The use of Multi-domain Electroencephalogram Representations in the building of Models based on Convolutional and Recurrent Neural Networks for Epilepsy Detection

TL;DR

This work addresses automated epilepsy detection from EEG by systematically comparing time-, frequency-, and time-frequency-domain representations using DL architectures (CNN, RNN, CRNN) on the CHB-MIT dataset. It demonstrates that frequency-domain features, especially PSD from Multitaper and Welch methods, yield the strongest performance with AUC values exceeding 0.99, and that longer windows enhance discriminative power in these representations. The study provides statistically validated guidance on representation choice and window sizing, highlights CNNs as a generally efficient and effective architecture, and discusses limitations including dataset scope and edge-case misclassifications. The findings offer practical implications for building robust, high-precision seizure-detection systems in clinical settings, with potential extensions to other neurological conditions and advanced architectures.

Abstract

Epilepsy, affecting approximately 50 million people globally, is characterized by abnormal brain activity and remains challenging to treat. The diagnosis of epilepsy relies heavily on electroencephalogram (EEG) data, where specialists manually analyze epileptiform patterns across pre-ictal, ictal, post-ictal, and interictal periods. However, the manual analysis of EEG signals is prone to variability between experts, emphasizing the need for automated solutions. Although previous studies have explored preprocessing techniques and machine learning approaches for seizure detection, there is a gap in understanding how the representation of EEG data (time, frequency, or time-frequency domains) impacts the predictive performance of deep learning models. This work addresses this gap by systematically comparing deep neural networks trained on EEG data in these three domains. Through the use of statistical tests, we identify the optimal data representation and model architecture for epileptic seizure detection. The results demonstrate that frequency-domain data achieves detection metrics exceeding 97\%, providing a robust foundation for more accurate and reliable seizure detection systems.

Figures (14)

• Figure 1: Example of electrode arrangement on the scalp and their nomenclature freeman2012imaging.
• Figure 2: Time signal
• Figure 3: PSD Welch signal
• Figure 4: PSD Multitaper signal
• Figure 5: Spectrogram signal