FASEDenseNet: A Module Embedded Convolutional Neural Network for Scalp EEG based Epilepsy Prediction

Epilepsy is considered a common neurological disorder. It not only has a major effect on the patient's life but also can be life-threatening. The cause of death in epilepsy patients is less likely by epileptic seizure but most commonly caused by accidents that happened during seizure onset such as car accidents, drowning and serious head injury caused by falling off. Therefore, it is important to predict the epileptic seizure to maximally prevent the serious effects of the patients. There are many existing models for helping to predict epilepsy. However, the most current models require complex preprocess and try to use deeper networks to enhance seizure prediction ability. Therefore, this research aims to design a better patient general model for seizure prediction using scalp EEG data with a simpler but efficient structure. The proposed model, Focal Adaption Squeeze Excite Densely Connected Convolutional Network (FASEDenseNet) has achieved outstanding results with an average sensitivity of 100% in classifying the normal EEG data and the data before seizure onset from different patients using CHB-MIT and Siena datasets. It used a channel focal adaptation module to enhance the model feature extraction ability and reduce the network depth. The proposed model shows outstanding robustness in achieving all correct predictions using the resampled data generated by different fixed windows, which shows that it is not restricted to preprocess and has better flexibility.