TPC-GCN: Deep learning for pulse pattern classification in traditional Chinese medicine

Area of Science:

• Computational Medicine
• Traditional Chinese Medicine (TCM)
• Artificial Intelligence in Healthcare

Background:

• Pulse diagnosis is a cornerstone of TCM diagnostics, relying on intricate pulse characteristics.
• Accurate classification of pulse patterns is crucial for the objectification and standardization of TCM.
• Existing methods face challenges in handling the complexity and variability of pulse data.

Purpose of the Study:

• To develop an enhanced methodology for accurate and objective TCM pulse pattern classification.
• To improve the robustness and accuracy of diagnostic models using advanced machine learning techniques.
• To provide new computational approaches for pulse signal analysis in TCM.

Main Methods:

• Utilized an enhanced SMOTE (Synthetic Minority Over-sampling Technique) approach for data augmentation.
• Implemented multi-domain feature extraction and constructed graph data structures for in-depth analysis.
• Devised a multi-channel lightweight graph convolutional network (GCN) with attention-weighted fusion.

Main Results:

• The proposed GCN model achieved high performance metrics on the pulse dataset.
• Achieved 91.68% accuracy, with mean F1 score, recall, and precision all at 92%.
• Demonstrated significant improvements in pulse classification accuracy and model robustness.

Conclusions:

• The enhanced SMOTE and GCN approach effectively improves TCM pulse classification accuracy.
• The study validates the efficacy of the proposed methodology for objective TCM diagnostics.
• Offers novel computational perspectives and methodologies for pulse signal classification research.