Date on Master's Thesis/Doctoral Dissertation

8-2024

Document Type

Master's Thesis

Degree Name

M.S.

Department

Electrical and Computer Engineering

Degree Program

Electrical Engineering, MS

Committee Chair

Amini, Amini

Committee Co-Chair (if applicable)

Inanc, Tamer

Committee Member

Inanc, Tamer

Committee Member

Elmaghraby, Adel

Author's Keywords

Physics informed neural networks; medical imaging; 4D flow MRI; CFD; deep learning

Abstract

In recent years, the use of 4D flow MRI has revolutionized cardiovascular imag- ing by providing comprehensive data on blood flow dynamics over time. However, the limited spatial and temporal resolution of this imaging modality can hinder the accurate assessment of complex hemodynamic phenomena. This thesis explores the application of Physics-Informed Neural Networks (PINNs) to enhance the resolution of 4D flow MRI data, thereby improving its clinical utility. PINNs are a class of neural networks that integrate physical laws into their training process. By embedding these physics equations, PINNs can discover the underlying physics of fluid dynamics to produce more accurate and realistic superresolved flow fields. For this application, we leverage the Navier Stokes momentum equations for three-dimensional unsteady flow. This approach not only enhances the spatial and temporal resolution of the 4D flow MRI data but also ensures the generated high- resolution data adheres to the fundamental principles governing blood flow. The methodology involves training PINNs using low-resolution Computational Fluid Dynamics (CFD) data along with the Navier-Stokes equations as regularizers of the network. The dataset used for training the PINNs is provided by the MIL Lab, and it consists of CFD data representing pulsatile flow at the peak systolic time points of flow. The CFD simulations are however performed at high spatial and temporal resolutions. The CFD data were subsampled to MRI resolutions. This and the original datasets allow for a comprehensive exploration of various hemodynamic conditions. The potential applicability of this method is vast. Enhanced 4D flow MRI data can provide more detailed and accurate assessments of blood flow patterns, aiding in the diagnosis and treatment of various cardiovascular conditions. By improving the resolution, clinicians can gain better insights into the flow dynamics, which would lead to improved patient outcomes. Furthermore, the transformation of this technology to real patient data involves several steps. Firstly, extensive validation is required using synthetic and phantom data to ensure the robustness and accurracy of PINNs. Synthetic data is the main focus of this thesis. Following this, the method must be tested on in-vitro and clinical datasets. Finally, a clinical validation study would be necessary to demonstrate the efficacy of superresolved data in a clinical setting. These steps will be performed as part of future research.

Download

Share

COinS