M. Fathi, A. Bakhshinejad, A. Baghaie, D. Saloner, R. Sacho, Denoising and Spatial Resolution Enhancement of 4D-Flow MRI Using POD and Lasso Regularization M. Fathi, A. Bakhshinejad, A. Baghaie, D. Saloner, R. Sacho, V. Rayz, R. D'souza
Introduction 4D-Flow MRI technique The issues of 4D-Flow MRI Non-invasive imaging of human cardio-vascular system In-vivo volumetric time-resolved three dimensional blood flow velocity measurements 4D means 3 spatial dimensions + time The issues of 4D-Flow MRI Velocity aliasing Phase offsets Random acquisition noise Low spatial and temporal resolution
Input Data 4D-Flow MRI Data Vascular geometry In-vivo Noisy Eddy current effect Vascular geometry Contrast-Enhanced MRA Time-of-Flight MRA
Running CFD Using ANSYS Fluent Creating patient-specific geometry Estimating the boundary conditions Rigid wall assumption No-slip boundary condition at walls About 10 snapshots per 4D-Flow time step
POD Basis Vectors Using SVD Method Find the basis vectors in CFD mesh resolution Downsample the basis vectors into 4D-Flow resolution
LASSO Regularization Using OWL-QN algorithm Finding the sparsest solution in the POD basis β controls the sparseness of c
Reconstruction Reconstructing all snapshots In CFD mesh resolution In 4D-Flow resolution
Summary This method is capable of reconstructing data to any arbitrary high resolution (depending on the resolution of the CFD mesh) and reconstruct many flow details. It can potentially improve the ability to accurately derive relevant secondary hemodynamic parameters such as wall shear stresses, and pressure gradients at much higher resolution than 4D-Flow resolution.