Automatic segmentation of brain structures from MRI integrating atlas-based labeling and level set method
Matineh Shaker, Hamid Soltanian‐Zadeh · Conference proceedings - Canadian Conference on Electrical and Computer Engineering · 2008
The purpose of this study is to segment hippocampus, amygdala and entorhinal cortex in magnetic resonance images (MRI) of temporal lobe epilepsy (TLE) patients. The proposed method consists of two separate parts. First, we use an atlas-based segmentation method to obtain initial segmentation results for desired structures. Using additional preprocessing steps for image registration and gray matter (GM) segmentation is the specification of this stage of the work. Then, all of the GM voxels are labeled using an anatomical atlas. In the next stage, variational level set formulation without re-initialization is applied on the images. We use the boundaries obtained by atlas-based segmentation as the contour for initialization of level set function. Automatic generation of initial contour makes the final segmentation results operator-independent. The proposed approaches are evaluated by comparing automatic and expertpsilas segmentation results and confirming their similarity.