Continuous video-based postural tracking for Parkinson's disease: a pose estimation approach with landmark calibration.
Yuki Kondo, Ippei Suzuki
An automated video tool for measuring postural angles in Parkinson's disease tracks forward-flexion angles well but overestimates lateral trunk flexion by about 5°.
Half the 10° threshold for Pisa syndrome, so it should not replace manual measurement for that diagnosis.
Key findings
1The automated system agreed well with manual measurements for forward-flexion angles (lATF ICC 0.96 unadjusted, 0.98 after calibration; tATF ICC 0.94, 0.96) and moderately for lateral trunk flexion (LTF ICC 0.79, 0.80).
2After operator-guided initial-frame calibration, the forward-flexion bias dropped to near zero (lATF mean difference −0.09°), but the lateral-flexion bias stayed at 4.99° — roughly half the 10° diagnostic threshold for Pisa syndrome — meaning patients with true angles of 5–10° could be misclassified as meeting that criterion.
3The lateral-flexion error grew systematically as forward trunk flexion increased (r = 0.563), so a single fixed subtraction would not correct the bias for patients with severe sagittal-plane postural abnormality.
Still to come
How it was doneWhat they foundWhat it means for PTs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
Only 14 participants, and no a priori sample-size calculation was performed; the authors acknowledge this is an exploratory validation. The 140 measurements used for reliability analysis were 10 repeated readings nested within each of the 14 people, and the ICC analysis did not model that within-person correlation, so confidence intervals may be slightly too narrow. Only 10 of roughly 120 available frames per participant were validated against manual measurement (about 8% of the continuous record). The persistent 4.99° lateral-flexion bias is clinically meaningful: it is about half the 10° Pisa syndrome threshold, and the proportional component means a single fixed correction will not work for patients with severe forward flexion. Pose-estimation accuracy degrades in severe postural deformity, with occlusion of C7 by the head and unreliable hip-joint detection being the main failure modes. Validation was done under realistic conditions (regular clothing, clinical background), which is a strength for generalisability but also explains wider limits of agreement than prior studies that used undergarments and uniform backgrounds.
Declared interests
Funded by the Yokohama Foundation for Advancement of Medical Science (grant to Y.K.). The authors declare no conflicts of interest. No company that sells the software or a competing product is named as funder, designer, or manuscript contributor.
The easy way to misread this
Do not read the 'good to excellent' reliability statement as clearance to use this tool for Pisa syndrome diagnosis. The lateral-trunk-flexion bias of 4.99° is about half the 10° diagnostic threshold, and because the error grows with forward flexion (r = 0.563), a patient with a true 6° lean and moderate camptocormia could easily be pushed past 10° by measurement error alone. The authors themselves state the LTF output should not substitute for manual measurement in diagnostic decision-making without a validated correction.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →