On the Detection of Possible Epileptic Seizure by Means of Explainable and Robust Deep Learning

Epilepsy is a neurological disorder characterized by recurrent seizures, necessitating accurate and timely diagnosis. In this paper we propose a deep learning-based approach for the detection of possible epileptic seizures from electroencephalography signals, by exploiting Convolutional Neural Networks to classify electroencephalography data into two categories: normal brain activity and possible epileptic seizures. We also take into account a set explainability algorithms through Class Activation Mapping methods to highlight the most relevant electroencephalography regions contributing to model predictions. Experimental analysis on a dataset of 45,400 electroencephalography-based images shows an accuracy of 87.1%. Moreover, the integration of explainability techniques further enhances the trustworthiness of the model, making it more suitable for clinical applications.