Epileptic Seizure Detection Using Single-Channel EEG and Artificial Intelligence Techniques

Epilepsy is a chronic neurological disorder characterized by recurrent involuntary seizures which can lead to serious injuries and death. Clinicians diagnose epilepsy by examining the electroencephalograms (EEG) of patients for seizures. However, early-stage diagnosis of epilepsy is difficult due to the labor-intensive nature of the procedure, lack of trained professionals, and inherent variability in the EEG of different patients. In this paper, we propose an automatic seizure detection system with a patient-independent approach in which the system analyzes and detects seizures in single-channel EEG from multiple patients. The system classifies an EEG instance as seizure or non-seizure based on learned features from the training set. For this task, we combine a comprehensive set of features extracted from the time, frequency, and time-frequency domains to capture information about EEG signals' amplitude, frequency, and morphology. A significant amount of experimental work has been conducted to establish the best algorithms and parameters for preprocessing, feature selection, and classification. Which achieved satisfactory results where the highest accuracy and specificity exceeded 99% through a random forest classifier and the highest sensitivity of 98.9% was achieved through the RBF-SVM classifier. We also demonstrate through comparative results that feature sets selected by the minimum redundancy maximum relevance (MRMR) algorithm outperform the ones obtained through principal component analysis (PCA). Our results indicate that an accurate, rapid, and robust automatic detection of epileptic seizures is possible; which contributes significantly to the efficiency of the epilepsy diagnosis process and the care of epileptic patients.