Interlimb Gait Trajectory Synchronization from Inertial Measurement Unit (IMU) with ResUNet.

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Machine Learning for Healthcare

Background:

• Gait rehabilitation is crucial for patients with neuronal injuries, significantly impacting quality of life.
• Current rehabilitation often neglects the coordinated role of all four limbs, separating upper limb and gait training.
• Interlimb coordination is fundamental to human gait physiology.

Purpose of the Study:

• To develop and evaluate autoencoder-based networks (ResNetAE, ResUNet) for interlimb gait trajectory synchronization.
• To enable active control in gait rehabilitation systems by synchronizing upper and lower limb movements.
• To improve the effectiveness of gait rehabilitation by incorporating a holistic, interlimb approach.

Main Methods:

• Collected gait data from 30 healthy subjects across three age groups (young, middle, elderly).
• Utilized the Xsens MVN Inertial Measurement Unit (IMU) system to extract and synchronize body segment positions.
• Employed autoencoder networks with residual blocks (ResNetAE, ResUNet) to predict interlimb trajectories using hand and foot coordinate data.

Main Results:

• Achieved a Mean Absolute Error (MAE) of 0.0472 meters in predicting interlimb trajectories.
• Demonstrated high phase synchronization with a Phase Synchronization Index (PSI) of 96.70%.
• Outperformed the LSTMAE baseline in accuracy and synchronization on unseen subjects.

Conclusions:

• The proposed ResNetAE and ResUNet models show significant promise for interlimb gait trajectory synchronization.
• The system's performance is adequate for advancing interlimb phase synchronization research.
• These findings support the potential integration of interlimb-aware systems in gait rehabilitation exoskeleton technology.