Compression-Enabled Joint Entropy Estimation for Seizure Detection on Human Intracortical Electroencephalography

Area of Science:

• Neuroscience
• Biomedical Signal Processing
• Epilepsy Research

Background:

• Epilepsy affects 1% of the global population, with one-third experiencing drug-resistant forms requiring surgical intervention.
• Current epilepsy diagnosis relies on manual review of electroencephalography (EEG) data, which can be time-consuming and subject to inter-observer variability.
• Quantitative EEG (qEEG) offers potential for objective, efficient, and standardized analysis of neurophysiological data.

Purpose of the Study:

• To introduce and evaluate the inverse compression ratio (ICR) as a novel qEEG method for automated seizure detection.
• To assess the performance of ICR in analyzing intracortical neurophysiological data from a large cohort of epilepsy patients.
• To explore the potential of ICR to enhance the speed, accuracy, and equity of epilepsy care.

Main Methods:

• The inverse compression ratio (ICR), an estimation of joint entropy using compression algorithms, was developed as a qEEG seizure detection tool.
• ICR was tested on a dataset of 10 kHz intracortical neurophysiological recordings from 30 participants, encompassing over 240 seizures.
• The study utilized a substantial volume of continuous, multi-day intracortical neuroelectrophysiology data (2,900+ hours).

Main Results:

• Single-electrode ICR achieved a F1 score of 0.80 and an area under the precision-recall curve of 0.69, surpassing conventional qEEG methods.
• Multielectrode ICR analysis performed within the top 2% of individual electrode performance, suggesting potential to bypass the need for electrode selection.
• This study represents the first demonstration of compression-based multidimensional estimation in a clinical application.

Conclusions:

• The inverse compression ratio (ICR) shows significant potential for automated seizure detection in epilepsy.
• Integrating ICR into clinical workflows could lead to substantial improvements in epilepsy diagnosis and patient care.
• ICR provides a model-free solution for high-dimensional joint entropy estimation, with potential applications beyond epilepsy in biomedical signal processing.