Automatic Classification of Epilepsy Lesions

Epilepsy is a common and diverse set of chronic neurological disorders characterized by seizures. Epileptic seizures result from abnormal, excessive or hypersynchronous neuronal activity in the brain. Seizure types are organized firstly according to whether the source of the seizure within the brain is localized or distributed. In this work, our objective is to validate the use of MRI (Magnetic Resonance Imaging) for localizing seizure focus for improved surgical planning. We apply computer vision and machine learning techniques to tackle the problem of epilepsy lesion classification. First datasets of digitized histology images from brain cortexes of different patients are obtained by medical imaging scientists and provided to us. Some of the images are pre-labeled as normal or lesion. We evaluate a variety of image feature types that are popular in computer vision community to find those features that are appropriate for the epilepsy lesion classification. Finally we test Boosting, Support Vector Machines (SVM) and the Nearest Neighbor machine learning methods to train and classify the images into normal and lesion ones. We obtain at least 90.0% of accuracy for most of the classification experiments and the best accuracy rate we get is 93.3%. We also automatically compute neuron densities. As far as we know, our work of performing histology image classification and automatic quantification of focal cortical dysplasia in the correlation study of MRI and epilepsy histopathology is the first of its kind. Our method could potentially provide useful information for surgical planning.