Detection of Temporal Lobe Epilepsy in Magnetic Resonance Imaging Using SPHARM-Based Shape Analysis of Hippocampus

There are evidences that temporal lobe epilepsy can cause some lateralized atrophy and deformation on hippocampus and other substructures of the brain. Multi scale 3D shape representation and analysis of hippocampus is useful for diagnosis temporal lobe epilepsy in magnetic resonance imaging. Spherical harmonics (SPHARM) is a powerful tool for representation and analysis of 3D closed shape surfaces. The SPHARM is known for its power in multi-scale representation of global shape differences. Therefore, it can be used in diagnosis ofepilepsy based on the deformations in hippocampus structure. Our purpose is to design, develop and test an algorithm for classification of magnetic resonance images based on hippocampal asymmetry detection using SPHARM coefficients as shape features. Our algorithm consisted of two main parts; shapefeature extraction, and image classification. In the first part, we selected a set of 14 discriminative SPHARM coefficients that detected the asymmetry of hippocampus in patients in the training set. Then we utilized a support vector machine classifier to classify the test set of the images to normal and epileptic using our selected coefficients. On a dataset of 25 images, 12 training images for feature extraction and 13 test images for classification, our algorithm leaded to an accuracy, specificity and sensitivity of 84%, 66%, and 90%, respectively. These preliminary results show that using SPHARM coefficients for hippocampal asymmetry detection could be helpful in diagnosis of thetemporal lobe epilepsy disease in magnetic resonance imaging.

Keywords:

temporal lobe epilepsy , spherical harmonics , hippocampus shape analysis , classification , SVM