround time was 6.8 days in 2012 and 5.3 days in 2013 (P < 0.05). For many cases immunostains were ordered. For certain cases, especially hematopathology, more than one round of immunostains was needed which delayed turnaround time. For 360 (47.5%) cases digital slides were sent for consultation without a primary diagnosis, whereas in 398 (52.5%) cases a primary diagnosis or impression was provided by the referring local hospital in China. Malignancies accounted for 66.4% of cases, benign entities for 19.9%, and borderline neoplasms for 13.7% of consults. In most (84.8%) cases there was a definite diagnosis provided, 9.4% a favored/suggestive diagnosis, and 5.8% an atypical/uncertain diagnosis. For the 398 cases where a primary diagnosis/impression was included, the final diagnoses rendered by UPMC was identical in 22.9% of cases, mildly modified in 23.4%, and significantly modified (treatment plan altered) in 53.8% of cases.



Conclusion

These results indicate that international telepathology consultation can improve patient care by facilitating access to pathology expertise. The success of this international digital consultation service was dependent on strong commitment and support from leadership, information technology expertise, and dedicated pathologists who understood the language and culture on both sides. Challenges included internet speed and firewalls, difference in cultures and health care systems, lack of clinical information and gross pathology descriptions with cases, insufficient tissue sections submitted for evaluation, and difficulties related to direct communication with Chinese clinicians.

Applied Pathology Informatics

Thursday, May 15, 2014


Kings Garden North

Automated Pathology Workflow for the 21 st Century

Navid Farahani1, Jeffrey L Fine 2

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

Content

Many whole slide image (WSI) systems claim to be highly automated, but largely emulate and maintain traditional surgical pathology workflow. Pathologists read laboratory information system (LIS) data and review slides manually, one at a time and in order, and then manually construct a report. A previously outlined "21 st century workflow proposal" presented a hypothetical, automated sign-out workflow. This project is a prototype mock-up of an advanced sign-out of a breast cancer specimen, intended to provide a concrete example of automated sign-out that can be used to guide and facilitate WSI adoption.

Technology

Slides and reports from two breast cancer cases were de-identified: A core biopsy and a lumpectomy with axillary sentinel lymph node biopsy. WSIs were created using an Aperio ScanScope XT, and screen captures were created using vendor-provided software (Leica, Vista CA USA). The advanced workflow prototype was constructed using PowerPoint software (Microsoft, Redmond WA USA).

Design

We modeled an interactive, computer-assisted workflow: The hypothetical computer previews the WSIs in the context of integrated LIS data (e.g. specimen type, gross description) and pre-defined diagnostic tasks. These tasks include: Correlation with previous tumor diagnosis; tumor grading and staging; margins assessment; etc. Relevant regions of interest (ROIs) are automatically identified and triaged by the computer. The pathologist's sign-out work consists of an interactive review of the ROIs that is driven by the diagnostic tasks. The results of this are then utilized to automatically construct a final report.

Results

The created mock-up models a possible automated workflow for a breast cancer resection specimen [Figure 1]. It is an animated slide show that presents the hypothetical sign-out stepwise, illustrating the various interactions and explaining how these steps can be automated (publicly available at http://pitt.edu/~jlf60/supplemental.html).



Conclusion

Our prototype demonstrates a radical rethinking of pathology workflow that focuses pathologists on work that only they can do. Partly based on hypothetical intelligent computer systems, it also relies on existing, under-utilized LIS and WSI data. There are precedents: imaged Pap testing and automated hematology are two microscopy-based tests that are now automated. We have presented a concrete step that may help digital pathology provide desperately needed and long-awaited automation for pathologists.

Interfacing Automated Immunostainers with an Anatomic Pathology Laboratory Information System

John Sinard M, Peter Gershkovich

Department of Pathology, Yale Medical School, Yale University School of Medicine, New Haven, CT. E-mail: john.sinard@yale.edu

Content

Integration of laboratory information systems (LIS) with automated equipment has only recently begun to occur in anatomic pathology laboratories. While many of the newer automated immunostainers have the ability to receive and send electronic information, few laboratories have taken full advantage of this capability. This has been due to both the lack of appropriate functionality in many LISs and to the lack of standardization in equipment requirements for integration. Integration is even more complicated in laboratories that maintain immunostainers from more than one vendor.

Technology

We have integrated our Histology Asset Tracking (HAT) software with the Leica Bond III and the Ventana Benchmark Ultra automated immunostainers. Messages outbound from the HAT system to the stainers were programmed into HAT. An additional standalone interface engine, dubbed Hermes, was written to receive and process messages from the stainers. All software was written using the Java programming language and Open Source frameworks (Apache Mina, Apache Commons, etc.). Our LIS is Cerner CoPath Plus.

Design

Stainer protocol codes were loaded into the "Stain/Process" dictionary of our LIS. As barcoded slides are scanned onto a stainer batch, an HL-7 message is sent from the HAT system to the appropriate stainer with the order information. For one vendor's stainer, a relabeling step is required. The stainer is then able to recognize the barcode on the slide, perform the appropriate stain, and inform the Hermes middleware (via another HL-7 message) when the stain is complete. Hermes updates stain status in the LIS.

Results

Electronic integration of automated immunostainers has greatly facilitated the workflow within the immunohistochemistry lab, improving staff efficiency and reducing errors. The interfaces integrate the instruments into the asset tracking process.

Conclusions

Integration of immunostainers reduces the manual labor or retyping staining orders and reprinting slide labels and, through that, eliminates opportunities for errors. While some degree of user customization is needed to integrate this equipment with anatomic pathology laboratory information systems, the newer stainers are increasingly able to support this integration, and vendors are recognizing the importance of helping customers with this process in order to maintain a competitive edge.

Analysis of the Impact and Value of a Specimen Tracking System Implementation for Anatomic Pathology

Buer Song 1 , Robert Stapp 2 , J. Mark Tuthill 2

1
Department of Pathology and Laboratory Medicine, The State University of New York at Buffalo, Buffalo, NY, 2 Henry Ford Health System, Detroit MI.

Email: buersong@buffalo.edu

Content

Accurate and effective specimen tracking and routing in a surgical pathology laboratory is critical for improving efficiency, reducing error, and improving patient care. Performing specimen tracking and routing in the traditional manner could cost significant time and resources, with minimal to no, real-time workflow monitoring. An automated system, integrated with the laboratory information system, would provide accurate routing, detailed tracking, real-time workflow monitoring, and reduce costs. Here, we analyzed the elements required to set up an automated tracking and routing system at Buffalo General Medical Center. We examined the needs and expectations of a medium-large scale pathology laboratory, established objectives, and planned a system implementation capable of reaching them.

Technology

The proposed system consists of Sunquest CoPath, as the laboratory information system, integrated with specimen management routing and tracking (SMART). Additional hardware requirements would include Windows based workstations, 2-D barcode scanners, slide label and cassette printers/etchers, and network with wireless capabilities.

Design

Specimen volume, workflow, existing problems, and desired improvements in the histology laboratory were analyzed. Required elements of the CoPath+SMART system were listed according to the needs of the laboratory. The system was then virtually run through the entire workflow to explore the potential for process improvements. Tentative solutions for anticipated problems were also proposed. A brief comparison with Vantage from Roche was also performed.

Results

An automated tracking and routing system could be successfully implemented with the necessary hardware and software. Compared to our current laboratory information system, CoPath+SMART could provide more accurate routing, detailed tracking, and valuable workflow monitoring functions. Potential issues include major workflow changes for the entire workforce, protocol compliance issues, and possible technical glitches that might arise.

Conclusion

As the major histology lab for a tertiary regional medical center, the lab would benefit from an automated specimen tracking and routing system. Increased work efficiency, reduced error, and better lab management would be expected. However, it would require major changes in laboratory culture, workforce adjustment, and a well-tailored plan to increase benefit/cost ratio.

Application of an Anatomic Pathology Tracking System in Specimen Management and Beyond

Robert Stapp 1 , Marc Levine 1 , Ron Brown 1 , Mehrvash Haghighi 2 , J. Mark Tuthill 1

1
Henry Ford Hospital, Detroit, Michigan, 2 Columbia University Medical Center, New York, NY. E-mail: rstapp1@hfhs.org

Content

Advances in anatomic pathology laboratory information systems (AP-LIS) have allowed us to greatly enhance and streamline our workflow processes. Historically laborious processes, have now been automated as new functionality in the AP-LIS have become available. The applicability and potential usefulness of automated tracking systems in anatomic pathology has been widely accepted. Herein, we examine how tracking data could be utilized with standard tools within the AP-LIS as well as applications beyond asset tracking.

Technology

CoPath Plus v6.0 was used as the AP-LIS (Sunquest Information Systems, Tuscon, AZ), along with the specimen management routing and tracking (SMART) module. Customized InfoMaker reports were created using PowerBuilder software (Sybase, Dublin, CA).

Design

Utilizing CoPath SMART, we defined specimen points of tracking (SPOTs) in the AP-LIS dictionary. Each SPOT was linked to a specific workstation as defined on the health system's internal network. As each asset was scanned and processed at designated tracking points; their location, scan time, status, and associated user data were automatically recorded. Standard tracking tools within the AP-LIS allowed us to monitor assets in real-time throughout the laboratory workflow. We then used this data to create customized reports, which could be leveraged in lab management and workflow monitoring.

Results

We utilized both tracking functions and customized reports for analyzing a variety of workflow processes. Standard AP-LIS tracking tools allowed us to evaluate each case's hierarchy of assets in a single view. This enabled individuals to evaluate each case asset and their statuses in the laboratory workflow. The tools also provided valuable error checking capability, in that individual assets could be reconciled for completeness and/or technical errors resulting from workflow defects. Beyond asset tracking, customized reports allowed technicians and lab managers to monitor workload distribution, productivity, defects in specimen processing, and the quality control of asset labeling.

Conclusion

An automated tracking system, integrated with the AP-LIS has potential to enhance laboratory efficiency, reduce errors, and add a component of accountability (i.e. chain-of-custody) as specimens travel through the laboratory. The availability of granular tracking data can enhance the management of laboratory practices in anatomic pathology.


   Electronic Poster Session Top


Thursday, May 15, 2014

Presented in the Grand Foyer, 2 nd Floor, Wyndham Grand Pittsburgh Hotel

Neuropathology Specialty Training Using Wikis for Case Presentation of Unknowns

James R. Hackney, Seung Park

Department of Pathology, Division of Informatics, University of Alabama at Birmingham. E-mail:jhackney@uab.edu

Content

While a few publicly available pathology image archives are now in existence, curated collections of digital neuropathology images, presented within their clinical context, are needed. The wiki format is ideal for the presentation of contextualized medical images in any number of interesting formats for instructional purposes. We have recently built an instructional neuropathology wiki for the education of Residents and Fellows using the case presentation format.

Technology

Server Hardware: Dell Precision T3600; Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Guest Operating System: Ubuntu Linux Server 12.04 LTS 64-bit; Web Server: Nginx 1.5; Database Management System: MariaDB 5.5; Programming Language: PHP-FPM 5.3; Rapid Publication Environment: MediaWiki 1.21; Image Editor: GNU Image Manipulation Program (GIMP) 2.8.

Design

A virtual machine with a standard LEMP (Linux, nginx, MariaDB, PHP) stack was generated and deployed; MediaWiki was installed atop this stack. A selection of interesting neuropathology cases was identified; case report pages for these cases were constructed using MediaWiki's markup language, WikiText. Histopathology and radiology images from these cases were then post-processed in GIMP, uploaded into MediaWiki, and added to the pertinent pages, completing the case reports.

Results

Our educational wiki for surgical neuropathology simulates the clinical approach to any difficult diagnostic case, including patient presentation with pertinent history, laboratory studies, radiological images, and histopathology. Where appropriate, we present images of the squash preparation and the frozen section, asking the observer to construct a differential diagnosis prior to moving on to the permanent sections, immunohistochemical stains, and case discussion.

Conclusion

Use of this wiki exposes residents to a much greater number of cases than they are likely to see during their rotation in Neuropathology. It also exposes them gently to informatics, as we teach our residents WikiText and have them upload a case per week over the course of their Neuropathology rotation. This wiki will therefore become an integral part of our Neuropathology rotation, giving our residents the opportunity to (a) Learn neuropathology by generating teaching content and (b) learn informatics by editing and administering a wiki in a well-supported and beginner-friendly environment.

Smartphone Adapters for Digital Photomicrography

Douglas J. Hartman, Somak Roy, Liron Pantanowitz, Milon Amin, Raja R. Seethala, Ishtiaque Ahmed, Samuel A. Yousem, Anil Parwani, Ioan Cucoranu

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

Content

Photomicrographs in Anatomic Pathology provide a means of quickly sharing information from a glass slide for consultation, education, documentation and publication. While static image acquisition historically involved the use of a permanently mounted camera unit on a microscope, such cameras may be expensive, need to be connected to a computer, and often require proprietary software to acquire and process images. Another novel approach for capturing digital microscopic images is to use smartphones coupled with the eyepiece of a microscope.

Technology

Recently, eight smartphone adapters have been developed that allow a user to capture digital images from an optical device (microscope, binoculars, telescope or gun scope). The aim of this study was to test the utility of smartphone adapters for digital photomicrography.

Design

We surveyed the market for adapters to attach smartphones to the ocular lens of a conventional light microscope. Three adapters (Magnifi, Skylight and Snapzoom) were tested. We assessed the designs of these adapters and their effectiveness at acquiring static microscopic digital images.

Results

All adapters facilitated the acquisition of digital microscopic images with a smartphone. The optimal adapter was dependent on the type of phone used. The Magnifi adapters for iPhone were incompatible when using a protective case. The Snapzoom adapter was easiest to use with iPhones and other smartphones even with protective cases.

Conclusion

Smartphone adapters are inexpensive and easy to use for acquiring digital microscopic images. However, they require some adjustment by the user in order to optimize focus and obtain good quality images. There are numerous possible applications for which acquisition of images via a smartphone may be ideal: (a) Capturing images for tumor board presentation, (b) sharing occasional interesting cases for educational purposes, (c) for rapid consultations, such as for frozen sections, (d) screening select cases to assess the need for submitting the entire case for consultation, (e) for quality assurance to document using static images, and (f) for submitting static images of single microscopic fields for teleconsultation. Smartphone microscope adapters provide an economically-feasible method of acquiring and sharing digital pathology photomicrographs.

Use of Time-of-Day Dependence of Running Averages as Input to Patient-Based Quality Control (PBQC): An Example Using K+ Data for a University Hospital

Bryon P. Jackson, Laura J. McCloskey, Douglas F. Stickle

Jefferson University Hospital, Philadelphia, PA. E-mail: bryonpjackson@gmail.com

Content

In many hospital laboratories, significant time-of-day (TOD) variation in running averages of hospital patient data may be operative for certain analytes, due to regular TOD-dependent variation in proportions of inpatient and outpatient subjects. Where such patterns are stable, it is anticipated that use of TOD-dependent running averages as inputs to PBQC algorithms could improve receiver-operator characteristics of PBQC. We examined this premise for PBQC applied to measurement of K+ in a university hospital.

Technology

Primary patient K+ datasets for two successive months (M1, M2) were obtained from the Sunquest LIS system. Data reduction and simulations of PBQC were performed using Visual Basic.

Design

Running means of length 30 (A30) were calculated across the datasets with input restricted to samples within the K+ reference range (3.5-5.0 mmol/L). For both M1 (n = 17342) and M2 (n = 17199), average A30 was 4.13 ± 0.075 mmol/L (standard deviation, s). Average A30 as a function of time-of-day (TOD) was calculated for 48 half-hour intervals across 24 hours. Low-frequency-filter Fourier transforms produced continuous, smooth functions (F1, F2) characterizing A30 vs. TOD. Simulations (n = 1000) to determine length-of-sequence of PBQC to detect 1x(2.5s) deviations of A30 from average A30 (false positives, assuming in-control data) were performed using random starting points within the sequence of data points for M2, when using either a time-invariant average A30 from M1 as input (control), or when using TOD-dependent F1 for average A30 from M1 as input (experiment).

Results

TOD pattern of variation of A30 was stable across M1 and M2, with high correlation of F2 vs. F1 (r2 = 0.929). Control simulations (using fixed A30 from M1) yielded a false-positive rate of 23.8 per month (1 per 1.3 days) for M2. Experiment simulations (using TOD-dependent A30 function F1 from M1) yielded a false-positive rate of 14.4 per month (1 per 2.1 days) for M2.

Conclusions

In this example, use of TOD-dependent data for A30 as input for PBQC analysis of K+ data decreased the false positive rate by 40%. In some settings, use of TOD-dependent patterns of running averages of patient data may improve overall receiver-operator characteristics of PBQC.

Data Explorer: A Tool for Data Exploration and Cohort Discovery in Massive Multidimensional Datasets

Ameen Kazerouni, Darryl Tharpe, Ashish Sharma

Department of Biomedical Informatics, Emory University, Atlanta, GA. E-mail: ameen.kazerouni@emory.edu

Content

Large collections of multidimensional data, accumulated from multiple sources, are integral components of integrative studies. Here we report on a system that provides interactive, visualization of multidimensional data that facilitates data exploration and cohort discovery within the datasets.

Technology

The system uses the Silverlight based PivotViewer Technology. PivotViewer is a framework to interactively explore massive multidimensional datasets. Such interactive explorations help expose hidden patterns and help create cohorts, non-trivial tasks if the data is not being explored interactively.

Design

Multidimensional biomedical datasets include clinical data, genomic data, imaging data etc. Interactive visualization of such collections can provide the information needed to formulate hypothesis, create cohorts for validation, and help make data driven decisions. The Data Explorer, exposes The Cancer Genome Atlas BRCA clinical dataset and images. A pivot collection requires two primary components, (1) Data for querying and filtering; (2) Images to visualize the data.

Results

The Data Explorer is used to filter the collection on one or more clinical attributes. It allows users to explore the dataset as an interactive whole rather than just one patient at a time. As exploration progresses, a query is formulated and maintained. Cohorts of interests can be pivoted on an additional attribute and the distribution of the cohort across that attribute can be visualized as histograms. As seen in [Figure 1], histogram bins can be clicked on to create additional data subsets. Individual tiles can be single clicked on to view all associated metadata. Fields within the metadata can be used to further expand the query. Double clicking on an individual tile will launch the digitized pathology image in our image viewer, caMicroscope, for further analysis.

Conclusion

The Data Explorer allows you to visualize large multidimensional datasets and allows efficient exploration and cohort discovery within multidimensional datasets. Future work on this system will include the integration of the Data Explorer into a data analysis workflow. The workflow would allow for the extraction of formulated queries and redirection of discovered cohorts to analysis engines.



Cytologically Yours: The Utility of a Wiki as a Teaching Tool in Cytopathology and Informatics

Jessica A Levesque, 1 Stephanie Simmons, 1 Isam-Eldin Eltoum, 1 Seung Park 2

1
Department of Pathology, Division of Cytopathology, University of Alabama at Birmingham, 2 Department of Pathology, Division of Informatics, University of Alabama at Birmingham. E-mail: jalevesque@uabmc.edu

Content

Informatics platforms - most notably rapid publication environments (e.g. wikis) - are poised to disrupt current models of resident and fellow education. The benefit of wikis in education is well-documented, but adoption of this technology in cytopathology has been slow. In our cytopathology fellowship program, we have built a wiki ecosystem in order to provide trainees with (a) A searchable archive of teaching material, (b) experience in programming and administering of a complex, stack computing-enabled web application, and (c) a curated virtual machine that can be arbitrarily redeployed on any x86-64 computer.

Technology

Server Hardware: Dell PowerEdge 2950; Server Host Virtualization Hypervisor: VMWare ESXi 4.1.0; Development Host Virtualization Hypervisor: Oracle VM Virtualbox 4.3; Guest Operating System: Ubuntu Linux Server 12.04 LTS 64-bit; Web Server: nginx 1.5; Database Management System: MariaDB 5.5; Programming Language: PHP-FPM 5.3; Rapid Publication Environment: MediaWiki 1.21.

Design

Under the guidance of a pathology informaticist, two cytopathology fellows, both without prior informatics experience, built a field-deployable LEMP (Linux, nginx, MariaDB, PHP) stack as a virtual machine; MediaWiki was then installed atop the stack. One fellow catalogued extant cytopathology teaching materials (organ based cytology lectures, "unknown" conferences, cytologic/histologic correlation conferences) and converted them directly into wiki-based case presentations. The other fellow programmed several PHP-based extensions for MediaWiki, one of which allows for the direct display of Microsoft PowerPoint slide shows in the wiki. This combination approach allowed for both (a) Rapid generation of teaching content and (b) easy reuse of current teaching content.

Results

The wiki has become an integral part of both the cytopathology fellowship and the resident cytopathology rotation. Case studies, didactic material, and "unknowns" are added to the wiki on a monthly basis.

Conclusion

The creation of a wiki page for teaching allows for a centralized location for residents, students, and fellows to read and learn about cases from our institution. Furthermore, it gives residents and fellows the opportunity to learn informatics by actually "doing informatics" - in this case administering and maintaining a relatively complex Web 2.0 platform. We will continue to add functionality (videos, live telecytology, whole slide imaging) in future versions.

Automated Registration of Multimodal Microscopy Images for Integrative Analysis of Tumor Tissue Sections

Giuseppe Lippolis,1 Anders Edsjö, 2 Leszek Helczynski, 3 Anders Bjartell, 1 Niels Chr Overgaard, 4

1
Department of Clinical Sciences, Division of Urological Cancers, Lund University, Malmö, Sweden, 2 Department of Pathology, University of Gothenburg, Gothenburg, Sweden, 3 University and Regional Laboratories Region Skåne, Clinical Pathology, Malmö, Sweden, 4 Centre for Mathematical Sciences, Lund University, Malmö, Sweden. E-mail: giuseppe.lippolis@med.lu.se

Content

Morphological and molecular information from tissue samples may be integrated by aligning microscopic histological images in a multiplex fashion. This process is usually time-consuming and exhibiting intra- and inter-user variability. Our aim is to investigate the feasibility of applying modern image analysis methods to automatic alignment of microscopic images from differently stained adjacent paraffin sections from tissue specimens.

Technology

The algorithm has been initially tested in Matlab and then implemented in a C++ framework using free libraries like OpenCV. Image pairs were aligned allowing for translation, rotation and scaling. The registration was performed automatically by first detecting landmarks in both images, using the scale invariant image transform (SIFT), followed by the well-known RANSAC protocol for finding point correspondences and finally aligned by Procrustes fit. The Registration results were evaluated using both visual and quantitative criteria.

Design

Tissue samples, obtained from biopsy or radical prostatectomy, were sectioned and stained with either hematoxylin and eosin (H and E), immunohistochemistry for p63 and AMACR or Time Resolved Fluorescence (TRF) for androgen receptor (AR). Another set of breast biopsies was also used for 12 biomarkers on 12 consecutive tissue sections.

Results

Images of consecutive tissue sections stained with H and E and p63/AMACR were successfully aligned in 85 of 88 cases (96.6%). The performance of the algorithm was evaluated in different Gleason scores classes. The failures occurred only in 3 out of 13 cores with highly aggressive cancer (Gleason score ≥8) but these cases showed also some damage in the tissue. Secondly, TRF and H and E image pairs were aligned correctly in 103 out of 106 cases (97%). The third experiment considered the alignment of image pairs from large breast biopsies progressively further away. The initial results show a success rate of ca. 80%.

Conclusion

The proposed method is both feasible and fast and well suited for automatic segmentation and analysis of specific areas of interest. It can enable multiplex biomarker analysis, TMA alignment and integration of morphological information with protein expression data from a stack of consecutive tissue sections.

Texture Analysis and Gleason Sub-Pattern Classification in Prostate Cancer Staging

Nicholas P. McCarthy, 1 Padraig Cunningham 1 , Jim Diamond 2 , Gillian O'Hurley 1

1
University College Dublin, Computer Science and Informatics, Dublin, Ireland, 2 PathXL Ltd., Belfast, Northern Ireland. E-mail: (nicholas.mccarthy@gmail.com)

Content

We investigate the identification and classification of tumour and Gleason patterns in prostate histopathology images using different texture methods. Most recent papers in this area have concentrated in the binary (tumour vs non-tumour) and multi-class (Gleason pattern) cases. Our dataset includes intermediary patterns where Gleason 3 is evolving into Gleason 4, and Gleason 4 into Gleason 5. We also investigate the use of clustering as a classification pre-processing step in the identification of global and intra-Grade tissue patterns, and how they relate to increases in accuracy in the multi-class case.

Technology

The 20 images used in the dataset were scanned at 40x on a Leica SCN400 Slide-Scanner. Features are extracted using an OpenCV and C++ based framework. Linear SVM models are used to classify image sub-regions as one of the described classes.

Design

We perform a discrete mapping of an image in a tile-based classification framework. Images were annotated with 5 classes: Gleason 3, 4, 5 and intermediary patterns. Annotated regions were validated by a panel of pathologists. Random-forest feature selection is performed to improve accuracy and determine good feature subsets.

Results

Choosing correct parameters for some feature sets will in general perform just as well as a feature subset chosen from a larger set using feature selection. Feature selection also improves accuracy of Gleason pattern classification, and lowers feature extraction time. Sub-pattern clustering for subsequent classification improves base model accuracy and can help the model and feature selection process by visualizing the different clusters present with given image parameters. When classifying tumour versus benign stromal regions we have achieved an accuracy of 86.9% accuracy. Additionally, we have reached 80.99% accuracy when discriminating Gleason 3 and Gleason 4 patterns, and 90.68% for Gleason 4 and Gleason 5.

Conclusions

Feature selection should be performed routinely in cases where there are many disparate feature sets and large feature vectors, as is often the case in automated histopathology analysis. This does not increase accuracy in all cases, but should reduce feature extraction time by removing redundant and highly-correlated features, and avoid overfitting a model to a particular set of images, thereby increasing model generalization on unseen images.

Diagnostic Differences between Glass Slide and Digital Slide Image Evaluation: A Morphological Feature Based Study

Anil V. Parwani 1, David Glinski 2 , Orly Aridor 2 , Liron Pantanowitz 1 , Leslie Anthony 2 , Jonhan Ho 3

1
Department of Pathology, University of Pittsburgh School of Medicine, Pittsburgh, PA, 2 Office of Sponsored Programs and Research Support, University of Pittsburgh Medical Center, Pittsburgh, PA, 3 Department of Dermatopathology, University of Pittsburgh Medical Center, Pittsburgh, PA. E-mail: (parwaniav@upmc.edu)

Content

Prior pathology validation studies comparing glass to digital slides report excellent diagnostic concordance. These studies focused largely on diagnoses and not specific morphology.

Technology

The aim of this study was to assess diagnostic differences between glass and digital slides focusing on morphologic features. A second intent of the study was to provide an opportunity for practical application of recently installed scanning equipment at five United States Air Force Medical Service (AFMS) pathology centers.

Design

Surgical pathology cases with H and E slides were randomly selected from AFMS and University of Pittsburgh Medical Center pathology labs. Representative slides (1-2 key slides/case) were scanned at 20x using Aperio AT scanners (Vista, CA). Of 131 digitized cases, 7 were excluded due to either levels/tissue fragments missed during scanning (n = 4) or pathologist deferral for subspecialist consultation (n = 3). Nine pathologists at six locations reviewed the remaining 124 cases. Case-related clinical histories were provided. Each participating pathologist viewed both digital and glass slides with no wash out period and documented detailed histomorphologic features required for diagnosis.

Results

Overall intra-pathologist concordance for morphologic features was 96.8% (120/124 cases). Pathologists reported a total of 7 cases for which digital slide evaluation of morphologic features was problematic; in addition to the 4 cases that pointed to specific differences in morphologic features, 3 cases referenced better color and resolution with glass slides viewed under the microscope. Three additional cases referenced multiple tissue fragments being easier to examine on the glass slide. Though it was not a focus of the study, diagnostic concordance was 95.9% (119/124); discordance ranged from minor (n = 4) to moderate (n = 1).

Conclusions

The current study is unique because no wash out period challenged study pathologists to compare and document differences in morphological features between these two modalities in real time. Limited AFMS digital pathology experience may have contributed to comments favoring glass slide examinations. Additional studies are needed to determine if intra-pathologist concordance improves with increased digital experience.

Utilization of Computerized Physician Order Entry and Electronic Medical Record In-Patient Documentation to Improve Fine Needle Aspiration Service

Matthew A. Smith 1 , Sara E. Monaco, 1 Joel Weinberg 2 , Anil V. Parwani, 1 Liron Pantanowitz, 1

1
Departments of Pathology and 2 Internal Medicine, University of Pittsburgh Medical Center, Pittsburgh, PA. E-mail: smithma@upmc.edu

Content

Benefits of a patient electronic medical record (EMR) include computerized physician order entry (CPOE) and electronic documentation. Currently, these features are typically not used in Anatomical Pathology practice. We describe our experience of customizing the EMR to facilitate fine needle aspiration (FNA) consultations for inpatients and the creation of a cytopathology-specific template for documenting these procedures directly in the patient's electronic chart.

Technology

EMR used is PowerChart (Cerner, Kansas City, MO).

Design

Requests for FNA consultations on inpatients were added to the CPOE menu in the EMR. The order was named "Cytology, Fine Needle Aspiration FNA Consult", with synonyms "biopsy and "fine needle aspiration". Data fields captured patient details, reason for FNA, anatomical site to be biopsied, and requesting physician's contact details in the order. An electronic template was subsequently designed to generate a note in the EMR after completing an inpatient FNA.

Results

Electronic orders for a bedside FNA triggered three simultaneous actions: (1) Automatic notification of the hospital operator to alert the cytopathology secretary to schedule the FNA, (2) printing of the order on a networked printer located in the cytopathology department, and (3) addition of the order to a daily cytology log. Offering FNA orders in the CPOE menu made it easier for clinicians to request these procedures and for their orders to contain helpful clinical information. Recording completed FNA procedures in the EMR by cytopathology staff promoted standardization of reports, ensured accurate and complete paperless documentation in the EMR, and reduced the number of calls to cytology from clinical house staff inquiring about preliminary diagnoses. Forms completed by cytopathology fellows participating in a bedside FNA were designed to be forwarded to their attending's queue to be endorsed.

Conclusion

Our experience demonstrates the benefits our cytopathology division derived by leveraging available features in the institutional EMR. Employing CPOE for clinicians to order FNAs on inpatients made it user-friendly and more in line with how they currently order other tests and consultations. Electronic templates for procedures notes on inpatient FNAs within the EMR improved documentation in the patient chart, streamlined the workflow, reduced the need for transcription, and made the preliminary cytology findings accessible to clinical teams caring for inpatients. Efforts are underway at our institution to add similar CPOE orders to our outpatient EMR (Epic) to promote our FNA clinic service.

Applications of Google Glass [GLΛSS TM ] in Autopsy Pathology

Muhammad A. Syed 1 , Jeffrey Nine 1 , Ishtiaque Ahmed 2 , Anil V. Parwani 1 , Chris Saylor 3 , Jeff McHugh 2 , Brian Kolowitz 2 , Jeffrey Taylor 2 , George Michalopoulos 1 , Liron Pantanowitz 1

1
Department of Pathology, University of Pittsburgh Medical Center, 2 Information Services Division, University of Pittsburgh Medical Center, 3 Center for Medical Innovation, University of Pittsburgh. E-mail: (syedma2@upmc.edu)

Content

Google Glass (GLΛSS TM) are glasses that consist of a head-mounted, wearable computer that can connect to the Internet and display information in a smartphone-like format. Only a few individuals in the United States have been selected to test Google Glass as an Explorer Program. Our aim was to explore the utilization of Google Glass in gross autopsy pathology.

Technology

Google Glass (explorer edition) runs the Android operating system with a 640×360 display and 5-megapixel camera capable of taking photos and recording video. Glass can use Wi-Fi 802.11b/g or Bluetooth for connectivity. The device has a built-in 16GB hard drive for storage with 682MB RAM. Unique features include a motion sensitive accelerometer for gestural commands, ambient light sensing, proximity sensor and a bone conduction transducer. An extended external battery pack and charger for USB Mobile Devices (9000 mAh) was employed during testing.

Design

After establishing a secure WiFi network connection Glass was tested in the autopsy suite [Figure 1]a. The pathologist initiated video calls using a shared Google+ hangout. Remote pathologists invited to join Google+ could view live feeds from the autopsy suite. The ergonomics during autopsy, ability to acquire static images and perform telepathology were evaluated.



Results

Google Glass use in the autopsy room had pros and cons. Streaming video and audio quality to participants in other hospitals was clear. Hands-free capturing of static images was easy and image quality of organs being examined was acceptable at a 2-3 foot distance from the specimen [Figure 1]b and c. Problems included distracting background noise in the autopsy suite from ventilation, echoing and presence of multiple people. Voice navigation failed when the wearer used a surgical mask/shield. Finger-swipe navigations were difficult because of bloody gloves.

Conclusions

Overall there are potential positive uses of Google Glass in the autopsy room. The best use of Glass was streaming video to others during an autopsy. Finger-swipe navigation of Glass is best avoided during an autopsy. Problems related to voice recognition could be alleviated by adding a headset microphone and installing clear eye lenses provided with the glasses instead of a protective face shield [Figure 1]d.

Mapping LOINC Codes: The UPMC Experience

Muhammad Syed 1 , Tony Gigliotti 2 , Michael Sendek 2 , Anil V. Parwani 1 , Liron Pantanowitz 1 ,

1
Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA, 2 Information Services Division, University of Pittsburgh Medical Center, Pittsburgh, PA. E-mail: syedma2@upmc.edu

Content

Under the Health Information Technology (HITECH) Act, Meaningful Use supports using Logical Observation Identifiers Names and Codes (LOINC) as a standard for electronic exchange of laboratory data between the laboratory information system (LIS) and electronic medical records (EMRs). As a result, pathology laboratories that do not routinely use LOINC need to map their lab catalogues to the LOINC standards. The aim of this communication is to share our experience with mapping LOINC codes at the Univeristy of Pittsburgh Medical Center (UPMC).

Technology

Tools used to identify LOINC definitions were the LOINC dictionary portal and RELMA desktop mapping program found on the LOINC homepage: http://loinc.org/LOINC mapping employed our Sunquest LIS (v7.1, Sunquest Information Systems, Tucson, AZ) and dbMotion™ clinical data repository (v5.0, dbMotion Inc., Pittsburgh, PA).

Design

All lab tests utilized during one year at UPMC recorded in the dbMotion™ clinical data repository were analyzed. We identified over 23,000 laboratory test results. The tests were sorted according to frequency of use. A Pareto analysis demonstrated that 80% of tests ordered occurred in the first 300 highest volume tests. LOINC codes for lab tests were mapped in two phases; first applying LOINC to the top 300 tests, then to the remainder of the data set. Each discrete code was manually evaluated utilizing LOINC web tools, except for send out tests where codes were supplied by our primary reference laboratory. Verified LOINC codes were maintained in the Sunquest LIS and mapped with all downstream EMRs receiving lab results.

Results

Phase 1 required 2 months of analyst time and phase 2 another 7 months to code the remaining tests. An estimated 1000 hours was needed. Test descriptors entered in the LOINC search tool often brought back many hits, each needing evaluation for "Best fit" of LOINC mapping. Various lab supervisors and clinical pathology directors were consulted to code difficult tests, specifically microbiology and molecular tests. For newer lab tests, such as molecular genetics and next generation sequencing, no LOINC codes were found.

Conclusion

LOINC mapping is a challenging task that requires many manual hours to accomplish. Difficulties were experienced with manual data entry errors, lack of analyst knowledge regarding the meaning of esoteric lab tests, and no codes for new molecular tests. While defined LOINC codes could be maintained in the Sunquest LIS to support interoperability with our multiple EMRs, the LIS was limited by a lack of hierarchy for test nomenclature. There were limited options for us to perform quality assurance on our LOINC database to verify the accuracy of codes. Successful LOINC mapping has enabled UPMC to not only fulfil the requirements required to comply with Meaningful Use, but to use these codes for electronic laboratory reporting with state agencies and enable enterprise data analytics ventures.




 

 
Top
  

    

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

 
  In this article
    Short Abstract P...
    Electronic Poste...
    Short Abstract P...
    Electronic Poste...
    Platform Short A...

 Article Access Statistics
    Viewed8990    
    Printed71    
    Emailed0    
    PDF Downloaded412    
    Comments [Add]    

Recommend this journal