Wednesday Scientific Sessions

Scientific Sessions
Wednesday, October 10, 2012
7:30 am – 8:55 am
 View Scientific Session Videos:
(there is no audio for the first half hour of the Acapulco Room session)
Toronto Room: Education
Scientific Imaging & Applications Development
Location: Regency Ballroom
Use of Automated Algorithms to Detect Early Acute Kidney Injury (AKI)
Jason M. Baron, MD (, Xingxing S. Cheng, MD2, Hasan Bazari, MD2, Jingyang Huang, MD1, Rosemary T. Jaromin1, Kent B. Lewandrowski, MD1,  Anand S. Dighe MD, PhD1
1Department of Pathology, Massachusetts General
    Hospital, Boston, MA
2Department of Medicine, Massachusetts General
    Hospital, Boston, MA
AKI is a common problem in hospitalized patients. Clinical criteria (e.g. RIFLE) allow a diagnosis of AKI to be made based on specified increases in blood creatinine. However, clinical laboratories only flag creatinine results exceeding the reference range and generally do not flag creatinine increases from an individual patient’s baseline. Consequently, AKI may be overlooked in patients with rising creatinine values that still fall within the reference range. In this report, we evaluate rule-based strategies, implementable within our laboratory information system that could be used to flag rising creatinine levels and facilitate early identification of AKI. 
Data was extracted from our hospital's laboratory information system and analyzed using relational database tools. 
We evaluated three different types of flagging systems. The first was a delta check, comparing the current creatinine value to the patient’s most recent value within the prior 72 hours. The second was based on comparing the patient’s current creatinine value to a 72 hour, tracked-minimum value. The third was a combination of a delta check and a tracked-minimum-based comparison. Rules were compared to RIFLE criteria using 42,299 inpatient creatinine results from our hospital.
After optimization of the specific creatinine change thresholds, rules combining delta checks with tracked minimum comparisons generally demonstrated improved performance with respect to RIFLE criteria in comparison to rules based on delta checks alone or on tracked minimum values alone. For example, one combination rule was more than 80% sensitive and 80% specific in identifying AKI (injury level criteria). Moreover, we found that certain rules may be useful in identifying patients at risk for proceeding to AKI before they meet formal criteria. 
Rule-based flagging systems within a laboratory information system may be useful in identifying patients at risk for AKI.
Subclassifying the Spectrum of Abnormal Endometrial Proliferation with Secretory Change: Supervised Automated Image Analysis as a Tool for Quantitative
Assessment of Ki67 Index
Grzegorz T. Gurda MD PhD (, Alexander S. Baras MD PhD, Robert J. Kurman MD, Toby C. Cornish, MD PhD
Department of Pathology, The Johns Hopkins Hospital, Baltimore, MD
Abnormal endometrial proliferation exists in a spectrum from focal crowding to hyperplasia to carcinoma. Diagnostic morphology, however, can be obscured by secretory changes. Here, we assess if Ki67 index correlates with morphologic diagnoses and how well human estimates and automated image analyses subclassify the entities within this spectrum versus the gold standard of time-intensive manual counts.
Cell profiler (Broad Institute); Image Scope (Aperio Technologies).
70 cases of endometrium with secretory changes were collected: baseline secretory endometrium, focal crowding, hyperplasia (simple/complex) and endometrial carcinoma. Corresponding Ki67 immunostains were photographed with 3 representative 100X fields per case. All images were analyzed with automated software and 1 per case counted manually or by averaged ‘eyeball’ estimate by 3 independent pathologists. Statistical analyses were carried out with Graph Pad Prism.
Ki67 index closely parallels the step-wise progression of abnormal endometrial proliferation with secretory change; similar rates are obtained by manual counting, estimates and automated image analyses (Table). Overall, there was high concordance between manual counts versus estimate and automated analysis (93%, 100%). At intermediate Ki67 rates (10-75%), automated analysis may be more accurate (100% vs 86%), likely due to human overestimate. Human eye may be superior at identifying secondary structures (glands, vessels, etc), whereas automated software more accurate at numerical counts. Time allotment per case: 10 minutes (manual), 30sec-1min (automated) and 15-20sec (estimate).         

Diagnostic Entity

Manual Count



Secretory Endometrium

1.2 + 1.1%

1.3 + 2.2%

3.5 + 3.6 %

Focal Gland Crowding

3.8 + 4.0%

5.0 + 3.7%

4.7 + 1.5%

Hyperplasia + atypia

30 + 18% ***

28.9 + 19%

26 + 26%

Endometrial Carcinoma

64 + 24% ***

49 + 20%

76 + 24%

***p<0.001 versus Secretory by one-way ANOVA. Non-significant for manual vs estimate vs automated.
Automated analysis is a fast and useful tool in assessment of Ki67 index, particularly at boundaries of overlapping, but differentially-managed diagnostic entities. Combination of human-automated methodology (supervised automated counting), may further improve its utility. Future Directions: Assess inter versus intra specimen variability as a diagnostic marker; crowdsourcing ( mTurk platform) as ancillary tool in supervised automated counting.
Automated Whole Slide Image Screening for Prostatic Adenocarcinoma by Use of Spatially-Invariant Vector Quantization
Jennifer Hipp, MD, PhD (, Lakshmi P. Kunju, MD. Ulysses J. Balis, MD
Department of Pathology, University of Michigan, Ann Arbor, MI
Spatially-Invariant Vector Quantization (SIVQ) has exhibited broad utility in the detection of subtle architectural and nuclear features, making a compelling case for the exploration of its utility to screen (identify) prostate cancer in core biopsies. Given the time-consuming task of screening, automated detection would be of significant utility to the diagnosing pathologist in transforming workflow from a time-intensive screening task to a directed-review/confirmatory activity.
Digital whole slide images were obtained from 6 prostate core biopsies with prostatic adenocarcinoma (Gleason score 3+3=6). Use of a region-of-interest extraction tool, dCore, followed by use of an image aggregation tool, ImageMicroArray Maker, allowed for the generation of a montage of diverse examples of prostatic adenocarcinoma, atrophy, adenosis and benign glands, and finally, high grade PIN. 
Since the morphologic criteria for prostate cancer includes both architectural features and the presence of cytologic atypia, image analysis was performed at low and high magnification (8x & 40x, respectively). Two vectors in tandem were utilized to identify malignant epithelial architectural features. Ring vectors specific for cytologic atypia were chosen to identify nuclear enlargement, hyperchromasia, prominent nucleoli on the high magnification field views. 
SIVQ-based ring vectors were able to efficiently identify the cytologic features of cancer (35/37 malignant glands with a sensitivity and specificity of 87% and 94%, respectively), as well as high-grade PIN (owing to both diagnostic entities having the same cytologic features). This same constellation of ring vectors was successful avoiding false-positive detection of areas of adenosis, atrophy or benign epithelium. Architecturally tuned vectors exhibited a high rate of detection of most small infiltrating glands (122/179 malignant glands) but also with a small false discovery rate for a minority of admixed atrophic glands (7/38 F.P. for benign glands).
SIVQ-based detection is able to efficiently identify the architectural and cytologic features of prostate cancer with a higher sensitivity for nuclear atypia. We attribute the lower sensitivity of architectural vectors detection to the variation in size and color of cancerous glands, and to the difficulty of distinguishing malignant from atrophic glands at low magnification.   This study provides pilot data that use of the SIVQ algorithm as a stand-alone solution holds promise for the development of automated WSI detection.
Development of a Computer-Aided Decision Support Tool for Diagnosis of Micropapillary Urothelial Carcinoma
Steven C Smith, MD, PhD1 (, Jason Hipp, MD, PhD1, Jerome Cheng, PhD1 James Monaco, PhD2,  Anant Madabhushi, PhD2,  Jesse McKenney, MD3,  Lakshmi P Kunju, MD1 and Ulysses J Balis, MD1
1Department of Pathology, University of Michigan,
    Ann Arbor, MI
2Rutgers University, Piscataway, MJ
3Stanford University, Stanford, CA
Micropapillary urothelial carcinoma (MPUC) is a distinctive, aggressive variant of urothelial carcinoma (UC). Recent studies have examined the prognostic value of micropapillary morphology in UC; however, the results have exhibited significant interobserver variability in both diagnosis and threshold for amount of MPUC versus non-MPUC morphology that is associated with poorer prognosis. Thus, we investigated the feasibility of a computer-aided approach to assist pathologists in the diagnosis and quantitation of this entity.
SIVQ (spatially invariant vector quantization) is an image analysis algorithm that uses circular textural vectors for pattern recognition in digital slides. Image microarrays are tiled digital images of constrained size and resolution containing histopathologic features of interest, prepared from digital whole slide images.
We iteratively developed vectors targeting two recently studied diagnostic features of MPUC: 1) retraction artifact, reportedly 100% sensitive, and 2) multiple nests within the same lacuna, reportedly 95% specific. We then applied the algorithm, using these vectors, to image microarrays constructed from quadruplicate representative MPUC and non-MPUC areas of digital whole slide images of 10 MPUC cases culled from a recent consensus report on MPUC morphology.
For detection of retraction artifact, the algorithm identified classic MPUC morphology, showing 12.0Description: plusmn5.6 [95%CI]-fold greater detection in MPUC as compared to non-MPUC regions. For the multiple nests within the same lacuna feature, the algorithm was more specific for MPUC, showing 18.8Description: plusmn8.9-fold greater detection in areas of MPUC.
Our findings suggest that this novel strategy may provide a useful decision support tool to pathologists in the diagnostically challenging setting of MPUC. The automated feature identification process is tailored to recapitulate the thought process of the pathologist, based on matching sensitive or specific features of MPUC. We are currently testing the utility of combining vectors for both features to optimize identification and quantitation of MPUC morphology. Future studies will test the utility of this approach for the standardization of interobserver reproducibility for diagnosis and quantitation of MPUC.
Scientific Imaging & Applications Development
Location: Acapulco Room

 MIGIS: High Performance Spatial Query System for Analytical Pathology Imaging

View PDF

Ablimit Aji, MS(, Qiaoling Liu, MS1, Fusheng Wang, PhD2,3, Tahsin Kurç PhD2, 3, Joel Saltz, MD, PhD2, 3
1 Department of Mathematics and Computer Science,
     Emory University, Atlanta, GA
2 Department of Biomedical Informatics, Emory
    University, Atlanta, GA
3 Center for Comprehensive Informatics, Emory
    University, Atlanta, GA
Systematic analysis of large-scale microscopy images can generate tremendous amount of spatially derived quantifications for microanatomic objects such as nuclei and blood vessels. Most queries on the results are spatially oriented, and highly computational and data intensive. For example, there are queries on spatial cross-matching of multiple sets of segmented objects and on spatial proximity between micro-anatomic objects and local nearest neighbor density. We report our recent progress on developing a highly scalable and efficient query system for supporting large scale analytical queries on pathology images.
We are developing a MapReduce based framework MIGIS to support cost effective high performance spatial queries on commodity clusters. The data is staged on Hadoop Distributed File System (HDFS) which supports reliable storage with replication. We build a spatial query engine by extending CGAL computational geometry library and SpatialIndex library. We take advantage of the open source MapReduce framework Hadoop for parallelizing and scheduling query execution.
MIGIS consists of a spatial query engine RESQUE which runs spatial query by building spatial indexes on the fly. Data files with segmented boundaries and extracted spatial features are merged and staged onto HDFS. Spatial queries are expressed as “map” and “reduce” functions and parallel execution is performed by Hadoop. Effective job optimization is performed to minimize query execution bottleneck due to data skew on MapReduce.
MIGIS supports following queries: spatial cross-matching query for algorithm evaluation, nearest neighbor queries to find closest blood vessel for each cell, and high density queries to find regions with high concentration of nuclei, such as pseudopalisades. We tested our experimental study with datasets of 18 and 50 whole slide images respectively. Our study demonstrates high scalability of the queries on MapReduce. 
The big data from medical imaging -- the vast amount spatially derived information generated from pathology image analysis -- shares similar requirements for high performance and scalability with enterprise data, but in a unique way. Our system MIGIS demonstrates an effective and highly scalable solution that can run on cost-effective commodity clusters. This can be used for algorithm sensitivity study, and whole slide image analysis.
Evaluation of 655 Patent Applications Related to Digital Pathology
Ioan C. Cucoranu, MD (,   Anil V. Parwani, MD, PhD1, Ronald Weinstein, MD2,  Jonhan Ho, MD1, Liron Pantanowitz, MD1
1Department of Pathology, University of Pittsburgh
     Medical Center, Pittsburgh, PA
2University of Arizona, Tucson, AZ
Digital pathology is reshaping the practice of pathology. Development of applications for telepathology and image analysis will play a major role in the future adoption of digital pathology. There have been major technological advances in this field over the last two decades. Most of these developments have been captured in the form of patents, to protect the intellectual property of inventors. Our aim was to systematically evaluate patents related to digital pathology.
Patents published in databases at (United States Patent and Trademark Office, Alexandria, Virginia, USA), were searched using Google Patents (Google Inc., Mountain View, California, USA). Data was extracted and analyzed with Papers software (Mekentosj BV, Aalsmeer, Netherlands).
The United States Patent and Trademark Office public database was queried using the following keywords: “computer assisted diagnosis and pathology”, “digital image analysis and pathology”, “digital imaging and pathology”, “digital imaging and pathology and data security”, “digital microscopy”, “digital pathology”, “telemicroscopy”, “telepathology”, “telepathology and data security”, “videomicroscopy”, “virtual microscopy”, “virtual pathology”, “whole slide imaging”, “whole slide scanner and pathology”. Google search engine was used to facilitate the download of data into the Papers application.
A total of 655 patent applications related to the field of digital pathology were identified, including 413 (63%) patents specific to pathology use, and 242 (37%) patents also usable beyond pathology. Of the 413 pathology specific filed patents, 297 (72%) were awarded and 63 of them expired (due to non-payment of maintenance fees or term end). There were also 52 abandoned applications, 63 pending patents and 1 rejected application. Several (28) initial telepathology patents were related to television. The number of pathology specific patent applications quadrupled in the last decade: 68 (1991-2000) versus 323 (2001-2010). The number of patents actually awarded almost tripled during these periods (figure): 66 versus 208.
This study shows that there have been many patent applications related to digital pathology in the last two decades. Of these, the number of patents specific to use in pathology almost quadrupled in the last decade. Most (2/3) of them are related to telepathology (including whole slide imaging, data management, networking, data security and user interface), while the rest (1/3) are related to image analysis. Based on these observations, technological advances related to digital pathology, especially tools for image analysis are expected to increase in the future. 
Early Experience With Optical Coherence Tomography (OCT), A Novel 3D Imaging Modality in Pathology
Jeffrey L Fine MD (, Ioan Cucoranu MD
Department of Pathology, University of Pittsburgh, Pittsburgh, PA
Optical coherence tomography (OCT) and related 3-dimensional microscopy techniques can potentially allow pathologic diagnosis directly from tissue, without glass microscope slides. Our early efforts have focused on correlation with histopathology and optimizing the imaging process, first with an ophthalmic system and now with a newer pathology system. We present our experience with OCT imaging as we begin systematic studies with the technique.
Modified full-field OCT (Light CT, LLTech, Paris, France), using a motorized stage to capture and stitch together multiple 10x fields of view into a single image or image stack. Its resolution is about 1 um (transversely and axially). Resulting images are viewed using included software. H&E slides were scanned using a WSI system (Aperio XT, Vista, California).
Fixed and fresh tissue was scanned several different imaging parameters. H&E slides from the imaged tissue were subsequently prepared and compared. In addition to evaluating images for diagnostic features, scans were evaluated for other features such as depth into tissue, scan time, and usefulness of quality parameters.
Tissue from 22 patients (about 50 tissue samples) was scanned. Large-area images were limited to one (or few) focal planes into the tissue, but thicker 3D image stacks were possible using fewer fields of view. Resolution appears adequate to discern small features (fat septa in breast, recognizably mucinous glands in an ovarian tumor), but images appear to have limited contrast in many tissue types. Both fixed and fresh tissue worked well, though there appear to be subtle differences in image quality related to firmness of fixed tissue.
OCT is an exciting imaging modality and many histopathologic features are recognizable in our images. Further studies are needed to better establish its utility in clinical settings such as “room temperature” frozen section or even in-vivo imaging. Image size, capture speed, and other technical issues exist but will likely be solved with hardware improvements if a market for clinical applications develops. Other medical specialties are already developing 3D microscopy applications, and if pathologists can become involved then we will be better able to adapt as future medical diagnostic work evolves.
A Method for Quantitative Analysis of Ki67 With Enlarged Follicles in Follicular Lymphoma
Bruce Levy, MD (, Yukako Yagi, PhD2, Abner Louissaint, Jr., MD., PhD3
1University of Illinois at Chicago, Department of
    Pathology, Chicago, IL
2Pathology Imaging and Communication Technology
     Lab, Massachusetts General Hospital, Boston, MA
3Department of Pathology, Massachusetts General
    Hospital, Boston, MA
In many cases the size of an area of interest or image analysis software may restrict simple quantitative analysis. For example, while analyzing cases of follicular lymphoma stained for Ki67, enlarged/coalesced follicles could not be analyzed from a single captured image.   A method was developed to subdivide these follicles into smaller segments that were analyzed individually and then mathematically reconstituted to determine Ki67 positivity for the entire follicle.
Used in this study was a Hamamatsu Nanozoomer 2.0 HT scanner with associated scanning and viewing software, ImmunoRatio image analysis software (, and a basic desktop computer and monitor.
Slides of follicular lymphoma stained with Ki67 were scanned with a Hamamatsu Nanozoomer 2.0 HT scanner with NDP.scan software at 20X objective magnification. A pathologist examined the slides using NDP.serve viewer and divided the follicles into two-dimensional grids of 0.1 mm2 squares that were individually exported. Each square was analyzed for the percentage of Ki67 positive nuclei using ImmunoRatio image analysis software. The pathologist defined the boundaries and determined the area of the follicle in each exported image, and the software calculated the percentage of nuclei positive for Ki67. The average percent positivity for the entire follicle was then calculated utilizing the formula:
Average Positivity = ∑(An*Pn)/Total area
Where: A=Area of follicle in each square
            P=Percent positivity for each square
            n=Number of squares making up a follicle
Comparing the analysis of follicles analyzed in their entirety and then divided into between two and four segments validated the method. A total of 23 cases of follicular lymphoma were evaluated using this method. Areas analyzed per case ranged from 1.3519 mm2 to 5.1404 mm2. The Ki67 positivity was calculated between 3.6% and 82.1%. 
The grid method presented in this paper demonstrates a process in which larger areas that cannot be quantitatively analyzed in one image can be broken down and evaluated in a systematic method. It could be automated, providing a quantitative result for an entire slide quickly and reliably. The process is not limited to Ki67 analysis, but could be applied to many different quantitative analyses. It can also be used to compare different areas from the same case for variations in positivity, which was noted in this study, but not further analyzed at this time.
Location: Toronto Room
Revitalization of Small Group Pathology Teaching in a Medical School Curriculum Incorporating Informatics Tools
Philip J. Boyer, MD, PhD ( , Dana M. Grzybicki MD, PhD, Robin L. Michaels, PhD, Robert L. Low, MD, PhD
Department of Pathology, University of Colorado Denver, Aurora, CO
Pathology instructional content is a cornerstone of the organ system-focused years 1 and 2 curriculum of the University of Colorado School of Medicine and includes 20 two-hour “small group” sessions (156 medical students divided into 8 to 10 groups) with faculty and residents serving as preceptors. The small group sessions had been essentially unchanged for approximately ten years. There was lack of uniformity of content and instructional model of the sessions; materials were outdated; student scores in the pathology component of standardized examinations lagged behind scores in other disciplines; and evaluation of the small group sessions by students, highly variable among instructors, ranged from high praise to disappointment to virulent diatribes.  
Distribution of the text- and image-based content of the curriculum incorporated Microsoft Word and PowerPoint documents, Adobe Acrobat Pro PDFs, standard HTML Web pages, and the Blackboard content management system. Specimens used for macroscopic instruction were acquired from surgical pathology and autopsy cases to supplement the extensive but outdated teaching collection.
Case-based educational modules were developed for each of the 20 small group sessions, incorporating didactic content and gross specimens to compliment and supplement material covered in lectures, guided by National Board of Medical Examiners (NBME), United States Medical Licensing Examination (USMLE) topic lists. Outcomes assessment included (1) tracking of student performance on pathology content of a customized version of the NBME Comprehensive Basic Science Exam (CBSE) taken after completion of the first year of medical school and on Step 1 of the USMLE taken after the second year and (2) evaluation of feedback from students for each “block” of the curriculum and on Graduation Questionnaire (GQ) surveys. 
There has been an improvement in pathology component scores on the CBSE and NBME Step 1 examinations; generally strongly positive comments about small group sessions and practice questions on course evaluation; and reduction in complaints about the pathology course on student GQ surveys.
Work to date has met the primary objectives of improving and standardizing content. Advances during the 2012-2013 academic year will include incorporation of required Blackboard-based “pre-test” quizzes for each session.
Active Learning for Pathology Image Analysis
Lee AD Cooper, PhD (, Christina Appin, MD2, Rami Yacoub, MD3, David A Gutman, MD, PhD4, Hyun Ju Choi, PhD4, Jun Kong, PhD1, Fusheng Wang, PhD4, Carlos S Moreno, PhD2, Robin Bostick, MD, MPH3, Daniel J Brat, MD, PhD2, Joel H Saltz, MD, PhD4
1Center for Comprehensive Informatics, Emory
    University, Atlanta, GA
2Department of Pathology and Laboratory Medicine,
    Emory University, Atlanta, GA
3Rollins School of Public Health, Emory University,
    Atlanta, GA
4Department of Biomedical Informatics. Emory
    University, Atlanta, GA
Object classification is a fundamental problem in pathology image analysis. We have developed an active learning system to perform object classification tasks in whole-slide imaging applications. This system enables users to iteratively refine classification rules through a series of computer-guided experiments. The classifier selects examples for user review, and feedback is collected to refine the classifier at each iteration. This machine-driven approach forces users to label ambiguous examples and also hides algorithm complexity from end users, enabling scientific collaborators to be engaged more easily.
Sections of glioblastoma tissue were stained with Hoescht and quantum dot immunohistochemistry to identify mTOR, phospho-Rb and Ki67. Slides were digitized at 20X objective magnification on a 3D Histech panoramic scanner. Nuclei were segmented using gradient flow tracking, and each cell/nucleus was represented with 73 features describing morphology and protein expression. LogitBoost was used to classify cells and provide classification confidence scores. Scores are used to select samples for review to create a uniformly distributed set of examples from obvious to ambiguous.
Examples are presented in groups of fifty, evenly split between classes, and classification and segmentation errors are recorded (random class assignment initially). Review continues until enough examples from each class have been identified to train the classifier. Two experiments were performed: 1. Identify cells with any single positive signal and 2. Identify cells with coexpression. Classifiers were trained until the pathologist was satisfied and then validated on 500 randomly selected cells.
The first experiment achieved 99% accuracy in discriminating single positive cells. This classifier was trained in 4 iterations reviewing 300 total cells. Of these, 76 required correction, with 73 corrected in the first iteration. The classifier for coexpression did not converge during the initial iteration due to a lack of coexpressing examples. 600 total cells were reviewed in this iteration.
The active learning framework is a promising approach to object classification problems, but requires further investigation to determine its limitations. The initial training iteration is problematic for applications with rarely encountered objects since examples are initially randomly chosen. Future studies will investigate improving sampling methods and more complex applications.
Preparation and Training for Interinstitutional WSI Clinical Consultation: A Process Analysis
Nicholas C. Jones (, Elena F. Brachtel, MD, Rosemary H. Tambouret MD, Chin-Lee Wu, MD, PhD, Eugene J. Mark MD, G. Petur Nielsen, MD, Rosalynn M. Nazarian, MD, Lyn M. Duncan, MD, Gregory Y. Lauwers, MD, David C. Wilbur, MD
Department of Pathology, Massachusetts General Hospital and Harvard Medical School, Boston, MA
Prior to beginning an interinstitutional Whole Slide Image (WSI) subspecialty consultation pilot, a number of volunteer pathologists and one WSI technician needed to gain experience with scanning and interpreting WSIs and emulating the clinical workflow with a digital pathology platform. This organic process of training new pathologists across multiple subspecialties was continuously evaluated.
The Corista Digital Pathology Platform (Corista LLC, Concord, MA, USA, web browser-based WSI viewer and case management software), Mirax MIDI WSI Scanner, networked PCs.
An imaging technician scanned and entered cases from the normal clinical workflow of volunteer subspecialist pathologists into the digital platform. The pathologists were asked to interpret the cases on WSI before their normal glass slide interpretations. No washout period was used in order to emphasize potential differences between digital and glass slide interpretations. Case selection was not random; a variety of smaller cases were selected in order to minimize the impact on the pathologist’s time. WSI and traditional interpretations were compared, pathologists were surveyed as to their thoughts on the process and technology, and technical data was gathered.
170 specimens were reviewed spread among 9 pathologists. 163 (95.9%) of the WSI interpretations correlated with the glass slide interpretations. Six (3.5%) had minor discrepancies (no patient care impact), and one (0.6%) case had a major discrepancy (potential patient care impact). Pathologist responses to the training methodology and platform were highly positive, with the time involved in WSI interpretation being cited as a concern, primarily due to computer speed, network speed, and input device limitations.
Though the goal was not to validate WSI use, the correlations between digital and glass interpretations were excellent. The individuals involved are comfortable with the technology and primed for WSI telepathology consultation. As computer hardware and network bandwidth increase at predictable rates and pathologist WSI screening speed increases with experience, it is expected that these limitations will decrease with time.
The History of Pathology Informatics:
A Global Perspective
Seung Park, MD (, Anil Parwani, MD, PhD1, Raymond D. Aller, MD2, Lech Banach, MD3, Michael J. Becich, MD, PhD4, Stephan Borkenfel5; Alexis B. Carter, MD6, Bruce A. Friedman, MD7, Marcial Garcia Rojo, MD8, Andrew Georgiou, PhD9, Gian Kayser, MD10, Klaus Kayser, MD11, Michael Legg, PhD12, Christopher Naugler, MD13, Takashi Sawai, MD14, Hal Weiner, MS15, Dennis Winsten, MS16, Liron Pantanowitz, MD1
1Department of Pathology, University of Pittsburgh
    Medical Center, Pittsburgh, PA
2Department of Pathology, University of Southern
    California, Keck School of Medicine,Los Angeles,CA
3Department of Anatomical Pathology, Walter Sisulu
     University, Mthatha, South Africa
4Department of Biomedical Informatics, University of
    Pittsburgh, Pittsburgh, PA
5International Academy of Telepathology, Heidelberg,
6Department of Pathology, Emory University,
    Atlanta, GA
7Department of Pathology, University of Michigan
     Medical School, Ann Arbor, MI
8Department of Pathology, Hospital General de Ciudad
    Real, Ciudad Real, Spain
9University of New South Wales, Centre for Health
    Systems and Safety Research, Australian Institute
    of Health Innovation, New South Wales, Australia
10University Freiburg, Institute of Pathology, Freiburg,
11Charite, Institute of Pathology, Berlin, Germany
12University of Wollongong, Centre for Health
    Informatics and e-Health Research, New South
    Wales, Australia
13Department of Pathology, University of Calgary,
    Calgary, Ontario, Canada
14Department of Pathology, Iwate Medical University
     School of Medicine, Iwate Perfecture, Japan
15Weiner Consulting Services, Florence, OR
16Dennis Winsten & Associates, Inc., Tucson, AZ
“If you don’t know where you came from how can you know where you’re going?” The history of pathology informatics is a tale with many dimensions. At first glance, it is the familiar story of individuals solving problems that arise in their clinical practice of medicine. Under the surface, however, lie powerful forces – technical, regulatory, societal and beyond – that have all played their part in molding our discipline into what it is today. The aim of this project was to solicit a historical account of the evolution of pathology informatics from informaticists around the world.
Email correspondence; Microsoft Word (Redmond, WA, USA).
Leading figures in pathology informatics from all over the world (Africa, the Americas, Asia, Australia, Europe) were requested to submit historical material (articles, monographs, personal remembrances, photographs) regarding local progress made in pathology informatics. 
Review of submitted materials over a one year period indicates the following trends: (1) early efforts in pathology informatics stemmed from the USA and Europe; (2) pathology informatics is now a global discipline, with almost all continents driving developments in the field; (3) the advent of the Internet was a disruptive event with significant impact on pathology informatics; (4) the prevalence of different technologies in various regions correlates with intangible factors (e.g., regulatory concerns) and tangible factors (e.g., operational costs).
This project resulted in the first global compilation of the history of pathology informatics. It clearly demonstrates that intimate knowledge of our past is a critical foundation for wise stewardship of our future. This historical document is intended to serve as a baseline for future developments and refinements in the field.    Further work is required to capture missing details to complete this treatise.
Clinical Imaging
Location: New Orleans Room 
Report from the 2nd International Scanner Contest 2012
Peter Hufnagl, PhD (,2, Norman Zerbe1,2, Alexander Alekseychuk, PhD3, Frederick Klauschen, MD, PhD1, Gian Kayser, MD4, Marcial Garcia Rojo, MD, PhD5, Arvydas Laurinavicius, MD6, Yukako Yagi, PhD7, Thomas Schrader, MD8
1Institute for Pathology, Charité – Universitätsmedizin
    Berlin, Berlin, Germany
2University of Applied Science Berlin, Berlin, Germany
3Computer Vision and Remote Sensing, TU Berlin,
    Berlin, Germany
4Institute of Pathology, University of Freiburg,
    Freiburg, Germany
5Servicio de Anatomia Patologica, Hospital General de
    Ciudad Real, Ciudad Real, Spain
6National Centre of Pathology, Vilnius, Lithuania
7Massachusetts General Hospital, Boston, MA
8Dept. Informatics and Media, University of Applied
    Sciences Brandenburg, Brandenburg, Germany
The International Scanner Contest understands its mission to promote digital pathology and virtual microscopy in research, education and diagnostics. 
Several algorithms and applications have been developed to measure and compare focus and image quality, speed, color fidelity and geometric distortions.
The 2nd ISC 2012 contained 5 domains to evaluate different capabilities of participating systems. Each vendor was free to choose which of the available domains to attend:
High Throughput:
Sets of 35 slides of different origin randomly mixed had to be scanned automatically under daily-use conditions. Preparation and scanning time have been measured and a focus-corrected speed was calculated.
The same 10 slides had to be scanned by each vendor in a limited time with best quality. Blinded quality evaluation was done by experienced pathologists applying a 9-tired scoring system.
Two artificially stained monochrome slides (quantum dots) had to be scanned in a limited amount of time. Contrast, contrast-noise ratio, signal-noise ratio and sharpness were measured to compare results.
Image Analysis:
Twenty spots located on a Ki67 stained breast cancer TMA had to be scanned automatically. Subsequently participants had to quantify the amount Ki67 positive tumor cells in selected regions. Results were compared to manual quantification by a reference panel of pathologists.
Two artificial slides, an IT 8.7/1 color target and a calibrated micrometer raster, had to be scanned. Resulting virtual slides were analyzed for color fidelity and geometric distortions.
In the contest we evaluated 7 scanning systems (Metafer-VSlide-SFx80, NanoZoomer-HT-2.0, Pannoramic-Desk, Pannoramic-250, TISSUEScope-4000, UltraFastScanner-UFS, VS120-S5) from 6 vendors (3DHistech, Hamamatsu, Huron, Metasystems, Olympus, Philips) participating in 32 tests that were all passed successfully.
The International Scanner Contest 2012 has a great benefit for the digital pathology community, because it offers a standardized and comprehensive means of evaluation and therefore a way for manufacturers to improve their devices and customers to choose the best device for their respective needs. The test domains represent requirements in routine pathology as well as education and research. Nevertheless vendors as well as pathologists requested additional domains, e.g. to determine capabilities of workflow integration, for the 3rd International Scanner Contest in 2013.
Design and Implementation of a Digital Pathology Network for Air Force Medical Service (AFMS) Pathology Practice
Nicholas Lancia, MD (; Jonhan Ho, MD2, Leslie Anthony3, Orly Aridor3, David W. Glinski3, Chris Saylor3, Ricky Bond3, Joseph P. Pelletier, MD4, Dale M. Selby, MD4, Emily Green, MD5, Kyle Rickard, MD6, Anil V. Parwani, PhD, MD7
1Department of Pathology, 81st Medical Group
    Hospital, Keesler Air Force Base, Biloxi, MS
2Department of Dermatopathology, University of
    Pittsburgh Medical Center, Pittsburgh, PA
3Office of Sponsored Programs and Research Support,
    University of Pittsburgh Medical Center, Pittsburgh,
 4Department of Pathology, Wilford Hall Ambulatory
    Surgical Center, Lackland Air Force Base, San
    Antonio, TX
5Department of Pathology, David Grant Medical
    Center, Travis Air Force Base, Fairfield, CA
6Department of Pathology, Wright-Patterson Medical
    Center, Wright-Patterson Air Force Base, Fairborn,
7Department of Pathology, University of Pittsburgh
    School of Medicine, Pittsburgh, PA
Due to recent advances in digital pathology (DP) and the resulting anticipated changes in pathology practice the Air Force Medical Service (AFMS) is exploring ways to introduce DP into its pathology labs.
A model DP network composed of commercially available whole slide imaging (WSI) scanners and their supporting software will be designed. The network should allow sharing of digital images between several geographically remote pathology labs via the AFMS IT network.
To support the design of a small scale model DP network, the unique needs and requirements of AFMS pathologists and pathology system were identified using the contextual inquiry method. Findings were utilized to determine work practices and clinical applications that will most benefit from DP implementation. US commercially available and serviced WSI scanners and supporting software were reviewed. The selected preferred WSI scanner/system had to support the relevant clinical application/workflows and be compatible with AFMS IT-related infrastructure. 
Contextual inquiry revealed that DP will be highly beneficial for AFMS mainly due to the large number and global distribution of its patient population and medical facilities and its unique pathologist/histotechnologist staffing pipeline. Clinical applications and workflows that would most readily benefit from DP implementation were consultations, quality assurance, and global workload distribution. WSI scanner evaluation criteria included a variety of objective technical features. Subjective criteria such as image quality of identical preselected glass slides scanned by each scanner/system and usability of each system were compared. The selected preferred DP system was purchased and installed at four regional AFMS pathology centers. Training processes to support scanning and digital image review were designed, as well as surveys to track pathologists’ adoption and use of DP. Following a successful connection of the WSI systems to the AFMS IT network, implementation of a clinical application such as a virtual dermatopathology consultation program within AFMS will be established. 
Introduction of new technology such as DP into a large healthcare organization such as AFMS requires careful design and implementation.
Computer-Assisted Analysis of Cytologists’ Exploration of Whole Slide Images
Liron Pantanowitz, MD (,2, Walis E. Khalbuss, MD, PhD1, Eugene Tseytlin, MS2, Sara E. Monaco, MD1, Chengquan Zhao, MD1, Jackie Cuda1, Anil V. Parwani, MD, PhD1, Claudia Mello-Thoms, PhD2
1Department of Pathology, University of Pittsburgh
    Medical Center, Pittsburgh, PA
2Department of Biomedical Informatics, University of
    Pittsburgh, Pittsburgh, PA
Current clinical practice in cytology involves a two-step process where slides are initially screened by a cytotechnologist and then sent for final review and diagnosis by a cytopathologist. With Pathology’s migration towards a digital environment it is important to determine whether these screening and diagnostic tasks can be performed as well on digital slides as they are with glass slides.
A custom-designed virtual microscope was used in this study. Cytologists had the ability to zoom in on the slides (up to 40x), pan, report diagnostic criteria and ultimate diagnosis (as well as differential). All interactions with the interface were time-stamped and recorded.
Three cytopathologists and five cytotechnologists voluntarily participated in this study. They read a case set of 12 cytology digital slides, of which 5 depicted Pap tests (conventional and liquid-based specimens), and 7 depicted non-gynecological samples, including 3 FNAs of lung, 1 FNA of liver, 1 FNA of thyroid, 1 FNA of breast, and 1 urine sample. Diagnoses covered a wide range of conditions, from benign (such as infections) to neoplasms (carcinoma, melanoma). Specialized software was developed to analyze the slide exploration strategy of cytologists. Exploration was divided by magnification range into “low” (1x < mag <= 10x), “medium” (10x < mag <= 20x) and “high” (20x < mag <= 40x).
Cytotechnologists took significantly longer to read the digital slides (average 11 min 13 sec) than cytopathologists (4 min 51 sec) (z=-11.628, P<0.0001). On average, cytologistssampled only about 66% of the entire slide at low magnification levels, and their exploration became even more “economical” as the magnification level was raised. Greater exploration of the slides at low magnification was statistically significantly correlated with correct diagnoses for cytotechnologists (for correct diagnosis, average coverage was 81%, whereas for incorrect diagnosis it was 65%, z=-2.52, P=0.0117). A high agreement among cytopathologists in the areas of the slide explored was also observed.
Our study suggests that cytotechnologists may have trouble with place keeping using whole slide images, and as a result they cover less than ¾ of the slide at low magnifications, and this has significant screening and diagnostic consequences. They also take significantly longer to use this medium, which could potentially have a significant impact on workflow. 
Recommendations for Validating Whole Slide Imaging for Diagnostic Purposes in Pathology:
College of American Pathologists Pathology and Laboratory Quality Center
John H. Sinard, MD,PhD (, Walter H. Henricks, MD2, Alexis B. Carter, MD3, Lydia Contis, MD4, Bruce A. Beckwith, MD5, Andrew J. Evans, MD, PhD6, Christopher N. Otis, MD7, Lisa A. Fatheree, BS SCT(ASCP)8, Avtar Lal, MD, PhD9, Anil V. Parwani, MD, PhD4, Liron Pantanowitz, MD4
1Department of Pathology, Yale University School of
     Medicine, New Haven, CT
2Pathology and Laboratory Medicine Institute,
    Cleveland Clinic, Cleveland, OH
3Department of Pathology and Laboratory Medicine,
    Emory University, Atlanta, GA
4Department of Pathology, University of Pittsburgh
    Medical Center, Pittsburgh, PA
5North Shore Medical Center Salem Hospital, Salem,
6Laboratory Medicine Program, University Health
    Network, Toronto, ON, Canada
7Department of Pathology, Baystate Medical Center,
    Tufts University School of Medicine, Springfield, MA
8College of American Pathologists, Northfield, IL
9University Hospital, London Health Science Center,
    London, ON, Canada
Whole slide imaging (WSI) is increasingly being used for diagnostic purposes, education and research. Concern has arisen whether WSI can replace the conventional light microscope as the method by which pathologists view patient samples and render diagnoses (primary and/or consultation). Validation of WSI is important to ensure that diagnoses rendered using digitized slides are as accurate as those obtained using glass slides. There are currently no standardized guidelines regarding validation of WSI for clinical diagnostic use.
In June 2010, the College of American Pathologists Pathology and Laboratory Quality Center convened a non-vendor panel from North America with expertise in digital pathology. The project scope was defined as: “To recommend validation requirements for WSI used for diagnostic purposes.”
A computerized search of published articles related to WSI was conducted. 767 studies met the search term requirements. 27 studies underwent detailed data extraction to capture evidence in support of validation recommendations. Assessment for strength of evidence, consisting of level of evidence, quantity, size of the effect, statistical precision and quality assessment (risk of bias) was performed. The panel convened eight times via conference calls and developed draft recommendations. These were made available for open comment in July 2011 in the form of a survey. Approximately 132 responses for each recommendation plus 531 comments were received over the four-week comment period. The panel had ten additional conference calls to review the feedback, grade the evidence, and finalize the draft recommendations.
Evidence-based review suggests that diagnoses made using WSI can be as accurate as using glass slides. Validation of the entire WSI system is important prior to using the system for any clinical purpose. The panel developed 12 final recommendations for the validation of WSI systems.
Validation of WSI is necessary to ensure that a pathologist using this technique to view digitized glass slides can consistently make the same clinical interpretation as they would from viewing the glass slides using a traditional bright field microscope. Validation should address both technical and interpretative components, and must be specific for the intended clinical use.
Implementation of an Enterprise-Wide Picture Archiving and Communication System (PACS) for Henry Ford Health System (HFHS) Department of Pathology and Laboratory Medicine (PALM)
Robert Stapp, DO ( , Mehrvash Haghighi, MD, Tina Caruana, Ron Brown, J. Mark Tuthill, MD
Department of Pathology and Laboratory Medicine, Henry Ford Hospital, Detroit, MI
We implemented an enterprise PACS throughout the laboratories of HFHS. PACS was deployed across HFHS-PALM that includes four hospitals. A database server was deployed in our data center with client software installed on workstations for pathologists, grossing, and autopsies.
The software (Apollo EPMM® v9.4.3, Falls Church, VA, USA) is a file agnostic image database responsible for managing a variety of image acquisition devices based upon the Microsoft .NET® framework and allows secure storage and distribution of digital files acquired from slides, gross specimens, X-rays, whole-slide imaging, electron microscopy, as well as document scanners. EPMM is capable of interfacing with information systems through its use of HL7, DICOM, and ADT interfaces. Database images can be accessed though the use of the client software or via a web-portal interface.
Staff were directed to acquire images in real-time during the grossing and sign-out process as well as ad hoc additions to cases. Legacy data was converted from our previous archive, requiring importing demographic information of 290,000 cases into the database. This was accomplished by extracting data from our anatomic pathology laboratory information system (LIS); the achieved files, approximately 512,000, were then matched and imported based upon case accession number. In the current live implementation of PACS, demographics are obtained through an ADT interface with our LIS.
Prior to the PACS, all files were hosted on a shared server based on a traditional file-folder hierarchy using accession number as the root for cases. This system became increasingly inefficient due to the volume of cases and resultant slow access time, limited controls to associate the correct image data to the correct patient, and lack of user and role based security. PACS provided improvements in speed, efficiency, and security. Inter hospital consultations are now performed in real-time. PACS is used to acquire all scanned requisitions and gross images through using automated file mover services. Tumor board presentations have become more efficient and accessible.
PACS system has shown growth in both usage and implementation with increases in efficiency, patient safety, security, and accessibility. As additional features are implemented, PACS will prove to be an invaluable for pathologists.
Usage of Inexpensive Consumer Electronics in the Three-Dimensional Capture and Visualization of Gross Specimens
Seung Park, MD (, Jeffrey Fine, MD
Department of Pathology, University of Pittsburgh Medical Center, Pittsburgh, PA
The recent explosion in commoditized imaging technologies has allowed the widespread availability of powerful three-dimensional (3D) imaging and scanning technologies. One such device, the Microsoft Kinect, has garnered a large amount of third-party support, currently enabling low-resolution 3D scans of real-world objects in (or near) real time. In this presentation, we investigate possible clinical uses of this technology in the anatomic pathology laboratory.
A Microsoft Kinect (Microsoft Corporation, Redmond, Washington, USA); a portable PC (HP ENVY 14; Hewlett-Packard Corporation, Palo Alto, California, USA); software RGBDemo (Manctl Labs) and MeshLab (Instituto di Scienza e Tecnologie dell’Informazione, Pisa, Italy).
In conjunction with a rotating platform and printed scan markers, RGBDemo and the Microsoft Kinect were used to capture point cloud and color data from two gross specimens: a mastectomy for breast cancer and a hystectomy for fibroids. These data were vertex-cleaned and mesh-reconstituted using MeshLab, and made viewable using the same (Figure, clockwise from top left: raw capture of point mesh, beginning; raw capture of point mesh, ending; final 3D model, zoomed in view; final 3D model, zoomed out and rotated view).
3D models were generated for each specimen, each requiring one slow manual rotation (less than one minute). Each model had 360 degree visibility, and could be rotated and viewed in 3D space. Post-processing enabled one to clean the models of extraneous artifacts (e.g., the “table” surface). Image quality was somewhat limited by resolution.
We captured near-realtime 3D models of gross specimens using cheap hardware and freely-available software. Although resolution-limited, these systems are under active development for other applications; we anticipate rapid improvement in image quality. Future higher quality 3D images may represent a leap over conventional photography for documenting gross specimens, teaching and for pathologist telepresence. Surgeons may also benefit, particularly those already using 3D surgical systems in the operating room. Finally, physicians may be better able to communicate with patients using these models. The possibilities are endless.



Copyright 2009 | Site by UMC Web Team