Publications

Abstract : Mammograms are customarily employed as one of the reliable computer-aided detection (CAD) techniques. We propose an efficient modified graph-cut (GC) segmented, grey-level co-occurrence matrix (GLCM)-based support vector machine (SVM) technique, for classification of tumour. In this work, SVM classification was carried out in single-view mammograms, subsequent to preprocessing, GC segmentation and GLCM feature extraction. Segmentation of pectoral muscles was done first, followed by segmentation of tumour, using kernel space mapped normalized GCs. We believe this process is the first of its kind used in mammograms. A suitably large number of features were extracted from GLCM, using Haralick method, which in turn increased the training efficiency. The proposed method was tested on 322 different mammograms from Mammographic Image Analysis Society (MIAS) and hence successfully verified to provide efficient results. High accuracy rates were achieved by combining best methods at each stage of diagnosis.