Digital Response Test in Epilepsy assesses interictal epileptiform discharge effects in real time

Abstract

Objective: Interictal epileptiform discharges (IEDs) in people with epilepsy (PWE) can impair cognitive functions and increase reaction time (RT) and the likelihood of missed reactions. These effects are not routinely assessed, because reliable methods for detecting IEDs of variable appearance in real time and suitable tests to measure IED effects do not yet exist. The objective was to assess different IED effects using new artificial intelligence and medical electronics.

Methods: The Digital Response Test in Epilepsy (DigRTEpi) consisted of a laptop and electronic circuits in a closed loop. Our model with Markov Transition Fields and a deep neural network (ResNet34) visualized the electroencephalogram (EEG) and classified the resulting images. IED detection triggered stimuli in a driving game or in a new cognitive assessment, the interictal Automated Responsiveness Test (iART). DigRTEpi was validated in a prospective case series with 20 people with focal and generalized epilepsies. During offline analysis, sensitivity, specificity, false-positive IED detection rate, latency of EEG classification, IED-induced RT prolongation, virtual crashes, and impaired responses to neuropsychological tasks were determined.

Results: The model detected IEDs with 84% sensitivity and 96% specificity in our training dataset. In the prospective study with 20 PWE, median sensitivity was 90% (95% confidence interval [CI] = .81-.95), and false-positive IED detection rate was 2.8 (95% CI 2.1-5.9). The ongoing EEG was classified window-by-window in a median 98.7 ms (95% CI = 98.0-99.4). Median RT prolongation and crash probability due to IEDs were 43.8 ms (95% CI = 20.3-64.7) and .9% (95% CI = 0-6.0) per person, respectively. Two patients (10%) had delays of >100 ms, found to be clinically relevant in our prior publication. IEDs caused four patients (20%) each to respond incorrectly or miss answers to neuropsychological tasks. The median false-positive IED detection rates were 2.8/min (95% CI = 2.1-5.9; driving game) and 2.1/min (95% CI = 1.5-3.2; iART).

Significance: By effectively detecting IEDs of variable morphology in real time, DigRTEpi assessed the severity of IED-associated transitory impairment of virtual driving and cognition to improve personalized care.

Keywords: EEG; driving safety; interictal epileptiform discharge effect; machine learning; transitory cognitive impairment.