Increasing the Accuracy of Mammogram Interpretation

Abstract

A computer-based decision-support system and automated report writer for mammography were refined and evaluated in a clinical setting. For each case, the computer solicits from a radiologist quantitative ratings of a checklist of perceptual features of mammograms that have been determined to be most diagnostic and most informative for a therapeutic recommendation. The ratings are converted both to a probability of malignancy by a statistical prediction rule and to a prose report of findings by computational linguistic techniques. Overall, the decision aids served to increase accuracy less than in previous laboratory studies, but a substantial gain was shown for the more difficult cases that present as calcifications. Other substantive and methodological advances show the continuing promise of this approach for further development and use in practice. The version of a report writer that was evaluated was seen to need further development along specifiable lines, but demonstrated its ability to improve on the usual dictated report in several respects. Together, the decision-support system and automated report writer are expected to find cost-effective use in a larger radiological and medical information system. The ranked list of certified, scaled perceptual features developed by this approach, and the predictive value of their merged ratings, should also be valuable as a teaching tool.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 01, 1998
Accession Number
ADA366631

Entities

People

  • John A. Swets

Organizations

  • BBN Technologies

Tags

Communities of Interest

  • Biomedical

DTIC Thesaurus Topics

  • Accuracy
  • Artificial Intelligence
  • Breast Cancer
  • Cancer
  • Computers
  • Databases
  • Decision Support Systems
  • Diagnostic Imaging
  • Health Care
  • Health Services
  • Information Science
  • Medical Personnel
  • Neoplasms
  • Physicians
  • Psychology
  • Regression Analysis
  • Statistical Analysis

Fields of Study

  • Medicine

Readers

  • Instructional Design and Training Evaluation.
  • Oncology and Biomarker-Based Cancer Detection.
  • Systems Analysis and Design