Proceedings of the NMR USERS MEETING
PROCEEDINGS OF THE 18ᵗʰ NMR USERS MEETING

In recent years, the increased performance of computers promoted Computational Fluid Dynamics (CFD), and CFD applications have diversified from industrial simulations into areas that depend on techniques such as Magnetic Resonance Imaging (MRI). In this case, image processing with an accurate denoising algorithm is essential to combine MRI and CFD. The purpose of this study is to analyze a denoising algorithm for hemodynamic simulations in brain arteries. First, we used simulated brain images from a public database called BrainWeb. We added complex random Gaussian noise with standard deviations varying from 1% to 10% of the mean intensity of image voxels to each image. The efficiency of a modified Non-local Means (NLM) filtering algorithm was evaluated with metrics such as Correlation Coefficient (CoC) and Mean Structure Similarity Index (MSSIM). Then, we combined NLM with Atropos segmentation algorithm, 3D reconstruction, and OpenFOAM on a 3T Time-of-Flight (TOF) MRI from a healthy female with a complete circle of Willis anatomy from the IXI dataset. Using blood flow rate values from a group of healthy women, we simulated four hemodynamic scenarios: normal flow, basilar artery (BA) occlusion, right internal carotid artery (RICA) occlusion, and left internal carotid artery (LICA) occlusion. For the BrainWeb data, we measured a maximum CoC improvement of 37%, 7%, and 39%, respectively, for T1, T2, and PD for the noisier condition. Also, MSSIM increased its values to 13%, 11%, and 21%, respectively, for T1, T2, and PD in the same noise condition.In practice, we observed the known property of NLM using the redundancies in degraded images to preserve the overall structure. For the IXI database data, the denoising improvements for the subject TOF generated a 4% increase in arteries volume and 5% in area, characterizing a more detailed domain for arteries extremities after segmentation.The simulated hemodynamics in occlusion scenarios presented a flow rate increase in posterior communication arteries from the circle of Willis.Besides the circle of Willis working as expected from physiological function, we measured a decrease of 65.75% on distal branches flow rate from RICA and LICA occlusions.In general, the BA occlusion had a smaller flow rate decrease in brain arteries except for posterior cerebral arteries that had a 35% drop compared to 29% from RICA and LICA occlusions. We modified an NLM algorithm for typical MRI noise distribution that replicated the known properties of the filter.The use of NLM in volunteer data improved the detail for the 3D reconstructed domain necessary for CFD simulations. Finally, the simulations recreated the hemodynamics of the circle of Willis anatomy and estimated cerebral artery branches flow rates in occlusion scenarios. Future studies will consider a patient-specific approach with CFD combined with CFD combined with arterial spin labeling and 4D phase-contrast angiography techniques.