Improving Diagnostic Accuracy of 4D Flow MRI with SOSVD Filtering
(Center for Precision Medicine Platform Based on Smart Hemo-Dynamic Index)
The Multiscale Heat & Fluid Flow Lab (MFL) utilizes 4D flow MRI to investigate complex flow phenomena in biological and engineering systems. While 4D flow MRI is a powerful modality for visualizing cardiovascular flows with complex geometries, its applications are often limited by measurement noise. To address this challenge, MFL has developed a novel denoising technique called the Split-and-Overlap Singular Value Decomposition (SOSVD) filter. Unlike conventional singular value decomposition methods, the SOSVD filter splits the velocity matrix into overlapping subdomains and performs SVD individually within each subdomain. By retaining only the dominant flow features—specifically the first singular mode—in each subdomain, this method effectively suppresses noise while maintaining physical fidelity. The SOSVD technique has been validated using both numerically simulated and experimentally acquired flow fields in a square duct, showing significant reductions in root mean square noise levels. Furthermore, its application to in vivo aortic flow data demonstrated substantial improvements in representative flow indices, including divergence, velocity gradients, streamline coherence, and overall flow consistency.
Human
Healthcare
Research Publication
1.
Analyzing Hanyang University Medical Center 4D Flow MRI
(Center for Precision Medicine Platform Base on Smart Hemo-Dynamic Index)
For more precise hemodynamic research, it is essential to perform analyses based on real patient-specific blood flow data. The Multiscale Heat and Fluid Flow Lab (MFL) conducts such analyses using 4D Flow MRI, a technique that enables time-resolved measurement of blood flow within the human body using a commercial MRI system.
In collaboration with Hanyang University Medical Center, we are acquiring carotid and vertebral artery 4D Flow MRI data from patients with cerebrovascular diseases. To ensure higher accuracy, post-processing techniques developed in our lab—such as denoising filters and correction for partial volume effects caused by limited spatial resolution—are applied to the acquired velocity fields.
Furthermore, we aim to establish a framework for patient-specific hemodynamic analysis by comparing computational simulation results using boundary conditions based on real patient data with those based on literature-reported flow profiles.
This research is being carried out within the Center for Precision Medicine Platform Based on Smart Hemo-Dynamic Index, with support from the National Research Foundation of Korea.
Human
Healthcare
Research Publication
1.
2.
Research Presentation Video
1.
S. Kang, D. An, H. Ha, D. Yang, I. Jang and S. Song, 2024, "Denoising four-dimensional flow magnetic resonance imaging data using a split-and-overlap approach via singular value decomposition", Physics of Fluids, 36, 011906
S. Ko, S. Lee, H. Ha, D. Yang, J. Lee, I. Jang, S. Song, 2025, "Refining wall shear stress measurements in four- dimensional flow magnetic resonance imaging through intra-voxel wall boundary consideration", Physics of Fluids, 37, 041907
S. Lee, Y. Lee, K. Choi, S. Song, "4D Flow MRI post-processing for CFD boundary condition", BESCO, 2024, June. 28, Daejeon, Korea
Integral parameters for hemodynamic analysis using 4D flow MRI
(Center for Precision Medicine Platform Based on Smart Hemo-Dynamic Index)
The Multiscale Heat & Fluid Flow Lab (MFL) is developing an integral parameter-based method to address the limitations of 4D Flow MRI caused by low spatial resolution. Conventional parameters such as wall shear stress (WSS) and oscillatory shear index (OSI) rely on spatial differentiation and are highly sensitive to noise, whereas the proposed integral parameters—time-averaged velocity magnitude (TA|V|) and oscillatory velocity index (OVI)—offer improved robustness and reliability. This method was validated using CFD simulations, in vitro 4D Flow MRI, and synthetic datasets. Even at reduced resolution, integral parameters maintained stronger agreement with CFD ground truth compared to traditional differential parameters. Empirical formulas were also established to estimate differential indices from integral values, enabling accurate prediction when direct calculation is unreliable. The approach has demonstrated strong potential for clinical application, offering a more stable and quantitative framework for hemodynamic analysis using 4D Flow MRI.
Human
Healthcare
Estimation Aorta Pressure Distribution Using Physics-Informed Neural Network
(Center for Precision Medicine Platform Based on Smart Hemo-Dynamic Index)
Pressure and wall shear stress constitute the dominant hemodynamic loads on the vascular endothelium but mapping the three‑dimensional intravascular pressure distribution remains difficult: direct measurement is invasive, and high‑fidelity CFD reconstructions, though reliable, demand hours to days of computational time. Physics‑informed neural networks (PINNs) offer a compelling alternative by embedding the Navier–Stokes equations into a deep‑learning framework, enabling near‑real‑time pressure predictions that are robust to noisy or sparse data, free from meshing overhead, and able to accommodate complex or patient‑specific boundary conditions. Leveraging these advantages, the Multiscale Heat & Fluid Flow Laboratory (MFL) is developing a PINN‑based model that ingests velocity measurements and instantly delivers spatially resolved pressure fields for patient‑specific vascular geometries, accelerating both research and clinical decision‑making
Artificial
Intelligence
Human
Healthcare
OpenFOAM Boundary Condition Generation Using 4D Flow MRI Data
(Center for Precision Medicine Platform Based on Smart Hemo-Dynamic Index)
To perform patient-specific CFD for a particular artery, it is extremely important to determine the patient-specific vascular geometry and boundary conditions. While most current studies use precise vascular geometries obtained from CT or MRA, they typically assume a parabolic velocity profile based on flowrate measured via ultrasound(or other methods) for the boundary conditions. However, for more accurate patient-specific simulations, the ideal approach is to apply the patient’s actual 3D velocity profile inside the patient’s blood vessel as the boundary condition. The Multiscale Heat & Fluid Flow Laboratory (MFL) has addressed this need by developing a post‑processing tool that converts 4D‑Flow MRI data into OpenFOAM‑ready boundary files. The tool extracts the measured 3D velocity field at the inlet (or any chosen cross‑section) and formats it as a spatially resolved velocity‑profile boundary condition, enabling CFD simulations to run with the patient’s actual inflow/outflow distributions rather than an idealized parabola.
Artificial
Intelligence
Human
Healthcare
MRV challenge 1: Benchmarking MRV for Turbulent Flow through a U-bend
(MRV Challenge)
The Multiscale Heat & Fluid Flow Lab (MFL) participated in the 2019 international MRV Challenge to evaluate the reliability and reproducibility of Magnetic Resonance Velocimetry (MRV) for complex flow measurements. The challenge involved multiple institutions acquiring 4D flow MRI data of turbulent water flow through a standardized U-bend geometry under consistent flow conditions, each using their own MRI systems and processing pipelines. MFL acquired high-resolution velocity fields using a 3.0T Philips scanner, employing 6-point phase-contrast encoding and divergence-free smoothing tailored for wall-bounded flows. The results accurately captured both primary and secondary flow structures. Despite differences in equipment and reconstruction methods, all participants reported highly consistent results, demonstrating the robustness of MRV in resolving intricate turbulent flow features. This collaborative study was conducted jointly with Stanford University and the U.S. Military Academy (USMA), the University of Rostock (Germany), and Mayo Clinic (USA), laying the groundwork for standardizing MRV as a reliable diagnostic tool in experimental fluid mechanics.
Energy &
Environment
Human
Healthcare
Research Publication
1.
Research Presentation Video
1.
M. J. Benson, A. J. Banko, C. J. Elkins, D. An, S. Song, M. Bruschewski, S. Grundmann, D. D. Borup and J.K. Eaton, 2020, "The 2019 MRV challenge: turbulent flow through a U-bend". Experiments in Fluids, 61, 148
D. An, S. Song, C. Im and S. Oh, "MRV Flow Visualization of U-bend Using 3T and 7T MRI and Comparison of Results", KSVI, 2019, December. 6, Daegu,Korea
MRV challenge 2: Phase-Locked MRV for Pulsatile Turbulent Flow
(MRV Challenge)
The Multiscale Heat & Fluid Flow Lab (MFL) participated in the second international MRV Challenge to explore pulsatile turbulent flow through a structured roughness array using phase-locked magnetic resonance velocimetry (MRV). Five global research teams performed synchronized 4D flow MRI measurements of a jet interacting with cube-shaped elements in a square channel, enabling direct cross-laboratory comparison under a shared experimental setup. MFL utilized a 3.0T Philips MRI scanner with 6-point phase-contrast encoding and a gating system that divided the pulsatile cycle into 20 time-resolved phases. Customized post-processing—including wall-bounded divergence-free filtering and distortion correction—ensured physical accuracy of the measured velocity fields. Despite varying hardware and protocols, all groups showed strong agreement in jet dynamics, wake behavior, and recirculation zones. This collaboration included Stanford University–USMA, University of Rostock (Germany), University of Illinois Urbana-Champaign (USA), and Seoul National University (Korea), highlighting MRV’s potential as a robust tool for benchmarking unsteady turbulent flows.
Energy &
Environment
Human
Healthcare
Research Publication
1.
Research Presentation Video
1.
2.
M. J. Benson, A. J. Banko, C. J. Elkins, D. An, S. Song, M. Bruschewski, S. Grundmann, T. Bandopadhyay, L. V. Roca, B. Sutton, K. Han, W. Hwang and J. K. Eaton, 2023, "MRV challenge 2: phase locked turbulent measurements in a roughness array", Experiments in Fluids, 64, 28
D.G. An, S. Song, "Magnetic Resonance Velocimetry for a Pulsatile Cross-Jet Flow Measurements in Building Structures (2021 MRV Challenge)", KSME Fluid Engineering Division Spring Conference, 2021, Aug. 18, (Online)
D.G. An, S. Song, "2021 MRV Challenge: Hanyang University Results", 74th APSDFD, 2021, November. 22, Arizona, USA
MRV challenge 3: Velocity and Passive Scalar Comparisons in a Highly 3D Turbulent Flow
(MRV challenge)
The Multiscale Heat & Fluid Flow Lab (MFL) actively participates in global collaborative efforts to advance magnetic resonance velocimetry (MRV) for complex engineering flows, including its involvement in the MRV Challenge, an international initiative that benchmarks 3D flow measurements across research teams using shared experimental setups. To expand the scope of MRI-based diagnostics in engineering flows, related methods such as magnetic resonance concentration (MRC), particle (MRP), and temperature (MRT) measurements have been developed. Despite its advantages, the adoption of MRV and its variants has lagged behind laser diagnostics due to perceived complexity. To address this, the MRV Challenge was launched in 2019, enabling global research teams to benchmark measurements across shared flow cases. The most recent MRV Challenge integrates MRV with MRC or MRT and aims to promote best practices, quantify uncertainties, and advance MRI-based fluid diagnostics in engineering applications.
Energy &
Environment
Human
Healthcare
Research Presentation Video
1.
