Uncovering Turbulent Dynamics in Stenotic Flows from 4D-flow MRI Measurements via Resolvent Analysis and Data Assimilation

A. Villié, H. Dillinger, S. Schmitter, S. Demange, K. Oberleithner · 2026

Read Uncovering Turbulent Dynamics in Stenotic Flows from 4D-flow MRI Measurements via Resolvent Analysis and Data Assimilation by A. Villié, H. Dillinger, S. Schmitter, S. Demange, K. Oberleithner (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-03838.txt.

Introduction

Villié and co-authors recover turbulent dynamics in narrowed-artery flow from 4D-flow MRI. The framework joins lab measurements with data assimilation and linear modeling. An idealized stenosis phantom carries a cosine-shaped narrowing. Three-dimensional mean velocity is measured at Reynolds number 3960 by MRI. Displacement artifacts mar the raw data, as usual for this method.

A physics-informed neural network corrects the artifacts in a first pass. It then extracts the unknown mean pressure and eddy viscosity fields. The cleaned mean flow feeds global stability and resolvent analyses. Stationary modes sit in the recirculation bubble with positive growth at two wavenumbers. Resolvent analysis finds a broadband pseudo-resonance from the separated shear layer. The work shows sparse MRI data plus physics models can expose mean fields and coherent structures. It is a first step toward linear analysis of cardiovascular flows.

Catalog record

CREATORA. Villié, H. Dillinger, S. Schmitter, S. Demange, K. Oberleithner
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.03838

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Uncovering Turbulent Dynamics in Stenotic Flows from 4D-flow MRI Measurements via Resolvent Analysis and Data Assimilation A. Villié, H. Dillinger, S. Schmitter, S. Demange, K. Oberleithner Abstract ABSTRACT This study presents a hybrid experimental and computational framework that couples in vitro 4D phase-contrast magnetic resonance imaging (4D-flow MRI) measurements with data assimilation and linear modeling to characterize the flow linear amplification mechanisms. We manufacture an idealized stenosis phantom with a cosine-shaped contraction and acquire three-dimensional (3D) mean velocity measurements at Reynolds number 3960 using 4D-flow MRI. To overcome the inherent displacement artifact, we perform data assimilation via a two-step optimization strategy using physics-informed neural network (PINN). This approach first corrects measurement artifacts before extracting the unknown mean pressure and eddy viscosity fields. The RANS-compatible mean flow then serves as the base state for global linear stability analysis (LSA) and resolvent analysis. The global LSA reveals stationary eigenmodes located in the recirculation bubble that exhibit a positive growth rate for azimuthal wavenumbers m=2 and m=3 . The forced dynamics of this eigenmode dominates the low-frequency dynamics. Resolvent analysis identifies a broadband pseudo-resonance associated with the convective instability of the separated shear-layer, with maximal amplification for m=0 . This methodology demonstrates how integrating sparse experimental MRI data with physics-based modeling enables the identification of mean fields and coherent structures. By leveraging the capabilities of 4D-flow MRI to non-invasively measure 3D velocity fields without requiring physical or optical access, this approach is a first step in the application of linear analysis to cardiovascular flows. INTRODUCTION The dynamics of blood flow past arterial stenoses centrally influence the development and progression of cardiovascular diseases. Atherosclerosis induces a localized narrowing, or stenosis, which can cause non-recoverable pressure loss and flow choking. The downstream disturbed flow regions exhibit low-magnitude and oscillatory shear stresses, which correlate with arterial wall thickening and endothelial dysfunction ( Ku 1997 ) . Experimental studies show that vortex shedding from the constriction throat drives the onset of unsteadiness and transition to turbulence Vétel et al. 2008 . Sherwin et al. 2005 carried out a global stability analysis and found a stationary eigenmode that becomes unstable above the critical Reynolds number Re_{c}=722 , that they interpret as a Coanda-type instability. Blackburn et al. 2008 later performed transient growth analyses of laminar base flows to characterized the onset of unsteadiness in stenotic flows. They have demonstrated that the primary route to transition is driven by convective instability at even smaller Reynolds numbers, identifying sinuous perturbations in the separated shear-layer as the most energetic structures. However, these stability studies did not investigate dynamics beyond a Reynolds number of 1000 . Given that peak systolic Reynolds numbers in the aorta can reach 4000 ( Ku 1997 ) , characterizing the persistent flow structures driven by continuous nonlinear forcing under turbulent conditions remains an unexplored yet essential step for deepening our understanding of atherosclerosis. 4D-flow Magnetic Resonance Imaging (4D-flow MRI) is a non-invasive measurement method that can help characterize complex hemodynamics in stenotic flows ( Markl et al. 2012 ) . It allows to resolve 3D velocity fields in complex geometries, but suffers from inherent resolution constraints that does not allow to capture the turbulent scales. Furthermore, the displacement artifact, or misregistration error, constitutes primary source of error ( Busch et al. 2013 ) . Because an encoding delay separates the acquisition of each velocity component, the moving fluid shifts the measured velocities spatially, corrupting the accuracy of the velocity field ( Dillinger et al. 2020 ) . For steady laminar flows, a correction method is available ( Thunberg et al. 2002 ) which, however, is not applicable to unsteady or turbulent flows. Physics-Informed Neural Networks (PINNs) provide a flexible physics-augmented framework for data assimilation ( Raissi et al. 2019 ) . It approximates data with a continuous function, the neural network, that additionally approximates a solution to a given set of PDEs. The training includes physical information into the network by optimizing composite loss functions which, in addition to the typical data loss term, contain an additional physical loss term composed of the PDEs residuals. Because the PINN evaluates flow quantities continuously across the spatial domain, they naturally handle sparse and low-resolution experimental data. PINN has demonstrated significant potential for enhancing fluid flow measurements through data assimilation, successfully assimilating turbulent mean flow data ( von Saldern et al. 2022 ) and specifically improving stenotic mean flow measurements ( Arzani et al. 2021 ) . In Villié et al. 2025 , a data assimilation framework using PINN was developed to correct the displacement artifact and to assimilate a high-quality mean field from noisy 4D-flow MRI data. Correcting the MRI measurement artifacts leads to a high-quality 3D mean field that can serve as a base state for linear modeling. It enables to extract the dominant coherent structures without requiring highly resolved time-series data, and potentially grants access to amplification mechanisms and driving instabilities. Global linear stability analysis (LSA) linearizes the Navier-Stokes equations around the mean flow to characterize the intrinsic modal instabilities of the unforced system ( Schmid et al. 2001 ) . However, in many separated flows, such as those found in severe stenoses, transition to turbulence arises from convective instabilities and non-modal transient growth rather than unstable global modes ( Blackburn et al. 2008 ; Griffith et al. 2008 ) . To capture these mechanisms, resolvent analysis (RA) formulates an input-output framework that models the nonlinear fluctuation terms as an external harmonic forcing acting on the mean flow ( McKeon et al. 2010 ) . Despite its success in various canonical shear flows, resolvent analysis has not yet been applied to 4D-flow MRI data, nor has it been used in the broader context of cardiovascular flows. This study leverages 4D-flow MRI data to investigate the stability properties and amplification mechanisms in a stenotic flow at a Reynolds number of 3960 with a 75\% area reduction. The paper first presents the experimental setup, then introduces the two-step data assimilation framework that corrects the artifact and assimilates the mean flow. Finally, the results of mean flow stability analysis are discussed. 4D-FLOW MRI MEASUREMENTS Experimental Setup For experimental data, an axisymmetric cosine-shaped stenosis phantom with an areal reduction of 75% (severe stenosis) was manufactured by 3D stereolithography printing. The fluid used was water with molecular viscosity \nu_{m}~=~1.0~\times~10^{-6}~ m 2 /s. A constant flow rate of \dot{m}~=~94\times 10^{-6} m 3 /s was set at the inlet by a gear pump (CardioFlow5000 MR, Shelly Medical Imaging Technologies). The resulting Reynolds number based on the unobstructed diameter was Re=3960 . In the following, we normalize the results by the obstructed diameter d_{t}=0.015 m and bulk velocity at the stenosis throat U_{t}=0.53 m/s. MRI data is acquired at 3 T (Cima.X, Siemens Healthineers, Erlangen, Germany) using a knee coil and a vendor-provided 4D-flow sequence for velocity measurement. The symmetry stenosis axis is aligned with the MRI longitudinal ( z ) axis, so that each measurement slice is perpendicular to the axial flow component. The data covers a field-of-view of 256 x 144 x 1~\mathrm{mm}^{3} with a resolution of 1 mm isotropic. The acquisition incorporates 24 averages for a total time of 76 min. Furthermore, the velocity encoding strength is set to 75 cm/s and

Read the complete text