Dynamic System Modeling for Reconstruction of Intracranial EEG Signals in Epilepsy Using Weighted Least Squares.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy affects 1% of the global population, causing seizures with potential for severe motor, cognitive, and life-threatening impairments.
• Accurate brain mapping and identification of epileptogenic zones are crucial for effective treatment.
• Dynamic linear time-invariant state-space models are valuable for reconstructing intracranial EEG (iEEG) activity and mapping brain regions.

Purpose of the Study:

• To investigate Weighted Least Squares (WLS) as a robust alternative to Ordinary Least Squares (OLS) for modeling iEEG signals.
• To evaluate the efficacy of WLS in mitigating outliers and improving long-term signal prediction compared to OLS.
• To assess the potential of WLS for enhancing clinical applications in epilepsy management.

Main Methods:

• Employed dynamic linear time-invariant state-space models for iEEG signal reconstruction.
• Compared WLS and OLS estimation methods using five patient iEEG datasets (2,300 windows) sampled at 1000 Hz.
• Utilized synthetic signals with introduced outliers to assess model susceptibility and employed root mean square error (RMSE) for performance evaluation.

Main Results:

• WLS demonstrated significantly lower RMSE (0.698 mV) compared to OLS (0.788 mV) in human iEEG recordings (p < 0.0001).
• WLS maintained low RMSE on synthetic data even with increasing outlier magnitudes, while OLS errors escalated significantly (p < 0.0001).
• WLS preserved short- and long-term signal structure more effectively than OLS, indicating superior robustness.

Conclusions:

• Weighted Least Squares (WLS) provides a robust framework for iEEG signal reconstruction, effectively mitigating outliers and stabilizing long-term estimates.
• This WLS approach can enhance clinical applications, including more accurate seizure localization and the design of neuromodulation therapies.
• The proposed data-driven dynamical models can serve as a predictive aid, augmenting clinical workflows for improved epilepsy diagnosis and treatment decisions.