Recognition Of Microcalcifications in Digital Mammograms Using High Order Markov Random Field Model

This paper studies the performance of MCs (microcalcifications) recognition in digital mammograms by using a modified autobinomial Markov random field (MRF) model. Fifty 85x85 sample images are selected from the MIAS (Mammographic Image Analysis Society) database for this study. Among these images, 25 samples arc! of MCs and the other 25 samples are of normal. We model these images by the modified autobinomial model of fourth order., and extract the 12 model parameters as the feature vector of the images by maximum bivariate pseudolikelihood parameter estimation method. The SVM (Support Vector Machine) and BPNN (backpropagation neural network) classifiers are both employed to test the performance of the proposed method. By applying the "leave one out" test approach, an impressive average recognition rate of about 82%, out of the 50 sample data, is achieved. This result demonstrates the potential power of MRF model in the recognition of MCs in digital mammograms.