EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Predicting Knee Contact Forces During Walking From Monocular Video and Demographic Information

Jere Lavikainen, HunMin Kim, David Pagnon, Jari Arokoski, Lauri Stenroth

Annals of Biomedical Engineering · 19.11.2026

DOI: 10.1007/s10439-026-04461-7 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Annals of Biomedical Engineering, online erschienen am 19.11.2026. Verfasst von Jere Lavikainen und 4 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Kuopion Yliopistollinen Sairaala, Academy of Finland, University of Eastern Finland.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Purpose

Motion capture conducted in motion laboratories can be used with musculoskeletal simulation to estimate knee contact forces (KCFs), which describe physical loading of the knee joint. Video-based markerless methods offer accessible alternatives to motion capture but cannot measure ground reaction forces, which are required alongside kinematics to estimate KCFs. If artificial neural networks could be trained to learn the mapping from kinematics to knee contact forces and then utilized on kinematics estimated from video recordings, the estimation of KCFs would become more accessible.

Methods

We trained artificial neural network models for predicting compressive total, medial, lateral, and patellofemoral KCFs during the stance phase of walking from sixteen joint angle time series and body mass, height, age, and sex. We then used an existing tool to estimate 3D kinematics from monocular video recordings and utilized the estimated kinematics to predict markerless KCFs. The markerless KCFs were compared against reference KCFs estimated using motion capture data and musculoskeletal simulation to quantify their accuracy.

Results

Root mean square error was 47, 40, 30, and 29 % of body weight for the predicted total, medial, lateral, and patellofemoral KCFs, respectively. Lateral KCFs were predicted poorly ( R = 0.64, negative coefficients of determination). Other KCFs correlated strongly with the reference forces ( R > 0.8).

Conclusion

We present a workflow for predicting KCF time series from single-camera video and subject demographics. While it is a promising first step toward out-of-lab estimation of KCFs, the workflow is currently insufficiently accurate for reliable individual-level loading assessment and requires further validation in larger and more diverse populations before clinical application can be considered.

Förderung

  • Kuopion Yliopistollinen Sairaala
  • Academy of Finland
  • University of Eastern Finland

Zitieren

APA

Lavikainen, J., Kim, H., Pagnon, D., Arokoski, J., & Stenroth, L. (2026). Predicting Knee Contact Forces During Walking From Monocular Video and Demographic Information. Annals of Biomedical Engineering. https://doi.org/10.1007/s10439-026-04461-7

Vancouver

Lavikainen J, Kim H, Pagnon D, Arokoski J, Stenroth L. Predicting Knee Contact Forces During Walking From Monocular Video and Demographic Information. Annals of Biomedical Engineering. 2026. doi:10.1007/s10439-026-04461-7

BibTeX
@article{Lavikainen2026,
  author = {Lavikainen, Jere and Kim, HunMin and Pagnon, David and Arokoski, Jari and Stenroth, Lauri},
  title = {Predicting Knee Contact Forces During Walking From Monocular Video and Demographic Information},
  journal = {Annals of Biomedical Engineering},
  year = {2026},
  doi = {10.1007/s10439-026-04461-7}
}
Kostenlos starten →