Estimation of Central Aortic Pressure Waveforms by Combination of a Meta-Learning Neural Network and a Physics-Driven...

Area of Science:

• Cardiovascular Physiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Accurate non-invasive estimation of central aortic pressure waveforms (CAPW) is critical for cardiovascular disease management.
• Current methods for CAPW estimation lack sufficient accuracy and practicality.
• Personalized physiological indicators are key to improving CAPW estimation.

Purpose of the Study:

• To develop and validate a novel method for accurate and practical non-invasive estimation of CAPW.
• To integrate a meta-learning neural network with a physics-driven approach for enhanced CAPW prediction.
• To utilize personalized physiological indicators for individualized CAPW estimation.

Main Methods:

• A meta-learning neural network (MAML) was trained on data from 260 patients undergoing catheterization.
• Inputs included measured CAPW and personalized indicators (weight, BMI, MAP, HR, CO, SBP, DBP).
• A physics-driven loss function constrained the neural network output (Gaussian parameters of CAPW) for improved accuracy.

Main Results:

• The model achieved high accuracy in estimating CAPW, with a normalized root mean square error (NRMSE) of 0.0206.
• Low biases were observed for systolic blood pressure (SBP), diastolic blood pressure (DBP), and mean arterial pressure (MAP).
• Validation in 52 patients confirmed the model's accuracy and practicality.

Conclusions:

• The combined meta-learning and physics-driven approach accurately estimates CAPW non-invasively.
• This method offers personalized parameters for calculating myocardial ischemia indicators (iFR, FFR).
• The approach holds potential for early monitoring and prevention of cardiovascular diseases.