Methods
We conducted a PRISMA-compliant systematic review of PubMed, IEEE Xplore, and Scopus (2015-2024). Included studies directly compared IMU and vision-based (e.g., Kinect, MediaPipe) systems for upper limb motion analysis in neurological populations, reporting quantitative accuracy metrics (RMSE, correlation coefficients) or clinical utility data. Two reviewers independently screened studies, extracted data, and assessed methodological quality using the QUADAS-2 tool. Meta-analysis was performed for comparable accuracy measures.
Results
From 1,038 records, 22 studies met inclusion criteria (n=687 participants). IMUs demonstrated superior accuracy for joint angle measurements (mean RMSE 4.3° vs 8.7° for vision systems, p<0.001), particularly in complex movements involving trunk compensation. Vision-based systems showed better usability for gross movement assessment (setup time 62% faster, p=0.01) but suffered 31% more tracking failures during occluded motions. Clinical implementation favored IMUs for home monitoring (78% of studies) versus vision systems for clinic-based evaluation (65%). Both technologies showed strong correlation with gold-standard motion capture (r>0.89) but differed in sensitivity to movement quality parameters.
Conclusion
IMUs provide more reliable kinematic data for detailed movement analysis, while vision systems offer practical advantages for rapid clinical assessment. A hybrid approach leveraging both technologies may optimize rehabilitation monitoring across care settings. Future development should address occlusion challenges in vision systems and improve IMU data interpretation for clinical decision-making.
