Constrained multiplicative graph cuts based active contour model for magnetic resonance brain image series segmentation

Graph cuts-based active contour model (GCACM) is often used in image segmentation, which can be categorized into additive GCACM and multiplicative GCACM. However, both the additive GCACM and multiplicative GCACM are insufficient for magnetic resonance (MR) brain image series segmentation. Considering the effectiveness of the multiplicative GCACM over the additive GCACM in local segmentation, we propose a new constrained multiplicative GCACM (CM-GCACM) for MR brain image series segmentation, in which, the constraint term is built based on the signed distance function, and can make the segmentation results obtained around the initialized contour. Generally, the deformations between adjacent slices in MR brain series are small, so we only need to give the initialized contour in one selected slice for constrained segmentation, and then the selected slice segmentation result can spread to the adjacent slices, in which case, the segmentation result of the current slicer can be served as the initialized contour for adjacent slices, and the constrained segmentation can be obtained again. By that analogy, we can realize the series segmentation. Experiments on putamen and caudate nucleus segmentation in MR brain image series demonstrate the effectiveness of proposed CM-GCACM over additive GCACM and multiplicative GCACM.