Subgrouping non-specific low back pain based on spinal marker trajectory data: An unsupervised machine learning approach.

BACKGROUND: Non-specific low back pain (LBP) is a heterogeneous condition. Therefore, it is important to investigate whether clinically feasible assessments can identify diverse movement patterns in individuals with LBP. PURPOSE: To identify distinct movement-based subgroups among individuals with non-specific LBP using thoraco-lumbo-pelvic marker trajectories during forward bending and to compare the resulting clusters with healthy controls. STUDY DESIGN: Cross-sectional study. METHODS: Kinematic data were collected from 127 individuals with non-specific LBP and 58 healthy controls during a forward bending task using a smartphone-based video recording system. Three markers were placed over T12, L2, and S2, and their x- and y-axis displacements were extracted using an open-source software. Unsupervised machine learning (K-means clustering) was applied to classify movement patterns within the LBP group based on six kinematic features (the horizontal and vertical displacements of the T12, L2, and S2 markers). RESULTS: Two clusters were identified within the LBP group: cluster 1 (large-excursion, 54 %) and cluster 2 (small-excursion, 46 %). Both clusters showed significant differences from healthy controls in marker displacement (p < 0.001). Cluster 2 reported a slightly higher pain intensity (p = 0.036), with no difference in disability scores. CONCLUSIONS: Unsupervised clustering revealed distinct spinal movement subgroups in individuals with non-specific LBP. These findings indicate that both excessive and limited movement may relate to pain-related adaptation, supporting the need for movement-based subgrouping to guide individualized management.