Mammographic mass detection based on robust learning algorithms

Abstract

This thesis provides an in-depth investigation to develop advanced machine learning algorithms for automatic breast mass detection in digitized mammograms. The work consists of the establishment of software system to process the digitized mammographic images automatically. According to the character of masses and the background breast tissue in digitized mammograms, two image segmentation algorithms based on information theory and a new classifier based on statistical learning theory are proposed. The main contributions of this thesis include: the proposal DACF method in the segmentation of circumscribed mass, the investigation of RIC algorithm in the segmentation of masses that are embedded in glandular or dense glandular breast tissue, the study of VSVM for mass pattern analysis that are embedded in fat, glandular or dense glandular breast tissue with various shapes by a semi-automatic approach. In summary, novel and robust learning algorithms for the approaches of fully and semi-automatic detection of breast masses in digitized mammograms are proposed. The achieved results are hoped to be useful for the further investigation of automatic processing of mammograms and facilitate the clinical decision.