Skip to Main Content
Breast cancer is a leading cause of deaths in women. Mammography is considered as the most effective technology presently available for breast cancer screening, being very effective in the detection of clustered microcalcifications which are considered as one of the most important findings associated to the existence of breast cancer. A computer aided diagnosis (CAD) system named ldquoHippocrates-mstrdquo has been already developed in the lab based on detailed analysis and evaluation of related features of microcalcifications (individually and in clusters). Preliminary evaluation results have shown that the system achieves high levels of sensitivity, while suffering from low specificity. For this reason, our current studies aim to a methodology refinement which will lead to optimized classification results. In this paper, we focus on obscure diagnostic cases classified by the radiologists as BI-RADS 3. In such cases, although short-term re-examination is normally advised, radiologists and physicians usually have strong doubts about their recommendations. We tested the performance of two classifiers embedded in the proposed CAD system using a dataset of 63 (57 benign and 6 malignant) mammograms, all classified as BI-RADS 3 and biopsy proven. The sensitivity achieved by the first one (the default Hippocrates-mst classifier) is as high as 100%, classifying correctly all the malignant cases. As far as the benign cases are concerned, systempsilas specificity is 35.09%. Using the second classifier (a rule based and SVM hybrid classifier) the specificity increases to 63.16% with a cost of sensitivity decrease to 66.67%.