Innovative Spatial-Temporal Attention Network (STAN) for Skeleton-Based Timed-Up-and-Go Analysis to Stratify Lower Back Pain Severity With Monocular RGB Camera

Lower back pain (LBP) significantly affects individuals’ quality of life and represents a considerable burden on society. Precise stratification of lower back pain (LBP) severity aids in the assessment and management of LBP, facilitating tailored treatment plans and enabling healthcare providers to track intervention effectiveness over time. In our research, we aim to introduce a novel human motion analysis technique focused on stratifying the severity of lower back pain (LBP) through the Timed-Up-and-Go (TUG) test. We present a novel spatial-temporal attention network (STAN) designed to extract intrinsic spatial-temporal features from skeletal data obtained from video recordings of TUG trials, leveraging the affordability of RGB cameras. Results: Our STAN model demonstrates remarkable efficacy in accurately (96.95%±1.15%) stratifying the severity of lower back pain from extensive evaluation results, which validate the superiority of our STAN-based approach over state-of-the-art machine learning models when analyzing video recordings for LBP stratification, from statistical evaluations of multiple metrics. The proposed study holds promise in improving the LBP assessment, which can further facilitate the related care providers' monitoring of the progress of LBP to set up effective treatment and rehabilitation plans.