Assessment of the Precision and Reproducibility of Ventricular Volume, Function and Mass Measurements with Ferumoxytol-Enhanced 4D Flow MRI

Abstract

Purpose

To compare the precision and inter-observer agreement of ventricular volume, function and mass quantification by three-dimensional time-resolved (4D) flow MRI relative to cine steady state free precession (SSFP).

Materials and Methods

With research board approval, informed consent, and HIPAA compliance, 22 consecutive patients with congenital heart disease (CHD) (10 males, 6.4±4.8 years) referred for 3T ferumoxytol-enhanced cardiac MRI were prospectively recruited. Complete ventricular coverage with standard 2D short-axis cine SSFP and whole chest coverage with axial 4D flow were obtained. Two blinded radiologists independently segmented images for left ventricular (LV) and right ventricular (RV) myocardium at end systole (ES) and end diastole (ED). Statistical analysis included linear regression, ANOVA, Bland-Altman (BA) analysis, and intra-class correlation (ICC).

Results

Significant positive correlations were found between 4D flow and SSFP for ventricular volumes (r = 0.808–0.972, p<0.001), ejection fraction (EF) (r = 0.900–928, p<0.001), and mass (r = 0.884–0.934, p<0.001). BA relative limits of agreement for both ventricles were between −52% to 34% for volumes, −29% to 27% for EF, and −41% to 48% for mass, with wider limits of agreement for the RV compared to the LV. There was no significant difference between techniques with respect to mean square difference of ED-ES mass for either LV (F=2.05, p=0.159) or RV (F=0.625, p=0.434). Inter-observer agreement was moderate to good with both 4D flow (ICC 0.523–0.993) and SSFP (ICC 0.619–0.982), with overlapping confidence intervals.

Conclusion

Quantification of ventricular volume, function and mass can be accomplished with 4D flow MRI with precision and inter-observer agreement comparable to that of cine SSFP.

INTRODUCTION

The quantification of ventricular volume, function and mass is an essential goal of many cardiac MRI examinations, typically achieved with multiple short-axis bright-blood cine steady-state free precession (SSFP) acquisitions. The accuracy of ventricular volume, ejection fraction (EF), and mass measurements by cardiac MRI has been validated (1–5), and cine SSFP is now widely used in practice as a clinical reference standard (6, 7). However, cardiac MRI exams often require advanced operator skills to identify appropriate scan planes and to optimize acquisition parameters, particularly in the setting of congenital heart disease (CHD). Thus, these exams are often lengthy and frequently require prolonged deep anesthesia or sedation in the pediatric population (8).

Three-dimensional time-resolved (4D) flow MRI is an evolving imaging technique that has the potential to simultaneously acquire both flow and function information in a single acquisition, even without detailed operator knowledge of cardiac anatomy (9, 10). The use of contrast agents in 4D flow MRI has been shown to improve signal-to-noise ratio in magnitude data and noise reduction in velocity data (11). However, until the recent implementation of under-sampling methods, including parallel imaging, the clinical utility of 4D flow MRI had been largely limited by prohibitively long image acquisition times (12). To date, although flow and function assessment have been described with 4D flow imaging (9, 13–15), assessment of ventricular mass quantification has not yet been reported to our knowledge. Left ventricular (LV) mass quantification is commonly performed in the setting of congenitally corrected transposition of the great arteries after pulmonary arterial banding to assess feasibility of an arterial switch operation (16). Similarly, right ventricular (RV) mass quantification may have prognostic value in the setting of repaired tetralogy of Fallot (TOF) and other CHD (17–19).

Our clinical CHD practice has been greatly simplified by the implementation of 4D flow imaging. We have anecdotally found that cardiovascular MRI exams can be rapidly performed with a single ferumoxytol-enhanced 4D flow sequence, thus reducing the frequency, depth, and/or duration of anesthesia. However, when indicated, we have had to perform additional sequences for the evaluation of ventricular mass. Here, we aim to compare the precision and inter-observer agreement of ventricular volume, function and mass quantification by 4D flow MRI relative to the gold standard cardiac-gated cine SSFP. We hypothesized that the precision and inter-observer agreement of LV and RV volume, function and mass measurements would not be different between 4D flow and cine SSFP acquisitions.

MATERIALS AND METHODS

Patient Population

With institutional review board approval, informed consent and HIPAA compliance, we prospectively recruited consecutive patients with suspected or known CHD who were referred for 3T ferumoxytol-enhanced cardiac MRI over a one-year period from April 2014 to March 2015. Twenty-two patients were included, 10 males (45%), age range 0.2–25 years (mean 6.4±4.8 years). The following data were abstracted for each patient through electronic patient records: demographics, height, weight, body surface area and medical history including details of the patient’s CHD (Table 1).

Table 1.

Patient Demographics

	Patients (n=22)
Age (years)	6.4±4.8
Sex (male)*	45.5% (n=10)
Height (cm)	112.6±26.1
Weight (kg)	23.2±12.6
BSA (m2)	0.83±0.31
Congenital heart disease*
Tetralogy of Fallot, corrected	36.4% (n=8)
Congenitally corrected TGA	13.6% (n=3)
Coarctation of the aorta	13.6% (n=3)
Atrioventricular canal defect	9.1% (n=2)
Ventricular septal defect	9.1% (n=2)
Hypoplastic right ventricle	4.5% (n=1)
Bicuspid aortic valve	4.5% (n=1)
Other	9.1% (n=2)

Except where indicated, data are mean ± standard deviation.

^*Data are percentage of patients, with number in parentheses.

BSA = body surface area; TGA = Transposition of the great arteries.

Image Acquisition

All imaging was performed on a 3T MRI scanner (GE MR750 3T, GE Healthcare, Waukesha, Wisconsin, USA) using a 32-channel cardiac coil (Invivo, Gainesville, Florida, USA), obtaining complete ventricular coverage with a standard two-dimensional (2D) short-axis multi-slice cine SSFP sequence and whole chest coverage with an axial 4D flow sequence.

The 4D flow sequence used minimum echo time (TE) flow-encoding gradients and a simple four-point encoding strategy (20, 21) in a cardiac synchronized 3D Cartesian RF-spoiled gradient echo sequence with variable density Poisson disk k-space under-sampling. Scan parameters included a flip angle of 15°, resolution of 0.8×0.8×1.4 mm³, minimum TE of 1.8–1.9 ms, minimum repetition time (TR) of 3.9–4.3 ms with an additional 5.1 ms for fat-saturation, sampling reduction factor of 2.4×4.4 before k-space corner cutting, views-per-segment of 2–6 depending on heart rate, temporal resolution of 72.9–225.6 ms, bandwidth of ±83.33 kHz, and velocity encoding range (VENC) of 150–300 cm/s. The VENC was selected empirically for each study based on the clinical indication. A combined parallel imaging and compressed sensing algorithm, l1-ESPIRiT, was used to reconstruct each cardiac phase separately (22). The total time for the l1-ESPIRiT reconstruction was less than one hour on a server with two 10-core CPUs. Scans were performed after injection of 0.1 mL/kg ferumoxytol (Feraheme; AMAG Pharmaceuticals, Waltham, Massachusetts, USA) for blood pool enhancement (23–25). General anesthesia was used in the majority of cases (n=17, 77.3%). Scan duration ranged from 4.6–13.0 min (8.5±2.0 min).

Standard 2D multi-slice short-axis cine images were prescribed off a vertical long axis image, and acquired using balanced SSFP (GE Healthcare: FIESTA) ensuring coverage from the cardiac base to apex. The protocol followed published guidelines for cardiac MRI in patients with CHD (26). Scan planes were prescribed by board-certified radiologists with dedicated training in pediatric cardiovascular MRI. In patients capable of breath-holding (n=4, 18.2%), a single signal average breath-hold acquisition was performed. Otherwise, a free-breathing acquisition with two to three signal averages was used to reduce respiratory artefacts (n=18, 81.8%). Scan parameters included a slice thickness of 6–8 mm depending on patient size, matrix 160 × 224, TR of 3.0–4.2 ms, TE of 1.5 ms, flip angle of 60°, views-per-segment of 10–18, and 3 signal averages with retrospective gating. Mean scan duration was 5.3±1.4min.

Image Analysis

Two cardiovascular radiologists (AK and KH, with 8 and 3 years of experience, respectively) independently segmented both SSFP and magnitude 4D flow images in a blinded fashion. LV and RV endocardial and epicardial borders were manually contoured by each radiologist to assess for end-diastolic (ED) and end-systolic (ES) volumes (EDV and ESV, respectively), stroke volume (SV), EF and ventricular mass (at both ED and ES), thus yielding two blinded measurements of ventricular volumes and EF, and four blinded measurements of ventricular mass for each ventricle, by each radiologist, in each subject. The phase of ED and ES were defined visually by the observer, as the phase with the largest and smallest ventricular volumes, respectively. Trabeculations and papillary muscles were included as part of the ventricular blood pool (27).

4D flow analysis was performed using dedicated software for reformatting and viewing of volumetric datasets (Arterys, Arterys Inc, San Francisco, CA). After user specification of anatomic landmarks on magnitude 4D flow images (mitral valve, aortic valve, left ventricle apex, tricuspid valve, pulmonic valve and right ventricle apex), images were reformatted to 6–12 short-axis slices of each ventricle, dynamically tracking the excursion of the apex and its orientation relative to the inlet valve. The valve plane and apex were cross-referenced on long-axis views to prevent errors in edge slice segmentation. Short-axis views were then manually segmented and summed to estimate volumes and mass at ED and ES. To facilitate consistent identification of ED and ES phases, as well as to visualize flow, velocity vector magnitudes were color-coded and blended over the magnitude data (Fig. 1).

Fig. 1.

9-year-old male with aortic coarctation. Screen captures of multiple short-axis 4D flow images demonstrating LV and RV endocardial (upper row) and epicardial (lower row) end-diastolic contours, with color velocity overlay to facilitate identification of the myocardial blood boundary. Left: LV contours; Right: RV contours.

SSFP image analyses were also performed using dedicated commercially available software (QMassMR; Medis, Inc, Leiden, NL) with each study manually processed without knowledge of the results from 4D flow. The valve plane and apex were cross-referenced on vertical long-axis images to prevent errors in segmentation.

Statistical Methods

Statistical analysis was performed using SPSS software version 22 (IBM SPSS, Chicago, IL). Continuous variables were described using mean and standard deviation and categorical variables using frequency and percentage. To compare 4D flow against SSFP, several null hypotheses were tested statistically. First, volume, EF and mass measurements are significantly different between 4D flow and SSFP. Second, there are significant differences in consistency between 4D flow and SSFP estimates of ventricular mass comparing ED and ES measurements. Given the conservation of mass, ED and ES estimations of ventricular mass should be equal, thus serving as an internal control for assessment of measurement precision of the two techniques. Third, there are significant differences in inter-observer agreement between 4D flow and SSFP estimates of ventricular volume, EF and mass.

To identify differences in measurements between 4D flow and SSFP techniques, two-tailed paired Student t tests were used. The sample size was calculated to detect a difference in indexed LV mass measurements between techniques of 8 g/m² with a standard deviation of 12 g/m² using a paired t-test (28). To detect this difference with a power of 80% and alpha error of 0.05, a total of 20 subjects were needed. The agreement between any two measurements was evaluated using both Pearson correlation coefficient (r) and Bland-Altman analysis (as a percentage of the mean). Bland-Altman limits of agreement were calculated as percentage differences from the mean of 1.96 SDs. To evaluate the variance in 4D flow and SSFP estimates of ventricular mass, the mean square difference of ED and ES measurements was calculated for each technique. The difference between ED and ES measurements was assumed to be approximately normal with a mean of zero. An F test was then performed on this statistic. Inter-observer agreement was assessed via the intra-class correlation coefficient (ICC). The ICC model used was two-way mixed and the type was absolute agreement. A two-tailed p-value <0.05 was considered statistically significant.

RESULTS

There was no difference in heart rate between 4D flow and SSFP acquisitions (92.0±23.4 vs. 92.2±23.7 bpm, p=0.637).

Comparison of Ventricular Volumes

EDVs, ESVs and SVs were well correlated between techniques for both LV (r = 0.964–0.972, p<0.001) and RV (r = 0.808–0.934, p<0.001), as shown in Table 2. No statistically significant difference was observed between 4D flow and SSFP measurements of LV volumes. RV measurements of EDV and ESV were significantly lower at 4D flow compared to SSFP (p<0.05), with differences of 8.8 mL/m² (8.5% of the mean) and 4.8 mL/m² (6.4% of the mean), respectively. The mean difference in volumes between techniques was within 10% of zero for all measurements, and Bland-Altman relative limits of agreement were between −22% to 20% for the LV and −52% to 34% for the RV (Fig 2 A–F). For all measurements, the limits of agreement were wider for the RV compared to the LV.

Table 2.

Comparisons between ventricular volume, ejection fraction and mass measurements.

	Paired t-test			Correlation
	4D flow	SSFP	p-value	r-value	p-value
Left ventricle
LV EDV (mL/m2)	71.4±23.1	72.5±23.9	0.365	0.972	<0.001
LV ESV (mL/m2)	31.8±9.5	31.6±10.3	0.738	0.971	<0.001
LV SV (mL/m2)	39.6±15.0	40.9±15.3	0.156	0.964	<0.001
LV EF (%)	54.8±6.1	55.9±6.9	0.058	0.928	<0.001
LV mass ED (g/m2)	40.2±11.3	40.9±11.1	0.479	0.910	<0.001
LV mass ES (g/m2)	41.0±11.7	39.2±10.5	0.05	0.934	<0.001
Right ventricle
RV EDV (mL/m2)	99.8±22.9	108.6±24.3	0.011	0.808	<0.001
RV ESV (mL/m2)	57.4±19.6	62.2±23.0	0.026	0.916	<0.001
RV SV (mL/m2)	42.4±13.1	46.4±13.0	0.087	0.683	<0.001
RV EF (%)	43.3±12.2	44.1±13.6	0.501	0.900	<0.001
RV mass ED (g/m2)	21.5±8.9	21.3±10.5	0.872	0.884	<0.001
RV mass ES (g/m2)	21.2±9.3	21.4±11.7	0.819	0.918	<0.001

Valves are indexed to body surface area other than ejection fractions. Data are mean ± standard deviation.

LV = left ventricle; RV = right ventricle; EDV = end-diastolic volume; ESV = end-systolic volume; SV = stroke volume; EF = ejection fraction; ED = end-diastole; ES = end-systole.

Fig. 2.

Bland–Altman plots of the mean difference for left ventricular (LV) and right ventricular (RV) end-diastolic (ED) and end-systolic (ES) volumes, stroke volume, ejection fraction (EF), and mass between 4D flow and SSFP. For each parameter the average of measurements from both techniques is plotted on the x-axis and the percent difference between techniques is plotted on the y-axis. The solid gray horizontal line plots the mean percent difference and the dotted gray lines indicated the limits of agreement (differences from the mean of 1.96 SDs) for each parameter. A, LV EDV; B RV EDV; C, LV ESV; D, RV ESV; E, LV SV; F RV SV; G, LVEF; H, RVEF; I, ED LV Mass; J, ED RV Mass; K, ES LV Mass; L, ES RV Mass.

Comparison of Ejection Fractions

Ejection fractions were well correlated between techniques for both LV (r = 0.928, p<0.001) and RV (r = 0.900, p<0.001). No statistically significant difference was observed between 4D flow and SSFP measurements of LVEF or RVEF. The mean difference in EFs between techniques was near zero, and Bland-Altman relative limits of agreement were −11% to 7% for the LV and −29% to 27% for the RV (Fig 2 G–H).

Comparison of Ventricular Mass

Ventricular mass measurements at both ED and ES were well correlated between techniques for both LV (r = 0.910–0.934, p<0.001) and RV (r = 0.884–0.918, p<0.001). No statistically significant difference was observed between 4D flow and SSFP measurements of LV and RV mass at both ED and ES. The mean difference in mass between techniques was near zero for all measurements, and Bland-Altman relative limits of agreement were between −23% to 26% for the LV and −41% to 48% for the RV (Fig 2 I–L). For all measurements, the limits of agreement were wider for the RV compared to the LV.

Mean bias between ED and ES mass measurements as a percentage of the mean for 4D flow and SSFP were −1.4±8.9% and 4.2±4.7%, respectively for the LV, and 1.7±8.1% and 1.1±10.0%, respectively for the RV (Fig 3). There was no significant difference between 4D flow and SSFP with respect to mean square difference of ED-ES mass for the LV (F=2.05, p=0.159) or RV (F=0.625, p=0.434).

Fig. 3.

Bland–Altman plots of the mean difference between end-diastolic (ED) and end-systolic (ES) ventricular mass measurements for both 4D flow and cine SSFP. The average of ED and ES measurements is plotted on the x-axis and the percent difference between measurements is plotted on the y-axis. The solid gray horizontal line plots the mean percent difference for 4D flow measurements, and the solid black horizontal line plots the mean percent difference for cine SSFP measurements. The dotted gray lines indicated the limits of agreement (differences from the mean of 1.96 SDs) for 4D flow, and the dashed black lines indicated the limits of agreement for SSFP. A, LV mass; B, RV mass.

Inter-observer Agreement

Inter-observer agreement was moderate to high with both 4D flow and SSFP for the LV (ICC 0.605–0.993 and 0.801–0.982, respectively) and RV (ICC 0.523–0.934 and 0.619–0.952, respectively), with overlapping confidence intervals (Table 3). For all measures, agreement was greater for the LV compared to the RV.

Table 3.

Inter-observer agreement

	4D flow	SSFP
Left ventricle
LV EDV	0.993 (0.984, 0.997)	0.982 (0.957, 0.992)
LV ESV	0.957 (0.899, 0.982)	0.974 (0.938, 0.989)
LV SV	0.957 (0.899, 0.982)	0.970 (0.930, 0.988)
LV EF	0.605 (0.049, 0.836)	0.801 (0.580, 0.912)
LV mass ED	0.948 (0.880, 0.978)	0.960 (0.906, 0.983)
LV mass ES	0.936 (0.851, 0.973)	0.953 (0.890, 0.980)
Right ventricle
RV EDV	0.934 (0.847, 0.972)	0.952 (0.885, 0.980)
RV ESV	0.800 (0.579, 0.912)	0.910 (0.796, 0.962)
RV SV	0.742 (0.474, 0.884)	0.848 (0.668, 0.934)
RV EF	0.523 (0.140, 0.770)	0.619 (0.276, 0.822)
RV mass ED	0.746 (0.481, 0.886)	0.629 (0.291, 0.827)
RV mass ES	0.883 (0.739, 0.950)	0.658 (0.337, 0.842)

Valves are intra-class correlation coefficient (ICC) with 95% confidence intervals in parentheses.

LV = left ventricle; RV = right ventricle; EDV = end-diastolic volume; ESV = end-systolic volume; SV = stroke volume; EF = ejection fraction; ED = end-diastole; ES = end-systole.

DISCUSSION

The results of this study demonstrate, for the first time to our knowledge, that 4D flow can be performed with sufficient resolution to allow assessment of left and right ventricular mass with precision and reproducibility comparable to that of cine SSFP. We also recapitulate results from earlier works showing that equivalent ventricular volume and function measurements can be obtained from 4D flow, but now with ferumoxytol enhancement.

The essential goals of many cardiac MRI exams include quantitative assessment of ventricular volumes, function and mass, in addition to evaluation of systemic and pulmonic flow rates (29). Given that 4D flow yields not only a temporally resolved velocity field allowing quantification of flow, but also concomitant magnitude images which may be used to extract ventricular volumes, function, and mass, this technique has the potential to fulfill all the essential goals of a cardiac MRI exam in a single free-breathing comprehensive 10–15 minute acquisition (9, 30).

Although short-axis stack cine SSFP MRI is the current clinical standard for ventricular volume and function quantification, acquisitions are often lengthy and even an experienced MRI technologist may require considerable oversight from a trained cardiovascular imager to acquire appropriate imaging planes, particularly in the setting of CHD. The volumetric nature of 4D flow acquisitions allows for separation of the processes of image acquisition and interpretation, which may improve clinical workflow and reduce total scan times compared with traditional multi-sequence approaches. The clinical utility of 4D flow MRI was previously limited by prohibitively long image acquisition times (13). However, with the implementation of under-sampling methods including parallel imaging and compressed sensing, near-isotropic images can now be acquired in a scan time practical for clinical practice (9). Recent advances in motion-compensation and compressed sensing may further improve the temporal resolution of 4D flow imaging (31). Under-sampling methods and nonlinear reconstruction approaches have also been applied to real-time, free-breathing and single-breath hold cine acquisitions with promising results (32–34). However, in the setting of CHD, flow information is often required in addition to assessment of ventricular volumes and function. Thus the benefit of the 4D flow analysis approach proposed in this study is that both phase contrast and magnitude data are acquired concurrently, allowing for evaluation of both flow and function with a single comprehensive acquisition.

Cardiac MRI assessment of ventricular size, function and mass has been shown to inform clinical decisions in many types of CHD (35, 36). Accurate and reproducible assessment of ventricular volume, function and mass is especially important in conditions in which the ventricle is subject to chronic volume or pressure overload (19, 37). Furthermore, RV mass has been shown to correlate with functional health status and exercise capacity in the setting of repaired TOF (38, 39), highlighting the importance of this measure with respect to both patient perception and clinical decision-making.

We have confirmed the results of previous work demonstrating that ventricular volume and EF measurements are highly correlated between 4D flow and cine SSFP acquisitions (9). In our study, ventricular volume and EF measurements did not differ significantly between the two techniques with the exception of RV EDV and ESV. RV EDV and ESV were underestimated by 4D flow compared to SSFP, possibly attributable to differences in basal slice selection. We also demonstrated that there is no significant difference in LV or RV mass quantification with 4D flow compared to SSFP acquisitions. However, Bland Altman limits of agreement were relatively wide for all measurements, particularly for the RV, although similar to prior published values (9).

Second, we demonstrated equal internal consistency between 4D flow and SSFP estimates of ventricular mass comparing ED and ES measurements as an internal control. Due to conservation of mass, ED and ES mass measurements should be equal. Previously, both LV and RV mass measurements by cardiac MRI have been shown to accurately reflect actual ventricular weight on autopsy (40, 41). We did not have pathologic specimens or a reference standard in this study, and thus it was not clear which imaging method was truly more accurate.

Finally, we demonstrated moderate to high inter-observer agreement of 4D flow and SSFP estimates of ventricular volume, EF and mass, with no significant differences in agreement between techniques. Similar to prior studies using cine SSFP, we confirmed that for all measurements, the limits of agreement are wider for the RV compared to the LV (18, 42–44). Because of its complex geometry, accurate and reliable quantification of RV volume can be challenging, particularly in the setting of abnormal RV anatomy in patients with CHD (45). Values for inter-observer agreement of LV volumes and mass in our study (ICC 0.957–0.993) are similar to previously reported values (ICC 0.94–0.99) (18). However, the values we report for RV volumes and mass (ICC 0.742–0.934) are slightly lower than previous reports (18, 45), possibly due to the fact that patients included in our study had a range of underlying CHD, with varying and complex anatomy. Similar to prior reports, greater inter-observer variability was noted for EF measurements (ICC 0.525–0.605), which may be explained by the fact that error in two independent volume measurements may be increased by dividing them (18). Results from an earlier study suggested that assessment of ventricular mass in systole may improve reproducibility over diastole (17). However, we did not find a significant difference in precision or inter-observer agreement between ES and ED estimates of ventricular mass, consistent with findings from a more recent study (39).

Several earlier reports have suggested that axial measurements of RV volumes may be more reproducible than those obtained in short-axis orientation (46–48). Given the volumetric nature of 4D flow imaging, data may be segmented in virtually any orientation following acquisition, potentially facilitating assessment of different analysis orientations and methods retrospectively without the need to acquire separate data sets. The effect of different analysis methods on the quantification of RV parameters with 4D flow data sets remains to be tested.

This study is primarily limited by the relatively small number of subjects included. The subjects included in this study were heterogenous, including both pre- and post-operative patients, and differing forms of CHD. In addition, we did not assess intra-observer or inter-study reproducibility. Cine SSFP imaging was performed using breath-holding in those patients who were able, and free-breathing in others, potentially confounding comparison of results. Unfortunately, respiratory navigation was not an available option on our scanner. Prior reports indicate good agreement between cardiac volumes acquired during free breathing and with breath hold commands (49).

All imaging was obtained following administration of ferumoxytol, likely resulting in improved blood-to-myocardium contrast and wall delineation (11). This is an off-label use of the agent, which is FDA approved for iron replacement therapy (24). In our institution, ferumoxytol has been used for over three years as part of our routine clinical practice, largely because it provides reliable high diagnostic image quality while reducing the depth, duration, and frequency of anesthesia. However, this agent must be used with caution, as it may produce significant hypotension, as well as serious hypersensitivity reactions (50). The FDA has recently issued a revised box warning emphasizing that serious, potentially fatal allergic reactions have occurred in patients treated with ferumoxytol (51). A prior study investigating potential differences in the assessment of LV parameters with cine SSFP found no significant difference between pre- and post-contrast acquisitions using gadolinium based contrast (52). However, it remains to be tested whether non-ferumoxytol enhanced 4D flow acquisitions would provide the same results as in the current study.

In conclusion, the results of this study demonstrate that ferumoxytol-enhanced 4D flow MRI determines cardiac ventricular volume, function and mass with comparable precision and inter-observer reproducibility relative to cine SSFP. 4D flow MRI enables rapid comprehensive imaging evaluation of patients with CHD in a single acquisition, potentially improving clinical workflow and reducing anesthesia-related costs due to shortened data acquisition times compared with traditional multi-sequence approaches. Further studies should assess inter-study reproducibility of measurements obtained with 4D flow imaging, which is particularly important in the setting of CHD, as patients may undergo repeated imaging and reliable comparison of measurements between studies is required.