An expert system for mammogram analysis

This thesis describes an expert system which analyses mammograms. The system uses image processing techniques and rule-based decision making scheme when analyzing digitized mammograms. The analysis of mammograms is performed in two stages. First, the expert system detects possible deformities (regions of interest) by using a multi-resolution image processing algorithm which performs image segmentation based on fuzzy pyramid linking. Second, The expert system classifies the detected regions into three classes of objects: nontumor, benign tumor, and malignant tumor. Classification is performed in a hierarchical fashion using a rule-based decision making scheme that considers size, shape, edge information, and textural characteristics of the detected deformities. Shape analysis employs a tumor model and compares that model with the detected deformities by measuring the similarity between the model and a detected deformity using least mean square fitting. Local edge processing, using the Marr-Hildreth edge detector, is used in this work to find the true edges of tumors. The distance variance between the true edge of a tumor and the model is used to classify benign and malignant tumors. Classification is attempted at each level of the hierarchy, and the process continues through the hierarchy only when the decision regarding the class of the objects cannot be made with certainty.

The proposed algorithms for mammogram analysis in this work are reliable Vll and efficient. The expert system can automatically analyze mammograms without any specific a priori information about mammograms.