CDS&E: Scalable Bayesian Reconstruction of Cardiovascular Hemodynamics via Information Field Theory

Purdue University, West Lafayette IN

Investigators

Abstract

Heart disease accounts for one-fifth of adult deaths with congenital heart conditions afflicting 1% of newborns and contributing to 4.2% of neonatal fatalities. How blood flows through the body is vital diagnostic information for heart disease. It can be determined from medical imaging techniques like phase-contrast (PC) magnetic resonance imaging (MRI), 4D Flow MRI, and Color Doppler echocardiography (Echo). These methods directly measure blood flow, but several limitations obscure results for practical use in clinical settings. This project will advance methods for increasing reconstruction accuracy and quantifying uncertainty for advanced medical imaging techniques. Success in both aspects can provide valuable information to clinicians, aiding treatment decisions. Quantifying uncertainty can also help improve imaging protocols for maximum information gain, leading to faster, cheaper, and more reliable scans. The approach promises to significantly enhance medical image analysis to inform clinical decision-making by improving accuracy, resolution, and quantifying uncertainties in hemodynamic fields, ultimately leading to more effective and personalized treatment plans for heart disease patients. This project develops a scalable Bayesian methodology to solve the inverse problem of reconstructing cardiovascular hemodynamic flow fields and cardiac structure from advanced medical imaging modalities (PC-MRI, 4D Flow MRI, and Echo). The research design seeks to overcome existing limitations, such as inaccurately reconstructed velocity flow, which often violates known physical laws, low signal-to-noise ratio, scanner-to-scanner variation, and lack of quantified uncertainty in reconstructions. Central to this effort is the application of information field theory (IFT), which allows for the fusion of governing physical laws with noisy, heterogeneous data. Through four specific aims, the project: 1) Formulates the inverse problem of identifying hemodynamic fields within the IFT framework; 2) Explores parameterizations of time-varying hemodynamic fields; 3) Devises scalable numerical algorithms to characterize the joint posterior distributions of these fields and relevant parameters; and 4) Verifies and validates the methodology using synthetic, in vitro, and in vivo data. The expected outcome is a next-generation statistical methodology that solves the hemodynamics field and physical parameter estimation problem with quantified uncertainty. The success of this project will provide a comprehensive and accurate picture of the complex blood flow structures within the physiology under study, which, in turn, will aid medical diagnosis, especially in the early asymptomatic phase, risk assessment, treatment planning, and pre/post-treatment management. The project will also impact biomedical research. The probabilistic quantification of hemodynamics parameters such as pressure drop, vorticity, turbulent kinetic energy, or viscous energy loss can lead biomedical researchers to identify biomarkers for various diseases. Furthermore, since it will contribute to the core theory of IFT, the project will impact other applications that reconstruct spatiotemporal fields in multi-physics settings. The open-source package the team will develop will make it easier for other domain experts to adopt IFT. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.