Classification of microcalcifications in mammograms using artificial neural networks

An advanced method is described for the classification of malignant and benign clustered microcalcifications in mammograms. The relevant microcalcification database contains 122 cases generated from 103 subjects. Quantitative and qualitative data was provided by the radiologists, and the pathology results were available. These data include age, six (6) qualitative parameters (shape, uniformity of size, uniformity of shape, uniformity of density, shape of cluster and distribution), and the overall impression by the radiologists. A trainable multilayer feedforward neural network has been designed to maximise collectively the sensitivity and the specificity of the classification using these qualitative parameters as inputs. Using the data set, a sensitivity of 86.1% and a specificity of 84.2% have been obtained