A Front End Processor for Mass Detection and Classification in Digital Mammograms

Abstract

Breast Cancer is known to be the most dangerous disease for women in the world. Early detection of the cancer is very important. The analysis of medical images, especially mammograms, is one of the early steps in cancer diagnosing Automatic computer aided imaging systems are highly required for analysis and diagnosing. This research work aims at constructing a front-end processor for mass classification in digital' mammograms. General adaptive algorithms are proposed to be used in the system. Stages of the system include noise equalization, noise removal, contrast enhancement, mass detection and mass classification. Every stage of the proposed front end system is surveyed to find the most general, considerably accurate and efficient algorithms. Chosen algorithms are then implemented and compared to select the best. A suggestion of a complete front end system is then proposed. The segmentation part of the proposed system includes a new adopted segmentation module based on the watershed of texture energy images. The research also introduces a novel classifier, namely, the discrete hidden Markov tree of Ranklet transforms, which proves a high reliability in training nad classification. A theoretical analysis of the effect of the system components on the classification accuracy is also included in the research.

The system is implemented and applied to the well-known MIAS database of mammograms. The classification part of the system gives 96% accuracy when classifying masses to normal and abnormal classes. The overall system reaches 83% successfulness in behaving like an expert physician; a result that outperforms other related systems