Classification of tumors and masses in mammograms using neural networks with shape and texture features

We propose an approach using artificial neural networks to classify masses in mammograms as malignant or benign. Single-layer and multi-layer perceptron networks were used in a study on perceptron topologies for pattern classification of breast masses. The boundaries of 108 breast masses and tumors were manually delineated and represented by polygonal models for shape analysis. Ribbons of pixels were extracted around the boundary of each mass. Three shape factor measures based on the contours, and fourteen texture features based on gray-level co-occurrence matrices of the pixels in the ribbons were computed. Various combinations of the features were used with perceptrons of several topologies for classification of benign masses and malignant tumors. The results were compared in terms of the area A/sub Z/ under the receiver operating characteristics curve. Values of A/sub Z/ up to 0.99 were obtained with the shape factors, whereas texture features provided A/sub z/ up to only 0.63.