A Mixed Filtering Approach for Real-Time Seizure State Tracking Using Multi-Channel Electroencephalography Data

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Real-time seizure state tracking is essential for developing effective feedback neuromodulation therapies.
• Electroencephalography (EEG) offers high temporal resolution, scalp coverage, and non-invasive applicability, making it suitable for seizure tracking.

Purpose of the Study:

• To develop a real-time, continuous seizure state estimation method using EEG.
• To create a seizure probability metric for quantitative analysis and closed-loop neuromodulation.

Main Methods:

• Utilized a mixed-filter approach with multiple EEG channels for initial seizure state estimations.
• Applied a Kalman filter to combine individual estimations into a single seizure state estimation.
• Employed a modified wrapper feature selection to identify optimal continuous and binary features.
• Modeled the hidden seizure state within a state-space framework, using expectation maximization for parameter estimation.
• Implemented a square-root Kalman filter for real-time seizure state estimation.

Main Results:

• Achieved an average accuracy of 92.7%, sensitivity of 92.8%, and specificity of 93.4% on the CHB-MIT EEG database.
• Successfully generated a single seizure probability estimation reflecting the likelihood of a seizure.
• Validated the method's applicability in a real-time setting.

Conclusions:

• The proposed EEG-based seizure estimation method is effective and accurate for real-time tracking.
• This approach provides a quantitative seizure probability, enabling advancements in closed-loop neuromodulation.
• Potential applications include long-term seizure treatment efficacy analysis and personalized neuromodulation strategies.