Predicting left ventricular contractile function via Gaussian process emulation in aortic-banded rats.

Area of Science:

• Computational biology
• Cardiac electrophysiology
• Cardiovascular modeling

Background:

• Cardiac contraction involves complex multi-scale interactions.
• Biophysical cardiac models are computationally expensive, hindering parameter fitting and global sensitivity analysis (GSA).

Purpose of the Study:

• To develop a machine learning framework for efficient cardiac model parameter inference and GSA.
• To apply this framework to healthy and failing rat heart models.

Main Methods:

• Utilized Gaussian process emulation to create probabilistic surrogate models.
• Employed Bayesian history matching (HM) for model parameter inference.
• Performed GSA on whole-organ cardiac mechanics.

Main Results:

• Surrogate models accurately predicted left ventricular pump function (R² = 0.92 for ejection fraction).
• Bayesian HM successfully fitted virtual rat heart models to experimental data.
• GSA identified Troponin C and cross-bridge kinetics as critical parameters for ventricular function.

Conclusions:

• The proposed machine learning approach enhances the efficiency of cardiac modeling.
• This method facilitates accurate prediction and parameter identification in cardiac mechanics.
• The findings contribute to a better understanding of heart failure mechanisms.