AI-Driven Health Monitoring: Integrating Transformer and Convolutional Fusion for Stroke Patient Posture Estimation.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Rehabilitation Science

Background:

• Traditional physical therapy for stroke patients is resource-intensive, subjective, and lacks real-time feedback.
• Effective motor recovery requires consistent and accurate monitoring of rehabilitative exercises.

Purpose of the Study:

• To propose an AI-driven network architecture for precise posture estimation in stroke patient rehabilitation.
• To enhance the real-time performance of rehabilitative motion posture recognition.
• To facilitate independent home-based rehabilitation exercises for stroke survivors.

Main Methods:

• Developed an AI-driven network architecture integrating spatial convolutional layers and an improved transformer module (ConvTrans).
• Incorporated lightweight multi-head self-attention (LMHSA) and inverted residual forward networks (IRFFN) within the ConvTrans block.
• The architecture captures local and global structural information for enhanced representational capabilities.

Main Results:

• The proposed ConvTrans architecture demonstrated strong performance on three human pose estimation (HPE) datasets.
• The integration of LMHSA and IRFFN reduced computational costs and improved processing efficiency.
• The AI system effectively generates posture skeleton content for feedback.

Conclusions:

• The AI-driven posture estimation network offers efficient and accurate feedback for stroke patient rehabilitation.
• This technology can significantly assist in monitoring and treating stroke patients, promoting independent home exercise.
• The ConvTrans architecture presents a promising advancement in AI-assisted physical therapy.