Adaptive Semi-Supervised Intent Inferral to Control a Powered Hand Orthosis for Stroke

Area of Science:

• Rehabilitation Engineering
• Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

• Wearable robotic orthoses require robust and intuitive controls for effective therapy.
• Existing EMG-based control methods for robotic hand orthoses can be burdensome due to training requirements and susceptibility to concept drift.

Purpose of the Study:

• To explore semi-supervised learning for controlling powered hand orthoses in stroke subjects.
• To introduce a novel disagreement-based semi-supervision algorithm to manage intrasession concept drift.
• To reduce the training burden on users by leveraging unlabeled data.

Main Methods:

• Development and application of a disagreement-based semi-supervision algorithm.
• Utilizing multimodal ipsilateral sensing for control.
• Evaluation on data from five stroke subjects, including functional task validation.

Main Results:

• The proposed algorithm effectively adapts to intrasession concept drift using unlabeled data.
• A significant reduction in user training burden was observed.
• Feasibility was validated through successful completion of pick-and-handover tasks by two subjects.

Conclusions:

• Semi-supervised learning is a viable and effective paradigm for controlling powered hand orthoses.
• The developed algorithm enhances adaptability and reduces user burden in robotic-assisted therapy.
• This approach shows promise for improving functional rehabilitation outcomes for stroke survivors.