Improved spatial-temporal graph convolutional networks for upper limb rehabilitation assessment based on precise...

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Artificial Intelligence in Healthcare

Background:

• Stroke often leads to limb movement disorders, necessitating effective rehabilitation.
• Quantitative rehabilitation assessment is crucial but currently lacks clinical application.
• Current methods for assessing upper limb motor function post-stroke are limited.

Purpose of the Study:

• To develop a precise and quantitative method for upper limb motor function assessment in stroke patients.
• To improve the accuracy and stability of rehabilitation assessment using advanced AI models.
• To bridge the gap between experimental quantitative assessment and clinical practice.

Main Methods:

• Utilized two Azure Kinect sensors for an expanded visual field in precise posture measurement.
• Developed a multi-degree-of-freedom rigid body model of the upper limb with optimized inverse kinematics.
• Implemented improved spatial-temporal graph convolutional networks with self-attention and Long Short-Term Memory (LSTM).

Main Results:

• The proposed posture measurement method enhanced accuracy, stability, and Signal Noise Ratio (SNR), reducing occlusion-induced position jumps.
• The developed rehabilitation assessment model demonstrated superior performance with the lowest mean absolute deviation, root mean square error, and mean absolute percentage error.
• The system effectively evaluated upper limb motor function, including distal reachable workspace and proximal self-care ability.

Conclusions:

• The novel approach offers a robust and accurate quantitative assessment for upper limb motor function in stroke rehabilitation.
• This method has the potential to be integrated into clinical practice, optimizing training programs and patient outcomes.
• The integration of precise posture measurement and advanced AI models represents a significant advancement in stroke rehabilitation assessment.