Video-estimated peak jump power using deep learning is associated with sarcopenia and low physical performance in adults
Performance · Agreement of vJP with gJP (test set)Result — go to the resultAverage difference 0.27 W/kg 95% limit of agreement CI - 5.01 to + 5.54Authors' conclusion
vJP was associated with sarcopenia, age, and muscle parameters in adults, with good agreement with ground truth jump power.
Study context
- Population
- Adults attending the osteoporosis clinic of Severance Hospital, Korea, March–August 2022; mean age 62 years, 77% women, sarcopenia 19%658 jump motion data
- Participants
- 220 adults
- Comparison
- Video-estimated peak jump power vs ground force plate jump power
- Outcome
- Agreement of vJP with gJP (test set)
Methodological detailsprospective, single-centre, deep learning pose estimation, 20% hold-out test set
- Prospective
- Single-centre
- Deep learning pose estimation
- 20% hold-out test set
Performance
As reported in the abstract
Average difference
0.27 W/kg
95% limit of agreement CI - 5.01 to + 5.54
By group
vJP
- Sensitivity for gJP-defined low jump power
- 81.8%
- Specificity for gJP-defined low jump power
- 91.3%
vJP
Sensitivity for gJP-defined low jump power81.8%
Specificity for gJP-defined low jump power91.3%
Further findings
vJP vs age and muscle function
- vJP showed a steep decline across age like gJP, with modest to strong correlation with handgrip strength (HGS) and CRT
- Most contributing features: eight landmarks (both shoulders, hip, knee joints, ears)
Association of vJP with sarcopenia, HGS and CRT (unadjusted and adjusted)
- Sarcopenia, unadjusted and adjusted
- - 3.95 and - 2.30 W/kg
- HGS, per 1 SD decrement, unadjusted and adjusted
- - 3.69 and - 1.96 W/kg
- log-CRT, per 1 SD decrement, unadjusted and adjusted
- - 2.79 and - 1.87 W/kg
- Similar to that of gJP
Original publication
Video-estimated peak jump power using deep learning is associated with sarcopenia and low physical performance in adults. Osteoporos Int. Published 15 May 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →