Background
Assessment of right ventricular (RV) volumes and function is important in patients after tetralogy of Fallot (TOF) repair. Currently, cardiac magnetic resonance imaging (MRI) is considered the clinical reference method for RV volume and function measurements. Three-dimensional (3D) knowledge-based reconstruction derived from two-dimensional echocardiographic imaging with magnetic tracking is a novel approach to RV volumetrics. The aim of this study was to assess the feasibility and reliability of this novel echocardiographic technique in patients after TOF repair. The accuracy of the method was assessed by comparison with measurements obtained by cardiac MRI.
Methods
Thirty patients (mean age, 13.7 ± 2.8 years) after TOF repair, referred for cardiac MRI, were included. Immediately after MRI, echocardiographic image acquisition was performed using a standard ultrasound scanner linked to a Ventripoint Medical Systems unit. Echocardiographic and MRI measurements were performed offline. Parameters analyzed were end-diastolic volume (EDV), end-systolic volume (ESV), and ejection fraction. Intraobserver, interobserver, and intertechnique variability was assessed using Pearson’s correlation analysis, coefficients of variation, and Bland-Altman analysis.
Results
Echocardiographic two-dimensionally based 3D reconstruction was highly feasible, with low intraobserver and interobserver variability for EDV and slightly higher variability for ESV and ejection fraction. The 3D reconstruction values for EDV, ESV, and ejection fraction were correlated highly with MRI values, with low coefficients of variation. The agreement between both methods was high. Three-dimensional reconstruction slightly underestimated RV volumes, by 2.5% for EDV and 4.6% for ESV compared with MRI volumes.
Conclusions
In patients after TOF repair, echocardiographic 3D reconstruction is highly feasible, with good reproducibility for measurements of RV EDV. There is good agreement with MRI measurements, with a small underestimation of RV volumes. The use of this method in clinical practice warrants further investigation.
Assessment of right ventricular (RV) volumes and function is important in the follow-up of cardiac lesions when the right ventricle is affected. This is particularly relevant in patients after surgical repair of tetralogy of Fallot (TOF), who often develop progressive RV dilatation and dysfunction, related to postoperative pulmonary regurgitation. Currently, cardiac magnetic resonance imaging (MRI) is considered the method of choice for the assessment of RV volumes and ejection fraction (EF). Quantitative three-dimensional (3D) echocardiography using real-time volumetric acquisition and dedicated RV volume analysis software has emerged as a promising clinical alternative to MRI. However, the feasibility of this method is problematic, especially in postoperative patients with significantly dilated RV volumes. Poor echocardiographic windows can limit RV endocardial border detection in about 20% to 50% of all patients, limiting the feasibility of the method. Additionally, 3D echocardiography was shown to underestimate RV volumes by up to 15% to 25%, with the underestimation being more important in more dilated ventricles. Therefore, alternative echocardiographic methods are worthwhile investigating. In this study, we used a two-dimensional (2D) method that localizes different anatomic landmarks in a magnetic 3D space and reconstructs the RV volumes using a database of various RV shapes and sizes obtained in patients after TOF repair.
The aim of the present study was to assess the feasibility, reliability, and accuracy of the echocardiographic knowledge-based 3D reconstruction (3DR) method to measure RV volumes and EFs in patients after TOF repair and compare the results with RV measurements obtained by cardiac MRI.
Methods
This was a single-center, prospective, observational study. The study was approved by the institutional research ethics board. Informed consent was obtained from all participants and/or their legal guardians before enrollment.
Patients after TOF repair, aged 8 to 17 years, who were scheduled to undergo cardiac MRI for clinical indications were included. Exclusion criteria were an inability to cooperate, known or detected arrhythmia interfering with image acquisition, and pacemakers or any contraindication to MRI. We also excluded patients after surgical repair with right ventricle–to–pulmonary artery conduits. The current TOF database used for RV volume reconstruction does not include patients with conduits. A different catalog for patients with conduits has been developed but was not used in the present study. Just before or immediately after cardiac MRI, a limited echocardiographic study was performed, including image acquisition for 3DR. All echocardiographic studies were performed using a GE Vivid 7 machine (GE Healthcare, Milwaukee, WI). Data were digitally stored and analyzed offline. RV end-diastolic volume (EDV), end-systolic volume (ESV), and EF were calculated. Intraobserver variability was performed using repeated analysis for the 3DR method, with ≥2 weeks between the analyses. For interobserver variability, two independent observers analyzed the 20 first echocardiographic data sets. Both observers were experienced with the technique.
3DR Method
The 3DR method enables 2D-based full volumetric reconstructions of cardiac chambers from 2D echocardiographic images localized in 3D space, using proprietary software and a catalog of heart shapes. This knowledge-based 3DR of ventricular volumes has been previously described and validated in vitro from 3D echocardiographic data sets using the piecewise smooth subdivision surface reconstruction method. It was also clinically validated for the left ventricle, and more recently, a catalog of fully traced MRI acquisitions of a wide range of RV shapes and sizes was shown to allow accurate knowledge-based reconstructions of RV volumes using MRI data sets. On the basis of the same principles, an echocardiographic method was developed. The method involves the acquisition of 2D echocardiographic images with a magnetic localizer attached to the ultrasound probe, allowing localization of the different 2D images in 3D space. The image acquisition protocol should provide good coverage of the different parts of the right ventricle. After acquisition, offline analysis is performed, including the identification of different anatomic landmarks on the 2D images. The data points (anatomic landmarks with their spatial localization) are matched to a catalog of 3D shapes obtained from cardiac MRI studies in patients after TOF repair, and the RV volumes and EFs are calculated. The different steps are explained in more detail.
Image Acquisition
The acquisition involves freehand 2D transthoracic scanning using a standard 2D probe with a magnetic field localizing system (Ascension Technology Corporation, Andover, MA) to track the position and orientation of the ultrasound probe in 3D space. The system consists of a magnetic field transmitter located under the exam bed. This generates different orthogonal magnetic fields that are sensed by the receiver, which is attached to the ultrasound probe. The position and orientation of the receiver and thus the plane of the 2D picture can be computed and placed within the volume created by the magnetic transmitter. Different 2D images of the right ventricle are obtained. To obtain maximal coverage of the right ventricle, 10 to 15 different echocardiographic views should be acquired in each patient. These are summarized in Table 1 and illustrated in Figure 1 . For each view, short clips of 2 to 4 sec are recorded to include at least three full cardiac cycles. Subjects are not allowed to change position on the exam bed during the acquisition process, and when possible, we obtained each clip using breath holding at end-expiration. The ultrasound scanner is linked to a computer (Ventripoint Medical Systems; Ventripoint, Inc., Seattle, WA) through the video output, and every clip is digitized at 30 frames/sec. The magnetic field data are entered at the same time, and 3D localizing data are linked in the computer online. The entire acquisition protocol generally takes about 5 min.
| View | Region of interest |
|---|---|
| 1. Apical four-chamber view | Tricuspid annulus, basal bulge, RV endocardium, RV apex, RV septum |
| 2. Apical two-chamber view of the right ventricle | Tricuspid annulus, subtricuspid inferior wall, RV endocardium, pulmonary annulus |
| 3. Foreshortened apical three-chamber view | Tricuspid annulus, RV endocardium, RV septum, RV outflow |
| 4. Modified oblique view of the true RV apex | RV apex |
| 5. Parasternal long-axis view, mitral valve | Interventricular septum, RV endocardium |
| 6. Parasternal long-axis view, tricuspid valve | Tricuspid annulus, RV endocardium |
| 7. Parasternal long-axis view, pulmonary valve | Pulmonary annulus, RV endocardium, RV outflow |
| 8. Parasternal short-axis view, pulmonary valve | Pulmonary annulus, subpulmonary area, RV endocardium |
| 9. Short-axis view, base | Interventricular septum, RV endocardium |
| 10. Short-axis view, mid/apex | Interventricular septum, RV endocardium |
| 11. Rotated apical views | Tricuspid annulus, basal bulge, RV endocardium, RV apex, RV septum |
| 12. Foreshortened apical view, RV outflow | RV endocardium, pulmonary annulus, RV outflow |
| 13. Subcostal four-chamber view | RV endocardium, RV apex, RV septum |
| 14. Subcostal sagittal view | RV endocardium, RV outflow |
Image Analysis
The images and spatial information are stored on the Ventripoint Medical Systems computer, which has an Internet connection to the centralized database. First, timing of the cardiac cycle needs to be performed, with identification of end-diastole and end-systole. End-diastole is automatically set to correspond to the electrocardiographic R wave but can be adjusted manually. End-systole is identified manually on one clip, and the time interval is automatically applied to all clips on the basis of the electrocardiogram. This requires the heart rate to be relatively stable during the acquisition process. After the user approves the timing, identification of the different anatomic landmarks is performed on the different images. A small number of key anatomic points need to be identified for both end-diastole and end-systole. This includes the tricuspid and pulmonary valve annuli, the RV apex, the RV side of the interventricular septum, the RV free wall, the subpulmonary area (just below the pulmonary valve), and the subtricuspid area (on the RV inferior wall). For full 3DR, 17 to 23 points need to be identified on the 2D images. No additional anatomic tracing or border detection is required.
Surface Generation Using Catalog-Based Method
Once the anatomic points have been identified and localized in the magnetic 3D space, the RV volumetric reconstruction is performed by sending the data to the online database. The “knowledge” MRI database consists of fully traced RV volumes of postoperative patients after TOF repair with a wide spectrum of different RV volumes and shapes. The identified landmarks on the 2D images with their location in 3D space are matched with the database, and the best fit is computed using a proprietary algorithm. Once a minimum number of points have been placed on the anatomic structures, ideally 15 to 20, the first reconstruction can be run. The reconstructed RV volume is projected on the 2D data set ( Figure 1 ), and on the basis of this quality check, the placement of points can be adjusted and more points can be added to refine the RV volumetric modeling and end up with the best fit for EDV ( Figure 2 ). For calculating ESV, the same process is repeated. RV volumes and EF are calculated by the system.
Cardiac MRI
MRI scans were performed on a 1.5-T scanner (Avanto; Siemens Medical Systems, Erlangen, Germany). For ventricular volumetry, a short-axis cine stack was acquired during breath hold, using a balanced steady-state free precession gradient-echo sequence with minimum echo and repetition times, a flip angle of 45°, bandwidth of 31.25 kHz, one excitation, views (lines) per segment to allow for 20 true reconstructed phases per cardiac cycle, 5 to 6 mm of slice thickness, 10 to 12 slices, gap adjusted to cover both ventricles, and in-plane spatial resolution of 1.5 to 2.5 mm. The MRI data were analyzed with commercially available software packages (Mass Analysis and CV Flow, Medis Medical Imaging Systems, Leiden, The Netherlands). Left ventricular and RV end-diastolic (maximal) and end-systolic (minimal) volumes, stroke volumes, and EFs were measured. The interobserver variability of the RV volumes was assessed by independent analysis of the first 20 studies by two experienced observers.
Statistical Analysis
Different statistical methods were used to analyze intraobserver, interobserver, and intertechnique variability. The agreement between the measurements is described using Bland-Altman statistics, including the calculation of mean bias (average difference between measurements), with standard deviation. The statistical significance of the mean bias was tested using a paired, two-tailed t test (the null hypothesis was zero bias). The lower and upper limits of agreement (95% limits of agreement of mean bias) were calculated. The percentage differences (the difference between paired measurements divided by the average of the two measurements times 100) and percentage bias were calculated for the different Bland-Altman analysis. The correlation between observers and techniques was calculated using Pearson’s correlations. Variability of the measurements was also assessed by calculating the coefficient of variation as the standard deviation of the difference of paired samples divided by the average of the paired samples.
Results
Feasibility
Thirty children (mean age, 13.7 ± 2.8 years) were included in the study. The 3DR analysis was successfully completed in all studies. Small respiratory artifacts were noted in a few patients, but they did not preclude the reconstructions. Generally, three to four reconstructions with point adjustments were required to obtain optimal reconstructions, with easier adjustments when fewer initial points were entered. Overall image acquisition took about 5 min, and the analysis took about 15 min.
Reproducibility
The results of the intraobserver and interobserver variability analysis are shown in Table 2 , and the Bland-Altman plots are depicted in Figure 3 . The intraobserver and interobserver variability of 3DR for calculating EDV was low, with excellent correlation coefficients, low coefficients of variation, no significant bias, and narrow limits of agreement. Intraobserver and interobserver variability was slightly higher for ESV and less good for EF. The differences between intraobserver and interobserver measurements were not significant for any of the parameters analyzed. The reproducibility of MRI measurements between observers is presented in Table 2 , with overall good agreement between the two observers, especially for EDV and less so for ESV and EF. A small but significant bias was detected between the two observers: a 2.4% (5.3 mL) difference for EDV ( P = .01) and a 4.4% (5.7 mL) difference for ESV ( P = .002).
