Clinical evaluation of compressed sensing 4D flow magnetic resonance imaging in patients with congenital heart disease

Introduction

Congenital heart disease (CHD) is a common birth defect. Its diagnosis and treatment rely heavily on the accurate evaluation of the structure and function of the heart and great vessels (1-3). Four-dimensional cardiovascular magnetic resonance flow imaging (4D flow CMR) can quantify blood flow in three dimensions and with time resolution. It provides key parameters such as wall shear stress, peak blood flow velocity, and effective valve area. This technique effectively reduces the rate of misdiagnosis and missed diagnosis in complex CHD (4-9). However, conventional 4D flow sequences have a long scan time while patients often have poor tolerance during examination. However, conventional 4D flow sequences have a long scan time, which patients often tolerate poorly (e.g., difficulty breath-holding, increased motion artifacts, or early scan termination). In addition, image artifacts can easily occur due to respiratory motion and involuntary body movement. These artifacts further affect the accuracy and reliability of hemodynamic parameter measurements (2,5). Compressed sensing (CS) technology is based on undersampling acquisition and iterative reconstruction algorithms. It can significantly shorten MRI scan time while maintaining image quality. This effectively addresses one of the major clinical limitations of conventional 4D flow sequences—the long scan time.

Recently, numerous studies have explored the feasibility of CS 4D flow MRI across various clinical scenarios. Neuhaus et al. demonstrated that six- to eight-fold acceleration of thoracic aortic 4D flow MRI is feasible in healthy subjects using acceleration factors ranging from 4 to 10 (10). Varga-Szemes et al. compared intracardiac flow measurements between conventional and CS-based highly accelerated 4D flow acquisitions, confirming that analysis of intracardiac flow patterns and diastolic function is feasible using the CS prototype, despite an approximately 10% underestimation of flow measurements. Notably, CS significantly reduced acquisition time compared with conventional accelerated acquisition (6.7 vs. 12.0 min) (11). Net forward flow for all valves showed good correlation (r>0.81) and agreement, with the CS technique underestimating flow by 3.3–8.3%. Evaluation of diastolic function exhibited errors of 3.2–17.6%. These findings support the application of CS 4D flow in intracardiac flow assessment, including valvular regurgitation. Panayiotou et al. tested the feasibility of CS 4D flow in healthy neonates without sedation, showing excellent inter- and intra-observer agreement and quantification comparable to validated two-dimensional phase-contrast (2D PC) imaging (12). Ma et al. validated CS 4D flow in patients with aortic disease, achieving a two-minute scan with less than five minutes for inline reconstruction, although they observed systematic underestimation of some hemodynamic parameters by the CS sequence (13). Blanken et al. confirmed the ability of CS sequences to assess valvular flow in patients with valve regurgitation, and Kilinc et al. provided validation data in aortic dissection, a specific aortic disease entity (14,15). In parallel, Vincenti et al. applied compressed sensing to single-breath-hold cardiac cine imaging to quantify left ventricular (LV) function, volumes, and mass. Their findings demonstrate that CS can not only accelerate flow acquisitions but also enable fast and accurate functional assessment, highlighting its broader clinical potential for comprehensive cardiac evaluation (16). Sodhi et al. reported the first clinical validation of CS 4D flow in children and adolescents with congenital or acquired heart disease (n=27). They found that CS sequences were equivalent to conventional sequences in measuring net flow and forward flow in the aorta and main pulmonary artery, but peak systolic velocity was significantly underestimated (aorta: 114 vs. 128 cm/s; main pulmonary artery: 106 vs. 112 cm/s). This study laid an important foundation for the application of CS technology in CHD (17).

However, previous studies have limitations in the scope of vascular assessment. Sodhi et al. only evaluated two vascular segments—the ascending aorta (AA) and main pulmonary artery (17). In CHD patients, however, blood flow abnormalities often involve a wider range of vessels, including the left and right pulmonary arteries (LPA and RPA), superior vena cava (SVC), and inferior vena cava (IVC). Assessment of these segments is crucial for comprehensive evaluation of conditions such as repaired tetralogy of Fallot, pulmonary atresia, and anomalous cava connections.

This study aims to investigate and validate the application value and clinical advantages of CS 4D flow sequences in the hemodynamic evaluation of patients with CHD. The core innovation of our study lies in its comprehensive assessment and unique patient population. We systematically evaluated the performance of CS 4D flow in six core vascular segments [AA, descending aorta (DA), LPA, RPA, SVC, and IVC], making this one of the most extensive CS 4D flow validation studies in the CHD population to date. Moreover, we validated this technique in a real-world adult/adolescent CHD cohort (mean age 20 ± 8 years), complementing the existing literature that has predominantly focused on pediatric patients. We present this article in accordance with the STARD reporting checklist (available at https://rx.amegroups.com/article/view/10.21037/rx-26-0034/rc).

Methods

Study population

We prospectively enrolled 14 consecutive patients with CHD who underwent CMR examination at our hospital between August 2024 and September 2025 (6 males; mean age 19.5±7.9 years). Selective enrollment based on subjective preferences of the researchers was not applied. Figure 1 presents the study flow diagram for participant enrollment. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the institutional review board of Shandong Second Provincial General Hospital (2025-002-01). Written informed consent was obtained from the patients (or their parents or legal guardians of children under 18 years old). Exclusion criteria were: (I) age <8 years; (II) no new CMR sequence collection; (III) inability to cooperate with breath-holding or poor electrocardiographic (ECG) gating; (IV) poor image quality for new sequence validation; (V) failed new sequence acquisition or presence of serious artifacts (e.g., motion, metal artifacts); (VI) missing clinical or reference standard data; and (VII) inability to obtain reliable flow parameters from conventional or CS sequences during post-processing due to technical issues, including low signal-to-noise ratio and failed vessel segmentation.

Figure 1 Study flow diagram for participant enrollment. 4D, four-dimensional; CMR, cardiovascular magnetic resonance imaging; ECG, electrocardiographic.

Data acquisitions and magnetic resonance imaging (MRI) protocol

All MRI scans were performed on a 3.0 T scanner (MAGNETOM Prisma; Siemens Healthcare, Erlangen, Germany). Each patient underwent consecutive examinations using both the conventional and CS accelerated 4D flow sequences within the same session. The scan coverage fully included the heart, SVC and IVC and pulmonary artery. The parameters of the two sequences are summarized in Table 1.

The conventional 4D flow MRI sequence served as the reference standard (2,4,10). Respiratory motion was prospectively gated using diaphragm navigator echoes to ensure measurement accuracy and image stability. This sequence employed GRAPPA parallel imaging with an acceleration factor of 3 in the phase-encoding direction and 1 in the three-dimensional direction. It has long been recognized as a standard for cardiovascular flow quantification and is widely used for great vessel hemodynamic analysis in patients with CHD (8,18,19).

The CS accelerated 4D flow MRI sequence achieves rapid imaging based on compressed sensing principles (the reconstruction workflow is shown in Table S1), with a total acceleration factor of 7.6 (WIP license option corresponding to WIP1106 from VE11). This technique also used respiratory navigation but relied on signal sparsity and random sampling to overcome the full-sampling limitation of conventional sequences. During k space acquisition, a pseudo-random undersampling pattern was adopted to substantially reduce raw data and shorten acquisition time (20). In the subsequent data reconstruction phase, undersampled data were accurately reconstructed using an optimized iterative reconstruction algorithm under sparse constraints and data consistency constraints to restore complete blood flow imaging data. While maintaining spatial resolution, flow signal contrast and measurement accuracy, this sequence achieved a substantial improvement in scan speed and effectively overcomes the clinical drawbacks of conventional sequences such as long scan time and poor patient tolerance. After each sequence scan was completed in all subjects, the actual scan time was accurately recorded for both sequences.

CMR analysis

All imaging data obtained from scanning were imported into Circle cvi42 (v.5.13, Circle Cardiovascular Imaging, Calgary, Alberta, Canada), a dedicated post-processing software for cardiovascular magnetic resonance, for subsequent quantitative analysis. To avoid measurement bias and ensure objective results, a double-blind design was used. Two attending radiologists with many years of experience in cardiovascular magnetic resonance image analysis (J.L. and C.T., with 12 and 8 years of CMR experience, respectively), who were completely unaware of the scan sequence type, independently performed quantitative measurements of all hemodynamic parameters. The measurement regions covered six core vascular segments including the AA, DA, LPA, RPA, SVC and IVC (8,19). The measured parameters included total flow volume, peak velocity, regurgitant fraction, time average axial wall shear stress (TAaWSS) and time average circumferential wall shear stress (TAcWSS). Peak velocity was defined as the temporal-spatial maximum velocity magnitude [|V| = √(Vx² + Vy² + Vz²), unit: cm/s] at any spatial location within the analyzed vessel segment and at any time point within the cardiac cycle, representing the maximum blood flow velocity in that segment over the entire cardiac cycle. TAWSS was calculated as the arithmetic mean of wall shear stress values across all time frames, representing the average level of wall shear stress during the cardiac cycle. Regurgitant fraction was derived by integrating the flow-time curve, which inherently relies on temporal information throughout the cardiac cycle. All parameter measurements were performed strictly according to a unified standard operating procedure to ensure consistent measurement positioning, analysis thresholds and other conditions.

Statistical analysis

Data analysis was performed using SPSSAU 20.0 (an online statistical tool) (21). The inter-physician consistency of hemodynamic parameter measurements at each anatomical site for both conventional and CS 4D flow sequences was evaluated using intraclass correlation coefficient (ICC) (22). Specifically, ICC >0.90 indicated excellent agreement; 0.75–0.90, good agreement; 0.50–<0.75, moderate agreement; and <0.50, poor agreement. For ICC >0.75, the senior physician’s measurement was used as the representative value; for 0.5< ICC ≤0.75, the average of the two physicians was reported; for ICC <0.5, the parameter is excluded from sequence comparisons. To compare the hemodynamic differences between the conventional and CS 4D flow sequences, and to assess their agreement, the limits of agreement (LoA), coefficient of repeatability (CR), and P value were used. Normally distributed data were expressed as mean ± standard deviation, and the above metrics were derived from Bland-Altman descriptive statistics. Non-normally distributed data were analyzed using non-parametric Bland-Altman analysis, with data expressed as median (interquartile range, IQR). LoA was based on the 2.5th and 97.5th percentiles of the differences, CR was defined as the difference between the 97.5th and 2.5th percentiles of the differences, and the P value was obtained from a paired sample Wilcoxon test.

All tests were two-tailed, and a P value <0.05 was considered statistically significant. Effect size (Cohen’s d for paired t-tests or the r value for Wilcoxon tests) were calculated to quantify the magnitude of the difference (23). To assess the agreement between peak flow velocities measured by the conventional and CS sequences, Pearson correlation analysis was performed. The Pearson correlation coefficient (r) and its 95% confidence interval were calculated, and an r value >0.80 was considered to indicate a strong correlation.

Results

Study population and scan feasibility

Table 2 showed the patient demographic and clinical characteristics. All 14 enrolled patients (6 males; mean age 19.5±7.9 years) tolerance both sequences well and successfully completed conventional and CS accelerated 4D flow examinations without interruption or failure. A total of 28 valid scans were completed (one per sequence per patient). All images were of sufficient quality for clinical and postprocessing analysis, with no significant motion artifacts, respiratory artifacts, or signal loss affecting data measurement. Subsequent post-processing was completed smoothly, and no data were lost or unavailable for quantitative analysis.

Scan time

The mean scan time of the conventional 4D flow sequence was 9.8±6.4 minutes, whereas that of the CS-accelerated sequence was 3.1±0.5 minutes (see Figure S1 for specific time data). The CS sequence reduced the mean scan time by 68.4% compared with the conventional sequence, and the difference was statistically significant (t=12.26, P<0.001, Cohen’s d =3.277).

Interobserver agreement

Table 3 summarizes the inter-observer agreement for the conventional 4D flow sequence. For total flow volume, regurgitant fraction, and most TAcWSS parameters, the ICC exceeded 0.75 (range, 0.82–0.99), indicating excellent agreement. Moderate agreement (ICC 0.50–0.75) was observed for peak velocity in DA (ICC =0.70), SVC (ICC =0.75), and IVC (ICC =0.61); for TAaWSS in the DA (ICC =0.72), SVC (ICC =0.66), and IVC (ICC =0.65); and for TAcWSS in IVC (ICC =0.73). All measured parameters had an ICC >0.50, and no parameter was excluded from subsequent sequence comparisons.

Agreement between conventional and CSaccelerated 4D flow sequences

Table 4 presents the comparisons between the two sequences. Across most vessel segments and hemodynamic parameters, the conventional and CS sequences showed good agreement, with no statistically significant differences (all P≥0.07 except for the few parameters noted below). Specifically, in LPA and RPA, SVC, and IVC, all five parameters (total flow volume, peak velocity, regurgitant fraction, TAaWSS, and TAcWSS) were not significantly different between the two sequences (all P≥0.15). Similarly, in DA, peak velocity (P=0.07) and regurgitant fraction (P=0.86) showed no significant differences. In the AA, total flow volume (P=0.48) and regurgitant fraction (P=0.72) also showed no significant differences.

Notably, statistically significant differences were observed for a limited set of parameters. In the AA, the CS sequence yielded significantly lower peak velocity than the conventional sequence (100.35±24.01 vs. 122.92±32.23 cm/s, P=0.001). In DA, mean axial WSS was significantly lower with the CS sequence (0.20±0.09 vs. 0.26±0.08 Pa, P=0.01). Mean circumferential WSS was also significantly lower with the CS sequence in both the AA [median (IQR): 0.06 (0–0.1) vs. 0.06 (0.05–0.11), P=0.03] and DA [0.04 (0–0.1) vs. 0.06 (0–0.1), P=0.03]. For all other comparisons across all sites and parameters, no statistically significant differences were observed (P≥0.07). Post-processing imaging results of the two sequences are shown in Figure 2. Figure 3 presents the consistency analysis of peak velocity in AA, LPA and RPA, in which all R-values were greater than 0.8.

Figure 2 The image quality from post-processing results of two sequences. (A) The conventional 4D flow. (B) The CS 4D flow sequence. Each group contains peak velocity at different sites. 4D, four-dimensional; AA, ascending aorta; CS, compressed sensing; MPA, main pulmonary arteries; IVC, inferior vena cava.

Figure 3 Consistency analysis of peak velocity in ascending aorta and left and right pulmonary artery. (A) Scatter plot. Horizontal axis: peak velocity of ascending aorta, left and right pulmonary artery in conventional 4D flow group; Vertical axis: the corresponding flow velocity in compressed sensing 4D flow group. (B) Bland-Altman plot. Horizontal axis: the average of two sets of data; Vertical axis: the measurement difference. Upper limit of agreement is 59.065, 24.898 and 57.959. 4D, four-dimensional; SD, standard deviation.

Discussion

This study demonstrated that the CS-accelerated 4D flow sequence reduced scan time by 68.4% compared with the conventional sequence while maintaining reliable measurement accuracy and consistency. This advantage is particularly valuable in CHD patients, especially young children, who often have poor compliance and impaired cardiopulmonary function (12,24). Long scan times are associated with a higher likelihood of patient movement, which can induce image artifacts, compromise the accuracy of data measurements, and may even lead to examination failure due to poor patient tolerance. Prolonged scan times increase the risk of motion artifacts, compromise data accuracy, and may lead to examination failure. By shortening the scan duration, the CS sequence reduces the examination burden, lowers motion artifact risk, and improves scan success rate and image quality—findings consistent with previous studies (25-28). Additionally, shorter scan time increase MRI throughput and reduce t clinical waiting lists, facilitating efficient workflows.

Accurate hemodynamic measurement is critical for CHD severity evaluation, treatment planning, and prognosis prediction. In this study, most hemodynamic parameters showed excellent inter-sequence agreement, with ICC >0.90 for most parameters, indicating that the CS sequence provides comparable accuracy to the conventional sequence (21,28). However, significant differences were observed in a limited set: peak velocity in AA, TAaWSS in DA, and TAcWSS in both the AA and DA (P=0.001, 0.01, 0.03, and 0.03, respectively). The AA, DA, and bilateral pulmonary arteries are the primary regions affected by hemodynamic abnormalities in CHD, making consistent measurement in these vessels particularly important.

Several factors may explain the observed differences. For peak velocity in the AA, the CS sequence reduces spatiotemporal resolution to achieve acceleration, and its iterative reconstruction may smooth early-systolic high-velocity jets. The AA, having the highest velocities, is most susceptible to this effect, whereas lower-velocity regions (pulmonary arteries, cava veins) did not show significant differences. For WSS (AA and DA), the parameter is highly sensitive to spatial resolution and edge detection. The undersampling and regularization-induced smoothing inherent to CS reconstruction reduce the near-wall velocity gradient, leading to systematic underestimation—consistent with Pathrose et al., who reported WSS as the most markedly underestimated parameter. The absence of significant differences in other sites (e.g., pulmonary arteries, cava veins) may be attributed to their low-velocity, thin-walled nature (especially in Fontan patients), where small absolute WSS values render any CS-induced bias potentially overwhelmed by inter-individual variability (29).

Importantly, statistical significance (P<0.05) does not directly imply clinical importance. Published data indicate that the inter-observer variability of 4D flow MRI is typically within 10% (4,28,30). In this study, the magnitude of differences between CS and conventional sequences (e.g., ~18% for AA peak velocity but with wide LoA; absolute WSS differences of 0.01–0.06 Pa) generally falls within this range. Furthermore, we observed that parameters with only moderate inter-observer agreement (e.g., peak velocity in DA and IVC, axial WSS in DA, SVC and IVC; ICC 0.61–0.75)—which may be influenced by respiratory variation and edge detection challenges—coincided with those showing the largest inter-sequence differences. Thus, the clinical relevance of these statistical differences is likely limited, and part of the observed discrepancy may stem from inherent measurement variability rather than systematic bias of the CS sequence. Given the ~7.6-fold reduction in scan time, CS 4D Flow remains a practical option for clinical screening and follow-up.

From a technical perspective, CS employs sparse sampling and iterative reconstruction algorithms to substantially reduce scan data while preserving key image information, thereby balancing shorter scan time and acceptable imaging quality (10,24,25). Quantitative comparisons from the literature support these trade-offs. Ma et al. achieved >70% scan time reduction with an acceleration factor of 7.7 but observed significant underestimation of peak velocities (9.4–16.2%) and peak flow (2.9–11.6%) (13). Neuhaus et al. systematically evaluated acceleration factors from 4 to 10 and found that 6- to 8-fold acceleration (scan time reduction ~50–70%) was feasible, albeit with trends toward increased artifacts and decreased sharpness, with peak velocity differences reaching –7.9±13.0 cm/s at R=8 (10). In all cases ultimately included in our study, images met clinical and post-processing requirements without significant artifacts, which may be attributable to the CS reconstruction algorithm—rooted in a nonlinear iterative reconstruction within a SENSE-type framework (incorporating coil sensitivity maps) combined with spatial and temporal regularization—effectively suppressing aliasing artifacts from sparse sampling. The clinical significance of this study lies in providing a more efficient and feasible method for hemodynamic evaluation in complex CHD patients. Although conventional 4D flow sequence is considered a reference standard, its long scan time limits its application in young children and critically ill patients. The CS accelerated sequence significantly shortens scan time while maintaining measurement accuracy, improving examination feasibility and safety, particularly suitable for patients requiring repeated examinations and dynamic monitoring of hemodynamic changes, such as postoperative follow-up and disease progression assessment (8,25).

This study also has several limitations. First, the sample size was relatively small (n=14), with an age range concentrated between 8 and 29 years. Larger and more diverse cohorts are needed to further validate CS sequence applicability. Second, only static hemodynamic parameters were evaluated; consistency in dynamic flow changes across the cardiac cycle was not analyzed and warrants future investigation. Third, objective image quality indicators (e.g., resolution, signal-to-noise ratio) were not compared; only subjective judgment was used to determine whether images met post-processing requirements. Finally, although no significant differences were observed in the LPA and RPA, the LoA for the RPA were wider than for the LPA, likely reflecting greater anatomical variability and more complex flow patterns at this site in CHD patients. Nevertheless, the CS sequence demonstrated acceptable equivalence to the conventional sequence for flow measurements in the pulmonary artery branches.

Conclusions

In clinical practice, even in patients with complex cardiac anatomy, this accelerated sequence has the potential to the conventional one. It improves inspection efficiency, and shortens scanning time with no detectable loss of accuracy within this study. The CS accelerated 4D flow sequence can serve as a promising alternative for evaluating aortic and pulmonary hemodynamic parameters in patients with CHD, with significant clinical value and broad application prospects.

Acknowledgments

We would like to express our gratitude to Dr. Juncheng Jiang for his assistance during the initial stage of writing the paper. The authors acknowledge the use of DeepSeek in language polishing and grammar checking during the revision of this manuscript. The AI was used to check grammar and improve clarity without generating core scientific content, data, or conclusions. All authors have reviewed and take responsibility for the content, accuracy, and originality of the manuscript.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://rx.amegroups.com/article/view/10.21037/rx-26-0034/rc

Data Sharing Statement: Available at https://rx.amegroups.com/article/view/10.21037/rx-26-0034/dss

Peer Review File: Available at https://rx.amegroups.com/article/view/10.21037/rx-26-0034/prf

Funding: This work was supported by the Natural Science Foundation of Shandong Province (No. ZR2024QA110); and the Shandong Province Medical Health Science and Technology Project (Nos. 202425020256 and 202403010254).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://rx.amegroups.com/article/view/10.21037/rx-26-0034/coif). J.X. is a current employee of Magnetic Resonance Research Collaboration, Siemens Healthineers Ltd., Shanghai, China. D.G. is a current employee of Magnetic Resonance Cardiovascular Team, Siemens Healthcare GmbH, Erlangen, Germany. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the institutional review board of Shandong Second Provincial General Hospital (2025-002-01). Written informed consent was obtained from the patients (or their parents or legal guardians of children under 18 years old).

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Rizk J. 4D flow MRI applications in congenital heart disease. Eur Radiol 2021;31:1160-74. [Crossref] [PubMed]
2. Jacobs K, Hahn L, Horowitz M, et al. Hemodynamic Assessment of Structural Heart Disease Using 4D Flow MRI: How We Do It. AJR Am J Roentgenol 2021;217:1322-32. [Crossref] [PubMed]
3. Demirkiran A, van Ooij P, Westenberg JJM, et al. Clinical intra-cardiac 4D flow CMR: acquisition, analysis, and clinical applications. Eur Heart J Cardiovasc Imaging 2022;23:154-65. [Crossref] [PubMed]
4. Bissell MM, Raimondi F, Ait Ali L, et al. 4D Flow cardiovascular magnetic resonance consensus statement: 2023 update. J Cardiovasc Magn Reson 2023;25:40. [Crossref] [PubMed]
5. Urmeneta Ulloa J, Martínez de Vega V, Molina Borao I, et al. 4D-Flow Cardiac Magnetic Resonance Imaging: An 8-Year Clinical Practice Review. Rev Cardiovasc Med 2026;27:46999. [Crossref] [PubMed]
6. Reiter G, Reiter C, Ovcina I, et al. Four-dimensional Flow MRI for a Dynamic Perspective on the Heart and Adjacent Great Vessels. Radiology 2025;316:e242972. [Crossref] [PubMed]
7. Grafton-Clarke C, Assadi H, Li R, et al. Four-dimensional flow provides incremental diagnostic value over echocardiography in aortic stenosis. Open Heart 2025;12:e003081. [Crossref] [PubMed]
8. Connor B, Takei M, Clark DE, et al. Improved quantification of aortic regurgitation with direct regurgitant jet measurement by four-dimensional flow cardiovascular magnetic resonance in complex congenital heart disease. J Cardiovasc Magn Reson 2025;27:101876. [Crossref] [PubMed]
9. Oyama-Manabe N, Aikawa T, Tsuneta S, et al. Clinical Applications of 4D Flow MR Imaging in Aortic Valvular and Congenital Heart Disease. Magn Reson Med Sci 2022;21:319-26. [Crossref] [PubMed]
10. Neuhaus E, Weiss K, Bastkowski R, et al. Accelerated aortic 4D flow cardiovascular magnetic resonance using compressed sensing: applicability, validation and clinical integration. J Cardiovasc Magn Reson 2019;21:65. [Crossref] [PubMed]
11. Varga-Szemes A, Halfmann M, Schoepf UJ, et al. Highly Accelerated Compressed-Sensing 4D Flow for Intracardiac Flow Assessment. J Magn Reson Imaging 2023;58:496-507. [Crossref] [PubMed]
12. Panayiotou HR, Mills LK, Broadbent DA, et al. Comprehensive Neonatal Cardiac, Feed and Wrap, Non-contrast, Non-sedated, Free-breathing Compressed Sensing 4D Flow MRI Assessment. J Magn Reson Imaging 2023;57:789-99. [Crossref] [PubMed]
13. Ma LE, Markl M, Chow K, et al. Aortic 4D flow MRI in 2 minutes using compressed sensing, respiratory controlled adaptive k-space reordering, and inline reconstruction. Magn Reson Med 2019;81:3675-90. [Crossref] [PubMed]
14. Blanken CPS, Gottwald LM, Westenberg JJM, et al. Whole-Heart 4D Flow MRI for Evaluation of Normal and Regurgitant Valvular Flow: A Quantitative Comparison Between Pseudo-Spiral Sampling and EPI Readout. J Magn Reson Imaging 2022;55:1120-30. [Crossref] [PubMed]
15. Kilinc O, Chu S, Baraboo J, et al. Hemodynamic Evaluation of Type B Aortic Dissection Using Compressed Sensing Accelerated 4D Flow MRI. J Magn Reson Imaging 2023;57:1752-63. [Crossref] [PubMed]
16. Vincenti G, Monney P, Chaptinel J, et al. Compressed sensing single-breath-hold CMR for fast quantification of LV function, volumes, and mass. JACC Cardiovasc Imaging 2014;7:882-92. [Crossref] [PubMed]
17. Sodhi A, Markl M, Popescu AR, et al. Highly accelerated compressed sensing 4D flow MRI in congenital and acquired heart disease: comparison of aorta and main pulmonary artery flow parameters with conventional 4D flow MRI in children and young adults. Pediatr Radiol 2023;53:2597-607. [Crossref] [PubMed]
18. Valdeolmillos E, Sakhi H, Tortigue M, et al. 4D flow cardiac MRI to assess pulmonary blood flow in patients with pulmonary arterial hypertension associated with congenital heart disease. Diagn Interv Imaging 2024;105:266-72. [Crossref] [PubMed]
19. Zamani-Aliabadi SM, Qanadli SD, Fatehi-Feyzabad SH, et al. Assessment of 4D flow MRI for quantification of left-to-right shunt in pediatric patients with ventricular septal defect: comparison with right heart catheterization. Front Cardiovasc Med 2024;11:1399110. [Crossref] [PubMed]
20. Basha TA, Akçakaya M, Goddu B, et al. Accelerated three-dimensional cine phase contrast imaging using randomly undersampled echo planar imaging with compressed sensing reconstruction. NMR Biomed 2015;28:30-9. [Crossref] [PubMed]
21. The SPSSAU project (2024). SPSSAU (Version 24.0) [Online application software]. Available online: https://www.spssau.com.
22. Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med 2016;15:155-63. [Crossref] [PubMed]
23. Fritz CO, Morris PE, Richler JJ. Effect size estimates: current use, calculations, and interpretation. J Exp Psychol Gen 2012;141:2-18. [Crossref] [PubMed]
24. Béland V, Goolaub DS, Portnoy S, et al. Rapid slice-to-volume four-dimensional flow in pediatric congenital heart disease: a feasibility study. J Cardiovasc Magn Reson 2025;27:101887. [Crossref] [PubMed]
25. Dong G, Rhee J, Kumar SJ, et al. Longitudinal Live Imaging-Derived 4D Hemodynamics and Dynamic Tissue Mechanics Across Outflow Tract Morphogenesis. Ann Biomed Eng 2026;54:2129-47. [Crossref] [PubMed]
26. Dirbach F, Tenisch E, Prša M, et al. Comparison of free-running whole-heart five-dimensional and four-dimensional flow imaging to standard two-dimensional flow in patients with congenital heart disease. J Cardiovasc Magn Reson 2026;28:102713. [Epub ahead of print]. [Crossref] [PubMed]
27. Lichthardt D, Schmidt M, Wetzl J, et al. Coronary Artery Blood Flow Imaging Using 3D Flow MRI. Magn Reson Med 2026;95:1545-59. [Crossref] [PubMed]
28. Burkhardt BEU, Kellenberger CJ, Callaghan FM, et al. Flow evaluation software for four-dimensional flow MRI: a reliability and validation study. Radiol Med 2023;128:1225-35. [Crossref] [PubMed]
29. Pathrose A, Ma L, Berhane H, et al. Highly accelerated aortic 4D flow MRI using compressed sensing: Performance at different acceleration factors in patients with aortic disease. Magn Reson Med 2021;85:2174-87. [Crossref] [PubMed]
30. Gottwald LM, Peper ES, Zhang Q, et al. Pseudo-spiral sampling and compressed sensing reconstruction provides flexibility of temporal resolution in accelerated aortic 4D flow MRI: A comparison with k-t principal component analysis. NMR Biomed 2020;33:e4255. [Crossref] [PubMed]