Machine learning-based prediction of heat pain sensitivity by using resting-state EEG

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Objective pain sensitivity assessment is vital for clinical pain management and precision medicine.
• Current methods often lack quantitative, objective predictors for pain sensitivity.

Purpose of the Study:

• To develop and validate objective signatures for classifying individual heat pain sensitivity.
• To combine neurophysiological and psychometric data for predicting pain sensitivity.

Main Methods:

• Recruited healthy individuals and collected heat pain sensitivity, psychometric, and resting-state electroencephalography (EEG) data.
• Utilized power spectral density (PSD) and functional connectivity (FC) from EEG, alongside psychometric scores, as features.
• Employed machine learning models, including support vector machine (SVM), for classification and prediction.

Main Results:

• Identified resting-state PSD and FC as reliable brain signatures for classifying heat pain sensitivity.
• Achieved classification accuracy of 86.2%-93.8% using SVM models.
• Attained 80% accuracy in predicting pain sensitivity on an independent dataset using a cubic SVM model with combined PSD and FC features.

Conclusions:

• SVM models demonstrated acceptable accuracy, indicating the potential for objective pain perception evaluation in clinical settings.
• The findings support the use of multimodal neurophysiological and psychometric signatures for pain sensitivity assessment.
• Further validation with larger datasets is recommended for the predictive model.