Abstract
Segmentation of noisy and low-contrast images remains one of the most challenging and difficult tasks, especially in the context of medical imaging. In this work, we propose an extension of the Active Shape Models (ASM) which is based on a priori knowledge about the shape and the deformation modes of the studied Region(s) of Interest (ROI). The main contribution of the proposed extension resides in the integration of a statistical directional relationship within the ASM, which is learned during a training phase. In particular, in order to force the active contour to move towards points in space that satisfy the spatial relationship, we propose a fuzzy directional constraint that allows a more robust localization of ROI. In fact, the learned a priori knowledge has been modeled using fuzzy logic in order to model uncertainty and ambiguity of the spatial representation. Realized tests on scintigraphic and MRI images proved the performance of the proposed model for the detection of multiple objects of interest in noisy and low-contrast images, even when real contours are ill-defined.
Keywords: Active shape models, a priori knowledge, fuzzy logic, medical image segmentation, spatial relationships.
Graphical Abstract
23
1
1
1