Evaluation of Valvular Insufficiency and Shunts with Parallel-imaging Compressed-sensing 4D Phase-contrast MR Imaging with Stereoscopic 3D Velocity-fusion Volume-rendered Visualization

Four-dimensional phase-contrast MR imaging may enable a more comprehensive cardiac examination by not only facilitating quantification of blood flow, but also enabling morphologic characterization of valvular insufficiency and shunts without additional imaging time.

Abstract

Purpose:

To assess the potential of compressed-sensing parallel-imaging four-dimensional (4D) phase-contrast magnetic resonance (MR) imaging and specialized imaging software in the evaluation of valvular insufficiency and intracardiac shunts in patients with congenital heart disease.

Materials and Methods:

Institutional review board approval was obtained for this HIPAA-compliant study. Thirty-four consecutive retrospectively identified patients in whom a compressed-sensing parallel-imaging 4D phase-contrast sequence was performed as part of routine clinical cardiac MR imaging between March 2010 and August 2011 and who had undergone echocardiography were included. Multiplanar, volume-rendered, and stereoscopic three-dimensional velocity-fusion visualization algorithms were developed and implemented in Java and OpenGL. Two radiologists independently reviewed 4D phase-contrast studies for each of 34 patients (mean age, 6 years; age range, 10 months to 21 years) and tabulated visible shunts and valvular regurgitation. These results were compared with color Doppler echocardiographic and cardiac MR imaging reports, which were generated without 4D phase-contrast visualization. Cohen κ statistics were computed to assess interobserver agreement and agreement with echocardiographic results.

Results:

The 4D phase-contrast acquisitions were performed, on average, in less than 10 minutes. Among 123 valves seen in 34 4D phase-contrast studies, 29 regurgitant valves were identified, with good agreement between observers (k = 0.85). There was also good agreement with the presence of at least mild regurgitation at echocardiography (observer 1, κ = 0.76; observer 2, κ = 0.77) with high sensitivity (observer 1, 75%; observer 2, 82%) and specificity (observer 1, 97%; observer 2, 95%) relative to the reference standard. Eight intracardiac shunts were identified, four of which were not visible with conventional cardiac MR imaging but were detected with echocardiography. No intracardiac shunts were found with echocardiography alone.

Conclusion:

With velocity-fusion visualization, the compressed-sensing parallel-imaging 4D phase-contrast sequence can augment conventional cardiac MR imaging by improving sensitivity for and depiction of hemodynamically significant shunts and valvular regurgitation.

© RSNA, 2012

Supplemental material: http://radiology.rsna.org/lookup/suppl/doi:10.1148/radiol.12120055/-/DC1

Introduction

Patients with congenital structural heart disease often exhibit complex pre- and postsurgical anatomy, which is typically evaluated by using a combination of imaging modalities—namely, echocardiography, magnetic resonance (MR) imaging, computed tomography, and conventional angiography. Cardiac MR imaging is increasingly essential for many of these patients, particularly because of (a) its proven quantitative accuracy and reproducibility in the evaluation of ventricular size and blood flow (1–5) and (b) its ability to delineate extracardiac structures without ionizing radiation. At present, this is usually accomplished by using a series of predominantly planar pulse sequences, including two-dimensional (2D) cine phase-contrast and 2D cine balanced steady-state free precession sequences. The exact combination of pulse sequences is often customized to match the clinical indication, which requires multiple localizers and sequentially prescribed acquisitions to achieve each of the necessary imaging planes.

Thus, congenital heart MR examinations are labor and time intensive, requiring highly trained imagers who are familiar with structural heart disease and its surgical management to directly supervise image acquisition. Each of these factors contributes to the cost and resources required for MR imaging. To reduce examination time, clinical MR imaging examinations are tailored to address diagnostic questions that have been incompletely addressed at echocardiography. As such, these examinations often rely on the ability of an imager to review cardiac echocardiograms to identify intracardiac shunts and insufficient valves, which can be further interrogated with MR imaging, if necessary. Thus, the role of MR imaging has not been to detect lesions but to quantify their hemodynamic significance. However, in situations where sonographic (US) windows do not permit adequate visualization or when echocardiographic findings are inconclusive, MR imaging may be called on to better delineate the nature of suspected shunts and valvular leaks to guide surgical decision making. While quantitative analysis routinely is used indirectly to characterize these lesions, direct visualization of the morphology of shunts and valvular leaks can be difficult with the planar pulse sequences that are typically used.

Four-dimensional (4D) phase-contrast MR imaging has the potential to address some of the limitations of conventional cardiac MR imaging by simultaneously capturing a volume of time-varying anatomic data and a vector field of motion, which can be used to evaluate blood flow (6). However, until the recent adaptation of parallel imaging, long acquisition times have prohibited the clinical use or thorough investigation of this pulse sequence, which previously required 20–30 minutes of imaging time unless spatial resolution or field of view were sacrificed. The combined compressed-sensing and parallel-imaging technique provides new opportunities to shorten acquisition time without the sacrifice in signal-to-noise ratio typically observed with high acceleration factors (7) and recently has been applied to 4D phase-contrast MR imaging (8). In addition, newly available blood pool contrast agents can further improve signal-to-noise ratio, thereby enabling higher acceleration. The spatial resolution and image quality of this data can be sufficient for quantification of ventricular volume, with accuracy and precision comparable with those of steady-state free precession imaging (8). Furthermore, with specialized software, a few groups have shown that the quantitative accuracy of 4D phase-contrast MR imaging in the measurement of blood flow is comparable with that of the clinical standard, 2D phase-contrast MR imaging (9), and that in certain circumstances, its precision may exceed that of 2D phase-contrast MR imaging (10,11).

In addition to its quantitative value, we hypothesize that the 4D phase-contrast acquisition may hold diagnostic information beyond the pulse sequences and imaging planes typically used in cardiac MR imaging. At present, it is both time consuming and technically challenging to thoroughly evaluate the function of all valves and check for intracardiac shunts with conventional cardiac MR imaging pulse sequences. Since 4D phase-contrast data contain temporally resolved hemodynamic information throughout the entire imaging volume, it may be possible with appropriate software to investigate the function of each valve at the workstation well after MR imaging data have been acquired. It may also be possible to investigate the integrity of the atrial and ventricular septum and to identify shunts. However, 4D phase-contrast data are challenging to interpret in the clinical environment with currently available methods of image review due to the sheer volume and high dimensionality of these data. Thus, we sought to assess the potential of compressed-sensing parallel-imaging 4D phase-contrast MR imaging and specialized visualization software to evaluate for valvular insufficiency and intracardiac shunts in patients with congenital heart disease.

Materials and Methods

We obtained Stanford University institutional review board approval for this Health Insurance Portability and Accountability Act–compliant study, and we retrospectively identified 34 consecutive patients in whom compressed-sensing 4D phase-contrast MR imaging was performed as part of a routine clinical cardiac MR examination between March 2010 and August 2011 and who had also undergone in-house echocardiography. In these patients, comprehensive two-dimensional transthoracic color Doppler echocardiography was performed to evaluate valvular flow patterns at each valve, evaluate cardiac function, and estimate pressure gradients. All patients who had undergone 4D phase-contrast MR imaging and in-house echocardiography were included. No additional selection or exclusion criteria were used. Patient characteristics are provided in Table E1 (online). Patients ranged in age from 10 months to 21 years (mean age, 6 years); body surface area, determined with the Mosteller equation, ranged from 0.45 m² to 1.90 m² (mean, 0.86 m²). Heart rates also spanned a wide range, stretching from 36 to 139 beats per minute (mean, 84 beats per minute).

All MR imaging was performed with a 1.5-T TwinSpeed MR imager with an eight-channel phased array cardiac coil (GE Healthcare, Milwaukee, Wis), a 150 T/msec maximum slew rate, 40 mT/min gradients, and vector echocardiographic gating. For the conventional portion of the examination, imaging planes were prescribed by radiologists with 5, 7 (S.S.V.), and more than 10 years of experience in pediatric cardiovascular MR imaging. Congenital heart MR examinations were tailored to the clinical indication and typically included multiple 8–10-mm 2D phase-contrast imaging planes at the outflow valves and at any additional sites of interest noted at the time of image acquisition. These were acquired with a gradient-echo sequence (FastCard; GE Healthcare) with four to 10 views per segment, depending on patient heart rate. Parallel imaging was not used. Steady-state free precession images were acquired as axial short-axis cine left- and right-sided two-, three-, and four-chamber views, with section thickness of 6–8 mm depending on patient size, and a 45° flip angle. For these latter views, single signal average breath-hold acquisitions were used in patients capable of breath holding. Otherwise, two or three signal average free-breathing acquisitions were used to reduce respiratory artifacts. Contrast material–enhanced MR angiograpy was performed with an in-house three-dimensional spoiled gradient-echo sequence that enabled both parallel-imaging and compressed-sensing reconstruction. By excluding the 4D phase-contrast component of the examination, cardiac MR imaging was performed in 24–103 minutes (mean, 41 minutes).

Each 4D phase-contrast acquisition was performed after contrast-enhanced MR angiography, which was performed with intravenous gadofosveset to enhance signal-to-noise ratio and support higher acceleration. The 4D phase-contrast MR imaging was performed by using a spoiled gradient-echo–based sequence with tetrahedral flow encoding and variable-density Poisson disk k-space undersampling (7,12), with total acceleration factors ranging from 1.4 × 1.4 to 2.2 × 2.2. Images were reconstructed for each cardiac temporal phase separately with a combined autocalibrating parallel-imaging compressed-sensing algorithm (L₁-SPIRiT) (7,13). Compressed sensing was implemented to take advantage of per-section 2D spatial sparsity, without enforcing temporal sparsity. K-space phase reordering was used for respiratory compensation (EXORCIST; GE Healthcare). A flip angle of 15° and two to four tetrahedral encodes per segment were used. Velocity-encoding parameters were selected to avoid aliasing and ranged from 120 to 350 cm/sec. Average spatial resolution was 1.02 × 1.34 × 2.30 mm and ranged from 0.78 to 1.37 mm in the right-to-left direction, 1.04 to 1.82 mm in the anteroposterior direction, and 1.20 to 3.40 mm in the superior-to-inferior direction. Mean repetition time and echo time were 4.8 and 1.9 msec, respectively. Temporal resolution ranged from 31 to 86 msec (mean, 61 msec). Acquisition time ranged from 5 to 15 minutes (mean, 9 minutes 43 seconds). Image reconstruction was performed with a general-purpose computing on graphics processing unit implementation of the aforementioned combined autocalibrating parallel-imaging compressed-sensing algorithm L₁-SPIRIT (14) at a 64-bit Linux workstation equipped with four Tesla C1060 graphics cards (NVIDIA, Santa Clara, Calif). Image data were corrected for Maxwell phase effects (15), encoding errors related to gradient field distortions (16), and eddy current–related phase offsets (17).

Multiplanar, volume-rendered, and stereoscopic three-dimensional (3D) velocity-fusion visualization algorithms were developed and implemented by the primary author (A.H.) in Java and OpenGL. For anatomic data, average intensity, maximal intensity projection, and surface-rendering algorithms were implemented by using a ray-casting approach and summing the contribution of each voxel along rays traced from the user’s perspective through the imaging volume. In the volume-rendering mode, the default slab thickness was set to 20 mm; however, this was dynamically adjustable at the time of image review. For velocity data, adjustable image filters were created to enable manual suppression of low-signal and relatively noisy low-velocity data. This was facilitated by the creation of two manual slider bars; the first was created to adjust a minimum magnitude signal threshold, and the second was created to adjust a minimum magnitude times speed product threshold. Velocity data were superimposed on anatomic data by using α blending. The penetration and conspicuity of high-velocity components was amplified over low-velocity components by weighting the opacity of α blending by the fourth power of the speed. Additional user controls were created to adjust interocular distance for stereoscopic 3D visualization. To enable real-time interactive navigation, implementation of each of these techniques was optimized for use with a workstation equipped with a Quadro 6000 graphics card (NVIDIA), a 3D monitor, and stereoscopic 3D active shutter glasses.

Two radiologists (A.H., S.S.V.) with 3 and 7 years of experience in the interpretation of congenital heart disease MR examinations independently reviewed each parallel-imaging compressed-sensing 4D phase-contrast study. For the sake of simplicity, each observer tabulated the presence or absence of valvular regurgitation without grading severity. Each observer also tabulated the presence of any intra- or extracardiac shunts. The custom software facilitated real-time interactive manual navigation of the imaging volume with multiplanar reformatted planes and volume-rendered slabs, enabling users to navigate into two- and three-chamber views of each ventricle, as well as into the four-chamber view. The stereoscopic 3D component of the software was used by each radiologist for the duration of image interpretation. The 4D phase-contrast images were reviewed solely with the computer system described previously and without awareness of prior history or of the results of other portions of cardiac MR imaging or echocardiography. These results were subsequently compared with color Doppler echocardiographic results and cardiac MR imaging reports, which were generated without the benefit of the 4D phase-contrast visualization tools.

Statistical Analysis

Data were analyzed with Excel 2003 software (Microsoft, Redmond, Wash) with in-house statistical macros. Cohen κ statistics were calculated to assess interobserver agreement in the identification of valvular regurgitation with 4D phase-contrast imaging. Cohen κ statistics were also calculated to assess agreement between observers for echocardiography. To assess intermodality agreement with echocardiography, the presence of valvular insufficiency at echocardiography was defined by the following two alternate minimum thresholds: at least mild regurgitation or more than mild regurgitation.

Results

Custom visualization software facilitated evaluation of the 34 consecutive parallel-imaging compressed-sensing 4D phase-contrast studies, providing real-time interactive navigation. Average frame rates during image review at 1920 × 1080-pixel monitor resolution exceeded 40 frames per second. Image quality of anatomic data and baseline velocity noise were noticeably improved at compressed-sensing reconstruction when compared with those at conventional parallel-imaging reconstruction, as shown by Figure E1 (online). For the first observer (A.H.) and developer of the software, mean interpretation time to assess each of the valves and to identify intra- or extracardiac shunts was 3 minutes. For the second observer (S.S.V.), mean interpretation time was approximately 8 minutes.

Valvular Insufficiency

Of the 123 valves investigated with the 4D phase-contrast examinations, including those in six patients with single-ventricle physiology, a total of 29 regurgitant valves were identified by at least one observer (Table E2 [online]) with a high degree of agreement between observers (k = 0.85). All but one of the regurgitant valves identified by the first observer were also found by the second more-experienced observer. Five regurgitant valves were detected by only the second observer. Altogether, valves with detectable insufficiency with this approach included five aortic valves, 11 pulmonary valves, three mitral valves, and 10 tricuspid valves. Typical cases of valvular regurgitation seen on 4D phase-contrast images are shown in Figure 1 (Movies 1–4 [online])

Figure 1:

Examples of valvular regurgitation in four different patients examined with parallel-imaging compressed-sensing 4D phase-contrast MR imaging and velocity-overlay slab volume rendering. A, Diastolic regurgitant jet from the aortic valve (arrow) is seen. B, Small jet from the mitral valve (arrow) is seen during systole. C, Free pulmonary regurgitation (arrow) is seen in a patient with repaired tetralogy of Fallot. D, Arrow points to one of two jets of tricuspid regurgitation.

In addition to high interobserver consistency, 4D phase-contrast findings of both observers were consistent with echocardiographic findings, irrespective of which minimum threshold was used (Table E3 [online]). With a threshold of at least mild regurgitation at echocardiography, there was good agreement between modalities (observer 1, κ = 0.76; observer 2, κ = 0.77). For this echocardiographic level of valvular insufficiency, both observers had considerable sensitivity (observer 1, 75%; observer 2, 82%) and high specificity (observer 1, 97%; observer 2, 95%) with the 4D phase-contrast technique. With a threshold of more than mild regurgitation at echocardiography, there was again good agreement between 4D phase-contrast imaging and echocardiography (observer 1, κ = 0.69; observer 2, κ = 0.61). At this level of valvular insufficiency, the 4D phase-contrast technique was both highly sensitive (100% for both observers) and highly specific (observer 1, 91%; observer 2, 87%).

There were only a few cases in which 4D phase-contrast findings and echocardiographic findings were not completely concordant. One pulmonary valve and two additional tricuspid valves had detectable insufficiency on 4D phase-contrast images that was seen by both observers; however, these valves were thought to have only trace regurgitation at echocardiography. Similarly, one mitral valve, two tricuspid valves, and one pulmonary valve were found to have mild regurgitation at cardiac echocardiography, but these were not detected by either observer at 4D phase-contrast imaging. In the first case, an aortic regurgitant jet was redirected posteriorly toward the valve and mimicked mitral regurgitation, but the mitral valve itself appeared competent at 4D phase-contrast imaging. In both cases of tricuspid regurgitation, tiny systolic jets were seen retrospectively across the tricuspid valve at 4D phase-contrast imaging. Initially, these may have been obscured by baseline velocity noise. In the last case, in a patient with repaired tetralogy of Fallot, the pulmonary regurgitant jet was visible in retrospect with 4D phase-contrast imaging, but it was oriented eccentrically, which possibly contributed to the initial failure to detect this lesion.

Intracardiac Shunts

Intracardiac shunts were also assessed by using the velocity-fusion volume-rendered visualization approach. Among the 34 patients in the study population, eight had echocardiographic evidence of intracardiac shunts. This included a patent foramen ovale, four atrial septal defects (ASDs), a ventricular septal defect (VSD), and two postsurgical leaks (Table). All of these lesions were identified with the 4D phase-contrast technique by both observers without any prior knowledge (Fig 2) (Movies 5–8 [online]). No additional intracardiac shunts were detected with either echocardiography or 4D phase-contrast imaging alone. In contrast, four of these eight intracardiac shunts could not be seen with conventional MR imaging sequences, including an ASD, a patent foramen ovale, an aortic baffle leak, and a VSD patch leak, each of which will be detailed further.

Characteristics of Intracardiac Shunts and Surgical Leaks Identified in the Study Population

Note.—Shunt fractions (Q_p/Q_s) are provided; however, several patients had multiple shunts. ASD = atrial septal defect, ECG = echocardiography, NA = not applicable, PAPVR = partial anomalous pulmonary venous return, TGA = transposition of the great arteries, VSD = ventricular septal defect.

Figure 2a:

Intracardiac shunts depicted with parallel-imaging compressed-sensing 4D phase-contrast MR imaging and velocity-overlay slab volume rendering. (a) A left-to-right shunt is seen across an ostium secundum ASD (arrow). (b) Blood is seen traversing a muscular VSD (arrow). (c) A residual VSD (arrow) is detected just below the sinuses of Valsalva at the superior margin of a VSD patch, shunting blood into the right ventricle. (d) Blood is shunted into the right ventricle through a leak (arrow) in an intraventricular aortic baffle.

Figure 2b:

Intracardiac shunts depicted with parallel-imaging compressed-sensing 4D phase-contrast MR imaging and velocity-overlay slab volume rendering. (a) A left-to-right shunt is seen across an ostium secundum ASD (arrow). (b) Blood is seen traversing a muscular VSD (arrow). (c) A residual VSD (arrow) is detected just below the sinuses of Valsalva at the superior margin of a VSD patch, shunting blood into the right ventricle. (d) Blood is shunted into the right ventricle through a leak (arrow) in an intraventricular aortic baffle.

In one such case in a patient with partial cor triatriatum, a 5-mm ostium secundum ASD was readily evident at 4D phase-contrast imaging and echocardiography, but it was difficult to see at conventional MR imaging. Since no 2D phase-contrast data were available, a shunt fraction (Q_p/Q_s, where Q_p is pulmonary blood flow and Q_s is systemic blood flow) of 1.3 was retrospectively computed from 4D phase-contrast data by measuring the ratio of blood flow at the pulmonary and aortic valves, as reported in earlier works (8,10). Because of the relatively small shunt fraction attributed to the ASD, no surgical intervention was planned immediately. In a second case, an abnormal jet of left-to-right flow was readily seen between the atria with 4D phase-contrast imaging and was identified by both observers. The echocardiographic findings were interpreted as a left-to-right shunt through a patent foramen ovale.

Postsurgical Leaks

A total of three postsurgical leaks were unexpectedly identified in the 34 clinical examinations included in this study. Two of these behaved as intracardiac shunts, as mentioned previously, while a third was functionally equivalent to mitral regurgitation. While the conventional planar 2D phase-contrast and steady-state free precession acquisitions were useful in the indirect quantification of the magnitude of the intracardiac shunts based on shunt fraction (Q_p/Q_s), the precise location of the shunts was not depicted. With 4D phase-contrast imaging, each postsurgical leak was readily visualized.

In the first case, in a patient with a prior repair of pulmonary atresia with VSD, an apparent surgical leak was identified at echocardiography, and the patient was referred for further evaluation with MR imaging. At 2D phase-contrast imaging, a negligible shunt fraction was attributed to the lesion, but the site of the leak itself could not be visualized. This surgical leak was independently identified at 4D phase-contrast imaging by both observers (Fig 2c); however, interpretation of this research data was not available in time to guide surgical management. The patient did undergo surgical repair on the basis of echocardiographic findings, which enabled us to confirm a small residual VSD at the margin of a VSD patch.

Figure 2c:

Intracardiac shunts depicted with parallel-imaging compressed-sensing 4D phase-contrast MR imaging and velocity-overlay slab volume rendering. (a) A left-to-right shunt is seen across an ostium secundum ASD (arrow). (b) Blood is seen traversing a muscular VSD (arrow). (c) A residual VSD (arrow) is detected just below the sinuses of Valsalva at the superior margin of a VSD patch, shunting blood into the right ventricle. (d) Blood is shunted into the right ventricle through a leak (arrow) in an intraventricular aortic baffle.

In the second case, a patient with dextrocardia, cor triatriatum, congenitally corrected transposition of the great arteries, and a double-outlet right ventricle who had previously undergone a Rastelli procedure and a hemi-Mustard atrial switch procedure was referred for cardiac MR imaging to evaluate a possible residual VSD seen at echocardiography. This was undetectable with the multiple steady-state free precession sequence and the 2D phase-contrast imaging planes used in the conventional MR examination. The 4D phase-contrast study was evaluated shortly thereafter; it showed a 5-mm leak from within an intraventricular aortic baffle that was originally placed to redirect blood flow from the morphologic left ventricle through a VSD into the aorta (Fig 2d). The 4D phase-contrast findings helped to guide surgical intervention, which enabled us to confirm the presence of a leak within the baffle itself. The baffle was removed and reconstructed.

Figure 2d:

Intracardiac shunts depicted with parallel-imaging compressed-sensing 4D phase-contrast MR imaging and velocity-overlay slab volume rendering. (a) A left-to-right shunt is seen across an ostium secundum ASD (arrow). (b) Blood is seen traversing a muscular VSD (arrow). (c) A residual VSD (arrow) is detected just below the sinuses of Valsalva at the superior margin of a VSD patch, shunting blood into the right ventricle. (d) Blood is shunted into the right ventricle through a leak (arrow) in an intraventricular aortic baffle.

In the third case, a 6-year-old boy had previously undergone repair for a complete atrioventricular canal defect at an outside institution. On the basis of echocardiograms obtained at the outside institution, it was thought that he had worsening mitral regurgitation with unexplained abnormal flow in the coronary sinus, and he was referred for cardiac MR imaging for further evaluation. Conventional MR images showed abnormal flow in a dilated coronary sinus, but it was not immediately clear what was causing this. Upon review of the 4D phase-contrast study, it became clear that an eccentric jet of blood was being ejected into the coronary sinus during systole around the mitral valve. Since the coronary sinus was kept on the left side of the atrioventricular patch during the original surgical repair, this behaved effectively as an inlet valve leak. Intraoperatively, surgeons identified a 1-cm linear tear in the mitral valve adjacent to the coronary sinus that was responsible for this leak.

Extracardiac Shunts

By using 4D phase-contrast sequences, both observers were able to identify each of the extracardiac shunts. These included five bidirectional Glenn shunts and two cases of partial anomalous pulmonary venous return (Fig 3). Both cases of partially anomalous pulmonary venous return were suspected but were not confirmed prior to cardiac MR imaging, and all of the Glenn shunts were previously known from each patient’s clinical history. No additional extracardiac shunts were identified with cardiac MR imaging or echocardiography.

Figure 3a:

(a, b) Extracardiac shunts depicted with parallel-imaging compressed-sensing 4D phase-contrast MR imaging and velocity-overlay slab volume rendering. Anomalous venous drainage (arrows) is seen entering the superior vena cava from multiple right-sided pulmonary veins. In b, blood flow from left-sided pulmonary veins traverses the left atrium and a sinus venosus ASD to enter the right atrium. A left-sided superior vena cava is also present.

Figure 3b:

(a, b) Extracardiac shunts depicted with parallel-imaging compressed-sensing 4D phase-contrast MR imaging and velocity-overlay slab volume rendering. Anomalous venous drainage (arrows) is seen entering the superior vena cava from multiple right-sided pulmonary veins. In b, blood flow from left-sided pulmonary veins traverses the left atrium and a sinus venosus ASD to enter the right atrium. A left-sided superior vena cava is also present.

Discussion

The real-time interactive stereoscopic 3D velocity-fusion volume-rendering technique is an effective method in the clinical evaluation of high-dimensional 4D phase-contrast MR imaging data. This visualization approach may facilitate more reliable detection and characterization of intracardiac shunts and regurgitation with MR imaging. In a small population of patients with congenital heart disease, we showed that these rendering methods for 4D phase-contrast data showed sensitivity comparable to that of echocardiography. Once identified, the same 4D phase-contrast data can be used to quantify the severity of insufficiency at each valve (9–11). Extracardiac shunts were also readily delineated with the 4D phase-contrast sequence; however, the number of cases encountered in this study was relatively small. Finally, we found that intracardiac shunts could also be identified with the 4D phase-contrast technique; however, in the few cases we encountered, the exact nature of these abnormalities was better characterized with the 4D phase-contrast technique than with echocardiography.

As acquisition techniques and visualization approaches for the 4D phase-contrast technique continue to evolve, it may become increasingly important to be able to characterize the morphology of valvular dysfunction to plan valve-sparing repairs. While we have not specifically addressed this point here, there is unique potential for MR imaging to play a role in surgical planning. For example, it may be possible to guide the surgical approach to tricuspid annuloplasty by assessing the morphology of the regurgitant jets. Future studies may be required to investigate this clinical indication further. We have also limited the scope of our study to pediatric patients with congenital heart defects who we thought would most benefit from direct MR imaging of structural abnormalities. These imaging approaches may also be applicable to adult patients with heart disease. For example, in patients with aortic and cerebrovascular disease, earlier works have explored pathline and streamline rendering of 4D phase-contrast data (18–21). In the past, the application of these rendering techniques to intracardiac shunts has been hampered by limited signal-to-noise ratio in the velocity data. With reduced baseline velocity noise that compressed-sensing and intravascular blood pool agents may now provide, it is possible that these rendering methods may be useful in the visualization of intracardiac shunts.

Although we used a blood pool contrast agent to improve the signal-to-noise ratio and facilitate high-acceleration factors, it is also possible to perform the acquisition with conventional contrast agents or even without intravenous contrast material. In this study, however, we did not specifically address image quality in the absence of blood pool contrast agents. In our experience, similar studies with conventional contrast agents resulted in weaker signal-to-noise ratio in both the anatomic data and the velocity data. It is possible that this could reduce the sensitivity and specificity for small shunts and lesser degrees of valvular insufficiency. Additional studies may be needed to address whether similar diagnostic accuracy for valvular insufficiency and shunts could be obtained with conventional agents.

For the sake of simplicity, we did not qualitatively grade the severity of valvular insufficiency in this study. We chose instead to tabulate the binary decision of whether a convincing regurgitant jet could be identified by each observer. It is certainly possible to qualitatively grade the severity of regurgitation; however, observers might require additional training to provide consistent grading. However, since 4D phase-contrast data are fundamentally quantitative, it is also possible to directly quantify the severity of regurgitation at each valve. Further studies should be performed to validate the accuracy of regurgitant fraction and volume once a leaky valve has been identified. We also found that it is possible to directly visualize and quantify the severity of valvular stenoses with the 4D phase-contrast technique; however, this is not specifically addressed in the current work. Future investigations should be directed toward the incremental-value 4D phase-contrast technique in the care of patients with inlet and outlet valve stenoses.

One limitation of the rendering method that we have developed and evaluated here is the lack of directionality information of the velocity overlay. For this study, each reader began to recognize that the directionality of flow could be readily inferred based on the phase of the cardiac cycle. In most cases, this was sufficient to identify inlet or outlet regurgitation, as well as most shunts, but it could have been aided by explicit directional information. One approach is to implement a virtual US probe and color the velocity data in a manner similar to color Doppler US, as suggested in an earlier work (22). This may be a useful technique with which to further investigate the flow field strength and supplement the rendering methods presented herein. Use of a vector arrow overlay may be another alternative approach, if it can be implemented in a manner that does not impair real-time navigation.

In this study, we applied compressed sensing only by exploiting sparsity of 2D sections separately. This was necessary to limit computational complexity and allow for complete image reconstruction within 2 hours of data acquisition. We have shown with this 4D phase-contrast implementation that it is already possible to perform qualitative evaluation of all valves and shunts, in addition to each of the quantitative assessments, with a single acquisition. Continued progress with compressed-sensing methods has the potential to further improve image quality without additional imaging time. By enforcing sparsity in time and space, even higher spatial or temporal resolution and/or higher velocity dynamic range might be possible.

Although it is possible to attempt full characterization of any shunts and to evaluate each of the valves with conventional cardiac MR imaging pulse sequences, such as the 2D phase-contrast sequence, this is too time consuming in practice. To do this would require immediate interpretation and direct physician-guided investigation at the time of image acquisition. This kind of comprehensive analysis would otherwise be too costly to perform. The findings presented here suggest the clinical potential of the 4D phase-contrast sequence for a more time-efficient comprehensive cardiac examination with MR imaging. Use of 4D phase-contrast MR imaging may enable a more comprehensive cardiac examination not only by facilitating quantification of blood flow, but also by enabling morphologic characterization of valvular insufficiency and shunts without additional imaging time. With shorter overall acquisition times than those required for the sequential steady-state free precession and 2D phase-contrast imaging planes, routine use of the 4D phase-contrast sequence may also reduce total time of cardiac MR imaging and thus the depth and duration of cardiac anesthesia required for younger pediatric patients. This may help increase the accessibility of cardiac MR imaging and lead to broader applicability to a wider range of patients and clinical indications.

Advances in Knowledge.

• • Combined parallel imaging and compressed sensing enables the clinically feasible acquisition of four-dimensional (4D) phase-contrast MR images with whole chest coverage and near-isotropic millimeter resolution, on average, in less than 10 minutes.

• • Stereoscopic three-dimensional velocity fusion volume rendering is a feasible method for real-time navigated visualization and interpretation of 4D phase-contrast MR imaging data.

• • The stereoscopic 3D velocity-fusion volume-rendering 4D phase-contrast sequence is feasible in the detection of valvular insufficiency at all cardiac valves with high interobserver agreement (k = 0.85; sensitivity, 75% and 82% for two radiologists) and high agreement with echocardiography.

• • Stereoscopic 3D velocity-fusion volume-rendering 4D phase-contrast MR imaging is a promising approach in the detection and evaluation of intracardiac shunts.

Implication for Patient Care.

• • A compressed-sensing parallel-imaging 4D phase-contrast acquisition may augment the conventional cardiac MR imaging examination by improving the sensitivity for and visualization of hemodynamically significant shunts and valvular regurgitation.

Disclosures of Potential Conflicts of Interest: A.H. No potential conflicts of interest to disclose. M.L. Financial activities related to the present article: institution received a grant from GE Healthcare. Financial activities not related to the present article: institution received a grant from GE Healthcare, receives royalties for patents from Stanford University. Other relationships: none to disclose. M.J.M. No potential conflicts of interest to disclose. M.T.A. Financial activities related to the present article: institution received a grant from GE Medical. Financial activities not related to the present article: institution received a grant from GE Medical. Other relationships: none to disclose. S.V.S. Financial activities related to the present article: institution received time on a research MR imager to enable sequence development from GE Healthcare, iinstitution received a computer and graphics cards from NVIDIA. Financial activities not related to the present article: none to disclose. Other relationships: none to disclose.