Expert Learning System Network for Diagnosis of Breast Calcifications

Publisher Website
Google Scholar
Abstract
Breast calcification diagnosis was studied by using clinical findings and computerized image processing of a mammogram in a network of trained expert learning systems (Outcome Advisor [OA]). The system was tested with records not used for training and performance was compared with radiologist. The network was 72% accurate in classifying clusters of calcifications as malignant or benign over a set of test cases radiologists had considered "hard-to-diagnose calcifications," and referred for biopsy. The radiologists had decided to conduct biopsy by selecting an equal number of positive and negative cases for the test group; thus the radiologists' performance with respect to categories of benign versus malignant was constrained to be 50/50. Statistical analysis shows only a 2% probability that the observed accuracy of 72% was a chance performance in recognizing whether a cluster is benign or malignant. The feasibility of developing a network of OAs for diagnosing breast cancer integrating digital image processing of mammograms is promising.

This publication has 0 references indexed in Scilit: