Classifying microcalcifications on digital mammography using morphological descriptors and artificial neural network

Microcalcifications are tiny calcium accumulations in the breast and are a warning sign of possible breast carcinoma in the early stages of its formation, so it is highly important for radiologists to identify them in a digital mammography and also to be able to discern in which category of BIRADS they belong. In this study the focus is to classify into BIRADS 2, 3 and 4, categories in which the advance of cancer can be prevented. Our method consists of selecting a region of interest (ROI), then as a preprocessing technique a High Pass Gaussian filter is applied to the image to enhance the lesions and then construct a binary mask to obtain its morphological descriptors. This data gets normalized and goes into an ANN, trained previously with a dataset acquired from the Medical Specialized Unit on Detection and Diagnosis of Breast Cancer (UNEME DEDICAM) in Querétaro, Mexico. This dataset includes its diagnosis into BIRADS. In this article it is used a two-layer feed-forward network, with sigmoid hidden and 10 hidden networks. The Neural Network is trained varying the inputs and the matrix of data is reduced using Principal Component Analysis (PCA) as working with less descriptors get us less accuracy, the use of 10 morphological descriptors is the best option. Classify BIRADS 2 and 3 into a sole category, is the best result as an 80% of the predictions are correct. In conclusion to discern between BIRADS 2 and BIRADS 3 more than only descriptors as measurement is needed.