Tumour Detection in Double Threshold Segmented Mammograms Using Optimized GLCM Features fed SVM

A mammogram is an x-ray image used for early stage breast cancer detection. In this work a novel and efficient method is proposed for tumour detection in mammograms using optimal GLCM features fed to SVM classifier. Optimal feature set from a set of GLCM features are selected using genetic algorithm. The proposed system consists of steps such as pre-processing, threshold segmentation, GLCM feature extraction, optimal feature subset selection using a genetic algorithm and classification using a support vector machine model. After pre-processing the image, binary thresholding is used to remove unwanted objects like high intensity labels from the image and get the required region of interest. Morphological dilation is performed on the segmented region and GLCM features are extracted. Using genetic algorithm, a subset is computed from a given set of features which gives the best classification rate. These features are used for the training of a support vector machine classifier. The trained SVM is used to classify new input images as normal or cancerous. The proposed methodology is tested on the mini MIAS database.