Graph Based Multichannel Feature Fusion for Wrist Pulse Diagnosis

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Traditional Chinese Medicine

Background:

• Traditional Chinese Medicine recognizes wrist pulse signals for health assessment.
• Computerized AI systems analyze wrist pulse signals but often neglect multichannel feature correlation and fusion.
• Effective fusion of multichannel wrist pulse features can significantly improve diagnostic performance.

Purpose of the Study:

• To propose a novel graph-based multichannel feature fusion (GBMFF) method for wrist pulse signal analysis.
• To effectively utilize and fuse features extracted from multiple wrist pulse channels.
• To enhance the accuracy and performance of AI-based wrist pulse diagnosis.

Main Methods:

• Utilized pressure and photoelectric sensors to capture three-channel wrist pulse signals.
• Applied stacked sparse autoencoder and wavelet scattering for feature extraction.
• Constructed graphs from feature vectors of each sample and employed graph convolutional networks for diagnosis.

Main Results:

• The proposed GBMFF method demonstrated superior performance compared to existing state-of-the-art approaches.
• Effective fusion of multichannel features led to improved wrist pulse diagnosis accuracy.
• The AI-based diagnostic system achieved enhanced performance in health status evaluation.

Conclusions:

• The GBMFF method offers a promising approach for advanced wrist pulse diagnosis by leveraging multichannel feature fusion.
• This AI-driven technique effectively integrates diverse data sources for more accurate health assessments.
• The study highlights the potential of graph-based methods in analyzing complex biomedical signals for improved healthcare outcomes.