AN OPTIMAL NEURAL NETWORK BASED CLASSIFICATION TECHNIQUE FOR BREAST CANCER DETECTION

Breast carcinoma is one of the most signifcant health diseases in the world. Early identifcation of breast carcinoma could be benefcial for in time treatment of the disease. This study presents an efcient classifcation method for benign and malignant breast cancer. The proposed method employs an optimal feature classifcation employing artifcial neural network. The proposed architecture has fve input nodes, two hidden layers with eight neurons each and one output node. Five features (cluster thickness, uniformity of {cell size, cell shape}, marginal attachment and radius of circle enclosing the abnormality) are nominated as input features to the ANN to predict the benign or malignant breast carcinoma. The network is trained, tested and validated on data bases that comprises of a set of previously extracted features provided by Wisconsin and Essex Universities. For the established neural networks comparative analysis is performed to study the optimum parameters required for prime mass classifcation. The execution of suggested methodology is estimated using ROC curve. The accuracy rate of developed method is 93.1% or 0.93 with sensitivity of 0.99 and specifcity of 0.83 according to the receiver operating characteristic (ROC)