How few channels suffice? A hierarchical validation framework for minimal-channel wearable sEMG in upper-limb rehabilitation gesture recognition.

Abstract

Background: Surface electromyography (sEMG) is widely adopted in upper-limb rehabilitation for decoding motor intent. However, current wearable systems typically require eight or more channels, imposing penalties on device complexity, power consumption, and patient comfort. Reducing channel count without sacrificing recognition accuracy remains an open challenge for user-independent rehabilitation scenarios. We hypothesize that systematically optimizing channel configurations under high-fidelity acquisition conditions can identify a minimal viable topology suitable for clinical deployment.

Methods: We present a hierarchical validation framework integrating three pillars of evidence. First, we developed a custom acquisition system achieving a common-mode rejection ratio exceeding 112 dB, ensuring that observed inter-channel differences reflect physiology rather than hardware artifacts. Second, we performed an exhaustive subset search across all 63 non-empty combinations of six channels in 24 healthy subjects under leave-one-subject-out (LOSO) cross-validation. Third, a dual-track sensitivity analysis was conducted by training on physiologically augmented data while testing under hardware-degraded conditions.

Results: Results reveal a tiered structure: the 3-channel configuration serves as the efficiency knee point (91.65%), exceeding the 90% clinical reliability threshold, although its cross-subject generalizability is limited by a maximum regret of 10.5% and only 8/24 subjects achieving zero regret under the global 3-channel topology. The 4-channel configuration acts as the recommended optimal (94.44%), eliminating worst-case blind spots ([Formula: see text]). An overlap ratio analysis confirms high alignment between global and individual optima, with 67% of subjects achieving zero regret under the global 4-channel topology. The dual-track sensitivity analysis demonstrates a 16% robustness gain under low-SNR conditions, validating that high-fidelity acquisition is the functional prerequisite for channel reduction.

Conclusions: Within the six-site candidate pool studied here, the globally fixed 4-channel topology achieved performance close to the 6-channel baseline under rigorous LOSO validation. While the 3-channel configuration meets the 90% clinical threshold on average, its substantial inter-subject regret (up to 10.5%) limits its suitability as a universal solution. The optimal 4-channel topology converges on four forearm muscles spanning the wrist's two kinematic degrees of freedom, providing an anatomically grounded, patient-independent design principle. These findings offer a practical pathway toward lightweight, comfortable, and clinically viable rehabilitation interfaces that could improve patient compliance and therapeutic outcomes.