Estimating Lower Extremity Running Gait Kinematics with a Single Accelerometer: A Deep Learning Approach.

Area of Science:

• Biomechanics
• Sports Science
• Machine Learning

Background:

• Abnormal running kinematics increase lower extremity injury risk.
• Accurate gait analysis is crucial for runner injury prevention.
• Existing inertial methods require complex setup or calibration.

Purpose of the Study:

• To validate a shoe-mounted accelerometer for measuring lower extremity angles during running.
• To utilize a deep learning approach for accurate gait analysis.
• To minimize sensor setup and participant-specific calibration requirements.

Main Methods:

• A convolutional neural network (CNN) was employed for regression analysis.
• Ten participants ran on a treadmill at five different speeds.
• Optical motion capture system provided reference joint angle data.

Main Results:

• CNN model achieved root mean squared error (RMSE) < 3.5° (intra-participant) and < 6.5° (inter-participant).
• Gait event estimation errors were < 2.5° (intra-participant) and < 6.5° (inter-participant).
• The approach demonstrated generalization across participants.

Conclusions:

• A shoe-mounted accelerometer with a deep learning model offers a minimal sensor setup for running gait analysis.
• This method accurately measures lower extremity angles and key gait events.
• It presents a promising solution for detecting gait abnormalities and preventing running injuries.