Effects of Confidence-Based Rejection on Usability and Error in Pattern Recognition-Based Myoelectric Control

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Rehabilitation Engineering

Background:

• Pattern recognition-based myoelectric control enhances prosthetic limb functionality.
• Movement rejection strategies improve system usability by filtering unreliable classifications.
• Previous studies determined rejection thresholds heuristically, lacking analysis of their impact on real-time control.

Purpose of the Study:

• To investigate the effect of varying rejection thresholds on the trade-off between error reduction and false rejections in myoelectric control.
• To determine the optimal confidence threshold for movement rejection in real-time closed-loop control.
• To differentiate and analyze operator versus systemic errors and their correlation with performance metrics.

Main Methods:

• A support vector machine classifier was used for real-time classification of myoelectric signals.
• Twenty-four able-bodied subjects performed a Fitts' law-style virtual cursor control task.
• Movement rejection was implemented at various confidence thresholds to assess its impact.

Main Results:

• Movement rejection improved information throughput across all tested thresholds.
• The highest performance was observed at rejection thresholds between 0.60 and 0.75.
• Increasing rejection thresholds reduced both operator and systemic errors.
• Systemic errors, unlike operator errors, strongly correlated with throughput and trial completion rates.

Conclusions:

• Movement rejection significantly enhances the usability of myoelectric control systems.
• A confidence threshold between 0.60 and 0.75 offers an optimal balance for real-time control.
• Usability encompasses more than just error prevention, as experienced users still committed errors despite better task performance.