Performance Analysis of Deep-Learning and Explainable AI Techniques for Detecting and Predicting Epileptic Seizures

Abstract

Epilepsy is one of the most common neurological diseases globally. Notably, people in low to middle-income nations could not get proper epilepsy treatment due to the cost and availability of medical infrastructure. The risk of sudden unpredicted death in Epilepsy is considerably high. Medical statistics reveal that people with Epilepsy die more prematurely than those without the disease. Early and accurately diagnosing diseases in the medical field is challenging due to the complex disease patterns and the need for time-sensitive medical responses to the patients. Even though numerous machine learning and advanced deep learning techniques have been employed for the seizure stages classification and prediction, understanding the causes behind the decision is difficult, termed a black box problem. Hence, doctors and patients are confronted with the black box decision-making to initiate the appropriate treatment and understand the disease patterns respectively. Owing to the scarcity of epileptic Electroencephalography (EEG) data, training the deep learning model with diversified epilepsy knowledge is still critical. Explainable Artificial intelligence has become a potential solution to provide the explanation and result interpretation of the learning models. By applying the explainable AI, there is a higher possibility of examining the features that influence the decision-making that either the patient recorded from epileptic or non-epileptic EEG signals. This paper reviews the various deep learning and Explainable AI techniques used for detecting and predicting epileptic seizures using EEG data. It provides a comparative analysis of the different techniques based on their performance.