Sulci detection in photos of the human cortex based on learned discriminative dictionaries

The use of discriminative dictionaries is exploited for the segmentation of sulci in digital photos of the human cortex. Manual segmentation of the geometry of sulci by an experienced physician on training data is taken into account to build pairs of such dictionaries. It is demonstrated that this approach allows a robust segmentation of these brain structures on photos of the brain as long as the training data contains sufficiently similar images. Concerning the methodology an improved minimization algorithm for the underlying variational approach is presented taking into account recent advances in orthogonal matching pursuit. Furthermore, the method is stable since it ensures an energy decay in the dictionary update.