A low computation cost method for seizure prediction.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Science

Background:

• Epileptic seizure prediction is crucial for patient care.
• Dynamic changes in electroencephalograph (EEG) signals precede seizures.
• Accurate seizure prediction requires efficient algorithms with low computational complexity.

Purpose of the Study:

• To develop and evaluate a low-computation algorithm for real-time epileptic seizure prediction.
• To utilize Higuchi fractal dimension (HFD) of EEG signals as a feature for seizure prediction.
• To assess the efficacy of Bayesian Linear Discriminant Analysis (BLDA) and Kalman filtering in enhancing prediction accuracy and reducing false alarms.

Main Methods:

• Extraction of Higuchi fractal dimension (HFD) from EEG signals.
• Classification of preictal and interictal states using Bayesian Linear Discriminant Analysis (BLDA).
• Application of a Kalman filter for smoothing classifier outputs and a thresholding procedure for final prediction.

Main Results:

• The algorithm achieved an average sensitivity of 86.95% (30 min) and 89.33% (50 min) prediction time.
• A low average false prediction rate of 0.20/h was recorded.
• The prediction times were 24.47 min (30 min) and 39.39 min (50 min), demonstrating effective preictal state identification.

Conclusions:

• Changes in HFD are reliable precursors to ictal activity, distinguishing preictal from interictal epochs.
• The combination of HFD and BLDA offers low computational complexity, suitable for real-time applications.
• The proposed algorithm demonstrates significant potential for practical, real-time epileptic seizure prediction systems.