Deep Learning-Based Ground Reaction Force Estimation for Real-Time Clinical Applications.

Area of Science:

• Biomechanics
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Clinical gait analysis traditionally relies on expensive force plates.
• Marker-based motion capture offers rich kinematic data.
• Accurate estimation of Ground Reaction Forces (GRF) and Center of Pressure (COP) is crucial for diagnosing and monitoring gait abnormalities.

Purpose of the Study:

• To develop and validate a cost-effective, real-time method for estimating GRF and COP using marker trajectories.
• To leverage recurrent neural networks (RNNs) for gait analysis in clinical settings.
• To reduce dependency on specialized force plate equipment.

Main Methods:

• Marker trajectories from a motion capture system were used as input.
• A recurrent neural network (RNN) was trained to predict GRF and COP.
• Real-time treadmill walking experiments were conducted with five healthy subjects.

Main Results:

• The RNN model demonstrated high accuracy in estimating COP (r=0.81) and GRF (r=0.96) compared to ground truth.
• The method proved efficient and cost-effective for real-time gait analysis.
• High correlations indicate the model's predictive power.

Conclusions:

• Marker-based RNNs offer a viable, economical alternative for real-time GRF and COP estimation in clinical gait analysis.
• This approach has the potential to enhance accessibility and efficiency of gait assessments.
• Further validation with pathological gait data is needed for broader clinical application.