External Validation of EEG-Based Machine Learning Models for Continuous Pain Prediction

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Machine learning (ML) shows promise for predicting pain intensity from electroencephalography (EEG) data.
• External validation of ML models for continuous pain prediction from EEG is limited.

Purpose of the Study:

• To externally validate ML regression models for predicting continuous pain intensity from single-trial EEG features.
• To assess the clinical utility of ML models in pain assessment using multi-stage validation.

Main Methods:

• Developed and validated Random Forest (RF) and Long Short-Term Memory (LSTM) models using EEG data from 91 participants across three samples.
• Utilized single-trial EEG time-frequency features and graded pneumatic pressure stimuli.
• Performed internal, cross-subject external, and within-subject temporal external validation.

Main Results:

• Both RF and LSTM models outperformed random prediction but not a baseline dummy model in continuous pain prediction.
• RF achieved mean absolute errors (MAE) between 18.90 and 21.29 across validation stages.
• RF classifier achieved up to 64% internal and 58% external validation accuracy for low/high pain classification, remaining below clinical thresholds.

Conclusions:

• Predicting continuous pain intensity from EEG oscillations using current ML methods is highly challenging.
• ML classification models show above-chance performance but lack clinical significance for pain assessment.
• Methodological advancements, including composite measures, are needed for ML to be clinically useful in pain assessment.