Frequency division technique on linear regression for robust simultaneous and proportional myoelectric control during medium and high contraction-level variation

Abstract

Myoelectric prosthesis are able to provide assistance to individuals with transradial amputations. Even though extensive research has been completed in machine learning techniques such as pattern recognition (PR) and regression, commercially available prostheses continue to use the control strategies prevailing since the 1960s. For clinical applications, the prosthesis should be simultaneous, proportional, intuitive and robust to the various non-stationaries in the EMG signal. In recent developments, frequency division technique (FDT) associated with PR has been capable of addressing contraction level variance as an EMG non-stationary. This study examined the performance of Linear Regression as a control scheme with FDT processing and investigated its robustness to contraction level variations. Twenty able-bodied and four individuals with trans-radial amputations performed wrist movements in 2 two degrees of freedom goal-oriented tasks, divided in three groups of Type I, Type II, and Type III. The performance of these tasks was assessed by the performance indices (time to reach, throughput, path efficiency, near miss and completion rate). Two different training contraction levels (medium and high) were performed for the two processing methods of Bandpass and FDT filtering. The results indicated that LR-FDT had an advantage over traditional methods in the tested real-time myoelectric control tasks. For control participants, LR-FDT had 95.33%, which was significantly higher than LR-Bandpass with a CR of 64.08% (p<0.001). For the clinical participants, all the individuals had a CR >90% using LR-FDT and had an average CR 69.8% using LR-Bandpass. Moreover, LR-FDT method performed significantly better in all the other performance indices in at least one target type. There was no significant difference in the performance of LR-FDT between medium and high runs. Also, the variability of LR-FDT was significantly smaller than LR-Bandpass. This study shows that LR-FDT provides advantages in online myoelectric control as it introduces a more accurate, robust and contraction level invariant control scheme for performing prosthetic hand movements.