Shape analysis of hippocampus in temporal lobe epilepsy using Signed Poisson Mapping

Complementary role of computer assisted models using machine learning methods in medical imaging has been a center of attention in recent years. Shape analysis of the brain structures can be used to evaluate their abnormalities and deformations, specifically in patients suffering from neurological diseases like epilepsy, Alzheimer, and Parkinson. We propose an automatic diagnosis and lateralization algorithm using Signed Poisson Mapping (SPoM), which has been recently proposed as a new framework for shape analysis of three-dimensional (3D) structures. In contrast to previous studies, we use a three-class classification to show the robustness of our algorithm in differentiating between normal, left temporal lobe epilepsy (LTLE), and right temporal lobe epilepsy (RTLE) subjects. We also use a support vector machine (SVM) classifier with a radial basic function (RBF) kernel for lateralization, i.e., differentiating between RTLE and LTLE patients. The classification accuracy for the three-class classifier is 94% and for the lateralization task is 95% which is superior to those reported in the related literature.