023-22 – External validation of machine learning and EEG for pain intensity classification in healthy individuals

Machine learning (ML) and electroencephalography (EEG) to classify pain intensity has significant potential for clinical applications. However, previous research has failed to assess model performance on novel data (e.g., external validation), hindering the interpretation of the method’s potential clinical utility. This study aimed to be the first to externally validate ML for pain intensity classification (low versus high pain) using EEG features. We conducted two independent experiments and used study one (n = 25) for cross-validation and study two (n = 15) for external validation, respectively. In both paradigms, pain sensations were experimentally induced using a pneumatic pressure stimulator (delivered to the fingernail bed) whilst EEG was recorded simultaneously. We calculated ML features from frontal, central and parietal regions using single-trial EEG epochs, which were used to train several established ML algorithms. The results showed that all algorithms performed than chance on both the cross-validation and external validation assessments. Moreover, the random forest (RF) model demonstrated the best performance, achieving accuracies of approximately 73 and 68% for cross-validation and external validation, respectively. Overall, this study is the first to externally validate ML and EEG for the classification of pain intensity. These models successfully generalised to both new populations and experimental paradigms, demonstrating the robustness of the results. Therefore, this study addresses one of the most significant limitations within the field and provides the best estimates of the clinical potential of ML and EEG for pain classification.