On detecting abnormalities in digital mammography

Abstract

Breast cancer is the most common cancer in many countries all over the world. Early detection of cancer, in either diagnosis or screening programs, decreases the mortality rates. Computer Aided Detection (CAD) is software that aids radiologists in detecting abnormalities in medical images. In this article we present our approach in detecting abnormalities in mammograms using digital mammography. Each mammogram in our dataset is manually processed — using software specially developed for that purpose — by a radiologist to mark and label different types of abnormalities. Once marked, processing henceforth is applied using computer algorithms. The majority of existing detection techniques relies on image processing (IP) to extract Regions of Interests (ROI) then extract features from those ROIs to be the input of a statistical learning machine (classifier). Detection, in this approach, is basically done at the IP phase; while the ultimate role of classifiers is to reduce the number of False Positives (FP) detected in the IP phase. In contrast, processing algorithms and classifiers, in pixel-based approach, work directly at the pixel level. We demonstrate the performance of some methods belonging to this approach and suggest an assessment metric in terms of the Mann Whitney statistic.