Enhancing Automated Seizure Detection via Self-Calibrating Spatial-Temporal EEG Features with SC-LSTM.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy diagnosis relies heavily on electroencephalography (EEG), but manual interpretation is challenging due to data complexity.
• Traditional machine learning models struggle with the high-dimensional, temporal nature of EEG data, limiting seizure detection accuracy.
• Automated seizure detection needs robust methods to handle patient-specific variability and signal artifacts.

Purpose of the Study:

• To introduce SC-LSTM, a novel hybrid deep learning architecture for enhanced automated seizure detection.
• To integrate dynamic spatial and temporal feature extraction for improved EEG signal analysis.
• To improve the accuracy, stability, and adaptability of seizure detection systems.

Main Methods:

• Developed SC-LSTM, a hybrid deep learning model combining a Self-Calibrated Reconstruction Module (SCConvNet) for spatial features and a Bidirectional Long Short-Term Memory (Bi-LSTM) network for temporal features.
• Evaluated SC-LSTM on two real-world neonatal EEG datasets using K-fold cross-validation and simulated single-channel signal loss.
• Compared SC-LSTM performance against Convolutional Neural Network (CNN) and CNN-LSTM models.

Main Results:

• SC-LSTM achieved 97% accuracy and an Area Under the Curve (AUC) of 0.99 in seizure detection.
• The model significantly outperformed existing CNN and CNN-LSTM approaches.
• SC-LSTM demonstrated high diagnostic performance resilience even with partial data loss, indicating robustness to clinical variability and artifacts.

Conclusions:

• SC-LSTM offers a significant advancement in automated seizure detection, improving accuracy and stability.
• The model's ability to handle complex EEG data and variability supports individualized diagnostics and precision medicine.
• Open-source availability of SC-LSTM promotes reproducibility and future applications in neurological disorder monitoring.