Subject

The development of knee osteoarthritis (OA) has been attributed to the overloading of articular cartilage. A common, non-invasive treatment strategy to reduce internal knee loads (i.e. Knee Contact Forces (KCFs)) is gait modification. The state of the art techniques for a non-invasive estimation of KCFs are based on complex, time-consuming musculoskeletal modelling and simulations, using software such as Opensim and Anybody. But recent studies (e.g. [1,2]) demonstrated that real time estimation of KFCs can be achieved using a Machine Learning (ML) or a Deep Learning (DL) approach instead of musculoskeletal modelling. These studies predict KFCs taking as input various data: joint trajectories, joint angles, EMG signals, GRFs and anthropometric data.

DL approaches thrive on diverse and extensive datasets: the more and varied the data input, the more accurate the AI's KCFs predictions. However, training datasets obtained using sensor-based knee prosthesis are scarce, and the generation of training data using modelling software is cumbersome.

So far, we have been working with the following open access datasets:

Grand Challenge Competition to Predict In Vivo Knee Loads [3]

CAMS Knee dataset [4]

Dataset of Lower Extremity Joint Angles, Moments and Forces in Distance Running [5]

While the first two contain each real data from 6 people walking with a prosthesis knee, the latter is a simulated dataset (motion data collected of 20 subjects walking, KFCs obtained using Opensim). A recent study [6] suggests using the concept of domain adaptation, particularly with CORAL (Correlation Alignment) layers, to bridge the gap between simulated and real datasets for the training and testing of KCFs prediction models.

Kind of work

Our goal is to standardize available datasets ([3-5]) and use them together to train a DL approach for KCF estimation. Once a baseline model is established, we aim to incorporate in the model CORAL layers for domain adaptation.

Description of Work

Literature review

Standardization of the dataset: definition of input and output features common to all the datasets

DL baseline model development and performance evaluation

Integration of CORAL layers and performance evaluation



The methods will be implemented in Python using open-source libraries and common deep learning frameworks. Additionally, for the standardization of the dataset, the use of musculoskeletal modelling software may be required (e.g. Opensim).

Framework of the Thesis

References?

?[1] G. Giarmatzis, et al. Real-Time Prediction of Joint Forces by Motion Capture and Machine Learning. Sensors. 2020.

[2] M. A. Boswell et al. A neural network to predict the knee adduction moment in patients with osteoarthritis using anatomical landmarks obtainable from 2D video analysis. Osteoarthritis and Cartilage. 2021

[3] https://simtk.org/projects/kneeloads

[4] https://orthoload.com/comprehensive-data-sample/

[5] Dataset of Lower Extremity Joint Angles, Moments and Forces in Distance Running

[6] I Loi et Al. Multi-Action Knee Contact Force Prediction by Domain Adaptation. IEEE Trans Neural Syst Rehabil Eng. 2024.

Number of Students

1