|J Pathol Inform 2017,
|Date of Web Publication||10-May-2017|
Source of Support: None, Conflict of Interest: None
|How to cite this article:|
. Abstracts. J Pathol Inform 2017;8:20
| The De Novo Process and Beyond|| |
1Emerging Businesses, Philips Digital Pathology Solutions E-mail: [email protected]
The DPA recently suggested that WSI device intended for primary diagnosis is recommended to be submitted for de novo applications to the Food and Drug Administration (FDA), which is a great step forward. The DPA is engaged with the FDA and discussing the differences of the pathways, such as general and special controls and change control process. What are the advantages, what does this mean to industry and users, what does this mean for WSI and computational pathology, what is next?
| Towards Better Digital Pathology Workflows: Quality Assessment of Whole Slide Image, 20 Trillion Pixels Later|| |
David Ameisen1,2,3,4, Julie Auger-Kantor1, Emmanuel Ameisen1
1Paris Biotech Santé, Université Paris Descartes, Paris, France, 2IRIF, 3CNRS, 4Université Paris Diderot, Paris, France.E-mail: [email protected]
Background: We have designed quality assurance tools to quantify blur of custom-sized tiles composing Whole Slide Images regardless of their dimensions, quantity or acquisition rate, generating global results, exchangeable logs and sharpness-maps. We have now integrated our programming libraries and standalone programs to existing workflows and software in order to improve quality assurance procedures. Methods: One notable implementation is inside the FlexMIm project, a collaboration that aims to improve the global workflow of digital pathology (27 pathology laboratories in the Paris area, research laboratories from 2 universities and 3 companies). All WSI are analyzed and mapped as they are stored in the platform. Quantified results are stored in a JSON file, in a database, or converted to be interoperable with existing software. Using thresholds' profiles, WSI are tagged as “accepted”, “to review”, “to rescan” and can be double-checked along their focus map. In order to only target the tissue samples inside each WSI, we built a dynamic blank-tile detection based on statistical analysis of each WSI. Results: This solution is extremely fast and easily scalable as storage and processing can be distributed amongst local or remote servers, and quantified results can be stored in remote databases: more than 20 trillion pixels have been analyzed at a 3.5 billion pixels per quad-core processor per minute rate. Conclusions: During acquisition, it can help re-acquire the WSI tiles that were not scanned properly. Once the WSI stored, systematic quality checks can detect de-calibration of the acquisition devices, list the WSI that must be re-acquired and notify users. During image analysis, the regions of interest can be automatically suggested according to their quality-map, reducing the amount of incoherent results and significantly accelerate the process. In a remote WSI viewer, tiles of best quality can be sent first, and magnification levels can be dynamically resampled for a better rendering. Finally, quantified quality scores should be integrated in the WSI's metadata for traceability purposes. Our tool can be used upstream or downstream, at every step of the image acquisition, management, display and analysis process for better medical imagery workflows and a more efficient quality-controlled care.
| Digital Pathology: Drivers for Full Adoption|| |
Sylvia L. Asa1, Zoya Volynskaya1
1Department of Pathology, University Health Network, University of Toronto, Toronto, Canada. E-mail: [email protected]
Background: Implementation of digital pathology is on the horizon for clinical use. However, full adoption will require the input of experts from multiple disciplines, including clinicians, histotechnologists, informatics experts, institutional information technology teams, and vendors. The factors impacting participation in digital pathology diagnostics are complex and remain to be determined. Methods: We evaluated the platforms and processes used to implement digital pathology in a large multisite tertiary care Department of Pathology over 10 years. Results: Adoption was dependent on three major factors: (1) Availability of reliable and user-friendly technology to meet the changing needs and workflow integration for pathologists and histologists, (2) a clear need for the solution and (3) a perceived benefit to the users. Improvements in technology and workflow increased the number of users and use cases, and technologic failures resulted in reduced participation. The benefits identified by users included time savings, after-hours access to images, ease of viewing, accuracy of annotation and academic benefit. Equally as common were complaints about cases requiring more time, technical limitations, and lack of perceived benefit. The addition of appropriate support increased slide scanning for various additional purposes including teaching, tumor boards and publications, thus increasing number of pathologists participating in digital review for non-clinical usage. However, this had a minimal impact on clinical use. Conclusions: Full adoption of digital pathology requires the appropriate infrastructure to provide pathologists with high quality images at high speed. Integration into clinical workflow is essential. The lack of a clear need for this technology will be a barrier in conventional pathology groups. However, the perceived benefits of adoption, including appropriate business cases for practice groups and academic advances must be developed to ensure success of this disruptive technology.
| Making Good on 30 Years of Hype. Digital Pathology Finally Comes of Age|| |
Ulysses G. J. Balis1
1Department of Pathology, University of Michigan Health System, Ann Arbor, Michigan, USA. E-mail: [email protected]
Digital Pathology has progressed through multiple generations of technologies and associated expectations, since its inception in the late 1980s. Through an evolutionary and incremental process over the past twenty or more years, the prior barriers in digital pathology have systematically given way, yielding contemporary solutions that, at least in theory, can now deliver on the promise of standing up an all-digital workflow model. In tandem with that, a number of additional operational and diagnostic benefits are poised to accompany the use of digital pathology. But, even with the underlying technology no longer being a substantive issue, widespread adoption in the US still waits on the heels of device validation. Fortunately, this too is now firmly in sight as a result of the FDAs recent de novo announcement. This presentation will provide salient history of this field, followed by a series of deep dives on how Digital Pathology is likely to be adopted for use in primary sign out settings in the very near term. Emphasis will be placed on use cases where this technology can offer solutions that are simply not possible with use of conventional microscopy-based workflow alone, including: image-based analytics and informed detection, productivity tools, real-time asset tracking, collaborative practice and finally, workflow models. As the final segment of this presentation, an overall adoption model will be explored – one in which there can be no doubt that Digital Pathology has truly come of age and is ready for the task at hand.
| Computer Aided Assessment of Sentinel Lymph Node Histology for Breast Cancer Patients|| |
Maschenka Balkenhol1, Francesco Ciompi1, Marcory van Dijk2, Babak Ehteshami Bejnordi3, Meyke Hermsen1, Peter Bult1, Nico Karssemeijer3, Geert Litjens1, Jeroen van der Laak1
1Departments of Pathology and 3Radiology and Nuclear Medicine, Radboud University Medical Center, Nijmegen, 2Department of Pathology, Rijnstate Hospital, Arnhem, The Netherlands. E-mail: [email protected]
Background: In the last decade, the sentinel lymph node (SLN) procedure resulted in decreased morbidity and a reduction in the number of lymph nodes examined by pathologists. Pathological evaluation of the SLN is a critical yet tedious procedure, making it a strong candidate for automation. Advanced pattern recognition techniques (e.g. deep learning) applied to high resolution, digitized microscopic sections (so-called whole slide images; WSI) will facilitate pathologists in detecting clinically relevant regions, optimizing their workflow. Study aims (1) develop deep learning techniques to detect SLN metastases in H&E stained WSI, and (2) to set up a workflow to routinely analyse SLN whole slide images. Materials and Methods: A convolutional neural network (convnet) was trained to recognize tumours using a large set of H&E stained WSI of SLN, in which all tumour was outlined. The convnet was subsequently evaluated using an independent test set. To study the state-of-the-art in this field, we also organized an ISBI Grand Challenge (“Camelyon16”) to enable researchers world-wide to work on this complex task, using our WSI. Next we designed a system analysing all new SLN in our department with the constructed convent, providing the pathologist these results during primary reading. Results: The convnet identified 90% of all metastases at the cost of one false positive detection per normal WSI. Forty percent of WSI without tumour in the test set was identified without missing any cancer case. Results from the Grand Challenge show that deep learning approaches the accuracy of a trained pathologist. In our hospital, we now automatically analyse every SLN (200 patients per year) and present the results to the pathologist, together with the glass slides for primary reading. Conclusions: Our deep learning technique could fully automatically identify SLN metastases and is currently used in clinical practice in our hospital.
| Combinatorial Approach with Digital Pathology and Social Media in Enhancing the Effectiveness of Pathology in Medical Undergraduate Training|| |
1International Medical School, Shah Alam, Selangor, Malaysia. E-mail: [email protected]
Background: The role of pathology in undergraduate medical training is invaluable. To date, there is limited research on the impact of combining digital pathology with social media. This study seeks to understand the role of combining digital pathology with social media in engaging, motivating and enhancing the effectiveness of pathology training among medical students. Methods: One hundred medical students were divided into two groups of 50 students each. One group was provided with the combined social media and digital pathology method of teaching whereas the control group had the regular glass slides and microscopes. Pathology slides used for instruction were photographed using mobile camera and video options. Selected Digiscan digital pathology slides were also used. Different areas of the slides were posted on social media such as Facebook for the test group. The test group was asked to collaborate and arrive at a conclusion using the digital slides, arrive at a diagnosis and provide additional comments about microscopic features that would affect patient care. The control group had regular glass slides for the same. A pre and post-test qualitative questionnaire and student responses were evaluated. Results: This study showed that the group using digitized slides with social media had better group dynamics on qualitative analysis and also developed deeper learning of concepts since they utilized the digitized slides to explore further about the topic on internet and similar slides posted Facebook. Thus this group was able to generate more ideas and insight about the various aspects of heterogeneity in tissues that have an impact on diagnosis. Both groups arrived at the same diagnosis. The control group which did not use the digitized slides presented their insight and ideas also as the test group but had not explored deeply into the wide range of microscopic features as the test group and were not as highly motivated as the digitized slide groups with the social media component. Conclusions: Digitized slides combined with social media are a powerful and effective tool in enhancing the quality of pathology education. This combinatorial approach encourages collaborative learning and enhances motivation among medical undergraduate students. Further research is required in this area.
| Getting Ready for College of American Pathologists Inspection in Digital Pathology|| |
Emmanuel Agosto-Arroyo1, Marilyn M. Bui1
1Moffitt Cancer Center, University of South Florida, Tampa, Florida, USA. E-mail: [email protected]
Digital pathology has been incorporated into daily anatomic pathology practice since digital cameras and digital images are being used. Quantitative image analysis of cancer biomarkers has made great impact on precision medicine. The emerging incorporation of whole slide imaging and telepathology created new ways for anatomic pathology laboratories to improve on gathering, managing and interpreting diagnostic, prognostic and predictive information. This represents a challenge to the laboratory personnel to appropriately utilize digital pathology for reliable and reproducible results. The College of American Pathologists (CAP) has developed an accreditation checklist for digital pathology, which encompasses various aspects from digital pathology practice. In this workshop, the current CAP digital pathology requirements for accreditation will be discussed. Institutional experience and literature review will be discussed to offer best practice solutions. This will not only prepare the pathology labs to be in compliance with the current CAP checklist but also to be better prepared for the wider use of whole slide image for primary diagnosis in the near future.
| Image Analysis of Tumor Microenvironment: A Customized Application Measuring Intratumoral and Extratumoral pH|| |
Tingan Chen1, Mark Lloyd1, Veronica Estrella1, Robert A. Gatenby1, Robert J. Gillies1, Marilyn Bui1
1Moffitt Cancer Center, University of South Florida, Tampa, Florida, USA. E-mail: [email protected]
Background: Dysregulated pH is emerging as a hallmark of cancer which enables cancer progression by promoting proliferation, apoptosis, metabolic adaptation, migration and invasion. Comprehension of mechanisms of pH regulation in tumors is of paramount importance for therapeutic implications. New advances, including advanced pH sensors and imaging and analysis methodologies, have provided insight into the molecular basis for pH-dependent cell behaviors that are relevant to cancer cell biology. Methods: Here we describe how to customize a commercially available image analysis platform to measure pH in images acquired with an Olympus FV1000 MPE multiphoton laser scanning microscope in conjunction with a window chamber (WC) affixed to the dorsal skin of Severe Combined Immunodeficiency (SCID) mice inoculated with HCT116/GFP colon cancer cells. The tumor pHe gradient in the WC was compared to normal cell pHi gradient. Results: The in vitro (cell culture) pHi range is from 7.1-7.2 consistent with our previously published report, which served as a validation for this method. The in vivo (WC) study showed the extratumoral pHe is lower (6.4-6.7) comparing to the non-tumoral area (7.3); the latter was consistent with the prior report of another investigator. Conclusion: Customize a commercially available image analysis platform to measure pH is feasible. Customization of image analysis software offers valuable option for biomedical image studies regardless of size, dimensionality or complexity.
| Exploring the Effects of Innovations in Pathology on Cancer Outcomes through Discrete Event Simulation Modeling|| |
Asmaa El-Banna1, Ian Cree2, Jason Madan1
1Warwick Medical School, University of Warwick, 2University Hospitals Coventry and Warwickshire, UK. E-mail: [email protected]
Background: The study explores how modelling, in particular discrete event simulation (DES) can be used to explore the impacts of innovations in pathology services. The breast cancer pathway was modelled and effects on patient outcomes explored. The DES model can also be used to determine the cost-effectiveness of these interventions. Methods: A structured literature review was carried out to guide how best to construct the breast cancer model. As well as identifying the types of modelling techniques previously used it was possible to summarise economic evaluations in this area. A DES model was built using the Simul8 software; it has been developed to simulate the real life breast cancer pathway. It represents the processes that take place from arrival to decision making and through to treatment initiation. It is based on UK guidelines and expert opinion. Data collected from a UK hospital further informed the model in to reflect as much as possible current practice. Parameters derived from the dataset are input into the model to identify where queues arise in the real-system and therefore allowing suitable interventions to be made. Results: The DES model can be used to analyse the role of interventions in pathology such as digital image analysis. The model itself can be manipulated through sensitivity analysis to explore how modifications in the pathway such as improved accuracy and speed to diagnosis as a result of the introduction of digital pathology will change decisions and treatment. Changes in patient outcomes and costs are consequently captured. The literature review highlighted the scarcity of modelling and economic evaluations in this area. Only 28 economic studies were identified of which 14 were on the introduction of genetic assays. Only this group presented any type of modelling, cost- effectiveness analysis or sensitivity analysis. The remaining presented poor economic evidence. Conclusions: The DES model can be used to explore how innovations in pathology can impact cancer outcomes. This will add to the limited but growing pool of studies that use modelling techniques to investigate pathology interventions. Specifically this is the first DES model to explore innovations in cancer pathology services.
| The Role of the Pathologist in the Coming Era of High Throughput Quantitative Technologies|| |
1Computational Pathology, 3Scan, California, USA. E-mail: [email protected]
Background: Since the mid-nineteenth century, light microscopy has remained the quintessential tool for the morphologic analysis of histologic specimens of human tissues. In essence, contemporary pathologists rely on low-throughput and antiquated techniques, which create nominal data sets (i.e. two-dimensional sub-micron tissue sections) that drive their clinical assessments. In the non-clinical domain, novel high throughput quantitative technologies are already being utilized for the generation of large quantitative data sets of architectural and cellular morphology. Herein, we examine HT quantitative technology through a clinical lens. Methods: A review of the existing surgical pathology workflow, for the majority of clinical laboratories, is presented. Next, a single HT quantitative histopathologic technology, knife-edge scanning microscopy (KESM), is reviewed and in order to assess it's potential in the alteration of the traditional workflow components. Finally, we extrapolate the potential implications of KESM to other emerging HT quantitative technologies. Results: Traditional surgical pathology workflow has remained principally unchanged since the mid-nineteenth century: (1) Specimens received by pathology are examined with the naked eye, (2) grossed surgically, (3) sent to histology labs for slide preparation, (4) glass slides are examined using a light microscope, and in some cases, (5) slides are digitized using whole-slide scanners. KESM combines robotics, rapid serial sectioning, and advanced imaging into a single platform for simultaneous tissue sectioning and imaging. Using KESM, steps (2) through (5) could be combined into a single process, thus enabling the generation of large, quantitative, histopathology data sets. Conclusions: Emerging HT technologies are currently used in the non-clinical domain for the creation of large quantitative data sets of architectural and cellular morphology. Many of these HT technologies, including KESM, can also be utilized in the clinical domain. Doing so will allow pathologists to integrate disparate and potentially novel data sets, from different modalities.
| Task-related Visual Patterns on Digital Pathology Media|| |
Sharon E. Fox1, C. Charles Law2, Beverly E. Faulkner-Jones1
1Beth Israel Deaconess Medical Center, 2Kitware, Inc., New York, USA. E-mail: [email protected]
Background: Digital pathology has been demonstrated to be a useful tool for both education and clinical diagnosis. In order to optimize digital platforms for these purposes, it is important to understand the visual processes by which pathologists arrive at image-based diagnoses. Several studies have examined medical visual expertise during search-based tasks, as well as image identification and interpretation. The factors which affect each of these diagnostic scenarios can be applied to the use of digital pathology in the clinical setting. Methods: Pathologists and trainees were recruited to view a variety of digital pathology media, including static digital images, whole slide images (WSIs), and multi- image views, with specimen presentation in both search-related screening tests and in image identification tasks. Slides were digitized on a Philips UFS scanner and the WSIs then viewed on SlideAtlas (https://slide-atlas.org), a high-performance web-based digital pathology system. Participants were instructed to explore the images in each format and render a diagnosis. For some images, participants were given a diagnosis, and instructed to assess a diagnostic feature within the image. A Tobii X2-60 eyetracker was used to collect gaze pattern data throughout the course of the experiment. The eye-tracking data acquired and analyzed included fixation count and duration, patterns of eye gaze and slide movements. Results: Gaze patterns during diagnostic exploration of digital media tend to fall into recognizable categories of search-related and holistic or image-recognition patterns. These patterns are optimized by training, and visual expertise therefore increases with experience. Patterns of search appear to be related to the type of diagnostic question (e.g. screening task versus diagnosis of a discrete lesion), and not to a specific digital media format. Conclusions: Eye- movements and their meaning have long been the subject of scientific study, and recent advances in technology have allowed for increasingly quantitative study of eye gaze during the acquisition of complex forms of visual expertise. Here we examine our ability to categorize diagnostic processes on digital media into subtypes of expertise-related task. The results of these studies may be useful in the optimization of digital media for diagnostic performance.
| The Benefits of Digital Integration of Radiology and Pathology|| |
1Director of Pathology Informatics, Medical College of Wisconsin, Milwaukee, WI, USA. E-mail: [email protected]
Pathology and Radiology are the backbone of cancer diagnostics, yet currently operate in “silos” throughout most of the country. Recently, a handful of healthcare systems have published papers describing their efforts in digitally integrating reporting and operational workflow between radiology and pathology. As Digital Pathology gains greater importance in clinical operations, new potential benefits occur in diagnostic work due to the digital integration of radiology and pathology. The goal of this presentation is to give an up to date picture of the landscape of Radiology/Pathology integration, as well as illustrate possible benefits to and from the use of Digital Pathology in diagnostic work.
| High-throughput Analysis of Biomarkers Using Machine Learning|| |
Niels Grabe1,2,3,4, Bernd Lahrmann3, Lilja Aprupe2,5,6, Geert Litjens2,3, Jeroen van der Laak7, Nicolas Wentzensen8
11Department of Medical Oncology National Center for Tumor Disease NCT, Heidelberg, Germany, 2Hamamatsu Tissue Imaging and Analysis Center, University Heidelberg, 3Steinbeis Center for Medical Systems Biology, Heidelberg, Germany, 4Institute of Pathology, University Hospital Heidelberg, 5German Cancer Research Center (DKFZ), Germany, 6German Cancer Consortium, 7Radboud MC, University Hospital Nijmegen, The Netherlands, 8Division of Epidemiology, National Cancer Institute, USA, E-mail: [email protected]
Image analysis is for a long time widely considered to realize the great potential of digital pathology in diagnostic application, pharmaceutical research and clinical trials. It is thus firstly discussed how the conventional image analysis methods are limiting this proposition through multiple challenges. In out view, image analysis in digital pathology will evolve towards ready-to-use, high-throughput digital image assays with full automation. Such digital assays could then be validated in large clinical trials in blinded settings with standardized performance. We present examples and strategies for addressing this challenge from the areas of diagnostic pathology and immune-oncology in clinical trials. Firstly, we present as an example of automation in diagnostic pathology, our fully automatic digital analysis of a cervical cancer screening trial performed by the US National Cancer Institute based on the cytological analysis kit Roche MTM CinTec Plus p16/Ki-67 discriminating between subsequent treatment options. So far evaluation of this test relies on manual evaluation. On 341 patients our automatic test slightly even outperformed manual reading. To our knowledge this is the so far most comprehensive validation of a digital pathology assay. As a second strategy we discuss how for immuno-oncology, deep learning based image analysis techniques, devoid of the need to detect specific features, can be used to perform fully automatic biomarker analysis on the IHC of solid tumors. Two application cases are presented in connection with automatic slide registration. We show at the example of prostate adenocarcinoma how on the basis of multiplex IHC stains prostate carcinoma can be reliably detected automatically. Deep learning based Image analysis of immune cells can then reliably detect immune cells even in complex multi-stain histology. For supporting immuno- therapy of lung, we demonstrate how deep learning allows to distinguish different growth patterns in lung adenocarcinoma morphology with precise micro-segmentation. We further show how deep learning allows to overcome the problem of anthracotic pigment artefacts, so far hindering the automatic analysis of lung cancer. Taken together we present an overview on the many challenges and our strategies towards fully automatic analysis of biomarkers cancers in diagnostic pathology and immune-oncological patient profiling.
| Telepathology Network in Ile de France: An 18-month Experiment Project for Frozen Sections (Telediagnosis) and Second Opinion Diagnosis (Teleexpertise)|| |
1Department of Pathology, University Paris Sud Xi- Faculty of Medicine, France. E-mail: [email protected]
Telepathology network in Ile de France: a 18-month experiment project for frozen sections (telediagnosis) and second opinion diagnosis (teleexpertise). Because of the policy of health structure restructuration in France, the shortage of pathologists and the increase in practicing pathologist workload, there is a real need for new medical organization in pathology. The Health Agency for Ile de France provided financial support to implement an experiment project of a regional Telepathology Network for 18 months (june 2014 - december 2015). The aim of the experiment is to prove the feasibility of Telepathology in the setting of a non dedicated information network. The network includes 17 Pathology structures (11 from academic hospitals of Assistance Publique-Hôpitaux de Paris, 5 from general hospitals and 1 private) and 3 Hospitals without Pathology lab. The project covers activities of telediagnosis for frozen sections (3 binomials) and teleexpertise for second opinion diagnosis (all structures) through a centralized telepathology platform. 747 intraoperative telediagnosis on frozen sections and 262 second opinion diagnosis have been performed from june 2014 to december 2015. The presentation will describe workflows and provide quantitative and qualitative results of this project.
| Connecting the Dots: Image Analysis Solutions for RNA In situ Hybridization|| |
1Biogen, Cambridge, Massachusetts, USA. E-mail: [email protected]
Although immunohistochemistry (IHC) is still considered the gold standard for analyzing biomarkers within the tissue context, the lack of available antibodies or proper validation thereof often limits the usefulness in research and clinic. Recent improvements in in RNA situ hybridization (RNA-ISH) techniques provide and exciting and viable alternative to study the morphological landscape of gene expression. In this context, RNA-ISH can be used as a stand-alone technology or to validate IHC results. Translating RNA-ISH results into accurate quantitative data and visually appealing information, however, is a challenging task for image analysis. This presentation offers ideas and concepts on how to avoid potential pitfalls, create innovative analysis tools and generally make the most out of your RNA-ISH experiment.
| Laboratory Workflow Considerations When Implementing Digital Pathology|| |
Douglas J. Hartman1
1Division of Pathology Informatics, University of Pittsburgh Medical Center, Pittsburgh, USA. E-mail: [email protected]
Background: Digital pathology can have a positive and negative impact on workflow. We have begun implementing digital pathology for routine diagnostic work at our institution. Our implementation process within the pathology laboratory environment required significant adjustments to workflow. The aim of this presentation is to review the workflow changes we executed and to point out how they simultaneously introduced safer and better quality practices for the histology lab. Methods: A digital pathology solution was implemented at our institution comprised of both academic and community hospitals. Digital pathology was deployed to satisfy routine clinical diagnostic work, research studies, and educational use cases. Results: Digital workflow can be broken down into pre-processing and post-processing phases. Pre-processing changes include barcoding slides, strategic placement of tissue sections on slides, and rapid tissue processing. Post-processing changes include case selection for digitization, interfacing the lab information system with the digital pathology system, and the layout of signout rooms to handle digital cases. Conclusions: Many changes need to be made within the laboratory besides the purchase of hardware and software to maximize implementation of digital pathology for routine pathology work. Addressing key pre- and post-processing changes should be part of a successful implementation.
| Fluorescent and Brightfield Imaging in Biomarker Development|| |
Tyna Hope1, Dan Wang1, Sharon Nofech-Mozes2, Kela Liu1, Sireesha Kaanu-malle3, Yousef Al-Kohafi3, Kashan Shaikh3, Robert Filkins3, Martin Yaffe1,4
1Biomarker Imaging Research Laboratory, Physical Sciences, Sunnybrook Research Institute, Toronto, Canada, 2Department of Anatomic Pathology, Sunnybrook Health Science Centre, Toronto, Canada, 3GE Global Research Center, Niskayuna, USA, 4Department of Medical Biophysics, University of Toronto, Toronto, Canada. E-mail: [email protected]
Background: The Biomarker Imaging Research Laboratory (BIRL) works with collaborators to develop new biomarker panels for diagnosis, prognosis, and as predictive markers for treatment planning. Through this work we have used fluorescent and brightfield tissue imaging methods and image analysis tools. We will discuss some issues faced in obtaining the images, deriving quantitative information and validating the signals obtained. Methods: General Electric Global Research Center has a platform that allows for multiplexing biomarker signals on individual cells through a sequential stain and bleach process of fluorescent markers directly conjugated to antibodies. The platform allows for staining, imaging and extracting segmented cellular compartment information. It is capable of allowing multiplexing of many signals, we have successfully stained with 8 to date. BIRL has used this platform to investigate a four biomarker panel, ER, PR, HER2 and Ki67 on invasive breast cancer. Given that assessing fluorescent images is not the norm for collaborating pathologists, we validated the fluorescent signals against traditional IHC bright field images. Since clinicians are less familiar with fluorescent images, we are also investigating the ability to multiplex biomarker signals in brightfield images. If we can identify a subset of a panel, derived through the fluorescent experimentation, we may be able to translate this to more traditional IHC methods for rapid translation to the clinic. BIRL has a Perkin Elmer Nuance multispectral camera and we are investigating non-traditional chromogens that may allow IHC-based multispectral imaging to provide comparable information to the fluorescent presentation. Results: As a case study, the work on the validation of the fluorescent signals for a DCIS panel will be presented, including the assessment methods and feedback from the pathologists on the acceptability of working with this unfamiliar imaging technique. We will share the quantitative methods explored, the difficulties in developing methods in a cutting edge research areas, and the lessons learned. Conclusions: Fluorescent and brightfield image analysis methods together can contribute to the development of new biomarker panels and given the intense interest in precision therapy for cancer, multiplexed imaging in pathology could become a rapidly-growing area of activity in the near future.
| Food and Drug Administration Regulation of Digital Pathology|| |
1Division of Molecular Genetics and Pathology, Office of In Vitro Diagnostics and Radiological Health, Center for Devices and Radiological Health, Food and Drug Administration, Silver Spring, MD. E-mail: [email protected]
FDA has cleared or approved a number of digital pathology devices for different intended uses, such as digital read and image analysis of immunohistochemically stained tissue slides, classification and enumeration of peripheral blood cells, and initial screening of cervical cytology slides. FDA has been actively engaged in discussions with various stakeholders regarding whole slide imaging systems intended for primary diagnosis of H&E stained slides. To ensure that the images produced from H&E stained slides are safe and effective for use as an aid in primary diagnosis of cancer, FDA has maintained that WSI manufacturers should assess technical performance of the components in the imaging chain, and demonstrate analytical and clinical performance using a sufficient number of specimens and readers representative of the intended use populations. FDA has provided WSI manufacturers with outlines of various validation studies and performance measures. This presentation is intended to provide an overview of regulatory challenges and opportunities for WSI.
| Practical Implementation of Digital Pathology in Dermatopathology Practice|| |
Michael N. Kent1,2, John C. Moad1,2, Mary Jo Kerns1,2, Sean Stephenson1, Michael Murchland1, Michael P. Conroy1, Katherine Tesno1, Isaac Huff1, Theresa Feeser1, Thomas G. Olsen1,2
1Laboratory Manager, Dermatopathology Laboratory of Central States, 2Dermatology Department, Boonshoft School of Medicine, Wright State University, Dayton, OH, USA. E-mail: [email protected]
Background: Digital pathology is poised to transform pathology in ways that have not been experienced since the light microscope was introduced. However, the practical implementation of digital pathology in the clinical setting poses complex challenges to achieve a patient- centered, cost-effective solution. Our solution must provide: (1) Seamless workflow via the Laboratory Information System (LIS), from ordering through reporting, archiving and retrieval of records. (2) Hardware and software to digitally manage specimens, and associated data, to increase efficiency and reduce errors compared to manual processes and conventional microscopy. (3) Intra and inter-laboratory workflow methods that allow for technical preparation and provision of digital images and data, including reporting capabilities, to remote dermatologists choosing to interpret their own cases. Methods: We created and implemented an in-house LIS platform (Clearpath™) that accommodates glass (GS) and Whole Slide Images (WSI) in a high- volume independent university-affiliated dermatopathology laboratory with 7 board certified dermatopathologists. Our solution seamlessly manages specimen workflow and data storage in a Cloud environment with reporting and EHR interface. Three dermatopathologists established consensus diagnosis of 499 previously diagnosed skin biopsies from cases reflecting the spectrum and complexity encountered during routine practice. Fifteen melanoma cases were added to ensure this diagnosis was not underweighted. Each case was interpreted by GS or WSI by one of 3 different dermatopathologists, followed by a 30 day washout. Cases were interpreted by the same dermatopathologist by the alternate modality (GS or WSI) and compared with both the consensus and intra-observer findings. We also compared pathologist reading rates (time/case % change) to determine effect on efficiency and effect of the transition from paper to digital in the lab workflow. Results and Conclusions: Digital pathology is a viable component and enhancement to an efficient dermatopathology workflow. Mean intra-observer GS to WSI agreement among 3 pathologists was 93%, while mean agreement with consensus diagnosis was 92% (GS) and 91% (WSI). Minor GS/WSI discrepancies accounted for 6% of disagreements. GS and WSI read rates were comparable commensurate with experience. Additional laboratory workflow gains demonstrate that implementing digital pathology in a high volume specialty pathology laboratory provides a viable future path.
| Microscopy with UV Surface Excitation: Novel, Fast, Simple Microscopy Method for Slide-Free Histology|| |
Farzad Fereidouni1, Zachary Harmany1, Miao Tian1, Mirna Lechpammer1, Richard Levenson1
1Department of Pathology and Laboratory Medicine, UC Davis Medical Center, UC Davis, USA. E-mail: [email protected]
Conventional brightfield or fluorescence microscopy requires thin specimens mounted on glass slides, but these demand hours of processing and the help of skilled personnel. MUSE (Microscopy with UV Surface Excitation) is a new, inexpensive approach that can generate highquality histology and histopathology images directly from cut surfaces of fresh (or fixed) samples of any thickness, within 1-2 minutes, without slides. The method is non-destructive, and eliminates need for fixation, paraffin-embedding, or thin-sectioning. Methods: Ultraviolet light at 280 nm excites the surface layer of tissue briefly stained with common fluorescent dyes, which emit signals in the visible. Images are captured using conventional microscope optics and a standard color camera. Results: High-quality MUSE images can be captured from normal and diseased tissues. These images can faithfully resemble those from conventional histology, but they also contain information including surface shading and depth cues that reveal surface profiles important in understanding 3D organization of complex specimens. Some of the resulting images can resemble those obtainable with scanning electron microscopy (SEM). It is possible to image arbitrarily large pieces of tissue even larger than will fit on conventional slidesso virtual slide imaging of extended specimens is feasible. Conclusion: We have developed a new form of optical microscopy that generates diagnostic-quality histological images, with enhanced content, from fresh or fixed, but unsectioned tissue, rapidly, with high resolution, simply and inexpensively. We anticipate adoption in research and clinical settings, and in high- as well as low-resource environments.
| WUPax: Valuing Diversity|| |
John D. Pfeifer1
1Department of Pathology, Washington University School of Medicine, St. Louis, Missouri, USA. E-mail: [email protected]
Whole slide imaging has been used for almost a decade to support telepathology and archiving of tissue sections. However, three logistical issues continue to hinder widespread adoption of WSI-based diagnosis in routine patient care, specifically the lack of integration between vendor platform and image formats, lack of pinned patient information and scalable reporting forms, and the lack scalable software architecture that is regulatory compliant. In order to address all three issues, we developed WUPax, a web-based telepathology platform for surgical pathology diagnosis which provides outside institutions with the ability to submit orders, upload WSI's with pertinent patient history and demographic information, track the status of cases, and generate a PDF of the surgical pathology report in a HIPAA compliant and CAP validated infrastructure. WUPax is integrated into the laboratory LIS, fully searchable, and designed to be viewer agnostic. WUPax has been implemented in several routine patient care settings, locally and remotely, most notably in support of telepathology-based real-time surgical pathology diagnosis of donor organ biopsies from remote sites. WUPax is a plug-and-play solution for laboratories that desire out-of-the box functionality with a wide variety of WSI platforms and clients.
| Multimodal Microscopy Improves Precision of Pathological Digital Image Analysis and Interpretation|| |
1Associate Director of the Microscopy Core Facility at the Institute for Genomic Biology, University of Illinois at Urbana-Champaign, Champaign, Illinois, USA. E-mail: [email protected]
Optical brightfield imaging in color is a hallmark microscopy technique in identifying or diagnosing a pathological onset of disease in digital images from biopsies and so on. In this talk focus is made to improve the precision of such diagnosis using additional microscopy modality such as fluorescence techniques in addition to the brightfield optical techniques. I use normal chicken muscle and muscle experienced an anomaly in muscle fiber shape (circular fibers) labeled with Mason's Trichrome staining. From the high resolution scanned images from four micron thick muscle sections in both brightfield and fluorescence modality and merged images of two modalities were compared against regular single modality brightfield images using a 2D automated morphometric image analysis program. The objective is to count the number of circular muscle fibers with increased precision in normal and hardened muscle phenotypes in digitally scanned images using multimodal techniques. Data were acquired with multiple modalities using two different optical systems and the advantages in ease, flexibility and range of analytical options are discussed. Results are further discussed in light of the advantages of adding an extra microscopy modality for improved feature identification which might help in diagnosis based on digital pathological biopsies with increased accuracy.
| Implementing and Integrating Digital Pathology and Whole Slide Imaging Across the Enterprise|| |
J. Mark Tuthill1
1Division Head of Pathology Informatics, Henry Ford Health System, Michigan, USA. E-mail: [email protected]
Background: Use of digital pathology and whole slide imaging technology solutions is already a critically required component of modern pathology practice. To be successful, such technical solutions must be integrated and harmonized across a given laboratory's environment. The presenter will discuss an array of digital imaging technologies used across the Pathology and Laboratory Medicine service line at Henry Ford Health System to support diagnostic pathology, operational workflow and reporting. This enterprise system is used across a service area covering 2800 comprising 5 hospitals and 32 medical centers. Methods: We integrated a wide variety of digital pathology solutions (Grossing, Milestone; Xrays, Faxitron; Camera on stick based systems, Multiple vendors; immunofluorescence, Olympus; Document Scanning, Scantron), with our LIS (Sunquest CoPath 6.3.1) using an enterprise media management solution (Apollo EPMM® v9.4.3) We also worked with our whole slide imaging vendor partners (Mikroscan, D2; Ventana-Roche, iScan Coreo, iScan HT) to create a suite of integrated solutions that leverage standardized pathology work for whole slide imaging. Key to such a solution is integration of bar code labeling that can be used on any platform using labels generated from the LIS, eliminated relabeling. Database servers were deployed in our data center with client software installed on workstations including histology, pathologists offices, grossing areas, and autopsy. Results: All digital pathology solutions deployed across the Pathology and Laboratory Medicine service line support digital documentation, education and training, clinical communication, and diagnostic support for intraoperative and collaborative case consultation. Whole slide imaging technology has recently been adopted for consultation replacing prior technology based on robotic tele-microscopy. Whole slide imaging technology has been integrated with LIS and EPMM to allow efficient case management and a uniform labeling process, using HL7 interfaces and integrated databases. Conclusions: Digital pathology solutions, including whole slide imaging, are becoming increasingly import to support diagnostic pathology workflow. A critical component to successful deployment integration of such systems is harmonization of the workflow between them. With the appropriate selection of implementation of technology this can be readily achieved resulting in a scalable, effective design that has direct impact on pathology workflow efficiency and patient care.
| Deep Learning Based Cancer Metastases Detection|| |
Dayong Wang1, Aditya Khosla2, Rishab Gargeya1, Humayun Irshad1, Andrew Beck1
1Harvard-Beth Israel Deaconess Medical Center, Boston, 2MIT Computer Science and Artificial Intelligence Laboratory, Cambridge, MA, USA E-mail: [email protected]
Background: In general, the cancer metastases detection in lymph node takes a long time for pathologists since they need to exam a huge whole slide image block by block. In this project, we aim to tackle this challenge by adopting the emerging deep learning technique, which can take advantage of the increasing computational power from GPU computing and the explosion of whole slide images. Methods: We tackled the problem of automatic cancer metastases detection by integrating deep learning and traditional machine learning algorithms. We first trained a deep convolutional neural network (CNN) on ~290K randomly selected tumor and normal patches that were extracted from hundreds of training whole slide images. We then applied the trained deep model to partially overlapping patches from each WSI to create tumor prediction heatmaps. After evaluating the accuracy of the heatmaps, we extracted additional training patches from the false positive regions, and we trained a final model to create tumor prediction heatmaps using the combined training data set. For the slide-based tumor classification task, we extracted a set of geometric features from each tumor probability heatmap in the training set, and trained a random forest classifier to estimate the probability that each slide contained metastatic cancer. We then applied these models (CNN and random forest classifier) to the test images to provide a slide- based estimate of the probability of cancer metastases. For the lesion- based tumor region segmentation task, we applied a threshold of 0.90 to the tumor probability heatmaps and predicted tumor location as the center of each predicted tumor region. Results: The proposed framework is evaluated in the ISBI challenge on cancer metastasis detection in lymph node, which aims to make automated detection of metatases in H&E stained whole-slide images of lymph node sections. There are 23 submissions from all over the world from many strong industry and academic groups. We won both challenges. Conclusions: Our framework could be adopted as a pre-screen tool to reduce the labor efforts for pathologists. Deep learning and other machine learning techniques would be adopted into more computational pathology based applications in future.
| Insights from Neuroscience: Formulating Digital Pathology Solutions Based on Vision and Decision Theory|| |
1Department of Pathology and Laboratory Medicine, Drexel University College of Medicine, Philadelphia, PA, USA. E-mail: [email protected]
Background: Imaging science supports pathology workflows through the coordinated benefits of telepathology, computer-aided diagnosis, and most recently with the advent of in vivo microscopy. The diagnostic potential of these techniques promises to transform pathology into a specialty that enables rapid and accurate diagnostic support and minimally-invasive diagnostics. Despite the many recent advances in digital pathology, diagnosis from experienced pathologists remains the gold standard and adoption of many of these novel techniques has been slow. We sought to identify improvements in pathology image analysis that can readily be integrated into existing diagnostic workflows by analyzing the decision processes used by pathologists and by developing advanced image analysis models using evidence gained from these insights. Methods: We employed a multidisciplinary approach to pathology image analysis that spanned neuroscience, computer science, and pathology. This collaborative methodology enabled us to first uncover the visual and decision processes that underlie whole-slide diagnosis, and then to generate algorithms modeled after the gold standard of pathologist review. We used eye tracking, neurophysiology, and computer modeling. Results: We studied the gaze patterns of pathologists as they actively viewed whole-slide images so that we could better understand the visual factors that led to diagnoses. These data directed the development of computer vision models to identify regions in whole-slide images with prognostic potential. Computational evaluation of these regions was guided by our own neurophysiological experiments in animal models that revealed circuits in the brain responsible for segmentation via feature extraction mechanisms that relied on feedback networks. We developed models to mimic the hierarchical nature of these brain circuits to parse H&E images at multiple scales in order to capture fine-scale nuclear and stromal features as well as broader information associated with tumor architecture. Finally, using models based on neuronal decision theory, we developed algorithms to predict breast cancer attributes from the extracted features. Conclusions: These results offer a new perspective on whole-slide image analysis that can reveal patterns that mimic the information used by pathologists. The intersection of neuroscience and pathology can therefore serve as the means to improve feature extraction and decision/classification tools for pathology diagnosis and prediction.
| Highly Multiplexed Immunofluorescence Using Spectral Imaging|| |
Meyke Hermsen1, Irene Otte1, Dagmar Verweij1, Maschenka Balkenhol1, Jeroen van der Laak1
1Department of Pathology, Radboud University Nijmegen Medical Centre, Nijmegen, The Netherlands. E-mail: [email protected]
Research into cancer biomarkers often comprises testing of a number of potentially relevant markers in tissue sections. Use of single antibody immunohistochemistry, however, limits the amount of information available from the analysis. One single component can be visualized at a time, prohibiting the simultaneous assessment of multiple markers. Immunofluorescence is a widely used alternative, enabling two or three simultaneous markers. In, for instance, the study of tumor infiltrating lymphocytes one may wish to use multiplex immunophenotyping to identify the relevant subtypes of T-cells. This may require more than three markers in a single section. Also, because of availability, it is not always possible to compile panels of markers raised in different species to prevent crossreactivity. This workshop will focus on highly multiplexed imaging, allowing up to 7 markers in a single section. The method comprises a staining procedure consisting of consecutive steps of IHC staining with tyramid signal amplification and microwave stripping of the antibody. Every antibody is labeled with a different fluorescent dye. Subsequent imaging is performed using a fully automated multispectral imaging setup. This approach enables the use of antibodies raised in the same species (e.g. mouse MAbs) and is capable of eliminating the effect of autofluorescence. The workshop will consist of two parts. First the specific staining techniques will be treated. In the second part we will focus on the imaging and analysis options. There will be ample opportunities for questions and discussion.
| Implementation and Validation of Whole Slide Imaging|| |
Doug Hartman1, Michael N. Kent2, Anil Parwani3
1University of Pittsburgh Medical Center, Pittsburgh, 2Laboratory Manager, Dermatopathology Laboratory of Central States, Dayton, 3Ohio State University, Wexner Medical Center, Columbus, USA. E-mail: [email protected]
As Digital Pathology is adopted in the routine laboratory practice, more and more labs are looking for effective solutions for implementing and validating whole slide imaging (WSI). In addition to selecting a system as well as validation, labs need to be aware of the necessary quality control parameters that will need to be developed and standardized. This session will discuss practical issues that arise when selecting a WSI system, implementing and validating them. The guidelines of CLIA, CAP and other regulatory agencies will be discussed. Three key aspects will be reviewed in this workshop: selection, implementation and validation of the WSI systems. Two early adopters will present their experience with WSI implementation in detail including an academic hospital network (UPMC) as well as a private subspecialty lab (DLCS). The work shop presenter will share experiences, challenges and lessons from the independent lab perspective as well as a busy academic pathology practice.
| PD-L1 Workshop|| |
| Part 1|| |
Virginia Burns1, Lidija Pestic-Dragovich2, David Rimm3
Bristol Myers Squibb, New York, 2Roche Tissue Diagnostics, 3Yale University, USA.
| Part 2|| |
Carine El-Sissy1, Mike Montalto2, Abhi Gholap3, Joe Krueger4, Amanda Lowe5
1Hopital Europeen Goerges Pompidou, INSERM, 2Bristol Myers Squibb, New York, 3OptraSCAN, Sunnyvale, CA, 4Flagship Biosciences, Westminster, CO, USA, 5Visiopharm.
Immune checkpoint modulators are rapidly changing the way cancer is treated. Biomarkers that can predict response to this class of drugs are becoming increasingly important in current pathology practice. Specifically, several PD-L1 immunohistochemistry (IHC) assays are approved by the FDA as companion and complimentary diagnostics for various immune checkpoint inhibitors. This workshop will examine the state of PD-L1 IHC assay use for determining therapeutic eligibility with emphasis on its robustness, inter-reader variability and inter-assay correlations. The results of the NCCN/BMS Assay Comparison study will be reported. Representatives from the diagnostic and pharmaceutical industry will discuss their respective views on complimentary/companion test development and how this relates to PD-L1 testing. Further, the potential of PD-L1 automated image analysis algorithms to augment PD-L1 IHC testing in the clinic will be addressed by various image analysis experts from academia and industry. Finally, an open forum discussion will be held to discuss the pros/cons of automated imagen analysis in PD-L1 testing.
| Successful Implementation of Digital Pathology Solutions for Clinical Use|| |
Junya Fukuoka1, Doug Hartman2, Joseph Sirintrapun3, Anil Parwani4
Department of Pathology, Nagasaki University Hospital and Nagasaki University, Japan, 2University of Pittsburgh Medical Center, Pittsburgh, 3Memorial Sloan Kettering Cancer Center, NY, 4Ohio State University, Wexner Medical Center, Columbus, OH, USA. E-mail: [email protected]
Whole slide imaging (WSI) is accelerating in its potential to provide powerful solutions for enhanced all-digital anatomic pathology workflow including the use of these tools for clinical use including on site fine needle consultations, remote second opinion consultations as well as education of future pathologists. Deployment of these tools for clinical use may require understanding of the regulatory as well as technology issues. The process of selecting and implementing a WSI platform/solution for consultations may be a daunting task in settings where pathologists involved in the selection process have insufficient technical background and operational experience. This workshop is focused on assembling a collection of current best practices, which are based upon the broad experiential input from actual first-adopters who are continually building solutions for their end users and finding ways to unleash the power of digital pathology tools.
| Use Cases for Digital Pathology: How to Save on Expenses and Build Your Portfolio|| |
Renee J. Slaw1, A. Scott Mackie1, Lisa Stephens1, Michelle Bower1, Thomas W. Bauer1
1Digital Pathology, Cleveland Clinic, Ohio, USA. E-mail: [email protected]
Digital whole slide scanners can be expensive, and even more so when one considers the associated high speed IT network connections, terabytes of storage, and committed employees that are required for optimum quality. Recognizing applications within the laboratory that might benefit from the use of whole slide imaging can help justify those costs. Methods: The Center for ePathology at Cleveland Clinic has been in place for over six years. Although we initially focused on the use of WSI for education and risk management, we quickly recognized the ultimate value of WSI for diagnostic use. Knowing that implementing WSI for primary diagnosis would take time, we have identified additional applications for WSI to help justify costs as diagnostic use is gradually validated and brought into practice. Results: Examples of cost-savings and “value-added” opportunities include: Stain controls, educational recuts, risk management, legal cases, training, frozen sections, stain validations, digital consultations, research collaborations, colleague review, tissue banking, competencies and more. At this time we offer nearly 30 different use cases which have saved more than $40,000 a year. Discussion: The use of WSI for research, education, and patient care offers many advantages, but installation of a WSI system is associated with high start-up and maintenance costs. Recognizing opportunities for potential savings along with additional enhancements to clinical support, research and education can help justify those costs.
| Doctor, We Shrunk the Cost (and the Tissue)! How Digitizing Control Slides Led to Cost Savings and Tissue Size Reduction|| |
Renee J. Slaw1, Lisa Stephens1, Linda McDonald1, Amy Posch1, A. Scott Mackie1, Gary Vitanye1, Thomas W. Bauer1
1Digital Pathology, Cleveland Clinic, Ohio, USA. E-mail: [email protected]
Background: Pathologists need to review special stain controls. The Cleveland Clinic System consists of a main campus hospital and 8 regional sites. All tissue processing is now at our Main Campus, but pathologists interpret cases in several different locations. We found that implementing whole slide imaging (WSI) for Her2 quantitative batch control (QBC) and special stain controls lowered cost and improved efficiency compared to duplicate control microscope slides. Over time we realized that the size of tissue used for controls was larger than necessary and hypothesized that smaller tissue would use less time and less storage thus saving money. Methods: In February 2015 we were scanning all Her2 QBC and all special stain controls. Pathology staff are alerted by e-mail when scanned controls are available for remote viewing. We compared the calculated cost of necessary duplicate glass controls with the costs of obtaining WSI. After 15 months we reviewed file sizes of the most commonly scanned special stains and saw file sizes up to 617mb. Smaller control tissue blocks were created, scanned and reviewed. Results: The cost of the Her2 controls via scanning versus cost of a glass Her2 control slide extrapolated to a total savings of $11,000 or 85%. The cost for glass special stains ordered from 8 different sites excluding slide distribution would have been approximately $14,000. The change to scanning yielded a savings of $9,140 or 60.4%. We project a combined annual savings of $59,000 in consumables, reagents and labor. Control tissue size was reduced by an average of 74% and we expect storage savings of at least $3,500 annually. Discussion: The use of WSI offers many advantages, but installation has high start-up and maintenance costs. Recognizing opportunities for potential savings can help justify those costs. Since sharing controls using WSI, we have achieved a savings of approximately $4,900 per month. The conversion has been strongly supported by staff, in part because multiple users can attain access to the digital image from any location more easily than to the microscope slide. The smaller control tissue has also been well received and has initiated additional changes in the histology lab.
| Development of a Cloud-based Histology Database for Collaborative Cancer Research|| |
1Histowiz, Inc., Brooklyn, NY, USA. E-mail: [email protected]
Background: The development of Whole Slide Imaging (WSI) technology has allowed for the digitization of histology data, but there is a pressing need to build a scalable and cost effective IT infrastructure to archive and manage the multiterabyte databases for archiving and data sharing. Development of a virtual slide database on a scalable IT infrastructure will improve histology data viewing and searching resulting in enhanced global scientific collaboration, allowing cancer to be fought cooperatively instead of individually. Design: The main repository for WizBase is a relational database powered by Amazon Web Services with online viewing, collaboration and long term archiving of WSIs. We will also create ontology parameters and tagging approach using 3 types metadata: Type I, classification parameters, specimen information and experimental details captured at order submission; Type II, ontology terms, annotation added upon visualization of the slide by HistoWiz and slide owners; and Type III, crowd sourced tags and collaborative comments not requiring formality or standardization. Results: We have developed a robust IT infrastructure allowing for viewing, tagging and searching of histology images. HistoWiz viewer is the first Cloud based viewer that allows users to instantly view their histology slides on any mobile device without the need to download any files, software or plugins. Furthermore, we have establish ontology parameters for the classification and tagging of digital pathology cancer images, and slide metadata are captured during online order submission. Finally, through defined field and free text tagging of the slide metadata, users can search for slides meeting specific criteria. Searches based on similarity to a particular subject slide metadata allows ranking slide relevancy. The crowdsourced database currently has >10,000 histology slides and is growing at a rate of 300% per year. Conclusion: WizBase TM is the first centralized Cloud-based WSI database for cancer histopathology with an image tagging web application to facilitate histology driven data mining. WizBase TM can be regarded as the scientific hybrid of a microscopy product with the viewing power of Google Earth combined with the search and crowdsourcing of Flickr.
| Challenges in Creating a Customized ePathology Program for Collaboration at the Centers for Disease Controls Infectious Diseases Pathology Branch|| |
J. M. Gary1, Y. Ermias2, J. M. Ritter2, W. Shieh2, S. Zaki2
1IHRC, Inc, Atlanta, GA, USA, 2IDPB, NCEZID, CDC. E-mail: [email protected]
Background: CDC's Infectious Diseases Pathology Branch (IDPB) is developing a multipronged digital pathology program to enhance its unique mission, which includes identifying and investigating emerging pathogens, conducting and supporting infectious disease research, contributing to public health work, and providing support and training to health departments and other health organizations. The need to communicate with private and public institutions, both stateside and overseas, within the framework of a secure network environment, has created unique challenges and provided an opportunity for across team cooperation.
| A Novel Approach for Automated Detection of Neutrophil Extracellular Traps|| |
Brandon Ginley1, Tiffany Emmons2, Constantin Urban3, Brahm H. Segal2,4, Pinaki Sarder1
1Department of Pathology and Anatomical Sciences, JSMBS, 2Department of Immunology, Roswell Park Cancer Institute, 4Department of Medicine, Division of Infectious Diseases, JSMBS, Buffalo, NY, USA, 3Department of Molecular Biology, Umeå University, Umeå, Sweden. E-mail: [email protected]
Background: Neutrophil extracellular trap (NET) formation is a distinct mode of cell death used by neutrophils to aid in controlling infection. NETs are composed of chromatin backbones and neutrophil granular constituents that target extracellular microbes. This process is highly antimicrobial, aids in preventing bacterial dissemination, is thrombogenic, and contributes to several other known and unknown processes. NETs have been shown to be involved in inflammatory injury, sepsis, autoimmune diseases, and cancer. Existing methods used to visually quantify NETotic versus non-NETotic neutrophils are extremely time consuming and subject to user bias. These limitations are obstacles to developing NETs as prognostic biomarkers and therapeutic targets. Methods: We have developed an automated method to quantify and categorize neutrophils and NET structures captured using a flow cytometry-imaging system known as ImageStream®. Our method first uses contrast limited adaptive histogram equalization to improve signal intensity, then thresholds at a fixed level. Structure quantification is performed on the resulting binary image by calculating region properties of the resulting foreground structures. NETs classification is performed using a support vector machine (SVM.). Results: We have applied our method to 387 isolated immunofluorescence images of neutrophils obtained using ImageStream®. Neutrophils were isolated from normal donor blood and stimulated for two hours with phorbol myristate acetate as a positive control for NET generation. The cells were then marked with DRAQ5 and analyzed on ImageStream®. Manual examination established ground-truth in these data, with 143 NETs and 244 non- NETs. Convex area and eccentricity were discovered to be the most predictive features, and NETotic structures were classified up to 2000x faster per sample than manual. Classification performance of the images was 0.97/0.96 sensitivity/specificity. Of all 387 images, ImageStream's® native classifier was unable to classify 48 ambiguous images, scoring.93/.92 sensitivity/specificity on the rest. A graphical user interface software, ExamiNET, has been developed to ease end-user application of NET quantification. Discussion: We present an automated and reproducible classification method for NETs. Our method will be extended to quantify and classify NETotic structures in vivo. Future work will also attempt to discriminate quantifiable differences between NETotic structures in the various diseases they influence.
| Automatic Labeling of Glomeruli to Aid Renal Histopathology|| |
Brandon Ginley1, John Tomaszewski1, Pinaki Sarder1
1Pathology and Anatomical Sciences, SUNY Buffalo, Buffalo, NY, USA. E-mail: [email protected]
Background: Proteinuria is the excessive loss of blood serum proteins to the urine, and is an indicator for many different structural diseases of the kidney; if unchecked, it often leads to renal failure and death. Harmful structural alteration typically occurs in the glomerulus. The existing clinical practice to diagnose a cause is examination of renal tissue under light microscopy. However, visual appraisal of glomeruli requires assessment of diverse intraglomerular compartments whose boundaries are poorly distinguished from the surrounding environment. We have developed a computational method to automatically detect the boundary of glomeruli in histology images. Establishing automated glomerular extraction enables rapid extraction of perturbations of structural distributions in glomerular micro-environment caused by pathophysiological states. Methods: Whole kidney tissue slices from healthy rat were stained with five common histopathological stains. Slices were mounted on slides and imaged as whole slide images, from which 200 rectangular sections containing glomeruli were extracted from each stain (n = 1000 total). An approximated image of each glomerulus is generated by Gaussian blurring, and Gabor texture segmentation is used on these approximations to accurately define glomerular boundaries. Gabor textural sensitivity is further boosted by averaging with a mask generated by statistical F-testing for intra-glomerular space, in addition to a spatial weighting map synthetically constructed from prior information of glomerular location. Results: Our method to label exact glomerular boundaries performs with average 0.95 ± 0.005 sensitivity and 0.83 ± 0.01 specificity as compared to manual, on all images from five stains. Of the five histological stains tested, Hematoxylin & Eosin and Congo Red displayed the highest mean sensitivities and specificities (0.97 & 0.85.) Silver staining appears to display the lowest sensitivity, and Trichrome appears to display the lowest specificity. Discussion: We aim to expand this method to the analysis of whole slide images themselves, to enable rapid digital access to glomerular structural compartments. Our computational method to label glomeruli boundary from renal microenvironment is a step towards equipping pathologists with streamlined access to quantified information of renal tissue, automatically.
| Xploring Digital Pathology Data. Web-based Data Integration for Pathology Informatics and Multisite Collaboration|| |
M. Stewart, D. McCleary, M. Taylor, P. W. Hamilton
Philips Digital Pathology, Belfast. E-mail: [email protected]
Background: Digital pathology is not just about images. In modern research, digital pathology needs to reflect the broad spectrum of data modalities that underpin tissue-based research and cellular biomarker discovery. This includes epidemiological, clinical, pathology, biomarker and molecular data, as well as the large digital images that capture the wealth of morphological information that tissue samples contain. Methods: Xplore represents the next generation of digital pathology management and analysis platforms than can transform our ability to use digital pathology in discovery. Providing essential web-based tools for scanner independent viewing, trial/project management, slide management, data management, integromics, image analysis integration, search and data exchange, and provides an essential backbone for all laboratories running digital pathology scanners. Results: Xplore now forms the key Cancer Research UK hub for the sharing and analysis of multimodal data within the UK Digital Pathology Accelerator Programme, running across a network of six Cancer Research UK Centres. Conclusions: The goal here is to drive standardisation and consistency of tissue biomarker research across collaborative Centres. The power of the Xplore platform in this context is immense and help transform digital pathology as a tool for biomarker discovery and precision medicine.
| Tissue Mark: Measuring Tumour Cell Content in Breast Cancer for Molecular Diagnostics|| |
A. Askew, D. McCleary, J. Diamond, N. Montgomery, P. W. Hamilton
Philips Digital Pathology, Belfast. E-mail: [email protected]
Background: In breast cancer, the precision of molecular tests rely on rapid tumor annotation, the precise estimation of % tumour cells from tissue sectons and in many settings macrodissection to ensure sufficiency of tumour DNA. However, visual estimation of % tumour nuclei shows gross variation between laboratories and between pathologists, potentially resulting in inaccurate molecular test results, lack of standards and impacting on research and potentially patient therapy. Methods: We have developed a framework call TissueMark for the automated analysis and identification of breast cancer from H&E stained tissue sections. Using whole slide scanning and high resolution digitisation of tissue samples, a range of tissue recognition algorithms have been developed to identify breast cancer patterns and distinguish these from non-tumor regions of the slide. This was based a large training set of over 12,000 image tiles showing the tumor and non-tumor patterns across over 300 digital slides of breast cancer. Using sophisticated learning algorithms and classifierbased analysis, the image can be digitally mapped and annotated for tumour. These annotations can be used for subsequent microdissection and tumor cell counting algorithms applied to quantitatively evaluate the % tumour cell content. In addition, independent machine vision algorithms have been developed to highlight ductal carcinoma in situ (DCIS) and distinguish this from invasive cancer. Results: Results show a classification accuracy of 94% of tumour tiles. This allows well defined tumor boundaries to be rapidly and precisely drawn on the digital slide. DCIS can be automatically identified when found adjacent to invasive cancer regions and separately analysed. Cell counting algorithms can mark-up individual tumour nuceli within identified boundaries and show strong concordance (87%) with manual nuclear counts. Conclusions: Image analysis and pattern recognition algorithms can be used to automatically and reliably identify tumour in H&E tissue sections. Quantitative identification of tumour nuclei in digital H&E images provides an objective, reproducible tumour cell percentage thresholds for molecular evaluation. Novel image analysis approaches such as this can provide objective indices of sample quality, improving the quality of molecular studies in breast cancer and other solid tumors and provides a framework for the rapid pre-analytical screening of samples for centralised molecular testing.
| Trial of Fostering Pathological Specialists using Digital Diagnostic Networking System|| |
Yukio Kashima1, Tomoo Itoh2, Junya Fukuoka3
1Nagasaki University Graduate School, 2Kobe University Hospital, 3Department of Pathology, Nagasaki University, Kobe, Japan. E-mail: [email protected]
Background: Shortage of physicians in pathology department has long been the serious problem of medical-service community in Japan. Methods: So as to improve the situation, an epoch-making project has been started by collaboration between Nagasaki Univ. and Kobe Univ. They send a candidate of pathological specialist with shallow experience of pathlological diagnosis (hereinafter referred to as “trainee” ) to a local hospital which desires to a get full-time doctor in order to make its status as the core hospital in the region. The local hospital introduces the devices for digital pathology, such as a scanner and a server with networking systems. The trainee starts working for the hospital as a fulltime job doctor. Routine diagnosis including biopsies and surgical specimen are drafted by him and finalized by specialists after discussing through digital pathology. Thus, including autopsy, CPC (clinicopathological conference), and other in-house discussions, the trainee perform the required tasks as a pathological specialist do alone in a core hospital, by getting strong backup from remote laboratories. Additionally, the trainee can have many chances to attend other laboratory's conferences, workshops, and journal clubs etc as necessary. Results: Sending a trainee and introducing digital pathology system to a local hospital can make it possible to foster the trainee through remote education and discussion, and also to provide reliable diagnoses for the local core hospital. By doing so, Laboratories as the suppliers of trainees, and local hospitals as demanders of full-time doctors, and trainees themselves can establish win-win relationship together. Conclusions: At this difficult moment for many pathologists, the utility of digital pathology is, as it were, limitless when we successfully build collaboration with each other. We would like to put digital pathology as one of the strongest strategy for fostering pathological specialists.
| CDK7: A Marker of Poor Prognosis and Therapeutic Target in Triple-negative Breast Cancer|| |
1CBT Lab, Conway Institute, University College Dublin, Dublin, Ireland. E-mail: [email protected]
Background: Triple-negative breast cancer (TNBC) represents a heterogeneous subgroup of breast cancer with substantial genotypic and phenotypic diversity. TNBC patients commonly exhibit poor prognosis and high relapse rates at early stages after chemotherapy treatment. Currently, there is a lack of biomarkers and targeted therapies for TNBC. Here, we aimed to identify novel kinase targets that may play a pivotal role in the progression of TNBC and, thus, offer new therapeutic prospects. Methods: Survival analysis was performed to investigate the association between CDK7 mRNA expression and clinical outcome of TNBC patients. CDK7 protein expression was evaluated via immunohistochemical staining of independent TNBC tissue microarray (TMA) cohorts, and survival analysis was performed. To therapeutically target CDK7, two highly specific CDK7 inhibitors BS-181 and THZ1 were tested in vitro. MTT and Annexin V/propidium iodide assays were performed to evaluate cell proliferation and apoptosis, respectively, following CDK7 inhibitor treatment. Dynamic BH3 profiling technology was utilized to evaluate survival dependency following THZ1 treatment of TNBC cells. Results: CDK7 mRNA expression was associated with poor recurrence-free survival in a public TNBC dataset (n = 383, P &360; 0.001, HR = 2.152, CI = 1.576 - 2.939) and with poor breast cancer specific survival in the METABRIC dataset (n = 217, P = 0.023, HR = 1.598, CI = 1.061 - 2.406). CDK7 protein expression was associated with poor breast cancer-specific survival in the RATHER TNBC TMA cohort (n = 109, P = 0.012, HR = 2.516, CI = 1.189-5.32) and the METABRIC TNBC TMA cohort (n = 203, P = 0.007, HR = 1.921, CI = 1.185-3.113). BS-181 and THZ1 reduced phosphorylation of RNA polymerase II, indicative of RNA transcription inhibition, and induced apoptosis in TNBC cells. The covalent CDK7 inhibitor, THZ1, demonstrated 1000- fold higher potency than BS-181. Dynamic BH3 profiling showed that THZ1 caused an increased survival dependency on BCL-2/BCL-XL. Moreover, a rational combination treatment for TNBC involving THZ1 and ABT-263/ABT-199 was identified. Conclusions: CDK7 is a promising marker of poor prognosis in TNBC. Targeting CDK7 alone or in combination with the BH3 mimetics ABT-263/ABT-199 may be a useful therapeutic strategy for TNBC.
| Automated Wholeslide Analysis of Fibroblast-activated Protein in Immunohistochemistry Simplex and Duplex Brightfield Imaging|| |
Auranuch Lorsakul1, Emilia Andersson2, Suzana Vega Harring2, Hadassah Sade2, Oliver Grimm2, Joerg Bredno1
1Ventana Medical Systems, Inc. (Roche Tissue Diagnostics), Digital Pathology, Mountain View, CA, USA, 2Pathology and Tissue Analytics, Pharma Research and Early Development, Roche Innovation Center Munich, Germany. E-mail: [email protected]
Background: Carcinoma-activated fibroblasts (CAFs) support cancer progression and are potential targets for therapy. CAFs exist in a majority of epithelial carcinomas, but automated-quantitative methods for the detection in immunohistochemistry (IHC) brightfield imaging are limited. We established an automated digital-pathology solution to quantify expression of Fibroblast-Activation-Protein (FAP) biomarkers on wholeslide analysis. Our method was verified against ground-truth provided by pathologists. Study of FAP-expressing areas on simplex- (BF-IHC-S) and duplex-stained slides (BF-IHC-D) demonstrated the ability to detect FAP presence even for brightfield-duplex assays. Methods: Fifty-nine human-tissue slides (2.5-μm-FFPE sections) of breast, colorectal, pancreas, non-small-cell-lung, and head-and-neck carcinomas were included. BF-IHC-S stained for FAP with Diaminobenzidine (DAB). BF-IHC-D stained for FAP and Pan-Cytokeratin antibody cocktails with novel purple- and yellow-detection kits, respectively. Slides were counterstained with Hematoxylin, imaged using VENTANA iScan-HT scanner, and analyzed by algorithms which: 1. Decomposed RGB image into individual biomarker based on colordeconvolution method. 2. Detected biomarkers, based on FAP intensities, using supervised-generation rule. Two pathologists determined ground truth (GT) from example images. Algorithm was trained based on GT to generate the best method to automatically identify FAP presence. 3. Determined FAP presence at high magnification and compiled wholeslide readouts to report percentage of FAP area relative to total analyzed-tissue area (FAP-ARA). Results: For training and verification of the BF-IHC-D algorithm, 58 field-of-view images (FOVs) were collected from 29 slides. Two pathologists individually determined FAP presence in each FOV. Observers agreed with R2=0.93 and FAP-ARA (mean ± sd) of 14.36 ± 13.73% and 13.55 ± 12.79%, respectively. In each FOV, algorithm results were compared to mean GT from two pathologists. Algorithm reproduced pathologists' FAP areas with R2=0.96 and FAP- ARA of 13.88 ± 15.03%. For comparison of BF-IHC-S and BF-IHC-D, consecutive sections from 15 blocks were included. Correlation of wholeslide FAP-positive areas in consecutive sections was R2 = 0.60 and mean of FAP-ARAs were 2.01 ± 3.67% and 6.94 ± 6.24% on BF-IHC-S and BF-IHC-D, respectively. Conclusions: Automated-wholeslide analysis provides quantitative readouts to support evaluation of staining protocols and cut-off determination for CAF-based-target therapies. Automated algorithms can be trained to reproduce expert assessment. Signal presence was detected with higher sensitivity using duplex-stained assays and novel chromogens in comparison to DAB-simplex-stained slides. Automated assessment on brightfield-duplex slides may be viable option for CAF-targeting therapeutically-predictive analyses.
| Unsupervised Multi-Scale Glomerular Compartmentalization in Renal Pathology|| |
Brendon Lutnick1, John Tomaszewski1, Pinaki Sarder1
1Department of Pathology and Anatomical Sciences, SUNY Buffalo, Buffalo, NY, USA. E-mail: [email protected]
Background: Clinical pathology relies on manual compartmentalization and quantification of biological structures, which is time consuming and often error-prone. Application of computer vision segmentation algorithms to histopathological image analysis, in contrast, can offer fast, reproducible, and accurate quantitative analysis to aid pathologists. Algorithms tunable to different biologically relevant structures can allow accurate, precise, and reproducible estimates of disease states. In this direction, we have developed a fast unsupervised computational method for simultaneously separating all biologically relevant structures from histopathological images in multi-scale. To establish clinical potential in renal pathology, we have extended our method to quantitatively visualize scale variant compartments of heterogeneous glomerular structures for the first time. Implications of the utility of our method extend to fields such as oncology, genomics, as well as non-biological fields. Methods: Segmentation is achieved by solving an energy optimization problem. Representing the image as a graph, nodes (pixels) are grouped by minimizing a Potts model Hamiltonian, adopted from theoretical physics, modeling interacting electron spins. Pixel relationships (modeled as edges) are used to update the energy of the partitioned graph. By iteratively improving the clustering, the optimal number of segments is revealed. Due to the large number of edges in the graph for a typical tissue histology image, a spatial distance cutoff is used to exclude edges and promote computational efficiency. This method uses only one optional resolution parameter to tune to important biological structures in varying scale. Results: To validate our method, we use histopathological glomerular images of healthy mice. The biological relevance of glomerular segmentation has been validated by resident pathologists. Differing resolution levels show segmentation of differing relevant structures, partitioning the nuclei, glomerular basement membrane, Bowman's space, mesangial matrix, and capillary lumen. Accurate segmentations of images with as many as 10^6 pixels can be completed in as little as 2 min. Discussion: We have developed a robust method for extracting image features from histopathological images with computational efficiency unparalleled by similar unsupervised methods. Our computational method is a step towards simultaneous and rapid quantification of pertinent structures in multi-scale for computational renal pathology.
| 5-year Experience of Web-based Whole Slide Imaging Primary Diagnosis of Rural Hospital in Japan|| |
Ichiro Mori1, Takashi Ozaki2, Yasuteru Muragaki2, Takatoshi Ibata3, Hiroshi Ueda2, Robert Y. Osamura1
1Department of Pathology, Mita Hospital, International University of Health and Welfare, 2Wakayama Medical University, 3Shingu Municipal Medical Center, Japan. E-mail: [email protected]
Background: Japan is short of pathologists. Most of the pathologists are aged, and concentrated in cities. Thus, only few pathologists are available in rural regions. To cover one rural hospital, we started web-based primary diagnosis system 5 years ago using WSI. Methods: Routine H&E stained pathology slides were scanned using 40x objective lens. We used LINCE (CLARO, Japan) as WSI scanner. We restricted the target to the biopsy specimens. Pathology technician stores WSI to the scanner controller PC together with application forms converted to PDF. For the security reason, patient's name is excluded from the WSI and application form. Case is identified using serial pathology number. We used Team Viewer to open software VPN, and the shared folder containing WSI data is viewed through internet from Tokyo, 500 Km from the hospital. Pathologists read application form, observe VS, make diagnosis report, and store the report to the same shared folder. Pathology technicians of the hospital copy and paste the diagnosis report to the pathology information system of the hospital. Result: We start this system on September 2011. About 4000 WSI primary diagnosis has been done without incident. Web-speed was slow and unstable in the beginning, especially in the evening, probably because the hospital was connected to the web through Cable-TV Company. After public optical line is introduced, the line speed was stabilized. Diagnosis of Helicobacter Pylori in gastric biopsy was usually difficult, but improved after introduction of 4K (3840x2160) display. Once, we were about to make serious incident. Our WSI viewer has a thumbnail image, but doesn't contain slide label area. We omit the branch number on cover glass from the scan area to prevent focus trouble. We number specimens from label side, and slides were usually set to the scanner with label right hand side. One day, slides were set opposite direction. One of 4 colon biopsy specimens was cancer, and the difference caused opposite direction of cancer location. We were lucky to find mistake before clinician read the report. Conclusions: Web-based remote diagnosis system is useful especially to the Japanese rural area where available pathologist is limited.
| Quantifying Immunohistochemical Expression of Programmed Death-1 on T-cells|| |
Yao Nie1, Emilia Andersson2
1Ventana Medical System, Inc. - Digital Pathology, Mountain View, CA, 2Roche Diagnostics GmbH, Roche Innovation Center Munich, Germany. E-mail: [email protected]
Background: The programmed death-1 (PD-1) receptor is expressed on T-cells and functions as an inhibitory receptor that down regulates the immune system. Both PD-1 and its ligand, PD-L1, are targets for cancer immunotherapy with the purpose of activating tumor-specific T-cell responses. Since expression levels of PD-1 on T-cells may dictate the efficacy of anti-PD-1/PD-L1 therapy, it is of great importance to know PD-1 expression levels in addition to the PD-1 expressing cell count. In this study, we sought to quantify the immunohistochemical expression of PD-1 on cell pellets from CD4+ T-cells. Method: Six purified CD4+ Tcell samples were exposed to tumor cell lysis to induce different levels of PD-1 expression. The cells were first analyzed by flow cytometry. In addition to cell count, PD-1 expression levels were measured by evaluating the intensity of the fluorescence emitted by labeled antibodies binding to the PD-1 receptors (i.e., binding capability). Then a cell pellet array was generated and paraffin embedded. Two slides were cut from the cell block and stained with anti- PD-1 in Diaminobenzene (DAB) followed by a hematoxylin counterstain. Digital images of the slides were obtained from whole-slide scanning, on which regions of different cell samples were annotated. Automatic image analysis was applied to extract the DAB image channel and detect the cells in each region. The median pixel intensity value within each cell was evaluated in the DAB image channel and used as the quantified measurement for PD-1 expression level. Group statistics, including the mean, standard deviation (STD) and histogram, of the measurements are reported for all the cells from the same sample. Results: Mean DAB intensity measurements for the cells from the six different cell samples are 0.62, 0.55, 0.46, 0.25, 0.18 and 0.09 on the first slide and 0.60, 0.49, 0.32, 0.46, 0.19 and 0.10, on the second slide, respectively. For both slides, the STDs of the measurements range between 0.18~ 0.35, and are generally smaller for cell samples with lower measurement means. The histograms also demonstrate the difference of measurement distribution between the different cell samples. Conclusions: The mean DAB intensity measurement on stained slides generally agrees with the corresponding antibody binding capability for each cell sample. Therefore, the level of PD-1 expression can be quantified by evaluating the immunohistochemical DAB image channel intensity value within the cells. We can use this method to evaluate PD-1 expression in stained tissues slides.
| Validation of Live Robotic Telepathology for Intraoperative Neuropathology Interpretations at a Tertiary Care Center|| |
Aisha Sethi1, Jose Otero1, Norman Lehman1, Bagirathan Somasundaram2, Patricia Goede3, Chrystal Adams3, Anil Parwani1, Lynn Schoenfield1
1Ohio State University Wexner Medical Center, Columbus, OH, 2Sakura Finetek USA, Inc., Torrance, 3XIFIN, Inc. USA. E-mail: [email protected]
Background: Digital microcopy (live or whole slide scanning) has significant utility in this era of rapid technological advancements, producing numerous advantages including remote slide review, portability, archiving, training sets which are easily accessible and shareable, for consults, tumor boards, teaching and research. The aim of our study was to validate the use of a robotic live digital microscope for reviewing of intraoperative neuropathology cases to enable remote slide interpretations within our tertiary care center. Methods: Frozen sections and touch prep/smear cytologic preparations from 60 de-identified neuropathology cases were first examined remotely using the VisionTek® M6 Digital Microscope (Sakura Finetek USA Inc.), then after 2 weeks of wash-out, reviewed using conventional microscopy. Two neuropathologists evaluated the cases and provided interpretations based on the following clinical information: basic clinical history, MRI report, specific site, age and gender. Any discrepancies between glass slide interpretations and live digital interpretations were categorized as major (significant clinical impact) or minor (no significant clinical impact). To date, a total of 42 cases have been reviewed by neuropathologist #1 and 18 cases by neuropathologist #2 using both methods. Results: There were no major discrepancies and only 2 minor discrepancies between live digital and glass slide interpretations. Conclusions: Our study validated the use of a robotic live digital microscope for remotely interpreting intraoperative neuropathology specimens at our institutions in different locations. With a concordance of 100% between both methods, live digital microcopy was equivalent to conventional microscopy for intraoperative slide interpretation of neuropathology cases.
| A Small Reduction in Tissue Size Yields a Significant Reduction in Digital Image Storage Cost|| |
Lisa Stephens1, Linda McDonald1, Scott Mackie1, Renee J. Slaw1, Thomas W. Bauer1
1Cleveland Clinic, Cleveland, Ohio, USA. E-mail: [email protected]
Background: As digital pathology is becoming more prevalent, the cost of image storage becomes factor to consider. The purpose of this study was to test the effect of reducing the tissue size of frequently scanned slides on digital whole slide image (WSI) file size and file storage cost. Methods: We investigated the file sizes of our most commonly scanned slides – the daily special stain controls. The average file size of these stains ranges from 109 mb to 476 mb. All of the Masson trichrome slides, Movat pentachrome slides and PAS slides were reviewed from 2015, and the tissue measurements and file sizes were recorded. New, smaller control tissue blocks were created. Once scanned, the tissue measurements and file sizes were also recorded. Results: Overall, we reduced the size of the tissue on the slide by an average of 74%. Conclusions: We calculate that for these three control tissues alone, $3465 in storage costs could be saved every year by reducing the size of the tissue on the slide.
| Building and Deploying Digital Pathology Infrastructure for a Heterogeneous User Base|| |
Silke Smeets1, Stijn Piessens1, Sabrina D'Haese1, Chris Groven1, Peter In't Veld1, Wim Waelput2, Ramses Forsyth2
1Department Experimental Pathology, Faculty of Medicine, Vrije Universiteit Brussel, 2University Hospital Brussels (UZ Brussel), Brussels, Belgium. E-mail: [email protected]
Background: Too often, image analysis and data/image mining projects remain stuck in micro-environments because they are limited by vendor-specific solutions that neither scale nor interact with material from other departments or institutions. Successful roll-out of digital histopathology therefore requires more than a whole slide scanner. At Brussels Free University (VUB), we wanted to provide a core digital pathology infrastructure that can support a range of different use cases Methods: The flexibility of digital pathology hardware and software solutions today allows institutions to create bespoke solutions to meet individual user needs. We built a centralized infrastructure that integrates a variety of imaging platforms (brightfield, fluorescence, multi-vendor formats). Results: On top of this diversified hardware, a hybrid solution was compiled that consists of commercial software, as well as open source packages. In order to mine these data efficiently, a top-down approach was employed to manage and integrate the various platforms. Custom coding was used to interact with various vendor-software and server applications, where needed. Conclusions: Digital pathology involves much more than the acquisition of a slide scanner. At the VUB, we have engaged five different imaging platforms onto a single architecture. We are now storing data from all modalities in a single storage facility, and can manage it through a single access point. The end-result is an interconnected network of heterogeneous scalable information silos. We currently have three main use cases for WSI: education, biobanking, and telepathology. We have about 7 terabytes of data online, and serve 50 users across different research groups and departments.
| Workshop Report: Bioinformatics Meets Digital Pathology|| |
Yves Sucaet1, Jeroen Van der Laak2, Marius Nap3, Zev Leifer4, Yukako Yagi5, Raphaël Marée6, David Ameisen7,8,9, Paul Van Diest10
1Vrije Universiteit Brussel, Belgium, 2UMC Radboud, Netherland, 3HistoGeneX, Belgium and Rigshospitalet Copenhagen, Denmark, 4New York College of Podiatric Medicine, USA, 5Harvard Medical School & Massachusetts General Hospital, USA, 6University of Liège, Belgium, 7IRIF, 8CNRS, 9Paris Diderot University, France, 10UMC Utrecht, Netherlands. E-mail: [email protected]
Background: Pathologists are the guardians of large quantities of high- resolution imaging material, that is barely noticed by bioinformatics today. Yet this (annotated) material can be of huge contributing value to any research project, as pathology preserves topological tissue features, and can help explain difficult phenomena such as tumor heterogeneity. The goal is then to bridge research methods and materials from the bioinformatics community with the (digital) pathology community. Methods: A one-day workshop was organized for the first time in the context of a major bioinformatics meeting (ECCB 2016) in Den Hague on September 3, 2016. Keynote speakers were Jeffrey Fine and Jeroen Van der Laak. Contributions to the workshop were accepted in the form of poster abstract, short oral talks, and conference proceeding papers. Results: The workshop consisted of keynotes, short oral talks, a poster session, and several networking opportunities throughout the day. This effectively facilitated bridging opportunities between the bioinformatics and tissue image analysis communities. Conclusions: Automated analysis of such large-scale datasets is challenging and their combination with omics data is not trivial. Our workshop wants to address two emerging opportunities: (1) There are real problems in (digital) microscopy today that deserve their attention and are at least as (if not more) interesting than the NGS and *seq stuff that people working on today. (2) Digital microscopy in its own right offers a new layer of data that can be added to and mixed with their current levels of *omics-datasets, and help them gain new insights into projects that they're already working on.
| Convolutional Neural Network Prescreening of Urine Cytology Whole Slide Images|| |
Kaitlin Sundling1, Suzanne Selvaggi2, Daniel Kurtycz1, Darya Buehler1
1University of Wisconsin Hospital and Clinics Department of Pathology, 2Department of Pathology and Laboratory Medicine, UW School of Medicine and Public Health, Madison, WI, USA . E-mail: [email protected]
Introduction: Urine cytology poses many diagnostic challenges in daily practice. These include identifying rare high grade urothelial carcinoma cells and separating reactive atypia from malignancy. With the recent introduction of the Paris system for reporting urinary cytology, attention has turned toward a standardized approach to urine cytology. Deep machine learning may be useful to assist cytotechnologists and pathologists in isolating rare diagnostic cells while increasing speed and reducing the number of benign cells viewed. We present a pilot study using convolutional neural networks to identify diagnostic cells on urine cytology whole slide images, utilizing the Paris system recommendations. Methods: Urine cytology slides representing Paris system categories (negative, atypical urothelial cells, low-grade urothelial neoplasia, suspicious for high-grade urothelial carcinoma and high-grade urothelial carcinoma) are scanned using an Aperio digital pathology slide scanner. The whole slide images are segmented to produce individual cells and small groups of cells (image segments) for evaluation by three pathologists. The individual segments are annotated by the pathologists based on Paris system criteria, including nucleus-to-cytoplasm ratios, cytoarchitectural features and intactness. The annotated images are then utilized to train the convolutional neural network. The whole slide images in their entirety are reviewed by three pathologists for diagnosis and categorization within the Paris system. A subset of the images are utilized to train a convolutional neural network. The convolutional neural network is then tested on the remaining whole slide images, with results including adequacy as well as the Paris system categories listed above. Agreement between neural network classification and pathologist assessment is measured using the kappa statistic weighted by diagnostic category. Results: Pathologists' diagnoses within the Paris system are compared to neural network categorization, both on image segments as well as complete whole slide images. Conclusions: This pilot study investigates the possible role of convolutional neural network analysis in identifying diagnostic cells for cytotechnologist and pathologist review, using the Paris system guidelines. In addition, we present methods for image segmentation and a training interface that may be broadly applicable to image analysis in other aspects of cytopathology.
| Multiphoton Microscopy with Clearing: Fast Histology for Primary Diagnosis in Un-embedded Specimens|| |
Eben Olson1, Michael J. Levene2, Richard Torres1
1Yale School of Medicine, Connecticut, USA, 2Applikate Technologies, LLC, WESTON, CT. E-mail: [email protected]
Background: The ability to provide fast and accurate diagnostic interpretations of tissue histology is limited by factors such as the relatively slow speed of traditional embedding and sectioning processes, the balance in small specimens between visualization and preserving tissue for ancillary studies, a restricted two-dimensional perspective, and artifacts of tissue freezing and routine slide preparation. Confocal and multiphoton microscopy (MPM) on ex vivo tissue have the potential to address these limitations but in their usual implementation suffer from low tissue penetration depth and incomplete 'in-focus' sections due to irregularities in cut tissue surfaces and high light scatter. They also do not natively produce images with the color characteristics to which pathologists are accustomed. In an effort to improve upon existing methods, we have employed a novel tissue processing method for MPM capable of producing complete high-resolution sections with contrast and resolution comparable to traditional histology. Methods: A variety of fresh human and mouse tissue specimens were optically cleared with a fast protocol using dehydration followed by clearing with total processing times between 15 minutes and 1 hour. A home-built multiphoton microscope with optics optimized for cleared-tissue imaging was used. Fluorescent dye combinations were applied during processing and collected fluorescence was digitally converted into H&E-like images using re-convolution. Results: Images obtained with the MPM-clearing approach are amenable to primary diagnosis with characteristics that faithfully reproduce optimally processed traditional histology sections under high power. Optical sectioning achieved was on the order of 1 μm, offering nuclear clarity that surpasses that of traditional sections. The capability for extended contrast and hue adjustments creates opportunities for enhanced visualization. Entire core biopsy specimens were viewable with less than 1 hour of processing, allowing complete three-dimensional evaluation without consuming tissue. Conclusions: Multiphoton microscopy with clearing is able to produce optical sections of superb quality from un-embedded specimens with minimal preparation and a fraction of the time of traditional processing, while preserving all tissue for ancillary studies. The combination of features of MPM with clearing could enable the practical implementation of digital pathology while further improving workflow and diagnostic accuracy.
| Digital Pathology in the Immune Oncology Field: A Roadmap for the Future|| |
Yannick Waumans1, Mark Kockx1, Christopher Ung1
1Campus Middelheim – Pathology, Laboratory building, HistoGeneX, Belgium. E-mail: [email protected]
Immunotherapy requires a sophisticated understanding of contextual immune information with oncology pathways, such as the relative positioning of immune to tumor cells or co-localization of biomarkers. While molecular biology can provide elaborate and copious data sets for foundational contexture, histopathology remains the indispensable method to deliver crucial morphological information paramount to characterizing the tumor microenvironment (TME). Robust digital histopathology platforms, especially when fully integrated with data-rich laboratory information management systems, are necessary to realize the full utility of histopathology, but its use is often impeded by technology-wariness, comfort-inertia with conventional optical systems, as well as the unclear regulatory positioning towards digital histopathology systems. Thus, the endeavor to create an efficient and well-validated digital histopathology system that is flexible to the fluctuations of technology innovations and continued discoveries in the TME is indeed highly complex, but worth undertaking. In this presentation, we describe such a digital histopathology system model that can be used within a clinical development context with particular emphasis for analyzing the TME that can provide delineation for immune oncology development programs. We describe a two-tier validation of the entire infrastructure combined with an intelligent compartmentalization of the digital histopathology system, using recommendations from both FDA and CAP to produce a system that can absorb changes without compromising quality practices. We also describe IT innovations that create a seamless workflow and quality-centered environment, specifically the concept of a virtual private cloud combined with innovative storage solutions and powerful software, that reduces cost, increases image availability and sets up quantitative analyses while complying with regulatory requirements. Finally, we demonstrate the use of this digital histopathology model within HistoOncoImmuneTM, our TME analysis methodology, as a way to integrate each patient's morphological and molecular information in global clinical trials which can then be correlated to patient outcomes.
| Automated Vascular Segmentation, Reconstruction and Classification on Whole-slide Histology|| |
Yiwen Xu1,2, J. Geoffrey Pickering1,2,3, Zengxuan Nong2, J. Sachi Elkerton1, Aaron Ward1,4
1Department of Medical Biophysics, 2Robarts Research Institute, 3Department of Medicine, 4Department of Oncology, Western University, London, Canada. E-mail: [email protected]
Background: Histology of the microvasculature depicts detailed characteristics relevant to tissue perfusion. One important histologic feature is the smooth muscle component of the microvessel wall, which is responsible for controlling vessel caliber. Abnormalities can cause disease and organ failure, as seen in hypertensive retinopathy, diabetic ischemia, Alzheimer's disease and improper cardiovascular development. However, assessments of smooth muscle cell content are conventionally performed on selected fields of view on 2D sections, which may lead to measurement bias. As well, microvessels are inherently 3D and include both the arterial and the venous sides. Therefore, there is a need for 3D reconstruction and separate analysis for different vessel types. We propose an automated method for 3D vascular reconstruction, detection/segmentation and classification of vascular subtypes for detection and measurement of differences between normal and regenerated vascular smooth muscle. Methods: Vessels were immunostained for smooth muscle a-actin using 3,3'-Diaminobenzidine and the nuclei stained with hematoxylin, assessing both normal (n = 6 mice) and regenerated vasculature (n = 5 mice). 2D locally adaptive segmentation involving vessel detection and fragment connection through skeletonization was reconstructed into 3D using a nucleus landmark-based registration. The vessels were categorized into arterioles and venules on 2D using supervised machine learning. Morphological measurements were performed in both 2D and 3D. Results: Vessel medial area and vessel wall thickness were found to be greater in the normal vasculature as compared to the regenerated vasculature, for both the 2D and 3D measurements (p < 0.01). Validation analyses yielded a Dice coefficient of 0.88 for the segmentations of the vascular smooth muscle, a 3D reconstruction target registration error of 4 microns, and area under the receiver operator curve of 0.89 for arteriole vs. venule vasculature discrimination. Conclusions: We developed an automated system to assess the arteriolar and venular smooth muscle distributions in the microvasculature, which allows for high throughput analysis of digital histology images. With microvascular measure visualization methodologies in 3D and automated segmentations, we are now capable of locating focal pathologies on a whole slide level using 3D histology reconstruction, and performing separate analyses on the arteriolar and venular side of the microvascular tree.
| Easy Application of 3D Histopathological Imaging|| |
Akira Yoshikawa1, Ruben Groen1, Emiko Udo1, Junya Fukuoka1
1Department of Pathology, Nagasaki University Hospital, Nagasaki, Japan. E-mail: [email protected]
Background: The utility of three-dimensional (3D) imaging has been appreciated during the application of medical imaging. Those technologies have enabled us to recognize small and complexed lesions and became a common clinical facility, especially in radiology. In histopathology, several 3D imaging modalities have also been developed and utilize for some studies. However, those technologies need a large expense, time, and skilled technologists. Here, we introduce a unique method to observe histology in 3D simply and quickly. Method: Formalin-fixed paraffin-embedded blocks of the lung were prepared. The blocks were heated at 40-50ºC in hot water for a minute. Sections of 100 to 1000 μm thick, depending on the purposes of the observation, were cut with a conventional sliding microtome. The sections were directly soaked into reagents including process of deparaffinization followed by staining. Titrations of each formulas/stains were optimized for each condition. Ordinary microscopes were used for the observation. ImageJ was used for 3D image reconstruction and analysis. Result: 3D structures of lung adenocarcinoma and emphysema cases were successfully obtained. Through the observation, we found that micropapillary clusters in lung adenocarcinoma formed 4 different structures. On the other hand, we also found a case of pseudo micropapillary by 3D observation in which floating-looking tumor cells indeed were sections of papillae brunching complexly. For cases with pulmonary emphysema, we found an increase in pores of Kohn and patterns of the lung structure destruction, which was almost impossible to recognize by plane image of the ordinary histology. Conclusion: 3D image reconstruction by thick sections is a simple, fast, and cost-saving technology to recognize 3D structure of lung diseases. We propose that this method is fairly useful for evaluation of the lung tissue of large cohorts and case-control studies.