Rehabilitation movement simulation via joint angle-based generative AI

Area of Science:

• Biomedical Engineering
• Computer Science
• Rehabilitation Science

Background:

• Generative models excel at synthesizing realistic human motion for animation and virtual reality.
• Their application in clinical settings, particularly for rehabilitation, is underexplored.
• Existing methods often use joint positions, which may not align with clinical assessment methods.

Purpose of the Study:

• To introduce a conditional diffusion-based generative framework for rehabilitation-oriented motion synthesis.
• To generate motion in a clinically meaningful joint-angle space, encoding joint range of motion.
• To enable subject-independent modeling and improve interpretability for clinical use.

Main Methods:

• Developed a conditional diffusion-based generative framework operating on joint-angle representations.
• Incorporated joint range of motion explicitly into the generation space.
• Employed a comprehensive evaluation protocol with qualitative and quantitative metrics, including visualizations and adherence assessments.

Main Results:

• The model successfully generated plausible and contextually accurate motion sequences.
• Joint-angle representations demonstrated improved generalization and superior performance over position-based methods.
• Cross-subject and leave-one-combination-out experiments validated the model's effectiveness.

Conclusions:

• The proposed framework is feasible for generating rehabilitation-oriented motion simulations.
• Joint-angle representations enhance motion synthesis for clinical relevance and interpretability.
• Future work should explore personalized rehabilitation scenarios and larger, more diverse datasets.