Real‑time motion onset recognition for robot‑assisted gait rehabilitation

Journal of NeuroEngineering and Rehabilitation, 2022 · DOI: https://doi.org/10.1186/s12984-022-00984-x · Published: January 1, 2022

Assistive Technology Biomechanics

Simple Explanation

Many patients with neurological movement disorders fear falling during postural transitions, limiting their daily activities. Multi-directional Body Weight Support (BWS) systems offer a safe training environment. These systems can assist patients in training gait-related tasks. A challenge is manually switching between task-dependent supports, which is error-prone and cumbersome. A real-time motion onset recognition model is proposed for automatic support switching between standing-up, sitting-down, and other gait-related tasks, totaling 8 classes.

Study Duration

Not specified

Participants

19 controls without neurological movement disorders and two individuals with incomplete Spinal Cord Injury (iSCI)

Evidence Level

Not specified

Key Findings

1
A single-layer neural network with 25 neurons achieved the best performance among five classifiers tested.
2
The neural network model achieved an F1-score of 86.83% ± 6.2% in Leave-One-Participant-Out Cross-Validation (LOPOCV).
3
Real-time classifier performance was nearly identical to the offline classifier, with a difference of only 0.08%.

Research Summary

This study presents a real-time motion onset recognition system for gait-related tasks, crucial for automated gait analysis and switching between task-dependent supports in rehabilitation robots. A wireless interface board was designed to enable synchronized real-time data acquisition from Bluetooth-based IMUs, attached to the sternum and thighs of participants. A neural network model with 25 neurons was selected as the recognition model due to its robust performance. The model was tested and showed feasibility for real-time implementation.

Practical Implications

Automated Support Switching

The real-time motion onset recognition model can automate switching between task-dependent supports, reducing errors and improving training workflow.

Enhanced Rehabilitation Devices

The algorithm can be applied to various rehabilitation devices like BWS systems/exoskeletons, improving the support provided during specific tasks.

Personalized Therapy

The system contributes to creating more personalized rehabilitation programs and can be adapted to individual patient needs.

Study Limitations

1
It was challenging to compare recognition results to other studies due to different experimental setup (number of sensors, sensor placements, and recognition rate) and recognition method (offline/real-time).
2
Classifier has still low performance on patient data (patient #2).
3
The iSCI participant #2 was not able to perform tasks without walking aids such as standing upright.

Your Feedback

Was this summary helpful?

Back to Assistive Technology