Classification accuracy of pain intensity induced by leg blood flow restriction during walking using machine learning based on electroencephalography

Pain assessment in clinical practice largely relies on patient-reported subjectivity. Although previous studies using fMRI and EEG have attempted objective pain evaluation, their focus has been limited to resting conditions. This study aimed to classify pain levels during movement using a wearable device with three forehead electrodes and advanced machine learning. Twenty-five healthy participants performed walking tasks under tourniquet-induced pain. It was confirmed that pain increased as walking time extended. Walking time was used as an index of pain stimulus intensity, and EEG data were collected to classify pain levels. Three machine learning algorithms-Random Forest, eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine-were employed. XGBoost achieved the highest classification performance among them. Classification accuracy for 2-, 3-, and 5-class classifications was evaluated and compared with and without BrainRate (BR), a metric indicating changes in the frequency spectrum and reflecting relative shifts across all frequency bands. Without BR, accuracies were 0.82 for 2-class, 0.60 for 3-class, and 0.40 for 5-class classification. Including BR improved accuracies to 0.96, 0.75, and 0.47, respectively. These findings highlight the significant role of BR in improving pain classification accuracy and the potential of this system for objective pain assessment even during movement.