A New Scheme of Mammographic Masses Classification Based on the BI-RADS Lexicon

Computer-aided diagnosis (CAD) systems are intended to assist the specialist in the interpretation of the medical images aiming to support the clinical conduct. In mammography breast tumor classification, the CAD systems commonly provide a benign and malignant classification disregarding provide more relevant information for the radiologist that explains the internal criteria that take the classifier to classify a tumor. Therefore, it is necessary to give more information compatible with the radiological language according to the BI-RADS standard. In this paper, we proposed a new scheme of mammographic tumors classification that includes the BI-RADS lexicon modeling based on malignancy probabilities to get a new feature vector to represent the tumor and explain the internal criterion considered by the classifier. For comparison purposes, a CAD system based on a typical pattern recognition scheme (called conventional CAD) is also evaluated. The results suggest that the proposed method obtained an accuracy of 90%, whereas the conventional CAD reached an accuracy of 88%. According to the results, the proposed method is suitable for the histopathological classification. Besides, it provides more information about the radiological language used by the radiologist in terms of the BI-RADS lexicon.