Computer-Aided Classification of Malignant and Benign Lesions on Mammograms.
Abstract
In the first year of our project, we have made progress in (1) database collection for mammograms containing masses and microcalcifications; (2) segmentation, transformation, and feature extraction from regions of interest on mammograms containing masses; (3) classifier design for the classification of lesions as malignant or benign; and (4) evaluation of algorithms for classification of masses on mammograms. We have shown that the new image transformation and feature extraction methods improve the mass classification accuracy significantly. We have also shown that, compared to standard feature selection methods, significant improvement can be obtained at the high-sensitivity region of the receiver operating characteristic curve by using a genetic algorithm-based feature selection method.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1997
- Accession Number
- ADA328057
Entities
People
- Berkman Sahiner
Organizations
- University of Michigan