Why Three-Dimensional CFD Imaging?

The current approach to the calculation of effective regurgitant orifice area (EROA) using the two-dimensional (2D) proximal isovelocity surface area (PISA) method has many limitations. From a practical standpoint, the assumption of a hemispheric flow convergence region (FCR) and the use of single-frame (largest) PISA are chief among them. The advantages of three-dimensional (3D) echocardiography, specifically 3D CFD, are widely recognized, and this technique is recommended to overcome these limitations of 2D CFD. However, there are technical and practical challenges with 3D CFD, which have hindered its routine use. In this issue of JASE , Pierce et al . and Tan et al . describe an approach using an algorithm called the 3D field optimization method (FOM), which generates a 3D flow vector velocity field in the FCR by comparing the modeled flow field with the observed spatial distribution of flow velocity vectors in the FCR proximal to the orifice. The FOM then computes regurgitant volume (RVol) using the instantaneous flow velocities integrated over the duration of MR. In their in vitro study, Pierce et al . examined the differences in RVol computed using the conventional 2D CFD PISA method and the 3D FOM when applied to circular and slit-shaped regurgitant orifices, using measures from a flow probe as a reference. Compared with 2D PISA results, they found that the 3D FOM was more accurate for slitlike orifices and for severe MR. For round orifices and mild MR, 3D FOM was no more accurate than 2D PISA, but for moderate MR with circular and slitlike orifice, 3D FOM was only modestly accurate and 2D PISA fared even worse. The accompanying in vivo animal study by Tan et al ., and the results of their application of 3D FOM to 3D transesophageal echocardiographic CFD data sets in patients with MR, largely confirm the results of the in vitro study by Pierce et al . There are many limitations in both studies, which the investigators have clearly elucidated. Chief among them is the lack of an independent reference standard in the work by Tan et al ., meaning that they could not verify the accuracy of 3D FOM. Even the accuracy data from the in vitro work is not compelling in the moderate MR category, which is where we need the most help for decision making. Taken together, the data from the two studies suggest that 3D FOM adds to the more established 3D CFD methods to quantify MR but that this approach will need further refinement and validation before routine clinical use.

Despite these limitations, 3D FOM is a novel approach in that it does not require data from continuous-wave spectral Doppler, which is often not optimal, especially in eccentric MR jets. Another notable feature is that velocities over the duration of the MR are used in computing EROA and RVol, thus avoiding the potential for overestimation of the degree of MR in dynamic and nonholosystolics MR. An alternative 3D CFD approach uses voxel segmentation to identify isovelocity in the FCR and automatically quantify the 3D surface area. Then, using the peak velocity and the velocity-time integral from continuous-wave spectral Doppler of the MR jet, 3D EROA and RVol are computed. When the 3D surface area of the frame with the largest FCR is used along with the peak velocity of the MR jet, we can determine “peak PISA” EROA, and when this is combined with the VTI of the MR jet, “peak instantaneous” RVol is obtained. Both of these variables overestimate the degree of MR compared with cardiac magnetic resonance (CMR)–based classification of MR severity, especially with dynamic MR, as is often seen with secondary MR, and when MR is not holosystolic in timing. However, the 3D surface area of the FCR at each time point over the duration of MR can be measured, and “peak instantaneous” RVol at each of these time points can be derived using the peak MR velocity from continuous-wave Doppler. By integrating these individual peak instantaneous RVol values over the duration of MR, “integrated PISA”–derived RVol is determined; this measure shows significantly better agreement with RVol measured by CMR. Thus, both automated 3D FOM and the 3D voxel segmentation approach to identify the FCR overcome the potential for overestimating MR severity using the single-frame approach. Integration of data from 2D CFD of the FCR over the duration of MR can be done, but it requires manual computation and still suffers from limitations of using 2D images to describe what is essentially a 3D shape.

The Benefits of Automated Quantitative 3D CFD

Perhaps the most important finding in the studies by Pierce et al . and Tan et al . is the value of automation. Three-dimensional data sets are inherently rich in data, and the need for manual extraction of quantitative data has been and continues to be one of the most important challenges to its routine use in everyday practice, even when the acoustic window is optimal. Specifically, with regard to quantification of MR, the cumulative science and common wisdom that 3D CFD is superior to 2D CFD is compelling enough for guidelines and standards to recommend its use, at least in challenging situations. But two key factors, among others, have discouraged the routine use of 3D CFD: the need for electrocardiographically gated imaging and manual (often tedious) interaction with the data to obtain parameters. Both of these are counterintuitive to work flow (all work for flow, or all work and no flow!) and reproducibility. It is now possible to perform real-time (nongated, nonstitched) CFD imaging at temporal resolution not significantly different from that of gated imaging, although there is clearly room for improvement. Furthermore, gated 3D CFD is virtually impossible in atrial fibrillation, which is not uncommon in chronic MR. Even if cardiac rhythm is not an issue, flow is instantaneous, and it makes sense to measure it beat by beat, which also makes averaging data feasible. One of the criticisms leveled at CMR is that it is a gated technique and hence “not real time”; if this is a true limitation, then it behooves us to embrace real-time (nongated) CFD imaging and to push for all platforms to provide this. In addition, we should welcome automation and not use the argument that an expert, extracting data manually, is irreplaceable. The problem in 3D echocardiography is not abuse of automation but rather underuse and often misplaced skepticism. The fact is that both accuracy and reproducibility are improved by automation, and reproducibility is considerably worse when experts perform tasks manually, whether in a fully manual manner or by making subtle “adjustments” to the automation. This is not to argue that automation is infallible and that it should be implemented without the ability to override what is obviously incorrect. However, we would note that if automation works in excess of 90% of the time, it will improve work flow significantly and encourage wider use. Furthermore, it is not unreasonable to anticipate that continuous refinement of technology and intelligence will improve automation to the levels that are seen in new airplanes and space technology, which have similar degrees of complexity. Work flow is one of the strengths of echocardiography, and given that it is the most commonly performed and recommended test for the quantification of MR, automated extraction of parameters from a complex 3D data will only enhance its value.

The Benefits of Automated Quantitative 3D CFD

Perhaps the most important finding in the studies by Pierce et al . and Tan et al . is the value of automation. Three-dimensional data sets are inherently rich in data, and the need for manual extraction of quantitative data has been and continues to be one of the most important challenges to its routine use in everyday practice, even when the acoustic window is optimal. Specifically, with regard to quantification of MR, the cumulative science and common wisdom that 3D CFD is superior to 2D CFD is compelling enough for guidelines and standards to recommend its use, at least in challenging situations. But two key factors, among others, have discouraged the routine use of 3D CFD: the need for electrocardiographically gated imaging and manual (often tedious) interaction with the data to obtain parameters. Both of these are counterintuitive to work flow (all work for flow, or all work and no flow!) and reproducibility. It is now possible to perform real-time (nongated, nonstitched) CFD imaging at temporal resolution not significantly different from that of gated imaging, although there is clearly room for improvement. Furthermore, gated 3D CFD is virtually impossible in atrial fibrillation, which is not uncommon in chronic MR. Even if cardiac rhythm is not an issue, flow is instantaneous, and it makes sense to measure it beat by beat, which also makes averaging data feasible. One of the criticisms leveled at CMR is that it is a gated technique and hence “not real time”; if this is a true limitation, then it behooves us to embrace real-time (nongated) CFD imaging and to push for all platforms to provide this. In addition, we should welcome automation and not use the argument that an expert, extracting data manually, is irreplaceable. The problem in 3D echocardiography is not abuse of automation but rather underuse and often misplaced skepticism. The fact is that both accuracy and reproducibility are improved by automation, and reproducibility is considerably worse when experts perform tasks manually, whether in a fully manual manner or by making subtle “adjustments” to the automation. This is not to argue that automation is infallible and that it should be implemented without the ability to override what is obviously incorrect. However, we would note that if automation works in excess of 90% of the time, it will improve work flow significantly and encourage wider use. Furthermore, it is not unreasonable to anticipate that continuous refinement of technology and intelligence will improve automation to the levels that are seen in new airplanes and space technology, which have similar degrees of complexity. Work flow is one of the strengths of echocardiography, and given that it is the most commonly performed and recommended test for the quantification of MR, automated extraction of parameters from a complex 3D data will only enhance its value.