Download A new automated method for analysis of gated

A new automated method for analysis of gated-SPECT
images based on a 3-dimensional heart shaped model
Milan Lomsky1 , Jens Richter2 , Lena Johansson1 , Henrik El-Ali3 , Karl Åström4 , Michael
Ljungberg3 , Lars Edenbrandt1, 2
1
Department of Clinical Physiology, Sahlgrenska University Hospital, Gothenburg, Sweden
2
Department of Clinical Physiology, Malmö University Hospital, Malmö, Sweden
3
Department of Radiation Physics, Lund University, Lund, Sweden
4
Department of Mathematics, Lund Institute of Technology, Lund, Sweden
Corresponding author:
Lars Edenbrandt
Department of Clinical Physiology
Malmö University Hospital
SE-205 02 Malmö
Sweden
E-mail: [email protected]
1
Abstract
A new automated method for quantification of left ventricular function from gated-SPECT
images has been developed. The method denoted CAFU is based on a heart shaped model and
the active shape algorithm. The model contains statistical information of the variability of left
ventricular shape. CAFU was adjusted based on the results from the analysis of five simulated
gated-SPECT studies with well defined volumes of the left ventricle. The digital phantom NCAT
and the Monte-Carlo method SIMIND were used to simulate the studies. Finally CAFU was
validated on ten rest studies from patients referred for routine stress/rest myocardial perfusion
scintigraphy and compared to QGS, a commercially available program for quantification of
gated-SPECT images. The maximal differences between the CAFU estimations and the true left
ventricular volumes of the digital phantom were 11 ml for the end-diastolic volume (EDV), 3 ml
for the end-systolic volume (ESV) and 3% for the ejection fraction (EF). The largest differences
were seen in the smallest heart. In the patient group the EDV calculated using QGS and CAFU
showed good agreement for large hearts and higher CAFU values compared to QGS for the
smaller hearts. In the larger hearts, ESV was much larger for QGS than for CAFU both in the
phantom and patient studies. In the smallest hearts there was good agreement between QGS and
CAFU. The findings of this study indicate that our new automated method for quantification of
gated-SPECT images can accurately measure left ventricular volumes and EF.
Key words: Computer-Assisted Image processing; Radionuclide imaging; Cardiac function
tests
2
Introduction
Left ventricular (LV) ejection fraction (EF) and volumes are valuable indices in clinical
decision making and powerful independent prognostic variables (Travin et al., Go et al.) ECGgating of myocardial perfusion single photon emission computed tomography (SPECT) images
allows a simultaneous evaluation of LV function (EF, volumes, wall thickening and wall motion)
and perfusion. There are presently several commercially available software packages for the
quantification of gated-SPECT (Germano et al., Faber et al.). They have all – to variable degree –
been validated using other nuclear medicine techniques (Everaert et al.) and other imaging
modalities (Nichols et al. 2000, Vallejo et al.). There have been mostly a good correlation
between the gated-SPECT methods and the reference methods, but some methods showed a
consistent underestimation of EF (ref) and the variability between the results from different
software packages could be substantial (Nakajima et al., Nichols et al. 2002). A recent meta
analysis of data from studies comparing LV end-diastolic volume (EDV), end-systolic volume
(ESV) and EF measured with gated-SPECT to the corresponding measurements from cardiac
magnetic resonance imaging showed high correlations between the methods but considerable
discrepancies for individual subjects (Ioannidis et al.). For example, as many as 11% of patients
would be misclassified with gated-SPECT when identifying subjects with EF of 40% or less.
Therefore, new approaches to quantify gated-SPECT images could be of value.
The aim of this study was to develop a new automated method for quantification of LV
function based on gated-SPECT images. We decided to use a detection algorithm, which is
based on a heart shaped model and adjusted using a digital heart phantom.
Material and Methods
CAFU
The new method for quantification of CArdiac FUnction – denoted CAFU – is based on the
active shape algorithm. The search and delineation of the left ventricle in the gated-SPECT
images is based on a heart model, which was developed using 50 selected patient studies. The
model contains statistical information of the variability of LV shape in this group of patients.
3
Step 1 – LV search in frame 1
The short axis slices are used as input to CAFU. The total count values of each of the slice
images are calculated. Seven adjacent slice images with high total count values are assumed to be
in the middle of the left ventricle. Each of these images is binarised by setting 15% of the pixels
with highest count values to 1 and the rest to 0. The seven images are thereafter summed and the
sum image is binarised by setting pixels greater than 50% to 1 and the rest to 0. This new image
is searched for LV lumen candidates, i.e. regions with “0-pixels” surrounded by “1-pixels”.
The cluster of candidates with positions closest to the upper right corner is defined as the LV
lumen position.
Thereafter, a search for the first estimate of the LV base is performed. In each slice image
starting from the most basal slice, a search from the LV lumen position in 16 directions was
performed. Maximum count values greater than 40% of maximum was considered to be LV wall
candidates. At least seven such candidates were required to define the first estimate of the LV
base. Thereafter, the first estimate of the LV apex was calculated in a similar way.
Step 2 – Delineation of mid-myocardial surface
In an iterative process, the heart shaped LV model is adjusted to optimize the fit with the
image data of frame 1. The model is positioned and scaled in order to fit with the first estimate of
LV base and apex from step 1. The model contains 272 landmarks distributed in 17 layers from
apex to base with 16 landmarks in each layer. In each landmark, a count profile normal to the
surface of the model is extracted. The position of the maximal count value in each profile is
noted provided that this value is at least 30% of the maximal myocardial count. The LV model is
then adjusted to minimize the distances between the landmarks and the corresponding positions
of the maximal count values. After this adjustment of the model a new iteration with search for
maximal count values and model adjustment is performed. This procedure is continued until the
changes of the LV model are very small or the maximum number of iterations (fifteen) has been
performed.
4
Step 3 – Epi- and endocardial delineation
In the 10% landmarks with the highest count values, count profiles normal to the surface of
the model are extracted. An average profile is calculated and the positions corresponding to 75 %
of the maximum count values in this profile are defined as the epi- and endocardial positions.
The distance between these positions is used as an estimation of the average thickness of the
myocardium in the region corresponding to the 10% landmarks used in this calculation. The
corresponding average count values for these landmarks is also calculated. The thickness in each
landmark point of the myocardium is then calculated as the product between this average
thickness and the ratio between the count value for the specific landmark and the calculated
average count value. Thickness values less than 8 mm are set to this value. Finally the epi- and
endocardial surfaces are used to calculate the myocardial volume.
Step 4 – Analysis of the following frames
The final position of the LV model in frame 1 is the first estimation of the position of the
model in the rest of the frames. The same iterative procedure is applied in the other frames in
order to find the midmyocardial surface. Thereafter the area of this midmyocardial surface is
calculated. The average thickness of the myocardium in the frame is determined based on the
assumption of constant myocardial volume of the left ventricle throughout the cardiac cycle. The
area of the midmyocardial surface from the specific frame and the volume estimation from frame
1 is used in this calculation. The thickness in each landmark point is adjusted with a factor
correlated to the relation between average count value of the myocardium in the frame and the
count value for the landmark.
Step 5 – Calculation of volumes
The LV volume is calculated using the endocardial surface and the LV valve plane. This
calculation is performed in all frames and the largest volume is defined as the EDV and the
smallest is defined as the ESV. The LV EF is calculated as the (EDV-ESV)/EDV. Frames with a
total count value for the myocardium less than 80% of that of frame 1 (tail drop) are not
considered in these calculations.
5
Step 6 – Wall motion and thickening
The wall motion measure for each landmark is calculated as the distance normal to the
surface of the myocardium between the position of each landmark in the ESV and EDV frames.
The thickening measure for each landmark is calculated as the ratio between the count value in
the ESV frame and the count value in the EDV frame.
Digital Phantom
The dynamic anthropomorphic computer phantom NCAT v1.12 (Segars et al.) was used.
This phantom includes possibilities of several alterations of the shape of the phantom and the
cardiac structure. For this simulation study, five different heart sizes were used in a male
geometry. We have used a standard 9 9Tcm -Sestamibi gated-SPECT study as a model for our
simulations where the heart cycle is divided into eight timeframes. The phantom also allows for
respiratory movements but this was not used in this work. The NCAT phantom gives an activity
map and a corresponding attenuation map for each time frame.
We used the Monte-Carlo program SIMIND (Ljungberg et al.) to simulate gated-SPECT
studies based on these activity and attenuation maps. The Monte-Carlo program was setup to
simulate a SPECT scintillation camera system with a low energy high resolution collimator with
energy resolution of 9.8% full width half maximum at 140 keV and an intrinsic resolution of 4
mm. The simulations mimic a common clinical procedure including acquisition of 48 projections
over 180 degree rotation mode starting at angle 315. The radius of rotation was kept to 26 cm
and the energy window was 20%.
Five perfusion studies with different heart sizes were obtained with the NCAT phantom and
SIMIND program. The EDV ranged between 57 and 186 ml and the ESV between 22 and 72
ml. The studies all had a motion pattern corresponding to a LV EF of 61%. The NCAT software
also provides the true chamber volume for each segment. These values were compared to the
values reported by the two analysis programs.
6
QGS
The Cedar-Sinai quantitative gated-SPECT (QGS) program (Germano et al.), which is the
most widely used program for quantification of LV function in gated-SPECT images, was used
for comparison to CAFU. Shortly, the QGS program operates in the three-dimensional space
and uses gated short axis slices as input. The automatic algorithm identifies the epi- and
endocardial contours for each of the sets of short axis slices in the cardiac cycle to calculate
volume changes. The largest and the smallest LV volumes correspond to the EDV and ESV,
respectively.
Patients
A total of 50 rest studies from patients referred for routine stress/rest myocardial perfusion
scintigraphy at Sahlgrenska University Hospital in Gothenburg, Sweden were used in the
construction of the heart model. The patients were selected to represent a variety of different
heart shapes. Ten other rest studies were used in the evaluation phase of CAFU. The patients
were selected to give a distribution of hearts sizes (EDV from approximately 50 ml to 200 ml).
Patients with large perfusion defects were not included. An experienced physician did the
selection of the studies.
Image acquisition
The rest and stress studies were performed using a two-day 99m Tc-sestamibi protocol.
Rest acquisition began at least 60 minutes after the injection of 600 MBq 99m Tc-sestamibi.
Images were acquired with a rotating dual-head SPECT camera equipped with low energy, high
resolution collimators. Acquisition was done with two different cameras, both using circular
acquisition and 64x64 matrix; in seven patients with camera A using 64 projections over 180
degrees for 40 s per projection and in three patients with camera B using 68 projections over 204
degrees for 40 s per projection. The patients were positioned supine on the SPECT table and
monitored with a three-lead ECG. The acceptance window was opened to 20% of the predefined
R-R interval. Other beats were rejected. Each R-R interval was divided into eight equal time
intervals. Gated-SPECT acquisition was performed at the same time as routine SPECT
7
acquisition. Projection images were reconstructed into transaxial images using filtered
backprojection. No attenuation or scatter correction was used.
Results
CAFU was first applied to the five phantom studies and after adjustment of parameters in
the algorithms the measurements calculated by CAFU were close to the true values. The true
volume of the myocardium for the five phantoms ranged between 89 and 301 ml and the
volumes calculated using CAFU were 1-8% lower. The LV EF and ESV for all five phantom
studies and the EDV for the two largest phantom studies were almost correctly calculated (Table
1). The EDV for the three smallest phantom studies were underestimated by CAFU by
approximately 10 ml. The epi- and endocardial boundaries calculated by CAFU for the smallest
and largest of the phantom studies are presented in Figure 1.
QGS calculated EDV (r = 0.99) and ESV (r = 0.99) values that correlated well with the true
volumes for the phantom studies (Figure 2). The absolute differences between the true volumes
and the volumes calculated using QGS were, however, substantial for the larger hearts, especially
for the ESV (Table 1). This relation resulted in a QGS measured EF as low as 39% compared to
the true EF of 61% for the largest phantom study. The overestimation of EDV by QGS for the
large phantom was caused by a false delineation of the basal part of the LV (Figure 1).
In the patient group the EDV calculated using QGS and CAFU showed good agreement for
large hearts and higher CAFU values compared to QGS for the smaller hearts (Figure 3). The
relation that QGS calculated much larger ESV than CAFU for large hearts, which was found in
the phantom studies, was also found in the patient studies (Figure 4). For the smallest hearts
there was good agreement between QGS and CAFU. The epi- and endocardial boundaries
calculated by CAFU and QGS for a small and a large heart of the patient studies are shown in
Figure 4.
Discussion
We have developed a new automated method for quantification of LV function based on
gated-SPECT images. The innovative approach with our method compared to previously
8
presented methods is the use of the active shape algorithm. With this technique a heart shaped
model is used instead of geometrical models such as an ellipsoid model or a hybrid cylindricspheric model in which the basal two thirds of the myocardium are modeled as cylindrical and
the apex is modeled as hemispheric. A heart shaped model is more likely to fit to myocardial
surfaces from patients with different shapes than a geometrical model.
During the initial development of a new method visual inspection of the epi- and endocardial
boundaries produced by the algorithm can be used. Serious mistakes by the algorithm are
possible to find from presentations such as those of Figure 4. This type of analysis is, however,
not accurate enough for a more precise adjustment of the algorithm. Independent examinations
such as gated blood-pool imaging, first pass studies, echocardiography or magnetic resonance
imaging are also difficult to use as reference methods for this purpose. The last of these methods
is probably the most accurate reference method, but this method has the disadvantage that the
examinations are not performed simultaneously. Differences in physiological states between the
two examinations may cause differences in LV volumes and EF. Therefore we used the digital
phantom NCAT and the Monte-Carlo method SIMIND to simulate gated-SPECT studies for the
development and adjustment of CAFU. This approach has the advantage that the true volumes
are known.
A limitation of the study is that CAFU was developed and validated on a limited number of
normal ventricles from different phantoms and patients. In the next step, the accuracy of the
method will need to be further evaluated in larger patient groups with independent methods to
assess, for example the LV EF. A large validation set that includes, for example, left ventricles
with large perfusion defects and cases with extra-cardiac activity will be of value in the evaluation
process. These type of cases have previously shown to cause problems for the quantitative
analysis.
The findings of this study indicate that our new automated method for quantification of
gated-SPECT images can accurately measure left ventricular volumes and EF. Further studies
are needed to evaluate the clinical value of CAFU.
9
Acknowledgments
This study was supported by grants from the Swedish Medical Research Council.
10
References
Everaert H, Bossuyt A, Franken PR. Left ventricular ejection fraction and volumes from gated
single photon emission tomographic myocardial perfusion images: comparison between two
algorithms working in three-dimensional space. J Nucl Cardiol. 1997;4:472-6
Faber TL, Cooke CD, Folks RD, Vansant JP, Nichols KJ, DePuey EG, Pettigrew RI, Garcia EV.
Left ventricular function and perfusion from gated SPECT perfusion images: an integrated
method. J Nucl Med. 1999;40:650-9
Germano G, Kiat H, Kavanagh PB, Moriel M, Mazzanti M, Su HT, Van Train KF, Berman DS.
Automatic quantification of ejection fraction from gated myocardial perfusion SPECT. J Nucl
Med. 1995;36:2138-47
Go V, Bhatt MR, Hendel RC. The diagnostic and prognostic value of ECG-gated SPECT
myocardial perfusion imaging. J Nucl Med. 2004;45:912-21
Ioannidis JP, Trikalinos TA, Danias PG. Electrocardiogram-gated single-photon emission
computed tomography versus cardiac magnetic resonance imaging for the assessment of left
ventricular volumes and ejection fraction: a meta-analysis. J Am Coll Cardiol. 2002;39:2059-68
Ljungberg M, Strand S-E. A Monte Carlo program simulating scintillation camera imaging.
Comp Meth Progr Biomed 1989;29:257-72
Nakajima K, Higuchi T, Taki J, Kawano M, Tonami N. Accuracy of ventricular volume and
ejection fraction measured by gated myocardial SPECT: comparison of 4 software programs. J
Nucl Med. 2001;42:1571-8
Nichols K, Lefkowitz D, Faber T, Folks R, Cooke D, Garcia EV, Yao SS, DePuey EG, Rozanski
A. Echocardiographic validation of gated SPECT ventricular function measurements. J Nucl
Med. 2000;41:1308-14
11
Nichols K, Santana CA, Folks R, Krawczynska E, Cooke CD, Faber TL, Bergmann SR, Garcia
EV. Comparison between ECTb and QGS for assessment of left ventricular function from gated
myocardial perfusion SPECT. J Nucl Cardiol. 2002;9:285-93
Segars WP, Lalush DS, Tsui BMW. A Realistic Spline-Based Dynamic Heart Phantom. IEEE
Trans Nucl Sci 1999;46:503-6
Travin MI, Heller GV, Johnson LL, Katten D, Ahlberg AW, Isasi CR, Kaplan RC, Taub CC,
Demus D. The prognostic value of ECG-gated SPECT imaging in patients undergoing stress
Tc-99m sestamibi myocardial perfusion imaging. J Nucl Cardiol. 2004;11:253-62
Vallejo E, Dione DP, Bruni WL, Constable RT, Borek PP, Soares JP, Carr JG, Condos SG,
Wackers FJ, Sinusas AJ. Reproducibility and accuracy of gated SPECT for determination of left
ventricular volumes and ejection fraction: experimental validation using MRI. J Nucl Med.
2000;41:874-82
12
Table 1
CAFU and QGS results in the phantom study
Phantom
EDV
ESV
EF
True
CAFU
QGS
True
CAFU
QGS
True
CAFU
QGS
1
57
46
45
22
21
21
61
58
54
2
80
71
70
31
28
37
61
61
47
3
109
101
104
43
43
57
61
58
45
4
144
142
145
56
59
83
61
59
43
5
186
185
201
72
71
122
61
62
EDV – end-diastolic volume; ESV – end-systolic volume; EF – ejection fraction
39
13
Figure Legends
Figure 1
Largest and smallest phantom quantified by CAFU and QGS.
Figure 2
Relation between true phantom volume and EDV (H) and ESV (Δ)measurements by
CAFU (filled) and QGS (open).
Figure 3
Relation between EDV ( ) and ESV (Δ)measurements by QGS and CAFU for 5
phantoms (open) and ten patients (filled).
Figure 4
Two patient studies quantified by CAFU and QGS.
14