An iterative possibilistic image segmentation system: Application to breast cancer detection

A novel approach for digital mammograms segmentation is proposed. This approach aims to segment the mammograms using an iterative fusion process of information obtained from multiple sources of knowledge (contextual, image processing algorithm, a priori knowledge, etc). Initial Fuzzy Membership Maps (IFMMs) of different thematic classes are first estimated using available information. These IFMM's are then interpreted as Possibility Distribution Maps (PDMs), which represent the possibility for each analyzed pixel to be one of the different thematic classes in the considered image, these possibility values are then iteratively updated using contextual (spatial) information. An additional class called “Rejection” is used to manage ambiguity and to delay the segmentation operation until the establishment of high level possibility degrees for these pixels. The segmentation results are given as a thematic map as well as a confidence curve evaluating the segmentation result quality.