Classification of Intraductal Breast Lesions Based on the Fuzzy Cognitive Map

Abstract

The diagnosis of intraductal breast lesions has always been considered one of the most challenging issues in breast cancer treatment. These lesions are divided into three groups: usual ductal hyperplasia (UDH), atypical ductal hyperplasia, and ductal carcinoma in situ. Investigations indicate that 50 % of surgeries performed on intraductal lesions have been unnecessary. Since patients afflicted with UDH do not need any surgery, appropriate classification of these lesions is very important. In this paper, a new method based on the fuzzy cognitive map (FCM) has been introduced for the screening and separation of UDH from the rest of intraductal lesions. For this purpose, 86 patients in the Shahid Beheshti Hospital of Isfahan were studied. Ten of the most significant characteristics needed for the screening of these lesions were extracted by three pathologists and used as the main concepts of FCM. The degree of impact of these features on one another and the impact of each feature on the final concept were extracted as linguistic expressions and in the form of fuzzy rules. After defuzzification, the matrix of FCM weights was obtained. An overall accuracy of 95.35 % was obtained on the entire database. The obtained results indicated that the proposed FCM not only enjoys a high level of accuracy, but also is capable of offering an acceptable FNR while reducing the value of FPR to zero. Here, by combining FCM with support vector machine, we have been able to improve the concordance among pathologists on inter- and intra-observations.