Three-dimensional left atrial blood flow characteristics in patients with atrial fibrillation assessed by 4D flow CMR

Abstract

Aims

To apply 4D flow cardiac magnetic resonance (CMR) for the volumetric measurement of 3D left atrial (LA) blood flow to evaluate its potential to detect altered LA flow in patients with atrial fibrillation (AF) and to investigate associations of changes in systolic and diastolic LA flow with the current clinical risk score (CHA₂DS₂-VASc) used for the assessment of thromboembolic risk in AF.

Methods and results

4D flow CMR was performed in 40 patients with a history of AF (in sinus rhythm during CMR scan, age = 61 ± 11 years), 20 age-appropriate controls (59 ± 7 years), and 10 young healthy volunteers (24 ± 2 years) to measure in vivo time-resolved 3D LA blood flow. LA velocities were characterized with respect to atrial function and timing by calculating normalized LA flow velocity histograms during ventricular systole, early diastole, mid-late diastole, and the entire cardiac cycle. Mean, median, and peak LA velocity steadily decreased when comparing young volunteers, age-appropriate controls, and AF patients by 10–44% and 8–26% for early diastole and the entire cardiac cycle, respectively (P < 0.01 for all comparisons except median velocity for young vs. older volunteers and peak velocity for older volunteers and AF patients). There were moderate but significant inverse relationships between increased CHA₂DS₂-VASc score and reduced mean LA velocity (early diastole: r = −0.37, P < 0.001; entire RR-interval: r = −0.33, P = 0.005), median LA velocity (r = −0.33, P = 0.003; r = −0.25, P = 0.017), and peak velocity (r = −0.36, P = 0.001; r = −0.45, P < 0.001). LA flow indices also correlated significantly with age and LA volume (R² = 0.44–0.62, P < 0.001), but not with left ventricular ejection fraction.

Conclusion

Left atrial 4D flow CMR demonstrated significantly reduced LA blood flow velocities in patients with AF. Further study is needed to determine whether these measures can improve upon the CHA₂DS₂-VASc score for stroke risk prediction and enhance individual decisions on anticoagulation in patients with AF.

Introduction

Atrial fibrillation (AF) is the most common cardiac arrhythmia, affecting ∼33.5 million patients worldwide.¹ Among adults 40 years or older, the lifetime risk of developing AF is ∼25%.² A frequent and serious complication from AF is stroke (15–20% of all strokes occur in patients with AF), which is attributed to embolism of thrombus from the left atrium (LA).^3,4 Currently, clinicians use risk models to estimate an AF patient's annual stroke risk. The most widely used algorithms in patients with AF have been the CHADS₂ score⁵ and more recently the CHA₂DS₂-VASc score which is now recommended.^6–9 However, these scores have limited predictive values for thromboembolism (C statistics 0.55–0.67), as they are based on upstream clinical factors (age, gender, diabetes, etc.) rather than individual physiologic factors, such as stasis, implicated in LA thrombus formation.^6,10

Individualization of the factors associated with atrial thrombus formation might thus help improve these population-based risk scores. Virchow's triad, consisting of endothelial/endocardial damage or dysfunction, stasis or reduced flow, and hypercoagulability, describes factors conducive to thrombogenesis. Although there is evidence supporting the presence of all three factors in AF, reduced left atrial appendage (LAA) flow velocity is thought to play a primary role as demonstrated by previous Doppler transoesophageal echocardiography (TEE) studies.^11,12 In these studies, however, the assessment of flow was based on 2D analysis planes and single direction velocity measurement in the LAA which does not provide complete evaluation of the complex 3D blood flow in the LA.

The aim of this study was therefore to apply 4D flow CMR for the comprehensive quantification of time-resolved, three-directional blood flow velocities with full volumetric coverage of the LA.^13–19 We hypothesized that the LA blood flow velocity distribution measured by 4D flow CMR is significantly altered in patients with a history of AF compared with controls. Furthermore, we hypothesized that metrics of LA haemodynamics, in particular, reduced mean and peak LA flow velocities, correlate with the CHA₂DS₂-VASc risk score.

Methods

Study population

A total of 70 subjects underwent 4D flow CMR for quantification of LA 3D blood flow velocities: 40 patients with a history of AF who were in sinus rhythm at the time of the CMR scan, 20 age-appropriate volunteers, and 10 healthy volunteers under age 30. Patients with documented history of atrial fibrillation—either paroxysmal or persistent—were eligible for inclusion. Patients with implantable devices, significant renal impairment, and either intolerance or contraindication to CMR were excluded. None of the volunteers had a history of cardiovascular disease. All subjects for this HIPAA compliant study were included in the study according to procedures approved by the Northwestern University Institutional Review Board

Magnetic resonance imaging

All patients in this study were in sinus rhythm during the CMR scan. CMR examinations were performed on both 1.5 T and 3 T MR systems (Espree, Aera, Avanto, and Skyra, Siemens, Erlangen, Germany). All patients underwent standard-of-care CMR including the acquisition of four-chamber, two-chamber, and short–axis, ECG-gated time-resolved 2D cine MR images using a steady-state free precession (SSFP) sequence. For the assessment of LA blood flow, time-resolved, 3D phase-contrast CMR with three-directional velocity encoding (4D flow CMR) was used to measure 3D blood flow velocities in the LA. The principle advantage of 4D flow CMR is that during a single, free-breathing acquisition, blood velocity can be measured in three orthogonal directions with full volumetric coverage of the LA throughout the cardiac cycle.²⁰ The data acquisition was synchronized with cardiac motion (prospective ECG gating) and subject's respiration by navigator gating of the diaphragm motion as described previously.²¹ Further 4D flow CMR pulse sequence parameters were as follows: flip angle = 15°, spatial resolution = 2.5–3.0 × 2.5–3.0 × 3.0–4.0 mm, temporal resolution = 37.6–41.6 ms, total acquisition time = 10–20 min depending on heart rate and navigator efficiency, velocity sensitivity = 100–150 cm/s.

Data analysis

A schematic of the 4D flow analysis workflow is shown in Figure 1. After noise filtering, Maxwell, and eddy current correction, 3D PC-MR angiography (3D-PC MRA), or time-averaged magnitude (tMag) data were derived from the 4D flow data.²¹ 3D-PC MRA or tMag data (depending on quality of atrial lumen contrast) were used to guide manual definition of the LA endocardial surface using dedicated 3D segmentation software (MIMIC's Materialise 16.0, USA). For each subject, the resulting 3D LA segmentation mask was used to isolate the velocity data inside the segmented LA volume for all atrial voxels. Next, velocity magnitudes for each subject were arranged in histograms for cardiac time frames corresponding to ventricular systole (4D flow data within the first 300 ms after the R-wave), early diastole, mid-to-late diastole (first and second 50% of the remainder of the cardiac cycle), as well as the entire cardiac cycle. All histograms were normalized by the total number of voxels in the atrium to prevent overweighting data from patients with large atria and to allow for the calculation of group-averaged LA velocity histograms and comparisons across subjects.

Figure 1.

Atrial 4D flow CMR for two AF patients (both imaged while in sinus rhythm) with similar LA volume and low CHA₂DS₂-VASc scores but different LA flow velocities. (A) Left atrial 3D blood flow visualization and a representative LA tomogram colour coded according to the velocity vector magnitude at a single time frame in diastole (B) 3D LA segmentation based on volumetric 3D PC-MRA data (grey-shaded iso-surface). (C) Normalized velocity histograms quantify the LA velocity distribution inside the segmented LA geometry. Quantitative indices of LA flow—mean, median, and peak LA velocity—are shown on the histograms. Note the reduced flow velocities in Subject no. 35 compared with Subject #no. 51 despite similar clinical factors.

In addition, for each subject, we calculated LA volume (in mL), mean velocity (in m/s), median velocity (in m/s), and peak velocity (in m/s) for all analysed time periods (systole, early diastole, mid-to-late diastole, entire cardiac cycle). A typical patient LA velocity histogram covering the entire RR interval contained >50 000 velocities (3D volume + time over the cardiac cycle). Peak velocity was calculated as the average of the top 5% of all LA velocities. To evaluate inter-observer variability of 4D flow-based LA flow analysis, a second independent observer, blinded to the other's results, independently analysed a subgroup of 17 subjects.

2D cine SSFP data were used to evaluate global cardiac function and left ventricular ejection fraction (LVEF).

CHA₂DS₂-VASc Risk Score

The current guidelines for management of patients with AF recommend the use of the CHA₂DS₂-VASc Risk Score for assessment of stroke risk.^6,7 Patient medical records documented prior to CMR acquisition were examined to document clinical risk factors for stroke or thromboembolism. Using the definitions described by Lip et al.,⁶ patients were given one point for congestive heart failure/left ventricular dysfunction, hypertension, aged 65–74, diabetes, vascular disease, sex category female, and two points for age ≥75 and stroke/transient ischaemic attack/thromboembolism.

Statistical analysis

All continuous data are presented as mean ± standard deviation. For each group (young volunteers, age-appropriate volunteers, and AF patients), a Shapiro–Wilk test was used to determine whether parameters were normally distributed. To compare parameters among the three groups, one-way ANOVA (Gaussian distribution) or Kruskal–Wallis (non-Gaussian distribution) was used. If these tests determined that a parameter was significantly different between groups (P < 0.05), multiple comparisons for all groups were performed using independent sample t-tests (Gaussian distribution) or Mann–Whitney U tests (non-Gaussian distribution). Bonferroni correction was used to adjust for multiple comparisons, and differences were considered significant for P < 0.0167. All analysis was performed using Matlab (version R2011a, The Mathworks, USA). To identify relationships between LA volume, age, and metrics of LA haemodynamics, linear regression was performed and Pearson's correlation coefficient was calculated; a correlation was considered significant for P < 0.05.

Results

Study population

Characteristics of the study population are described in Tables 1 and 2. The AF patients and age-appropriate controls were similar in age (61.3 ± 11.1 vs. 59.2 ± 7.4, P = 0.45) and gender distribution (68 vs. 65% male, P = 0.70). Overall, LVEF was well preserved in all groups with no difference between young (63.7 ± 3.7%) and age-appropriate volunteers (62.1 ± 4.2%, P = 0.31). Compared with AF patients (58.1 ± 8.1%), there was a trend towards higher LVEF in age-appropriate volunteers (P = 0.063) and a significantly higher LVEF in young volunteers (P = 0.016). Mean LA volume steadily increased when comparing young controls, age-appropriate controls, and AF patients (P < 0.01 for all comparisons). Heart rate was similar among all groups.

Table 1.

Demographics of study cohorts

	n	Age (years)	LVEF (%)	LA volume (mL)	Heart rate (bpm)
Young volunteers	10	24 ± 2	64 ± 4	25 ± 5	63 ± 8
Older volunteers	20	59 ± 7	62 ± 1	37 ± 12	68 ± 11
AF patients in sinus	40	61 ± 11	58 ± 8	61 ± 24	68 ± 15

Table 2.

Statistical analysis of differences between groups

	Repeated measures	Young controls vs. age-appropriate controls	Young controls vs. AF pts. in sinus	Age-appropriate controls vs. AF pts. in sinus
Age	ANOVA P < 0.0001	P < 0.0001a	P < 0.0001a	P = 0.4469a
LVEF	Kruskal–Wallis P = 0.02	P = 0.3068a	P = 0.0161b	P = 0.0625b
LA volume	ANOVA P < 0.0001	P = 0.0049a	P < 0.0001a	P = 0.0001a
Heart rate	Kruskal–Wallis P = 0.40	N/A	N/A	N/A

Significant differences (after Bonferroni correction, P < 0.0167) between individual groups are indicated by bold type. N/A, multiple comparisons between groups were not performed due to non-significant differences in repeated-measures test.

^aTwo-sided t-test.

^bMann–Whitney test.

Additional clinical information is provided for the AF patients (n = 40) in Table 3. Most AF patients previously underwent rhythm control procedures: 13 underwent both cardioversion and AF ablation, 10 only had prior cardioversion, 6 only had prior ablation, and 11 had no prior rhythm control procedures. Of the 19 patients with prior ablation, 12 had prior catheter AF ablation, 3 had prior surgical AF ablation, and 3 had both. The median time interval between rhythm control procedure and CMR was 184 days (349 ± 429 days, range = 12–1704 days). Only two (5%) patients underwent CMR <30 days after a rhythm control procedure. One patient had cardioversion 13 days before CMR (mean LA velocity 0.13 m/s); another had cardioversion 12 days before CMR (mean LA velocity 0.10 m/s). All patients received some thromboembolism prophylaxis: 8 (20%) aspirin only, 12 (30%) warfarin, and 20 (50%) novel oral anticoagulant (NOAC). None of the patients had severe mitral stenosis, and three patients had mild (n = 2) and severe (n = 1) mitral regurgitation.

Table 3.

AF patient clinical information

	Cardioversion	Ablation	Both CV and ABL	None
Prior rhythm control procedures	10 (25.0%)	6 (15.0%)	13 (32.5%)	11 (27.5%)
	Median (d)	Range (d)	Patients with <30-day interval
Time from rhythm control procedure to CMR	184	12–1704	2 (5%)
Risk factors			Antiarrhythmic medications
CHF/LV dysfunction	2 (5%)		Class IA	0 (0%)
Hypertension	21 (52.5%)		Class IC	9 (22.5%)
Aged 65–74	3 (7.5%)		Class II (BB)	22 (55%)
Diabetes	3 (7.5%)		Class III	14 (35%)
Stroke	6 (15%)		Class IV (CCB)	8 (20%)
Vascular disease	8 (20%)		Class V	6 (15%)
Aged 75 or over	3 (7.5%)
Sex category female	11 (27.5%)		Thromboembolism prophylaxis
			Aspirin only	8 (20%)
Hyperlipidaemia	22 (55%)		Warfarin	12 (30%)
eGFR >60	27 (67.5%)		NOAC	20 (50%)
eGFR 45–59	12 (30%)
eGFR 30–44	1 (2.5%)		ACE/ARB	11 (27.5%)
eGFR <30	0 (0%)		Statin	20 (50%)

CV, cardioversion; ABL, ablation; CHF, congestive heart failure; LV, left ventricular; eGFR, estimated glomerular filtration rate; BB, β-blocker; NOAC, novel oral anticoagulant; ACE, ACE inhibitor; ARB, angiotensin receptor blocker; CCB, calcium channel blocker.

Left atrial 4D flow CMR examples

Figure 1 illustrates LA flow visualization (Figure 1A), LA segmentation based on 3D PC-MRA data (Figure 1B), and resultant LA velocity histograms (Figure 1C) for two AF patients with similar LA volumes and low CHA₂DS₂-VASc scores but different LA velocities. LA blood flow visualization (Figure 1A) showed overall higher blood flow velocities for Subject no. 35 (top row) compared with Subject no. 51 (bottom row) as corroborated by the velocity histograms (Figure 1C). Accordingly, quantitative indices of overall LA flow (mean velocity, median velocity, and peak velocity based on velocity data from entire cardiac cycle) were reduced by 38, 25, and 40%, respectively, for Subject no. 51, indicating the potential of 4D flow CMR to detect poor LA flow despite similar LA volume and low CHA₂DS₂-VASc risk score.

LA velocities with respect to atrial function and timing—group comparisons

All 4D flow CMR data were of sufficient quality for LA segmentation and velocity quantification. Group-averaged LA velocity histograms during systole, early diastole, and mid-to-late diastole for control groups and AF patients are shown in Figure 2 and illustrate overall lower LA velocities (more compact histogram centred at lower velocities) in AF patients compared with controls. Resulting LA flow metrics (mean, median, and peak LA velocities) for all time periods and statistical differences between groups are summarized in Table 4. For AF patients, all indices were significantly reduced (P < 0.0167) compared with young and age-appropriate controls during systole and early diastole (except for peak LA velocity compared with age-appropriate controls). Findings were most pronounced for the early diastolic time period. If LA velocity data over the entire cardiac cycle were taken into account, all three measures of LA blood flow detected statistically significant differences (P < 0.0167) between young volunteers, older volunteers, and AF patients except median velocity for young vs. older volunteers and peak velocity for older volunteers vs. AF patients.

Figure 2.

Group-averaged left atrial velocity histograms for n = 10 young healthy volunteers (A), n = 20 older controls (B), and n = 40 AF patients (C) during systole, early diastole, and late diastole. Mean, median, and peak left atrial velocities are marked on each panel.

Table 4.

Descriptive statistics of metrics of LA 3D blood flow for control cohorts and AF patients

	Mean LA vel. (m/s)	Median LA vel. (m/s)	Peak velocity (m/s)
Entire RR interval
Young volunteers	0.18 ± 0.02	0.15 ± 0.01	0.43 ± 0.02
Age-appropriate controls	0.16 ± 0.02*	0.14 ± 0.02	0.37 ± 0.04*
AF patients in sinus	0.13 ± 0.02#,+	0.12 ± 0.02#,+	0.34 ± 0.05#
Ventricular systole
Young volunteers	0.15 ± 0.02	0.13 ± 0.02	0.36 ± 0.04
Age-appropriate controls	0.14 ± 0.03	0.13 ± 0.03	0.34 ± 0.05
AF patients	0.12 ± 0.02#,+	0.11 ± 0.02#,+	0.30 ± 0.05#,+
Early diastole
Young volunteers	0.24 ± 0.03	0.21 ± 0.03	0.47 ± 0.03
Age-appropriate controls	0.16 ± 0.03*	0.15 ± 0.03*	0.36 ± 0.05*
AF patients	0.14 ± 0.03#+	0.12 ± 0.03#+	0.32 ± 0.07#
Mid-late diastole
Young volunteers	0.15 ± 0.02	0.13 ± 0.02	0.36 ± 0.05
Age-appropriate controls	0.17 ± 0.04	0.15 ± 0.04	0.38 ± 0.06
AF patients	0.15 ± 0.04	0.14 ± 0.04	0.34 ± 0.09

*Significant difference age-appropriate controls vs. young volunteers, P < 0.0167.

^#Significant difference AF patients vs. young volunteers, P < 0.0167.

⁺Significant difference AF patients vs. age-appropriate controls, P < 0.0167.

The distribution of individual patient LA flow metrics for time periods with most pronounced changes in LA velocities (early diastole, entire cardiac cycle) is illustrated in Figure 3. All three indices steadily decreased when comparing young volunteers, age-appropriate controls, and AF patients by 10–44% and 8–26% for early diastole and the entire RR interval, respectively. However, for both time periods, individual patient values illustrate a wide distribution within each group.

Figure 3.

Group-wise comparisons of LA mean (A), median (B), and peak velocities (C) representing LA flow dynamics over the entire cardiac cycle (top row) and during early diastole (lower row). The individual box plots illustrate the median (central mark) and the 25th and 75th percentiles (edges), the whiskers extend to the most extreme data points not considered outliers, and outliers are plotted individually as ‘+’.

Correlation with clinical factors

For all 70 subjects, scatter plots illustrating the relationship between metrics of LA flow based on velocity data from entire cardiac cycle and known clinical risk factors for stroke (age,²² LA volume,²³ and LVEF²⁴) are shown in Figure 4. Mean LA velocities correlated negatively with age (R² = 0.31, P < 0.0001) and LA volume (R² = 0.30, P < 0.0001). Similarly, reduced median LA velocities were significantly associated with increased age (R² = 0.19, P < 0.0001) and increased LA volume (R² = 0.28, P = 0.0001). In addition, significant inverse relationships were found for peak LA velocities with age (R² = 0.39, P < 0.0001) and with LA volume (R² = 0.22, P < 0.0001).

Figure 4.

Correlation analysis between metrics of LA flow (A: mean velocity, B: median velocity, C: peak velocity based on velocity data from entire cardiac cycle) and LA volume, subject age, and LVEF for all n = 70 subjects included in the study.

As summarized in Table 5, there were moderate significant (P < 0.05) relationships between increased CHA₂DS₂-VASc score (i.e. higher thromboembolic risk) and reduced mean LA velocity, median LA velocity, and peak LA velocity during early diastole and based on data from the entire cardiac cycle (Figure 5).

Table 5.

Results of correlation analysis comparing metrics of LA flow with the clinical risk score for AF patients (n = 40)

	CHA2DS2-VASc score vs.
	Mean LA vel.		Median LA vel.		Peak velocity
	r	P	r	P	r	P
Entire RR interval	−0.331	0.005	−0.255	0.017	−0.446	<0.001
Ventricular systole	−0.133	NS	−0.136	NS	−0.124	NS
Early diastole	−0.368	<0.001	−0.331	0.003	−0.361	0.001
Mid-late diastole	−0.214	0.038	−0.174	NS	−0.140	NS

NS, not significant.

Figure 5.

Relationships between metrics of LA flow (mean, median, and peak LA velocity based on velocity data from entire cardiac cycle) with stroke risk estimated by CHA₂DS₂-VASc score for n = 40 AF patients.

Inter-observer variability

Bland–Altman plots demonstrate high inter-observer agreement for measurements of mean, median, and peak LA velocity (Figure 6). Observer variability resulted in average inter-observer differences of 5.5% for mean velocity, 5.9% for median velocity, and 6.8% for peak velocity.

Figure 6.

Bland–Altman analysis of inter-observer variability in a subgroup of n = 17 subjects with LA segmentation by two observers, blinded to each other's results.

Discussion

The findings of this study demonstrate the feasibility of 4D flow CMR for the comprehensive quantification of left atrial 3D blood flow. 4D flow CMR detected significant changes in LA haemodynamics and reduced LA velocities in patients with a history of AF compared with young and age-appropriate control groups. Moreover, reduction in LA velocities was monetary but significantly associated with age and LA volume in the entire cohort, and there was a significant relationship with the standard-of-care clinical CHA₂DS₂-VASc risk score in patients with history of AF. These results indicate the potential of the technique for the evaluation of changes in LA blood flow. Specifically, the correlation of LA flow metrics with the CHA₂DS₂-VASc score (a population-based measure for thromboembolic risk) suggests that individual physiologic LA flow measurements, as obtained by 4D flow CMR, may be sensitive markers for stroke risk stratification in patients with AF. However, the observed correlation was moderate indicating the potential added value of 4D flow-derived metrics of LA flow dynamics for the identification of patients with altered LA velocities. In this context, our findings also demonstrated that AF patients can express substantially different LA velocities despite similar CHA₂DS₂-VASc score as illustrated in the examples in Figure 1 and evident by the overlap in metrics of LA flow in Figures 3 and 5. We speculate that a number of factors (including LA volume, age, blood pressure, etc.) will influence LA flow and a direct measurement of AF induced changes LA velocities may be the most optimal method rather than relying on surrogate metrics or risk scores.

Currently, risk models such as the CHA₂DS₂-VASc score used in this study are recommended to help clinicians weigh the benefit of stroke reduction against the risk of bleeding. These models assign points for each of several risk factors for stroke held by the patient (e.g. age, hypertension, heart failure/reduced LVEF, and prior thromboembolism), and the total point score corresponds to the estimated stroke risk.^6–9 Risk models have played a vital role in helping clinicians estimate stroke risk, to weigh the benefit of therapy aimed at stroke risk reduction against the concomitant increased risk of major bleeding. However, a comparison of these risk models, including those from Atrial Fibrillation Investigators (AFI), Stroke Prevention in Atrial Fibrillation Investigators (SPAF), Framingham Heart Study, and the most commonly used CHADS₂ and CHA₂DS₂-VASc Risk Scores, demonstrated suboptimal C statistics ranging from 0.55 to 0.67.⁶

A major limitation of risk scores is that they are based on upstream clinical factors that are associated with stroke on a population basis. Predictive accuracy can potentially be improved by using physiologic data that are specific to the individual patient. In this context, studies utilizing TEE have shown that decreased LAA peak flow velocities are independent risk factors for stroke in AF.^11,25 Furthermore, the presence of spontaneous echo contrast, a marker of reduced blood flow velocity or stasis, has been shown to be an independent predictor for thromboembolism.^12,26

In this study, we used 4D flow CMR for the assessment of LA blood flow velocities, which has several potential advantages compared with TEE including full volumetric coverage of the LA, quantification of the complete time-resolved, three-directional velocity field inside the LA, and no need for semi-invasive oesophageal intubation. It should be noted that previous TEE studies focused only on the assessment of LAA velocities, while our study was based on the quantification of flow within the entire LA that limits the comparability of findings from both modalities.

Using 4D flow CMR, we found significant differences in the LA flow profiles of young volunteers, older volunteers, and patients with a history of AF who were in sinus rhythm at the time of the scan. LA velocities were significantly impaired in patients with a history of AF compared with older volunteers, while young volunteers predictably demonstrated the most robust LA flow patterns. It is notable that there was a wide spread in the LA flow metrics of individual patients in each group. We speculate that based on LA flow patterns, some of the AF patients may have stroke risk similar to controls, while those with impaired LA flow may have higher stroke risk.

Future longitudinal studies are needed to confirm this hypothesis and investigate whether the combination of clinical scores (e.g. CHA₂DS₂-VASc) with patient-specific 4D flow-derived LA flow velocities can provide improved stroke risk stratification. Indeed, we found that LA flow metrics correlated with other known predictors for the development of stroke and the CHA₂DS₂-VASc score. In the population-based longitudinal Framingham Heart Study, age²² and LA size²³ were found to be significant risk factors for stroke. In our study, all three indices of LA flow (mean, median, and peak LA velocity) were associated with LA volume and age. LA velocities did not correlate with LVEF; however, the correlation between impaired LV function and stroke risk in AF is not seen as consistently across trials. In analysis of the Euro Heart Survey for AF⁸ (used to derive the CHA₂DS₂-VASc score) and in the SPORTIF III/V trials,²⁷ heart failure and LVEF <40% were not significant univariate predictors of thromboembolic events.

It should be noted that all 4D flow CMR data were acquired over multiple heartbeats, and the resulting images represent a composite of blood flow over the entire acquisition time. As all subjects with a history of AF were in sinus rhythm at the time of CMR scan, the 4D flow data can accurately capture the magnitude and timing of LA velocities over the cardiac cycle. Mitral regurgitation that could affect LA velocities was not analysed on our study cohort. However, mitral regurgitation would result in high atrial velocities (as a result of a regurgitant jet) and may thus result in generally elevated LA velocities compared with subjects without regurgitation. Nevertheless, our data show reduced LA velocities in AF patients during all cardiac phases and thus indicate that, even in the presence of mitral regurgitation, LA velocities are still significantly lower in patient with AF compared with age-appropriate controls. Furthermore, our study evaluated blood flow velocities in the entire LA, while previous TEE studies focused on the quantification of metrics of LAA flow. Further studies are thus warranted to systematically evaluate blood flow velocities in both the LA and LAA and assess the impact of the mitral regurgitation on 4D flow data.

Our study has important limitations. The small size of our patient cohort (n = 40) underlines the feasibility nature of our study and limits the conclusions regarding the diagnostic value of metrics of LA flow dynamics that can be drawn from our findings. Specifically, the heterogeneity of our study cohort (inclusion of patients with mitral regurgitation or shortly after cardioversion and thus stunned atrium and mitral valve) may have additionally influenced LA flow and velocities. It should be noted, however, that 4D flow data were acquired with velocity sensitivities (venc) ranging from 100 to 150 cm/s. As a result, blood flow velocities exceeding venc and the normal range of atrial velocities (e.g. high regurgitant jet velocities in patients with mitral regurgitation) will undergo velocity aliasing and are thus mapped back into the range of ±venc. The venc setting in 4D flow CMR therefore acts as a low-pass filter for high blood flow velocities. We expect that even the presence of severe mitral regurgitation and high regurgitant jet velocities will thus only moderately impact LA velocity quantification. Nevertheless, future studies should include larger cohorts and control for cardiovascular abnormalities that may additionally affect LA flow dynamics.

We did not have concurrent TEE data in these patients, precluding assessment of the correlation between peak LAA emptying velocity by TEE and 4D flow CMR metrics of LA flow. Analysis of velocity data is based upon the composite velocity histograms for each patient. More sophisticated analyses including residence time, vorticity, and shear force may improve identification of patients with poor LA flow characteristics that increase risk of thrombus formation. Four-dimensional flow data were acquired at both 1.5 T and 3 T MRI systems which may have resulted in different image quality (signal-to-noise levels). Future studies should include a systematic comparison of image quality and flow metrics between field strengths. We have previously investigated inter-study and inter-observer reproducibility for flow and velocity measurements using 4D flow CMR in the aorta and liver vasculature.^28,29 Findings from these studies have demonstrated good scan–rescan repeatability. However, the evaluation of inter-study repeatability for LA flow analysis was beyond the scope of this study and should be investigated in future studies. Finally, the assessment of atrial delayed enhancement was not part of our study protocol but would be an important addition for future studies.

Conclusions

Four-Dimensional flow CMR quantification of LA flow characteristics demonstrated impaired LA flow in patients with a history of atrial fibrillation compared with controls. Reduced LA mean, median, and peak LA velocities correlated with clinical predictors of stroke, including age, LA volume, and CHA₂DS₂-VASc score. The findings demonstrate the sensitivity of the technique to detect LA flow abnormalities associated with AF. Further studies are warranted to investigate whether the combination of clinical scores (e.g. CHA₂DS₂-VASc) with patient-specific 4D flow-derived LA flow velocities can provide improved stroke risk stratification.

Conflict of interest: None declared.

Funding

This work was supported by the American Heart Association (12GRNT12080032) and the National Institutes of Health (1R21HL113895-01A1).