Computer-vision system for the detection and characterization of masses for use in mammographic screening programs

Published

Journal Article

Studies have shown that early detection of breast cancer through periodic mammographic screening of asymptomatic women could reduce breast cancer mortality by 30-50%. However, screening yields a high volume of mammograms requiring interpretation. In addition, accurate characterization of detected masses is an important task of radiologists in order to reduce the number of unnecessary biopsies. Considerable misclassification of masses still occurs. In fact, on average, only 20-30% of masses referred for surgical breast biopsy are actually malignant. As a potential aid to radiologists in mammographic screening programs, we are developing a computer-vision system for the detection and characterization of masses in digital mammograms. This system includes a detection subsystem and a characterization subsystem. Motivated by the systematic methods of viewing mammograms used by radiologists, the detection system is designed to analyze the deviation from the architectural symmetry of normal right and left breasts, and employs gray-level histogram analysis, a bilateral-subtraction technique and run-length linking of multiple subtraction images to locate potential masses. False-positive detections are further reduced by various feature-extraction techniques. The characterization system employs various image analysis techniques, such as the measurement of margin spiculation of masses, in order to estimate the likelihood of malignancy.

Duke Authors

Cited Authors

  • Yin, FF; Giger, ML; Doi, K; Vyborny, CJ; Schmidt, RA; Metz, CE

Published Date

  • January 1, 1991

Published In

Volume / Issue

  • 1396 /

Start / End Page

  • 2 - 4

International Standard Serial Number (ISSN)

  • 0277-786X

Citation Source

  • Scopus