Multidimensional fetal flow imaging with cardiovascular magnetic resonance: a feasibility study
To image multidimensional flow in fetuses using golden-angle radial phase contrast cardiovascular magnetic resonance (PC-CMR) with motion correction and retrospective gating.
A novel PC-CMR method was developed using an ungated golden-angle radial acquisition with continuously incremented velocity encoding. Healthy subjects (n = 5, 27 ± 3 years, males) and pregnant females (n = 5, 34 ± 2 weeks gestation) were imaged at 3 T using the proposed sequence. Real-time reconstructions were first performed for retrospective motion correction and cardiac gating (using metric optimized gating, MOG). CINE reconstructions of multidimensional flow were then performed using the corrected and gated data.
In adults, flows obtained using the proposed method agreed strongly with those obtained using a conventionally gated Cartesian acquisition. Across the five adults, bias and limits of agreement were − 1.0 cm/s and [− 5.1, 3.2] cm/s for mean velocities and − 1.1 cm/s and [− 6.5, 4.3] cm/s for peak velocities. Temporal correlation between corresponding waveforms was also high (R~ 0.98). Calculated timing errors between MOG and pulse-gating RR intervals were low (~ 20 ms). First insights into multidimensional fetal blood flows were achieved. Inter-subject consistency in fetal descending aortic flows (n = 3) was strong with an average velocity of 27.1 ± 0.4 cm/s, peak systolic velocity of 70.0 ± 1.8 cm/s and an intra-class correlation coefficient of 0.95 between the velocity waveforms. In one fetal case, high flow waveform reproducibility was demonstrated in the ascending aorta (R = 0.97) and main pulmonary artery (R = 0.99).
Multidimensional PC-CMR of fetal flow was developed and validated, incorporating retrospective motion compensation and cardiac gating. Using this method, the first quantification and visualization of multidimensional fetal blood flow was achieved using CMR.
KeywordsFetal Golden angle radial Motion correction Retrospective gating Compressed sensing Phase contrast MRI
congenital heart disease
cardiovascular magnetic resonance
gradient recalled echo
intraclass correlation coefficient
metric optimised gating
main pulmonary artery
phase contrast cardiovascular magnetic resonance
Pearson correlation coefficient
region of interest
spatial total variation
temporal Fourier transform
temporal total variation
Disruption of the fetal circulation by congenital heart disease (CHD) can result in injury to critical organs and possible fetal death . In such cases, premature delivery with subsequent intervention may be recommended, but this option must be weighed against the risks of prematurity such as infection, impaired organ development and cognitive delay. Emerging treatments intended to improve the fetal circulation include drugs provided to the fetus through the maternal circulation, maternal oxygen supplementation, and percutaneous surgical correction of fetal cardiovascular anatomy [2, 3, 4]. Selecting the most appropriate therapy and monitoring its efficacy, however, requires an accurate assessment of the fetal circulation.
Fetal vascular and cardiac structures are small with complex morphology. To quantify flow accurately in the heart and major vessels, a high spatial resolution (≤ 1 × 1 mm2) is beneficial.
Fetal heart rates are high (~ 110–180 bpm), requiring acquisitions at high temporal resolution. Based on adult studies using PC-CMR, a minimum of 15 cardiac phases are required to resolve blood flow, which corresponds to a temporal resolution of approximately 30 ms in fetal applications .
Motion from maternal respiration or gross fetal movements will corrupt portions of the data. Thus, the fetal PC-CMR acquisition and analysis must monitor and correct for these motions.
Conventional electrocardiogram (ECG) or peripheral pulse gating methods cannot be used to synchronize fetal data acquisition and reconstruction . Thus, a method for extracting this signal non-invasively is needed.
Previous PC-CMR studies of fetal blood flow have included Cartesian sampling with retrospective gating [9, 10, 11]. However, these were sensitive only to flow through the prescribed slice and did not correct for motion, often requiring repeat acquisitions to avoid image artifacts. Here, we propose, validate, and demonstrate the feasibility of a novel PC-CMR approach for measuring complex fetal hemodynamics that compensates for motion while meeting the spatiotemporal demands of fetal applications. This builds on previous work using radial golden-angle sampling with compressed sensing (CS) for fetal CMR [12, 13]. A strength of the golden-angle approach is its ability to provide real-time reconstructions for motion compensation and retrospective gating [12, 13]. We adopt a similar approach but incorporate multi-dimensional flow encoding for a single-slice acquisition, hereafter referred to as “single slice vector flow”. By acquiring all three velocity components, complex flows (such as those passing through the fetal heart) may be visualized and measured. The theory behind the sampling and reconstruction of such data is described in the next section, followed by an experimental validation of the method in a flow phantom and 5 healthy adult subjects. Finally, feasibility of the method is demonstrated in scans of complex blood flow through the fetal heart and major fetal vessels of 5 healthy pregnancies.
Golden-angle radial acquisitions provide reasonably uniform k-space coverage over flexible temporal windows, which enables image reconstruction at temporal resolutions appropriate for motion correction and image-based gating . Extending this approach to multi-dimensional PC-CMR, however, can dramatically reduce its temporal resolution. This limitation arises from having each direction of flow encoding acquired consecutively for a given radial spoke angle, such that the temporal resolution of the real-time reconstruction is four-fold lower than that of a similar anatomical scan. Using Ns spokes to reconstruct each real-time frame, the temporal resolution is 4·Ns·TR. For typical parameters (Ns~ 4, TR~ 7 ms), this leads to a temporal resolution of 112 ms, which is too long for fetal image-based gating.
Stage 1: Motion detection and correction
The first stage involves reconstruction of a real-time image series for retrospective detection and compensation of motion (Fig. 2b). For this purpose, we have found a real-time temporal resolution of approximately 200 ms is adequate for motion tracking. The reconstruction of such real-time frames is performed using CS, exploiting sparsity in the spatial and temporal total variation domains [15, 16]. CS is used here to remove undersampling artifacts that could impair motion detection. Data acquired during periods of through-plane motion (typically gross fetal movements) are discarded. This process uses a template matching approach, as follows. First, a reference frame is defined as the real-time frame with the highest mutual information with all other frames. Next, a region of interest (ROI) is placed within that reference frame around a target structure (e.g. fetal chest), and all other frames are registered to this ROI by rigid translation. The average mutual information between a frame and all others is then calculated over the ROI. This is repeated for each frame to identify outliers (frames with an average mutual information less than the mean of the distribution by 1.5× the interquartile range). These frames are deemed to be corrupted by through-plane motion and thus discarded from further analysis.
The remaining data are then corrected for in-plane displacements (typically maternal respiration), based on the ROI translations measured in the previous section. These ROI translations are interpolated to the repetition time (TR) of the original PC-CMR acquisition and used to motion-correct each spoke of the retained raw data (by applying the inverse phase offset corresponding to each translation) . In the proposed approach, only translational motions of fetal structures are considered since these are attributed to maternal respiration. Other types of motion (such as rotation, shear or dilations) of fetal structures are not considered since they are associated with gross fetal movements which include components of through-plane displacement (and are thus excluded from the reconstruction using the data rejection scheme described above).
Stage 2: Fetal cardiac gating
The second stage of the reconstruction involves retrospective detection of the fetal cardiac cycle (RR intervals) for subsequent grouping of data into cardiac phases. This involves a second real-time reconstruction, but at a higher temporal resolution for detection of high fetal heart rates. Using the motion compensated data from the previous stage, images are reconstructed at an effective temporal resolution of 25 ms (Fig. 2c). This results in high radial undersampling (R = 100), but which can again be reconstructed with CS using spatial total variation and temporal total variation sparsity.
Stage 3: CINE reconstruction
In this final stage, the calculated RR intervals are used to sort the motion compensated k-space data into cardiac phases, and a high-quality CINE PC-CMR reconstruction is performed (Fig. 2d). The CINE reconstruction uses CS with sparsity in the spatial total variation, temporal Fourier transform and complex difference domains. Images with periodic phenomena are sparse in the temporal Fourier domain, and flow-encoded images are sparse in the complex difference domain [18, 19]. These approaches have been previously shown to reconstruct cardiac time series images  and to quantify blood flow from highly undersampled data by Kwak et al. . Three separate CINE reconstructions are performed, corresponding to velocities x, y and z dimensions. Sparsity in reconstruction is minimized as follows. The complex difference is computed by performing a framewise-subtraction of the flow compensated reconstruction from the flow encoded reconstruction. The temporal Fourier and spatial total variation transforms are applied to the flow encoded and flow compensation reconstructions separately. As is typical of CS, consistency of the reconstruction is maintained by minimising the error between the multichannel data and the coil-sensitivity weighted, non-uniform fast Fourier transform of the reconstruction.
All experiments were performed using a commercial 3 T CMR system (PrismaFIT, Siemens Healthineers, Erlangen, Germany). The radial PC-CMR approach described in the previous section was first validated in a phantom experiment and 5 healthy adult subjects. Its feasibility was then demonstrated in 5 healthy pregnancies. Acquisitions were performed using a gradient recalled echo (GRE) sequence with a multi-channel receiver coil. All acquisitions were performed as part of an ethically approved study, and informed consent was obtained from all subjects. All reconstructions were performed in MATLAB (MathWorks, Natick, Massachusetts, USA), modifying code originally developed by Lustig et al. and Otazo et al., and using conjugate gradient descent for optimization during CS reconstruction [15, 16]. Computer specifications were: RAM 32GB, processor Intel® Core™ i7–6700 (3.40 GHz, 8 cores). NUFFT  was performed on a GPU (Nvidia Geforce GTX 960, 2GB and 1024 CUDA cores). Motion compensation was based on the normalized mutual information, calculated with elastix (Image Sciences Institute, University Medical Center Ultrecht, The Netherlands) .
PC-CMR acquisition protocol – Phantom
To validate the radial PC-CMR sequence, flow was measured in a phantom consisting of a computer-controlled pump connected to coiled tubing (10 mm diameter) positioned within the scanner. Flows measured using the proposed sampling strategy were compared to reference flows obtained using a conventional multi-dimensional flow encoded Cartesian PC-CMR sequence. Measurements were made along a double oblique plane to obtain velocity components along all three spatial dimensions. Constant flows (ranging from 10 to 60 mL/s) were measured in eight sections of the coiled tubing. Agreements between the radial and Cartesian measurements for the mean and peak spatial velocities across the cross-section of the tubing were quantified through linear regression for each velocity encoding dimension. Comparisons between radial and Cartesian PC-CMR measurements were made instead of validation against programmed flow for direct comparison of complex multidimensional flow vectors.
PC-CMR acquisition protocol – Adult
PCMR imaging parameters
A1, A2, A3, A4, A5
P1, P2, P3, P4, P5
Field of View (mm2)
256 × 256
256 × 256
1 × 1 × 4
1 × 1 × 4
Number of Flow Encodes
Spokes per Slice
4000 (1000 per encode)
3600–4800 (900–1200 per flow encode)
TR / TE (ms)
6.5 / 4.0
6.5 / 4.0
Scan Length (s)
During acquisition of the radial PC-CMR data, pulse gating waveforms were recorded for later comparison with MOG-based gating. Two CINEs were reconstructed from the radial data (one with MOG and the other with pulse gating) and measured velocities were compared. To test reproducibility of the flow measurements and gating performance, scans were repeated three times consecutively in each subject.
PC-CMR acquisition protocol – Fetal
Five healthy pregnant women (subject identifiers P1-P5; gestational ages 32–36 weeks) were recruited and scanned under free breathing conditions using the proposed PC-CMR approach. In three subjects, four-chamber views of the fetal heart were prescribed (subjects P1, P4 and P5). In the two other subjects, standard three vessel views of the fetus were prescribed (subjects P2 and P3). The imaging parameters are summarized in Table 1.
Data analysis – Adults
Adult flow data, acquired using the radial PC-CMR acquisition, were reconstructed using the pipeline described in the Theory section and depicted in Fig. 2, but without the need for gross motion correction (Fig. 2b).
First, real-time series with four consecutive spokes per frame were reconstructed using regularization weights of 0.05 and 0.001 for temporal and spatial total variation respectively (temporal resolution of 26 ms; acceleration of 100, Fig. 2c). These images were used to detect the RR intervals for each acquisition, using MOG with an ROI placed over the heart and surrounding vasculature. Based on these RR intervals, retrospective CINE reconstructions were performed using CS (regularization weights: 0.01 for spatial total variation, 0.01 for complex difference and 0.05 for temporal Fourier transform) with the minimisation procedure alternating between flow compensation and flow encoding (Fig. 2d). This resulted in single slice vector flow measurements that were then corrected for background phase .
The accuracy of the MOG-based reconstruction of radial PC-CMR data was evaluated based on two quantities. First, the retrospective gating signal obtained by MOG was compared with that from the recorded pulse gating log. Specifically, the timing error was defined as the standard deviation of the differences between the corresponding RR intervals from each method. Second, quantitative evaluation of the proposed sampling scheme was performed by running Wilcoxon signed rank test on and comparing Bland-Altman plots of the mean and the peak velocities between reconstructions from the MOG and the logged pulse-gated reconstructions. Agreement between the reconstructions was analyzed by looking at the interclass Pearson correlation coefficients (R) between the flow curves.
Data analysis – Fetal
Real-time frames for motion compensation were reconstructed using 32 consecutive spokes (temporal resolution of 208 ms; acceleration of 12) with CS (regularization weights: 0.05 for temporal total variation and 0.001 for spatial total variation). Motion compensation was performed on these real-time frames as described in the Theory. A second real-time reconstruction was then performed for gating, based on 4 consecutive spokes (temporal resolution of 26 ms; acceleration of 100) with CS (regularization weights: 0.05 for temporal total variation and 0.001 for spatial total variation). MOG was performed and the data was binned into 15 cardiac phases. The acceleration factor ranged between 5 and 12 in these final reconstructions, depending on the amount of data rejection due to through-plane motion. CINE reconstructions were performed using CS (regularization weights: 0.0025 for spatial total variation, 0.0025 for complex difference and 0.025 for temporal Fourier transform) with the minimisation procedure alternating between the flow compensation and flow encoding.
With the current pipeline, real-time reconstructions used for fetal motion compensation required approximately 5 min/slice and those used for gating required approximately 45 min/slice. Both reconstructions ran for 50 iterations. Image registration and MOG analysis each required approximately 10 min/slice. CINE reconstructions lasted approximately 30 min/slice with 80 iterations. Accounting for 2 min for manual ROI selections, the total processing time with this pipeline was approximately 102 min per slice.
Consistency in fetal reconstructions
As a simple test of the reproducibility of this acquisition and reconstruction, fetal data were acquired twice consecutively from one participant (P3). After reconstruction, pulsatile flows though the fetal aorta were measured. The consistency of these measurements was quantified based on the intraclass correlation coefficient (ICC) and the peak velocity error between the two flow curves.
PC-CMR validation – Phantom
PC-CMR validation – Adult
Table shows the MOG trigger timing error and mean Pearson correlation between pulse-gated and MOG derived flow curves for each subject
17 ± 3
25 ± 4
19 ± 4
17 ± 5
22 ± 4
Fetal motion compensation
Representative real-time fetal images used for motion compensation and MOG for subject P1 are shown in supporting Additional file 4: Video 4. Although the quality of these intermediate reconstructions is not diagnostic as a result of extremely high undersampling factors (13 and 100 respectively), they are sufficient for extracting translational motion and the gating signal, as demonstrated by the final high-quality CINE output from the pipeline (final frame of V4).
Fetal CINE PC-CMR
In this study, we devised an approach for quantifying multidimensional fetal blood flow. A golden angle radial sampling scheme with multidimensional flow encoding was developed whereby both the golden-angle radial trajectory and the velocity encoding direction were continuously updated in synchrony. Reconstruction of intermediate real-time images allowed for motion correction and gating, leading to CINE reconstruction and single slice vector flow maps at high spatial and temporal resolution. Validation of the proposed sequence was performed in a flow phantom and validation of the retrospective gating and CINE reconstruction was performed in healthy adult subjects. Feasibility of the approach was tested in healthy pregnancies. Through this approach, we have generated the first multidimensional PC-CMR visualizations of complex hemodynamics in the human fetal heart.
In clinical practice, ultrasound is the primary imaging modality for fetal anatomy and flow. However, ultrasound scanning is susceptible to artifact from acoustic shadowing, or low levels of amniotic fluid (oligohydramnios), particularly in late gestation. In cases where ultrasound examinations are insufficient, Fetal CMR provides complementary information which may improve management of at-risk pregnancies. Along with improved tissue contrast, CMR can measure fetal blood flow in arbitrary planes and quantify blood oxygen saturation and hematocrit . These measurements can provide a map of fetal flow and oxygen, helping us improve our understanding of how CHD affects fetal development, as well as the efficacy of CHD treatments.
A strength of the proposed approach is its ability to track motion in the real-time reconstructions and then correct for that motion in the acquired k-space data. Previous fetal PC-CMR studies have relied on lengthy Cartesian scans in which gross motion was evident only as significant artifact in the final images. When faced with obvious motion artifact, Cartesian scans are typically repeated. Our current work using radial PC-CMR demonstrated that periods of fetal through-plane motion can be identified and removed, and that residual in-plane motion from maternal breathing can be corrected. As a result, repeat measurements can be avoided, providing the vessel of interest remains in the slice plane for at least a portion of the scan.
Past implementations of fetal PC-CMR with MOG have used a simple two-parameter heart rate model. In this current work, a multiparameter MOG model was used instead, which accounts for variations in fetal heart rate throughout the acquisition. This multiparameter model was previously implemented and validated for fetal anatomical CMR . Here, the gating approach was validated for PC-CMR in adults and showed high accuracy when compared with pulse-gating RR intervals (timing error ~ 20 ms). Furthermore, flow curves obtained using this approach were highly correlated with those using pulse-gating (R = 0.98). Finally, fetal flows obtained using this approach were consistent with those reported in the literature using through-plane Cartesian PC-CMR, and dynamic fetal cardiac structures were well visualized [9, 10]. Alternative strategies for retrospective fetal cardiac gating have recently been published using self-gating [28, 29] and iterative outlier rejection with cardiac synchronisation , while prospective gating using an external ultrasound probe has also been demonstrated .
In the proposed approach, two real-time reconstructions were performed with one at lower temporal resolution for motion correction of the acquired data and the other at higher temporal resolution for gating of the motion corrected data. Another approach would have been to perform one real-time reconstruction of the acquired data at the higher temporal resolution, and then to average frames together to improve image quality for motion correction. However, since CS is a non-linear process, such averaging is corrupted by structural noise which reduces accuracy of the registration. Another consideration would have been to combine the MOG and the CINE reconstruction steps. This would have avoided the need to reconstruct real-time images at high temporal resolution. However, this approach would have required CS in each iteration of the MOG optimization, resulting in a significant computational burden and impractical reconstruction times.
In this work, ROIs were set manually and in general, the reconstruction is robust to small deviations in ROI placement provided that the fetal chest is encompassed for motion correction and gating. An alternative approach proposed by Demesmaeker et al. 2017, uses the frequency information derived from the temporal Fourier transform of real-time reconstructions to automatically detect an ROI containing the fetal heart in anatomical images using a balanced steady state free precessionsequence . Translating this approach to the PC-CMR sequence used in this work may alleviate the need for manual steps in the reconstruction, strengthening the clinical feasibility of our reconstruction pipeline.
Previous PC-CMR studies of fetal blood flow have used Cartesian sampling [9, 10], requiring relatively long scan times (~ 30 s for through-slice flows) at limited spatial resolution (1.25 × 1.25 × 5 mm3). The proposed radial PCMR approach provides much higher spatial resolution (1 × 1 × 4 mm3) and lower scan times, improving quantification and visualization of fetal flow by reducing partial volume effects. Further improvements may be achieved using more efficient sampling schemes, such as spiral PC-CMR . Regardless of the trajectory that is chosen, additional applications of the approach to multi-dimensional flow proposed in this work may include uncooperative postnatal subjects such as neonates or the elderly, where motion corruption can be problematic.
Despite the success of the proposed pipeline, there are limitations with fetal PC-CMR. First, our study was performed at 3 T whereas many clinical scanners are 1.5 T, and the reduced signal-to-noise ratio (SNR) at 1.5 T could affect the accuracy of the proposed pipeline. PCMR at earlier gestation demands even higher spatial resolution because fetal structures are less developed, further affecting SNR. Finally, coil coverage over the maternal abdomen will influence image quality, depending on fetal and maternal habitus. Collectively, these limitations will require designing coils targeted for pregnancy scans and re-optimising the parameters used in CS for the lower SNR.
In conclusion, we have proposed a novel radial sampling strategy for multidimensional PC-CMR. This approach provides real-time reconstructions, used to correct motion and retrospectively gate CINE reconstructions. Validation of the pipeline was performed in a flow phantom and healthy adult subjects, and its feasibility was tested in healthy pregnancies. This work resulted in the first quantification and visualisation of multidimensional fetal blood flow using PC-CMR, providing future tools for studying fetal hemodynamics, and new ways to manage patients with suspected fetal CHD.
Funded by Canadian Institutes of Health Research Grant PJT148505.
Availability of data and materials
The datasets analyzed during the current study are available from the corresponding author on reasonable request.
DSG prepared the manuscript, study design, imaging sequence, image reconstruction and analysis. CWR, ES and CKM contributed to the study design, image reconstruction and analysis. DM, MS and DS were involved in volunteer recruitment and imaging. All authors contributed to and approved the final manuscript.
Ethics approval and consent to participate
All subjects provided informed written consent, including permission to publish anonymized data, as part of a study approved by the Hospital for Sick Children institutional review board.
Consent for publication
Written consent and permission for publication was obtained from each participant in this study.
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
- 1.Sun L, Macgowan CK, Sled JG, Yoo S-J, Manlhiot C, Porayette P, Grosse-Wortmann L, Jaeggi E, McCrindle BW, Hickey E, et al. Reduced fetal cerebral oxygen consumption is associated with smaller brain size in fetuses with congenital heart disease. Circulation. 2015;137(10):1313–23.CrossRefGoogle Scholar
- 8.Sameni R, Clifford GD. A review of fetal ecg signal processing; issues and promising directions. The open pacing, electrophysiology & therapy journal. 2010;3:4.Google Scholar
- 10.Prsa M, Sun L, van Amerom J, Yoo S-J, Grosse-Wortmann L, Jaeggi E, Macgowan C, Seed M. Reference ranges of blood flow in the major vessels of the normal human fetal circulation at term by phase-contrast magnetic resonance imagingclinical perspective. Circ Cardiovasc Imaging. 2014;7(4):663–70.CrossRefGoogle Scholar
- 14.Feng L, Grimm R, Block KT, Chandarana H, Kim S, Xu J, Axel L, Sodickson DK, Otazo R. Golden-angle radial sparse parallel mri: combination of compressed sensing, parallel imaging, and golden-angle radial sampling for fast and flexible dynamic volumetric mri. Magn Reson Med. 2014;72(3):707–17.CrossRefGoogle Scholar
- 20.Knoll F, Schwarzl A, Diwoky C, Sodickson DK. Gpunufft-an open source gpu library for 3d regridding with direct matlab interface. In: Proceedings of the 22nd annual meeting of ISMRM, Milan, Italy; 2014. p. 4297.Google Scholar
- 23.Fadnes S, Nyrnes SA, Wigen M, Tegnander E, Lovstakken L. Detailed flow visualization in fetal and neonatal hearts using 2-d speckle tracking. In: Ultrasonics Symposium (IUS), 2016 IEEE International. Tours: IEEE; 2016. p. 1–4.Google Scholar
- 30.van Amerom JF, Lloyd DF, Price AN, Kuklisova Murgasova M, Aljabar P, Malik SJ, Lohezic M, Rutherford MA, Pushparajah K, Razavi R, et al. Fetal cardiac cine imaging using highly accelerated dynamic mri with retrospective motion correction and outlier rejection. Magn Reson Med. 2018;79(1):327–38.CrossRefGoogle Scholar
- 32.Demesmaeker R, Kober T, Yerly J, Chaptinel J, Prsa M, Mivelaz Y, Alamo L, Vial Y, Berchier G, Rohner C, et al. Automated heartbeat detection for self-gated fetal cardiac mri. Proc Int Soc Magn Reson Med; Honolulu, HI, USA. 2017. Abstract No 0633.Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.