Mitral valve regurgitation assessed by intraventricular CMR 4D-flow: a systematic review on the technological aspects and potential clinical applications

Baseline characteristics of the study cohorts, aim of the studies, publication year, and 4D-flow quantification methods are depicted in Table 1. The main objectives behind these studies were (1) to assess the accuracy and reproducibility of using 4D-flow CMR for quantifying MVR volume (n = 12, 67%), (2) to investigate the association of characteristics of the MVR jet with hemodynamic parameters (n = 3, 17%), and (3) to evaluate LV kinetic energy in patients with underlying cardiac disease and MVR (n = 3, 17%). Additionally, 11 studies (61%) compared patients with underlying cardiac disease and MVR to healthy volunteers for internal validity assessments. Across studies, underlying cardiac diseases such as mitral valve prolapse (MVP) [16], atrial fibrillation (AF) [17], and HCM [18] were included.

MVR volume quantification methods require the assessment of stroke volume (SV) either by volumetrically using cine CMR images or by calculation from phase-contrast data. Figure 2 summarizes all the MVR volume quantification methods. (1) The “4D-intraventricular annular inflow method” (4D-flow_AIM) calculates the regurgitant volume by subtracting the SV derived from aortic forward flow (SV_AAo) from the SV derived from the forward flow through the mitral valve (SV_MV), both derived from a single 4D-flow CMR dataset (available in n = 18 studies, 100%). The SV_AAo and SV_MV are calculated by integrating flows derived from the phase-contrast CMR images over the duration of one cardiac cycle. Additionally, (2) the clinical “2-dimensional phase-contrast standard method” (2D-PC_standard) is used to indirectly measure the MVR volume by subtracting the SV derived from PC imaging of SV_AAo from volumetrically assed LV SV from cine CMR images (n = 8 studies, 44%). The LV SV is calculated by subtracting LV end-diastolic volume (EDV) from LV end-systolic volume (ESV) as derived from short axis cine images of the heart. The remaining methods are (3) “the volumetric method”, which calculates the deviation of the LV SV and right ventricular SV from cine CMR images in 2 (11%) studies, and (4) the 4D-flow_jet method directly quantifying the flow and regurgitant volume of the regurgitant jet using 4D-flow CMR in 5 (28%) studies. No study assessed the MVR volume with (5) the “2D-PC mitral valve method” (2D-PC_MV), which quantifies the MVR volume by subtracting SV_MV from LV SV using 2D-PC and cine CMR images, analogous to the 2D-PC_standard method. It is important to note, that all quantification approaches, with the exception of the 2D-PC_standard method and 4D-flow_jet method, require adaptation when significant aortic regurgitation is present. The replacement of the SV of the ascending aorta (AAo) or aortic valve (AoV) by the “net forward flow” through the AAo or AoV (calculated as the SV minus the volume of aortic regurgitation) allows proper quantification of MVR in these cases. Additionally, it is important to note that these methods have limited utility when there is interventricular shunting.

Table 2 shows the technical parameters used in the reviewed studies. Scanners magnetic field strengths were 1.5 T (n = 11) and 3 T (n = 11). In all studies, the positioning of the FOV of the 4D-flow sequence was adapted to match a whole heart coverage, especially the entire left-sided cavities and the aortic root. The velocity encoding range (VENC) was set to values around 150 cm/s by default in most studies except in special cases such as congenital heart disease (CHD) [19]. The image resolution was ranging between 0.8 and 4.2 mm³, while most studies used a resolution of around 2.5 mm³, and a temporal resolution of around 40 ms (21–86 ms). Further acquisition parameters were as follows: echo time (TE) of 1–3 ms, repetition time (TR) of 5–15 ms, the flip angle was mostly 10° (7°–15°), and the mean image acquisition time was generally around 10 min (5–15 min). All studies administrated contrast agents before the 4D-flow acquisition, without specification of the exact timing, and used ECG triggering and respiratory gating. For flow analysis, retrospective valve tracking using post-processing software such as MASS (n = 6) (Leiden University Medical Center, The Netherlands) [9‐11, 17, 20, 21] and cvi42 (n = 2) (Circle Cardiovascular Imaging, Calgary, Canada) [8, 18] was common. All studies visually assessed the quality of images and performed pre-processing for de-noising and anti-aliasing.

Nine studies investigated the MVR quantification reproducibility of 4D-flow_AIM, 7 studies (78%) reported good to excellent intra- and inter-reader reproducibility (ICC > 0.8) (Table 3), and the remaining 2 studies described good to excellent intra- but only moderate inter-reader reproducibility [8, 22]. None of the included studies have investigated the inter- and intra-scan reproducibility of 4D-flow acquisition. Seven studies (39%) investigated the agreement of 4D-flow_AIM to other MVR acquisition methods [10, 16, 19, 20, 22‐24] (Table 4). Inter-modality correlation among the 4 quantification methods was heterogeneous across studies, ranging from moderate to excellent correlation (r > 0.51). In direct comparison to 2D-PC, 4D-flow_AIM measurements showed similar intra- and inter-observer agreement [16, 20]. Agreement of these techniques was also associated with the etiology of MVR. In primary MVR, a lower agreement (P > 0.05) was found compared to secondary MVR (P < 0.0001) [20]. When compared to the 2D-PC standard method, 4D-flow_jet provided higher MVR volumes (P < 0.05) [20]. Two studies compared 4D-flow_AIM to echocardiographic assessment of MVR volumes by the proximal isovelocity surface area (PISA) method with moderate correlation between the two modalities and systematically yielded higher MVR volumes as compared to CMR techniques (mean difference of 15.8 ml) [16].

Due to its widespread availability, simplicity, and affordability, echocardiography by visual assessment and PISA method, remains the most popular modality to evaluate MVR severity. However, echocardiography has some constraints such as variable velocity assessment caused by beam alignment with non-optimal flow convergence, dynamic changes in orifice, limited acoustic window and operator experience. Further, in cases of multiple regurgitant orifices the PISA method is limited. Additionally, when complex flow patterns or complex vessel geometries are present, the calculation of mean velocities and net flow is frequently based on assumptions about the vessel's cross-sectional area or flow profile, which can lead to inaccurate flow quantifications, especially as the regurgitant orifice is not round, but rather oval or irregular in shape [7]. As a result, estimated echo velocity values have a moderate correlation with CMR quantitative measurements. Moreover, among CMR 4D-flow quantification methods might provide additional information with higher reproducibility and robustness in borderline moderate to severe MVR.

2D-PC CMR has become the reference gold standard for clinical aortic forward and backward flow (regurgitation) quantifications because of its high spatial and temporal resolution, simplicity in acquisition and post-processing, and good prognostic and diagnostic outcome data [27]. However, when used for MVR analysis, 2D-PC overestimates the MVR volume by 15% when compared to 4D-flow_AIM [28] and is prone to errors because of the two different types of acquisition, 2D-PC and cine images [27]. Besides, concomitant valve disease might impact the accuracy of these measurements. Additionally, the 2D-PC imaging plane should be orthogonal to the flow direction, as stated by Vermes et al. in their study that the misalignment of the 2D-PC imaging plane prevents measuring the aortic peak velocity precisely and reduces the accuracy of flow measurements [29]. The CMR volumetric method based on one cine image acquisition allows a fast and easy assessment of MVR volumes and is a good method for quantifying solitary MVR. However, it is an indirect MVR quantification method, which has poor precision and high segmentation variability for right ventricle SV, and cannot be used in other valves incoherencies [27].

4D-flow CMR acquisitions allow for post-procedural adaptation of the angle and the position of the evaluation planes. 4D-flow has been used frequently for aortic diseases [30, 31], however, using the method in mitral valve disease is more complicated due to the saddle shape and significant through-plane motion of the mitral valve. To directly quantify the regurgitation jet volume with 4D-flow_jet, proper cine image acquisitions and retrospective valve tracking (RVT) are required. Another advantage of 4D-flow quantification methods is their ability to enable direct valve tracking throughout the cardiac cycle, which is not feasible with 2D-flow imaging due to the motion of the valve annulus. This direct measurement capability is a significant advantage for assessing mitral regurgitation and allows for high reproducibility that might be superior to that of 2D PC methods [13, 23]. Nevertheless, the preferable MVR quantification method by CMR still has to be determined by systematic comparisons of reproducibility and robustness in intra- and inter-reader variability. Moreover, kinetic energy and wall shear stress are some advanced novel 4D-flow intraventricular hemodynamic parameters. For example, Gupta et al. [18] reported that left atrial kinetic energy assessed by 4D-flow is associated with LV obstruction in HCM patients. Whether these novel parameters maybe of advantage and may provide additional information in MVR with a potential clinical impact has to be evaluated in the future. Furthermore, there is no gold-standard MVR grading system by 4D-flow CMR, and the cut-off values are usually decided by the experts at each center. The consensus statement on assessing MVR by CMR suggested a grading system presented in Table 5 [27], however, further studies are required to compare the cut-off values for different quantification methods directly with outcomes.

Across the reviewed studies, several limitations of 4D-flow CMR require attention, such as long acquisition time [11], using static time-averaged cine images for segmentations [8, 9, 11, 16, 18, 19, 26], difficulties in capturing the exact position of the peak MVR jet [10, 18, 19, 22], low temporal resolution in comparison to other CMR sequences, such as cine bSSFP [8, 20, 32], and the presence of image artifacts in patients with implanted devices [12].

Segmenting 4D-flow images based on time-averaged cine images requires an extra acquisition leading to misalignment between 4D-flow data and the cine images due to heart and patient movements [33]. Unfortunately, the blood-tissue contrast in 4D-flow is very low, which is why an accurate LV segmentation is difficult to perform on the 4D-flow data directly. Current approaches such as in Corrado et al. [34] register automated cine segmentations onto the 4D-flow data for faster analysis. Others, such as in Bustamante et al. [35] use atlas-based segmentations, that means a general segmentation mask is registered onto the 4D-flow CMR data and adapted to the scan. That atlas-based segmentation methods have been used to also train a U-net for direct LV segmentation of cardiac 4D-flow [36]. Prior research has shown that placing the atrioventricular plane at the position of the peak inflow velocity rather than at the height of the valvular plane improves the accuracy of 4D-flow_AIM flow velocity estimation [9].

In Garcia et al. [37] a machine learning tool was developed to automatically detect evaluation planes following the mitral valve motion in cine data, which then were interpolated onto 4D-flow data. The need for a measuring plane perpendicular to valvular inflow likely extends to jet planes, which may explain the relatively poor correlation between mitral regurgitation fraction measurements using the volumetric, 4D-flow_jet, and 4D-flow_AIM techniques [19]. Moreover, the limited temporal resolution reduces the overall 4D-flow SNR [32] and affects the velocity profile quality [20] and the measured KE [38].

4D-flow scanning parameters are dependent on many factors, such as the vendor, sequence, and patient’s hemodynamics, as indicated by the 4D-flow consensus statement [7]. The VENC (in cm/s) is often set to be 10% higher than the highest predicted velocity to achieve an acceptable velocity-to-noise ratio (VNR) and avoid aliasing. It is typically about 150 cm/s for MVR quantifications, ranging from 120 to 550 cm/s in the evaluated studies. Aliasing occurs when the VENC value is less than the highest flow velocity, and a high VENC results in a reduced VNR. The FOV of 4D-flow ideally covers the whole heart with the aortic arch. However, it is sufficient to cover the region of interest to decrease scan time, which in the case of MVR quantification is the left ventricle and left atrium. Since the spatial and temporal resolutions impact the accuracy of the flow acquisition, it is best to set them to the highest resolution if there is no time constraint. The temporal resolution is recommended to be lower than 40 ms as stated in the consensus [7], with a range of 21–86 ms. All the reviewed studies used retrospective ECG triggering to cover the whole cardiac cycle and avoid sequence interruptions. However, novel 4D-flow acquisitions use cardiac self-gating techniques [7]. All studies also used respiratory gating to decrease breathing artifacts and scan duration by positioning the navigator on the liver-diaphragm interface. Also, the flip angle varies from 5° to 15°. Overall, it can be concluded that variations in 4D-flow image quality might not be related to technique itself, rather to an inappropriate use of imaging parameters. A consensus of 4D-flow parameters for MVR is still needed.

As opposed to 2D-PC CMR, the 4D-flow analysis uses RVT to quantify eccentric regurgitation jets and correct for annular valve plane motions [10, 13, 26, 28]. In the net forward flow evaluation through cardiac valves, RVT has demonstrated greater accuracy with lesser variance when compared to 2D-PC CMR methods [10, 26, 28]. A multi-center study on assessing the consistency of automated RVT demonstrated that valvular flow measurement can be independent of local CMR scanners and protocols [25].

Even though the optimal setting for MVR quantification remains to be determined, currently used scanners and protocols, still allow for a consistent acquisition of 4D flow sequences [25].

Data on the clinical value of MVR quantification by 4D-flow CMR is scarce and based on small observational studies. To the best of our knowledge, no study exists that links MVR characteristics determined by 4D-flow CMR to the long-term outcome or hard clinical endpoints such as mortality or heart failure events, or remodeling after mitral valve replacement. Conflicting data from large randomized clinical trials on the value of transcatheter mitral valve edge-to-edge repair [39, 40] underline the urgent need for a reproducible and robust quantification of MVR severity that correlates with outcomes and can be used to guide therapeutic decisions [41].