Early Seizure Detection Using Neuronal Potential Similarity: A Generalized Low-Complexity and Robust Measure

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Refractory partial epilepsy poses challenges for seizure management.
• Existing automated seizure detection methods often lack sufficient accuracy or speed.
• Early detection is crucial for effective intervention and potential seizure suppression.

Purpose of the Study:

• To introduce a novel automated method for early seizure detection using neuronal potential similarity (NPS).
• To evaluate the efficacy of NPS in identifying seizure initiation trends from intracranial EEG (iEEG) signals.
• To assess the suitability of the NPS method for closed-loop seizure suppression systems.

Main Methods:

• Utilized spectral analysis of space-differential iEEG signals to compute NPS.
• Investigated the ratio of NPS values across specific frequency bands as a seizure indicator.
• Developed a threshold-based classifier for automated alarm generation.
• Validated the method on a large clinical dataset of 183 seizure onsets from 11 patients (1785 hours of iEEG).

Main Results:

• Achieved a high sensitivity of 86.9% (159/183 seizure onsets).
• Demonstrated a low false detection rate of 1.4 per day.
• Reported a mean detection latency of 13.1 seconds from electrographic onset, preceding clinical onset by 6.3 seconds.
• The method exhibited very low computational cost.

Conclusions:

• The NPS-based approach provides a robust and accurate method for automated early seizure detection in refractory partial epilepsy.
• The short detection latency and high performance make it suitable for real-time applications.
• Its low computational demands are advantageous for integration into implantable seizure suppression devices.