From Force Plates to AI: Establishing Validity and Reliability of a 2D AI-Camera System for Quantifying Sit-to-Stand...

Muscle power is a critical determinant of functional capacity and overall health, particularly in aging and athletic populations. The 30-second Sit to Stand Power Test (30STSPT) offers a practical means of assessing lower limb power, yet its widespread clinical adoption is limited by the need for specialized equipment. Emerging technologies, such as 2D Artificial Intelligence (AI)-based camera systems, may offer scalable and accessible alternatives for power assessment. # PURPOSEThe purpose of this study was to determine (1) the concurrent validity against a dual force plate system and (2) the test-retest reliability of a 2D AI-camera for capturing and calculating the muscle power for a 30STSPT. It was hypothesized that the 2D AI-camera would have high test-retest reliability and strong concurrent validity with the power measured by a dual force plate system and inertial measurement unit (IMU). # STUDY DESIGNValidation and reliability study # METHODSA convenience sample of 24 healthy adults (20-55 years) completed two maximal-effort trials of the 30-second Sit-to-Stand Power Test (30STSPT). During each trial, repetitions were counted by research personnel, the AI-based camera system, and the criterion system (dual force plates synchronized with an inertial measurement unit [IMU]). The AI system automatically calculated trial-level mean power (W·kg^-1) using body mass, stature, chair height, and performance time via a validated equation. The criterion method computed power from average peak vertical ground-reaction forces and IMU-derived vertical displacement. Concurrent validity between AI and criterion power was assessed using Pearson's correlation coefficient (r) with 95% confidence intervals (CI) and Bland-Altman analysis. Test-retest reliability for AI and criterion measures was evaluated using a two-way mixed-effects intraclass correlation coefficient (ICC(3,1)) with 95% CI, and measurement error was quantified via the standard error of measurement (SEM) and minimal detectable change at 95% confidence (MDC₉₅).

A total of 24 individuals (M:F, 9:15) with a mean age of 34.4 ± 9.4 years and an average BMI of 24.9 ± 4.1 kg·m^-2 completed two trials of the 30STSPT. AI-derived power demonstrated excellent correlation with the criterion method for Trial 1 (r = 0.945, 95% CI 0.861-0.979) and Trial 2 (r = 0.934, 95% CI 0.833-0.975). Bland-Altman analysis showed a mean bias of +0.66 W·kg^-1 (LoA: -0.51 to +1.83) for Trial 1 and +0.53 W·kg^-1 (LoA: -1.07 to +2.12) for Trial 2, with proportional bias evident in both trials (Trial 1 slope = -0.195, p = 0.027; Trial 2 slope = -0.285, p = 0.0049). Test-retest reliability of AI-derived power was excellent (ICC(3,1) = 0.942, 95% CI 0.860-0.977), with SEM = 0.362 W·kg^-1 (7.13%) and MDC₉₅ = 1.004 W·kg^-1 (19.8%). Criterion reliability was good-to-excellent (ICC(3,1) = 0.916, 95% CI 0.790-0.968), with SEM = 0.489 W·kg^-1 (11.4%) and MDC₉₅ = 1.355 W·kg^-1 (31.6%).

The findings of this study support the use of a 2D AI-camera system as a valid and highly reliable tool for quantifying muscle power during the 30-second Sit to Stand Power Test. The 2D AI-camera system offers a promising solution for scalable, objective performance testing in clinical and remote settings.