Fuzzy Cognitive Maps and a New Region Growing Algorithm for Classification of Mammography Images

Mammography is one the efficient ways to diagnose breast cancer in its early stages. In this paper we have proposed a novel approach based on fuzzy cognitive maps (FCMs) in order to classify breast tumors in one of the malignant or benign types. First of all, mammography images are preprocessed for reduction of noise and removing artifacts. In the next step the tumor is segmented with a new segmentation algorithm which is based on region growing method. After segmentation, 27 features describing texture and boundaries of segmented area are extracted and feature selection is performed with respect to the classifier validation error. Finally, FCM is trained based on square error of training set. To assess the generalization ability of our proposed method a dataset of digital database for screening mammography (DDSM) containing 149 benign and 148 malignant cases is used. The obtained AUC for test set is 0.851.