Pulse-Wave-Pattern Classification with a Convolutional Neural Network.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiovascular Diagnostics

Background:

• Pulse-wave morphology diversity complicates pulse-based diagnosis, particularly pulse-wave pattern classification (PWPC).
• Existing PWPC methods lack clear correlations between pulse patterns and specific disease types, limiting clinical applicability.
• Convolutional neural networks (CNNs) show promise for PWPC due to superior feature extraction capabilities.

Purpose of the Study:

• To develop an optimized CNN model for accurate pulse-wave pattern classification (PWPC).
• To establish a clear one-to-one correspondence between pulse patterns and cardiovascular disease (CVD) categories.
• To enhance the clinical practicability of pulse-based diagnostic methods.

Main Methods:

• Constructed two datasets for PWPC: Dataset 1 used five cardiovascular diseases (CVD) and complications; Dataset 2 used four related physiological parameters.
• Proposed an optimized CNN model with enhanced feature extraction for pulse signals.
• Evaluated the CNN model's accuracy on both datasets for PWPC.

Main Results:

• The optimized CNN model achieved 95% accuracy in PWPC using CVD and complication categories (Dataset 1).
• The model attained 89% accuracy in PWPC using physiological parameters (Dataset 2).
• Results indicate pulse waves result from multiple physiological parameters, highlighting limitations of single-parameter characterization.

Conclusions:

• The proposed CNN model demonstrates high accuracy for pulse-wave pattern classification (PWPC).
• Using CVD and complication categories as classification criteria is effective for CNN-based PWPC.
• The study underscores the complexity of pulse wave formation and the potential of AI in cardiovascular diagnostics.