Improving robustness against electrode shift of high density EMG for myoelectric control through common spatial...

Area of Science:

• Biomedical Engineering
• Neuroprosthetics
• Signal Processing

Background:

• Low-density electromyography (LD EMG) has been standard for myoelectric control due to clinical applicability and lower computational demands.
• Recent advancements in high-density electromyography (HD EMG) electrode wearability make it a viable alternative for myoelectric prostheses.
• Electrode shift during prosthetic use significantly degrades classification accuracy (CA) in myoelectric control systems.

Purpose of the Study:

• To evaluate the effectiveness of common spatial patterns (CSP) feature extraction methods for improving robustness against electrode shift in HD EMG-based myoelectric control.
• To compare the performance of CSP features (CSP-OvO and CSP-OvR) against traditional features like time-domain (TD), time-domain autoregressive (TDAR), and variogram (Variog).

Main Methods:

• Acquired HD EMG signals from the forearm of nine intact-limb subjects performing eleven hand and wrist motions.
• Applied multiclass CSP feature extraction using one-versus-one (CSP-OvO) and one-versus-rest (CSP-OvR) schemes.
• Assessed classification accuracy under transversal (ST1, ST2) and longitudinal (SL1, SL2) electrode shift configurations.

Main Results:

• CSP features significantly improved CA by over 10% compared to TD features across all shift configurations.
• CSP-OvO and CSP-OvR showed significant CA improvements over TDAR features (over 5% average) in various shift scenarios.
• CSP features demonstrated significant CA gains over Variog features, particularly in longitudinal shift conditions (SL1, SL2).

Conclusions:

• Common spatial patterns (CSP) features offer a substantial improvement in the robustness of HD EMG-based myoelectric control against electrode shifts.
• CSP-based feature extraction presents a promising approach to overcome performance degradation caused by electrode displacement in prosthetic devices.