Subject

Determining contact forces in the knee joint during walking is relevant for the study of knee pathology and rehabilitation thereof. In particular, the knee joint loading, which the force at the joint over a period of time, is relevant. However, it is obviously completely impossible to directly measure such forces in the knee in vivo in a walking human. There exist however advanced musculoskeletal simulation models in biomechanics that try to predict muscle forces from the kinematic data and ground reaction force data captured in the gait lab. Such models can be validated on a few available datasets from subjects with a knee prosthesis in which a force sensor is embedded [Fregly et al, 2012].

Kind of work

With the recent availability of several benchmark datasets for this problem and the adoption of machine learning methods in biomechanics, there are now initial attempts to simply train regression models (varying from simple linear methods to deep neural networks) to output knee contact force based on the joint positions as determined from motion capture technology. This approach is promising, but a large variety of research questions remain to be addressed. Your goal is to implement and replicate a state-of-the-art method [Rane et al, 2019] to build on top of it.
In a discussion between the student and the promotor, we can explore different avenues for research in the master thesis. Example topics are:
• What is the impact of additional training data on the performance of the mode? Additional databases became available and could be used.
• What is the importance of the different joint positions and measured ground reaction forces in the model? If certain information is not found to contribute to the model’s performance, then this data should not be collected. This can potentially result in a lower number of reflective markers to be positioned on the patient.
• In how far can the knee contact forces be computed without using ground reaction forces and from low-quality joint positions? If that would be possible, knee joint loading can be estimated from low-cost solutions such as the Kinect camera.
• In how far is the model patient specific and in how far does the model generalize over multiple subjects? For subject specific models, recent machine learning methods still allow to use training data from other subjects. This will once more increase the robustness of the model.
• A far more practical question that could be included in the study is the integration of the developed model into an existing state-of-the-art biomechanics software.
Fregly, B.J., Besier, T.F., Lloyd, D.G., Delp, S.L., Banks, S.A., Pandy, M.G., and D'Lima, D.D. (2012) Grand challenge competition to predict in vivo knee loads. Journal of Orthopaedic Research 30, 503-513.
Rane, L., Ding, Z., McGregor, A.H. et al. Deep Learning for Musculoskeletal Force Prediction. Ann Biomed Eng 47, 778–789 (2019). https://doi.org/10.1007/s10439-018-02190-0