Accurate Measurement of Left Ventricular Ejection Fraction by Three-dimensional Echocardiography
A Comparison With Radionuclide Angiography
Background Three-dimensional echocardiography is a promising technique for calculation of left ventricular ejection fraction, because it allows its measurement without geometric assumptions. However, few data exist that study its reproducibility and accuracy in patients.
Methods and Results Twenty-five patients underwent radionuclide angiography and three-dimensional echocardiography that used the rotational technique (2° interval and ECG and respiratory gating). Left ventricular volume and ejection fraction were calculated by use of Simpson's rule at a slice thickness of 3 mm. Analyses were performed to define the largest slice thickness required for accurate calculation of left ventricular volume and ejection fraction. Three-dimensional echocardiography showed excellent correlation with radionuclide angiography for calculation of left ventricular ejection fraction (mean±SD, 38.9±19.8 and 38.5±18.0, respectively; r=.99); their mean difference was not significant (0.03±0.17; P=.3), and they had a close limit of agreement (−0.385, 0.315). Intraobserver variability for radionuclide angiography and three-dimensional echocardiography was 4.2% and 2.6%, respectively, whereas interobserver variability was 6.2% and 5.3%, respectively. There was no significant difference between left ventricular volume and ejection fraction calculated at a slice thickness of 3 mm and that calculated at different slice thicknesses up to 24 mm. However, the standard deviation of the mean difference showed a stepwise increase, particularly at thicknesses >15 mm. At a slice thickness of 15 mm, the probability of three-dimensional echocardiography to detect ≥6% difference in ejection fraction was 80%.
Conclusions Three-dimensional echocardiography has excellent correlation with radionuclide angiography for calculation of left ventricular ejection fraction in patients and has an observer variability similar to that of radionuclide angiography. We recommend the use of a 15-mm-thick slice for accurate and rapid measurement of left ventricular volume and ejection fraction.
Radionuclide angiography is an accepted method for the measurement of LVEF.3 4 However, because it is rather expensive and necessitates the exposure of the patient to radiation, it is a suboptimal test when serial calculations of LVEF are required.
2D echocardiography is a widespread technique for clinical evaluation of LVEF. However, assessment of LV performance by 2D echocardiographic techniques is based on geometric assumptions.5 6 Accurate measurement of LV volume and function requires the reconstruction of the true geometry of the heart, particularly in patients with distorted LV geometry and impaired LV function.7 8
The 3D echocardiographic technique reduces the limitations of 2D echocardiography and allows quantification of LV volumes and ejection fraction without geometric assumptions.9 3D echocardiography has been shown to be highly accurate in determining volume in vitro. In studies that used phantoms and excised ventricles, LV volumes calculated by 3D echocardiography agreed closely with the actual volumes.10 Few data have been published thus far on the comparison between 3D echocardiographic calculation of LV volumes and ejection fraction with other techniques in humans.11 12
The aim of the present study was to determine the feasibility and accuracy of 3D echocardiography for calculation of LVEF in comparison with radionuclide angiography. Reproducibility of both techniques was compared in terms of intraobserver and interobserver variability. In addition, LV volumes and ejection fraction were assessed by use of different slice thicknesses to determine the largest thickness required for calculation of LV volumes and ejection fraction without loss of accuracy.
3D echocardiography was performed in 25 patients undergoing multigated radionuclide angiography for evaluation of LVEF. Patients were not selected clinically but rather for echocardiographic quality: patients in whom it was possible to visualize the entire left ventricle in all standard apical echocardiographic views were included in the present study. The 25 patients (15 men and 10 women) ranged in age from 25 to 82 years, with a mean age of 53±16 years. Eleven patients had ischemic heart disease (10 with previous myocardial infarction and 1 with angina pectoris), 5 patients had dilated cardiomyopathy, 8 patients were evaluated during chemotherapy for cancer, and 1 normal volunteer was also studied.
Informed consent was obtained from each patient after they were given a full explanation of the procedure. In each patient, a multigated radionuclide angiogram for evaluation of LVEF was performed. This was followed by a 3D echocardiographic study on the same day in 17 patients and at an interval of 1 to 9 days (mean, 3.5 days) in 8 patients. The clinical condition of the patient and medical therapy remained stable between the two studies.
Multigated Radionuclide Angiography
Radionuclide angiography was performed in the 45° left anterior oblique view after in vivo labeling of the red blood cells with 15 mCi (540 MBq) of 99mTc. Acquisition was performed during a 6-minute period with a Siemens (Orbiter) gamma camera equipped with a low-energy, all-purpose collimator. The data were processed with standard software and background correction, and the LVEF was computed from the end-systolic and end-diastolic images.
Echocardiographic studies were performed with a transducer system in the apical position while the patient lay comfortably in the 45° left recumbent position. To acquire cross-sectional images for reconstruction, the operator must find the center axis around which the imaging plane is rotated to encompass the entire LV cavity. Because the spatial coordinate system changes with transducer movement, motion of the transducer must be avoided. Inadvertent patient movement during image acquisition can be prevented for the most part by thoroughly explaining the procedure to the patient before the study. The examination, including the calibration procedures, selection of the optimal gain settings and conical volume with a few test runs, and the actual image acquisition, requires approximately 8 to 10 minutes in patients with sinus rhythm.
Precordial Transducer Assembly and Ultrasound System
We used a newly developed, custom-built, handheld transducer assembly that can be rotated with a step motor via a wheel-work interface.13 14 A commercially available 3.75-MHz sector-scanning transducer (Toshiba Sonolayer SSH-140A system) is mounted in the probe assembly (Fig 1⇓). The step motor is commanded by a steering logic for controlled image acquisition (Echo-scan, Tom Tec GmbH).
Reconstruction of the left ventricle by 3D echocardiography requires three basic steps: image acquisition, image processing, and data analysis.
A software-based steering logic activates the step motor in the transducer assembly, which controls the image acquisition in a given plane by an algorithm that considers both heart rate variations (ECG gating) and respiratory phase by thoracic-impedance measurement. Before the actual image acquisition, R-R intervals were predetermined with an acceptable variability of 150 ms or less, and respiration was gated at the end-expiratory phase. On the basis of this information, the step motor was commanded by the steering logic to acquire cross sections of cardiac cycles that fell within the preset ranges. This allowed optimal temporal and spatial registration of the cardiac images. After a cardiac cycle was selected by the steering logic, the cardiac images were sampled at 40-ms intervals (25 frames/s), digitized, and stored in the computer memory. Then, the step motor was activated and used to rotate the transducer 2° to the next scanning plane, where the same steering logic was followed. Cycles that did not meet the preset ranges were rejected. To fill the conical data volume, 90 sequential cross sections from 0° to 178° had to be obtained, each during a complete cardiac cycle.
The recorded images were formatted in their correct rotational sequence according to their ECG phase in volumetric data sets (256×256×256 pixels for each eight bit). Postprocessing of the data sets was performed off line by use of the analysis program of the system. To fill the gaps in the far fields, a trilinear cylindrical interpolation algorithm was used.
3D echocardiographic trilinear cylindrical interpolation algorithm
When a rotational device is used to acquire an image, the rotation axis is assumed to be parallel to the vertical axis (y axis) of each acquired image (the x position of the rotation axis is defined as “axpos”). Each voxel x,y in the acquired image (n) of the rotational series may then be considered as a point in a cylindrical coordinate system with R=x−axpos, Φ=n* angular stepwidth +0° or +180° (depending on the sign of x−axpos), and Z=y. If the above parameters and a pixel resolution of 1 mm for the acquired image are assumed, the maximal gap width will be ≈6.7 mm. It is obvious that some kind of interpolation algorithm has to be applied to fill the gaps. A trilinear interpolation in the cylindrical space gives acceptable results. Each cartesian voxel coordinate x, y, z of the volume to be reconstructed is transformed into the cylindrical coordinate R.Φ.Z. The gray values of the eight points in the cylindrical coordinate system of the acquired images that come closest to R.Φ.Z contribute to a weighted sum that makes up the gray value at voxels x, y, z. Weights are inversely proportional to the distances of point r.φ.z to its neighbors R(i), Φ(i), Z(i) (i=1…8).
LVEF was calculated from the 3D data sets by use of Simpson's method. This method used LV manual tracing of sequential short-axis views of the left ventricle from the apex to the mitral annulus to calculate LV volume. After the long-axis view of the left ventricle was selected, the end-diastolic (the first frame before closure of the mitral valve) and then the end-systolic (the first frame before the opening of the mitral valve) data sets were selected. We then adjusted the parallel slicing through the data sets at 3-mm intervals. This resulted in generation of equidistant cross-sections of the left ventricle. The computer displayed the corresponding short-axis view in (1) a dynamic display for better identification of the endocardium in a digitized complete cardiac cycle and (2) a static display for manual endocardial tracing. When manual tracing of the displayed short axis was completed, the system calculated the volume by summing the voxels included in the traced area in a 3-mm-thick slice. Slice by slice, the system summed the corresponding subvolumes and finally calculated the end-diastolic and end-systolic LV volumes. The system then calculated and displayed the values of stroke volume and ejection fraction (Fig 2⇓).
For each technique, measurement of LVEF was performed by two experienced observers (Y.F.M.N., A.S., J.T.P., and A.E.M.R.) each blinded to the other's results. In addition, the first observer repeated the measurement after 7 days. Intraobserver variability was calculated and expressed as the SD of the difference of the two readings divided by the average value. To determine the interobserver variability, the average value of the first observer was compared with the reading of the second. Observer variabilities were analyzed by paired Student's t test. A probability level of P<.05 was considered significant. The probability value, the mean difference and 95% CI, and the limits of agreement15 are reported.
We also performed a paired t test to compare radionuclide angiography and 3D echocardiography using the average values of the first observer obtained at the 3-mm slice thickness. Pearson correlation coefficients are presented.
In addition, LV end-diastolic and end-systolic volumes and ejection fraction were calculated by the first observer from 3D echocardiography by use of slice thicknesses that ranged from 6 to 24 mm, with stepwise increments of 3 mm. Repeated7 comparisons were made with the value obtained at a slice thickness of 3 mm by paired t tests with Bonferroni correction. Power analysis was performed to determine the possibilities of a β error in the comparison with the 15-mm slice thickness.
3D echocardiographic acquisition and reconstruction could be performed without difficulty in all patients recruited in the present study. 3D echocardiographic acquisition was repeated in 1 patient because of an error in the calibration procedure of the rotational axis. All patients included in the present study were in sinus rhythm. The difference between the mean±SD of the patients' heart rates during radionuclide angiography and 3D echocardiography was not significant (82±11 and 81±10 bpm, respectively; P=.7). Echocardiographic examination revealed that 10 patients had segmental wall motion abnormalities, 5 had global hypokinesis, and 10 had normal wall motion. In each case, the examination required approximately 8 to 10 minutes, including the calibration procedures, selection of optimal gain setting and conical volume, and image acquisition. Calibration and storage of data in the computer memory required approximately 3 minutes. Off-line image processing and analysis required on average 50 minutes.
Intraobserver and Interobserver Variability of Radionuclide Angiography
There was no significant difference in the measurement of LVEF obtained by the same observer in two different settings (difference, 0.03±0.016; P=.07; intraobserver variability of 4.2%). There was no significant difference in the measurement of LVEF obtained by the two independent observers (difference, 0.007±0.024; P=.8; interobserver variability of 6.2%). In addition, there were close limits of agreement and 95% CIs between both intraobserver and interobserver measurements obtained by radionuclide angiography. (See Table 1⇓.)
Intraobserver and Interobserver Variability of 3D Echocardiography (Simpson's Method)
There was no significant difference in the measurement of LVEF obtained by the same observer in two different settings (difference, 0.001±0.01; P=.9; intraobserver variability of 2.6%). There was no significant difference in the measurement of LVEF obtained by the two independent observers (difference, 0.02±0.02; P=.3; interobserver variability of 5.3%). In addition, there were close limits of agreement and 95% CIs between both intraobserver and interobserver measurements obtained by 3D echocardiography (true Simpson's method). (See Table 1⇑.)
Comparison of Radionuclide Angiography and 3D Echocardiography
The mean±SD values of LVEF obtained by radionuclide angiography and 3D echocardiography are presented in Table 1⇑. There was excellent correlation in the measurements of LVEF obtained by radionuclide angiography and 3D echocardiography (r=.99) (Fig 3⇓). No significant difference existed in the mean difference between the average values of LVEF obtained by radionuclide angiography and 3D echocardiography (Fig 4⇓; Table 2⇓). In addition, there were close limits of agreement and 95% CIs between the measurements of radionuclide angiography and 3D echocardiography by Simpson's method (Table 2⇓).
3D Echocardiographic Measurement of LV Volumes and Ejection Fraction at Different Slice Thicknesses
The mean±SD values of LV end-diastolic and end-systolic volumes and ejection fraction measurements when different slice thicknesses were used are presented in Table 3⇓. There was no significant difference between the mean difference of LV end-diastolic and end-systolic volume and ejection fraction calculated at a 3-mm slice thickness and those calculated at different slice thicknesses ranging from 6 to 24 mm with stepwise increments of 3 mm (Table 3⇓). The difference between LVEF calculated by radionuclide angiography and by 3D echocardiography at different slice thicknesses ranging from 3 to 24 mm with stepwise increments of 3 mm was not significant for the entire group of patients or for subgroups of patients with normal and abnormal LV wall motion (Table 4⇓; Fig 5⇓). At a slice thickness of 15 mm with this number of patients, 3D echocardiography by Simpson's method was able to detect a difference of 6% in LVEF with a power of 80%, whereas this power was 99% for detecting a difference of 10% (Table 5⇓).
Calculation of LVEF is frequently needed for the evaluation of patients with heart disease. Radionuclide angiography is an established method for the noninvasive measurement of LVEF.3 4 This technique is rather expensive and becomes impractical if repeated evaluation of LV performance is needed.3 4 5 6 7 8 9 10 11 12 13 14 15 16
2D echocardiography has gained widespread use in the clinical evaluation of LVEF because it allows comprehensive evaluation of anatomy and function in a short period of time and is noninvasive, mobile, and relatively inexpensive compared with radionuclide or other emerging imaging techniques such as MRI. However, assessment of LV performance by 2D echocardiographic techniques has suffered a reputation of limited accuracy and reproducibility.17 18 Although fairly accurate estimates of volumes and ejection fraction can be made by use of assumed methods, 2D echocardiography continues to be based on geometric assumption.5 6 Moreover, quantitative evaluation of LV volume and function requires reconstruction of the true geometry of the heart, particularly in patients with a distorted LV cavity.7 8 3D echocardiography provides accurate measurement of LV volume and function by the reconstruction of true LV geometry.19
The present study of LVEF in patients as calculated by 3D echocardiography, with use of precordial rotational technique with ECG and respiratory gating, is the first comparison with that calculated by an accepted clinical method, radionuclide angiography. Our results demonstrate an excellent correlation of LVEF calculated by 3D echocardiography by use of Simpson's method with that of radionuclide angiography. There were close limits of agreement and 95% CIs between radionuclide angiography and 3D echocardiography by use of Simpson's method.20
Comparison With Other Studies
Our findings are in agreement with the study conducted by Gopal et al,21 who compared LVEF calculated by radionuclide angiography and 3D echocardiography in 51 patients with suspected heart disease. They calculated LV volumes from a series of real-time, parasternal, short-axis images (7 to 10 images) acquired with a line-of-intersection display as a guide. This line is computed and displayed in each image to indicate the position and orientation of that image with respect to the other image. All images for ventricular reconstruction are acquired during suspended respiration. A polyhedral surface reconstruction algorithm has been adapted for LV volume computation that uses the traced endocardial borders of the short-axis image. In patients in the study by Gopal et al,21 LVEF measured by radionuclide angiography ranged from 9% to 75%, with a mean of 47%. They reported a very good correlation between 3D echocardiography and radionuclide angiography (r=.94 and r=.98 for the two 3D echocardiographic observers, respectively, in their study).
Sapin et al22 achieved higher correlation between LV end-diastolic and end-systolic volume measured by 3D echocardiography using the same method described by Gopal et al21 and cineventriculography (for the two observers in the study, r=.97 and r=.98, respectively) than between 2D echocardiography and cineventriculography (for the two observers, r=.85 and r=.91, respectively). The same authors22 did not achieve a corresponding improvement in the calculation of LVEF by 3D echocardiography over 2D echocardiography (for the two observers in the study, r=.82 and r=.80, respectively). In the same study,22 the agreement (mean difference±2 SD) of 3D echocardiographic end-diastolic and end-systolic volume measurements with that of cineventriculography (12.9±25.4 and −0.7±24.8, respectively) was better than that of 2D echocardiography (21.1±54 and 2.4±44.8, respectively), whereas for measurement of the ejection fraction, a corresponding improvement in the agreement between 3D and 2D echocardiography with that of cineventriculography (6.6±19.6 and 7.5±21.4, respectively) was not obtained. Sapin et al explained this discrepancy by the possible balance of errors in measurement of end-diastolic and end-systolic volumes by 2D echocardiography and cineventriculography. These errors were nullified when ejection fraction was calculated. On the other hand, errors in measurement of end-diastolic and end-systolic volume by 3D echocardiography were not correlated because they were derived from multiple cardiac cycles from which errors of ejection fractions were obtained.
Sapin et al22 found a lower correlation and wider limit of agreement for LVEF calculation compared with the study by Gopal et al21 and the results of the present study. This may be related to the different reference method used by Sapin et al (cineventriculography). Differences in methodology may also explain in part the discrepancy between our results and those of Sapin et al.22 We performed LV imaging for 3D reconstruction from the apical window, which allows the simultaneous identification of basal and apical landmarks for the long-axis determination. Sapin et al used the parasternal window, from which it is often impossible to define these landmarks in a single cross section and which requires two different long-axis views. 3D image acquisition is then guided by a constructed line of intersection that is derived from two temporally dispersed short-axis views selected from the two parasternal long-axis views. Each acquired short axis is then readjusted by displaying this line of intersection several times. We used the standard apical four-chamber view in which the longest LV axis for rotational acquisition is considered to encompass the entire LV cavity. In our technique, image acquisition is ECG and respiratory gated, so that the cardiac images fall at the same moment in each accepted cardiac cycle. This allows more accurate 3D LV reconstruction and volume calculation. Sapin et al22 acquired their basic images at suspended end expiration, which is likely to produce motion artifacts. The data set must then be discarded and the procedure repeated. We reconstructed the heart from 90 cross-section planes, which allows faithful reconstruction of the true geometry, particularly in the presence of asynergy. Sapin et al22 used eight or nine slices with wider gaps and hence the possibility of missing asynergic segments.
In the present study, the intraobserver and interobserver variabilities for 3D echocardiography by Simpson's method were at least similar to that of radionuclide angiography. This can be attributed to the fact that we controlled the image acquisition by ECG- and respiratory-gating techniques and improved image selection by identifying the rotational long axis.
In the study by Gopal et al,21 intraobserver and interobserver standard errors of the estimate of LVEF measurements obtained by 3D echocardiography were one third to one half that of 2D echocardiography (3.4% to 5.5% versus 7.5% to 9.0%, respectively). The limits of agreement also showed no systematic overestimation or underestimation of LVEF by 3D echocardiography.
The present data showed that there were no significant differences between LV end-diastolic and end-systolic volumes and ejection fraction calculated by 3D echocardiography at a slice thickness of 3 mm and those calculated from slices that were 6 to 24 mm thick (with stepwise increments of 3 mm). Nevertheless, with an increase in slice thickness, there was a correlated, corresponding increase in the SD of the mean differences of LV end-diastolic and end-systolic volumes and ejection fraction, particularly at thicknesses >15 mm (Table 3⇑). There was no significant difference between LVEF calculated by radionuclide angiography and that calculated by 3D echocardiography at different slice thicknesses from 3 to 24 mm (with stepwise increments of 3 mm). Again, with an increase in slice thickness, there was a correlated, corresponding increase in the SD of the mean differences of LVEF, particularly at thicknesses >15 mm (Table 4⇑; Fig 5⇑). Accordingly, use of a slice thickness of 15 mm instead of 3 mm will overcome the major limitation for the routine use of 3D echocardiography for calculation of LV volume and ejection fraction, because it reduces the number of short-axis slices from 20 to 30 (for a 3-mm-thick slice) to 5 to 8 (for a 15-mm-thick slice), which consequently dramatically reduces the analysis time from 40 minutes to 10 minutes on average.
In our group of patients, we calculated the probability of 3D echocardiography by Simpson's method at a 15-mm slice thickness to detect several differences of LVEF (Table 5⇑). There was an acceptable probability of 80% to detect clinically relevant (>5%) differences. Obviously, a larger study population would be required to detect small differences (≤5%). By using a 15-mm-thick slice, we reduced the analysis time because a lower number of short-axis slices were used. The analysis time for a 15-mm-thick slice was 10 minutes versus the 40 minutes needed when a 3-mm-thick slice was used. This eliminated the limitation imposed on 3D echocardiography and provides a rapid and accurate method for calculating LVEF.
Advantages and Limitations of the Study
3D echocardiography allows calculation of LV volumes and ejection fraction without any geometric assumption. Image processing of ultrasonic data is complicated by the presence of artifacts and a substantial amount of noise in the image. With our technique, image conditioning is performed with an ROSA (Reduction of Spatial Artifacts) filter. Image acquisition in a given plane is controlled by an algorithm that considers both ECG- and respiratory phase–gated technology so that the computer accepts cycles that fall in the preset range. On the basis of this information, cycles that do not meet the preset range are rejected, which avoids rotational and movement artifacts and allows optimal temporal and spatial registration of the cardiac images. From the volumetric data set, it is possible to adjust the reference image with the longest apical long axes that guide the short-axis series and allow more accurate calculation of LV volume. The time factor is the most important limiting factor that currently restricts the routine use of 3D echocardiography. Development of faster computers and application of automated border-recognition software for area- and volume-analysis techniques will shorten the time needed for image acquisition, postprocessing, and data analysis. In addition, the use of a slice thickness up to 15 mm will reduce the analysis time. Prolonged acquisition time increases the chance of patient motion or rotation artifacts, although these can be prevented by thoroughly explaining the procedure to the patient.
Radionuclide angiography was not used in the present study to calculate LV volumes. Because LVEF is more tolerant of volume errors than is absolute volume, there is still a need to validate this technique for volume measurement with other well-established techniques. In the present study, patients with good echocardiographic image quality were selected. Thus, the value of 3D echocardiography independent of image quality remains to be proved. A larger population is required for the probability study of 3D echocardiography by Simpson's method with the use of 15-mm slice thickness to detect smaller differences (<5%) in LVEF.
3D echocardiography shows excellent correlation with radionuclide angiography for the calculation of LVEF. 3D echocardiography by Simpson's method has an intraobserver and interobserver variability that is at least equivalent to that of radionuclide angiography. Therefore, 3D echocardiography may be a preferable test when serial assessment of LVEF is requested. On the basis of the data obtained with different slice thicknesses, we suggest the use of a slice thickness up to 15 mm for accurate and rapid measurement of LV volume and ejection fraction.
Dr Nosir is supported by the NUFFIC, The Hague, Netherlands, and the Cardiology Department, Al-Hussein University Hospital, Al-Azhar University, Cairo, Egypt. We thank Jan H. Cornel, MD, for his cooperation throughout the study, Rene´ Frowijn for his technical assistance throughout the work and for preparing the images for the manuscript, and David Kean, MD, for reviewing the manuscript.
Selected Abbreviations and Acronyms
|LVEF||=||left ventricular ejection fraction|
- Received September 19, 1995.
- Revision received December 27, 1995.
- Accepted February 3, 1996.
- Copyright © 1996 by American Heart Association
Becker LC, Silverman KS, Bulkley BH, Kallman CH, Mellits ED, Weisfeldt M. Comparison of early thallium-201 scintigraphy and gated blood pool imaging for predicting mortality in patients with acute myocardial infarction. Circulation. 1983;67:1272-1282.
Pfeffer MA, Braunwald E, Moye LA, Basta L, Brown EJ, Cuddy TE, Davis BR, Geltman EM, Goldman S, Flaker GC, Klein M, Lamas GA, Packer M, Rouleau J, Rouleau JL, Rutherford J, Wertheimer JH, Hawkins CM. Effect of captopril on mortality and morbidity in patients with left ventricular dysfunction after myocardial infarction. N Engl J Med. 1992;327:669-677.
Hains AD, Khawaja IA, Lahiri A, Raftery FB. Radionuclide left ventricular ejection fraction: a comparison of three methods. Br Heart J. 1987;57:232-236.
Wyatt HL, Heng MK, Meerbaum S, Gueret P, Hestenes J, Dula E, Corday E. Cross-sectional echocardiography: analysis of mathematic models for quantifying volume of the formalin-fixed left ventricle. Circulation. 1980;61:1119-1125.
Garrison JB, Weiss JL, Maughan WL, Tuck OM, Guier WH, Fortuin NJ. Quantifying regional wall motion and thickening in two-dimensional echocardiography with a computer-aided contouring system. In: Ostrow H, Ripley K, eds. Proceedings in Computerized Cardiology. Long Beach, CA: 1977:25.
Siu SC, Rivera JM, Guerrero JL, Handschumacher MD, Lethor JP, Weyman AE, Levine RA, Picard MH. Three-dimensional echocardiography: in vivo validation for left ventricular volume and function. Circulation. 1993;88:1715-1723.
Handschumacher MD, Lethor JP, Siu SC, Mele D, Rivera JM, Picard MH, Weyman AE, Levine RA. A new integrated system for three-dimensional echocardiographic reconstruction: development and validation for ventricular volume with application in human subjects. J Am Coll Cardiol. 1993;21:743-753.
King DL, King DL Jr, Shao MY-C. Evaluation of in vitro measurement accuracy of a three-dimensional ultrasound scanner. J Ultrasound Med. 1991;10:77-82.
Roelandt J, Salustri A, Vletter W, Nosir Y, Bruining N. Precordial multiplane echocardiography for dynamic anyplane, paraplane and three-dimensional imaging of the heart. Thoraxcentre J. 1994;6:4-13.
Nosir YFM, Salustri A, Vletter WB, Cornel JH, Geleijnse ML, Krenning E, Fioretti P. Accurate measurements of left ventricular ejection fraction: radionuclide angiography versus three-dimensional echocardiography. Circulation. 1995;92:278. Abstract.
Gopal AS, Sapin PM, Shen Z, Sapin PM, Keller AM, Schnellbaecher MJ, Leibowitz DW, Akinboboye OO, Rodney RA, Blood DK, King DL. Assessment of cardiac function by three-dimensional echocardiography compared with conventional noninvasive methods. Circulation. 1995;92:842-852.