Central Arterial Dynamic Evaluation from Peripheral Blood Pressure Waveforms Using CycleGAN: An In Silico Approach

Area of Science:

• Cardiovascular physiology and biomedical engineering.
• Application of artificial intelligence in medical diagnostics.

Background:

• Arterial stiffness is a key indicator of cardiovascular disease risk.
• Current methods for assessing arterial stiffness, like pulse wave velocity (PWV), primarily use peripheral pressure signals.
• Assessing arterial stiffness via the pressure-strain hysteresis loop offers a more comprehensive analysis.

Purpose of the Study:

• To explore the capability of generative adversarial networks (GANs) in transferring peripheral arterial pressure signals to central arterial pressure and area signals.
• To reconstruct and evaluate the pressure-strain hysteresis loop for arterial stiffness assessment using deep learning.
• To compare the performance of different GAN loss functions, specifically Least-Square GAN (LSGAN) and Wasserstein GAN with gradient penalty (WGAN-GP).

Main Methods:

• Utilized a public, validated virtual database of arterial signals.
• Employed deep learning models, specifically GANs (LSGAN and WGAN-GP), for signal transfer and reconstruction.
• Reconstructed the pressure-strain hysteresis loop from peripheral signals.
• Evaluated the reconstructed loop using machine learning metrics and clinical parameters.

Main Results:

• LSGAN demonstrated superior performance compared to WGAN-GP in reconstructing central arterial pressure and area waveforms.
• LSGAN achieved a mean error of 0.8 ± 0.4 mmHg for pressure waveforms and 0.1 ± 0.1 cm² for area waveforms.
• The pressure-strain elastic modulus was estimated with a mean absolute percentage error of 6.5 ± 5.1% using LSGAN.

Conclusions:

• GAN-based deep learning models can effectively recover the pressure-strain loop characteristics of central arteries by analyzing peripheral pressure signals.
• This methodology provides a promising non-invasive approach for assessing arterial stiffness and cardiovascular health.
• The findings suggest LSGAN is a suitable model for this complex signal transfer and reconstruction task in cardiovascular analysis.