Lisa Yamada

A compression-enabled approach to analyze seizures for people with refractory epilepsy

Advisor: Paul Nuyujukian

Refreshments at 9AM

Epilepsy affects 1% of the world population and, if left untreated, can lead to progressively impaired cognition and function, brain damage, and other neurologic deficits. For approximately a third of patients, antiepileptic drugs are not effective, and their treatment options are often limited to surgical interventions to resect seizure-inducing regions of the brain. A promising alternative is to leverage closed-loop devices that disrupt epileptiform activity in the brain upon detection; however, this requires robust, quantitative EEG (qEEG) methods to automate seizure detection. In today’s epileptology, in which multi-day, multielectrode EEGs are manually reviewed by trained clinicians, data compression may be a useful tool for analyzing seizure activity. 

In my talk, I will first present a minimally disruptive, scalable platform to acquire 10 kHz research-quality, intracranial EEG in a hospital setting. This infrastructure is used to build our EEG repository for quantitative analyses. Then, I will introduce our information theoretic measure called the inverse compression ratio (ICR)–an unbiased estimate of joint entropy–and demonstrate its potential for seizure detection. Finally, video compressors were explored, and performance was improved when leveraging time-dependence and lossy encodings. Information theoretic measures operate without linear assumptions or parametric modeling and provide an unbiased measure of signal complexity to perform qEEG analyses.