Data-Driven Model for Improving MEG Epileptic Spike Detection

Epilepsy is a neurological disorder characterized by spontaneous recurrent seizures, affecting over 50 million people worldwide. The prompt and accurate detection of epileptic events is crucial for effective treatment and management. Traditional methods such as Electroencephalogram (EEG) have been complemented by Magnetoencephalography (MEG), which offers superior spatial resolution due to its insensitivity to the distortive effects of the skull and scalp. This study advances the analysis of MEG data using Complex Continuous Morlet Wavelet Transform (CCMWT) for feature extraction, coupled with innovative machine learning architectures for classifying epileptic versus non-epileptic signal segments. We developed and compared a 3D Convolutional Neural Network (3D-CNN) and a VGG16 model with the Transfer learning strategy, in terms of accuracy, computational efficiency, and error rates. Our results demonstrate the VGG16 model with the Transfer learning strategy's superior performance in all cases except the computation times.