Epileptic Seizure Detection Using Hyperdimensional Computing and Binary Naive Bayes Classifier.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Epileptic seizure (ES) detection is crucial for epilepsy management.
• Intracranial EEG (iEEG) offers high-quality data but current detection methods are often data-intensive, computationally complex, or perform poorly with limited data.
• Developing efficient and generalizable ES detection methods is a significant clinical need.

Purpose of the Study:

• To propose a lightweight, data-efficient, and high-performance approach for ES detection using hyperdimensional computing (HDC).
• To enable accurate ES detection in low-data settings and facilitate hardware implementation.
• To reduce computational complexity and latency in ES detection.

Main Methods:

• Utilizing local binary patterns (LBPs) to extract temporal-spatial dynamics from iEEG signals.
• Employing hyperdimensional computing (HDC) for robust, high-dimensional data representation.
• Implementing a binary Naive Bayes classifier for ictal and inter-ictal state discrimination.

Main Results:

• Achieved 100% sensitivity and specificity in one-shot learning for most patients on the SWEC-ETHZ iEEG dataset.
• Maintained high performance in few-shot learning with an average sensitivity of 98.88% and specificity of 93.09%.
• Demonstrated an average latency of 4.31 seconds, significantly outperforming state-of-the-art methods.

Conclusions:

• The proposed HDC-based method offers an efficient, low-resource, and high-performance solution for epileptic seizure detection.
• The approach shows significant potential for real-time clinical applications, especially in data-scarce scenarios.
• The lightweight and hardware-friendly design facilitates practical implementation in epilepsy management tools.