Week of 2026-03-20 — 26 new papers

Cardiovascular CFD, hemodynamics, and AI modelling (PINN · neural operators · surrogates · digital twins). Auto-curated from OpenAlex + arXiv, classified with Claude.

At a glance

Tier	Topic	Count
A	Boundary conditions (inlet/outlet, Windkessel)	1
B	Turbulence modelling (RANS, LES, DNS)	0
C	V&V and uncertainty quantification	1
D	Physiology & scaling laws	2
E	Imaging & WSS measurement	11
F	AI, ML & digital-twin pipelines	11

Highlights this week

Physics-informed graph neural networks for real-time prediction of wall shear stress in stenotic coronary arteries — Ting-Ting Luo et al. · Scientific Reports
Accelerated Patient-Specific Hemodynamic Simulations with Hybrid Physics-Based Neural Surrogates — Natalia L. Rubio et al. · arXiv · cs.CE
Unraveling Flow-Dynamic Characterization of the Ascending Aorta and Left Ventricle in Bicuspid Aortic Valve Pathologies via 4D Flow MRI — Shirin Aliabadi · University of Calgary

Tier A — Boundary conditions (inlet/outlet, Windkessel)

Semi-Automated Generation and Hemodynamic Assessment of Surgical Baffle Geometry for Biventricular Repair

Elena Sabdy Martinez et al. · arXiv · cs.CE · 2026 · arXiv:2603.25207

Semi-automated framework generates patient-specific intraventricular baffle geometry for DORV biventricular repair using centerline-based area constraints and produces CFD-ready models for hemodynamic assessment. Retrospective validation on four patients shows predicted pressure gradients align with postoperative echocardiography, demonstrating feasibility for preoperative surgical planning.

Tier C — V&V and uncertainty quantification

Hemodynamic Performance and Blood Damage of the Intra-Aortic Pumps: A CFD-Based Investigation

Osman Aycan et al. · arXiv · physics.med-ph · 2026 · DOI · arXiv:2603.26911

CFD study comparing three intra-aortic pump designs using wall-modeled LES, with grid convergence validation via GCI and introduction of a dimensionless hemolytic number (HN) for standardized hemocompatibility assessment. Impeller-driven pump achieved superior hydraulic efficiency (~6%) and lowest hemolysis (NIH = 0.0035 g/100L), with HN < 1 across physiologic flow rates.

Tier D — Physiology & scaling laws

In silico analysis of the haemodynamic disturbances caused by the subaortic membrane pathology

Alessandra Monteleone & G. Burriesci · Computers in Biology and Medicine · 2026 · DOI

FSI simulation of subaortic membrane (SAM) pathology using smoothed particle hydrodynamics investigates hemodynamic disturbances and aortic valve leaflet fluttering in two SAM morphologies versus healthy controls. Stiffer membranes doubled oscillation amplitude and increased frequency by 8%, with detailed vortex dynamics and shear stress analysis revealing mechanisms of pathological fluttering.

Influence of Plaque Characteristics on Stent Biomechanical Outcomes - A Case Study on Double Kissing Crush Coronary Stenting

Andrea Colombo et al. · arXiv · physics.bio-ph · 2026 · arXiv:2604.07800

A finite-element and CFD study of double-kissing crush stenting in coronary bifurcations, modeling plaque composition (lipid vs. fibrous) and quantifying wall stress, malapposition, and post-deployment hemodynamics (shear stress exposure). Results show plaque stiffness significantly alters lumen restoration, wall mechanics, and relative stent performance.

Tier E — Imaging & WSS measurement

Unraveling Flow-Dynamic Characterization of the Ascending Aorta and Left Ventricle in Bicuspid Aortic Valve Pathologies via 4D Flow MRI

Shirin Aliabadi · University of Calgary · 2026 · DOI

4D flow MRI study of bicuspid aortic valve (BAV) disease characterizing hemodynamics in the ascending aorta and left ventricle across valve phenotypes and severities. Links regional wall shear stress and intracardiac pressure gradients to aortic root dilation and early diastolic dysfunction, proposing flow-derived biomarkers for risk stratification.

Patient-Specific CFD Analysis of Carotid Artery Haemodynamics: Impact of Anatomical Variations on Atherosclerotic Risk

Abhilash Hebbandi Ningappa et al. · Computation · 2026 · DOI

Patient-specific CFD of four carotid arteries using non-Newtonian pulsatile flow and Carreau–Yasuda rheology, quantifying wall shear stress (TAWSS) and oscillatory shear index (OSI) as markers of atherosclerotic risk zones. Results show low-shear regions (<1 Pa) and directional oscillations concentrate near bifurcations and bulbs, with geometry (angle, tortuosity, asymmetry) controlling flow patterns.

Integrating 4D-flow MRI and computational fluid hemodynamics to noninvasively estimate trans-stenotic pressure gradient in idiopathic intracranial hypertension

Yupeng Zhang et al. · Journal of Cerebral Blood Flow & Metabolism · 2026 · DOI

This study combines 4D-flow MRI with patient-specific CFD to noninvasively estimate trans-stenotic pressure gradients in idiopathic intracranial hypertension, validating against invasive manometry in 48 patients. Transient and steady-state simulations showed excellent agreement with manometry (bias <0.4 mmHg), and sensitivity analysis confirmed that omitting small venous branches does not degrade diagnostic accuracy.

Aortic Valve Phenotype Shapes Helical Flow in the Ascending Aorta: Insights from 4D Flow MRI

Karol Calò et al. · Annals of Biomedical Engineering · 2026 · DOI

4D flow MRI study characterizing how aortic valve phenotype (bicuspid vs. tricuspid) modulates helical flow topology in the ascending aorta, showing that BAV eccentric jets reduce helicity intensity. Authors propose a morpho-hemodynamic signature integrating anatomy, conventional hemodynamics, and helical flow metrics for BAV diagnosis and prognosis.

Patient Image-Based Hemodynamics of Intracranial Aneurysms: An In Silico Study

Algirdas Maknickas & Jurinda Merkevičiūtė · Applied Sciences · 2026 · DOI

Patient-specific CFD of intracranial aneurysms extracted from DICOM imaging, using pulsatile inlet velocity and pressure outlet BCs to compute WSS-based rupture indicators (TAWSS, OSI, RRT) and correlate hemodynamics with aspect ratio. Results align with literature and suggest hemodynamic profiling could inform clinical treatment selection.

Wall Shear Stress Hemodynamics and Morphological Data from CFD Simulations for 76 Intracranial Aneurysms from the Aneurisk Repository

Iago Lessa Oliveira · Zenodo (CERN European Organization for Nuclear Research) · 2026 · DOI

A dataset of 76 patient-specific intracranial aneurysm geometries with OpenFOAM-derived hemodynamic fields (WSS, velocity) and morphological metrics, using pulsatile laminar flow with physiologically scaled inlet conditions and zero-pressure outlets. Provides a curated resource for hemodynamic studies and machine-learning training without requiring independent CFD runs.

Wall Shear Stress Hemodynamics and Morphological Data from CFD Simulations for 76 Intracranial Aneurysms from the Aneurisk Repository

Iago Lessa Oliveira · Zenodo (CERN European Organization for Nuclear Research) · 2026 · DOI

A public dataset of 76 patient-specific intracranial aneurysm geometries with pre-computed CFD hemodynamic fields (wall shear stress, morphological metrics) from OpenFOAM simulations using pulsatile, age-adjusted inlet conditions and zero-pressure outlets. Data provided in VTP format for direct use in ParaView and further analysis.

A Patient-Specific CFD Study of Carotid Webs: Hemodynamic Analysis and the Role of Blood Viscosity

Xuning Zhao et al. · medRxiv · 2026 · DOI

Patient-specific CFD of carotid webs reconstructed from CT angiography examines wall shear stress (TAWSS, OSI, RRT) differences between symptomatic, asymptomatic, and normal bifurcations across three non-Newtonian viscosity models. Symptomatic webs show lower TAWSS and higher oscillatory indices, but hemodynamic metrics alone lack discriminatory power for clinical risk stratification.

Case Report: Unveiling segmental hemodynamic heterogeneity in internal jugular vein stenosis: a patient-specific CFD analysis

Hui Li et al. · Frontiers in Cardiovascular Medicine · 2026 · DOI

Patient-specific CFD of internal jugular vein stenosis reconstructed from CT venography reveals segmental hemodynamic heterogeneity and morphological variation. The authors propose CFD-derived metrics as potential clinical decision-support tools, though validation remains limited to case-level evidence.

Reproducibility of 4D Flow MRI-Derived Diastolic Function Testing by Mitral and Pulmonary Venous Flow Indices in Healthy Volunteers

Thomas in de Braekt et al. · Applied Sciences · 2026 · DOI

This study assesses scan–rescan reproducibility of 4D Flow MRI-derived diastolic function indices (mitral and pulmonary venous flows) in 21 healthy volunteers using two respiratory compensation strategies, finding moderate-to-strong agreement for most parameters but poor interobserver reproducibility for deceleration time and pulmonary venous metrics. The work identifies systematic measurement variability that limits clinical translation of 4D Flow MRI for diastolic assessment.

Wall shear stress and oscillatory shear index after carotid endarterectomy in the postoperative period

Gavrilenko Av et al. · Medical alphabet · 2026 · DOI

Clinical study measuring wall shear stress and oscillatory shear index in the internal carotid artery before and after endarterectomy using Doppler ultrasound (V-flow mode) in 26 patients over 6–14 months. Post-operative normalization of WSS, flow velocity, and OSI correlated with surgical success and reduced turbulence.

Tier F — AI, ML & digital-twin pipelines

Physics-informed graph neural networks for real-time prediction of wall shear stress in stenotic coronary arteries

Ting-Ting Luo et al. · Scientific Reports · 2026 · DOI

Physics-informed graph neural networks trained on 1000 synthetic stenotic coronary geometries predict wall shear stress with MAE=1.05 Pa and R=0.94, achieving second-timescale inference versus hours for CFD. The approach uses statistical shape modeling to span morphological variation and validates local accuracy via Bland-Altman analysis across severe and dual-lesion cases.

Accelerated Patient-Specific Hemodynamic Simulations with Hybrid Physics-Based Neural Surrogates

Natalia L. Rubio et al. · arXiv · cs.CE · 2026 · arXiv:2604.01549

Hybrid physics-informed neural network approach learns 0D model parameters (resistance, inductance) from 3D CFD data to accelerate patient-specific hemodynamic prediction while preserving interpretability. Validated on aortic, aortofemoral, and pulmonary anatomies, achieving 50%+ error reduction and sub-2-second runtime on consumer hardware.

Unsupervised 4D Flow MRI Velocity Enhancement and Unwrapping Using Divergence-Free Neural Networks

Javier Bisbal et al. · ArXiv.org · 2026

DAF-FlowNet is an unsupervised neural network that enforces divergence-free constraints by construction (curl of vector potential) to simultaneously denoise and unwrap phase artifacts in 4D Flow MRI. On synthetic CFD-generated aortic data and patient HCM cases, it reduces velocity RMSE by ~11% and wrapped voxels by 18–72% versus existing methods while preserving flow features.

Unsupervised 4D Flow MRI Velocity Enhancement and Unwrapping Using Divergence-Free Neural Networks

Javier Bisbal et al. · arXiv (Cornell University) · 2026 · DOI

DAF-FlowNet is an unsupervised divergence-free neural network that jointly denoises and unwraps phase-wrapped velocities in 4D Flow MRI by parameterizing flow as curl of a vector potential, enforcing mass conservation by construction. On synthetic CFD-generated and patient aortic data, it achieves 11% lower velocity RMSE and 44% lower divergence than alternatives, with robust handling of aliasing and segmentation noise.

Fast and Accurate Inverse Blood Flow Modeling from Minimal Cuff-Pressure Data via PINNs

Sokratis J. Anagnostopoulos et al. · arXiv · physics.comp-ph · 2026 · arXiv:2604.03221

A physics-informed neural network (PINN) solves the inverse problem of reconstructing patient-specific flow and pressure fields in a 1-D arterial tree model from noninvasive cuff pressure alone, achieving 10× speedup over prior methods and validating central hemodynamic metrics (CO, cSBP) against clinical data. The framework learns terminal boundary condition parameters (resistance, compliance) during training, enabling fast personalized hemodynamic assessment.

Real-Time Surrogate Modeling for Personalized Blood Flow Prediction and Hemodynamic Analysis

Sokratis J. Anagnostopoulos et al. · arXiv · cs.LG · 2026 · arXiv:2604.03197

A deep neural surrogate model trained on synthetic 1D arterial hemodynamics enables real-time prediction of pressure and cardiac output from clinical parameters, with built-in physiological filtering to reject non-physiological parameter combinations. The framework leverages multivariate correlations from the Asklepios cohort to generate realistic virtual populations and demonstrates application to central aortic hemodynamics estimation from clinical data.

SLE-FNO: Single-Layer Extensions for Task-Agnostic Continual Learning in Fourier Neural Operators

Mahmoud Elhadidy et al. · arXiv · cs.LG · 2026 · arXiv:2603.20410

Proposes SLE-FNO, a continual learning architecture combining Single-Layer Extensions with Fourier Neural Operators to enable surrogate models to adapt to distribution shifts (geometry, BCs, flow regimes) without catastrophic forgetting. Demonstrates the method on mapping transient concentration fields to time-averaged wall shear stress in pulsatile aneurysmal blood flow across 230 CFD simulations.

Cardiovascular digital twins using a Windkessel physics informed neural network

Deen Osman et al. · npj Digital Medicine · 2026 · DOI

Physics-informed neural networks (PINNs) combined with Windkessel models estimate personalized cardiovascular parameters (compliance, resistance) and predict blood pressure from noninvasive bioimpedance wearables. The framework reduces prediction error by 12–25% versus standard deep learning and achieves 0.77–6.07% accuracy in parameter estimation on synthetic datasets.

Asymptotic-Preserving Neural Networks for Viscoelastic Parameter Identification in Multiscale Blood Flow Modeling

Giulia Bertaglia & Raffaella Fiamma Cabini · arXiv · cs.LG · 2026 · arXiv:2604.06287

The paper uses asymptotic-preserving neural networks (physics-informed ML) to identify viscoelastic arterial wall parameters from 1D multiscale blood flow models, inferring pressure waveforms from Doppler ultrasound measurements of cross-sectional area and velocity. The method embeds governing equations into the learning procedure and is validated on synthetic and patient-specific data.

Toward Generalizable Graph Learning for 3D Engineering AI: Explainable Workflows for CAE Mode Shape Classification and CFD Field Prediction

Tong Duy Son et al. · arXiv · eess.SY · 2026 · arXiv:2604.07781

A graph neural network framework converts heterogeneous 3D engineering data (FE models, CAD, CFD meshes) into physics-aware graph representations for surrogate modeling of aerodynamic fields and vibration modes. The CFD component uses GNNs to predict pressure and WSS fields across body shape variants with symmetry-preserving downsampling.

PINN-ing the Balloon: A Physically Informed Neural Network Modelling the Nonlinear Haemodynamic Response Function in Functional Magnetic Resonance Imaging

Rodrigo H. Avaria-Saldias et al. · bioRxiv (Cold Spring Harbor Laboratory) · 2026 · DOI

A physics-informed neural network embeds the Balloon-Windkessel model into a multi-headed architecture to estimate cerebral hemodynamic state variables (blood inflow, oxygen consumption, volume, deoxyhaemoglobin) from BOLD fMRI signals. The method recovers ground-truth variables with R² > 0.99 in simulation and yields physiologically plausible subject-specific HRF estimates from clinical stroke data.

Methodology, tier definitions and scope caveats: see the Paper Digest landing page.

Found an issue or have a suggestion for this page?

Open a GitHub issue