Applying a visual segmentation algorithm to brain structures MR images

A variation of the neural algorithm LEGION (Locally Excitatory Globally Inhibitory Network) has been developed for the automatic visual segmentation of T1-weighted 2D head magnetic resonance images. The network obtains good performances in segmenting the skull, the brain in all its ramifications as other structures within the skull, like cerebellum, Corpus Callosum and Brain Stem. These results can be used for MEG source modeling. Putting together the results on all the processed 2D images of one volume we will be able to have 3D segmentation results which can be used to generate surface and volume tessellations suitable for FEM (finite element method) forward field calculations. We have applied the algorithm to several MRI images. Despite the diversity of the imagesm the neural network shows good robustness.