Machine learning analysis of kinematic movement features during functional tasks to discriminate chronic neck pain patients from asymptomatic controls.

This study evaluated the discriminative potential of a machine learning model using movement features during functional tasks to distinguish between patients with non-traumatic chronic neck pain and asymptomatic controls. The study included patients with chronic mechanical neck pain and asymptomatic controls. Inertial sensors analyzed kinematics during two tasks: elevated weight transfer task and water drinking. Movement was characterized using fifteen features, incorporated into machine learning models to assess how movement patterns relate to patient condition. Features included range of motion, peak velocity, smoothness, spatiotemporal inter-plane coordination, energy distribution by frequencies, and movement heterogeneity. Fifty-three patients with neck pain (36.27 ± 14.3 years; 14 men and 39 women) and 53 asymptomatic participants (35.43 ± 14.65 years; 32 men and 21 women) completed the study. Permutation tests evaluated the discriminative potential of neck movement features between groups. The elevated weight transfer task showed significant discriminative power (P = .0337 ± .0239; Accuracy = 0.618 ± 0.02), while the water drinking task did not (P = .215 ± .202). Movement heterogeneity was the most important discriminative feature, with chronic neck pain patients showing higher movement intensity fluctuations over time. Although the elevated weight transfer task showed statistically significant discriminative potential between asymptomatic individuals and those with chronic neck pain, the models correctly classified participants only 61.8% of the time. This result questions the potential of kinematic analysis to identify patients with chronic neck pain. Future research should investigate these models during more challenging tasks in samples of patients with higher neck pain intensity or disability levels.