Cardiac PET Image Segmentation by a Deformable Model with a Force Field Driven Speed Term

Automatic, unsupervised image segmentation is playing an important role in medical imaging, but remains a challenging task in positron emission tomography (PET) due to unpredictable object shapes and inconsistent image quality resulting from noise and sampling artifacts. In this paper, we present an application of the class of deformable models, the so-called level sets method, where the speed term in the evolution equation of motion has been modified to allow the boundary curves of a detected structure to move forward and backward. This was performed by introducing a gravitational-like force field that is proportional to the gradient and inversely proportional to the distance in order to control the direction and the speed of the contours evolution. The model was applied to the automatic segmentation of the PET images of a mouse myocardium as measured by the Sherbrooke LabPET scanner.