Recognition of Microcalcifications in Digital Mammograms Based on Markov Random Field and Deterministic Fractal Modeling

We studied the performance of microcalcifications (MCs) recognition in digital mammograms using the combination of 4^th order Markov random field (MRF) and deterministic fractal model (FM). Fifty 88times88 ROI samples are selected for computer experiments, among which 25 samples are attributed to class MCs and another 25 samples to class normal. We employed a three-layer backpropagation neural network (BPNN) and the leave-one-out method to test the proposed method. The receiver operating-characteristic (ROC) analysis and the area under the ROC curve (AUC) are used to analyze the performance of recognition. We first demonstrated that 4 order MRF has a superior performance with AUC=0.87 to MRF of order lower than 4. By combining the 4^th order MRF and FM modeling, the AUC is further elevated to 0.90. Our results highlight the potential benefit of MRF and FM modeling in the recognition of MCs in digital mammograms.