Can Machines Identify Pain Effects? A Machine Learning Proof of Concept to Identify EMG Pain Signature

This study introduces a machine-learning-based approach for identifying "pain signatures" using electromyography data from volunteers undergoing acute pain. Leveraging the XGBoost algorithm, our method analyzes electromyography features (variance, mean absolute deviation, integral, peak, and entropy) to classify muscle contractions as painful or non-painful. Fifteen participants performed controlled elbow flexion tasks under three conditions: during painful and painless conditions. The results revealed that electromyographic peak and integral activity were key predictors of pain states, with the model achieving 73% sensitivity in distinguishing painful from painless conditions. Interestingly, placebo-induced responses with less intense pain exhibited muscular adaptations similar to, but less extensive than, those observed under actual pain. These findings underscore the potential of machine learning to enhance pain assessment by providing a non-verbal, objective method for analyzing neuromuscular adaptations, paving the way for personalized pain management and more accurate monitoring of musculoskeletal health.