Mammographic risk assessment and local greylevel appearance histograms

It has been shown that the probability to develop breast cancer is strongly correlated with the appearance of tissue in mammographic images. This appearance incorporates both greylevel and tissue pattern aspects and models of local texture information, which incorporate both greylevel and spatial aspects, can as such be related to mammographic risk assessment. Here we represent texture by the variation in local greylevel configuration/appearance in histogram format for which the distribution varies with texture appearance. The histogram information can be directly used to classify mammograms according to standard mammographic risk estimation models (e.g. BIRADS). Results on the MIAS database indicate a correct classification of 70% (which increases to 84% for high/low risk classification), which is comparable with existing methods. Variation of the classification results with respect to some of the model parameters are discussed, which indicate the robustness of the methodology. In addition future directions, to improve the classification results are discussed.