Learning the Optimal Contrast Enhancement of Mammographie Breast Masses

We propose a methodology to predict the optimal contrast enhancement for a mammogram by learning the non-linear mapping of lesion grey scale statistics with enhancement and segmentation performance. For the unseen mammogram on test, we identify a suspicious region and using a series of Artificial Neural Networks (ANN), determine the enhancement that will maximise segmentation performance. For accurately defined suspicious regions on test we obtain a rank-ordered correlation coefficient of 0.701 compared with the target enhancement.