Answers are grounded in this paper and cite their sources.
Smart telerehabilitation requires accurate motion assessment while protecting patient privacy and supporting real-time interaction between patients and clinicians. RGB-based approaches can expose identifying appearance information, whereas conventional depth-based methods may insufficiently represent long-range temporal dependencies and heterogeneous motion features. This study presents NeuroFusion-Rehab, an IoT-enabled depth-centric framework for privacy-preserving telerehabilitation and cloud-assisted monitoring. Depth silhouettes are processed using geometry-constrained CoTracker for dense motion tracking, depth-adapted DISK for shape-aware keypoints, DINOv2 for structural self-attention features, and 3D point clouds for geometric representation. A Hybrid Adaptive Spectro-Frequency Encoding (HASFE) module combines wavelet scattering and empirical mode decomposition to characterize multi-scale motion dynamics. The heterogeneous features are fused, optimized using Grey Wolf Optimization, and classified with a Graph Neural Network to model spatiotemporal relationships. An edge–cloud IoT architecture supports structured acquisition, transmission, visualization, and clinician-assisted monitoring. The framework achieved classification accuracy of 94.1% on KiMoRe, 95.9% on UT-Kinect, and 95.1% on NTU RGB-D 120. These results indicate that integrating depth-based geometric, structural, temporal, and spectral representations can provide accurate appearance-agnostic motion analysis for scalable remote rehabilitation and is compatible with emerging 6G-oriented intelligent healthcare architectures.
No takes yet. Share an insight, caveat, or question.
Khan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: