An intelligent aortic valve model for complete cardiac cycle.

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Computational Fluid Dynamics

Background:

• The aortic valve (AV) is critical for cardiovascular (CV) hemodynamics, but its nonlinear behavior complicates accurate prediction of cardiac output (CO) and left ventricular volumetric flow rate (LVQ).
• Existing methods struggle to precisely model AV function, especially under conditions like aortic stenosis (AS), limiting diagnostic and prognostic capabilities.

Purpose of the Study:

• To develop and validate a novel, nonlinear model for the aortic valve (AV) within the cardiovascular system.
• To improve the accuracy of estimating crucial hemodynamic parameters like left ventricular volumetric flow rate (LVQ) and cardiac output (CO) across various AV conditions, including AS.
• To integrate advanced computational techniques, including machine learning, for enhanced real-time CV signal analysis.

Main Methods:

• A new AV model was developed, incorporating time-varying AV resistance (TV-AVR) parameterized by pressure ratio and LVQ, simulating healthy and AS states.
• In vitro validation was performed using a hybrid mock circulatory loop, comparing model predictions with experimental measurements.
• A convolutional neural network (CNN) was unconventionally employed for model correction, transforming 1-D CV signals into 2-D tensors without requiring labeled datasets.

Main Results:

• The novel AV model demonstrated significantly improved accuracy in LVQ and CO estimation, with error rates below 3.41 ± 4.84% for healthy conditions and 2.83 ± 1.35% for AS.
• This represents a 33.13% enhancement in accuracy compared to traditional linear diode models.
• The CNN-based correction effectively improved model performance, highlighting the utility of real-time learning and signal transformation.

Conclusions:

• The proposed nonlinear AV modeling approach offers a substantial advancement in estimating cardiovascular hemodynamic parameters, particularly LVQ and CO.
• This method shows significant potential for enhancing the noninvasive diagnosis, prediction, and treatment strategies for AV diseases like AS.
• The integration of EKF and CNN without labeled data provides a flexible and powerful framework for future cardiovascular research and clinical applications.