GMAC-A Simple Measure to Quantify Upper Limb Use From Wrist-Worn Accelerometers.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Wearable Sensors

Background:

• Quantifying upper-limb use is crucial for rehabilitation and monitoring.
• Existing methods like thresholded activity counts (TAC) and gross movement (GM) have limitations in sensitivity and specificity.
• A previous hybrid measure, GMAC, improved detection but required further optimization.

Purpose of the Study:

• To develop a modified GMAC measure using only accelerometer data.
• To optimize GMAC parameters for both generic (limb- and subject-independent) and limb-specific (subject-independent) applications.
• To evaluate the performance of the optimized GMAC against previous methods and machine learning models.

Main Methods:

• A modified GMAC algorithm was developed utilizing solely accelerometer data from wrist-worn inertial measurement units.
• Parameters for the GMAC were optimized to create generic and limb-specific versions.
• Performance was assessed using hemiparetic data, comparing the optimized GMAC to the original GMAC and a random forest machine learning model.

Main Results:

• The optimized GMAC demonstrated superior detection performance compared to the original GMAC.
• The optimized generic GMAC achieved performance comparable to a leading machine learning model (random forest inter-subject model).
• In hemiparetic data, the limb-specific optimized GMAC outperformed the generic version and matched the performance of the random forest model.

Conclusions:

• The optimized limb-specific GMAC provides a simple, interpretable, and effective alternative to complex machine learning models for upper-limb use detection.
• This measure holds significant potential for both offline and real-time applications in monitoring and feedback systems.
• Further validation on larger datasets is recommended to confirm the generalizability of these findings.