MoCap-Referenced Neck–Shoulder sEMG–IMU Decoding for Discrete Assistive Commands: A Pilot Study

Hands-free command interfaces are essential for users who cannot reliably operate joysticks or upper-limb myoelectric control. Neck–shoulder surface electromyography (sEMG) is a promising alternative; however, performance is often reported using window-level validation which can overestimate accuracy due to overlap and trial leakage, and false-trigger behavior is not always quantified when an idle REST state is included. This pilot study presents a motion-capture (MoCap)-referenced decoding framework that uses four bilateral upper trapezius (UT) and sternocleidomastoid (SCM) sEMG channels with integrated inertial measurement units (IMUs). Optical MoCap was used as an external kinematic reference to support baseline-posture assessment and movement-execution quality control. Seven commands were decoded (shrug L/R, double shrug, rotation L/R, rotation + shrug L/R). To enable an eight-class formulation, a REST class was defined using low-activity segments extracted from baseline recordings and included in the evaluation. Computationally efficient time-domain sEMG features, pattern/symmetry descriptors, and baseline-referenced IMU kinematics (including an SCM yaw-range indicator) were classified using linear discriminant analysis (LDA), k-nearest neighbors (kNN), and linear support vector machine (SVM), evaluated using within-subject testing, trial-wise grouped cross-validation, and leave-one-subject-out (LOSO) testing. Across six participants, within-subject mean best-per-subject accuracy was 96.02% (seven-class) and 96.35% (eight-class); and pooled trial-wise accuracy reached 92.1% and 90.5%, respectively. Under LOSO, best-configuration accuracy decreased to 60.4% and 63.8% for the seven-class and eight-class formulations, respectively. Across the top LOSO configurations, REST FAR ranged from approximately 9.8% to 25.6%. These findings demonstrate controlled offline pilot feasibility and quantify key generalization and REST false-activation trade-offs, providing a foundation for future validation in larger, more diverse, and clinically relevant populations.