An AI-based assessment of musculoskeletal health
Motion Age
A cross-sectional study of 1,139 adults examining the biomechanical features that contribute most strongly to an age-equivalent movement index — measured entirely from markerless motion analysis.
1,139
Participants
2 × 3
Tasks × repetitions
Markerless
Capture
Why Motion Age
Movement is a whole-system signal.
Movement is a whole-system expression of joint range, balance, coordination, neuromuscular control and strength. Conventional assessments can be subjective or isolate a single parameter. Motion Age combines multiple kinematic features and expresses the resulting model output as an age-equivalent value that is easier to communicate.
Motion Age is an AI-derived index of how an individual’s recorded movement resembles the movement patterns represented in the study dataset.
Working definition — what it is, and what it is not
The output
An age-equivalent index generated by a supervised model trained on markerless movement features.
The evidence
Feasibility, feature attribution and age-estimation performance within this cross-sectional dataset.
Not established
Diagnosis, prediction of future decline, treatment response or a direct measure of biological ageing.
A Short, Standardised Protocol
Two everyday movements, captured cleanly.
Participants completed three sit-to-stand repetitions and three squats in a controlled indoor environment. From standard video, MAI Motion generated markerless 3D kinematics, a supervised model then estimated Motion Age and ranked feature attribution.
Sit-to-stand
Natural, self-selected pace from a chair with backrest and armrests.
Three consecutive repetitions
Squat
Shoulder-width stance and descent to a comfortable, self-selected depth.
Three consecutive repetitions
Record
Standard RGB camera; 1920 × 1080 resolution at 60 frames per second.
Extract
MAI Motion® generated markerless 3D joint and segment kinematics.
Characterise
Angles, range of motion, ab/adduction, smoothness and asymmetry were calculated.
Model
Google AutoML estimated Motion Age and ranked relative feature attribution.
What Drove The Model
Knee mechanics carried the strongest signal.
Knee-specific mechanics — particularly frontal-plane range and asymmetry — carried the strongest attribution scores. Scores describe relative influence inside this model. They are not clinical thresholds, causal effects or measures of disease severity.
Leading feature attribution
- Right knee ab/adduction — average RoM7
- Left knee angle — average RoM6
- Knee asymmetry4
- Left shoulder–knee segment — average RoM3
- Trunk–core smoothness / core asymmetry2
Signal, Error And Calibration
A real trend, reported with honest limits.
The model captured a broad age-related trend but retained substantial unexplained variation across the full cohort. These metrics do not support using Motion Age as a precise individual age estimate or diagnostic result.
13.4 y
MAE
Mean absolute error
16.1 y
RMSE
Root-mean-square error
32.6%
MAPE
Mean absolute percentage error
0.37
R²
Variance explained
Age-band calibration
Prediction error was lowest in densely represented mid-life and early older-age bands. The reported MAE in the 60–65-year band was approximately 3.3 years; this band-specific result must not be generalised to the full population.
Calibration finding
Predictions compressed towards the late-fifties to early-sixties: younger ages tended to be overestimated and older ages underestimated. Chronological band slope was 0.991 against a mean Motion Age slope of 0.364.
Evidence Status
A research index — not a diagnostic test.
Motion Age is a feasibility and feature-attribution study. It demonstrates that a supervised model can express markerless movement patterns as an age-equivalent value, and identifies which biomechanical features drive that output.
Prospective validation against clinical outcomes is required before any diagnostic, prognostic or treatment-response use. The index is not a measure of biological ageing.
Motion Age
An AI-based assessment of musculoskeletal health from markerless motion analysis.
Background
Functional mobility declines with age and is associated with frailty, falls and poor health outcomes. This study evaluated the biomechanical features most predictive of an age-equivalent Motion Age index using MAI Motion.
Methods
This cross-sectional study analysed 1,139 adults performing three sit-to-stand and three squat repetitions. Markerless 3D kinematic features were extracted and a supervised AI model estimated Motion Age and ranked feature attribution.
Results
Full-cohort MAE was 13.4 years (RMSE 16.1; R² 0.37). Knee ab/adduction range, knee angle range and knee asymmetry were the leading contributors. Error was lowest in densely represented mid-life bands, but predictions were compressed towards the sample mean.
Conclusion
Motion Age is a feasible, interpretable movement index with potential for screening and longitudinal monitoring. Prospective validation against clinical outcomes is required before diagnostic or treatment-response use.
From Age Estimation To Meaningful Change
The research agenda ahead.
The next phase is to establish whether Motion Age is reliable, responsive and associated with outcomes that matter to patients.
External validation
Test performance in independent and demographically balanced cohorts.
Reliability
Quantify repeatability across sessions, settings, cameras and assessors.
Clinical validity
Relate Motion Age to pain, function, falls, frailty and imaging phenotypes.
Responsiveness
Determine whether meaningful change follows rehabilitation or treatment.
Bias correction
Address regression towards the mean and improve calibration at age extremes.
Study Identity
Transparent by design.
Design
Cross-sectional observational study
Study team
Amin Sohani · Tanvi Verma · Yan Wen · Paul Lee
Platform
MAI Motion® markerless 3D movement analysis
Modelling
Google AutoML supervised machine-learning model
Governance
IRB M2500170001; approved 17 January 2025
This overview summarises the supplied manuscript. Motion Age is not established as a diagnostic test, biological-age measure or treatment-response endpoint. The complete manuscript contains the full literature context and reference list.