Upper and Lower-Limb Motor Decoding for Adaptive and Generalized Neural Rehabilitation

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Electromyographic (EMG) signal decoding is crucial for neurorehabilitation and assistive devices.
• Current methods face limitations in limb specificity, generalization, and data requirements.

Purpose of the Study:

• To develop a unified, cross-limb framework for motor intention decoding.
• To overcome limitations of existing EMG decoding approaches, enhancing generalization and reducing data dependency.

Main Methods:

• A principled feature selection identifies domain-invariant EMG representations.
• Reptile-based meta-learning with self-supervised pseudo-labeling enables few-shot adaptation.
• Bias-aware sampling stabilizes adaptation in low-data scenarios.

Main Results:

• Achieved state-of-the-art decoding accuracy across upper and lower limbs.
• Demonstrated superior performance in inter-session, inter-subject, and inter-dataset conditions.
• Framework shows adaptability without repeated recalibration.

Conclusions:

• The proposed framework offers a scalable and clinically viable solution for motor intention decoding.
• It significantly advances the potential of neurorehabilitation and assistive technologies.
• The approach reduces the need for extensive labeled data, facilitating real-world application.