Quantitative evaluation of shoulder motor compensation in stroke patients based on multimodal feature analysis

Currently, the clinical assessment of upper-limb motor function in stroke survivors relies heavily on subjective, time-consuming standardized scales, such as the Fugl-Meyer Assessment (FMA). To address this limitation, this study validates a simplified quantitative evaluation approach that integrates markerless motion capture with surface electromyography (sEMG) to assess motor recovery objectively.

To evaluate this approach, 25 stroke patients were recruited to perform a targeted shoulder abduction task. During the task, a developed Markerless Upper Limb Motion Measurement System (MULMMS) synchronously captured kinematic data via Azure Kinect DK and sEMG signals from the Upper Trapezius and Middle Deltoid. Following trajectory smoothing via Singular Spectrum Analysis (SSA), a multidimensional feature set, including Maximum Joint Angle, smoothness metrics, and frequency-domain muscle activity, was extracted and validated against clinical scores using Spearman's correlation analysis.

Specific multimodal features demonstrated strong correlations with clinical scores. Maximum Joint Angle (Ang_max) yielded the strongest kinematic correlation (ρ = 0.932, p < 0.001). For electromyographic features, Median Frequency (ρ = 0.806, p < 0.001) showed a highly significant correlation with motor recovery levels. Statistical analysis confirmed this single-task approach effectively distinguished patients across different clinical score stages (p < 0.001).

The integration of markerless motion capture and sEMG provides an effective means to evaluate upper-limb motor impairment. Multimodal quantitative metrics based on a single core movement strongly correlate with traditional scales. Ultimately, this data-driven method for assessing shoulder motor compensation offers a standardized, simplified alternative that effectively complements subjective clinical observations.