Database-Aided Diagnosis in Digital Mammography
Abstract
The purpose of this project is to refine and test a "visual neural network," combined with a mammographic database, for its ability too help radiologists reduce the number of benign biopsies in mammographic screening without increasing the number of missed cancers. We refer to this as a "mapped database diagnostic system." The unique features of this system are a)its exploitation of a clinically proved database of mammograms for enhanced diagnosis, b)automatic selection of highly discriminating mammographic features, and c) a two- dimensional "relational map" for enhanced browsing through the mammographic database. Years 1 and 2 of this project will be devoted primarily to retrospective studies, Years 3 and 4 primarily to clinical studies. In Year 1, reported here, we completed our tests of the mapped database diagnostic system on regions of interest containing microcalcifications. In these tests four radiologists, acting independently on 100 ROIs, recommended biopsies for 18, 15, 28, and 18 benign ROIs, and misdiagnosed cancers in 11, 12, 7, and 8 ROIs, respectively. Interaction with the mapped-database system reduced the numbers of benign biopsies to 11, 8, 18, and 10 cases, and the numbers of misdiagnosed cancers to 8, 7, 4, and 3 cases, respectively. Statistical analysis indicated that 3 of the radiologists achieved significant improvements at p < 0.02; the fourth achieved an improvement atp < 0.07. These results suggest that radiologists using this system can statistically significantly reduce the numbers of benign biopsies and misdiagnosed cancers. The enhanced diagnostic benefits seems to apply to mammographically certified radiologists over a wide range of capabilities and durations of post-certified experience.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2000
- Accession Number
- ADA386621
Entities
People
- Jack Sklansky
Organizations
- Charles R. Drew University of Medicine and Science