Embedded Files
Vision-Based Sarcopenia Risk Assessment from Tandem Gait
Introduction
Introduction
Sarcopenia, the age-related loss of muscle mass and function, is treatable if it is caught early, but the standard diagnostic tools work against early detection. They need specialized instrumentation and trained personnel, which makes them fine in a hospital and impractical for screening a whole community.
This work asks whether an ordinary RGB camera can do a useful first pass. The clinical task chosen is tandem gait: walking heel-to-toe along a line. Tandem gait narrows the base of support to under 5 cm, which makes balance deficits far more visible than they would be in a normal walk. A person compensating for weakness has nowhere to hide it.
Framework
Framework
The pipeline is contact-free end to end. From monocular video, 2D pose is estimated and lifted to a 3D skeleton sequence with PoseFormerV2. SkateFormer, a skeleton-based classifier with factorized spatiotemporal attention, then labels each short clip as stable or unstable. A sliding window over the whole walk turns those clip-level decisions into an instability rate, the fraction of the walk spent unstable, which serves as an exploratory subject-level risk indicator.
Results
Results
The study used 124 older adults recruited from Taipei Medical University Hospital: 93 normal controls and 31 with sarcopenia. Across three subject-wise stratified data partitions, the segment-level stability classifier reached a mean accuracy of 83.7% and a mean ROC-AUC of 0.897, outperforming ST-GCN.
At the subject level, instability-based risk groups were significantly associated with sarcopenia status (odds ratio 5.14, 95% CI 1.70 to 15.62, p = 0.003). These are preliminary results on a modest cohort, but they indicate that a single camera and a short tandem-gait walk may be enough to support contact-free community prescreening.
Page updated
Google Sites
Report abuse