Epileptic Seizure Detection Using Energy Thresholding

Epilepsy is a brain disorder characterized by frequent and unpredictable disruptions in the brain causing the known Epileptic Seizures (ES). One method to diagnose these seizures consists in analyzing the ElectroEncephaloGraphic (EEG) signals. However, these signals are complex and quickly-changing. They require extensive clinical experience to avoid diagnostic errors and confirm the presence of seizure onset. In recent years, handcrafted and Artificial Intelligence (AI) methods have been utilized to aid clinicians reduce the efforts needed to diagnose ES. Handcrafted methods are used to extract useful features from raw EEG data that can later on be fed into AI models to perform classification. One common feature to describe ES in the time domain is the energy where it significantly differs during ictal state. Hence, in this paper, a handcrafted method based on time-domain energy to detect ES in EEG recordings is investigated. The method was tested on the publicly available CHB-MIT dataset. The method was able to classify ES and non-seizure events by a sensitivity of 74.41%.