Transcript View Slide Presentation - Association for Pathology Informatics

Automated Annotation of a Novel Virtual
Tumor Neuropathology Report
Database:
A Paradigm for Creating Patient-Friendly
Reports
Bei Hu1, Tomer Schechori1, Steven S. Silver B.Sc.1,2,
Alberto Marchevsky M.D.2, X. Fan M.D.2, William H.
Yong M.D.1,2; Department of Pathology
(Neuropathology), UCLA Medical Center1 and
Department of Pathology, Cedars-Sinai Medical
Center2; David Geffen School of Medicine at UCLA,
Los Angeles, CA 90048
Goals
• Create a pathology report generator that
writes simulated brain tumor reports (with
fictional patients) suitable for informatics
experimentation
• Demonstrate that annotating these
simulated reports in an automated
fashion is feasible
The Need for Annotated Pathology
Reports
•Pathology reports contain obscure terminology. For
example…
•Dysembryoplastic neuroepithelial tumor (WHO Grade I)
•Atypical teratoid/rhabdoid tumor (WHO Grade IV)
•Automated addition of explanations oriented to the reader,
e.g. clinicians and patients, may improve understanding
IS THIS PATHOLOGY REPORT IN ENGLISH OR QUOKKA, MATE??
QuokkaA nocturnal marsupial
unique to Western Australia.
Utility of a Simulated Pathology Report
Database
THE PATIENTS ARE FICTIONAL, NON-EXISTENT
• Test software safely (De-identification, data mining,
search engines etc)
• Provide cases for implementation of pathology
information systems
• Education of medical personnel
• Experimentation with pathology report formats and
functions
Pros and Cons of a Simulated Pathology
Report Database
PROS (“IN VITRO ENVIRONMENT”)
- No real patient data and privacy (e.g.
HIPAA) concerns
- Control the content in the reports
- Scalable to large numbers of reports
CONS
- Different complexity than a real database
Pathology Report Generation
• Data elements are entered into MS Access
• Programming in Visual Basic for Applications (VBA)
pulls together data elements
• Final output as Microsoft Word files
Microsoft
Access DB
VBA
Simulated reports
in MS Word
Data Elements for Synthesizing
Simulated Brain Tumor Reports
• 83 brain tumors types from the WHO classification
• Microscopic descriptions and Comments (Standard)
• Tumor layman explanations (Annotated Reports)
• Locations in brain (e.g. Left occipital lobe)
• Median or mean age of each tumor
• World Health Organization (WHO) grades
• Incidence of tumors (Central Brain Tumor Registry of
the United States 2005-2006 statistical report)
A READER-APPROPRIATE EXPLANATION IS
CREATED FOR EACH TUMOR
THE 83 TUMOR TYPES ARE MAPPED AGAINST 84 BRAIN SITES
Pathology Report Generator
Menu
# of reports
to generate
Vary
proportions
of tumors
Annotation
feature, can
be turned
on or off
Simulated Report
Annotations
Writing in blue indicates annotations
Large databases of reports are easily
created
•13,999 SIMULATED BRAIN TUMOR PATHOLOGY
REPORTS (1.0-3.5 minutes/report)
•CBTRUS Proportionate Database: 10,000
•Proportionate plus rare cases forced in: 1,000
•Proportionate with/without annotation: 2,002
(1001 standard, 1001 annotated)
•Equal distribution Database: 997
Distribution of Tumor Types in the Proportionate
Database is appropriate
Greek
Conclusions
1. This prototype Simulated Pathology Report Generator
can produce simulated brain tumor reports in large
numbers.
2. The brain tumor reports mimic real pathology reports in
content but without the privacy concerns.
3. The automated conversion of data elements in the
standard report to a reader-appropriate version is
feasible.
4. We suggest that Simulated Report Generators and
Simulated Report databases have substantial potential
for many types of informatics experimentation
Acknowledgments
•
•
•
•
•
Bei Hu
Tomer Schechori
Xuemo Fan
Alberto Marchevsky
Steven Silver the Computer Cowboy
Sponsored in part by NCI U01 CA91429
William Yong M.D.
Dept of Pathology and Laboratory Medicine
(Neuropathology)
UCLA Medical Center
David Geffen School of Medicine at UCLA
[email protected]
(310) 825-0825