Unlocking Gait Analysis Beyond the Gait Lab: High-Fidelity Replication of Knee Kinematics Using Inertial Motion Units...

Area of Science:

• Biomechanics
• Human Movement Analysis
• Wearable Technology

Background:

• Three-dimensional (3D) motion capture is valuable for gait analysis but is expensive and inaccessible.
• Wearable inertial motion sensors (IMUs) are cost-effective but typically measure only simple temporospatial variables.
• Complex kinematic data from IMUs often requires machine learning (ML) for accurate analysis.

Purpose of the Study:

• To investigate if ML algorithms can accurately replicate knee kinematic data typically obtained from 3D motion capture using data from IMUs.
• To determine the feasibility of using wearable sensors and ML for advanced gait analysis.

Main Methods:

• Collected gait data from 40 healthy participants during walking, stair climbing, and sit-to-stand tasks using both 3D motion capture and IMUs.
• Trained sequence-to-sequence convolutional neural networks (CNNs) to map IMU data to right knee angle, right knee angular velocity, and right hip angle.
• Assessed model performance using mean absolute error (MAE).

Main Results:

• CNN models demonstrated high accuracy in replicating 3D motion capture-derived kinematic variables.
• MAE for right knee angle ranged from 4.30 ± 1.55 to 5.79 ± 2.93 degrees.
• MAE for right knee angular velocity ranged from 7.82 ± 3.01 to 22.16 ± 9.52 degrees/second, and for right hip angle from 4.82 ± 2.29 to 8.63 ± 4.73 degrees.
• Observed task-specific variations in model accuracy.

Conclusions:

• Leveraging raw data from wearable IMUs and ML algorithms shows potential for reproducing lab-quality kinematic data.
• This approach enables gait analysis outside traditional laboratory settings.
• The findings are significant for studying knee function in conditions like joint injury, post-surgery, or degenerative joint diseases.