Epileptic Seizure Detection with Hybrid Time-Frequency EEG Input: A Deep Learning Approach

Area of Science:

• Computational Neuroscience
• Medical Signal Processing
• Artificial Intelligence in Medicine

Background:

• Accurate epileptic seizure detection is crucial for patient safety and management.
• Electroencephalogram (EEG) is a primary tool for monitoring brain activity and detecting seizures.
• Deep learning methods show promise for automated feature extraction from EEG, but performance varies with input data format.

Purpose of the Study:

• To develop a deep learning-based epileptic seizure detection method utilizing hybrid EEG signal input formats.
• To investigate the impact of different EEG signal transformations on seizure detection performance.
• To improve seizure detection accuracy, especially in few-shot learning scenarios.

Main Methods:

• Proposed a deep learning model employing Convolutional Neural Networks (CNNs) for feature extraction.
• Utilized hybrid input formats: original EEG, Fourier Transform (FT), Short-Time Fourier Transform (STFT), and Wavelet Transform (WT) of EEG signals.
• Implemented a feature fusion mechanism to integrate extracted features for a robust seizure detection representation.

Main Results:

• The proposed hybrid input method demonstrated effective seizure detection capabilities.
• Feature fusion of multiple EEG signal formats led to more stable and accurate feature extraction.
• Significant performance improvement was observed in few-shot learning scenarios compared to single-format inputs.

Conclusions:

• Hybrid input formats combined with CNNs and feature fusion enhance epileptic seizure detection.
• This approach offers a promising solution for improving seizure detection accuracy, particularly with limited data.
• The method provides a more stable and reliable system for clinical applications of EEG-based seizure detection.