Automatic Seizure Detection Based on Stockwell Transform and Transformer.

Area of Science:

• Neurology
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy is a chronic neurological disorder characterized by abnormal brain activity.
• Automated seizure detection from electroencephalogram (EEG) is crucial for clinical practice and research.
• Current methods require significant manual review, highlighting the need for efficient algorithms.

Purpose of the Study:

• To develop a novel automatic epileptic EEG detection method.
• To improve the accuracy and efficiency of seizure detection algorithms.
• To assess the clinical feasibility of the proposed method.

Main Methods:

• Application of Stockwell transform (S-transform) for time-frequency representation of EEG signals.
• Feature extraction by grouping time-frequency matrices into EEG rhythm blocks and compressing them into vectors.
• Classification using a Transformer network with integrated feature selection and post-processing techniques.

Main Results:

• Achieved high performance on the CHB-MIT database: 96.15% accuracy, 96.11% sensitivity, 96.38% specificity, 96.33% precision, and 0.98 AUC in segment-based tests.
• Event-based results showed 96.57% sensitivity with a low false detection rate (0.38/h) and a 20.62s delay.
• Demonstrated significant improvements over existing seizure detection methods.

Conclusions:

• The proposed S-transform and Transformer-based method offers a highly accurate and efficient approach for automatic epileptic seizure detection.
• The algorithm's performance indicates strong potential for real-world clinical application.
• This method can significantly reduce the burden on medical professionals in analyzing EEG data.