Despite being one of the most common and incapacitating musculoskeletal disorders in the world, low back pain (LBP) is difficult to diagnose and treat since 85–95% of cases are non-specific in nature. This study uses sagittal plane gait kinematics taken from 2D video recordings to offer a novel machine learning framework for objectively estimating pain severity and pain-related disability in LBP patients. Traditional gait analysis approaches, though effective, are typically resource-intensive and inaccessible in low-resource situations. The proposed method makes use of spatiotemporal and kinematic gait characteristics, such as stride length, cadence, joint angles, asymmetry indices, and step variability that are derived from recorded gait sequences using pose estimation techniques in order to get around these restrictions. A dataset of 100 LBP patients was used to train a Gradient Boosting regression model (XGBoost), which included self-reported pain and disability scores in addition to gait-derived features. The model demonstrated strong predictive performance, with impairment scores predicted with a comparably low error on a 100-point scale and pain intensity estimations on a 10-point scale falling within ±1.0 point of the self-reported values. In line with the established biomechanical consequences of LBP, feature importance analysis revealed that the most informative predictors were cadence, stride length, knee angle, and hip range of motion. The outcomes confirm that AI-enabled gait analysis has the potential to be a scalable and affordable substitute for traditional pain assessment instruments.
Santra et al. (Thu,) studied this question.