An Ensemble of Hyperdimensional Classifiers: Hardware-Friendly Short-Latency Seizure Detection With Automatic iEEG...

Area of Science:

• Neurology
• Computer Science
• Biomedical Engineering

Background:

• Epileptic seizures require accurate and timely detection for patient management.
• Existing seizure detection algorithms face challenges in latency, accuracy, and computational efficiency.

Purpose of the Study:

• To develop and evaluate a novel algorithm for epileptic seizure detection using hyperdimensional computing.
• To assess the algorithm's performance in terms of latency, accuracy, and specificity compared to state-of-the-art methods.

Main Methods:

• Feature extraction (mean amplitude, line length, local binary patterns) and embedding into hyperdimensional computing prototype vectors.
• Classification of ictal and interictal brain states using ensemble classifiers at various spatial scales.
• Validation on the SWEC-ETHZ iEEG dataset using k-fold cross-validation.

Main Results:

• The algorithm achieved significantly shorter latency (8.81 s vs. 11.57 s), higher specificity (97.31% vs. 94.84%), and accuracy (96.85% vs. 95.42%) compared to existing methods.
• Latency was further reduced to 3.74 s with a minor loss in specificity.
• Effective seizure detection was maintained using only the top 10% of electrodes identified by the algorithm.

Conclusions:

• The proposed hyperdimensional computing-based algorithm demonstrates superior performance for epileptic seizure detection.
• The algorithm's efficiency, reduced latency, and ability to identify critical electrodes make it suitable for real-time applications and deployment on embedded systems.