Video-based gait analysis for clinical monitoring of genotype-specific functional patterns in osteogenesis imperfecta.

backgroundOsteogenesis imperfecta (OI) is characterized by substantial genetic and functional heterogeneity, yet low-burden objective tools for repeated gait monitoring in routine outpatient care remain limited. This study investigated whether routine smartphone videos combined with markerless pose estimation could capture genotype-associated gait patterns in children with IFITM5- and WNT1-related OI, with COL1-related OI included as a cross-sectional reference cohort.

methodsThis single-center retrospective cohort study screened 237 pediatric patients with genetically confirmed IFITM5- or WNT1-related OI between 2014 and 2024. The main analytic cohort included 71 patients: 48 with IFITM5 mutations and 23 with WNT1 mutations. During the same period, 107 patients with COL1-related OI were screened, of whom 26 were included as a cross-sectional reference cohort. MediaPipe Pose was used to extract spatiotemporal and postural gait parameters from outpatient walking videos. Algorithmic consistency under simulated video perturbations was assessed, and gait phenotypes were evaluated using multivariable models, linear mixed-effects models, Kaplan-Meier analysis, and exploratory machine-learning analysis.

resultsStep_Width and Trunk_Lean showed high algorithmic consistency under perturbation testing, with ICC ranges of 0.966-0.980 and 0.923-0.983, respectively. At baseline, WNT1 patients showed greater Step_Width (0.265 ± 0.073 vs. 0.203 ± 0.062 m), Trunk_Lean (4.53° ± 2.10° vs. 2.72° ± 1.63°), and Normalized_Step_Width (0.377 ± 0.103 vs. 0.270 ± 0.083) than IFITM5 patients (all P < 0.001). These differences remained after adjustment for age, sex, documented medication exposure, and surgical history, with adjusted differences of 0.046 m for Step_Width, 1.612° for Trunk_Lean, and 0.087 for Normalized_Step_Width. In the COL1-related reference analysis, WNT1 patients retained higher Step_Width and Normalized_Step_Width than COL1-related OI patients, whereas Trunk_Lean did not differ significantly between these groups. Longitudinal analysis showed no significant age-related gait changes in the WNT1 group, whereas IFITM5 patients showed mild annual decreases in Step_Width (- 0.0026 m/year) and Normalized_Step_Width (- 0.0046/year). Surgery-free survival declined earlier in the WNT1 group than in the IFITM5 group (log-rank P = 0.007). The exploratory genotype classification model achieved an AUC of 0.804 (95% CI, 0.678-0.914).

conclusionsRoutine outpatient smartphone videos combined with markerless pose estimation can capture genotype-associated gait patterns in pediatric OI. WNT1-related OI was characterized by a wider base of support and coronal-plane compensation, whereas IFITM5-related OI showed mild longitudinal changes in spatial gait metrics. Video-derived gait parameters may provide low-burden adjunctive measures for outpatient functional monitoring, although prospective reference-standard validation and larger multi-genotype cohorts are needed.