Abstract Details

Epilepsy affects 1% of the world population and results in a major burden in seizure-related disabilities, comorbidities, mortalities, and stigmas. Furthermore, one-third of people with epilepsy have pharmaco-resistant epilepsy and often turn to surgical interventions to manage their seizures. The field of epilepsy may benefit from incorporating qEEG methods to further our understanding of epilepsy and improve therapies.

Reliable quantitative electroencephalography (quantitative EEG; qEEG) methods can simplify the rich information contained in EEGs into objective properties and potentially transform the current practice of epilepsy assessment that relies on visual inspection of EEGs by epileptologists. In this work, we present an information theoretic qEEG method for seizure detection and compare its performance to those of conventional methods.

The inverse compression ratio (ICR) is a potential qEEG method for seizure detection that leverages compression to estimate the joint entropy (total information contained in a signal) of multi-channel EEG. ICR was calculated on 10 kHz intracortical EEGs of up to two weeks across 30 participants (15 adults and 15 children, 244 total seizures).

At seizure onset, we observed a sharp peak in ICR, followed by a dip before returning to baseline activity. Using a threshold to classify seizures from non-seizure states, ICR achieved a F1 score of 0.80 and area of precision-recall curve of 0.69, outperforming conventional qEEG methods: variance, Shannon entropy, approximate entropy, and sample entropy.

ICR may be a useful tool for automated seizure detection and suggests the potential of using information theoretic approaches to evaluate seizure activity. Implementing this qEEG method to clinical practice may offload labor-intensive tasks from clinicians, uncover EEG features undetectable to the eye, and provide efficient, consistent, and accessible care.