Comparison of free-running whole-heart five-dimensional and four-dimensional flow imaging to standard two-dimensional flow in patients with congenital heart disease

Abstract

Background

Flow quantification by cardiovascular magnetic resonance (CMR) is essential for diagnosis-making and follow-up in patients with congenital heart disease (CHD). Standard two-dimensional phase-contrast (2D flow) sequences require, however, precise image plane prescription, resulting in long exam times. Free-breathing three-dimensional (3D) whole-heart flow sequences (four-dimensional [4D] flow) offer faster acquisition but require respiratory navigation, rendering exam duration unpredictable. Free-running radial, fully self-gated, respiratory and cardiac motion-resolved five-dimensional (5D) flow promises ease of use and stable scan duration. This study compared net flow volumes (NFV) between 2D, 4D, and 5D flow and maximum peak velocity (Vmax) to transthoracic echocardiography (TTE).

Methods

CHD patients were scanned on a 1.5T scanner. 2D flow was performed in the ascending (AAo), descending aorta, main (MPA), right, and left pulmonary arteries, superior vena cava, and pulmonary veins. By an in-house prototype whole-heart free-running 3D radial phase-contrast CMR sequence, end-expiratory 5D flow images were obtained. Finally, 4D flow data were collected. 3D vessel segmentation was performed for 4D and 5D flow datasets. NFV were compared between 2D, 4D, and 5D flow and Vmax to TTE.

Results

Fifty-nine CHD patients (median age 28, interquartile range 17.5, 42% (25/59) women) with right- and/or left-sided CHD were included. Scan duration did not differ significantly between methods. 4D and 5D flow underestimated NFV relative to 2D flow by 4–17 mL per beat which was less prominent for 5D flow and larger vessels. Internal agreement was best for 4D flow, followed by 5D flow, while 2D flow showed the largest variability: MPA vs AAo (bias mL, 95% limits of agreement): 2D flow: −4.84 (−34.18, 24.5), 4D flow −0.84 (−17.85, 16.18), and 5D flow −6.4 (−28.53, 15.73). 2D and 4D but not 5D flow underestimated Vmax compared to TTE: AAo: TTE 120 ± 46; 2D 108 ± 36; 4D 107 ± 36; 5D 115 ± 40 cm/s; p < 0.03 TTE vs 2D and 4D flow; p = 0.253 TTE vs 5D flow.

Conclusion

4D and 5D flow provided measurements comparable to 2D flow. Despite systematic underestimation of NFV, internal agreement of whole-heart methods was superior, while scan time did not increase significantly. 5D flow yielded the most accurate Vmax compared to TTE, supporting its potential for comprehensive non-invasive hemodynamic evaluation.

Graphical abstract

1. Introduction

Patients born with congenital heart disease (CHD) require lifelong follow-up care due to their elevated risk of cardiovascular complications [1], [2]. Cardiovascular magnetic resonance (CMR) plays an important role in the evaluation of these patients, providing non-invasive assessment of cardiac anatomy, ventricular function, tissue characterization, and blood flow [3], [4], [5], [6]. Flow quantification is a fundamental component of congenital CMR protocols, allowing diagnosis, guiding treatment, and follow-up care.

Two-dimensional phase-contrast (2D flow) CMR imaging is considered as the clinical gold standard for non-invasive quantification of blood flow and velocity patterns. Despite its clinical robustness, this technique presents several drawbacks [7], [8]. 2D flow imaging requires meticulous prescription of imaging planes, a process that is both time-consuming and operator-dependent, necessitating the expertise of trained CMR technicians to accurately define scan planes and adapt acquisition parameters [9], [10], [11]. Additionally, 2D flow is sensitive to respiratory motion, requiring repeated breath-holds or respiratory averaging, which can considerably prolong total scan time [9], [10]. Furthermore, the unidirectional nature of 2D flow measurements restricts the assessment of complex, multidirectional flow patterns, frequently observed in CHD, particularly in cases involving turbulent or eccentric flow due to anatomical malformations, valvulopathies, or postsurgical alterations [12], [13].

To overcome these challenges, three-dimensional (3D) whole-heart sequences that enable 3D flow measurements, also referred to as four-dimensional (4D) flow CMR, were developed, providing comprehensive volumetric and multidirectional velocity encoding (VENC). This technique enables retrospective blood flow quantification at any vascular location from a single acquisition and can be performed under free-breathing conditions, thereby improving patient comfort and reducing overall scan time. However, 4D flow typically relies on respiratory navigator gating, which is operator-dependent and often limited by low navigator efficiency in patients with deep or irregular breathing patterns, thus leading to unpredictable and often prolonged scan durations.

In routine clinical practice, where CMR examination time is limited, knowledge of scan duration is essential to ensure acquisition of all clinically relevant data within the available time. To address the limitations of 4D flow, recent advances in acquisition and reconstruction techniques have enabled the development of free-running, whole-heart five-dimensional (5D) flow CMR. This method simultaneously resolves cardiac and respiratory motion within a single, continuous acquisition, eliminating the need for external electrocardiogram (ECG) gating or respiratory navigators [30] . By incorporating two temporal dimensions (cardiac and respiratory phases), 5D flow provides fully motion-resolved hemodynamic data, enabling comprehensive assessment of cardiorespiratory interactions throughout the entire heart and great vessels. Its ease of use, predictable scan time, and motion robustness make it particularly attractive for imaging patients with CHD.

While prior studies have compared net flow volumes (NFV) between 2D and 4D flow, direct comparisons including 5D flow remain scarce [8], [15], [16], [17], [18]. Moreover, the potential of advanced flow imaging techniques to evaluate smaller vessels, such as pulmonary veins, has received limited attention, despite their important role in CHD. Furthermore, maximum peak velocity (Vmax) is a crucial parameter for hemodynamic evaluation. However, data comparing 2D, 4D, and 5D flow-derived NFV and Vmax against transthoracic echocardiography (TTE) are currently lacking.

This study, therefore, compared NFV between 2D, 4D, and 5D flow in right- and/or left-sided CHD patients, by evaluating accuracy and reproducibility across large and small thoracic vessels [14], [19], [20]. Vmax acquired by 2D, 4D, and 5D flow was compared to the Vmax of TTE.

2. Methods

2.1. Patient population

This is a prospective, single-center study including consecutive pediatric and adult CHD patients, followed in a tertiary center who were referred for a clinically indicated CMR exam. Inclusion criteria included all comers between 2020 and 2022 with any type of CHD, whether operated or not, consecutive patients of any age and sex, and a clinical indication for CMR. Exclusion criteria were inability to perform the sequences for blood flow measurements, inability to give consent, pregnancy, and any known contraindication for CMR. The study was approved by the local Ethics Committee (CER-VD 2019-02164) and informed consent was signed. The study complies with the Declaration of Helsinki.

2.2. Data acquisition and reconstruction

All patients were scanned on a 1.5T scanner (Magnetom Sola, Siemens Healthineers, Erlangen, Germany). The choice of contrast regimen (none, gadolinium [Gadovist, Bayer Healthcare, Leverkusen, Germany], and ferumoxytol [Sandoz Inc., Princeton, New Jersey]) was selected by the responsible CMR physician according to the clinical indication of each patient.

Typical technical parameters for 2D, 4D, and 5D flow are shown in Table 1. Scan duration was determined from the timestamps of the first and last acquired images.

Table 1.

Technical parameters.

	2D flow	4D flow	5D flow
TR/TE, ms	5.25/3.04	4.25/2.24	(4.7–5.3)/(2.9–3.5)
Averages, N	2	1	1
Slice thickness, mm	6	2.5	2.1–2.5
Inplane resolution, mm	1.4–1.9	2.5	2.1–2.5
Temporal resolution, ms	31.5	51	40
Flip angle, °	20	7	7
Bandwidth, Hz/Px	491	558	599
FOV read, mm	340*	200–400*	220–240
Matrix, N	256 × 256	256 × 256	96 × 96 × 96
Scan time, mean± SD, min	All: 7.3 ± 2.7 With PV: 6.5 ± 2.6 w/o PV: 8.9 ± 2.2	9.7 ± 3.6	7.3 ± 2.7†

Data are absolute numbers if not otherwise indicated

TR time of repetition, TE time of echo, FOV field of view, PV pulmonary veins, 2D two-dimensional, 4D four-dimensional, 5D five-dimensional, SD standard deviation

^*Phase oversampling added as needed to avoid wrap-around

^†p < 0.02 4D vs 2D and 4D vs 5D flow

2.2.1. 2D flow

Free-breathing 2D flow was performed according to published guidelines [21]. A conventional phase-sensitive gradient-echo sequence was prescribed on two planes perpendicular to the vessel axis in the ascending and descending aorta (AAo, DAo), the main, right, and left pulmonary artery (MPA, RPA, LPA), the superior vena cava (SVC), and in right and left pulmonary veins (RPV, LPV), where clinically relevant.

The imaging plane for the MPA was typically positioned 1–2 cm above the valvular level but was adjusted when necessary to avoid artifacts, for example, in patients with right ventricular outflow tract conduits or stents. In the RPA and LPA, the imaging plane was placed before the first branch and approximately 1 cm distal to the bifurcation, in the AAo between the sinotubular junction and the level of the RPA, and for the DAo at the diaphragmatic level [9]. SVC flow was measured distal to the azygos vein. Pulmonary venous (PV) flow was measured in the superior and inferior LPV and RPV, and mid RPV if present.

A VENC of 150 and 100 cm/s was set for arterial and venous flow, respectively, and adapted to 20% above the Vmax if required. Each dataset was reconstructed into 25 magnitude (anatomic) and phase (velocity) images per cardiac cycle.

2.2.2. 4D flow

A compressed-sensed 4D flow sequence (provided by Siemens Healthineers as a Works-in-Progress package) was acquired using retrospective ECG gating [14]. Navigators were placed on the hepatic dome to detect respiratory cardiac motion [14]. A scout sequence was run to determine the acquisition window.

2.2.3. 5D flow

A custom prototype free-running sequence was used to acquire 3D radial phase-contrast CMR data [14], [19], [20]. A box was placed on axial, coronal, and sagittal single-shot images. No ECG or respiratory triggering was necessary.

For 4D and 5D flow, the 3D volume covered the heart, the thoracic aorta, MPA, RPA, LPA, SVC, and proximal pulmonary veins, and VENC was set to 200 cm/s. The field of view (FOV) was chosen accordingly for 2D and 4D flow to avoid wrap-around artifacts with minor impact on scan time (Table 1), while for 5D flow the FOV was 220–240 × 220–240 × 220–240 mm, according to the patient’s size. For 4D and 5D flow, each dataset was reconstructed into 25 magnitude and phase images per cardiac cycle.

2.3. Postprocessing

Postprocessing was performed using Cvi42 (Version 5.12.1, Circle Cardiovascular Imaging, Calgary, Alberta, Canada). For 2D flow, offset and anti-aliasing corrections, if necessary, were applied. NFV was obtained by subtracting retrograde from antegrade flow. Vmax was determined.

For 4D and 5D flow, 3D vessel segmentation was performed based on computed phase-contrast angiography images. Regions of interest (ROIs) were placed at the same anatomical locations used for 2D flow, and each ROI was oriented perpendicular to the vessel using two orthogonal views and adapted throughout the cardiac cycle, according to published consensus recommendations [22]. Images were analyzed at the end-expiratory phase. NFV were compared between 2D, 4D, and 5D flow, and Vmax was compared with the TTE (GE Healthcare Vivid E90, Chicago, Illinois) study closest to the corresponding CMR date (range: 3–198 days). TTEs were performed during a routine outpatient follow-up appointment in adult patients by two consultants (J.B. and T.R., both >20 years of experience), and in pediatric patients by four consultants (S.D.B., N.S., M.P., and S.B.D., all >10 years of experience). Aortic peak flow was assessed by apical five- and three-chamber views with continuous wave (CW) Doppler measurements. Pulmonary artery flow was assessed on parasternal long- and short-axis views with measures of peak velocity by CW Doppler according to guidelines [23], [24].

Flow measurements were performed primarily by one investigator (F.D., 1 year of training), with assistance for 2D flow by another investigator (S.F., 3 years of training), and for 4D and 5D flow by two further investigators (E.T. and T.R., both >15 years of training). All investigators were blinded to the results of the respective other flow modalities, the clinical situation, and type of CHD. Inter- and intraobserver variability analyses were performed by selecting 15 random cases. For intraobserver analysis, the time between the first and second analysis was >8 weeks (Tables 2 and 3).

Table 2.

Patient characteristics.

	Patient population
Age, median (IQR)	28 (18)
Sex, female, N (%)	25 (42)
Mean systemic EF, %	57
Mean non-systemic EF, %	49
Aortopathy, N (%)	10 (17)
RVOT obstruction/right-sided CHD (Fallot, pulmonary atresia, pulmonary stenosis), N (%)	18 (31)
Ross, N (%)	10 (17)
TGA, N (%)  Atrial switch  Arterial switch	8 (14) 5 (63) 3 (37)
Shunt lesion, N (%)	4 (7)
Other (Ebstein, Fontan, cyanotic heart disease), N (%)	9 (15)
Use of contrast  None, N (%)  Gadolinium, N (%)  Ferumoxytol, N (%)	18 (30) 33 (56) 8 (14)

Date shown are mean (IQR), number or percentage

CHD congenital heart disease, EF ejection fraction, IQR interquartile range, N number, RVOT right ventricular outflow tract, TGA transposition of the great arteries

Table 3.

Comparison of NFV bias between CMR-based methods.

NFV (mL/beat)	2D vs 4D flow	2D vs 5D flow	4D vs 5D flow	p
AAo	−11.0 (−32.0, 8.9)*	−4.67 (−29.44, 20.10)*	6.78 (−21.14, 34.70)*	<0.001
DAo	−6.03 (−19.24, 7.18)†	−12.49 (−29.95, 4.96)	−6.35 (−18.97, 6.27)	<0.001
MPA‡	−8.62 (−41.24, 24.01)	−6.9 (−39.41, 25.61)	1.87 (−26, 29.74)§	0.003
RPA	−10.25 (−26.1, 5.59)	−8.64 (−25.18, 7.91)	1.87 (−10.86, 14.59)¶	<0.001
LPA	−8.08 (−28.47, 12.32)	−8.02 (−30.07, 14.03)	0.19 (−15.48, 15.86)||	<0.001
SVC	−5.33 (−17.44, 6.37)‡	−7.09 (−16.74, 2.56)‡	−1.38 (−9.35, 6.6)	<0.001
RPV	−6.44 (−20.63, 7.75)	−17.69 (−38.97, 3.58)**	−8.91 (−25.85, 8.02)	<0.001
LPV	−9.47 (−36.11, 17.18)	−10.70 (−32.06, 10.66)††	−3.61 (−21.07, 13.84)	0.035

Data shown are bias (mL/beat) (95% lower, upper LOA). P-values refer to one-way analyses of variance with Tukey post-hoc analyses for comparison of bias between the three sequences

2D two-dimensional, 4D four-dimensional, 5D five-dimensional, AAo ascending aorta, CMR cardiac magnetic resonance, DAo descending aorta, LOA limits of agreement, LPA left pulmonary artery, LPV left pulmonary vein, MPA main pulmonary artery, NFV net flow volume, RPA right pulmonary artery, RPV right pulmonary vein, SVC superior vena cava, vs versus

^‡Six patients were excluded due to stent artifact in the MPA (2), or Fontan-type physiology (4)

^*AAo: p < 0.001 for all post-hoc comparisons

^†DAo: p < 0.001 2Dvs4D vs 2Dvs5D and 2Dvs5D vs 4Dvs5D

^‡SVC: p < 0.04 2Dvs4D vs 2Dvs5D and 4Dvs5D; and 2Dvs5D vs 4Dvs5D

^§MPA: p < 0.018 4Dvs5D vs 2Dvs4D and 2Dvs5D

^¶RPA: p < 0.001 2Dvs4D vs 4Dvs5D and 2Dvs5D vs 4Dvs5D

^||LPA: p < 0.001 2Dvs4D vs 4Dvs5D and 2Dvs5D vs 4Dvs5D

^**RPV: p < 0.004 2Dvs5D vs 2Dvs4D and 4Dvs5D

^††LPV: p = 0.023 2Dvs5D vs 4Dvs5D

2.4. Statistical analysis

Normality was assessed using the Shapiro-Wilk test. Data are presented as mean ± standard deviation for normally distributed parameters, as median (interquartile range [IQR]) for non-normally distributed parameters, or as number and percentage, as appropriate. The level of significance was set to a p-value <0.05. Bland-Altman analysis was used to assess internal agreement between 2D, 4D, and 5D flow for NFV, backward volume, regurgitation fraction (RF), and for inter- and intraobserver analyses. For assessment of internal agreement, patients with shunt and/or Fontan-type physiology were excluded, the latter to avoid the influence of potential arterial and/or venous collateral flow. Paired t-tests and analyses of variance (ANOVA) were used for comparisons of continuous variables. Tukey and Bonferroni post-hoc analyses were used after a significant ANOVA to identify specific pairwise differences. Pearson linear regression analyses were performed for the evaluation of the relationship of Vmax as obtained by TTE and 4D and 5D flow. Microsoft Excel (Microsoft, Redmond, Washington) was used as a database. Statistical analyses were performed using SPSS (v29.0.2; IBM, Armonk, New York) and GraphPad Prism (v10.4.1; GraphPad Software, San Diego, California).

3. Results

Fifty-nine patients (median age 28, IQR 17.5, 42% women) with right- and/or left-sided CHD were prospectively included. Table 2 summarizes patient characteristics and the use of contrast agents. Scan time did not differ significantly between methods (Table 1). Owing to time constraints, 2D PV flow was acquired only when clinically relevant, which applied to 15 patients. Due to a technical issue, 4D flow scan times were available for only 42 cases.

Fig. 1 shows NFV and Table 4 compares the bias of 2D, 4D, and 5D flow. 4D and 5D flow systematically underestimated NFV compared to 2D flow, which was more pronounced for 4D flow, in DAo and in smaller vessels such as pulmonary veins.

Fig. 1.

Comparison of NFV between 2D, 4D, and 5D flow. Data shown are mean and SD, p-values refer to analyses of variance with Bonferroni post-hoc analyses for comparison of bias between the three sequences. *p < 0.03, **p < 0.009, ***p < 0.001. 2D two-dimensional, 4D four-dimensional, 5D five-dimensional, AAo ascending aorta, DAo descending aorta, LPA left pulmonary artery, LPV left pulmonary vein, MPA main pulmonary artery, NFV net flow volume, RPA right pulmonary artery, RPV right pulmonary vein, SVC superior vena cava, SD standard deviation

Table 4.

Comparison of internal agreement (patients without shunt or Fontan-type physiology (51)).

NFV (mL/beat)	2D flow	4D flow	5D flow	p
MPA-AAo	−4.84 (−34.18, 24.5)	−0.84 (−17.85, 16.18)	−6.4 (−28.53, 15.73)	0.079
(DAo+SVC)-AAo	−0.99 (−14.99, 13.0)	−0.72 (−18.3, 16.85)	−16.52 (−45.36, 12.33)*	<0.001
(RPA+LPA)-MPA	7.5 (−21.97, 36.96)†	−6.25 (−22.65, 10.14)	−6.36 (−27.55, 14.82)	<0.001
(RPA+LPA)-AAo	1.21 (−23.39, 25.81)‡	−7.22 (−28.88, 14.43)	−12.83 (−39.08, 13.41)	<0.001
(RPA+LPA)-(DAo+SVC)	2.17 (−19.48, 23.81)	−5.57 (−23.48, 12.34)§	3.68 (−20.16, 27.53)	<0.001

Data shown are bias (mL/beat) (95% lower, upper LOA). P-values refer to one-way analyses of variance with Tukey post-hoc analyses for comparison of bias between the three sequences

2D two-dimensional, 4D four-dimensional, 5D five-dimensional, AAo ascending aorta, DAo descending aorta, LOA limits of agreement, LPA left pulmonary artery, MPA main pulmonary artery, NFV net flow volume, RPA right pulmonary artery, SVC superior vena cava, vs versus

^*(DAo+SVC)-AAo: p < 0.001 2D and 4D vs 5D

^†(RPA+LPA)-MPA: p < 0.001 2D vs 4D and 5D

^‡(RPA+LPA)-AA: p < 0.005 2D vs 4D and 5D

^§(RPA+LPA)-(DAo+SVC): p < 0.003 2D vs 4D and 4D vs 5D

Biases were generally small to moderate, with the largest discrepancies observed in pulmonary veins and the DAo. When directly comparing 4D and 5D flow, differences were smaller, indicating better agreement between the two whole-heart techniques.

Table 4 presents the comparison of internal agreement in patients without shunt or Fontan-type physiology. The latter was excluded to avoid the potential influence of any type of collaterals. 4D flow showed the most consistent agreement according to the 95% limits of agreement (LOA), when compared to 2D and 5D flow. 5D flow demonstrated larger biases in some systemic comparisons, particularly (DAo+SVC)-AAo. Fig. 2 shows Bland-Altman analyses of internal agreement assessments for MPA-AAo and (RPA+LPA)-MPA. In the pulmonary circulation, 4D and 5D flow demonstrated smaller biases and tighter LOA than 2D flow. Additional Bland-Altman plots illustrating further internal agreement comparisons are provided in Fig. S1.

Fig. 2.

Bland-Altman analyses for comparison of internal agreement (patients without shunt or Fontan-type physiology). Data shown are bias (mL/beat) (95% lower, upper LOA). P-values refer to one-way ANOVA with Tukey post-hoc analyses for comparison of bias between the three sequences. MPA-AAo: 2D vs 4D vs 5D: p = 0.079; 2D vs 4D p = 0.283; 2D vs 5D p = 0.785; 4D vs 5D p = 0.064 (RPA+LPA)-MPA: 2D vs 4D vs 5D, 2D vs 4D and 2D vs 5D: p = <0.001; 4D vs 5D: p = 0.863. 2D two-dimensional, 4D four-dimensional, 5D five-dimensional, AAo ascending aorta, ANOVA analysis of variance, LOA limits of agreement, LPA left pulmonary artery, MPA main pulmonary artery, NFV net flow volume, RPA right pulmonary artery

Fig. 3 shows Bland-Altman analyses for comparison of internal agreement for NFV bias across pulmonary branch arteries and veins (RPV-RPA, LPV-LPA). Compared to 4D flow, 5D flow demonstrated wider LOA, particularly in the RPV-RPA comparison, indicating slightly less internal agreement than 4D flow. The only significant difference between methods was observed in the RPV-RPA comparison between 4D and 5D flow (p < 0.001). LPV-LPA comparisons showed broadly similar performance across all three methods.

Fig. 3.

Bland-Altman analyses for comparison of internal agreement in the pulmonary circulation (patients without shunt or Fontan-type physiology). Data shown are bias (mL/beat), (95% lower, upper LOA). P-values refer to one-way ANOVA with Tukey post-hoc analyses for comparison of bias between the three sequences. RPV-RPA: p < 0.001 4D vs 5D flow. 2D two-dimensional, 4D four-dimensional, 5D five-dimensional, ANOVA analysis of variance, LOA limits of agreement, LPA left pulmonary artery, LPV left pulmonary vein, NFV net flow volume, RPA right pulmonary artery, RPV right pulmonary vein

Table 5 presents backward volumes and regurgitant fraction (RF) and Table S4 provides Bland-Altman analyses comparing RF between 2D, 4D, and 5D flow, demonstrating no significant differences among the three methods.

Table 5.

Comparison of backward volume and regurgitation fraction.

	2D flow	4D flow	5D flow	p
Backward volume, mL (mean±SD)
AAo	−4.01±8.91	−3.77±8.61	−2.39±5.66	0.487
MPA	−9.14±14.35	−5.72±12.20	−6.35±12.80	0.366
RPA	−2.73±4.78	−1.93±3.48	−1.59±3.37	0.275
LPA	−3.88±7.22	−2.97±5.71	−2.59±4.72	0.492
Regurgitation fraction, % (mean±SD)
AAo	4.56±9.27	5.04±10.19	3.06±6.53	0.448
MPA	11.30±18.14	6.67±11.54	8.04±14.48	0.269
RPA	5.92±9.53	5.54±8.63	4.78±9.33	0.792
LPA	9.78±16.93	8.57±12.87	8.84±15.26	0.904

Data shown are regurgitant volume (in mL and %, respectively) ± SD . P-values refer to one-way analyses of variance

2D two-dimensional, 4D four-dimensional, 5D five-dimensional, AAo ascending aorta; LOA limits of agreement, LPA left pulmonary artery, MPA main pulmonary artery, RPA right pulmonary artery, SD standard deviation

The comparison of Vmax in the AAo and MPA between CMR-based methods and TTE showed no difference between 5D flow and TTE but significant underestimation by 2D and 4D flow: Vmax AAo: TTE 120 ± 46 cm/s, 2D flow 108 ± 36 cm/s, 4D flow 107 ± 36 cm/s, 5D flow 115 ± 40 cm/s (p < 0.03 for TTE vs 2D flow and 4D flow and p = 0.253 for TTE vs 5D flow).

In the MPA, Vmax were TTE 168 ± 86 cm/s, 2D flow 153 ± 78 cm/s, 4D flow 142 ± 62 cm/s, 5D flow 145 ± 72 cm/s (p < 0.03 for TTE vs 2D flow and 4D flow; p = 0.071 for TTE vs 5D flow). Differences in Vmax between TTE and all three CMR-based methods increased progressively with rising Vmax (Fig. 4).

Fig. 4.

Correlation of bias (cm/s) and maximal peak velocity between TTE versus 2D, 4D, and 5D flow. 2D two-dimensional, 4D four-dimensional, 5D five-dimensional, R correlation coefficient, TTE transthoracic echocardiography, Vmax maximal peak velocity

Bias for 2D, 4D, and 5D flow NFV did not vary significantly based on the contrast regimen used in both larger and smaller vessels (no contrast, gadolinium, or ferumoxytol, Table S1).

Interobserver agreement (Table S2) was high across all three techniques, with small biases and acceptable LOAs, with most p-values being non-significant and only minimal bias between observers. Intraobserver variability was very low across all three methods (Table S3).

4. Discussion

In this study, including 59 CHD patients representing a heterogeneous but representative CHD population for a tertiary center, 4D and 5D flow were compared to 2D flow for flow determination in thoracic and cardiac vessels. Evaluation of flow is crucial in CHD patients for diagnosing and monitoring shunt lesions, and for assessing stenosis or regurgitation. While quantification of absolute NFV is an appealing way to compare different techniques, internal agreement is particularly important for the quantification of shunt or collateral flow.

When comparing NFV between the three methods, both whole-heart techniques showed systematic underestimation relative to 2D flow, a well-known observation which has already been discussed by several studies [12], [13], [17], [19], [25]. Notably, the degree of underestimation was smaller with 5D flow than with 4D flow. The discrepancy was most pronounced in the DAo and in the pulmonary veins.

Interestingly, despite the observed underestimation of NFV, no significant differences were found in backward volume or RF across 2D, 4D, and 5D flow, underscoring the clinical potential of these techniques for the assessment of valvular regurgitation.

Internal agreement comparisons showed, however, that whole-heart techniques outperformed 2D flow, with 5D flow falling slightly behind 4D flow for the agreement across systemic and pulmonary flow assessment. Although 5D flow exhibited slightly larger biases and LOA, the two whole-heart methods showed no significant differences in bias except in the (RPA+LPA)-(DAo+SVC) comparison. A possible explanation is the flow quantification in the DAo, which constituted a major source of underestimation in 5D flow.

4.1. The descending aorta issue

The DAo was the vessel with the largest biases and LOA across whole-heart techniques. A plausible explanation is the susceptibility-prone location of the DAo, situated adjacent to the pulmonary parenchyma and liver. The DAo was segmented in all three CMR-based methods at the upper diaphragmatic level, where respiratory motion and susceptibility artifacts due to lung-liver interaction are amplified. These factors are particularly relevant for self-navigated free-breathing sequences such as 5D flow, leading to signal loss and resulting in underestimation of NFV. Another possible reason could be an insufficient background correction in the 5D flow sequence. These observations underscore the need for sequence optimization at this level.

4.2. Further potential sources of bias

Several technical factors may contribute to the underestimation of NFV by 4D and 5D flow. In particular, the use of compressed sensing and the lower temporal resolution of these techniques might partially explain the lower NFV observed with 4D and 5D flow. Furthermore, vessel segmentation strategies differ intrinsically between 2D flow and whole-heart approaches. In 2D flow, analysis is constrained to a predefined, operator-selected imaging plane, whereas in 4D and 5D flow, planes are retrospectively placed and may be oriented either using magnitude images or by following the flow jet direction, depending on what was defined by the team. This difference in practice may be particularly relevant in highly pulsatile vessels such as the AAo and MPA, or in the presence of eccentric jets caused by valvular disease, both of which are common in CHD. Casciaro et al. discussed that ROI size and plane angulation could produce errors in NFV quantification in the AAo up to 20% by 4D flow, especially when dilatation is present [26]. The 2023 4D flow CMR consensus statement recommends measuring flow by using multiple 2D planes oriented orthogonally to the vessel and adjusting these planes throughout the cardiac cycle [22]. However, there is still no agreement on whether, in cases where the jet direction diverges from the vessel axis (such as with eccentric jets), segmentation should be aligned perpendicular to the jet or to the vessel. How these segmentation choices affect NFV in 4D and 5D flow datasets remains insufficiently explored.

In contrast, in 2D flow, plane prescription by the CMR technician is performed by orientating the plane perpendicular to the vessel of interest, thus orientating from static multislice balanced steady-state free precession images in three orthogonal planes. Therefore, the jet direction in the case of eccentric jets will not be taken into consideration when performing the plane prescription. This could lead to potential overestimation of NFV due to malalignment in 2D flow.

Furthermore, postprocessing algorithms (offset correction, anti-aliasing strategies) are not openly available and probably remain heterogeneous among different software platforms. Although flow measurements across different postprocessing software generally show acceptable agreements, variations in how these filters are implemented may partially explain technique-specific discrepancies [25].

4.3. Comparison of Vmax between CMR and TTE

When comparing Vmax between the three CMR-based techniques and TTE, 5D flow showed no significant difference from TTE in the AAo and MPA, whereas 2D and 4D flow significantly underestimated Vmax. This is an important finding in favor of 5D flow, as velocity estimation is crucial for assessing stenosis severity, especially relevant in CHD. Furthermore, in some patients, body morphology or cardiac anatomy precludes Vmax measurement by TTE due to Doppler misalignment or suboptimal acoustic windows, underscoring the value of CMR flow assessment.

Interestingly, the discrepancy between TTE and all CMR-based methods increased with higher Vmax (Fig. 4), suggesting that underestimation of Vmax by CMR may increase with the degree of stenosis or jet turbulence and which probably holds true for all three methods. This observation remains, however, hypothesis-generating, as TTE and CMR were not acquired the same day, therefore equivalent hemodynamic conditions cannot be guaranteed, and only few patients had very high-velocity jets. Further studies, including same-day TTE and CMR, would be helpful. Furthermore, the question remains whether TTE really can be considered as a gold standard, which has been challenged by a publication showing an overestimation of peak gradients by TTE when compared to catheter-derived measures. A comparison of CMR to invasively derived parameters could help to answer this question [27].

Additionally, implementation of dual-VENC techniques could potentially improve the determination of velocities [28].

4.4. Inter- and intraobserver variability

Inter- and intraobserver variabilities were very low and comparable across all CMR-based techniques (Tables S2 and S3) and unlikely to be clinically relevant. 2D flow did not demonstrate consistent superiority over 4D or 5D flow. This result was unexpected, as 2D flow benefits from fixed image planes whereas 4D and 5D flow require manual placement of planes by the observers. Nonetheless, whole-heart techniques did not provide inferior reproducibility of measures, suggesting that plane placement may be less critical than assumed when readers are adequately trained. These findings support the robustness and reproducibility of whole-heart techniques for routine clinical use and challenge the assumption that 2D flow inherently provides more reproducible results.

4.5. Small-vessel flow quantification

PV flow quantification remains challenging across all modalities. However, especially in CHD, assessment of PV flow is crucial in certain situations (e.g., impossibility of flow measurement in pulmonary branch arteries due to the presence of stents, anomalous PV return, and accurate shunt assessment in Fontan patients).

Small-vessel size, low velocities, limited signal-to-noise ratio, and often suboptimal contrast‐to‐noise ratio make PV flow particularly sensitive to noise and segmentation variability [29]. In our cohort, all three methods showed large biases and wide LOA, with no consistent trend in favor of one method. We therefore realized that there may be a lack of robust techniques to assess PV flow by CMR.

These limitations might be partly attributable to the lack of dedicated postprocessing tools optimized for small venous structures. In addition, only a few centers regularly acquire 2D flow of pulmonary veins, contributing to limited expertise and literature.

Of note that, due to significant prolongation of scan time when acquiring PV NFV by 2D flow, it was only assessed in 15 patients, and where clinically relevant. Therefore, our study is limited by a small amount of 2D PV flow data due to scan time constraints.

Future improvements, such as dual-VENC implementations, adaptive noise reduction, and dedicated segmentation algorithms, will be necessary to achieve robust PV quantification.

4.6. Choice of contrast medium

Although included only as a secondary analysis, contrast regimen (none, gadolinium, or ferumoxytol) did not significantly affect NFV bias in either large or small vessels. These results should be interpreted cautiously due to the unequal distribution of contrast types and the limited sample size in the non-contrast and ferumoxytol subgroups. Nonetheless, the findings support that whole-heart techniques may be reasonably robust across both contrast and non-contrast regimens.

4.7. Exam duration

Overall, 2D flow remains the fastest option when only a few vessels are evaluated. But exam duration increases with the number of vessels imaged and particularly when pulmonary veins (Table 2) or Fontan pathway need to be evaluated.

4D flow had the longest and most variable scan time, largely due to fluctuating navigator efficiency and respiratory motion, while acquisition of 5D flow was slightly longer than 2D flow but markedly more predictable and consistent.

Of note that the scan time does not reflect the whole time needed for flow acquisition. Plane prescription, which however has not been measured in this study, requires approximately 1 minute per vessel for 2D flow, while in 4D flow navigator placement and acquisition of the scout sequence easily takes 2–4 min. Placement of the box for the 5D flow, in contrast, only takes 30 s.

Unfortunately, postprocessing duration was not measured. Although we cannot provide numbers and statistical analyses, in our experience, for an experienced reader, 2D postprocessing with the software used takes about 20–30 min for great vessels (AAo DAo, MPA, RPA, LPA, and SVC) and 30 min for 4D and 5D flow. Adding analysis of pulmonary veins adds, for all methods, easily 15–20 min. We therefore believe that postprocessing should not differ importantly between methods.

5. Limitations

In addition to the already above-mentioned technical limitations, this study was also limited by its single-center design, with a relatively moderate sample size and a heterogeneous, but typical CHD population. The available total exam time was limited for each patient; therefore, assessment of pulmonary veins in 2D flow could not be performed for all patients. This speaks, however, in favor of 4D and 5D flow in which pulmonary veins can always be retrospectively measured.

6. Conclusion

This study, providing data from a heterogeneous CHD population, reflects the complexity typically seen in tertiary centers. Across a broad range of measurements, 4D and 5D flow showed systematic underestimation of NFV, with 5D flow being comparable to 4D flow, close agreement with TTE for Vmax, and comparable inter- and intraobserver variabilities. Both whole-heart techniques demonstrated better internal agreement than 2D flow, despite the underestimation of NFV, and despite limitations in the DAo and small vessels. Importantly, backward volume and RF did not differ significantly among 2D, 4D, and 5D flow, supporting the clinical applicability of these techniques for the assessment of valvular regurgitation.

While 5D flow remains a developing technology, our findings support its clinical potential for routine follow-up and decision-making in CHD, particularly when complemented with 2D flow in select cases. Further technical advances, especially dual-VENC, optimized postprocessing, and multicenter evaluations across scanner platforms, will be essential to refine accuracy and ultimately determine whether whole-heart flow imaging could replace 2D flow as the clinical standard.

Funding

F.D. and P.G. are research fellows (PhD students) funded by grants of the Swiss National Science Foundation. F.D. received funding from the Solis Foundation. Tobias Rutz is the principal investigator on the Swiss National Science Foundation Grant 320030_231872 that funded part of this research. Christopher W. Roy is the principal investigator on a Swiss National Science Foundation Grant PZ00P3_202140 that funded part of this research. Matthias Stuber is the principal investigator on the Swiss National Science Foundation Grants 320030_173129 and 201292 that funded part of this research.

Author contributions

Jürg Schwitter: Writing – review & editing. Estelle Tenisch: Writing – review & editing, Methodology. Fabienne Dirbach: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Data curation. Tobias Rutz: Writing – review & editing, Supervision, Methodology, Investigation, Funding acquisition, Conceptualization. Mariana B.L. Falcão: Resources. David Rodrigues: Resources. Matthias Stuber: Writing – review & editing. Michael Markl: Writing – review & editing. Christopher W. Roy: Writing – review & editing, Investigation. Milan Prša: Writing – review & editing. Paolo Garelli: Writing – review & editing. Sara FÄssler: Data curation.

Declaration of generative AI and AI-assisted technologies in the writing process

The authors used ChatGPT to support English language editing, specifically for grammar correction, wording refinement, and improvement of sentence fluency. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Declaration of competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

Figure S1

Supplementary material

.

Figure S2

Supplementary material

.