19
Developing a Validated Tool For Evaluating Whole Slide Images

Developing a Validated Tool For Evaluating Whole Slide Images

Embed Size (px)

Citation preview

Page 1: Developing a Validated Tool For Evaluating Whole Slide Images

Developing a Validated Tool For Evaluating Whole Slide Images

Page 2: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Authors

Dana Grzybicki, MD, PhDEvaluation Team – Principal Investigator

Russell Silowash, BSEvaluation Team – Research Analyst

Robb Wilson, MAEvaluation Team – Project Manager

Leslie Anthony, MAUPMC IMITS Telepathology Project – Project Manager

Page 3: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Background• Through appropriations in the

defense-spending bills for 2002 and 2004, the University of Pittsburgh Medical Center (UPMC) and the United State Air Force Medical Service (AFMS) created a partnership called the Integrated Medical Information Technology System (IMITS) Program

• Telepathology is a branch of the IMITS program that implements and validates digital pathology practices

Page 4: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Introduction

• Purpose of evaluation research– Independent examination of questions related to:

• Technical validity• Feasibility• Effectiveness

• To our knowledge, there are no generally available validated tools for evaluating WSI cases

• The UPMC Digital Pathology Imaging Group is working on the validation of a unique evaluation tool

Page 5: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Participants

• The validation of this tool was part of our telepathology evaluation project involving 5 UPMC pathologists– 2 pathology fellows– 3 staff pathologists with training in GU pathology

Page 6: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Case Selection

• 30 difficult prostate biopsy foci

Diagnosis Number of Cases

Adenocarcinoma 12

Atypical 6

Atypical PIN (ATYPIN) 1

High Grade PIN 3

Benign 8

Page 7: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Page 8: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Page 9: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Hypotheses

• Content Validity– The assessment is asking the proper questions

for the study at hand

• Internal Validity– There will be a positive correlation between the

number of slides/images in a case and the time needed to complete the case

Page 10: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Hypotheses (continued)

• External Validity– Whole slide image quality will be positively

correlated with glass slide quality

• Construct Validity– There will be a negative correlation between the

diagnostic confidence of a participant and the case complexity rating

Page 11: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Content Validity Results

• Content validity has been obtained by gaining feedback from pathologists that are part of the Digital Pathology Imaging Group (DPIG)

Page 12: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Internal Validity Results

• Statistically significant for only one participant (r2=0.327, p<0.01)

• Our inability to demonstrate internal validity for most of the participants was most likely due to time categorical variables that were too broad

Page 13: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

External Validity Results

Average Image vs Slide Quality Correlation

y = 0.5505x + 0.6673

0

0.5

1

1.5

2

2.5

3

3.5

0 0.5 1 1.5 2 2.5 3 3.5

Image Quality Rating

Slid

e Q

ual

ity

Rat

ing

• A positive correlation exists between WSI quality and glass slide quality

•There were statistically significant positive correlations for 3 of the 5 subjects

Page 14: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Construct Validity ResultsWSI Average Case Complexity vs Diagnostic Confidence

y = -0.6775x + 3.7803

1

1.5

2

2.5

3

3.5

0.9 1.4 1.9 2.4 2.9

Case Complexity

Co

nfi

de

nc

e In

Dia

gn

os

is

• Negative correlation exists in WSI phase of study between confidence in diagnosis and case complexity

• 3 of 5 pathologists had statistically significant positive correlations in the WSI Phase

Page 15: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Construct Validity Results (continued)

Glass Average Case Complextiy vs Diagnostic Confidence

y = -0.4087x + 3.3959

0.9

1.4

1.9

2.4

2.9

3.4

0.9 1.4 1.9 2.4 2.9

Case Complexity

Co

nfi

de

nc

e in

Dia

gn

os

is

• Negative correlation exists in the Glass phase between case complexity and diagnostic confidence

• 3 of 5 pathologists had statistically significant positive correlations in the Glass Phase

Page 16: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Summary

• We are able to establish content, external and construct validity, however internal validity at this point has not been definitively established.– Low internal validity could be due to time

categories being too broad

Page 17: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Next Steps

• Change the categories for the time variableTime to Complete (Minus

Interruptions)

•Less Than 15 Minutes

•15 – 30 Minutes

•30 – 45 Minutes

•45 – 60 Minutes

•Over 60 Minutes

Time to Complete (Minus Interruptions)

•Less Than 30 Seconds •7 – 10 Minutes

•30 Seconds – 1 Minute •10 – 12 Minutes

•1 – 2 Minutes •12 – 15 Minutes

•2 – 5 Minutes •Over 15 Minutes

•5 – 7 Minutes

• Implement an automatic timing solution

Page 18: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Next Steps (continued)

• Continue the validation of this tool by obtaining additional data and testing internal validity utilizing the modified time variables

Page 19: Developing a Validated Tool For Evaluating Whole Slide Images

University of PittsburghDepartment of Biomedical Informatics

Funding

• This work was supported by funding from the U.S. Air Force administered by the U.S. Army Medical Research Acquisition Activity, Fort Detrick, Maryland (Award No. W81XWH-04-2-0030 and Contract No. DAMD 17-0302-0017). The content of the information does not imply U.S. Air Force or Government endorsement of factual accuracy or opinion