An adaptive frequency partitioning framework for epileptic seizure detection using TransseizNet

Area of Science:

• Medical technology
• Artificial intelligence in healthcare
• Signal processing

Background:

• Epilepsy is characterized by recurrent seizures due to abnormal brain activity detected via electroencephalography (EEG).
• Conventional epilepsy seizure detection methods often suffer from low accuracy and high computational demands.
• There is a need for advanced, efficient, and accurate seizure detection systems.

Purpose of the Study:

• To propose a novel TransseizNet framework for enhanced epilepsy seizure detection from EEG signals.
• To improve the accuracy and computational efficiency of seizure detection compared to existing approaches.
• To develop an interpretable and practical tool for healthcare.

Main Methods:

• EEG data pre-processing using the Savitzky-Golay filter.
• Signal decomposition via Empirical Tunable Q-Wavelet Transform for improved time-frequency resolution.
• Epilepsy seizure detection and classification using a Wavelet-Graph Convolutional Network Vision Transformer.

Main Results:

• TransseizNet achieved an average accuracy of 98.65%, precision of 98.59%, F1-score of 98.45%, and recall of 98.30% across three datasets.
• The framework demonstrated a low computational time of 17 seconds and a detection latency of 2.5 seconds.
• Performance metrics significantly outperformed baseline approaches in epileptic seizure detection.

Conclusions:

• The TransseizNet framework offers superior performance for epilepsy seizure detection by integrating adaptive frequency decomposition and hybrid deep learning.
• Its high accuracy, minimal detection latency, and interpretability make it a promising solution for clinical applications.
• This novel approach addresses the limitations of conventional methods, paving the way for improved epilepsy management.