COMPUTER ANALYSIS OF THE MAMMOGRAMS ORIENTED ON BREAST CANCER DETECTION

Abstract: The aim of our study is to create a computer diagnostic system (CDS) for breast cancer (BC) recognizing. As a software diagnostic tool we have used Fahlman’s cascade correlation neural network (FNN). The FNN was trained by the vector of features – parameters extracted from mammographic images of healthy tissue (H), tissue with benign (BT) and malignant tumour (M). To prepare digital data for the NN new, original methods of transformation the mammograms were proposed: Algorithm of Summing up the Rows (ASR) operating on the binarized picture, and analysis of extracted features from Region Of Interest (ROI). There were lots of parameters optimized in previous research; in this work we present discussion of the level of image binarization in ASR method, and discussion of shape, size and number of analyzed features in ROI method. As the input data for neural network decision making system we have used six parameters calculated from a region of interest (ROI method), and four parameters calculated by the “summing up the rows algorithm ” (ASR). We have used all mentioned parameters to create the best combination of features and find the best representative vector, and get the highest correctness of recognition. The final diagnostic system could us obtain correctness of the mammogram interpretation about 92 % for healthy tissue, 89 % for benign and 91 % for malignant tumors