HOKF: High Order Kalman Filter for Epilepsy Forecasting Modeling.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy forecasting relies on high-order time series from electroencephalography (EEG).
• Accurate seizure prediction can improve patient quality of life and enable new therapeutic strategies.
• Existing methods have limitations in modeling complex EEG data properties.

Purpose of the Study:

• To propose an improved Kalman Filter (KF) algorithm, termed High-Order Kalman Filter (HOKF), for epilepsy forecasting.
• To effectively model noise, temporal smoothness, and tensor structure in high-order EEG time series.
• To enhance the accuracy and scalability of seizure prediction from neural activity.

Main Methods:

• Developed the High-Order Kalman Filter (HOKF) as an extension of the standard Kalman filter.
• Incorporated modeling of noise, temporal smoothness, and tensor structure within the HOKF.
• Evaluated HOKF effectiveness and scalability using a real epilepsy EEG dataset with forecasting and scalability experiments.

Main Results:

• HOKF effectively captures evolving trends in neural activity, even with missing EEG data.
• HOKF demonstrates linear scalability with respect to the number of time-slices.
• The proposed HOKF method significantly outperforms the original Kalman Filter and other existing methods in epilepsy forecasting.

Conclusions:

• The HOKF algorithm offers an effective and scalable solution for epilepsy forecasting.
• HOKF's ability to model complex EEG properties leads to superior seizure prediction.
• This advancement holds promise for improved epilepsy management and patient care.