A Quantitative Assessment of Time, Frequency, and Time-frequency Algorithms for Automated Seizure Detection and Monitoring

Epilepsy is a chronic brain disorder impacting more than 65 million people worldwide (1% of the population). Its primary symptoms, seizures, can occur without warning and can be deadly. Each year, over 100,000 patients die from Sudden Unexpected Death in Epilepsy (SUDEP). A reliable seizure warning system can help patients stay safe. This work presents a comprehensive, comparative analysis of three different signal processing algorithms for automated seizure/ictal detection. The methods perform feature extraction and seizure detection on scalp electroencephalogram (EEG) signals. The first optimized mathematical model, Approximate Entropy, performed statistical time domain analysis using a new sliding window protocol. The second algorithm performed seizure-specific spectral energy binning using the Fast Fourier Transform in the frequency domain. The third method applied signal decomposition to extract ictal features by implementing a time-frequency Discrete Wavelet Transform method. Each epileptic seizure detection algorithm was successfully validated using >75 hours of recordings from the Boston Children’s Hospital’s CHB-MIT scalp EEG clinical database. Results indicated that the Discrete Wavelet Transform algorithm performed the best, achieving a seizure detection sensitivity of 92% and a specificity of 98%. The experimental results show that the proposed methods can be effective for accurate automated seizure detection and monitoring in clinical care.