Enterprise AI Analysis

AI-assisted video analysis of the Trendelenburg test: a feasibility study

Kieran O'Sullivan, Tom Doyle, Eoghain Quinn, Conor J. Kilkenny, Gordan Daly & Aiden Devitt

Background: The Trendelenburg test is widely used to assess hip abductor function, but interpretation is typically subjective and only moderately reliable. Compensatory trunk lean can mask subtle pelvic drop, further limiting diagnostic accuracy. Artificial intelligence (AI) based markerless motion analysis allows objective quantification of pelvic, trunk, and knee angles using standard video recordings.

Methods: This single-centre cross-sectional feasibility study was conducted in an Irish orthopaedic unit. Twelve adults were enrolled: seven post-total hip arthroplasty (THA) and five with native hip pathology. Each patient performed a standardised single-leg Trendelenburg test on both legs while being recorded with a single posteriorly placed smartphone camera. Videos were analysed offline using an AI-based markerless motion application (OnForm) to derive coronal-plane pelvic obliquity, trunk lean, and knee angle change between bipedal and single-leg stance. Continuous outcomes were summarised as medians with interquartile ranges (IQR) and ranges. Pre-specified thresholds (pelvic drop ≥4°, trunk lean ≥5°, knee angle change ≥3°) were used to describe the frequency of marked deviations.

Results: All patients completed the protocol with analysable recordings. The median video capture time was 32.5 seconds (IQR 23.5–36.0; range 19–42) and the median analysis time was 184.5 seconds (IQR 178.5–196.5; range 168–207), giving a median total workflow time of 215.5 seconds (IQR 203.5-232.5; range 193-244) per patient. Median worst contralateral pelvic obliquity was 0.0° (IQR –1.0° to +1.5°; range -5° to +6°). Median maximum trunk lean was 4.5° (IQR 2.8°-9.0°; range 2°-10°). Median coronal-plane knee angle change was 3.0° (IQR 2.0°-4.0°; range 1°-8°). Post-THA patients showed greater trunk compensation than those with native hips (median maximal trunk lean 9.0° vs 3.0°; median difference 6.0°), with trunk lean ≥5° in 5/7 post-THA and 1/5 native-hip patients. Knee deviations ≥3° were seen in 8 patients (67%).

Conclusions: AI-assisted single-camera analysis of the Trendelenburg test is feasible, rapid, and clinically informative. The method consistently quantified pelvic, trunk, and knee angles and demonstrated that post-THA patients frequently compensate with trunk lean rather than contralateral pelvic drop. This approach could enhance objective documentation of Trendelenburg performance and support postoperative rehabilitation monitoring. These findings are preliminary and hypothesis-generating; larger controlled studies with asymptomatic controls and reference standards are required to validate accuracy and clinical utility.