- Email: [email protected]
Transthoracic three-dimensional echocardiographic volumetry of distorted left ventricles using rotational scanning
Transthoracic Three-Dimensional Echocardiographic Volumetry of Distorted Left Ventricles Using Rotational Scanning Iri Kupferwasser, MD, Susanne Mohr-Kahaly, MD, Peter St~ihr, MD, Hans-Jtirgen Rupprecht, MD, Uwe Nixdorff, MD, Matthias Fenster, Thomas Voigtlg~nder, MD, Raimund Erbel, and J~irgen Meyer, MD,* Mainz and Essen, Germany
The purpose of this study was to evaluate the relation of transthoracic three- and two-dimensional echocardiographic left ventricular volumetry to cineventriculographic volumetry. Twenty-five patients with distorted left ventricles were included in the study. To demonstrate the impact of acquiring data by rotational scanning, we performed three- and two-dimensional echocardiography in 36 latex ventricles with data acquisition in different areas of the ultrasound sectors. Interobserver and intraobserver variability were calculated to test for reproducibility. The three-dimensional imaging system consisted of a rotation motor device, a transthoracic 2.5 M H z transducer, a conventional ultrasound unit, and a workstation (TomTec) which provides data acquisition, postprocessing, and two- or three-dimensional visualization of digitized data. The transducer moved automatically at 2-degree increments with data acquisition at each tomographic level. The mean investigation time for threedimensional echocardiography was 21 - 6 minutes. In the central near field of the transducer, differences from
Two-dimensionalechocardiographic volumetry has become a major tool in the noninvasive, serial assessm e n t of myocardial function. 1-6 Nevertheless, the technique has several limitations. Endocardial border detection problems, the foreshortening o f the left ventricular apex, the requirement o f geometric assumptions o f the left ventricle, image plane positioning, and volume computation algorithms may lead echocardiographic volumetry to provide erroneous results, especially in patients with a distorted left ventricle. 4 Cineventriculography represents another frequently used quantitative approach in left ventricular volume determination but is limited, on the other hand, by its invasive nature. The m e t h o d also
From the II Medical Clinic, University of Mainz, and Division of Cardiology, University of Essen.* Reprint requests: Iri Kupferwasser,MD, Adult Infectious Diseases, Harbor-UCLA Medical Center, Bldg. RB-2, 1000 West Carson, Torrance, CA 90509. Copyright © 1997 by the American Society of Echocardiography. 0894-7317/97 $5.00+ 0 27/1/83467
840
true volumes in latex ventricles were remarkably smaller for three-dimensional compared with two-dimensional echocardiography (root mean square percent error: threedimensional echocardiography = 5.3% versus two-dimensional echocardiography = 14.6%). In three-dimensional echocardiography, there was considerable overestimation of volumes in the lateral far field (root mean square percent error = 13.2%) of the ultrasound sector. Differences between two-dimensional echocardiographic human left ventrioflar volumes and cineventriculography increased with larger volumes. In three-dimensional echocardiography the differences remained constant. Interobserver and intraobserver variability is reduced nearly twofold by three-dimensional echocardiography. Three-dimensional echocardiographic volumetry provides fewer discrepancies to cineventriculography and lesser variability than two-dimensional echocardiography. With the use of rotational scanning, the ventricle has to be positioned in the central near field of the transducer. (J Am Soc Echocardiogr 1997;10:840-52.)
relies on geometric assumptions concerning the left ventricular shape which may not be valid in ventricular distortion. Considerable measurement errors that have had an impact on clinical decision making have been observed in comparisons o f two-dimensional echocardiographic and cineventriculographic volume ~ y 5 , 7-10
Echocardiographic scanning in multiple planes had been suggested early but could not be realized in the clinical setting because of the demands it placed on computation. 3,n-is Since the recent development o f sophisticated hardware and improved software techniques, several three-dimensional echocardiographic systems with different approaches to data acquisition and volume determination have been introduced. 16-27 Previous studies have reported a high accuracy and reproducibility for three-dimensional echocardiographic volume determinationfl 8-31 This study was undertaken to determine the relation o f three-dimensional echocardiographic left ventricular volumetry to cineventriculographic volumetry. To this end we compared (1) two- and three-dimensional echo-
lournal of the ~nerican Society of Echocardiography Volume 10 Number 8
cardiograpNc and cineventriculographic volume measurements to true volumes in vitro, (2) transthoracic two- and three-dimensional echocardiographic to cineventriculographic volume measurements in vivo in distorted h u m a n lcft ventricles, and (3) the interobserver and intraobserver variability o f the different imaging methods. The: impact o f the data acquisition technique o n the volume measurements has also been evaluated.
Kupferwasser et al.
841
were made to avoid foreshortening of the ventricles. The magnification factor was determined by filming a metal ball with a volume of 180 ml. The ventricular outlines were traced, and in human beings the silhouettes of the papillary muscles were included in the left ventricular cavity. The end-diastolic, end-systolic, and model volumes were calculated with the method of discs. In human subjects the ejection fraction was also calculated. 32,a3
T w o - D i m e n s i o n a l Echocardiographic V o l u m e Calculation METHODS All two- and three-dimensional echocardiographic and the cineventriculographic measurements were performed in each heart model and patient indepcndcntly by two different observers. M1 the measurements were repeated by the first observer at least 2 weeks after the first measurement.
Heart Models A total of 36 latex ventricles with a volume ranging between 45 and 300 ml in 18 heart models were investigated in a water bath. Eighteen models represented nondistorted ventricles, the other 18 represented symmetrically or asymmetrically distorted ventricles. The models were filled with water which was measured before and after each investigation to ensure that the true volume remained constant during the imaging procedure.
Patients A total of 28 patients ( 16 men and 12 women, mesas age 54 years, range 29 to 72 years) were investigated between October 1994 and December 1994 by two- and three-dimensional transthoracic echocardiography and heart catheterization after informed consent. In three patients three-dimensionaldata sets were not of diagnostic quality; these patients were excluded from the study. All patients were referred to the catheterization laboratory for clinically indicated examinations. Six patients had a dilated cardiomyopathy, eight had a hypertrophic cardiomyopathy or left ventricularwall thickening caused by arterial hypertension, and seven had a left ventricular aneurysm after myocardial infarction. In four patients regional wall motion abnormalities were obvious. The mean time between echocardiography and cineventriculographywas 1 day (range 0 to 3 days). Patients in whom the left ventricular shape was obviously physiologic and free of abnormalities were not included in the study.
Cineventriculography After positioning a pigtail catheter in a human left ventricle, contrast medium was applied automatically by an injection device. Images of the left ventricle in patients were obtained in the 30-degree right anterior oblique projection and in the 60-degree left anterior oblique projection. The heart models were filled with the defined volume of contrast medium and positioned in the field of the cineangiographic system in away that the same projections were used in the latex ventricles as were in the patients. Special attempts
Two-dimensionalimages were obtained completely separately from thrce-dimensionalechocardiographic data acquisition. A conventional real-time, hand-held, two-dimensional scanner (Vingmed) with a 2.5 MHz transducer was used. Ventricular borders were traced manually at the black-white interface betwecn the endocardium/latex and the cavity. In human beings, the papillary muscles were excluded from the ventricular volume when continuous to the endocardium. Tracings were performed in the apical two-chamber and four-chamber views and in the parasternal short-axis view at the papillary muscle level. The corresponding images were obtained in the latex ventricles. End-diastolic and end-systolic volumes in patients and static volumes in the heart models were then calculated with the biplane summation of discs method (modified Simpson's role) according to the recommendations of the American Society of Echocardiography. 3,s4,ss In human subjects the ejection fraction was calculated. Heart models were investigated at three different points of the ultrasound sector: first in the central near field (2 to 9 cm), second in the central far field (9 to 18 cm), and third in the lateral far field.
T h r e e - D i m e n s i o n a l Echocardiographic Data Acquisition The system consists of a conventional transthoracic realtime, two-dimensional scanner with a 2.5 MHz transducer and an ultrasound unit (Vingmed CFM 800) linked to a workstation (TomTec) capable of three-dimensional, electrocardiographic, and respiration gated data acquisition, postprocessing, dynamic three-dimensional reconstruction, and volumctry. Before starting automatic data acquisition, a manual maneuver was performed that represented the movement of the transducer during a scan to ensure that a maximum amount of the left ventricular chamber or heart model was visualized and areas of lateral dropout in the near field of the transducer were avoided. The transducer was positioned in a manner that the apex as well as the basis of a ventricle was imaged during the scan. The transducer was inside a stepper motor which was manually held stable at one point in the water bath or on the chest wall, providing apical views of a ventricle. The motor advanced the transducer automatically at 2-degree over a span of 180 degrees. Data of a complete heart cycle were obtained when the length of a heart and respiration cycle was within operator defined trigger limits,s6 Depth of the ultrasound sector was chosen so that the left ventricle could be fully visualized. Heart models were investigated at the same locations in the ultrasound sector as in two-dimen-
842
Kupferwasser et al.
Journalof the AmericanSocietyof Echocardiography October 1997
Figure 1 Three-dimensional reconstruction of a heart model. The upper panel (a) shows photographs of the original model; the lower part (b) shows a high-quality three-dimensional reconstruction.
sional echocardiography. For in vivo investigations the transducer was positioned in a way that the ventricle was localized in the central near field of the ultrasound sector during the scan.
Three-Dimensional Echocardiographic Reconstruction and Volumetry Postprocessing was performed automatically after data acquisition was terminated. During postprocessing the rotated images are geometrically converted into cubic data sets. The gaps between the original tomographic planes are filled by interpolation techniques with the digitized data of the neighboring planes. After postprocessing was completed, a two-dimensional apical view of a ventricle or a heart model was selected and a long axis from the apex to the mitral anulus was defined by the operator. In a variable number of cross-sectional planes which were automatically visualized and exactly orthogonal to the long axis, the endocardial border of a ventricle was manually traced analogous to the border tracing conventions in two-dimen-
sional echocardiographic volumetry. The number of crosssectional planes is dependent on the distance between these planes, which is chosen by the operator. The distance was 8 mm in this study. This distance also determines the height of the cross-sectional slices. In this way, a series of contiguous slices were achieved, each forming part of the cavity of the ventricle. According to the disk method the volume of each slice is calculated by multiplying the traced area of a cross-sectional plane by the height of a cross-sectional slice. End-diastolic, end-systolic, or the static volumes of heart models volume are obtained by adding the volumes of the slices. The ejection fraction was calculated in the standard manner. A ventricle was visualized in shaded surface techniques and as a wire frame diagram with the traced borders forming the shape of a ventricle.
Statistical Analysis Lincar rcgrcssion analysis was used for comparing two- and three-dimensional echocardiographic and cineventriculographic measurements with true volumes in vitro and for
Journal of the American Society of Echocardiography Volume 10 Number 8
Kupferwasser et al.
843
Figure 2 Aneurysmatic left ventricle. Visualization by cineventriculography (a) and threedimensional echocardiography by shaded surface techniques (b). x, Small aneurysm at thc left ventricular apex; LV, left ventricle; MV, mitral valve; LA, left atrium.
comparing two- and three-dimensional echocardiography with cineventriculography in vivo. The mean difference and its standard deviation (SD) were calculated as a function of increasing volume. Different methods were compared by means of the analysis described by Bland and Altman. 37 In this analys!s the difference between two in vivo measurements is plotted against their mean, For in vitro measurements, the true volume is used as the standard. The mean difference between two methods is regarded as the systematic error (bias) and the mean -+ 2 SD defines the limits of agreement.i To assess the difference between measured and true volumes and between measured volumes of two different methods , the Wilcoxon test was used. A p value < 0.05 was considered statistically significant. Accuracy was calculated as the root mean square percent error between meat sured and true volumes in vitro. Interobserver (intraobserver) variability for each method was expressed as the SD of differences between the measurements of two independent obseNers (between the repeated measurements of one observer) divided by the mean of the measured values expressed a~ percentage. One observer was an experienced echocardiographer, one was less experienced. Differences between variabilities were compared by means of the Wilcoxon test.
RESULTS
High-quality, three-dimensional images were obtained in all o f the latex ventricles and in 25 o f 28 patients. Figures 1 and 2 show three-dimensional visualizations o f a heart m o d e l and a h u m a n left
ventricle in shaded-surface technique. L o w resolution problems occurred w h e n ventricles were localized in the lateral far field o f the transducer. T h e mean investigation time for three-dimensional echocardiography in h u m a n beings including data acquisition (mean time 3 minutes), postprocessing (mean time 9 minutes), and end-diastolic and end-systolic volumetry (mean time 9 minutes) was 21 -+ 6 minutes (range 16 to 29 minutes). In Vitro
The regression equations for the in vitro measurements are shown in Table 1. For three-dimensional echocardiographic volumetry, the lowest standard error o f the estimate and difference from true volumes were obtained in the central near field o f the ultrasound sector. Here, differences from zero were n o t statistically significant. The scatterplots in Figure 3 s h o w the larger spread o f values for two-dimensional echocardiographic measurements, which increased for true volumes a l t h o u g h the differences from zero were n o t statistically significant. The spread o f values for three-dimensional measurements was lower and remained constant with increasing true volumes. For three-dimensional volumetric measurements in the central far and lateral far field o f the ultrasound sector (Figure 4), overestimation increased with larger true volumes. H e r e differences from zero were statistically significant. Overestimation was largest in
Journal of the American Society of Echocardiography
844
Kupferwasser et al.
Table 1
October1997
In vitro measuremcnts: regression equations, correlation, and difference from true volumes
x
y
True
3 d - near field
True True
Regression
r
SEE (nil)
Percent error (%)
p Value
y = 1.04x-7
0.99
8.2
5.3
0.72
3 d - central far field
y = 1.14×-9.2
0.99
9.8
8.6
0.003
3 d - lateral
y = 1.14×-2.7
0.99
12.8
13.2
0.001
far field True
2 d - near field
y = 1.03×-10.5
0.96
20.1
14.6
0.29
True
3d - central
y = 1.05X-4.8
0.95
19.4
14.1
0.17
y = 1.02× +2.5
0.96
18.9
13.5
0.21
y = 1.03×-0.4
0.98
12.1
12.3
0.05I
far field True
2 d - lateral far field
True
Cine
r, Correlation parameter; SEE, standard error of estimate; true, true volume of heart models; 3d/2d, three-/two-dimensional echocardiographic volumetry; Cine, cineventriculographic volumetry; Percent error (%), root mean square percent error between measurement and standard; p, p value difference from zero.
the lateral far field of the ultrasound sector. Measurements obtained by two-dimensional echocardiography in the central far and lateral far field showed a comparable regression curve and spread of values as in the central near field. Although the root mean square percent error were comparable with threedimensional echocardiography in these areas of the ultrasound sector, no single trend toward overestimation or underestimation could be observed for twodimensional echocardiography. This is in contrast to three-dimensional echocardiography. For two-dimensional echocardiography differences from zero were not statistically significant in these parts of the ultrasound sector. Cineventriculographic measurements showed a constant spread of values with increasing true volumes. The root mean square percent error was comparable with two-dimensional echocardiography. Differences from zero were at the threshold of significance. In Figure 5 the vertical dimension o f the box indicates the spread o f values. The relation of a box to the zero line and the position o f the mean o f differences show a trend toward overestimation or underestimation. Two-dimensional echocardiography showed a trend toward undercstimation. Three-dimensional echocardiography and cineventriculography showed slight overestimation. The lowest spread of values is reached by three-dimensional echocardiography, the largest by two-dimensional echocardiography. The cineventriculographic measurements showed a slightly larger spread o f values than threedimensional measurements.
In Vivo
For end-diastolic measurements, the values for interobscrvcr/intraobserver variability were lower for three-dimensional volumetry (4.7% _+ 1.7%/2.4% _+ 1.7%) than for two-dimensional volumetry (9.2% _+ 3.1%/7.1% + 4.8%; p < 0.05) and cineventriculography (6.1% + 2 £ % / 4 . 9 % -+ 2.1%; not significant). The same is true for end-systolic volumes threedimensional echocardiography (4.4% _+ 2.5%/3.1% +- 3.6%), which offers lower variability than twodimensional echocardiography (8.6% -+ 5.6%/5.7% -+ 5%; p < 0 . 0 5 ) a n d cineventriculography (6.6% +4.1%/5.7% + 2.8%; not significant). The regression equations and the mean differences arc shown in Table 2. An excellent correlation betwecn three-dimensional echocardiographic measurcmcnts and cineventriculography was obtained for end-diastolic volumes, end-systolic volumes, and ejection fraction. Mean differences were significantly lower for three-dimensional echocardiographic enddiastolic (p < 0.05) and end-systolic (p < 0.01) volumes than for two-dimensional echocardiography. The scatterplots in Figure 6 show a systematic undcrestimation o f the cnd-diastolic cineventriculographic volumes by both of the echocardiographic methods. For small volumes, underestimation was low and comparable for both o f the methods. Underestimation o f large volumes was rcmarkably lower for the three-dimensional method. The degree o f underestimation remained constant with increasing volumes, which is in contrast to the two-dimensional mcthod, where increasing volumes were accompanied by increasing underestimation and a
The purpose of this study was to evaluate the relation of transthoracic three- and two-dimensional echocardiographic left ventricular volumetry to cineventriculographic volumetry. Twenty-five patients with distorted left ventricles were included in the study. To demonstrate the impact of acquiring data by rotational scanning, we performed three- and two-dimensional echocardiography in 36 latex ventricles with data acquisition in different areas of the ultrasound sectors. Interobserver and intraobserver variability were calculated to test for reproducibility. The three-dimensional imaging system consisted of a rotation motor device, a transthoracic 2.5 M H z transducer, a conventional ultrasound unit, and a workstation (TomTec) which provides data acquisition, postprocessing, and two- or three-dimensional visualization of digitized data. The transducer moved automatically at 2-degree increments with data acquisition at each tomographic level. The mean investigation time for threedimensional echocardiography was 21 - 6 minutes. In the central near field of the transducer, differences from
Two-dimensionalechocardiographic volumetry has become a major tool in the noninvasive, serial assessm e n t of myocardial function. 1-6 Nevertheless, the technique has several limitations. Endocardial border detection problems, the foreshortening o f the left ventricular apex, the requirement o f geometric assumptions o f the left ventricle, image plane positioning, and volume computation algorithms may lead echocardiographic volumetry to provide erroneous results, especially in patients with a distorted left ventricle. 4 Cineventriculography represents another frequently used quantitative approach in left ventricular volume determination but is limited, on the other hand, by its invasive nature. The m e t h o d also
From the II Medical Clinic, University of Mainz, and Division of Cardiology, University of Essen.* Reprint requests: Iri Kupferwasser,MD, Adult Infectious Diseases, Harbor-UCLA Medical Center, Bldg. RB-2, 1000 West Carson, Torrance, CA 90509. Copyright © 1997 by the American Society of Echocardiography. 0894-7317/97 $5.00+ 0 27/1/83467
840
true volumes in latex ventricles were remarkably smaller for three-dimensional compared with two-dimensional echocardiography (root mean square percent error: threedimensional echocardiography = 5.3% versus two-dimensional echocardiography = 14.6%). In three-dimensional echocardiography, there was considerable overestimation of volumes in the lateral far field (root mean square percent error = 13.2%) of the ultrasound sector. Differences between two-dimensional echocardiographic human left ventrioflar volumes and cineventriculography increased with larger volumes. In three-dimensional echocardiography the differences remained constant. Interobserver and intraobserver variability is reduced nearly twofold by three-dimensional echocardiography. Three-dimensional echocardiographic volumetry provides fewer discrepancies to cineventriculography and lesser variability than two-dimensional echocardiography. With the use of rotational scanning, the ventricle has to be positioned in the central near field of the transducer. (J Am Soc Echocardiogr 1997;10:840-52.)
relies on geometric assumptions concerning the left ventricular shape which may not be valid in ventricular distortion. Considerable measurement errors that have had an impact on clinical decision making have been observed in comparisons o f two-dimensional echocardiographic and cineventriculographic volume ~ y 5 , 7-10
Echocardiographic scanning in multiple planes had been suggested early but could not be realized in the clinical setting because of the demands it placed on computation. 3,n-is Since the recent development o f sophisticated hardware and improved software techniques, several three-dimensional echocardiographic systems with different approaches to data acquisition and volume determination have been introduced. 16-27 Previous studies have reported a high accuracy and reproducibility for three-dimensional echocardiographic volume determinationfl 8-31 This study was undertaken to determine the relation o f three-dimensional echocardiographic left ventricular volumetry to cineventriculographic volumetry. To this end we compared (1) two- and three-dimensional echo-
lournal of the ~nerican Society of Echocardiography Volume 10 Number 8
cardiograpNc and cineventriculographic volume measurements to true volumes in vitro, (2) transthoracic two- and three-dimensional echocardiographic to cineventriculographic volume measurements in vivo in distorted h u m a n lcft ventricles, and (3) the interobserver and intraobserver variability o f the different imaging methods. The: impact o f the data acquisition technique o n the volume measurements has also been evaluated.
Kupferwasser et al.
841
were made to avoid foreshortening of the ventricles. The magnification factor was determined by filming a metal ball with a volume of 180 ml. The ventricular outlines were traced, and in human beings the silhouettes of the papillary muscles were included in the left ventricular cavity. The end-diastolic, end-systolic, and model volumes were calculated with the method of discs. In human subjects the ejection fraction was also calculated. 32,a3
T w o - D i m e n s i o n a l Echocardiographic V o l u m e Calculation METHODS All two- and three-dimensional echocardiographic and the cineventriculographic measurements were performed in each heart model and patient indepcndcntly by two different observers. M1 the measurements were repeated by the first observer at least 2 weeks after the first measurement.
Heart Models A total of 36 latex ventricles with a volume ranging between 45 and 300 ml in 18 heart models were investigated in a water bath. Eighteen models represented nondistorted ventricles, the other 18 represented symmetrically or asymmetrically distorted ventricles. The models were filled with water which was measured before and after each investigation to ensure that the true volume remained constant during the imaging procedure.
Patients A total of 28 patients ( 16 men and 12 women, mesas age 54 years, range 29 to 72 years) were investigated between October 1994 and December 1994 by two- and three-dimensional transthoracic echocardiography and heart catheterization after informed consent. In three patients three-dimensionaldata sets were not of diagnostic quality; these patients were excluded from the study. All patients were referred to the catheterization laboratory for clinically indicated examinations. Six patients had a dilated cardiomyopathy, eight had a hypertrophic cardiomyopathy or left ventricularwall thickening caused by arterial hypertension, and seven had a left ventricular aneurysm after myocardial infarction. In four patients regional wall motion abnormalities were obvious. The mean time between echocardiography and cineventriculographywas 1 day (range 0 to 3 days). Patients in whom the left ventricular shape was obviously physiologic and free of abnormalities were not included in the study.
Cineventriculography After positioning a pigtail catheter in a human left ventricle, contrast medium was applied automatically by an injection device. Images of the left ventricle in patients were obtained in the 30-degree right anterior oblique projection and in the 60-degree left anterior oblique projection. The heart models were filled with the defined volume of contrast medium and positioned in the field of the cineangiographic system in away that the same projections were used in the latex ventricles as were in the patients. Special attempts
Two-dimensionalimages were obtained completely separately from thrce-dimensionalechocardiographic data acquisition. A conventional real-time, hand-held, two-dimensional scanner (Vingmed) with a 2.5 MHz transducer was used. Ventricular borders were traced manually at the black-white interface betwecn the endocardium/latex and the cavity. In human beings, the papillary muscles were excluded from the ventricular volume when continuous to the endocardium. Tracings were performed in the apical two-chamber and four-chamber views and in the parasternal short-axis view at the papillary muscle level. The corresponding images were obtained in the latex ventricles. End-diastolic and end-systolic volumes in patients and static volumes in the heart models were then calculated with the biplane summation of discs method (modified Simpson's role) according to the recommendations of the American Society of Echocardiography. 3,s4,ss In human subjects the ejection fraction was calculated. Heart models were investigated at three different points of the ultrasound sector: first in the central near field (2 to 9 cm), second in the central far field (9 to 18 cm), and third in the lateral far field.
T h r e e - D i m e n s i o n a l Echocardiographic Data Acquisition The system consists of a conventional transthoracic realtime, two-dimensional scanner with a 2.5 MHz transducer and an ultrasound unit (Vingmed CFM 800) linked to a workstation (TomTec) capable of three-dimensional, electrocardiographic, and respiration gated data acquisition, postprocessing, dynamic three-dimensional reconstruction, and volumctry. Before starting automatic data acquisition, a manual maneuver was performed that represented the movement of the transducer during a scan to ensure that a maximum amount of the left ventricular chamber or heart model was visualized and areas of lateral dropout in the near field of the transducer were avoided. The transducer was positioned in a manner that the apex as well as the basis of a ventricle was imaged during the scan. The transducer was inside a stepper motor which was manually held stable at one point in the water bath or on the chest wall, providing apical views of a ventricle. The motor advanced the transducer automatically at 2-degree over a span of 180 degrees. Data of a complete heart cycle were obtained when the length of a heart and respiration cycle was within operator defined trigger limits,s6 Depth of the ultrasound sector was chosen so that the left ventricle could be fully visualized. Heart models were investigated at the same locations in the ultrasound sector as in two-dimen-
842
Kupferwasser et al.
Journalof the AmericanSocietyof Echocardiography October 1997
Figure 1 Three-dimensional reconstruction of a heart model. The upper panel (a) shows photographs of the original model; the lower part (b) shows a high-quality three-dimensional reconstruction.
sional echocardiography. For in vivo investigations the transducer was positioned in a way that the ventricle was localized in the central near field of the ultrasound sector during the scan.
Three-Dimensional Echocardiographic Reconstruction and Volumetry Postprocessing was performed automatically after data acquisition was terminated. During postprocessing the rotated images are geometrically converted into cubic data sets. The gaps between the original tomographic planes are filled by interpolation techniques with the digitized data of the neighboring planes. After postprocessing was completed, a two-dimensional apical view of a ventricle or a heart model was selected and a long axis from the apex to the mitral anulus was defined by the operator. In a variable number of cross-sectional planes which were automatically visualized and exactly orthogonal to the long axis, the endocardial border of a ventricle was manually traced analogous to the border tracing conventions in two-dimen-
sional echocardiographic volumetry. The number of crosssectional planes is dependent on the distance between these planes, which is chosen by the operator. The distance was 8 mm in this study. This distance also determines the height of the cross-sectional slices. In this way, a series of contiguous slices were achieved, each forming part of the cavity of the ventricle. According to the disk method the volume of each slice is calculated by multiplying the traced area of a cross-sectional plane by the height of a cross-sectional slice. End-diastolic, end-systolic, or the static volumes of heart models volume are obtained by adding the volumes of the slices. The ejection fraction was calculated in the standard manner. A ventricle was visualized in shaded surface techniques and as a wire frame diagram with the traced borders forming the shape of a ventricle.
Statistical Analysis Lincar rcgrcssion analysis was used for comparing two- and three-dimensional echocardiographic and cineventriculographic measurements with true volumes in vitro and for
Journal of the American Society of Echocardiography Volume 10 Number 8
Kupferwasser et al.
843
Figure 2 Aneurysmatic left ventricle. Visualization by cineventriculography (a) and threedimensional echocardiography by shaded surface techniques (b). x, Small aneurysm at thc left ventricular apex; LV, left ventricle; MV, mitral valve; LA, left atrium.
comparing two- and three-dimensional echocardiography with cineventriculography in vivo. The mean difference and its standard deviation (SD) were calculated as a function of increasing volume. Different methods were compared by means of the analysis described by Bland and Altman. 37 In this analys!s the difference between two in vivo measurements is plotted against their mean, For in vitro measurements, the true volume is used as the standard. The mean difference between two methods is regarded as the systematic error (bias) and the mean -+ 2 SD defines the limits of agreement.i To assess the difference between measured and true volumes and between measured volumes of two different methods , the Wilcoxon test was used. A p value < 0.05 was considered statistically significant. Accuracy was calculated as the root mean square percent error between meat sured and true volumes in vitro. Interobserver (intraobserver) variability for each method was expressed as the SD of differences between the measurements of two independent obseNers (between the repeated measurements of one observer) divided by the mean of the measured values expressed a~ percentage. One observer was an experienced echocardiographer, one was less experienced. Differences between variabilities were compared by means of the Wilcoxon test.
RESULTS
High-quality, three-dimensional images were obtained in all o f the latex ventricles and in 25 o f 28 patients. Figures 1 and 2 show three-dimensional visualizations o f a heart m o d e l and a h u m a n left
ventricle in shaded-surface technique. L o w resolution problems occurred w h e n ventricles were localized in the lateral far field o f the transducer. T h e mean investigation time for three-dimensional echocardiography in h u m a n beings including data acquisition (mean time 3 minutes), postprocessing (mean time 9 minutes), and end-diastolic and end-systolic volumetry (mean time 9 minutes) was 21 -+ 6 minutes (range 16 to 29 minutes). In Vitro
The regression equations for the in vitro measurements are shown in Table 1. For three-dimensional echocardiographic volumetry, the lowest standard error o f the estimate and difference from true volumes were obtained in the central near field o f the ultrasound sector. Here, differences from zero were n o t statistically significant. The scatterplots in Figure 3 s h o w the larger spread o f values for two-dimensional echocardiographic measurements, which increased for true volumes a l t h o u g h the differences from zero were n o t statistically significant. The spread o f values for three-dimensional measurements was lower and remained constant with increasing true volumes. For three-dimensional volumetric measurements in the central far and lateral far field o f the ultrasound sector (Figure 4), overestimation increased with larger true volumes. H e r e differences from zero were statistically significant. Overestimation was largest in
Journal of the American Society of Echocardiography
844
Kupferwasser et al.
Table 1
October1997
In vitro measuremcnts: regression equations, correlation, and difference from true volumes
x
y
True
3 d - near field
True True
Regression
r
SEE (nil)
Percent error (%)
p Value
y = 1.04x-7
0.99
8.2
5.3
0.72
3 d - central far field
y = 1.14×-9.2
0.99
9.8
8.6
0.003
3 d - lateral
y = 1.14×-2.7
0.99
12.8
13.2
0.001
far field True
2 d - near field
y = 1.03×-10.5
0.96
20.1
14.6
0.29
True
3d - central
y = 1.05X-4.8
0.95
19.4
14.1
0.17
y = 1.02× +2.5
0.96
18.9
13.5
0.21
y = 1.03×-0.4
0.98
12.1
12.3
0.05I
far field True
2 d - lateral far field
True
Cine
r, Correlation parameter; SEE, standard error of estimate; true, true volume of heart models; 3d/2d, three-/two-dimensional echocardiographic volumetry; Cine, cineventriculographic volumetry; Percent error (%), root mean square percent error between measurement and standard; p, p value difference from zero.
the lateral far field of the ultrasound sector. Measurements obtained by two-dimensional echocardiography in the central far and lateral far field showed a comparable regression curve and spread of values as in the central near field. Although the root mean square percent error were comparable with threedimensional echocardiography in these areas of the ultrasound sector, no single trend toward overestimation or underestimation could be observed for twodimensional echocardiography. This is in contrast to three-dimensional echocardiography. For two-dimensional echocardiography differences from zero were not statistically significant in these parts of the ultrasound sector. Cineventriculographic measurements showed a constant spread of values with increasing true volumes. The root mean square percent error was comparable with two-dimensional echocardiography. Differences from zero were at the threshold of significance. In Figure 5 the vertical dimension o f the box indicates the spread o f values. The relation of a box to the zero line and the position o f the mean o f differences show a trend toward overestimation or underestimation. Two-dimensional echocardiography showed a trend toward undercstimation. Three-dimensional echocardiography and cineventriculography showed slight overestimation. The lowest spread of values is reached by three-dimensional echocardiography, the largest by two-dimensional echocardiography. The cineventriculographic measurements showed a slightly larger spread o f values than threedimensional measurements.
In Vivo
For end-diastolic measurements, the values for interobscrvcr/intraobserver variability were lower for three-dimensional volumetry (4.7% _+ 1.7%/2.4% _+ 1.7%) than for two-dimensional volumetry (9.2% _+ 3.1%/7.1% + 4.8%; p < 0.05) and cineventriculography (6.1% + 2 £ % / 4 . 9 % -+ 2.1%; not significant). The same is true for end-systolic volumes threedimensional echocardiography (4.4% _+ 2.5%/3.1% +- 3.6%), which offers lower variability than twodimensional echocardiography (8.6% -+ 5.6%/5.7% -+ 5%; p < 0 . 0 5 ) a n d cineventriculography (6.6% +4.1%/5.7% + 2.8%; not significant). The regression equations and the mean differences arc shown in Table 2. An excellent correlation betwecn three-dimensional echocardiographic measurcmcnts and cineventriculography was obtained for end-diastolic volumes, end-systolic volumes, and ejection fraction. Mean differences were significantly lower for three-dimensional echocardiographic enddiastolic (p < 0.05) and end-systolic (p < 0.01) volumes than for two-dimensional echocardiography. The scatterplots in Figure 6 show a systematic undcrestimation o f the cnd-diastolic cineventriculographic volumes by both of the echocardiographic methods. For small volumes, underestimation was low and comparable for both o f the methods. Underestimation o f large volumes was rcmarkably lower for the three-dimensional method. The degree o f underestimation remained constant with increasing volumes, which is in contrast to the two-dimensional mcthod, where increasing volumes were accompanied by increasing underestimation and a