Pectoral muscle segmentation on mammographic images based on radial lengths
Sevastianos Chatzistergos, Ioannis Andreadis, Konstantina S. Nikita · 2016
Mammography is the main imaging technique for breast cancer diagnosis and prevention. Many image processing techniques though require the breast region to be adequately defined in order to provide reliable results. Pectoral muscle segmentation is one of the most challenging tasks in this domain since the limits between the muscle and the actual breast region are sometimes quite difficult to distinguish. In the current work, a method to perform pectoral muscle segmentation on mammographic images based on the notion of Radial Lengths (RL) is presented. The mean value of RLs is used to reveal edge regions in mammograms. The edges produced would be discontinued and resemble mostly to edge segments. Points at those segments are then randomly selected and candidate edge lines start propagating having those points as a start. A number of criteria are set to define how well the line follows the actual muscle edge. A single line representing the pectoral muscle edge is finally selected based on a number of fitting criteria. The proposed method is compared to state of the art methods in the field and found to clearly outperform them.