Mammography classification using modified hybrid SVM-KNN

Today leading cause of cancer deaths for women is the Breast cancer. For early and accurate detection of breast cancer, mammography is found to be the most reliable and effective technique. In this context, computer aided diagnosis of breast cancer from mammograms is gaining high importance and priority for many researchers. In this paper machine learning based mammogram classification using modified hybrid SVM-KNN is proposed. The idea is to map the feature points to kernel space using kernel and find the K nearest neighbors among the training dataset for a given test data point. Doing this we narrow down the search for support vectors. Mammogram images are preprocessed and region of interest is extracted using Fuzzy C Means clustering and Active Counter technique. GLCM (grey level covariance matrix) based texture features are extracted from segmented ROI. These features are used to train modified hybrid SVM-KNN classifier proposed by authors. The trained classifier is used to classify breast tissues in normal/abnormal classes and further abnormal class into benign/malignant. Proposed technique is experimented on two standard MIAS and DDSM databases. The proposed classifier reports classification accuracy of 100 % for DDSM and 94% for MIAS database for benign and malignant class. Results are compared with SVM, KNN and Random Forest classifiers.