Automatic classification of seizure adequacy in electroconvulsive therapy: A proof-of-concept study

Read the full article See related articles

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Electroconvulsive therapy (ECT) is the most effective treatment of a range of severe mental dis-orders and is annually given to more than a million patients worldwide. Dosing of the electrical charge is critical for efficacy and side-effect burden; however, it is currently based on manual assessment of seizure quality and exhibits substantial clinical heterogeneity. Here, we collected 1,000 ECT session electroencephalograms (EEG) from 130 individual ECT patients and made an expert-level classification of the EEGs into actionable seizure quality categories (null-response [0], threshold [T], and adequate [A]). We then trained a range of machine learning algorithms on the raw EEG data to classify the EEGs into these categories and tested for classi-fication accuracy. The highest performing algorithm was able to classify the seizures with an area under the receiver operating characteristic curves of >0.9 for both the 0 vs. T and the T vs. A categories corresponding to excellent discrimination. In contrast, the accuracy of predicting seizure quality in subsequent treatments was only moderate. The addition of clinical variables to the models, such as demographics, anesthesia and anticonvulsant exposure, did not improve model performance. These results provide proof-of-concept of machine learning classification of seizure adequacy in ECT based on raw EEG data. The findings could pave the way for the de-velopment of more objective and personalized decision-support tools in ECT.

Article activity feed