Download Comparison of Two Systems for Determining Left Ventricular Function: Identifying Errors and more Lecture notes Medicine in PDF only on Docsity!
most 18 yr after the introduction of multi-gated blood-pool imaging (MUGA) the predominant use re
mains the determination of left ventricular ejection
fraction (LVEF) (1). If properly performed, the left
ventricular volume curve generated as part of the
MUGA study contains much of the information that
has traditionally been obtained by more invasive tech
niques (2).
As computer manufacturers and users attempt to
extract this data, it becomes obvious that interinstitu
tional and intermanufacturer variables exist that may
render the data unreproducible and therefore nearly
useless. In this paper we will examine the sources of
error that lead to this lack of reproducibility.
Recently, efforts have been made to derive left yen
tricular functional parameters from MUGA data, and
correlate them with various clinico-pathologic states (3—
6). Beforeattemptingto correlatethesevalueswith
pathologic states, we evaluated existing commercial
software as well as our own left ventricular functional
software for the derivation of left ventricular values
beyond EF and identified some of the variations that
occur between commercial packages.
We report below on our findings and the identified
sources of error making the comparison of LV func
tional parameters between the various commercial soft
Received Nov. 28, 1988;revision accepted July 18, 1989:For reprints contact: Robert H. Wagner, MD, Dept. of Nucl. Medi
cine,LoyolaUniversity,2160S. 1stAve.,Maywood,IL 60153.
ware packages difficult. We will report separately on
clinical applications of the LV parameters.
MATERIALSAND METHODS
All patients studied were referred for clinically indicated gated blood-pool studies. No normal volunteers were em ployed. Each patient's red cells were labeled using an in vivo technique with 25 mCi oftechnetium-99m pertechnetate (2). All MUGA studies were acquired in a modified 45-degree
LAO projection positioned to achieve the best ventricular
separation(best septal view). Each study consisted of32 frames with a minimum of200k counts per frame. These were filtered temporally with a three point 1-2-1 ifiter, which is standard
procedure at our institution. Left ventricularedgeswere de
termined by two commercially available software packages on different computer systems. In each case 32 independent ROIs were generated for each study. During this generation, a spatial nine point smooth was performed within the user defined area of interest. Variable region background correction systems were employed. The area for background correction was de termined automatically. The resultswere evaluated for the calculation of functional left ventricular values by processing patients on both systems and comparing the results. The patient studies were acquired on one system and then transferred for parallel processing to the other system. Initially there was agood correlation between the two systems for LVEF. We then developed our own postacquisition EF processing software for one system, and modified the available software on the other so that both
yieldedthe same final ventricularfunction parameterscalcu
lated with the same algorithms. The functional parameters derived from the study were the peak ejection rate (PER),
1870 Wagner,Halama,Henkinet al TheJournalof NudearMedicine
Errors in the Determination of Left Ventricular
Functional Parameters
Robert H. Wagner, James R. Halama, Robert E. Henkin, Gary L. Dillehay,
and Paul A. Sobotka
Section on Nuclear Medicine, Department ofRadiology, and the Section on Cardiology, Department oflnternal Medicine, Loyola University, Stritch School of Medicine, Maywood, Illinois
Gated blood-pool scans of the left ventricle are routinely employed for determination of the left ventricularejectionfraction.Recently,attemptshave been made to evaluateother left
ventricularfunctionalparameters.Thesevaluesincludepeakemptyingrate (PER),timeto
peak emptying rate (TPER), peak fillingrate (PFR), and time to peak fillingrate (TPFR). In studying these parameters clinically, we identified many software errors and assumptions that
impact on these values. These errors may also affect the determination of left ventricular
ejection fraction (EF). We condude that before any serious investigation of left ventricular functional parameters is undertaken, a detailed evaluation and standardization of the acquisition and edge detection algorithms must be performed.
J Nucl Med 30:1870—1874, 1989
peak filling rate (PFR), the time-to-peak ejection rate (TPER)
and the time-to-peakfillingrate (TPFR).
Eight patient studies were then processed four times on
each systemby the same operator to determine the precision
of the two systems. Each ventricular parameter for each pa tient was averaged. A standard deviation was calculated for each ventricular parameter for each patient and expressed as a percent of the average. These percents were then averaged for a given software protocol to give an indication of the variability in calculating each value. This process was identical for both computer systems. The only difference in processing was the commercially supplied edge detection algorithm.
Methods of Edge Detection System A (Siemens Microdelta/Maxdelta, Version 6.2). A zero crossing, second derivative edge tracking algorithm was
used. Edgeenhancement was with a spatial invariant second
derivative Laplacian operator after application ofa nine-point smoothing filter. Edges were searched radially outward from the center of the ventricle. A gradient threshold is applied, either low, medium, or high, which must be surpassed before an edge point is considered. The edge search is limited to
withina rectangularregiondefinedby the user.
System B (Medical Data Systems A3 with MIPS Software). This system used a zero crossing, second derivative edge
trackingalgorithmwithan outwardradialas describedabove.
An edge point may satisfyone of two conditions; either of
zero crossing,or when the pixelcount levelfallsbelowa user
selected threshold within one of four quadrants placed over the ventricular region. The zero crossing takes precedence,
and the threshold point is used if a zero crossingcannot be
identified.Again,a gradientthreshold,low, medium, or high
can be selected.
Calculation of the Derivative Curve Once the points of systole and diastole are identified and a left ventricular volume curve generated, the derivative curve may be created. This curve will identify the areas of peak ejection and peak filling and the times to their occurrence. To better representcontinuous curves, both the ventricular vol ume and derivative curves were interpolated by a factor of two. Interpolation and differentiation were obtained utilizing a
Fouriertransformtechnique(7). To obtain interpolation,the
Fourier transform of the ventricular volume curve was cx
tended in frequencyto the Nyquist frequencyof the desired
sampling interval, followed by inverse Fourier transformation at the new sampling interval. The cutoff frequency beyond which the Fourier transform is zero is determined by the
signal-to-noiseratio at the dominant frequencyas prescribed
by Bacharach et al. (8). For interpolation by two, the sampling interval is halved and the Nyquist frequency is doubled. The derivative ofthe volume curve is obtained by multiplying the Fourier transform of the ventricular volume curve by a fre quency ramp ifiter scaled by 2pii, where i is the imaginary numberdefined by the squareroot of —1,followed by inverse Fourier transformation. The ventricular volume curve was first replicated to three full cardiac cycles before Fourier transformation.
Finding PER, TPER, PFR, and TPFR
Once the derivativecurve wasgenerated,the peak ejection
rate (PER) and peak filling rates (PFR) were identified by
searching for the greatest slope. The search for peak ejection
and peak filling was restricted to their portion of the left
ventricular volume curve. Once these points were identified,
the time-to-peakejectionrate (TPER)and time-to-peakfilling
rate (TPFR) were obtained by knowing the number of frames between events and the time per frame.
RESULTS AND DISCUSSION
There were considerable differences in the reproduc
ibility of the calculation of left ventricular ejection
fraction (LVEF), peak ejection rate (PER), time-to-peak ejection rate (TPER), peak filling rate (PFR) and time to-peak filling rate (TPFR) between the two systems.
As can be seen in Figure 1, each of the values deter
mined on system A had greater variability than on
system B. The greatest variations were in the calculation
of TPER in both systems. The smallest variations were
in determination of PER in system B and of the EF in
system B. In both cases the PER was more precise than
the PFR. Similarly, the TPFR was more precise than
the TPER.
In order to better understand the errors that may
occur, a brief review of electrical and mechanical car
diac events is necessary. Figure 2 shows the cardiac
volume curve superimposed on the ECG. The first
event is electrical depolarization of the left ventricle.
The ventricles immediately begin to contract and the
A-V valves close. Approximately 20—30msec is re
quired for the intraventricular pressure to increase to
the level required to open the aortic valve. This is the
period ofisovolumetric contraction. Once the intraven
tricular pressure exceeds the aortic pressure, the aortic
valve opens and the blood is propelled into the systemic
circulation. Ejection is initially rapid and continues
despite the initiation ofventricular relaxation. Repolar ization begins and can be seen as the T-wave on the
ECG.
>. p. -J 4
I- z (U
IZIEF @PER@TPER@PFR@TPFR
FIGURE 1
A graphic representation of the percent variabilitywhen
generating the diastolic parameters on two systems.
SYSTEMA SYST@B
Volume 30 •Number 11 •November1989 1871
Errors in Determination of Heart Rate and Frame
Length
When a study is acquired, the time per frame is
determined from the preacquisition heart rate and this
is used throughout the study, despite any variations in the heart rate. It was anticipated that this information
would be stored by the computer. In both systems
evaluated, this parameter was not available retrospec
tively. It is this time per frame value that is essential in
the accurate calculation for PER, PFR, TPER, and TPFR.
The frame length is an essential part ofthe calculation
of ventricular parameters as can be seen from the Eq.
(2) through (5) below.
PER = Max Ef]
PFR = Max [@]
TPER = (# of Frames between
End Diast. and PER) (Time/Frame)
TPFR = (# of Frames between
End Syst. and PFR) (Time/Frame)
The simplest method to assure accuracy of frame
length data is to store it at the time the study is acquired.
If this is not possible, the original heart rate and the
true number of frames can be used to calculate it.
Time/Frame =
(HR) (# of Frames per R —R interval) (6)
This gives the result in minutes per frame. To get more usable numbers, multiply by 60 sec per minute
to get sec/frame.
The display ofthe regularity ofthe heartbeat and the
user defined windows of acceptable beats is an impor
tant piece of data for both study quality control and
computations. This R-R histogram is also an essential
component of the LV analysis package. It may also be
possible to determine an average heart rate from this data if the initial heart rate was not stored.
One erroridentified was that the use ofall heartbeats
to calculate the average heart rate may be erroneous
and does not correspond to that represented by the
image data. Premature ventricular contractions will
greatly affect this average. This problem is more evident
in the face of an arrythmia. To calculate an average
heart rate, we suggest using two standard deviations
from the mean of the obtained heart rate to calculate
the average heart rate. This insures that the average heart rate is as accurate as possible.
If the original heart rate or the time per frame is not
stored, any calculation of frame length will be in error.
An average heart beat derived from the histogram as
described above should not be used to calculate the
frame length since millisecond differences from the
actual frame length will cause the introduction of sub stantial error.
Errors in Detection of Systole
MUGA studies are acquired by gating the frame
acquisition to the R wave of the QRS complex on the
ECG. The assumption is made that systole begins with
the first frame after detection of the R wave. There is
in fact a lag of 20 to 30 msec between depolarization
and onset of ejection as can be seen in Figure 2. This
is, as noted above, the period of isovolumetric contrac
(2) tion. This may be prolonged in states where there is a
conduction defect.
For greatestreliability,the softwareshould search for
(3) the highest count obtained in the first 60% ofthe study rather than assume that the first frame of the study is the onset of systole. This will eliminate the isovolume (4) tric contraction phase, and give a more accurate impres sion of the time of contraction.
(5) The other assumption is that the acquisition begins
instantaneously with the R wave. Depending on the
hardware, this may not be true. A further time lag may
occur between the sensing of the R wave and the
beginning of the first frames acquisition due to hard
ware-software interaction. A long lag in starting acqui
sition may result in missing end diastole altogether.
Knowledge of the precise onset of systole is necessary
to calculate ejection fraction and the times to end
systole and peak ejection rate. If the gating is delayed,
the identification ofend diastole may be erroneous and
the ejection fraction may be several ejection fraction
units lower than the actual as can be seen in Figure 4.
T@-@_ /
E@ E
e liIIlTlllrtI1llS@lI1@II I
- 23MSEC'FRAME 236 EF: @34 HRz 828PM PEP: 333 EDV'SEC AVGR-RTIME: 745MIEC TPEP:fl MIEC 95'@P.RWIDTH:48MIEC @ ACCEPTEDTOL: PEP: 2.12 EDV'$EC R-R HIST @ 127 MSEC 82
FIGURE 4
Delayedgating will shift the curve to the left and the
determination of the onset of systole will be in error.
Volume30 •Number11 •November1989 1873
Errors in Detection of Diastole
The onset of diastole is the point at which the left
ventricular volume is smallest. The search for the small
est number of counts (end systolic frame) must be restricted to the first 75% of the cardiac cycle. If later
frames are included in the calculations, the onset of
diastole may erroneously be determined to be one of
the lower count terminal frames due to gating errors.
This will result in an erroneous ejection fraction. To
circumvent this problem it is advisable to include in
the display of the ejection fraction curve, a shaded area
or line as in Figure 5 indicating where systole and
diastole began. This will ensure the interpreter that the
correct assumptions were made for systole and diastole.
Proposed Methods for Determination of LV
Parameters
The most consistent edge detection was performed
using system B in combination with the variable back
ground placement and filtering as described above. The
detection of the onset of systole is the frame with the
largest number of counts in the region of interest. End
systole is the frame with the least number of counts.
Frame length should be recorded at the time of acqui
sition. After the derivative curve is calculated by the above method, the peak emptying and filling rates are identified. From these points, the times-to-peak ejection
and filling can be calculated.
CONCLUSION
The calculation of ventricular parameters other than
ejection fraction is becoming more useful as these pa
rameters are identified and related to various disease
states. The importance of the method of their calcula
tion lies in the fact that such great variation can occur
as to render these values nonreproducible and therefore
clinically useless if done without close attention to
detail. It is imperative that research done on these values
be preceded by an evaluation of software design. With
out a consistent standard by which to judge, it will be
impossible to reproduce results among institutions or
in a given patient.
NOTE
Since this study has been completed, System A's software (Version 87A) has been modified to be similar to that of System B.
REFERENCES
- Strauss HW, Zaret BL, Hurley PJ, Natarajan TK, Pitt
B. A scintiphotographicmethod for measuring left
ventricular ejection fraction in man without cardiac
catheterization. Am Journ Cardiol 1971; 28:575—580.
- Berman DS, Garcia EV, Maddahi J, Rozanski A. Thafflum-201 myocardial perfusion scintigraphy. In:
Freeman LM, ed. Freeman and Johnson's Clinical
Radionuclide Imaging. 3rd ed. florida: Grune and
Stratton; 1975:364—367.
- Miller TR, Goldman KJ, Sampathkumaran KS, Biello
DR, Ludbrook PA, Sobel BE. Analysis of cardiac
diastolic function: application in coronary artery dis
ease. JNuclMed 1983; 24:2—7.
- Schaffer EM, Rocchim AP, Spicer RL, Ct al. Effects of
verapamil on left ventricular diastolic filling in chil
then with hypertrophic cardimoypathy. Am J Cardiol
1988;61:4l3—417.
- Lee BH, Goodenday LS, Muswick GJ, Yasnoff WA,
Leighton RF, Skeel RT. Alterations in left ventricular
diastolic function with doxorubicin therapy. JAm Coil
Cardiol1987;9:184—188.
6. Miller TR, Gorssman SJ, Schectman KB, Biello DR.
Ludbrook PA, Ehsam AA. Left ventricular diastolic
@ fillingand its association with age.Am J Cardiol 1986;
58:531—535.
- Bracewell RN. In: The Fourier transform and its ap plications. 2nd ed. New York: McGraw Hill, 1978:117,194.
- Bacharach SL, Green MV, Vitale D, et al. Optimum
fourier filtering of cardiac data: a minimum-error
method: concise communication. J Nucl Med 1983;
24:1176—1184.
- Guyton AC. Heart muscle; the heart as a pump. In: Guyton AC, ed. Textbook ofmedicalphysiology. 6th
ed. Philadelphia: WB Saunders, 1981:154—155.
El
EF'53@HR.@:eE@PEP' P-PTIME:TPER'119M$EC95'@63EDY'SECAVG p-i @Ø@E:PFRs2.@ElY/SECP.R ACCEPTEDWIDTH:TOL:4@ HI5@TPFR'153MSEC
FIGURE 5
A normal ejection fraction curve with shaded areas depict
ing onset of systole-to-peakejection rate (It. gray) and
onset of diastole-to-peak fillingrate (dk. gray).
1874 Wagner,Halama,Henkinetal The Journal of Nuclear Medicine
