A Novel Approach for Mammogram Classification

One of main reason for increase in mortality rate of woman is breast cancer. Accurate early detection of abnormalities in breast tissues seems to be the only solution for diagnosis. These detected abnormalities need to be classified for further investigation. In breast cancer research various classification techniques have been proposed to reduce the diagnostic test false positives because of the subtle appearance of the breast cancer tissues. Accurate early classification of mammogram is essential to reduce diagnostic test false positives. Hence, an attempt is made to presents an automated classification technique for mammogram. This work is divided into four steps: Pre-processing, Feature Extraction, Feature Selection and classification. In this approach cropping and resizing is used as a preprocessing technique. In next step Texture, wavelet and PCA features are extracted form mammogram, cuckoo search algorithm is enhanced to select optimal feature split points. Finally, classification is achieved using association rule agreement-based classifier model. It classifies mammogram as normal or cancerous using association rules generated from selected features. The experiments are carried out on Mammographic Image Analysis Society (MIAS) database and performance of proposed method is validated over existing work using accuracy, sensitivity and specificity measures. The proposed technique has achieved accuracy 0.8289, sensitivity 0.9333 and specificity 0.7273. Results indicate the usefulness of proposed method in classification task.