Non-invasive assessment of tibial torsion and lower limb alignment: A comprehensive evaluation of tibial torsion and femorotibial angles using motion capture technique in comparison to medical imaging

Aim and Research Question(s)

The aim of this thesis is to evaluate whether  motion capture data can be used to estimate tibial torsion (TT) and femotibial (FT) angles (genu valgum/varum) typically obtained via medical imaging. Using a combination of statistical and machine learning approaches, this work investigates:

• Can motion capture techniques provide an alternative to medical imaging for measuring TT?
• How does gender, age and body mass index (BMI) influence the accuracy of motion capture in assessing TT and FT angles?
• Does the data analysis confirm the previously established accuracy of motion capture, as demonstrated in the study by Stief et al. (2020), as a valid alternative to radiographs for assessing FT angles?

Background

TT and FT follow age-dependent normal trajectories, but deviations are frequent in pediatric populations and linked to patellofemoral instability, pain syndromes, and early-onset osteoarthritis. While CT and radiographs are the diagnostic gold standard, they expose patients to ionizing radiation. Motion capture represents a promising, radiation-free alternative that requires validation against imaging.

Methods

A retrospective dataset of 5,276 patients was filtered to 882 TT and 122 FT cases (<18 years). Motion capture derived values were compared to imaging using regression (linear, multiple) and machine learning models (ridge, lasso, random forest). Age, BMI, gender, and anthropometric features were included as predictors. Performance was evaluated with R², MAE, RMSE, and agreement assessed via Bland–Altman plots and error distributions.

Results and Discussion

For TT, motion capture showed only moderate agreement with imaging (r ≈ –0.58, R² ≈ 0.35–0.39, MAE ≈ 6.4°). Systematic deviations exceeded the ±2° threshold, excluding clinical interchangeability. Age improved prediction, while BMI and gender had no consistent influence. Machine learning approaches did not outperform standard regression. For FT, correlations were somewhat stronger (r ≈ 0.65–0.70, R² ≈ 0.49, MAE ≈ 1.5°).  Previously reported high validity (r ≥ 0.80, bias ±0.7°) could not be reproduced. Limitations: The applied BMI cut-off (<25) may not be appropriate for paediatric cohorts. A single train–test split was used without cross-validation, limiting generalizability. Finally, paediatric definition relied on chronological age (<18 years) rather than skeletal maturity, which may not fully reflect developmental differences.

Conclusion

Overall, motion capture reflects general alignment trends but does not achieve the precision required for clinical diagnostics. It may serve as a complementary, radiation-free tool for research and screening, while medical imaging remains indispensable for individual diagnostics.

References

• Albersheim et al., Curr Rev Musculoskelet Med, 15(6), 2022
• Kutanzi et al., International Journal of Environmental Research and Public Health, 13(11), 2016
• Marques Luís et al.,  EFORT Open Rev, 6(6), 2021
• Rerucha et al., Am Fam Physician, 96(4), 2017
• Stief uet al., Gait Posture, 79, 2020