A Computational Pipeline for Advanced Analysis of 4D Flow MRI in the Left Atrium

TL;DR

This study tackles the challenge of analyzing left atrial hemodynamics with 4D Flow MRI by introducing the first fully open-source, Python-based pipeline tailored to the LA. The framework denoises and upsamples multicenter data, automatically segments the LA with nnU-Net, and computes a comprehensive set of hemodynamic indices including velocity spectrograms, energy metrics, vorticity with Q-criterion, and relative pressure via vWERP, while enabling rich visualizations. The authors demonstrate robust segmentation across centers (Dice ≈ 0.92 for 3T; HD95 ≈ 2.6 mm) and provide inaugural metrics of LA flow, energy, and vortex dynamics across controls and several LVDD-related pathologies, highlighting potential prognostic biomarkers. By openly sharing the code and emphasizing multicenter applicability, this work lays groundwork for standardized LA 4D Flow MRI analysis and broader clinical adoption.

Abstract

The left atrium (LA) plays a pivotal role in modulating left ventricular filling, but our comprehension of its hemodynamics is significantly limited by the constraints of conventional ultrasound analysis. 4D flow magnetic resonance imaging (4D Flow MRI) holds promise for enhancing our understanding of atrial hemodynamics. However, the low velocities within the LA and the limited spatial resolution of 4D Flow MRI make analyzing this chamber challenging. Furthermore, the absence of dedicated computational frameworks, combined with diverse acquisition protocols and vendors, complicates gathering large cohorts for studying the prognostic value of hemodynamic parameters provided by 4D Flow MRI. In this study, we introduce the first open-source computational framework tailored for the analysis of 4D Flow MRI in the LA, enabling comprehensive qualitative and quantitative analysis of advanced hemodynamic parameters. Our framework proves robust to data from different centers of varying quality, producing high-accuracy automated segmentations (Dice $>$ 0.9 and Hausdorff 95 $<$ 3 mm), even with limited training data. Additionally, we conducted the first comprehensive assessment of energy, vorticity, and pressure parameters in the LA across a spectrum of disorders to investigate their potential as prognostic biomarkers.

Figures (14)

• Figure 1: Overview of the computational framework for the advanced analysis of 4D flow magnetic resonance imaging of the left atrium. Only data from lower magnetic field strength acquisitions undergoes denoising and upsampling [1]4dflownet. Next, a 3D PC-MRA is computed [2]Bustamante2017, followed by automatic segmentation [3]nnunet. The resulting segmentation mask facilitates the isolation of the structure of interest for subsequent quantitative and qualitative hemodynamic characterization. The entire pipeline relies exclusively on open-source software and code, ensuring accessibility and reproducibility. PC-MRA: Phase-contrast Magnetic Resonance Angiogram; Q-crit$_{500}$: Ratio of Q-criterion $>$ 500 s$^{-2}$.
• Figure 2: At the center, a volumetric rendering of the 4D flow magnetic resonance imaging velocity field in the left heart is shown, encompassing the left atrium, left ventricle, and ascending aorta. Sample volumes analogous to those placed during pulsed-wave Doppler were approximated by spheres with a 6mm diameter, shown in red, displayed alongside the vessel cross-sections, depicted as gray planes. The corresponding flow rates ($ml/s$) and velocity spectrograms ($m/s)$ for the mitral valve and the four pulmonary veins are shown left and right, respectively. LA: Left atrium, LV: Left ventricle, Asc. Aorta: Ascending aorta, MV: Mitral valve, RS: Right superior, LS: Left superior, RI: Right inferior, LI: Left inferior.
• Figure 3: Dice score and Hausdorff 95 distance (mm) for the segmentation experiments of the left atrium. For Experiment 1, the x-axis is the total number of training cases from dataset 3 T (in red), while in Experiment 2, it is the amount of 1.5 T cases (in blue) added on top of the complete 3 T training dataset.
• Figure 4: Side-by-side comparison of conventional transthoracic echocardiography (TTE) pulsed-wave Doppler acquisitions and the 4D flow magnetic resonance imaging derived velocity spectrograms of the mitral valve. 4D Flow MRI returns a single cardiac cycle spectrogram that has been repeated for comparative purposes only. HCM: Hypertrophic cardiomyopathy; SAM: Systolic anterior motion.
• Figure 5: Conventional transthoracic echocardiography (TTE) pulsed-wave Doppler acquisitions and 4D flow magnetic resonance imaging derived velocity spectrograms of the pulmonary veins (PV) from two hypertrophic cardiomyopathy patients. Both patients underwent TTE-based diastolic dysfunction grading Nagueh2016 and were found to be Grade I (impaired relaxation) and Grade II with a pseudonormal mitral flow pattern, shown in Figure \ref{['fig:Spectro_MV']}. Typically, only the right superior PV can be acquired in TTE, and the image can be of poor quality, as seen in the pseudonormal case. On the contrary, 4D Flow MRI allows easy measurement of the four PVs. 4D Flow MRI returns a single cardiac cycle spectrogram that has been repeated for comparative purposes only. RS: Right superior, LS: Left superior, RI: Right inferior, LI: Left inferior.