Automated Annotation of Epileptiform Burden and Its Association with Outcomes

Abstract

This study was undertaken to determine the dose-response relation between epileptiform activity burden and outcomes in acutely ill patients. Methods: A single center retrospective analysis was made of 1,967 neurologic, medical, and surgical patients who underwent >16 hours of continuous electroencephalography (EEG) between 2011 and 2017. We developed an artificial intelligence algorithm to annotate 11.02 terabytes of EEG and quantify epileptiform activity burden within 72 hours of recording. We evaluated burden (1) in the first 24 hours of recording, (2) in the 12-hours epoch with highest burden (peak burden), and (3) cumulatively through the first 72 hours of monitoring. Machine learning was applied to estimate the effect of epileptiform burden on outcome. Outcome measure was discharge modified Rankin Scale, dichotomized as good (0-4) versus poor (5-6). Results: Peak epileptiform burden was independently associated with poor outcomes (p < 0.0001). Other independent associations included age, Acute Physiology and Chronic Health Evaluation II score, seizure on presentation, and diagnosis of hypoxic-ischemic encephalopathy. Model calibration error was calculated across 3 strata based on the time interval between last EEG measurement (up to 72 hours of monitoring) and discharge: (1) <5 days between last measurement and discharge, 0.0941 (95% confidence interval [CI] = 0.0706-0.1191); 5 to 10 days between last measurement and discharge, 0.0946 (95% CI = 0.0631-0.1290); >10 days between last measurement and discharge, 0.0998 (95% CI = 0.0698-0.1335). After adjusting for covariates, increase in peak epileptiform activity burden from 0 to 100% increased the probability of poor outcome by 35%. Interpretation: Automated measurement of peak epileptiform activity burden affords a convenient, consistent, and quantifiable target for future multicenter randomized trials investigating whether suppressing epileptiform activity improves outcomes.

Diagnostic and prognostic EEG analysis of critically ill patients: A deep learning study

“…Therefore, we compared the network output with the EEG given as input. Visual examination of epochs classified with high confidence confirmed the presence of features used also by human specialists, such as a continuous rhythmic background ( Beuchat et al, 2021 , Hofmeijer et al, 2015 , Westhall et al, 2016 ), sleep spindles ( Vassallo et al, 2021 ), G-RDA ( Beuchat et al, 2021 ) in epochs classified as favorable outcome; or such as suppressed background and burst-suppression ( Hofmeijer et al, 2015 , Ruijter et al, 2019 , Westhall et al, 2016 ), or epileptiform activity ( Zafar et al, 2021 ) in epochs classified as unfavorable outcome.…”

“…Most studies using computational methods for EEG-based prognostication in the ICU are conducted on homogeneous groups, typically on patients with HIE ( Beudel et al, 2014 , Tzovara et al, 2016 , Zubler et al, 2017 , Tjepkema-Cloostermans et al, 2017 , Tjepkema-Cloostermans et al, 2019 , Jonas et al, 2019 , Alnes et al, 2021 ). Only a handful have been conducted on patients with impaired consciousness of various etiologies, using predefined visual ( Müller et al, 2020 , Zhang et al, 2011 , Vassallo et al, 2021 , Kang et al, 2015 , Selioutski et al, 2019 ) or quantitative ( You et al, 2018 , Zubler et al, 2016 ) or mixed ( Zafar et al, 2021 ) features. To the best of our knowledge, it is the first time that DL is used for this task.…”

“…DL has been successfully used for EEG analysis for brain computer interfaces or for specific neurologic or psychiatric conditions (for comprehensive reviews see ( Craik et al, 2019 , Roy et al, 2019 )). In the context of critical care, EEG has been mainly used for prognostication in patients with hypoxic-ischemic encephalopathy after cardiac arrest ( Jonas et al, 2019 , Tjepkema-Cloostermans et al, 2019 ) or with epileptiform activity ( Zafar et al, 2021 ). However, all these applications are somehow artificial in that only a specific question or a single pathology is considered at the time.…”

Why Interpretable Causal Inference is Important for High-Stakes Decision Making for Critically Ill Patients and How To Do It

“…To do this, we need to first identify segments of the EEG signal containing seizure-like EA behavior. Doing this using human annotators would be extremely time consuming, so we use a convolutional neural network (CNN) trained on human annotators' classifications of 10 second windows into non-EA and EA in a semi-supervised fashion (Ge et al, 2021b;Zafar et al, 2021;Jing et al, 2016). We use the predictions to compute EA time series (Z ω t ).…”

The Past, Present, and Future of Tele-EEG

“…The dissemination of these guidelines has a clear impact on cEEG utilization. 13 While the greatest strength of evidence remains for patients with convulsive status epilepticus, the burden of epileptiform discharges and seizures in other conditions also correlates with outcomes at hospital discharge—severe acute ischemic stroke in adults 47 and various other indications in critically ill adults 48 and children, 49 50 for instance. cEEG recordings are frequently deployed in pediatric and neonatal ICUs as well as cardiac and surgical ICUs and any other area such as the emergency department where patients may present with some type of altered consciousness or seizure-like events.…”