AI-Supported Functional Pain Phenotyping in Low Back Pain Using Sensor-Based Gait and Neuromuscular Biomarkers

Low back pain (LBP) is associated with altered motor control that may be reflected in sensor-derived features. This exploratory development study analysed 3567 gait-analysis variables from 36 participants using a 15-feature pipeline and leave-one-subject-out evaluation. The primary outcome contrasted pain-free with high pain intensity after 11 participants in the intermediate mild-pain stratum were excluded by design (25 participants; 50 trials). Stacking achieved an accuracy of 0.780 (95% CI 0.620–0.920) and balanced accuracy of 0.731 (95% CI 0.573–0.893). Recurrent features involved the feet, left thigh and multifidus-specific EMG asymmetry. An exploratory two-cluster analysis showed clinical associations but weak separation (silhouette 0.172). As model configuration and comparison were developed within the available cohort, the estimates represent internal exploratory performance rather than validation of a clinical prediction model. The findings are hypothesis-generating and require independent external validation.