Scientific Session
Thursday, October 6, 2011
7:30 am – 8:55 am
Locations:
Grand Ballroom 1
Kings Garden North
Kings Garden South/LeBateau
1
Thursday, October 6, 2011
Session #1 - IMAGING & COMPUTATIONAL
Grand Ballroom 1
3D Prostate Histology Reconstruction Informed
by Quantified Tissue Cutting and Deformation
Parameters
models (rigid, rigid+scale, affine, and thin-platespline) were assessed by aligning homologous
fiducials using each model and subsequently
measuring misalignment of landmarks using a leaveone-out cross-validation. These quantifications
informed the design of a 3D reconstruction algorithm
to place histology in the context of the MR images,
which was evaluated using homologous landmarks.
Eli Gibson, MAsc (egibson@imaging.robarts.ca)1,
Cathie Crukley, MLT1,2, José A. Gómez, MD FRCPC1,
Madeleine Moussa, MBBCh, FRCPC1, Glenn Bauman,
MD, FRCPC1,2, Aaron Fenster, PhD1,2, Aaron D. Ward,
PhD1
Results:
Histology sections had a mean±std depth of
1.0±0.5mm and orientation of 1.7±1.1°. Rigid,
rigid+scale, affine and thin-plate spline deformation
models yielded mean±std misalignment of 1.4±0.7,
0.6±0.3, 0.5±0.3 and 0.4±0.3mm, respectively. The
3D reconstruction yielded a mean±std target
registration error of 0.69±0.36mm.
1The
University of Western Ontario, Robarts Research
Institute, London, Ontario, Canada
2Lawson Health Research Institute, London, Ontario,
Canada
Content :
3D reconstruction of digitized 2D histology sections
sparsely sampled from grossed tissue blocks depends
on knowledge about the position, orientation and
deformation of tissue during histological processing.
Many 3D reconstruction methods make the
assumption that histology sections are taken from
equally spaced, parallel planes at the front faces of
tissue blocks. This work quantified aspects of
histological processing and applied the results to
inform the design of a reconstruction algorithm that
aligns histology to ex vivo magnetic resonance (MR)
images.
Conclusions :
Variability in the position and orientation of sections
within tissue blocks could contribute substantial
(>2mm) 3D reconstruction error. The deformation
models for 3D reconstruction more flexible than
rigid+scale yielded small improvements in accuracy
(<0.2mm). Our 3D reconstruction algorithm achieved
sub-millimeter(0.69mm) reconstruction error.
DICOM Compliant Histopathology Software
Technology:
We acquired MR images with 0.27 0.27 0.20mm
voxels using a 3T Discovery MR750 (GE Healthcare,
Waukesha, USA), and histology images with 30x30µm
pixels using a ScanScope GL (Aperio Technologies,
Vista, USA) bright field slide scanning system.
Statistical analysis was performed using Prism 5.04
(Graphpad Software, Inc., San Diego, USA). We
developed the 3D reconstruction using MATLAB 7.8.0
(The Mathworks Inc., Natick, USA).
Danoush Hosseinzadeh, B.Eng, MASc
(dan.zadeh@pathcore.ca)1,3, Anne L. Martel1, 2,3
1Imaging
Research, Sunnybrook Research Institute,
Toronto, Ontario, Canada
2University of Toronto, Department of Medical
Biophysics, Toronto, Ontario, Canada
3PathCore Inc., Toronto, Ontario, Canada
Content:
Recent advancements in high resolution whole slide
scanners and the DICOM standard will spur a wave of
modernization in pathology. Just as whole slide digital
scanners are becoming available from several
vendors, the DICOM standard has also been amended
to include pathological imagery. The latter has the
potential to affect pathology much like it has done for
radiology. This paper discusses newly developed
software that implements DICOM for histopathology
Design:
MR images of 7 radical prostatectomy specimens
were acquired before and after gross sectioning into
4.4mm tissue blocks. One section taken from each
block was stained and digitized. 7-15 homologous
landmarks per midgland section (204 in total) were
identified on MR and digitized histology images.
Positions and orientations of sections within the tissue
blocks were calculated using the best-fit plane to
landmarks identified in MR images, and 4 deformation
2
images and discusses the changes in DICOM which
have allowed the realization of such software.
UMDNJ – Robert Wood Johnson Medical School,
Department of Pathology and Laboratory Medicine,
Center for Bioimaging and Informatics, New
Brunswick, NJ
Technology:
The DICOM standard recently defined two
supplements specifically for pathology: DICOM
supplement 122 which deals with the particularities of
pathological samples (tissue processing, specimen
information, etc.) and DICOM supplement 145 which
standardizes storage and data access methods to
overcome the challenges of high resolution
histological images.
Content:
In modern pathology imaging, the task of precisely
tracing the boundaries of cells and other objects of
interest is required for performing annotations of
specimens, establishing gold-standard image archives
for educational and training purposes and for
preparing ground-truth training sets to test new
quantitative imaging algorithms in computer vision
research applications such as segmentation and
classification. Given the potential impact of
inaccurate renderings, our team has performed a
systematic performance study to investigate the
efficacy of using a commercial, off-the-shelf tablet to
perform these annotations.
Design:
Adoption of DICOM in pathology can simplify the
workflow for Pathologists. Current day workflows
require Pathologists to physically sign out and handle
slides. These challenges can be overcome by DICOM
since it enables viewing and management of digital
slides from any location using computers. Telepathology for instance could help many smaller
hospitals which do not have resident pathologists and
it would allow Pathologists to review cases away from
the lab. Automated computer algorithms could also be
used to assist Pathologists in a variety of tasks such
as tumour margin estimation and disease grading.
Technology:
The onscreen interactive capabilities of tablets
provide a much more intuitive interface for end-users
when compared to conventional keyboard and mouse.
In this study we explore the feasibility of tracing
computer-generated geometric shapes exhibiting
well-defined spatial characteristics and a range of
salient biological objects by comparing tracing
accuracy and repeatability of a standard stylus and
mouse.
Results:
Software has been developed which creates
histopathology DICOM images. The software works
with whole slide scanners and can also be operated
manually by lab technicians. It produces DICOM files
containing all the meta-information available about
the digitized slide (tissue processing steps, specimen
collection processes, specimen type, sampling
methods, stains, fixatives and more). Like other
DICOM images, patient, study, and imaging
parameters are also embedded in the file.
Design:
The first experiment utilized a team of volunteers who
were asked to trace 1-pixel-wide outlines of angular
(triangle and nonagon) and curved (circle and ellipse)
shapes. Each shape was retraced at three different
scales. A Java based program was developed to
record the tracing time and the coordinates of the
traced boundaries. Tracing accuracy is measured by
computing differences between the generated and
traced contours as well as relative error in shape
measurements. In the reproducibility experiment,
several medical professionals were asked to precisely
outline a well-defined microscopic image region five
times.
Conclusions:
Digital pathology enables the conveniences of modern
technologies. Recent availability of whole slide digital
scanners along with recent supplements in DICOM
have made advances in clinical pathology possible.
We have developed software that creates DICOM
compliant histological images to promote adoption of
digital pathology.
Results:
Preliminary results showed that tracing the images
with stylus was not only significantly more accurate in
all measurements than using mouse (p<0.05); it was
also significantly faster (p=5e-11). The stylus also
reduced error in tracing perimeter by 12%. In
addition, tracing with stylus was more reproducible
compared to mouse (p<0.05). Our results also
indicate that it is possible to achieve optimum
combination of speed and accuracy by selecting the
optimum magnification in annotating pathology
images.
We acknowledge support from the Ontario Institute of
Cancer Research (OICR).
Efficacy Studies Investigating the Use of a
Tablet PC to Perform Image Annotations on
Digitized Pathology Specimens
Evita T. Sadimin, MD (sadimiet@umdnj.edu),Wenjin
Chen, PhD, David J. Foran, PhD
3
Conclusion:
We conclude that medical annotation using stylus is
not only feasible but also more intuitive, accurate and
reproducible than using mouse. Further work in this
area will include implementation of tablet-based
pathology annotation applications and better postprocessing techniques to improve shape fidelity.
Design:
Unstained sections of formalin-fixed, paraffinembedded prostate tissue containing discrete areas of
prostatic adenocarcinoma of different histologic
grades were obtained with approval from the
University of Minnesota Institutional Review Board.
Six adjacent 4 micron sections were cut from a tissue
block. One section was stained with hematoxylin and
eosin (H&E). IHC was performed for MKI67, ENO2,
CD34 and ACPP and alignment and protein expression
with IHCMap was computed.
Development of Multigene Expression
Signature Maps at the Protein Level from
Digitized Immunohistochemistry Slides
Results:
The study pathologist created annotations of prostate
cancer areas by Gleason grades present in the block,
separately annotating 3+3, 3+4 and 4+3 areas within
different virtual planes (“layers”) of the reference
slide image file. Additional information used by the
software included a table of IHC stains and their
respective user-input weighting factors. In this study,
the weights for MKI67, EN02, CD34 and ACPP were
chosen as 0.473, 0.035, 0.035 and -0.014,
respectively. The sign of the weights reflects the
current understanding as to each protein’s up- or
down-regulation in aggressive prostate cancer, but
the current magnitudes of the weights are somewhat
arbitrarily chosen to provide a proof a concept.
Stephen C. Schmechel, MD, PhD2,3
(schme004@umn.edu), Gregory J. Metzger, PhD,1
Stephen C. Dankbar, BS,2 Jonathan Henriksen, BS,2,3
Anthony E. Rizzardi, BS,2 Nikolaus K. Rosener2 and
Departments of 1Radiology and 2Laboratory Medicine
and Pathology, University of Minnesota, Minneapolis,
MN
3BioNet, University of Minnesota, Minneapolis, MN
Content:
Molecular classification of diseases based on
multigene expression signatures is increasingly used
for diagnosis, prognosis, and prediction of response
to therapy. Immunohistochemistry (IHC) is an optimal
method for validating expression signatures obtained
using high-throughput genomics techniques since IHC
allows a pathologist to examine gene expression at
the protein level within the context of histologically
interpretable tissue sections. Additionally, validated
IHC assays may be readily implemented as clinical
tests since IHC is performed on routinely processed
clinical tissue samples. However, methods have not
been available for automated n-gene expression
profiling at the protein level using IHC data. We have
developed the methods to compute expression level
maps (signature maps) of multiple genes from IHC
data digitized on a commercial whole slide imaging
system. Areas of cancer for these expression level
maps are defined by a pathologist on adjacent, coregistered H&E slides allowing assessment of IHC
statistics and heterogeneity within the diseased
tissue. This novel way of representing multiple IHC
assays as signature maps will allow the development
of n-gene expression profiling databases in three
dimensions throughout virtual whole organ
reconstructions.
Conclusion:
SigMap is a unique approach to multiplexed analysis
in IHC, and uniquely leverages whole slide imaging.
We successfully developed this method and applied it
to a prostate cancer signature. Work is ongoing to
determine utility in a larger clinical dataset.
In Silico Analysis of Diffuse Gliomas Identifies
Microenvironmental Influence on Key
Transcription Factors and Morphological
Signatures of Glioblastoma
Lee A.D. Cooper, PhD (lee.cooper@emory.edu), Jun
Kong, Fusheng Wang, David A. Gutman, Sharath
Cholleti, Tahsin Kurc, Carlos S. Moreno, Daniel J. Brat,
Joel H. Saltz, MD, PhD
Emory University, Center for Comprehensive
Informatics, Atlanta, GA
Content
Emerging multi-modal datasets that link histology to
genetic and patient endpoints are creating new
frontiers for pathology research. Using data from The
Cancer Genome Atlas (TCGA) and REMBRANDT
projects, we have developed In Silico methods and
informatics tools to investigate the role of tumor
microenvironment on transcription and the
associations of histology and genetics on patient
outcome. We have identified both microenvironmental
Technology:
A software interface, which will be referred to as
SigMap, was written in the Java programming
language, generated IHC signature maps through a
multistep image analysis and image registration
process.
4
influences on the expression of key transcription
factors and the existence of morphological subtypes
of glioblastoma.
Technology
Whole slide imaging presents opportunities to
quantitatively study morphology at large scales. We
have developed a suite of image analysis tools to
segment and characterize millions of nuclei in
gliomas. These tools generate statistical models of
patient morphology that can be analyzed for
comparison with patient outcome and genomics. We
have also investigated necrosis in gliomas using a
human-computer interface to identify necrotic tissue.
Extent of necrosis can then be compared with gene
expression arrays using Significance Analysis of
Microarray to identify genes significantly correlated
with necrosis. All segmentation and markup data are
managed through the Pathology Analytical Imaging
Standards database.
Design
Percentage of necrosis was analyzed in whole slide
images of frozen TCGA GBM tissues. Percentage
necrosis was correlated with gene expression to
identify transcripts that are significantly correlated
with extent of necrosis in 177 slides from 91 patients.
The morphologies of 200 million nuclei were analyzed
in images of permanent TCGA GBM tissues from 162
patients. Nuclear morphology was aggregated into
patient-level profiles that were clustered to identify
groups of patients with similar morphology. Patient
outcome and genetic alterations were analyzed across
clusters to determine cluster characteristics.
Results
The transcription factor CEBP-B/D, known for its role
as a master regulator of the aggressive Mesenchymal
GBM phenotype, was identified as significantly
correlated with extent of necrosis.
Immunohistochemistry analysis revealed that CEBPB/D are hypoxia inducible, suggesting that GBM
phenotype may be a local phenomenon driven by
tumor microenvironment. Morphological analysis of
nuclei revealed several patient clusters.
Conclusions
We have developed a suite of bioinformatics tools for
multi-modal glioblastoma data analysis and
integration and identified morphological signatures of
glioblastoma and microenvironmental influence on
key transcription factors.
5
Thursday, October 6, 2011
Session #2 Decision Support – Natural Language Processing
Kings Garden North
A Clinically Intuitive, Non-Parametric Method
Using Multidimensional Polyhedrons to
Combine the Results of Multiple Laboratory
Tests and Potential Implications for Clinical
Laboratory Decision Support
illustrate potential utility in improving decision support
information provided to clinicians and preventing
unnecessary biopsies.
Results:
tTG IgA and age were important predictors of biopsy
positivity with sufficient data for subsequent analysis.
1652 individuals in our clinical sample had both of
these measures. Probability of a positive biopsy
ranged from 0 to 1. Accuracy of prediction varied
with the density of past data near the test values,
with median difference between upper and lower
95% confidence limits being 0.11.
Brian H. Shirts, MD, PhD
(brian.shirts@hsc.utah.edu)1, Sterling T. Bennett, MD,
MS1,2, Brian R. Jackson, MD, MS1, 3
1University
of Utah, Department of Pathology, Salt
Lake City, UT
2Intermountain Healthcare, Salt Lake City, UH
3ARUP Laboratories, Salt Lake City, UT
Conclusions:
The presented method for multivariate analysis of
clinical results is transparent, conceptually simple, and
provides results that are easy to interpret. A
limitation of this method is the requirement of very
large data sets when many parameters are used. An
advantage is that a diagnostic laboratory could
continuously integrate data from local clinical
encounters into prediction databases, enabling more
precise probability estimates that are tailored to the
local population.
Content:
Realizing the potential of personalized medicine will
require statistical methods to integrate traditional
laboratory data with genetic and clinical information
for diagnosis and risk prediction. Many methods of
multivariate analysis have been applied to
diagnostics, but they are practically opaque and none
has been widely adopted. We evaluate the method of
counting disease and non-disease cases in
multidimensional polyhedral partitions of the
multivariate result space to calculate likelihood ratios
and probabilities of diagnosis. In principle, this
method is a non-parametric analog of multivariate
Gaussian-based likelihood ratios used for maternal
serum screening.
Design and Implementation of Custom
Middleware Based Chemistry Lab
Autoverfication Rules
Technology:
Simple tables of clinical data were used with each
parameter defining a separate dimension in a
multidimensional clinical data space. We used R
statistical software to evaluate the distribution of
cases and controls near specific points in this
multidimensional space defined by a specific set of
patient test values.
William J. Lane, MD, PhD (wlane@partners.org) ,
Frank Kuo, MD, PhD, Neal Lindeman, MD
Brigham and Women’s Hospital, Pathology
Department, Boston, MA
Content:
The accuracy of clinical laboratory test results must
be verified before release. Verification is intended to
detect analytical errors, by comparing the results with
expected values in both health and disease. In many
laboratories, this is done manually by medical
technologists, at great labor cost and with varying
degrees of expertise. Alternatively, a computer may
be programmed for automated review
(autoverification) of results.
Design:
We generated multidimensional polyhedrons,
centered on specified test values and extending a
specified multiple of the median average deviation
from this center for each testing dimension. Counts
of cases and controls contained in this polyhedron
defined the probability of the ‘patient’ being a case or
control. We used a sample of 2053 individuals with
celiac disease workup at Intermountain Healthcare to
6
Technology:
The Chemistry Lab at Brigham and Women’s Hospital
(BWH), which performs ~4.5 million per year,
recently implemented autoverification of results
generated on Cobas 6000 Analyzers (Roche
Diagnostics, Indianapolis, IN), using middleware
(Data Innovation, South Burlington, VT) that connects
to a legacy LIS.
2Madras
Christian College (Autonomous), Tambaram,
Chennai, India
Content:
Malaria is a significant cause of morbidity and
mortality in tropical countries. According to WHO, in
2009, more than 50% of confirmed reported malaria
cases were from India. The gold standard for the
diagnosis of malaria continues to be the manual
microscopic examination of a stained thick smear or
thin smear where at least 100 fields should be
screened at 100x (oil immersion) with at least 8-10
minutes spent per slide. Given the high prevalence
the numbers for primary screening and quality
assurance (10-15% repeat screening) is a mammoth
task requiring scarce resources. Therefore there is a
fear of underreporting and difficulty in quality control
of positive cases. Using technology to assist in
screening of slides by image analysis will introduce a
paradigm shift in the current scenario.
Design:
For each analyte, an autoverification rule decision tree
was designed based on knowledge of that analyte in
health and in disease. The rules were simulated using
real patient data accrued over one month. When
needed, rules were adjusted and re-simulated. Rules
that passed the simulation were encoded in
middleware and tested in silico with “cases” of results
designed to assess the performance of the rules.
Once all testing was complete, rules were
implemented and monitored in production for several
days.
Technology:
MATLAB 2009b (MathWorks Inc, MA, USA)
Results:
Analyte-specific autoverification hold rules were
created that identify if a particular result should be
held. The analyte-specific approach allowed
piecemeal deployment of rules for each analyte (most
common analytes first). The rules evaluate delta
checks (difference between two successive
measurements of the same analyte in the same
patient), instrument error flags, values >3SD beyond
the population mean, and values inconsistent with
other analytes assessing complementary states of
disease or health in a given patient. Currently, 13
different rule prototypes are used, for 48 analytes. In
the absence of hardware errors detected by the
instrument, these rules autoverify 95% of analyte
results and 85-90% of all chemistry tests, resulting in
a significant savings in labor and cost. Rules for the
remaining analytes are in development.
Design:
In India, the common species seen are Plasmodium
falciparum and Plasmodium vivax.
Automatic segmentation techniques were applied to
tiff images of peripheral blood smears
acquired using Leica DFC camera, in order to identify
gametocytes. To identify the
gametocytes, as a first step, the gray image of the
source image were inversed and converted into a
binary image with a proper threshold value, in order
to obtain all the objects in the image (I1). The
objects other than WBCs and RBCs are eliminated to
obtain image I2. The difference between images
I1 and I2 is obtained. The application then used
morphological operations and granulometric
analysis to segment out only the gametocytes. The
number of segmented gametocytes is also
obtained from the segmented binary image.
Conclusions:
Custom chemistry analyte autoverification rules were
developed and implemented in middleware, enabling
a significant savings in labor costs for the laboratory.
These rules should be usable by other institutions
after adjusting the rule parameters to match their
own patient populations.
Results:
We have a prototype application that can identify p.
falciparum gametocytes. The results of the
preliminary validation study will be discussed. The
quality of the images and the presence of artifacts
affects the analysis and these issues should be taken
in order to obtain better results.
Making Malarial Diagnosis More Reliable: Using
Image Analysis for Identification of
Plasmodium Falciparum Gameotcytes
Conclusions:
The above data shows that the possibility to use
image analysis techniques for screening of blood
smears for malaria parasites is possible. More work is
required to refine the algorithms and the methods
used. Initially this technique may be used for quality
control.
Joy J. Mammen MD (joymammen@cmcvellore.ac.in)1,
Maqlin P.2, Feminna Sheeba, MCA2, T. Robinson PhD2
1Christian
Medical College, Transfusion Medicine,
Vellore, India
7
Extraction and Analysis of Data Elements from
Text-based Prostate Cancer Pathology Reports
(<25%) required modification by the user prior to
commitment to the database.
Kavous Roumina, PhD (roumink@ccf.org), Eugene
Farber; Walter H. Henricks, MD
Conclusion:
A simple yet robust, object-oriented system has
enabled the transformation of textual diagnostic,
staging, and prognostic data in prostatectomy
pathology reports into discrete data elements to
enable clinical and translational research and other
analyses. The design paradigm should be deployable
to other types of textual checklist-formatted
pathology reports.
Cleveland Clinic, Center for Pathology Informatics,
Cleveland, OH
Content:
Pathology reports in laboratory information systems
are inherently textual and do not easily lend
themselves to data mining activities. The capability
to extract discrete data elements from text-based
pathology reports would be of great value for
research and analysis. We describe a heuristic-based
approach to categorize and compartmentalize
prostatectomy cancer reports into discrete data
elements ready for further analysis.
Advantages of Structured Data Reporting Using
the CAP Electronic Cancer Checklists (eCC):
The Cancer Care Ontario Experience
Samantha Spencer, MD (sspence@cap.org)1, Gemma
Lee, BSc, PMP2, Jaleh Mirza, MD, MPH1, John R.
Srigley, MD, FRCPC2,3, Tim Yardley, HND2, Aleem
Bhanji, BSc, PMP2, Jeffery Karp, BSc1, Gregory
Gleason, MBA1, Richard Moldwin, MD, PhD1
Technology:
Data analysis component (.NET Framework 3.5,
Microsoft); relational database (Access 2003,
Microsoft); laboratory information system
(CoPathPlus, Cerner).
1The
College of American Pathologists, Deerfield, IL
Care Ontario, Toronto, Ontario, Canada
3McMaster University, Hamilton, Ontario, Canada
2Cancer
Design:
The system consists of two components:
extract/analysis and presentation modules. The
extract/analysis module identifies Key-Value pairs
from textual, checklist-formatted prostatectomy
reports extracted from the laboratory information
system. Keys are checklist headers, e.g. “Gleason
Score”. Values are respective observations in the
report, e.g. “7”. The system analyzes Keys and
Values, extracts pertinent Values, and assigns them to
Keys as discrete elements in the database. The logic
accounts for variations in the tumor report checklist
over the years analyzed. Employing object-oriented
design, Key-Value pairs are organized as reusable
programming codes (“classes”). The system presents
the Key-Value pairings to a reviewer who verifies
assignments by the system and resolves potentially
inaccurate matches resulting from ambiguities in the
source report. Each Key-Value pair is assigned a
color-coded level of “uncertainty”. The reviewer may
access the original report within the application.
Following review, the user “commits” the case to the
permanent database.
Content:
A standard informatics approach for recording and
reporting cancer pathology reports has the potential
to prevent diagnostic errors and omissions, thereby
improving patient care and research. To this end, the
College of American Pathologists produces the
electronic Cancer Checklists (eCC) based on the CAP
Cancer Protocols, widely-recognized as a gold
standard in cancer pathology data collection. The
eCC informatics model allows for standardized data
transfer from anatomic pathology software to central
cancer registries.
Technology:
Checklist questions and answers are entered into an
editing tool for storage in a SQL Server database.
Each checklist is exported as a single XML file from
the eCC database. Vendors use these XML files to
create standardized data-entry forms and reports for
pathologists. The standardized eCC codes stored in
the vendor database can be sent to central registries,
using HL7 messages for mandated cancer reporting.
Results:
The system has processed 3456 historical
prostatectomy reports (2003-2011) from the
laboratory information system. Use of the system has
transformed the text-based, non-discrete diagnostic
and staging data in these reports into discrete data
elements available for analysis and research. A
typical prostatectomy report contained 28 Key-Value
pairs. An average of 6.1 Key-Value pairs per report
Design:
Cancer Care Ontario has adopted the eCC into its
cancer registry data collection system to improve
interoperability and the quality of data collection for
cancer surveillance. Error reduction and increased
timeliness are additional factors driving eCC uptake.
For 2011/2012, Cancer Care Ontario has mandated
8
implementation of 63 eCC templates, involving 110
pathology laboratories throughout the province.
determination of semantic type.
Design:
To have the broadest concept coverage possible, we
configured MojoMapper to use the Unified Medical
Language System. We extracted 232 diagnostic
phrases from 100 sequential pathology reports.
Phrases only containing temporal concepts (“cycle day
24-25”), lacking biomedical concepts (“see
comments”, “no further findings”, “histologic grade,
low grade”), or conceptual content of low retrieval
value (“negative surgical margins”) were excluded
from scoring resulting in 215 phrases used to
evaluate performance. Two physicians (M.H. and
E.G.) evaluated the performance of the system.
Individual points were awarded for each correctly
identified disease relevant concept and negation of
concept.
Results:
As of August 2011, 90/110 hospitals (82%) have
implemented eCC-based synoptic reporting. Ten to
fifteen additional hospitals are slated to participate by
late 2011. In July 2011, nearly 75% of cancer
pathology resection reports were sent to the central
registry using the eCC model. A survey of 970
clinicians found that pathologists, surgeons and
oncologists expressed high satisfaction with the eCCbased reports compared to traditional narrative
reports. The adoption of the eCC has led to more
complete reports, as well as the automated capture of
cancer staging and related data. Detailed reports are
shared with hospitals to provide feedback on
workflow and to assist with quality assurance efforts.
Conclusion:
The Cancer Care Ontario eCC integration experience
is an informatics success story. It demonstrates that
comprehensive implementation of eCC-based
standardized structured reporting improves
information system interoperability, data reporting,
quality assessment, cancer research and cancer care.
Results:
MojoMapper correctly identified 82% relevant
concepts. Inter-rater reliability between the two
physicians was moderate to good (kappa 58%). In
analyzing the MojoMapper errors, we identified a
number of optimizations that will improve the
system’s ability to correctly concept-tag biomedical
concepts for information retrieval of surgical
pathology cases.
Evaluation of A Natural Language Processing
Platform in Concept Tagging of Surgical
Pathology Reports for Information Retrieval
Conclusions:
MojoMapper was built to concept-tag causes of death
in electronic death records and was used as is without
any surgical pathology optimizations. Given this we
find that this system is still able to correctly match a
high percentage of pathology concepts. Using
MojoMapper’s architecture to integrate new
annotators, we plan to add pathology annotators to
address many of the pathology lexicon specific issues
that caused matching failures in the current analysis.
We also plan to implement improved semantic
processing so that “presence of inflammatory cells” is
conceptually equivalent to “presence of an
inflammatory process”.
Radhika Srinivasan, PhD (rsrinivasan@ucdavis.edu)1,
Albert Riedl, MS1, Estella Geraghty, MD2, Michael
Hogarth, MD1,2
1UC
Davis School of Medicine, Departments of
and Laboratory Medicine and 2Internal
Medicine, Sacramento, CA
1Pathology
Content:
Diagnoses in surgical pathology reports are primarily
contained in narrative, “free text” sections that limit
the ability to implement concept-based searching.
This work evaluates MojoMapper, a Natural Language
Processing (NLP) framework, in the task of concepttagging biomedically relevant concepts in surgical
pathology reports to support searching cases by
concept.
Technology:
MojoMapper is a java-based, web services-oriented
NLP platform developed at UC Davis. It implements
stochastic and rule-based text processing that adjusts
behavior in response to semantic and syntactic
features of the input text. The architecture is pipelinebased with annotators operating on pre-processed
text for linguistic manipulation, parts-of-speech
discrimination, negation detection and
9
Thursday, October 6, 2011
Session #3 - Lab Automation
Kings Garden South/LeBateau
Automation of Parasitized Erythrocyte Count by
Microorganisms in Wild Animals
Results:
The automated counting of cells by Digital Image
Processing facilitates the work of professional in the
field of hematology, to be effective, accurate and fast,
reducing costs to the laboratory. Upon completion,
the final product will be used to assess the
interference of parasitic microorganisms in
erythrocytes in relation to anemia, which interferes
with the biodiversity of local fauna.
Denise F. P. Costa (Denise.Costa2@unioeste.br),
Leonilda Santos, MSc, Fabiana Peres, MSc
State University of West of Paraná, Engineering and
Science, Foz do Iguacu, Brazil
Content:
To keeping the biodiversity of wildlife is required to
check the presence of microorganisms on the surface
of erythrocytes can cause anemia if the animal is with
impaired immune systems. The wild animals in
captivity, be influenced by stress and due to their
behavior, many of the diseases can be diagnosed only
by laboratory tests. The quantification of
microorganisms is performed manually and this
activity is an exhaustive, time consuming and more
error-prone, depends on physician skill. In order to
facilitate this activity, a procedure is being developed
using techniques of Digital Image Processing to
perform the quantification of microorganisms
automatically.
Conclusion:
The discovery of hemotropics microorganisms in wild
animals is recent, development manual techniques,
and the development of innovative automated
quantification of microorganisms parasites of
erythrocytes. The monitoring of parasitic
microorganisms is important in maintaining a healthy
population of wild animals which, in turn, helps in
reducing losses to the species, avoiding the imbalance
of regional biodiversity.
Application of Tracking Technology in Anatomic
Pathology
Troy Brown (troy.brown@orionbiosystems.com),
James Dobbs
Technology:
Blood samples of wild animals are collected with
anticoagulant and examined fresh; the capture of
images is made using a microscope Olympus BX-41,
objective 40 x and eyepiece 10 x, an Olympus DP-12
digital camera coupled to the biological microscope
and the camera software to transfer images to the
computer via USB 2.0 for application of the
techniques of Digital Image Processing.
Orion Biosystems, Rolling Meadows, IL
Content:
Misidentification errors in the anatomic pathology
laboratory, while relatively infrequent, can have
disastrous consequences. Such errors can cause
patient inconvenience or harm when additional tissue
collection is needed. Misidentification of cases,
specimens, blocks and slides can delay diagnosis and
treatment or cause treatment to be administered
inappropriately. Technology solutions such as
barcode labeling and tracking are being used but
have not been fully developed. This paper describes
misidentification errors and possible technology
solutions.
Design:
After defining a protocol for capturing images of
Digital Image Processing techniques are applied to
classify the constituent objects. Through a
representation and description to put in evidence the
characteristics of the objects of interest will be held
the counting of each class: total erythrocytes and
microorganisms, being defined the percentage
between the quantitative constituent objects.
Technology:
We analyzed the use of relational databases and
barcoding in pathology laboratories. We also studied
the application of Lean and Six Sigma technologies in
the pathology laboratory. We studied the current and
10
potential effectiveness of electronic cross-checking
and electronic tracking systems using Standard Query
Language databases.
logical next step to interface such orders to the
instrument platform eliminating dual order entry and
errors, improving efficiency, and thereby increasing
capacity.
Design:
We conducted a review of current literature regarding
misidentification of cases, specimens, blocks and
slides in pathology laboratories. We focused on
specific vulnerabilities associated with the anatomic
pathology laboratory setting, and potential remedies
for each. We reviewed the use of barcode technology
in all types of laboratories, including anatomic
pathology.
Technology:
Sunquest CoPathPlus (Sunquest Information Systems,
Tuscon, AZ), Dako Automatic Immunohistochemistry
Stainer with DakoLink interface server (Denmark)
Design:
A bidirectional HL 7 interface was implemented
between Sunquest CoPathPlus version 4.1 (SQCP) and
a Dako Automated Immunostain Platform allowing
immunohistochemistry orders place in CoPath to be
received in the DakoLink instrument control software
markedly simplifying run setup. Slides are required to
be labeled with vendor proprietary asset tags.
Results:
Barcode technology is well-established in a number of
health care settings, including laboratories. Some
anatomic pathology laboratories use barcode
technology, but it is not fully integrated, often
requiring the use of handwriting at the point of tissue
collection and accessioning. Many anatomic
pathology laboratories rely on an unintegrated, multitiered and nonsynchronous approach to maintaining
quality and integrity throughout the tissue handling
process. Paper requisition forms, which can be lost or
placed with the wrong tissue, are still used.
Results:
While slides still need to be manually labeled for runs,
the elimination of dual order entry by automation of
order entry markedly decreases assay run time saving
upward of 360 hour of manual effort per year,
eliminating errors, and improving laboratory
throughput.
Conclusions:
Anatomic pathology laboratories are particularly
vulnerable to misidentification errors because tissue
undergoes several processes from the time of
collection through transcription. Anatomic pathology
laboratories would benefit from a barcode system that
tracks tissue from the time of collection through
transcription. Such a system would reduce human
error and identify misplaced tissue. A paperless
requisition system would also reduce misidentification
errors by reducing the potential for misplaced
requisition forms and errors related to handwritten
information.
Conclusions:
Implementation of an interface between the AP LIS
and automated staining platforms save times and
eliminates errors increasing laboratory capacity and
throughput as well improving patient safety.
The First Conversion of Pathology Assets
to/from an APLIS - Lessons Learned
Lyman T. Garniss, BS (LGarniss@Partners.org), James
Floyd, Ling-Ling
Massachusetts General Hospital, Department:
Pathology Informatics, Boston, MA
Deployment of an Orders Interface Between
CoPathPlus and an Automated
Immunoperoxidase Staining Platform
Content:
Massachusetts General Hospital (MGH) Pathology
Service recently implemented Sunquest's CoPath Plus
version 5.0 This was not the first implementation of
Sunquest CoPath Plus v 5.0 but it was the first time
that large numbers of unique cassettes and slides
were converted from a foreign system into CoPath
Plus (or any APLIS) and could be recognized and
processed in the new APLIS.
J. Mark Tuthill, MD (mtuthil1@hfhs.org), Michael
Czechowski, Kathleen M. Roszka, HTL, ASCP,
Mehrvash Haghighi MD
Henry Ford Hospital, Division of Pathology
Informatics, Detroit, MI
Content:
Immunoperoxidase staining of tissues has become an
important routine aspect of pathology practice
resulting in the development of automation
technology to perform these assays. As orders are
typically created in the Anatomic Pathology LIS, it is a
During the implementation a team of dedicated
specialists from Sunquest, Massachusetts General
Hospital and Partners Healthcare Systems Information
Systems converted over 20 years of Anatomic
Pathology data. This included over 2.4 million cases,
11
1.8 million unique tissue cassette numbers and 4.2
million unique slides numbers.
Technology:
The technology to accomplish this project included;
The existing PowerPath APLIS database with unique
asset identifiers for specimen cassettes and slides;
programs and formatting rules supplied by Sunquest,
several programs, filters and conversion utilities
developed at Partners Healthcare Systems
Information Systems and the receiving CoPath Plus v
5.0 database.
Design:
Sunquest's CoPath Plus version 5.0 has the ability to
accept, store and make available " foreign identifiers
". In the case of the MGH data conversion the
PowerPath native identifiers for cassettes and slides
were moved to the foreign identifier fields in the case
records of CoPath.
Results:
A newly installed APLIS (CoPath Plus) was able to
open, access and process new tests and orders on
cassettes and slides from a foreign APLIS for the first
time. The process was not easy and many lessons
were learned while converting the assets and cases.
Conclusion:
This was the first time that a new APLIS was used
convert, store, access and order new tests on
cassettes and slides from a foreign APLIS. The
"converted " assets can be used to order new stains,
new slides, order whole slide imaging, and order
molecular tests years after the case has been
finalized. Slides and cassettes can also be more easily
tracked and retrieved.
12