EEG Temporal-Spatial Feature Learning for Automated Selection of Stimulus Parameters in Electroconvulsive Therapy

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Technology

Background:

• Electroconvulsive Therapy (ECT) requires precise stimulus to induce generalized seizures (GS), balancing efficacy and adverse effects like cognitive impairment.
• Current automation for ECT stimulus parameter selection is insufficient for clinical needs.
• Predicting GS induction probability is crucial for optimizing ECT treatment.

Purpose of the Study:

• To develop and validate ECTnet, a two-stage learning model for predicting GS induction probability.
• To automate the selection of optimal ECT stimulus parameters.
• To minimize stimulus charge while ensuring successful GS induction.

Main Methods:

• A two-stage learning model, ECTnet, was developed.
• Stage 1: Temporal-Spatial Feature Learning using channel-wise convolution and ConvLSTM on EEG data.
• Stage 2: Generalized Seizure (GS) Prediction by fusing EEG features with stimulus parameters.

Main Results:

• ECTnet achieved an AUC of 0.746, F1-score of 0.90, 89% precision, and 93% recall in predicting seizure induction.
• The model outperformed existing state-of-the-art methods.
• Incorporating stimulus parameters improved the F1-score by 0.054.

Conclusions:

• ECTnet effectively automates the selection of optimal stimulus parameters for ECT.
• The model demonstrates significant potential for improving ECT safety and efficacy.
• This AI-driven approach offers a promising solution for clinical ECT practice.