Real-Time Three-Dimensional Echocardiographic Acquisition and Quantification of Left Ventricular Indices in Children and Young Adults with Congenital Heart Disease: Comparison with Magnetic Resonance Imaging

Study Population 

Children with congenital heart disease with intact, functional left ventricles who were referred for a cardiac MRI at Children’s Healthcare of Atlanta between February and April of 2005 were considered for enrollment. Fourteen consecutive patients were enrolled. In two subjects, the acoustic windows were insufficient; thus, the study population consisted of 12 patients (5 female, 7 male; median age = 15.9 years, range 1-33 years). Seven subjects had previous right ventricular outflow tract reconstruction (repaired tetralogy of Fallot n = 5, pulmonary stenosis status/postpulmonary valvotomy n = 2), and five subjects had left-sided cardiac lesions (aortic coarctation n = 2, midaortic stenosis n = 1, aortic valve stenosis n = 1, and anomalous left coronary artery n = 1). Three patients with left-sided lesions had LV hypertrophy. Patients were enrolled in accordance with the standards of the Emory University and Children’s Healthcare of Atlanta institutional review boards. Cardiac MRI and RT3DE evaluations were performed on the same day in each patient.

Real-Time Three-Dimensional Echocardiography Assessment 

RT3DE was performed on a commercial ultrasound system (iE33 intelligent Echocardiography, Philips, Andover, MA). Pyramidal 3D volume datasets were obtained in the apical view, using the 1- to 3-MHz X3-1 matrix array transducer with 2400 elements and a scan aperture of 18 × 12.2 mm. Gain and compression controls were adjusted, and fundamental imaging was used to improve image quality. Views were optimized to include the entire LV cavity and walls using 2D biplane; then several full-volume 3D datasets with field views of 75 degrees were acquired using medium-line density. Each conical dataset was composed of four 75 × 19-degree subvolumes that were obtained from separate cardiac cycles, triggered to the R-wave of the electrocardiogram of every other heartbeat. Thus, total acquisition time was eight heartbeats. The four subvolumes were obtained from every other heartbeat. The temporal resolution was approximately 20 frames per second. Acquisition of the subvolumes was steered electronically without transducer movement. Breath holding was not used.

The full-volume images were manipulated with commercial software equipped with the RT3DE system (3DQLAB, Philips, Inc, Carlsbad, CA). Each volumetric dataset was displayed in a four-tile screen consisting of the pyramidal view and three color-coded planar cross-sections (Figure 1): four-chamber long-axis (green, top left), two-chamber long-axis (red, top right), and short-axis (blue, bottom left). The three planer images were manipulated using multiplanar reconstruction, as described by Mor-Avi et al.1 to select anatomically correct two- and four-chamber views with the largest long-axis dimensions. End diastole was marked on the cine-loop as the frame of mitral valve closure, and end systole was the frame just before mitral valve opening.

View Large Image Download to PowerPoint

Figure 1. Cross-sections of anatomically correct LV apical (top) and short-axis (bottom left), from the RT3D pyramidal data set (bottom right). Color figure online.

Biplane calculations were made from the anatomically correct 3D volumetric data using algorithms incorporated into the analysis software (3DQ, QLAB, Philips, Inc). After identification of the apex and mitral annulus in both the two- and four-chamber views, the LV border template automatically traced the endocardial and epicardial contours in both orthogonal views. (Figure 2A, B). LV trabeculations and papillary muscles were included within the LV cavity and excluded from mass calculation. Single or multiple border points were manually repositioned using a “force-field” click/drag system to optimize borders. LVEDV and left ventricular end-systolic volume (LVESV) were automatically calculated using biplane Simpson’s formula with 20 slices of the endocardial contour.17 Stroke volume (SV), ejection fraction (EF), and mass were calculated in accordance with the American Society of Echocardiography guidelines.18

View Large Image Download to PowerPoint

Figure 2. Online analysis of real-time 3-dimensional echocardiography (RT3DE) of the anatomically correct 4-chamber LV of two patients: (A, B): Patient 1, with endocardial tracing using Biplane method-of-discs in diastole (A) and systole (B). (C, D): Patient 2, with endocardial tracing of full volume dataset, using SABD in diastole (C) and systole (D). Color figure online.

Next, LV volumes were calculated using 3D full-volume algorithms on the RT3DE system (3DQ Advanced, QLAB, Philips, Inc). 3DQ Advanced used a semiautomated border detection (SABD) system to create a full-volume 3D endocardial contour from five user-defined points: four at the mitral annuli (two- and four-chamber views) and one apical point. An advanced parallel processing algorithm rapidly generated 3D wire-mesh endocardial volumes for end diastole and end systole, and calculated EF (Figure 2C, D). These borders were then manually manipulated for optimization in all three planes, and the left ventricular end-diastolic volume (LVEDV), LVESV, SV, and EF were automatically updated. The SABD algorithm used a true 3D deformable border to find endocardial edges in three dimensions from all dataset voxels. Thus, the algorithm did not use intersecting two-dimensional planes.

Magnetic Resonance Imaging Assessment 

All patients had cardiac MRI on a commercially available GE Twinspeed 1.5 T MR scanners (GE Medical Systems, Milwaukee, WI) using an eight-channel phased array cardiac coil appropriate for subject size. Five patients were sedated with propofol, per an institutionally approved sedation protocol. Ventricular volume analysis was performed on a GE Advantage Windows workstation equipped with MASS v4.1 analysis software. Images processed were obtained with electrocardiographically triggered, steady-state free precession (FIESTA) sequences (GE Medical Systems). Eight to 12 contiguous slices were obtained in a short-axis, two-chamber projection, perpendicular to the long axis of the right and left ventricles. Each slice had 20 phases per cardiac cycle and was 6 to 8 mm thick. Multiple signal averages were performed. Images corresponding to observer-selected end diastole (largest) and end systole (smallest) were volumetrically analyzed. Endocardial borders were traced using a commercially available software package (MASS v4.1, Medis, Leiden, The Netherlands). The papillary muscles were included in the LV volumetric calculations (Figure 3). Volumes, EF, and mass were determined for each ventricle.

View Large Image Download to PowerPoint

Figure 3. MRI short-axis slices with endocardial and epicardial tracings. Color figure online.

The approximate time required to acquire the 3D volume set using RT3DE was 2 minutes. Analysis of LV volume and mass measurements with biplane calculations required 20 minutes on average. Calculation of LV volume using 3DQ Advanced was performed in 10 minutes or less. MRI imaging of the left ventricle requires 15 minutes. Image processing to determine LV volume and mass from MRI images required 30 to 40 minutes.

Interobserver Variability 

To determine interobserver variability of the RT3DE measurements, the LVEDV measurements were performed by two independent observers (T.R., D.A.F.) blinded to the results of each other. Interobserver percent variability is expressed as the absolute value of the difference between measurements divided by the mean of the measurements, multiplied by 100.

Intraobserver Variability 

To determine intraobserver variability of separate measurements of LVEDV were performed by a single observer (T.R.). As above, the intraobserver percent variability is expressed as the absolute value of the difference between measurements divided by the mean of the measurements, multiplied by 100.

Statistical Analysis 

Statistical analysis of the LV volumes, mass, SV, and EF was performed with commercial software with statistical analysis tool add-ins (Excel, Microsoft Office 2000, Microsoft Corporation, Redmond, WA). Pearson’s regression and Bland-Altman analyses were used to assess correlation and agreement between the quantification methods.19 First, the two RT3DE techniques (biplane vs. SABD) were compared, using the true-volume SABD system as the reference standard. Then, the true-volume SABD volumetric measurements were compared with the gold-standard MRI calculations. Statistical significance was defined as P < .05.

Real-Time Three-Dimensional Echocardiography Techniques 

In patients in whom it was difficult to view the full long-axis LV cavity in a single 2D apical view because of cardiac abnormalities or tissue interference, planar manipulation of the RT3DE volume dataset enabled the visualization of the LV borders to measure more accurately the LV dimensions. Thus, LV volumes and mass could be quantified by both biplane and SABD techniques on all patients.

The biplane technique of quantifying LV volumes and EF was compared with the full-volume SABD technique. The mean measurement values and statistical analysis are given in Table 1. The large standard deviation in the mean values of LV indices was attributable to the wide range of patient age and size. Figure 4 shows the Pearson regression and Bland-Altman analysis, with 95% limits of agreement, of LVEDV using these techniques. There was excellent correlation for end-diastolic volume (EDV) and end-systolic volume (ESV) between biplane and SABD techniques (r = 0.99 and r = 0.98, respectively, P < .001). As volumes increased, SABD insignificantly underestimated EDV and ESV, compared with biplane method. This underestimation was within the clinically acceptable range of variability. Both methods had similar interobserver variability for LVEDV measurement: 7% ± 5% (range 1%-15%) for the biplane RT3DE method and 6% ± 3% (range 2%-11%) for the SABD method. The mean intraobserver variability for measurement of EDV by the biplane technique was a variability of 2.9% with a standard deviation of variability of 3.0%. The mean intraobserver variability for measurement of EDV by the SABD system was a variability of 5.2% with a standard deviation of variability of 4.6%.

Table 1. Quantification of left ventricular indices: Mean measured values, correlation, and agreement analysis data for all methods— biplane real-time three-dimensional echocardiography, semiautomated border detection real-time three-dimensional echocardiography, and magnetic resonance imaging

		Biplane vs. SABD RT3DE techniques				RT3DE vs. MRI
		Mean value ± SD		Regression coefficient, r	Bland-Altman mean difference ± SD	Mean value ± SD		Regression coefficient, r	Bland-Altman mean difference ± SD
		Biplane	SABD			RT3DE	MRI
	EDV (cc)	103±92	101±42	0.99⁎	−1.9±5.2†	101±42	105±44	0.99⁎	−4.1±5.2†
	ESV (cc)	48±25	47±25	0.98⁎	−1.5±5.5†	47±25	42±24	0.93⁎	4.9±9.3†
	SV (cc)	55±20	54±20	0.94⁎	−0.3±7.0†	54±20	63±24	0.94⁎	−9.0±8.4†
	EF (%)	55±8	56±9	0.85⁎	0.8±4.8†	56±9	62±9	0.69	−6.1±7.1†
	Mass (g)					72±31	69±33	0.96⁎	2.3±9.4†

EDV, End-diastolic volume; EF, ejection fraction; ESV, end-systolic volume, SV, stroke volume; SD, standard deviation; SABD, semiautomated border detection; MRI, magnetic resonance imaging; RT3DE, real-time three-dimensional echocardiography.

⁎ All Pearson’s regression correlation are significant, P < .001. † No statistically significant disagreement by Bland-Altman analysis.

View Large Image Download to PowerPoint

Figure 4. Pearson’s regression and Bland-Altman analysis of LVEDV by biplane and SABD methods of on-line real-time 3-dimensional echocardiography (RT3DE) analysis.

Real-Time Three-Dimensional Echocardiography Versus Magnetic Resonance Imaging 

Because results from both RT3DE techniques were similar, the full-volume SABD measurements were used to compare RT3DE with MRI volumetric data. The biplane technique was the only echocardiographic method available to measure mass; thus, it was compared with MRI mass. The RT3DE and MRI values for mean LV indices and statistical analysis are given in Table 1. Figure 5 shows the correlation and agreement graphs for LVEDV between RT3DE and MRI.

View Large Image Download to PowerPoint

Figure 5. Pearson’s regression and Bland-Altman analysis of LVEDV by SABD method of RT3DE compared with MRI.

All volume measures by SABD had excellent correlation to MRI values (r = 0.93-0.99, P < .001), with EF having the lowest correlation (r = 0.69, P = .013). As values of the LV indices increased, RT3DE insignificantly underestimated LVEDV, SV, and EF. Conversely, with increasing ESV and mass, RT3DE insignificantly overestimated LVESV and mass. All of the limits of agreement were clinically acceptable. Thus, RT3DE did not significantly underestimate or overestimate MRI values for LV indices.

Real-Time Three-Dimensional Echocardiography Compared with Magnetic Resonance Imaging 

Overall, there was excellent correlation between RT3DE and MRI measurements of LV volumes, EF, and mass. Our correlation results are similar to those of Bu and colleagues,20 who used separately reconstructed 3DE volume datasets to compare LV parameters in normal children with MRI. The Pearson’s regression analysis of correlation is lower for EF than for other indices because EF is a derived calculation, dependent on the accuracy of EDV and ESV values. In this study, there was slight underestimation of LVEDV, SV, and EF, and overestimation of ESV and mass. These results may be attributable to inherent technical differences between echocardiography and MRI. First, image quality is an important limiting factor in any echocardiographic quantification technique. Poor acoustic windows, such as occur because of air interference, ventricles that do not fit inside the aperture angles, and artifacts all lead to suboptimal 3DE volume datasets. Furthermore, in echocardiography, diastolic endocardial and epicardial borders are less distinct especially in the apical region because of near field noise. Inaccurate border detection can also result when the endocardial surface remains parallel to the beam when imaged from the apex. Conversely, MRI has “bright blood” contrast between the LV cavity and endocardial border, which facilitates border detection. Another difference is that MRI uses short-axis views for measurements, not long-axis views as in echocardiography. Despite these technique differences, unlike Bu and colleagues,20 we found no statistically significant disagreement between methods as LV indices increased in value.

In regard to mass, current results have no significant difference or bias between RT3DE and MRI. Mor-Avi and colleagues1 used the new online biplane quantification software to assess LV mass in normal adult hearts. These investigators found better correlation and agreement of 3DE to MRI than for 2DE. This same study reported interobserver variability that was similar to the current study (7%) and less than for 2DE (37%). Thus, the new online measurement of mass using a 3D dataset is more accurate and less prone to error than traditional 2DE biplane measurements.

Real-Time Three-Dimensional Echocardiography Improvements 

The new RT3DE online software addresses two issues associated with all previous methods of 3DE LV quantification. First, the problem of foreshortening ventricular length is decreased with image optimization. Manipulation of the 3D volume dataset simultaneously in three planes using multiplanar reconstruction allows the greatest long-axis LV length to be chosen, thereby minimizing quantification errors associated with foreshortening of the apical views. Improved views, automated LV border template, and force-field border adjustments aided the rapid LV volumes and mass calculations by Simpson’s biplane method of discs.

Second, geometric assumptions are eliminated in the new SABD algorithm, which does not rely on intersecting 2D planes to create a 3D volume. SABD was developed to decrease analysis time associated with tracing the LV contour.2, 21, 22, 23 However, all methods still traced LV endocardial borders on a finite number of intersecting 2D planes and then approximated the area between the slices to create a 3D volume.2 Volumetric calculations using these SABD methods had significant underestimation of EDV and ESV by approximately 14%.2 The updated SABD algorithm in the current study uses all voxels in the 3D dataset to create a true-volume, 3D deformable border, thereby eliminating errors associated with interpolation and geometric assumptions.

In this study, the measurements were identical between biplane and SABD, even in patients in whom the LV cavities were abnormal because of the concurrent presence of heart disease, such as a volume-overloaded right ventricle after Tetralogy repair. The advantage of using true volumes may be most appreciated in patients with abnormally shaped hearts, in whom geometric assumptions would lead to greater inaccuracy. There have been several previous publications that have examined LV volume and ejection using a 3D method. We previously reported our results with an internally rotating 5-MHz “omniplane” transthoracic transducer. This study showed a strong correlation between the two techniques for measurement of end-systolic and end-diastolic volumes (R2 = 0.91 and R2 = 0.90).11 More recently, van den Bosch and colleagues24 reported a good correlation between RT3DE and MRI for assessment of LVEDV and EF, R2 = 0.97 and R2 = 0.94, respectively. The latter study was performed in an older population with congenital heart disease (mean age of 31 ± 9 years). Their protocol used breath holding in all subjects. Breath holding can be difficult for many children with congenital heart disease to perform. We were encouraged to find that in our protocol, in which breath holding was not performed, RT3DE still had a strong correlation to MRI measurements. Given these findings, we are hopeful these factors will permit RT3DE to play an increasingly important role in care of children with congenital heart disease.