Probabilistic Cardiac Digital Twins for Robust Patient-Specific Modeling.

Area of Science:

• Computational biology
• Biomedical engineering
• Statistical modeling

Background:

• Uncertainty quantification (UQ) is crucial for reliable cardiac digital twins (DTs) in personalized medicine.
• Traditional UQ methods struggle with high-dimensional, nonlinear cardiac models.

Purpose of the Study:

• To introduce a novel framework for learning joint probability densities in cardiac DTs.
• To enable efficient uncertainty propagation and quantification in patient-specific cardiac models.

Main Methods:

• A geometry-aware generative learning framework is employed.
• Identifies low-dimensional latent representations of cardiac data.
• Uses stochastic differential equations in latent space for efficient sampling and mapping to high-dimensional spaces.

Main Results:

• The framework learns joint probability densities over cardiac observables and parameters.
• Enables construction of predictive distributions through sampling and conditioning.
• Demonstrated effectiveness on a ventricular arrhythmia prediction benchmark with fewer model evaluations than conventional methods.

Conclusions:

• The proposed framework offers a powerful approach for UQ in cardiac DTs.
• Facilitates characterization of statistical dependencies and exploration of interdependencies among observables.
• Advances the development of reliable and personalized cardiac digital twins for clinical applications.