Bayesian inference, machine learning in Cardiac Mechanics for real-time clinical decision support

Ge, Yuzhang (2025) Bayesian inference, machine learning in Cardiac Mechanics for real-time clinical decision support. PhD thesis, University of Glasgow.

Abstract

This thesis develops a Bayesian computational framework for parameter estimation and uncertainty quantification in ventricular biomechanics, based on a Gaussian-process (GP) emulator to accelerate finite-element simulations. First, a Kronecker product-structured GP surrogate model is constructed to generate a robust and computationally efficient map from input space to output space. Here, the input space is the combination of material parameters with a low-dimensional, PCA-based geometric representation of the left ventricle. The output space represents the diastolic inflation of the left ventricle, characterized by ventricular volume–time series. This approach exploits the fact that a tensor-product decomposition of the covariance matrix reduces the cost of covariance-matrix inversion by several orders of magnitude. Next, the emulator is embedded within a Markov-chain Monte Carlo sampling scheme conditional on in vivo MRI–derived ventricular volume data, yielding full posterior distributions of the myocardium material parameters and revealing how the number of measurement time points affects parameter identifiability and posterior uncertainty. To further constrain highly nonlinear, large-strain behaviour, ex vivo pressure–volume (Klotz) curve data are incorporated as complementary information, enabling principled fusion of multi-source measurements. Finally, the approach is applied to patients with hypertrophic cardiomyopathy, where individualized ventricular geometries are reconstructed via PCA, and the GP–MCMC pipeline delivers parameter posterior distributions and stiffness–strain curves for the respective subjects. The resulting workflow maintains high fidelity while reducing per-sample evaluation time from minutes to milliseconds, laying the groundwork for real-time clinical decision support.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Due to copyright issues this thesis is not available for viewing.
• Subjects: Q Science > QA Mathematics
• Colleges/Schools: College of Science and Engineering > School of Mathematics and Statistics
• Funder's Name: Engineering and Physical Sciences Research Council (EPSRC)
• Supervisor's Name: Husmeier, Professor Dirk and Gao, Dr. Hao
• Date of Award: 2025
• Depositing User: Theses Team
• Unique ID: glathesis:2025-85588
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 12 Nov 2025 16:44
• Last Modified: 12 Nov 2025 16:44
• Thesis DOI: 10.5525/gla.thesis.85588
• URI: https://theses.gla.ac.uk/id/eprint/85588