Exploiting accelerometers to improve movement classification for prosthetics.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Myoelectric control of prostheses aims to restore lost limb function.
• Surface electromyography (sEMG) is a common input modality, but has limitations.
• Integrating additional sensor data can enhance prosthetic control.

Purpose of the Study:

• To evaluate the effectiveness of arm dynamics, measured by accelerometers, as an input modality for myoelectric prostheses.
• To compare the performance of accelerometer data against sEMG for movement classification.
• To investigate the benefits of multi-modal classification combining both sensor types.

Main Methods:

• A large-scale movement classification task involving 40 distinct hand and wrist movements.
• Data collected from 20 human subjects.
• Machine learning classifiers were trained and evaluated using accelerometer signals, sEMG signals, and a combination of both.

Main Results:

• Accelerometer signals were found to be highly informative for movement classification.
• Accelerometer-based classification achieved higher accuracy than sEMG-based classification.
• The highest classification accuracy was achieved by integrating both accelerometer and sEMG modalities in a multi-modal classifier.

Conclusions:

• Arm dynamics measured by accelerometers represent a valuable input modality for improving myoelectric prosthesis control.
• Accelerometer data alone can outperform traditional sEMG for classifying a wide range of hand and wrist movements.
• Multi-modal approaches combining accelerometers and sEMG offer the most robust and accurate solution for advanced prosthetic control.