Focal cortical dysplasia classification for refractory epilepsy detection using artificial neural network

Refractory epilepsy is a condition characterised by epileptic seizure occurrence, which cannot be controlled with antiepileptic drugs. Focal Cortical Dysplasia (FCD) was detected as one of the main causes of refractory epilepsy. Surgical intervention is necessary to minimise or eliminate seizure occurrences, although it is only indicated in cases where there is complete certainty of the FCD region. Therefore, this paper addresses the development of a classification method to detect FCD on MRI based on morphological and textural features from a voxel-level perspective. The experiments were based on a voxel classification assessment as well as on a patient-based assessment. Multiple classifiers were tested using both voxel-based and patient-based approaches, and the best results were achieved using Multi-Layer Perceptron, AdaBoost and Support Vector Machine classifiers, which yielded close accuracy values. For voxel-based assessment, the best accuracy was 96.81% by Multi-Layer Perceptron and for patient-based assessment, the best accuracy was 90.64% by Support Vector Machine.