JACC Vol. 22, No. 1
Juiy 1993:258-70
258
DONALD L. KING, JR., MS, DONALD L. KING, MD
New York, New York and Danbury, Connecticut
Objectives. We evaluated a threemdi e~io~al echocardiographic method for ventricular volume and surface area determim
nation that uses polyhedral surface reconstruction. Six to eight
nonparallel, unequally spaced, nonintersecting short-axis planes
were p&tioned with a line of intersection display to overcome
limitations associated with twoqdimensional echocard
Of
~~kg~u~. Twedimensional echocardi
ventricular volume and surface area determi
by
assumptions about ventricular shape snd image plane pa&ion.
ventricular enddiastolic and end-systolic volardiai surface areas determined by threedimensional echocardiography and nuclear magnetic resonance
) imaging were compared in 15 normal subjects (7 men, 8
(
women, aged 23 to 41 years, body surface area 1.38 to 2.17 m’).
‘l’eenof these subjects also underwent tw~dime~ion~
echocardiograpby; and er;:d-oliastoiic and end.systolic volumes were determined by the apical biplane summation of discs meth
and
compared with results of NMR imaging.
Resuk. Interobserver variability was 5% to 8% for threedimensiunai echocardiography and 6% to 9% for NMR imaging.
Left ventricular
vohme is an important clinical variable to
monitor in patients with valvular disease, cardiomyopathy
and myocardial infarction. Therefore, an accurate, repeatable, noninvasive method of determining left ventricular
end-diastolicand end-systolic volumes would be useful, but
to date this has not been widely available. Determinationof
endocardia!surface areas would also be useful for quantifying infarct size and for studying the effects of ventricular
remodeling. Cineangiography has several limitations, the
most important of which is a requirement far geometric
assumptions in the calculation of volume (I). Radisnuclide
methods are subject to other problems related to detection of
edges and end planes as well as determination of background
activity (2).
Attempts to use two-dimensional echocardiography to
determine left ventricular volume identified several limitations, particularlyin patients with regions ofasynergy (3-6).
The ability to predict absolute volume for a given patient has
been hindered by broad variability in results (7,g). Endocardial surface area, likewise, can be estimated by twodimensional echocardiographyonly with the help of geometric assumptions at the current time (3-i ij. Factors
From the Division of Card;ology, Colutnhia
NPW VO; New
..-.- Ilniverritv
-....-.
contributing to the low predictive accuracy (6) of twoYork: and *DivisionofCardiology.
Danburl r Hosoital.
.__ _r . . .._. _Danhttrv.
-_____, , (Innnertir~~t
___.____..__..
dimensional echlocardiograpbicvolume measurements inThis study was presented in part at the 4~ Annual Scicotific Session of the
clude geometric assumptions (i2), image plane positioning
American
College
ofCardiology, D&s, Texas, Amil 1992.
hk~~~script received July i6, 1992; revised manuscript received Decemerrors (13,14)and imprecise endocardial boundary demarcaber 29. 1992, accepted January 7. 1993.
tion.
M&&QrJQP
n,swndence:
Aasha S. Gopal, MD, Columbia University,
A variety of three-dimensional echocardiomphic scanDivision ofCardio!ogy, P&S Building !&Ml,630 West 168 Street, New York,
New York 10032.
ners have been developed primarily to address the issue of
I
01993 by the American College of Cardiology
..,,
.._..
_”
.
.
.
.
0735-1097/P3/$6.00
JAW Vol. 22, No. B
July !!993:258-70
polyhedral surface reconstrucused an algorithm bas
from a series of short-axiscross
tion (34-36)of the ven
sections that are neither parallel nor intersecting. Although
the accuracy of this a~go~tbrnhas been
for other organ systems by ultworaogra
tomography using para.lleIimajges (37):
ventricular volume computati~oniii
our previous in vitro st
method for !eR ventricular volu
to validate our threeend-diastolic and end-systolic volume and surface area determination using NMR imaging as a standard of compari-
computer memory a reference
image is moved, the line of intersection moves in bot
JACC Vol. 22, No. 1
July 1593:258-70
GOPAL ET AL.
THREE-DIMENSIONAL ECHOCARDIOGRAPHY
Figure 1. Three-dimensionalechocardiographicmethod of ventricular volumeand surface area determinationfrom unequallyspaced,
nonparallel,nonintersectingshort-axiscross sections.Top left, First
parastemallong-axisreference image with guided short-axis cross
sections positionedusingthe line of intersectiondisplay. Top right,
Second parasternallong-axisreference image, which includes the
apex. Bottomleft, Real time short-axis image used for boundary
tracingwiththe lineof intersectionof both reference imagesshown.
Bottomright, Traced boundary of digitized real time short-axis
imageto be used for calculationof ventricularvolumeand endocardial surfacearea by polyhedral surface reconstruction.
three-dimensional
echocardiography.
Subjects
were
screened according to the quality of the routine echocardiographic study and selected to provide a wide range of
chamber size. One subject was excluded owing to poor
three-dimensional echocardiographic images. All subjects
were voluntary participants and gave verbal informed consent before the study. All examinations were easily performed it: the left lateral decubrtus position and were of
minimal tt:hnical difficulty. High quality images with easy
and completeidentificationof all structures were availablein
all cases. Images were always acquired at end-expiration
using a defined three-dimensional echocardiographic image
acquisition protocol. If patient motion occurred during image acquisitton, the data set was discarded and the proccdure repeated. First, two temporary short-axis images, one
at the levelof the aortic valve and the other at the apex, were
obtained to define the long axis of the ventricle. Adjusting
the ~arasterna~long-axisimage such that its line of intersection bisected the temporary short-axis images ensured that
the optimal long-axis image had been obtained. This
parasternal long-axis image then served as the reference
image. However, because it is rarely possible to visualize
both the base and the apex using a single parasternal
long-axis reference image, a second long-axis reference
image that included the apex was opttmally positioned using
the line of intersection display and the temporary short-axis
image of the apex as a guide. The temporary short-axis
images were then discarded, and both long-axis reference
images, one excluding the apex and other including it, were
used to position real time short-axis images. A series of eight
to nine nonparallel real time short-axis images were positioned and acquired in digital format, paying careful attention to the end-planeimagesand using the line of intersection
display to ensure that the image planes did not intersect
within the left ventricular cavity (Fig. 1). The first image
plane passed through the inferior surface of the aortic valve
cusps. The second passed tbrough the center of the mitral
anulus. The third was positioned at the posterior mitral
anulus at its junction with the posteroi~ferior myocardium.
The subsequent image planes were spaced variably in the
midventricle and were optimized for endocardial boundary
definition. The last image plane was positioned at the endocardial ape%without any remaining lumen using the line of
the vemtricle were co
struction using traced
was defined between two
the same contour and a sin
at is, each tile bo~~da~ was
ur element” and by two “spans”
Hence, there were 360 triangks
area between adjacent cross secttons.
endocardial surface area could be
areas of all surface triangles. To
ume, a centroid was defined for eat ce, and vectors were
two separate observers without knowle
the other observer or of the echsca
two slices was decomposed into 180X 3 =
total volume of the ventricle.
NueIe
tie
ance
subjects
nt
ng o
system (GE Medical). One subject was excluded owing to
poor NMR images. All subjects were volunteers and gave
verbal informed consent before the study, in accordance
with Institutional Revieiv Board guidelines. Subjects were
connected to telemetered electrocardiographic(ECG)gating
with respiratory compensation and then placed oo the imaging table in a 30*left anterior oblique (left shoulder down)
was obtained using a
position. Initial “scout”
echo sequence (pulse
multislice, single-phase,
mterval; echo delay time
repetition time [TR] = one
[TN = 20 ms). These cotomal images (long axis) were
ac&red inroa256 X 128 matrix.The 3coci:imagedisplaying
the longest vertical axis of the heart was used to define the
location for the subsequent slices used to calculate left
ventricular vohrme. Five contiguous siices pei-a:endicutarto
Apical two- and four-chamber views, along witb a paraster-
nat sho&x& view at the rnidve~t~c~~arlevel were obtained. Images were digitized into a video display analysis
method as recommended by the guidelinesof tbe American
§I?
of Echocardiography
ysis. Conventions for
dimm$.mal echo
vention was chosen for the papillary muscles because the
current version of the poiyhedrai surface reconst
262
IACC Vol. 22. Na. 1
July 1993:258-90
GOPAL ET AL.
THREE-DIMENSIONAL ECHOCARDIOGRAPHY
algorithm does not accept discontinuous data. The convention for boundary tracing by NMR imagingalways included
the papiikry suscles as part of the cavity. Using the mean of
two observers, left ventrici;!z end-diastolic and end-systolic
volumes and en6dAstolic and end-systoiic cr.dncardial sur-
face areas, as determined by three-dimensional echocardiography and NMR imaging, were compared with one another using Pearson’s correlation coefficient and simple
regression. The F test was used to test the null hypothesis
that NMR imaging and echocardiographic measurements
were identical, thus yielding the equation y = x. The
regression estimates of the slope and intercept were simultaneously tested against the values of 1 and 0 (values for the
line of identity) by the F test (45). Interobserver variability
for selectionof images and boundary tracing was determined
for each method, as shown below, wheat: s, = values of
observer 1; x2 = values of observer 2, a& w = 15:
ale II. Comparisonof Left Ventricular End-Diastolicand Endstolic Volumesand EndocardialSurface
DimensionalEchocard~ogra~hyand Nuciea
ResonanceImaging
Interobserver
Variability
Volume
End-diastolic
3D Echo
NMR imaging
End-systolic
3D Echo
NMR imaging
Endocardial surface area
End-diastolic
3D Echo
NMRimaging
End-systolic
3D Echo
NMR imaging
SEE
r*
p Vaiue
4.99%
6.5%
0.92
6.99 ml
<: 0.001
8.01%
9.02%
0.81
4.01 ml
-
0.84
8.25 cm’
< 0.001
0.84
4.89 cm2
c: 0.001
r* = Pearson’s correlation coefficient. NMR = nuclear magnetic resonance; 3D Echo = three-dimensional echocardiography.
Statistical analysis was performed using separate comparisons of two- and three-dimensional echocardiography with
NMR imaging(Pearson’s correlation coefficient and simple
regression)for the subset of 10 subjects who returned at a
later date for a two-dimensionalechocardiographicstudy.
The results for ventricular volumes are shown in Table I.
Values for interobserver variability for three-dimensional
echocardiographic end-diastolic volume (4.99%) and for
three-dimensional echocardiographic end-systolic volume
(8.01%)were lower than those obtained for NMR imaging
end-diastolic and end-systolic volumes (6.59% and 9.02%,
respectively). Excellent (p < 0.001)correlations were obtained for both end-diastolic and end-systolic volumes (r =
0.92 and 0.81, respectively), with a low SEE (6.99 and
4.01 ml, respectively). The values for end-diastolic and
end-systolic volumes by both methods and their regression
equations are plotted in Figures 2 and 3. Three-dimensional
echocardiography resulted in systematically higher enddiastolic and end-systolic volumes than those obtained with
NMR imaging. The regression equation for end-diastolic
volumes did not differ significantlyfrom the line of identity.
The regression equation for end-systolic volumes did differ
significantlyfrom the line of identity and was associated with
a slightlygreater spread of values.
The results for ventricular endocardial surface areas are
shown in Table 1. Interobserver variability was not determined because the traced boundaries by both methods were
used io simu&&neouslydetermine both volume an6 s&ace
I) correlations were
area. Once again, excellent
obtained for end-diastoli
nd-systolic endocardial surface areas (r = 0.84 and
espectively), with low SEE
values (8.25 and 4.89 cm’, respectively). Plots of enddiastolic and end-systolic endocardial surface areas
methods, along with their regression equations, are shown in
re 2. End-diastolic
volume (EDV) determined by threedimensional echocardiography (3D ECHO) plotted against enddiastolic volume by nuclear magnetic resonance (NMR) imaging.
Dashed lines represent 95% confidence intervals. Interobs.
Variab.
= interobserver
variability.
160
3 D ECHO Interobs.Variab. = 4.69Y0
NMR Interobs.Variab. = 6.69%
125
E
t
I
r = 0.92
SEE = 6.99 ml
pc
0.001
B
Nwt End-diilic
Volume (mu
MAR End-systohc
Voiume (ml)
End-systolic voiume (WV) determines by threeocardiography (3D ECHO) plotted
by nuclear magnetic resonance @I
8 represent 95%confidenceintervals. C&he
as in Figure 2.
F@mz
3.
For a saabsetof 1
three-d~rne~sio~~~
ech
imaging values (Table 2).
dimensional echocardio
agmg were ag-
r
tworeasons.
aging (r = 0.48, SEE =
echocardiography and
volume; r = 0.70, SEE =
20.5 ml. p = NS for en
5.6 ml, p < 0.025 for end-systolic volume).
between corresponding systolic
left ventricular end-diastolic and
‘QF~
orgeometric assumptions and
are possible by the elimhatr,..
image plane positionhg error. Larger end-diastolic and
designed to minimizesources of error in ~~trasol~~d
SYS~CCII%
264
.'ACCVol.22,No. 1
July1993:258-70
GOPALETAL.
THREE-DIMENSIONAL ECHOCARDIOGRAPHY
NMR End-systolic
Endocardial
Sutface Area (cm2,
Figure5. End-systolic endocardial surface area (ESSA) determined
4l. three-dimensional echocardiography (3D ECHO) plotted against
end-systolic endocardial surface area by nuclear magnetic resonance !NMR) imaging. Dashed lines represent 95% confidence
intervals.
utilized parallel imaging planes and yielded a mean error of
1.6%(0.64+ 0.72 ml) (35).The capabilityof the polyhedral
surface reconstruction algorithm to utilize nonparallel, unequally spaced, nonintersecting short-axis image planes for
volume computation was validated using nonuniform balloon phantoms and the results were compared with true
volumes by water displacement and volumes obtained by
NMR imaging.We obtained the followingresults for threedimensional echocardiography: accuracy = 2.27%. interobserver variability = 4.33%, r = 0.999, SEE = 2.45 ml;
p -C 0.0001.Those for magnetic resonance imaging were
accuracy = 8.01%, interobserver variability = 13.78%,r =
0.995, SEE = 7.01 ml; p < 0.001.There was no statistically
Table 2. Comparison of Two- ar.d Thret-Dimensional
Ventricular
Volumes With Nuclear Magnetic Resonance Imaging
r
SEE
Value
(ml)
D Value
0.48
0.70
20.5
5.6
NS
< 0.025
0.90
7.0
3.1
<O.nnl
“.“_”
<G.oOi
2D Echovs.NMRimaging
EDV
ESV
3D Echovs.NMR
EDV
ESV
imaging
0.88
-.-
E;iV= end-diastolic
volume:
ESV = end-systolic
volume:
2D
Echo
two-dimensional echocardiography; other abbr&ttbr,s as in Table 1.
=
significant difference between the two met
of this study indicatedthat by eliminatingg
tions and image plane positiooi~g error, high accuracy and
reproducibility for volume computation were achievable in
vitro and that NMR imaging wo~!d serve as an in vivo
standard of comparison (40).The ability of the technique to
accurately predict true volume (r = 0.99, SEE = 2.72 ml;
p < O.OOOl)
in the presence of irregular endocardial borders
with regions of echo dropout was further demonstrate
glutaraldehyde-fixedsheep and pig hcartts(46).
Lower correlations between two-dimensional cc
R imagingwere observed in this
ose in previous two- ensio in
ies (5-7); however, severa ifferences in study design make corn~a~~so~s
di
of cineangiography as a reference standard in previous
studies has inherent geometric assumptions in the cakulation of volume and thus the result cannot be strictly compared with those in a study of this kind, ac :!,hicha standard
free of geometric assumptions was used. Another important
difference between previous studies and the current study is
related to the range of volumes studied. The inclusion of
patients with wall motion abnormalities in previous studies
provided a greater ra
higher correlation co
study, which include
was particularly true for end-systolic volumes and surface
areas, which varied over a relatively narrow range.
Results obtained for surface area are also very encourag
ing and confirm our p vious in vitro results. Once again,,
three-dimensional e ocardiographic values for enddiastolic and end-systolic endocardial surface areas were
higher than NMR imagingvalues, largely owing to inclusion
of the mitral anulus and left ventricular outflow tract in the
three-dimensional echocardiographic method. The in vitro
model used to validate the three-dimensional echocardiographic method for total endocardial surface area and infarct
surface area was a pin model, in which pinheads mounted in
three parallel arcs served as the acoustic targets. K’hisstudy
revealed values for intraobserver and interobserver variability and accuracy to be 0.54%,0.4% and 1.36%, respectively,
for total surface area; the same values for infarct area were
I .40%, 1.27%and 2.13%, respectively. Interobserver variability in vitro was primarily due to boundary tracing error,
not to imaging (49). This finding indicates that the threedimensional echocardiographic system is robust and performs as inktied, accurately computing volume from nonparallel data presented to the system. We believe that errors
that might result from use of the system will arise from the
presentation of incorrect data to it by the operator, not from
the three-dimensionalechocardiographic system itself.
Elim~~at~~n of geamdric assumptions. The threedimensional echocardiographicmethod of volume computation spatially registers images and is therefore free of any
geometric assumptions of left ventricular shape or image
relation. This represents a significant improvement over a
JACC Vol. 22. No.
July 1993:256-70
I
lane
ventricular volume cm
isnassumes an 0
operators
must rely on image
tive sense and a ~~0w~edgeOf
t the correct image position. A
study of two-d~me~sio~a~
echocardi
rn~~at~On
by Erbej et al. (13) sho
two-chamber views were displaced
to the apex, resulting in a fores~o~e~i~gof the ventricular
imageand underestimation of ventricular volume.
study using our three-dimensional echocardiogra
ner, we showed that approximately 50% of standard twodimensional echocardiographic imaging views by experienced operators are not optimally positioned (32). Tracing
improperly positioned short-axis images will contribute to
errors in volume computation. The line of intersection
display allows Operators to accurately and re
position image planes for volume com~~tatio
reference image to provide additional land
otherkse nonvisualized dimension. Using this
nirsue,we were able to improve the regroduc
,-~~sae
positkning appximately threefold (32), in turn leading ‘:o a threefold improvement in reprOduCibilityof left
ventricular end-diastolic dimension (33).
because
the polyhedral surface reccdnstrucusly c~eC~~~~the acsuracy Of the
ycle of data ac~~is~t~o~,
the lengths
angles formed by the three soimd emi
standards results in automatic rejection bf the data and
repetitiCn of image acquisition until satisfacto~ data at-e
obtained. $%rm random measurement variation may Occur
266
GOPAL ET AL.
THREE-DIMENSIONAL
ECHOCARDIt%RAPHY
becgdse echocardiographic and NMR images are acquired
over a seties of respiratory cycles. Accurate image positioning using the line of intersection display also has been tested
by measuringdistances between definedimage planes on the
pin model phantom. The results show that errors fioin ihe
;iue of intersection display ‘are ~0.4% of the distance between imageplanes (35).The polyhedral suifacc reconstruction algorithm slightly underestimates the true vohme Of
convex structures such as the left ventricle because the
r.lV.UJdrh-i~nbetween points
surfaceis reconstructed from rhfirAn
on the boundary contours. These chords lie inside the arc
definingthe convex surfaceand thus omit from the calculation a very small volume found between the chord and the
arc, Similarly, the calculation of surface area is slightly
underestimated because each triangle of the polyhedral
reconstructed surface is a flat surface. The mag;:iiudeof this
omission is minimizedby choosing a large number of points
on the boundary contours and by increasingthe frequency of
spacing of the cross sections. In the current study, 360
triangles/slicewere used to approximate the surface. The
average distance between images or average slice thickness
depends on the depth. It is estimated that the mean distance
between imagesat the midventricularlevel was -1 cm. The
accuracy of the polyhedral surface reconstruction algorithm
for volume and surface area computation has been validated
by us (40,46,49)and by Cook et al. (37). Additionally,
operator-dependent errors may occur owing to interpretive
factors. In addition to identifying the first and last cross
sections of an object, the operator must space sections close
together when the surface curvature is changing rapidly. If
the slices are too thick relative to the changing surface
curvature, sigrificant volume may be excluded by the polyhedral surface reconstruction algorithm. The use of eight or
nine short-axisimages in this study was based on the work of
Weiss et al. (3) and ou our in vitro validation studies
(40,46,49).We believe that all of these errors associated with
the three-dimensionalechocardiographic method are negligible compared with inherent errors of two-dimensional
echocardiographicsystems.
ndary tracing error. Boundary tracing errors of twodimensionalechocardiographicsystems are also a persistent
major source of volume computation error in threedimensional echocardiographic systems by presenting imprecise data. These errors are caused prim&y by inadequate definition of the boundary and by relatively poor
lateral resolution of the ultrasound beam within the image
plane and perpendicular to it. Furthermore, lateral resolution errors are exaggerated by using more than the minimal
necessary amplification.Curvilineargeometry and irregular
trabeculationof the ventricle produce echo boundaries ofthe
chamber that are spatially Lmbiguous.Typically, an echo
composed of a large number of pixels represents a single
point in the chamber boundary. Selectingthe correctboundarYpixel is very difficultand almost invariably results in
some volume computation error. This problem was addressed in the present study by using the minimal amplifica-
tion necessary to visualize the boundary and by using t
three-dimensionalecbocardiogra~bicline ofi~tersectio~ dlsplay to direct the image plane or ~~traso~~dbeam as nearly
perpendicular to the endocardia! scrface as possibi
nonparallel cross sections permitted the seleciion
axis images that were optimized for endocard~a~b
definition, Precise boundary tracing, however,
difficult problem that may be addressed in the future by
computer-based semiautcmaied boundary recognition pro-
specific objective of
validate the three-d
and surface area computation
another tomographic me
studies have been judged
niques may require geometric assumptions or
other limitations and do not serve as ideal
comparison.
irnag~~~
Left ventricular volume computation by N
has been validated in vitro and in vivo previousl
,39)and
is believed to be the best available clinical modality for
quantitative study of the ve
e that is free of
assumptions (51). The use
e cross sections
previously validatedfor in vivo volume and mass c
(39,52),and increasing the number of slices further adds to
the imagingtime, with little improvement in technique. The
method for surface area determination has not been formally
validated; however, the calculation is based on measurements of circumference that have been validated previously
and thus should be a reasonable first approximation.
The limitations of N
imaging with regard to patient
movement and the disti on between the blood pool and
the myocardium, particularly in conditions of slow flow, are
well known. In this study, one subject was excluded because
of suboptimal images resulting from patient movement. To
eliminate the majority of the slow-flowingbloodlendocardial
interface artifact, saturation pulses were applied outside the
z selected slicejust before each acquisition. Other sources of
error include slice thickness-dependent partial volume effects and variableend-slice positioning in addition to boundary tracing error.
Study limitations. Three-dimensional echocardiography
and NMR imaging have important differehlcesin the algorithms used for volume and surface area computation. Methodologic differences involved in the computation of left
ventricular volumes are due to inclusion of the left
ular outflow tract by three-dimcnsiomalechocardi
and exclusion of this volume by NMR imaging. This limitation was recognized at the outset of the study; however, the
objective of the study was to compare the volumemeawrement by three-dimensionalechocard~ogra~hywith the roost
accurate tomographic method available as it is used clinically. Our intent was not to closely match the two techniques
but, rather, to develop a clinically practical threy-
ducer position limits t
transthoracic windows.
lanes are often acqui
three-dimensional echocardiography at the lower end of the
face area by tbree- ens~ona~echoca
innately75% J~ger than the
lower end of the scale is
imaging. This discrepancy
corresponding value by
involving the y intercepts of the volume and surface area
diBerences in tbe algorithms
regression equations is du
ce,
the discrepancy is due to
used by the two methods.
methodologic bias rather than to random meas!~rc~e~tvariation. Random errors of true measuremenr would have
is ca
268
JACCVat.
22,No.1
July 19!3.3:258-70
GOPALETAL.
THIW,E-DIMENSIONAL
ECI%.?CAPU)IOGRAP~Y
unnecessary boundary tracing. Thr: concept of using intergz&g
images to calculate image plane position has been
revio&y
used by Geiser et al. (19). Our system continuP
ously displays the three-dimensional spatial information
encoded in the acoustic spatial locater data so that it can he
used in an on-line, interactive, prospective fashion to guide
and orient image plane positioning in real time (28). This
permits optimal endocardial boundary definition while still
acquiring the minhmalnecessary short-axis images to adequately represent the ventricle. The volume aigotithm used
in this study differs from that of Siu et al. (16) and from
Moritz et al. (25) in that areas where the endocardial
boundary is poorly defined, manual rather than computer
interpolation is required.
implications.
The three-dimensionalechocardiographic method described in this study offers an excellent,
repeatable, noninvasive clinical method of left ventricular
end-diastolicand end-systolic volume cieterminationfree of
geometric assumptions. Tedious boundary tracing may be
minimized by carefully guiding and selecting a representative set of imageplanes using the line of intersection display.
This method may be of enormous value in closely monitoring
patients with valvuiarheart disease and cardiomyopathy and
after myocardial infarction. The method can also be extended to compute myocardiai mass (46)free of geometric
assumptions and may be especially important in monitoring
the effects of hypertension, antihypertensive therapy and
ventricular remodeling after myocardiai infarction. The
method also offersa direct measurementof total endocardial
surface area and infarct surface area without geometric
assumptions (49).as well as the possibilityof analyzing wall
motion in a much more systematic and quantitative fashion.
Both of these may be of great value in the thrombolytic era.
Conclusions.
Three-dimensionalechocardiography using
the line of intersection display for glJidanceof image positioning and the polyhedral surface reconstruction algorithm
is an excellent method for determining left vent;ricuhr enddiastolic and end-systolic volume as well as endocardial
surface area computation in vivo. The results compare
favorably with NMR imagingand seem to be superior to the
two-dimensionalechocardiographic apical biplane summation of discs method of volume determination. A good
correlation with NMR imaging is achieved through the
eliminationof geometric assumptions and image plane positioning error.
A
Accuracy of Discrete Method qf
ata Acquisition
It has been suggested in a report published after acceptance
of this study that the method of data acquisition used in our
investigationis subject to potential inaccuraciesthat will not
be found in an alternative form of data acquisition (53).The
method of data acquisition used in our study is to place and
holdtheimaging transducer ir! a
desired location, ae
single, discrete set of spatial
o images over the next
quent@ acquire a series of 16
second. The single spatialdata set is assigne
The alternative method is to
coordinate data during imagea
of spatial data to each video image as it is acquired. It is
asserted that the latter continuous method overcomes a
data acquisition (53). This assertion is
both methods, in the absence of respi
data from these cau
a acquisition is discrete or continuous.
The same authors (53) also suggest that the discrete
method of acquisition is subject to error due to transducer
motion occurring during the l-s acquisition
frames. We performed a study to determine if
ity, in fact, produced significant error. Typically, the 16
captured video frames encompass two ventricular systoles
and the diastolic interval between them. On the same 15
subjects reported in the present inve
tion, we c
uted
end-diastolic volume twice, using
first end
tslic
image of each data set and then the second end-diastolic
image of the data set. We hypothesized that if the
significanterror due to transducer motion during the
of discrete acquisition then there would be a si
difference between the first and secon
For the 15subjects the first and second volumes traced by a
single observer were compared using the paired t test. Fci
end-diastoie f = -0.16 and p = 0.87. First and second
volumes for end-systole were also computed. For these, I =
0.55 and p = 0.59. The results of the paired t test show that
there is no significantdifferencebetween the first and second
volume determinations. From these results we infer that the
possibility of error by transducer motion during image acquisition by the discrete method does not introduce a significant error into the results of volume computation. We
believe that three-dimensional echocardiography significantly reduces volume computation errors by eliminating
geometric assumptions and reducing image plane position
errors and that the most significant errors for volume computation using three-dimensional echocardiography are introduced by manual tracing of endocardial boundaries.
Dodge HT, Sandler
H, Baxley WA, Hawley RR. Us’efulnessand limitations of radiographic methods for determining left ventricular volume. Am
J Cardiol 1966;18:10-24.
Maurer AH, Siegel JA. Denenberg BS, et ai. Absolute left ventricular
volume from gated blood pool imaging with use of esophageal transmission measurement. Am J Cardiol 1983;51:853-8.
Weiss JL, Eaton LW. Kallman CH, Maughan WL. Accuracy of volume
JACC Vol. 22, No.
July 1993258-70
I
dcte~rn~~at~0~ by ~wo-~~rne~s~o~a~ echocardiography: defining sequiremenls under coatrolled conditions in the ejecting canine left ventricle.
Circulation 1953:67:889-9.5.
4.
L, Meerbaum S, Feng MB<, Gueret P, Corday E. Cross-sectional
echocardiography. III. Analysis of mathemakic models for qtoanbifying
volume of symmetric and asymme!ric left ventric&es.Am Heart J BYSO:
1OO:821-8.
5. Scbnittger 1. Fitzgerald PJ, DaugbPers GT, e; al. Ein;iretions ofcompar&
left ventrtcular volumes by two-dimensional echocardiography, myocardial markers and c~~~~~~o~~a~~y.Am J Cardiol 1982;50:512-9.
6. Foiiand ED, Parisi AF. Moynihan PF, Jones DR. Feldman CL, Tow DE.
Assessment ofleft ventricular ejection fraction and volumes by real-time
t#o Qmensionai ecbocardiography.
A comparison of ci~ea~g~~~a~b~c
s. Circulation 1979;60:760-6.
7. Schiller NB, Acquate’i
, PortsTA. et al. Left ventricular volume from
paired biplane two-d+nensional echocardiography. Circulation 1979;60:
547-55.
8. Gueret P, Meerbaum S, Wq’;iit Hk, Uchiyama T, Lang T, Corday E.
Two-~irne~s~o~a~ ec~~ocar~iogra~bic ~~ant~ta~~o~ of left ientricular volumes and eiection fraction. Circulation 1980:62:1308-18.
9. Guyer DE,bibson Tc”, Gillam LD, et al. A new ecbocard~og~a~~~c model
for quantifying three-dimensional endocardiai surface area. .I Am Coil
Cardiol 1986:8:819-29.
10. Wilkins GT, Southern JF, Choong Cl’, et al. Correlation between
echocardiographic endocardial surface mapping of abnormal wall motion
and pathologic infarct size in autgpsied hearts. Circulation 1988,47:978 87.
11.
, Wilkins GT, Ray PA, Weyman AE. Natural history of left
ventricular size and function after acute myocardiai infarction: assessment and prediction by echocardiographic endocardial surface mapping.
Circulation 1~~82:484-94,
1%. Teichbolz LE, Kreulen T, Herman MV, Gorlin R. Problems in ecbocardiographic volume determinations: ecbocardiographic-a~giographic
correlations in the presence or absence of asynergy. Am J Cardiol 1976;37:
7-11.
13. Erbel R, Schweizer P, Lambertz H, et al. Echoventsiculography-a
simultaneous analysis of two-dimensional echocardiography and cineventricuiography. Circulation 1983;1:205-17.
14. Wong M, Shail PM, Taylor RD. Reproducibility of left ventricular internal
dimensions with M-mode echocardiography: effecrs of heart size, body
oosition and transducer anauiat:on. Am J Cardioi 1981:47: 1068-94.
Smith SW, Kisslo J. Real-Time, three15. ion Ramm OT, Pavy I&,
dimensional echocardiography: the first human images (abstr). Circulation 1991;84(suppl Il):II.685.
16. Siu SC, Rivera JM, Guerrero JL, et al. Three-dimensional echocardiography: in vivo validation for left ventricular volume and function (abstr).
J Am Coli Cardiol 1992;19(suppl A):3.
17. Matsumoto M, Inoue M, Tamura S, Tanaka K, Abe H. Threedimensional echocardiography for spatial visualization and volume caiculation of cardiac structures. J Clia Ultrasound 1981;9:357-65.
18. Dekker DL, Piziali R. Dong E. A system for ultrasonic&y imaging the
human heart in three dimensions. Comput Biomed Res 1974:7:544-53.
19. Geiser EA, Ariet M, Conetta DA, Lupkiewicz SM, Christie LG, Conti
CR. Dynamic three.dimensional echocardingraphic reconstruction of the
intact human left ventricle: technique and intial observation in patients.
Am Heart J 1982;103:1056-65.
20. McGaughey MD, Guier WN, Hausknecht MJ, Herskowitz A, Walters B,
Weiss JL. Three-dimensional echo reconstruction of the left ventricle in
cardiac transplant patients: direct validation of mass determination
(abstr). Circulation 1984;7O(suppi II):&397.
21. King DL, A1Banna SJ, Larach DL. A new three-dimensional random
scanner for ultrasonic/computergraphic
imaging of the heart. In: White
DN, Barnes R, ed. Ultrasound in Medicine. New York: Plenum, 1976;2:
363.
22. McPherson @pD, Skorios DJ, Kodiyahm S, et al, Finite element analysis
of myocardiai diastolic function using three-dimensional echocardiographic reconstructions: application of a new method for study of acute
ischemia in dogs. Circ Res 1987;60:674-82.
23. Raichlen JS, Trivedi SS, Werman GT, St. John Sutton MG, Reichek N.
Dynamic three-dimensbnai
reconstruction of the left ventricle frc’m
two-dimensional echocardiograms. J Am CoII Cardiol 1986;8:364-70.
24. Ghosh A, Nanda NC, Maurer G. Three-dimensional reconstruction of
~cb~car~i~gra~~~c images using the roladon method. Ultrass?&
BioQ 1982;8:655-61.
‘5.
MQ&ZWE, Pearlman AS, McCabe DN, e; 21.An ukrasorti,;technique for
imaging the ventricle in three dimensions and calc&fj;?g its vojome,
SEE!?Trans Biomed Eng i~~3:3~:482-92.
1.mc
26. Nixon J\?. Sage:. SI, d&mlb
Y,. %3Bl~iiiStCG. ~~~~~-~~rn~~sjo~a~
ec~ove~~r~c~~ogra~by. Am Heart 9 19$3;106:435-43.
27. Martin RW, Bashein G. ~e~~r~me~~
of stroke vo~~rn~? with threedimensional iransesopbageai ultrasonic scanning. Afles!beskologY 1989;
70:470-6.
28.
images. J Ultrasound Med P
29. Harrison MR, King DL, King
Jr, Kwan OL, Mahoney 5, DeMaria
ricular diameter measurement
at chordal level by three-dimensional echocardiography (abstr). Circula.
tion 199@;82(suppi 11I):iI1-68.
30.
arrison MR, King DL, King DL Jr, Kwan OL, Mahoney S, &Maria
uence of lefl ventricular dilation upon reprod~cibi~~y ofdiame~e~
measurements: evaiuation by three-dimensional echo (abstr). J Am Sot
Echocardiogr 1991;4:292.
31. King DL. 6~pal AS, Harrison
R, King DL Jr, DeMaria AN. Improved
reproducibility of anferoposterior left atrial measurement by guided
three-dimensional ec~ocar~iog~pby
(abstr). J Am Sot Ecbocardiogr
32.
three-dimensional
1990;::22s.
echocard~~g~ph~~
(abstr).
bY
J Am Sot Echocardiogr
33. King DE, Harrison MR, King DL Jr, Gopal AS
1RP,
a&AN.
Improved reproducibility of kft atrial and left
lrements
utar
by guided three-dimensional echocardiography. J Am Co!l Carcliol 1992;
20:iL38-4s.
34. Cook LT, Cook PN, Lee KR, e: ~1. An al~~~itbrn for volume estimation
based on polyhedral approximation. lEEE Trans Biomed Eng 1980;27:
493-500.
35. King DL, King DL Jr, Shao MYC. Evaluation of in vitro measurement
accuracy of a three-dimensional scanner. J Ultrasound Med I
82.
36 Ariet WI,Geiser EA. Lupkiewicz SM. Conetta DA, Canti CR. ~vai~at~o~
of a three-dimensional reconstruction to compute left ventricular volume
and mass. Am J Cardiol 1984;54:415-20.
37. Cook LT, Cook PN. Volume 2nd surface area estimators. Automedica
38.
39.
40.
41.
42.
43.
44.
45.
46.
4? .
Filipchuck NG. et al. Left ventricular volumes
ing. Radiology 1985;156:7117-9.
Markiewicz W, Se&tern U, Kirby R, Derugin N, Caputo GC
CD. Measurement of ventricular volumes in the dog by nuclear
resonance imaging. J Am Coil CardioI 1987;10:170-1.
Gopal AS, King DL, Katz J, Boxt LM, King DL Jr, Shao hn.
Three-dimensional echocardiograpbic volume computation by polyhedral
surface reconstruction: in vitro validation and comparison to magnetic
resonance imaging. J Am Sot Echocardiogr 11992;5:115-24.
American Society of Echocardiography Corn&tee on Skm&ds, subcommittee OII Quantitation of Two-dimensional
Echocardiograms:
Schilier NB, Shah FM, Crawford M, et al. Recommendations for qUantitation of the left ventricle by two-dimensional echocardiography. J Am
Sot Ecbocardiogr 1989;2:358-67.
Moritz WE, Shreve PL, Lace LE. Analysis of m &as.~nic
spatial
locating system. IEEE Trans Instrum Mras 1976:25:43-30.
&for&z WE, &eve PL. A micro-processor based sp
for use with diagnostic ultrasound. Proc 1EEE 1976
M&z
WE, Shreve PL. A system for locating point
space. IEEE Trans Instrum Meas 1977;26:5-IO.
heter j, Wdaserman W. Applied Linear Statistical Models. HOmeWood,
IL: Richard D Irwin Inc. 19X140-3.
Gopal AS, King DL, Bietefeld MR, Cabreriza S,
MYC, Left ventricular mass and vr?!urZe of fix
dimensional echocardiography (abstr). Circulation 1991;84:1[I-684.
Schroeder K. Sapin PM, King DL, Smith MK, DeMaris i%?l. 1s threedimensional echocardiography superior to two-dimensiona’ ecbocardiog-
GOXAL ET AL.
THREE-DIMENSIONAL ECkIOCARDIOGRAPHY
raphyfor volumedetermination?(abstr).J Am SW Echocardiogr1992;
5:3.
48. SapinPM,SchroederK, KingDL, SmithMK,DeMariaAN. Comparison
of ihree-dimensionalechocardiography,two-dimensionalecbocardiography and biplane angiographyfor left ventricularvolume determination
(abstr).J Am Sot Echocardiogr199253.
49. King DL, GopalAS, King DL Jr, Shao MYC.Three-dimensionalechocardiography:in vitro validation for quantitative measurementof total
and “infarct” surfacearea. J Am Sot Bchocardiogr1992;6:69-76.
&.:,I. 11.11,
rm RWOll
V...“^ “L,
A. n.
a”._z_
CaninI.
W ‘, *U‘.e
K;.n D’Y, “1I”U.
50. .Schwder KM, ‘Cr-.
&Jr,lvLd#la
AN. Improvedreproducibilityof left ventricularvolume measurements
SACC Vol. 22, No. I
July 1993:2%-70
using guided two dimexione: and three dimensionalecbocardiogra~by
(abstr). Circulation1992;86(suppI
IhI-271.
L._
+L,
.T.T,.$.l
,I:._
51. HigginsCR. Which_;..cdrr.4
~~~~~~~~
L1aa
LLI._
6Y,U
. rickisA).
: Am Co!:Csrdio:
1992;19:1608-9.
52. Keller AM, Peshock RM, Malloy CR, et al. In vivo measurement of
myocardialmass using nuclear magnetic resonanceimaging.J Am Cog
cardiol 1986;g:lD-7.
53. HandschumacherMD,LethorJP, Siu SC, et al. A new integratedsystem
for three-dimensionalechocardiographicreconstruction: development
__>.._,I
I!__IWI
__..__
r-z_..r-...
Lr.“^..:A
I- .nr+:,.n..,
;n ‘...-s””
uw
vaudatlu.1
YCIIUIW~I
YU~U~.~C
nilUla,,!;,,.,.,,.
IW,,,,. 8u$~cts.
J Am Coil Cardiol 1993;21:743-53.