Automatic Seizure Detection Based on S-Transform and Deep Convolutional Neural Network.

Area of Science:

• * Neuroscience and Biomedical Engineering
• * Focus on advanced signal processing and machine learning for neurological disorder diagnosis.

Background:

• * Epilepsy diagnosis relies heavily on electroencephalogram (EEG) analysis, a labor-intensive process.
• * Automated seizure detection is crucial for efficient diagnosis and patient management.
• * Existing methods often struggle with accuracy and false detection rates in long-term recordings.

Purpose of the Study:

• * To propose and evaluate a novel automated seizure detection system.
• * To combine Stockwell transform (S-transform) with deep Convolutional Neural Networks (CNNs) for enhanced seizure onset detection.
• * To improve accuracy and reduce false detection rates in long-term intracranial EEG.

Main Methods:

• * EEG data preprocessing using wavelet decomposition.
• * Time-frequency representation generation via S-transform.
• * A 15-layer deep CNN with dropout and batch normalization for feature extraction and classification.
• * Post-processing with smoothing and collar techniques to refine detection.

Main Results:

• * High performance on a public EEG database (21 patients).
• * Segment-based sensitivity of 97.01% and specificity of 98.12%.
• * Event-based sensitivity of 95.45% with a low false detection rate (FDR) of 0.36/h.

Conclusions:

• * The proposed S-transform and deep CNN approach offers a robust and accurate method for automatic seizure detection.
• * This technique can significantly aid in epilepsy diagnosis and reduce the burden on clinicians.
• * The system demonstrates high efficacy in identifying seizure onsets in long-term intracranial EEG recordings.