Determining the optimal window length for pattern recognition-based myoelectric control: balancing the competing...

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Neuroprosthetics

Background:

• Pattern recognition in myoelectric prostheses shows promise but requires clinical optimization.
• Understanding the interplay between classification error, controller delay, and real-time usability is crucial for clinical translation.

Purpose of the Study:

• To investigate the relationship between classification error, controller delay, and real-time controllability in pattern recognition-based myoelectric prostheses.
• To determine optimal parameters for myoelectric prosthesis control in a clinical setting.

Main Methods:

• 13 able-bodied subjects controlled a virtual upper-limb prosthesis using electromyogram (EMG) pattern recognition.
• Classification error and controller delay were manipulated by varying analysis window lengths (50-550 ms) and EMG channel count (2 or 4).
• Real-time controllability was assessed using the Target Achievement Control (TAC) test.

Main Results:

• Offline analysis revealed that longer window lengths significantly decreased classification error (p < 0.01).
• User performance on the TAC test improved with lower classification error (p < 0.01) and decreased with longer controller delay (p < 0.01).
• The optimal window length for the studied system was identified as 150-250 ms, balancing error reduction and acceptable delay.

Conclusions:

• Both classification error and controller delay are critical factors influencing real-time controllability of myoelectric prostheses.
• Increasing analysis window length can be beneficial if it reduces classification error, even with increased delay.
• The 150-250 ms window length range offers a practical compromise for effective clinical implementation of EMG-based control.