Unsupervised fuzzy cognitive map in diagnosis of breast epithelial lesions

Breast cancer can be classified in terms of histology, considering their structural and cellular pattern. Histological misdiagnosis will cause wrong treatment and in some cases, it will lead to patient´s death. In this paper, application of Fuzzy Cognitive Maps (FCMs) as a modeling and classifying tool to diagnose the type of breast lesions is studied. Based on ten major histopathological features, 86 subjects were classified by expert pathologists according to World Health Organization (WHO) global system into three groups including UDH, ADH and DCIS. In this study, considering the inherent nature of FCM, the physician´s knowledge is used to construct and modify FCM model in diagnosing the type of the lesion and the resulted FCM is trained by Nonlinear Hebbian Learning (NHL) method. The classification is made based on histopathological features, which are the same concepts of FCM model. The classification accuracy for UDH is 88% and for ADH & DCIS is 86%.