Predicting Knee Joint Contact Forces During Normal Walking Using Kinematic Inputs With a Long-Short Term Neural...

Area of Science:

• Biomechanics
• Biomedical Engineering
• Computational Science

Background:

• Knee joint contact forces are crucial for understanding knee function and pathology.
• Current estimation methods like external moments and musculoskeletal modeling have limitations in accessibility and clinical applicability.
• A simplified, accurate method for predicting knee forces is needed.

Purpose of the Study:

• To develop and validate a novel, simplistic prediction method for knee joint contact forces using minimal kinematic inputs.
• To compare the accuracy of the novel method against traditional musculoskeletal modeling.

Main Methods:

• Utilized marker trajectories and instrumented knee forces from the Grand Challenge and CAMS datasets.
• Derived lower limb and trunk kinematics via inverse kinematics.
• Developed a long-short term memory (LSTM) network to predict medial and lateral knee forces using kinematic data.
• Derived forces using OpenSim musculoskeletal modeling for comparison.

Main Results:

• The LSTM network achieved high accuracy for medial knee forces (R2 = 0.77), requiring only frontal hip/knee and sagittal hip/ankle kinematics.
• Musculoskeletal modeling also showed good medial force prediction (R2 = 0.77).
• Both methods demonstrated poor accuracy for lateral knee forces (LSTM R2 = 0.18, Modeling R2 = 0.21).

Conclusions:

• A deep learning approach using LSTM networks can accurately predict knee joint medial contact forces with simplified kinematic inputs.
• This method shows promise as a more accessible alternative to complex musculoskeletal modeling for estimating knee joint forces.
• Further research is needed to improve lateral force prediction accuracy.