Refinement of left ventricle segmentation in MRI based on simulated annealing

In this work we present a methodology to refine the segmentation of 2D and 3D images with applications to MR images. Basically it consists of inclusion and exclusion of border voxels according to an energy function that encompasses contrast, texture and shape information. The decision process follows simulated annealing approach, and with proper energy function, precise segmentation can be achieved. Since there is an initial segmentation, some statistical properties and normalizing coefficients can be derived. This approach can be easily extended to other modalities.