Associations between 3D-based Left Atrial Volumetric and Blood Flow Parameters in a Single-Site Cohort of the Multi-Ethnic Study of Atherosclerosis

Abstract

Purpose

To investigate associations between left atrial volume (LAV) and function with impaired three-dimensional hemodynamics from four-dimensional flow MRI.

Materials and Methods

A subcohort of participants from the Multi-Ethnic Study of Atherosclerosis from Northwestern University underwent prospective 1.5-T cardiac MRI including whole-heart four-dimensional flow and short-axis cine imaging between 2019 and 2020. Four-dimensional flow MRI analysis included manual three-dimensional segmentations of the LA and LA appendage (LAA), which were used to quantify LA and LAA peak velocity and blood stasis (% voxels < 0.1 m/sec). Short-axis cine data were used to delineate LA contours on all cardiac time points, and the resulting three-dimensional–based LAVs were extracted for calculation of LA emptying fractions (LAEF_total, LAEF_active, LAEF_passive). Stepwise multivariable linear models were calculated for each flow parameter (LA stasis, LA peak velocity, LAA stasis, LAA peak velocity) to determine associations with LAV and LAEF.

Results

This study included 158 participants (mean age, 73 years ± 7 [SD]; 83 [52.5%] female and 75 [47.4%] male participants). In multivariable models, a 1-unit increase of LAEF_total was associated with decreased LA stasis (β coefficient, −0.47%; P < .001), while increased LAEF_active was associated with increased LA peak velocity (β coefficient, 0.21 cm/sec; P < .001). Furthermore, increased minimum LAV indexed was most associated with impaired LAA flow (higher LAA stasis [β coefficient, 0.65%; P < .001] and lower LAA peak velocity [β coefficient, −0.35 cm/sec; P < .001]).

Conclusion

Higher minimum LAV and reduced LA function were associated with impaired flow characteristics in the LA and LAA. LAV assessment might therefore be a surrogate measure for LA and LAA flow abnormalities.

Keywords: Atherosclerosis, Left Atrial Volume, Left Atrial Blood Flow, 4D Flow MRI

Supplemental material is available for this article.

© RSNA, 2024

Summary

Increased left atrial minimum volume, which causes reduced left atrial function, was associated with impaired flow hemodynamics in the left atrium and left atrial appendage.

Key Points

• ■ In a cohort of participants from the Multi-Ethnic Study of Atherosclerosis, higher indexed minimum left atrial (LA) volume derived from cardiac MRI was associated with impaired flow characteristics in the left atrial appendage (increased stasis [β coefficient, 0.65; P < .001] and decreased peak velocity [β coefficient, −0.35; P < .001]).

• ■ Increased total LA emptying fraction was associated with decreased LA stasis (β coefficient, −0.47; P < .001) while increased active LA emptying fraction was associated with increased LA peak velocity (β coefficient, 0.21; P < .001).

Introduction

Left atrium (LA) enlargement was first associated with stroke and death in the Framingham Heart Study (1). Over the years, associations of LA volume (LAV) and LA function with outcomes after myocardial infarctions, or diseases such as stroke or heart failure, were established and strengthened the importance of LA volumetric assessment (2,3).

More recently, assessment of four-dimensional (4D) flow MRI enabled the quantification of three-dimensional (3D) blood flow dynamics in the LA and LA appendage (LAA) in vivo in patients with atrial fibrillation (AF) who are known to be at increased risk for stroke (4,5). Studies have shown that 4D flow–based quantitative assessment of blood stasis and peak velocity in the LA and LAA may help to characterize impaired flow characteristics and, therefore, risk for thrombus formation (6). However, the associations between conventional volumetric and functional LA parameters with flow characteristics in the LA and LAA have not been systematically investigated in detail, with a lack of non-AF data from community-based cohorts.

In this study, we aimed to investigate the associations between 4D flow parameters and conventional volumetric-based LA assessment in a cohort of the Multi-Ethnic Study of Atherosclerosis (MESA) participants. We hypothesized that potential associations between these techniques could help to better characterize participants with impaired flow characteristics and enhance the understanding of mechanistic relationships between LA flow and volumetric parameters.

Materials and Methods

Study Cohort

MESA is a prospective Health Insurance Portability and Accountability Act–compliant multicenter study to investigate the development and progression of subclinical cardiovascular (CV) disease in participants free from known CV disease at baseline (2000–2002) (7). The current study is an ancillary study on MESA participants from our institution who underwent an additional research cardiac MRI (2019–2020). This ancillary study was approved by the institutional review board, and all participants gave written informed consent to participate in this study. There were no formal exclusion criteria.

Medical information about participants, including demographics and CV risk factors, was collected at the last MESA examination (2016–2018). Of the participants included in this study, 155 have been previously reported (8). This prior study dealt with comparison of different techniques for LAV assessment, whereas in this manuscript, we report on associations between LAV assessment and hemodynamics in the LA and LAA from 4D flow MRI.

Cardiac MRI

Cardiac MRI examinations were performed on a 1.5-T MRI system (Aera; Siemens Healthcare). All participants underwent balanced steady-state free precession time-resolved (cine) imaging in short-axis orientation covering the left ventricle (LV) and entire LA. Scan parameters of the cine series were as follows: echo time msec/repetition time msec, 1.2/35.5; flip angle, 59°; spatial resolution, 1.8 × 1.8 mm²; slice thickness, 6 mm; slice gap, 3 mm; interslice distance, 9 mm; field of view, 340 × 308 mm; matrix size, 192 × 180; and number of cardiac time points, 25.

Furthermore, retrospective electrocardiographically gated cine 3D phase-contrast MRI with 3D velocity encoding (4D flow) MRI acquisitions were acquired in coronal orientation during free breathing. The 4D-flow parameters were as follows: echo time msec/repetition time msec, 2.44/5.16; flip angle, 7°; spatial resolution, 2.5 × 2.5 × 2.5 mm³; field of view, 480 × 400 mm; matrix size, 192 × 120; number of cardiac time points, 22; velocity encoding, 120 cm/sec; and compressed sensing with R (R Foundation for Statistical Computing), 10. Of note, participants did not receive contrast media.

LAV and Function Data Analysis

Analysis was performed by a CV radiologist (M.P.) with 3 years’ experience using a self-programmed software in MATLAB (MathWorks). LA endocardial borders were contoured on each short-axis slice and all cardiac time points (Fig 1A). At the atrioventricular junction, the LA was contoured only if there was less than 50% myocardium visible. The contours from each plane were combined into a cine 3D volume data set for further analysis (Fig 1B).

Figure 1:

Analysis of left atrial (LA) volume based on three-dimensional cine MRI. (A) Three-dimensional cine MR images show LA segmentation on all short-axis slice images (A1–A8) and at all cardiac time points (images represent a single time point). (B) Segmentations were combined into a three-dimensional object for each time point, and the respective volumes were extracted to generate (C) volume-time curves. (D) Based on those curves, total, active, and passive LA emptying fractions (LAEF_total, LAEF_active, and LAEF_passive, respectively) were calculated. LAV = LA volume, LAV_max = maximum LAV, LAV_min = minimum LAV, LAV_preA = LAV before atrial contraction.

The 3D volumes for each time point of the cardiac cycle were calculated and a volume-time curve was created. Hence, maximum LAV (LAV_max) and minimum LAV (LAV_min) were extracted automatically. Furthermore, the LAV at the time point before atrial contraction (LAV_preA) was manually selected (Fig 1C). These three data points were used to calculate the active, passive, and total LA emptying fractions (LAEF_active, LAEF_passive, LAEF_total, respectively) for each participant (Fig 1D). In addition, the LAVs were indexed to the body surface area (LAVI_max, LAVI_min, and LAVI_preA).

4D Flow MRI Data Analysis

Data analysis was based on previously reported strategies for LA flow quantification (6). Briefly, 4D flow MRI data were corrected for Maxwell terms and eddy currents. A 3D phase-contrast MR angiography was calculated from the 4D flow MRI data and used to derive a manual 3D segmentation of the LA and LAA (Mimics; Materialise) (Fig 2A). The LA and LAA 3D segmentations were used to mask the 4D flow data to calculate blood stasis as the relative number of voxels with less than 10 cm/sec per time point normalized by the total number of time points for the LA and the LAA. Peak velocity was calculated as the top 5% of voxels per cardiac time point for both the LA and LAA. Furthermore, parametric maps of blood stasis and peak velocity for the LA and LAA were created (Fig 2B, 2C).

Figure 2:

Analysis of left atrial (LA) and LA appendage (LAA) hemodynamics based on four-dimensional (4D) flow MRI. After preprocessing the 4D flow data, (A) LA and LAA were segmented on phase-contrast MR angiographic images excluding the pulmonary veins. These segmentations were used to calculate parametric maps for (B) stasis and (C) peak velocity in the LA and LAA. AAo = ascending aorta, DAo = descending aorta, LPA = left pulmonary artery, LPVs = left pulmonary veins, RA = right atrium, RPA = right pulmonary artery, RPVs = right pulmonary veins.

Statistical Analysis

Statistical analysis was performed in Python (version 3.8; Python Software Foundation). If normally distributed, numbers were reported as means and SDs. Otherwise, data were reported as medians and IQRs. First, volumetry-based parameters of the LA and LV were compared with the four 4D flow–based parameters (stasis in the LA and LAA and peak velocity in the LA and LAA) by Pearson correlation coefficient. Multiple comparisons correction was performed by applying Bonferroni correction. Furthermore, to determine impact of volumetric parameters on impaired flow (defined as higher stasis and lower peak velocity), multivariable linear regression analyses with automatic stepwise forward and backward selection considering all volumetric parameters were performed for each 4D flow–based parameter. In addition, we added multivariable linear regression analyses controlled for CV disease risk factors (alcohol consumption, diabetes, history of AF, hypertension, smoking) and added box and whisker plots for each flow parameter and CV disease risk factor (see Appendix S1). For the resulting β coefficients, for every 1-unit increase (or decrease) in the predictor variable, the outcome variable will increase (or decrease) by the β value. A P value less than .05 was considered significant for all statistical analyses except for the Pearson correlation coefficients in Table 3 where a value of .00096 (.05/52) was required.

Table 3:

Pearson Correlation Coefficients between Four-dimensional Flow–based Left Atrial and Left Atrial Appendage Hemodynamic Parameters and Three-dimensional–based Volumetric and Functional Parameters

Results

Study Cohort

Of all 240 participants invited, 76 individuals declined participation, mainly because of concerns of being exposed to COVID-19 at our hospital-affiliated imaging center. Of the 164 participants who underwent cardiac MRI, two had to be excluded due to missing LA cine imaging and another four due to insufficient 4D flow imaging quality because of image noise. In total, 158 participants were finally included in this study (mean age, 73 years ± 7 [SD]; 83 [52.5%] female and 75 [47.4%] male participants) (Fig 3). The majority of the study group were White participants (70 of 158 [44.3%]), followed by Chinese (52 of 158 [32.9%]) and Black participants (36 of 158 [22.8%]). The average body mass index of all participants was slightly above the normal range (mean, 26.4 kg/m² ± 4.7). Hypertension was common (57.6%), and some participants had diabetes (14.0%) and history of AF (12.0%). About half of participants reported regular consumption of alcohol (51.2%) or had current or former smoking status (48.0%). Baseline characteristics of study participants are summarized in Table 1.

Figure 3:

Flowchart of study cohort. MESA = Multi-Ethnic Study of Atherosclerosis.

Table 1:

Baseline Characteristics of Study Cohort

Associations between LAV, LA Function, and LA Flow

The average values for all LAV and flow parameters are summarized in Table 2. As shown in Table 3, a moderate correlation between volumetric and flow parameters was found for LA stasis which showed a significant negative correlation with LAEF_total (r = −0.53; P < .00001). This relationship was predominantly caused by association with LAV_min (r = 0.39; P < .00001) rather than LAV_max (r = 0.14; P = .09). A similar association was found for the indexed volume parameter LAVI_min (r = 0.45; P < .00001).

Table 2:

Baseline MRI Volumetric and Flow Parameters for the Left Atrium and Left Ventricle

In addition, increased LA peak velocity was significantly correlated with higher LAEF_total (r = 0.56; P < .00001). This relationship was driven primarily by LAV_min (r = −0.46; P < .00001) rather than LAV_max (r = −0.22; P = .006). As for LA stasis, the indexed volume parameter LAVI_min showed similar associations (r = −0.51; P < .00001).

For the LAA, stasis showed the most pronounced correlation with LAVI_min (r = 0.43; P < .00001), followed by LAEF_total (r = −0.37; P < .00001) and LAV_min (r = 0.36; P < .00001). In addition, increased LAA peak velocity was associated with decreased LAVI_min (r = -0.59; P < .00001), decreased LAV_min (r = −0.56; P < .00001), and increased LAEF_total (r = 0.46; P < .00001).

LV parameters showed only weak correlation with LA stasis (LV_end-diastolic volume indexed and LV_stroke volume indexed) and LA peak velocity (LV_stroke volume indexed) and no significant correlations with any LAA parameters (Table 3).

Multivariable Linear Regression Analysis

The models for LA stasis and peak velocity showed that LAEF_total was the only independent parameter associated with LA stasis (β coefficient, −0.47; P < .001), and LAEF_active (β coefficient, 0.21; P < .001) was the only independent factor associated with LA peak velocity.

For the LAA, the model for LAA stasis revealed that both LAVI_min (β coefficient, 0.65; P < .001) and LAV_max (β coefficient, −0.18; P = .002) were independently associated with higher stasis. Similarly, LAVI_min (β coefficient, −0.35; P < .001) and LAV_max (β coefficient, 0.07; P = .005) were independently associated with higher LAA peak velocity (Table 4).

Table 4:

Results from the Four Multivariable Models with Automatic Stepwise Forward and Backward Selection

In multivariable regression analysis controlled for CV disease risk factors, similar results were found for LA hemodynamics. LAEF_total was associated with LA stasis (β coefficient, 0.47; P < .001), while LAEF_active was associated with LA peak velocity (β coefficient, 0.21; P < .001).

For the LAA, LAA stasis was associated with LAVI_min (β coefficient, 0.46; P < .001) and LV_stroke volume indexed (β coefficient, −0.28; P = .001). LAVI_min (β coefficient, −0.96; P < .001) as well as LAVI_preA (β coefficient, 0.55; P < .001) and LAEF_active (β coefficient, −0.20; P = .007) were all associated with LAA peak velocity (Table S1).

Discussion

In this study, we investigated the associations of volumetric 3D-based cine parameters with flow parameters derived from 4D flow MRI. In univariable analyses, we found moderate associations between volumetric and functional parameters with flow, indicating that increased LAV (maximum and especially minimum) and reduced LA function correlated with impaired flow (defined as higher stasis and lower peak velocity). Stepwise multivariable linear regression models revealed that reduced LA function was independently associated with impaired flow in the LA (LA stasis [LAEF_total: β coefficient, −0.47; P < .001]; LA peak velocity [LAEF_active: β coefficient, 0.21; P < .001]), while increased minimum LAV was associated with impaired LAA flow (LAA stasis [LAVI_min: β coefficient, 0.46; P < .001]; LAA peak velocity [LAVI_min: β coefficient, −0.35, P < .001]).

Impaired blood flow is thought to be a risk factor of thrombus formation in the LA and therefore is potentially associated with stroke (9–11). In more than 90% of patients who experienced a stroke, a thrombus in the LAA is found (12,13). Since its initiation more than 20 years ago, MESA has contributed to the understanding of the importance of LAV and LA functional assessments in diseases like AF, supraventricular ectopy, diabetes, or other cardiovascular risk factors (14–17). To our knowledge, this study is the first to quantify LA and LAA flow in a MESA subcohort using 4D flow MRI. This is of relevance since most studies on atrial 4D flow were performed in AF cohorts rather than general population cohorts (4,6,9,18,19). We found that blood stasis and peak velocity were mainly associated with changes of minimum LAV (indexed or absolute) in both the LA and LAA, which was revealed in multivariable models. Furthermore, in regard to impaired LA flow, we found associations of stasis and peak velocity with reduced LAEF_total and LAEF_active, respectively.

Lee at al (18) investigated atrial blood flow characteristics in patients with AF and reported positive correlations with a 3D-based LAV. However, they used a static time-averaged LAV from 4D flow–based 3D MR angiography which might represent LAV closer to LAV_max than LAV_min (20). Demirkiran et at (19) reported LA and LAA stasis and peak velocity in a small cohort of 15 participants (10 with AF in sinus rhythm and five age-matched controls) but also using short-axis–based time-resolved LAV assessment. They observed lower stasis for both LA and LAA, which might be explained by a lower mean age of their participants. Interestingly, they observed lower peak velocities (between 20 and 30 cm/sec for both groups), which is even below the expected range of normal LAA peak velocities (21). Another study investigated differences between LA and LAA flow parameters in 60 patients with AF (and 15 healthy controls) (4). Similar to us, they found lower stasis in the LA than LAA. However, we observed higher peak velocities in the LAA than the LA. The latter might be explained by the weak correlation of LA and LAA velocities, which was reported previously (22). Of note, while we found weak correlations of LV parameters with LA flow in univariable analyses, no LV parameter was associated with impaired flow in the LA or LAA in the multivariable models. This was also reported in other studies (18,19). In the multivariable models corrected for CV disease risk factors, results for LA stasis and peak velocity did not change significantly and the association between LA functional parameters and impaired hemodynamics remained. For LAA hemodynamics, the association between impaired flow characteristics with LAVI_min remained while we additionally found associations with LV parameters (LV stroke volume and LAA stasis), as well as LAVI_preA and LAEF_active (each with LAA peak velocity). In contrast, LAV_max did not remain associated with LAA hemodynamics.

The importance of LAV_min assessment has increased over the recent years. There have been few studies describing the association between LAV_min (rather than LAV_max) and reduced LA function in AF populations (23,24). Furthermore, increased LAV_min and reduced LA function has been shown to be associated with incident cerebrovascular events, supraventricular ectopy, heart failure and diastolic dysfunction in community-based cohorts (3,16,25–27). Our findings link increased LAV_min and reduced LA function to impaired flow characteristics which might be a contributing factor to LA or LAA thrombus formation. Furthermore, an increase in LAV_min, as the denominator in the equations to calculate the two functional parameters, is bound to a decrease in both LAEF_total and LAEF_active. Therefore, our study points toward the importance of LAV_min alterations on changes in hemodynamics in both LA and LAA. The relationship between structural (LAV) and functional (hemodynamics) parameters is thereby potentially influenced by atriopathy (also known as atrial myopathy), the latter of which is thought to be associated with LA enlargement (11).

While we found significant associations between LAV metrics and flow characteristics, it is important to consider that our results do not suggest that LAV-based analyses can replace flow assessment. Further studies are necessary to investigate whether LAV metrics could be used to identify patients who are at risk for impaired flow and would benefit from acquisition of advanced 4D flow imaging. In such a scenario, echocardiography might be useful to identify patients with or at risk for impaired LA and/or LAA flow to trigger advanced cardiac MRI. This gatekeeping approach might be useful to achieve a translation into clinics in the foreseeable future.

For the assessment of LAV, we used delineations of LA contours from a stack of short-axis cine images which were combined into a time-resolved 3D data set. While this analysis was more time-consuming, requiring contouring of LA boundaries on multiple slices and time points, we decided to use this method over the clinically used biplane assessment for LAV. This was due to recent studies reporting a lack of accuracy by the biplane method to assess LAV. The biplane method was primarily developed and evaluated for the assessment of LAV_max, and there is growing evidence on potential underestimation of LAV_min and overestimation of LA functional parameters (8,28,29). While 3D-based LAV assessment remains tedious, recent developments in automated segmentation methods using deep learning networks might enhance usage of 3D-based assessment as an imaging marker in the future when 4D flow MRI–based flow assessment is not available or impractical due to scan time constraints (30).

This study had several limitations. First, it was a single-center study including a small subcohort of the MESA study population, and MRI data were acquired on one scanner from a single vendor. While cine imaging is well standardized, the impact of different scanners might influence the 4D flow results. The MESA cohort represents only older individuals and only White, Black, and Chinese participants were represented in our study cohort. The results of our study might not be generalizable on younger participants or patient cohorts. This should be investigated in further studies. Also, as most of our participants had a normal LV function, additional studies should investigate patients with reduced LV function to better understand potential relationships between LV function and LA flow hemodynamics. Regarding the analyses of LA hemodynamics, we used a custom software in MATLAB. While this technique was established and has been implemented by other research groups, there is currently no commercial software solution available, which might limit translatability. Moreover, atriopathy could be a contributing factor to LAV, which we did not investigate in this study. Regarding statistical tests used in our study, the sample size of 158 participants is borderline for multiple linear regression analyses with 13 potential predictors. Last, we did not perform LA strain analysis, which is another parameter of LA function.

In conclusion, our study demonstrates associations between volumetric assessment with 4D flow MRI–based blood flow dynamics in the LA and LAA in a subcohort of MESA participants. Increased absolute and indexed minimum LAV was most closely associated with impaired LA flow, which is a known risk factor for LA thrombus formation and potentially ischemic stroke. Thus, 3D-based LAV quantification may be a useful surrogate measure for LA and LAA flow abnormalities in future studies.