Content uploaded by Markus S Juchems
Author content
All content in this area was uploaded by Markus S Juchems on Jul 03, 2015
Content may be subject to copyright.
ORIGINAL PAPER
Coronary plaque imaging with 256-slice multidetector
computed tomography: interobserver variability
of volumetric lesion parameters with semiautomatic plaque
analysis software
Oliver Klass •Susanne Kleinhans •Matthew J. Walker •
Mark Olszewski •Sebastian Feuerlein •
Markus Juchems •Martin H. K. Hoffmann
Received: 29 August 2009 / Accepted: 8 March 2010 / Published online: 26 March 2010
!Springer Science+Business Media, B.V. 2010
Abstract The purpose of this study was to evaluate
the potential clinical value of coronary plaque imaging
with a new generation CT scanner and the interob-
server variability of coronary plaque assessment with a
new semiautomatic plaque analysis application.
Thirty-five isolated plaques of the left anterior
descending coronary artery from 35 patients were
evaluated with a new semiautomatic plaque analysis
application. All patients were scanned with a 256-slice
MDCT scanner (Brilliance iCT, Philips Healthcare,
Cleveland OH, USA). Two independent observers
evaluated lesion volume, maximum plaque burden,
lesion CT number mean and standard deviation, and
relative lesion composition. We found 10 noncalcified,
16 mixed, and 9 calcified lesions in our study cohort.
Relative interobserver bias and variability for lesion
volume were -37%, -13%, -49%, -44% and 28%,
16%, 37%, and 90% for all, noncalcified, mixed, and
calcified lesions, respectively. Absolute interobserver
bias and variability for relative lesion composition
were 1.2%, 0.5%, 1.5%, 1.3% and 3.3%, 4.5%, 7.0%,
and 4.4% for all, noncalcified, mixed, and calcified
lesions, respectively. While mixed and calcified
lesions demonstrated a high degree of lesion volume
interobserver variability, noncalcified lesions had a
lower degree of lesion volume interobserver variabil-
ity. In addition, relative noncalcified lesion composi-
tion had a very low interobserver variability.
Therefore, there may a role for MDCT in serial
noncalcified plaque assessment with semiautomatic
analysis software.
Keywords Atherosclerosis !Plaque analysis !
Multidetector computed tomography !
Noninvasive imaging !Interobserver variability !
256-slice MDCT !Gaussian mixture model
Introduction
The ability of multidetector computed tomography
(MDCT) to noninvasively visualize coronary pla-
ques—both calcified and noncalcified—is an advan-
tage over luminography with conventional coronary
angiography [1,2]. Previous studies have shown the
reproducibility of MDCT-derived plaque indices with
those from intravascular ultrasound (IVUS) [3,4].
This paper has not yet been submitted for publication
elsewhere. The abstract is accepted for presentation in the
scientific sessions of the ESCR 2009.
O. Klass (&)!S. Kleinhans !S. Feuerlein !
M. Juchems !M. H. K. Hoffmann
Department of Diagnostic and Interventional Radiology,
University Hospital of Ulm, Steinho
¨velstr. 9,
89075 Ulm, Germany
e-mail: oliver.klass@uniklinik-ulm.de
M. J. Walker !M. Olszewski
Philips Healthcare, CT Clinical Science, Cleveland, OH,
USA
123
Int J Cardiovasc Imaging (2010) 26:711–720
DOI 10.1007/s10554-010-9614-3
Author's personal copy
At the same time, retrospective studies have shown
an association between plaques containing noncalci-
fied components and presentation with acute coronary
syndrome [5–8]. These observations have led to the
notion that noncalcified plaque may play a causal role
in the development of coronary events [7]; however,
while the quantification of calcified plaque is highly
standardized [9], no validated approach currently
exists to identify and quantify noncalcified plaque. In
addition, to translate the observation of noncalcified
plaque into diagnostically and potentially prognosti-
cally relevant information to support clinical decision
making, the sources of variation that may limit
MDCT from being used for either single-point or
longitudinal assessment of noncalcified plaque require
further investigation. The purpose of this study was to
evaluate the potential clinical value of coronary plaque
imaging with a new generation CT scanner and the
interobserver variability of semiautomatic coronary
plaque assessment with a new software application.
Methods and materials
Study group
The study population included 35 retrospectively
selected patients (28 male, mean age: 58 ±11 years)
with a single, circumscript lesion in the proximal left
anterior descending (LAD) coronary artery. Subjects
with isolated proximal LAD plaques were selected
retrospectively from our research database to allow
the study to focus on the reproducibility of the
algorithms and analysis techniques without the con-
founding effects of vessel caliber, plaque size, or
potential motion artifact. All patients were referred for
the exclusion of coronary artery disease (CAD) and
were imaged with a 256-slice MDCT scanner (Bril-
liance iCT, Philips Healthcare, Cleveland, OH, USA)
between November 2008 and March 2009. Patients
with a stable heart rate (HR) less than 65 bpm
underwent CT coronary angiography (CTCA) using
a prospectively triggered axial CTCA scan mode
(Step & Shoot Cardiac, Philips Healthcare, Cleveland,
OH, USA), and patients with heart rates greater than
65 bpm underwent retrospectively gated helical
CTCA. Patients with renal insufficiency (creatinine
B120 lmol/L), non-sinus rhythm, stents, or coronary
artery bypass grafts were excluded. Informed consent
was not required as CT imaging was clinically
indicated for the exclusion of coronary artery disease.
Patient preparation
All patients received nitrate at a dose of 0.8 mg
(glycerol trinitrate; Nitrolingual"N-Spray; Pohl-
Boskamp, Hohenlockstedt, Germany) sublingually
immediately before starting the scan. Patients with a
heart rate greater than 60 bpm received up to 25 mg
ß-blockade (metoprolol; Beloc", AstraZeneca GmbH,
Germany) intravenously. All patients were connected
to ECG leads placed in standard position to enable CT
synchronization with the ECG.
Injection protocols
For all patients, a patient weight and scan time
dependent volume (90–120 mL) of iodinated contrast
agent (iomeprol; Imeron 400, ALTANA Pharma
Deutschland, Germany) was administered followed
by a 50 mL saline flush. Contrast and saline were
injected at a flow rate of 6 mL/s into the antecubital
vein via an 18-gauge catheter. Automatic bolus
tracking was employed with a region of interest
(ROI) placed in the descending aorta. All scans of
were initiated 7 s after mean ROI contrast reached a
pre-set threshold of 150 Hounsfield Units (HU).
Image acquisition and reconstruction
Prospective triggered was performed with a gantry
rotation time of 270 ms and a scan-length dependent
adaptive z-collimation [10] at a maximum scan FOV
of 250 mm. For typical scan lengths, 256 overlapping
0.625 mm slices using a 2 9128 90.625 mm colli-
mation with dynamic z-focal spot and a z-increment of
62.4 mm between axial acquisitions was used. A peak
tube voltage of 120–140 kVp and a tube current–time
product of 150–250 mAs were used, both adjusted
depending on patient’s size. X-rays were prospec-
tively triggered at 75% of the R–R interval, corre-
sponding to the quiescent cardiac phase during
ventricular diastasis. A previously described X-ray
acquisition buffer of ±90 ms was used to ensure
sufficient data for reconstruction in the presence of
heart rate variation [11].
Identical scan parameters were used for helical
gating when applicable, with the exception of the
712 Int J Cardiovasc Imaging (2010) 26:711–720
123
Author's personal copy
effective tube current–time product, which was 600–
1,000 mAs, depending on body weight. A heart rate-
dependent pitch of 0.14–0.18 was used. To provide
the best possible image quality and the option of
reconstructing at any desired phase of the cardiac
cycle, ECG-based tube current modulation was
turned off for all patients.
All images were reconstructed with a thickness of
0.9 mm and an increment of 0.45 mm using an
enhanced semi-smooth reconstruction kernel (XCB).
For both prospectively triggered axial and retrospec-
tively gated helical acquisitions, images were recon-
structed in mid-diastole (75% of the R–R interval).
Additional images were reconstructed at other cardiac
phases during retrospectively gated helical acquisi-
tions as needed.
Plaque analysis
Image review was performed using standard axial and
multiplanar reformats as well as with dedicated cardiac
analysis software (Cardiac Viewer and Comprehensive
Cardiac Analysis, Brilliance Workspace, Philips
Healthcare, Cleveland, OH, USA). A new plaque
analysis application, integrated with the dedicated
cardiac analysis software, was used for semiautomatic
lesion identification, characterization, and quantifica-
tion. Plaque assessment was performed by two expe-
rienced, independent observers who were blinded to
the other’s results, and all cases were reviewed in
randomized order. Both observers were free to adjust
window width and window level as needed.
Image analysis included multiple, serial steps:
automatic, model-based, whole-heart segmentation
[12,13]; automatic coronary tree and centerline
extraction; automatic luminal and vessel wall contour-
ing; and semiautomatic plaque analysis. Semiauto-
matic plaque analysis included interactive lesion
identification (single user click within a visually
identified lesion), automatic lesion segmentation and
characterization (software-identified lesion voxels and
their classification), and quantification (software-cal-
culated cross sectional and volumetric lesion measure-
mnts). In addition, interactive manual and
semiautomatic correction tools, including those for
centerline, contour, and lesion segmentation, were
available to each observer. Corrections to vessel
centerlines, contours, and lesion segmentation were
made as necessary based on the individual expert
opinions of the observers. The reported quantitative
morphological parameters included cross sectional
(lumen, wall, and plaque diameters and areas, remod-
eling index, plaque burden) and volumetric indices
(plaque volume) for each lesion, vessel, and the entire
coronary tree. Cross sectional measurements, includ-
ing diameters, areas, and remodeling index were
calculated in planes orthogonal to the vascular center-
line and in accordance with definitions utilized in
intravascular ultrasound (IVUS) studies [14]. Figure 1
illustrates the relationships among cross sectional area
and plaque burden measurements. Here, plaque burden
effectively represents the percentage of the total vessel
cross sectional area occupied by plaque (Eq. 1). This is
a slight modification of the IVUS definition, which
regards plaque burden as the ratio of the luminal cross
sectional area to the vessel cross sectional area, often
expressed as a percentage [14]. In addition, reported
quantitative lesion characteristics included descriptive
Hounsfield Unit (HU) statistics and relative composi-
tion, expressed as a percentage of calcified and
noncalcified content. For the purposes of this study,
we chose to record and compare the resulting lesion
volume, maximum plaque burden (defined as the
largest ratio of plaque area to vessel area in a given
cross section—i.e., plaque burden—along the length of
the lesion), mean and standard deviation of HU, and
relative composition between observers.
Equation 1: Definition of Plaque Burden.
Plaque burden
¼100 #ðPlaque cross sectional area =
Vessel cross sectional areaÞð1Þ
Fig. 1 Morphological vascular measurements demonstrated in
a cross section of a vessel. Vessel cross sectional area is
encircled by the solid black line. The luminal cross sectional
area is encircled by the dashed black line. The plaque cross
sectional area (cross-hatched area) is the area characterized as
plaque between the vessel cross sectional area and the luminal
cross sectional. The plaque burden represents the ratio of this
plaque cross sectional area to the vessel cross sectional, as
shown in Eq. 1
Int J Cardiovasc Imaging (2010) 26:711–720 713
123
Author's personal copy
Two methods of plaque characterization were
available in the software application: a threshold-
based method and a Gaussian mixture model based
method [15]. For the purposes of this study, the
Gaussian method was selected. This method assumed
that within the lesion there were one or more different
components that could each be represented by a
Gaussian attenuation distribution. Each attenuation
distribution was then linearly combined to model the
whole lesion, with the relative lesion composition
determined by the ratio of the attenuation distribu-
tions. Voxels that belonged to attenuation distribu-
tions with means greater than a threshold set
empirically and fixed at 200 HU for all patients and
for both observers, were labeled as ‘‘higher-attenua-
tion’’ voxels. Lesions with a relative composition
(mean between observers) of greater than or equal to
90% higher-attenuation voxels were classified as
calcified, those less than or equal to 10% higher-
attenuation voxels were classified as noncalcified,
and all others were classified as mixed. A relative
composition of 10% (rather than 0%) or 90% (rather
than 100%) higher-attenuation voxels to classify a
lesion as noncalcified or calcified, respectively, was
chosen to overcome limitations of HU-threshold only
classification methods. By basing the lesion classifi-
cation on the relative number of higher-attenuation
voxels within a lesion, rather than a fixed attenuation
HU threshold (i.e., not taking into account the relative
number of higher-attenuation voxels, but only their
attenuation), this method avoids lesion misclassifi-
cation due to spurious inclusion of voxels associated
with intravenous contrast agent, noise, partial volume
averaging, etc.
Statistical analysis
Continuous values and proportions were reported as
mean ±standard deviation (range: minimum–
maximum). All noncomparative values were reported
as the mean of both observers, and observer com-
parisons are stated as absolute or relative differences.
Comparisons of continuous values were made using
analysis of variance (ANOVA), and comparisons of
proportional data were made using Fisher’s Exact test.
Bland–Altman analysis was used to compare the
interobserver variability for all reported values]. In
this way, agreement was reported as a bias (mean
difference of lesion measurements between observers)
and limits of agreement (random fluctuations of
differences about this mean). The bias and limits of
agreement were reported as a relative percentage of the
mean lesion volume measurement. Statistical signifi-
cance was determined with P\0.05. Statistical anal-
yses were performed using R (version 2.9.0, The R
Foundation for Statistical Computing).
Additionally, to determine if interobserver differ-
ences in lesion volume were due to relative content (%
calcification) or lesion volume (ANOVA, P\0.05),
we began with the largest lesion and iteratively
excluded the next largest lesion until no statistically
significant difference in lesion volume (ANOVA,
P[0.05) existed between noncalcified, mixed, and
calcified groups. Lastly, to further investigate the effect
of relative lesion composition on interobserver agree-
ment of lesion volume, each lesion was cumulatively
included, in order of increasing relative calcium
content, to assess interobserver variability.
Results
The demographics and scan characteristics of our study
group are shown in Table 1. No adverse reactions to
the sublingual nitrate, such as tachycardia or a
clinically relevant drop in blood pressure, or contrast
were observed. The distribution of scan protocols in
our study cohort was 27 (77%) prospectively triggered
axial scans and 8 (23%) retrospectively gated helical
scans. Scan length was 140 ±7 mm (range: 125–
155 mm), with a reconstructed transaxial field-of-view
of 195 ±20 mm (range: 160–229 mm). The dose-
length product (DLP) for the prospectively triggered
axial scans was 252 ±51 mGy cm (range: 69–
352 mGy cm) and 1,008 ±248 mGy cm (range:
Table 1 Summary of patient demographics and scan char-
acteristics
N35
Age (years) 58 ±11 [38–80]
Weight (kg) 81 ±13 [51–100]
Male gender 28 (80)
Mean heart rate (bpm) 61 ±7 [45–75]
Prospective gating 27 (75)
Note: All continuous data reported as mean ±standard
deviation [minimum–maximum]. All categorical data reported
as count (%)
714 Int J Cardiovasc Imaging (2010) 26:711–720
123
Author's personal copy
798–1,556 mGy cm) for the retrospectively gated
helical scans. The effective radiation dose of
CTCA was calculated as the product of the DLP
and a conversion coefficient for the chest (k=
0.014 mSv mGy
-1
cm
-1
[16,17]). The resulting
effective radiation dose was 3.5 ±0.7 mSv (range:
0.96–4.9 mSv) in prospectively triggered axial scans
and 14.1 ±3.5 mSv (range: 11.2–21.8 mSv) in retro-
spectively gated helical scans.
Based on relative lesion composition thresholds of
C90 and B10% higher attenuation content to clas-
sify calcified and noncalcified lesions, respectively,
we found 10 (28.6%) noncalcified, 16 (45.7%) mixed,
and 9 (25.7%) calcified lesions in our study cohort.
Table 2summarizes the morphological characteristics
of lesion volume and maximum plaque burden for
each of the lesion types. Table 3summarizes the
attenuation characteristics of each lesion type in terms
of lesion mean HU, lesion HU standard deviation,
and relative composition. Noncalcified, mixed, and
calcified lesions had significantly different volumes
(P\0.001) and maximum plaque burden (P\0.01),
with noncalcified lesions being smallest and calcified
lesions being largest on average. In addition, lesion
mean HU (P\0.001) and lesion HU standard
deviation (P\0.001) were significantly different
among types. Relative composition (percent higher
attenuation content) was not compared statistically
since their grouping was defined by their composition.
Bland–Altman analysis was used to compare
interobserver variability for each measured parameter.
Tables 4,5,6,7illustrate the interobserver variability
of lesion volume, maximum plaque burden, lesion
mean HU, and relative composition (percent calcifi-
cation) for all lesions and by lesion classification.
Additionally, following iterative lesion exclusion
based on decreasing lesion volume, it was determined
that by excluding lesions with a mean observer
volume C90 mm
3
, no statistically significant differ-
ence in lesion volume existed among noncalcified,
mixed, and calcified groups (P=0.07). Subsequent
analysis on this subgroup of volume distribution-
matched lesions showed that interobserver differences
Table 2 Lesion morphological characteristics: volume and
maximum plaque burden
Lesion
classification
Volume (mm
3
) Maximum plaque
burden (%)
All 79.5 ±80.7 [11.8–322.6] 51 ±17 [26–92]
Noncalcified 29.9 ±13.4 [11.8–59.7] 38 ±13 [26–70]
Mixed 64.0 ±38.0 [21.8–172.4] 51 ±12 [33–74]
Calcified 162.2 ±116.4 [20.5–322.6] 64 ±21 [33–92]
Note: All data reported as mean ±standard deviation
[minimum–maximum]
Table 3 Lesion attenuation characteristics: lesion mean HU, lesion HU standard deviation, and relative composition
Lesion
classification
Mean
HU (HU)
HU standard
deviation (HU)
Relative composition
(% high attenuation)
Noncalcified 96 ±35 [50–149] 64 ±18 [28–96] 1 ±3 [0–8]
Mixed 344 ±138 [139–599] 237 ±93 [66–374] 65 ±21 [31–87]
Calcified 449 ±168 [241–806] 249 ±104 [124–442] 98 ±4 [90–100]
Note: All data reported as mean ±standard deviation [minimum–maximum]
Table 4 Bland–Altman analysis comparing interobserver variability of lesion volume for all lesions and by lesion classification
Lesion classification Lesion volume
Relative bias Lower LOA Upper LOA
All -37 [-53, -22] -128 [-155, -100] 53 [25, 80]
Noncalcified -13 [-22, -3.6] -38 [-54, -22] 13 [-3.5, 29]
Mixed -49 [-71, -27] -129 [-166, -91] 30 [-7, 68]
Calcified -44 [-97, 8.5] -178 [-269, -87] 90 [-1.1, 181]
Note: all interobserver variability measures reported in relative terms as percentages [95% confidence intervals]. LOA is limit of
agreement
Int J Cardiovasc Imaging (2010) 26:711–720 715
123
Author's personal copy
in measured lesion volume were only dependent on
relative lesion content (P\0.01), not mean lesion
volume (P=0.34). Bland–Altman analysis of the
agreement between observers revealed a biasin lesion
volume measurementof -13% [95% CI: -22%,
-3.6%], -55% [95% CI: -83%, -28%], -98%
[95% CI: -319%, 122%] with lower limits of
agreement (LOA) of -38% [95% CI: -54%,
-22%], -140% [95% CI: -188%, -93%], -272%
[95% CI: -655%, 110%] and upper LOA of 13%
[95% CI: -3.5%, 29%], 29% [95% CI: -18%, 77%],
76% [95% CI: -306%, 458%] for noncalcified,
mixed, and calcified lesions, respectively in these
volume-matched groups. Figure 2shows a plot of
cumulative interobserver variability as a function of
relative lesion composition.
Discussion
A range of lesion volumes was observed, with mixed
lesions tending to be larger than noncalcified lesions,
and calcified lesions being larger than both other
classifications. Maximum plaque burden followed a
Table 5 Bland–Altman analysis comparing interobserver variability of maximum plaque burden for all lesions and by lesion
classification
Lesion classification Maximum plaque burden
Relative bias Lower LOA Upper LOA
All -25 [-35, -15] -81 [-98, -64] 31 [14, 48]
Noncalcified -18 [-23, -12] -33 [-43, -23] -2.3 [12, 7.4]
Mixed -24 [-40, -7.3] -83 [-112, -55] 36 [8, 64]
Calcified -36 [-65, -6.5] -111[-162, -60] 39 [-12, 90]
Note: All interobserver variability measures reported in terms as percentages [95% confidence intervals]. LOA is limit of agreement
Table 6 Bland–Altman analysis comparing interobserver variability of lesion mean HU for all lesions and by lesion classification
Lesion classification Lesion mean HU
Relative bias Lower LOA Upper LOA
All 3.4 [-6.4, 13] -52 [-69, -35] 59 [42, 76]
Noncalcified -3.4 [10, 17] -41 [-65, -17] 34 [10, 58]
Mixed -3[-19, 13] -60 [-87, -33] 54 [27, 81]
Calcified 22 [-0.24, 45] -35 [-74, 3.9] 80 [41, 119]
Note: All interobserver variability measures reported in relative terms as percentages [95% confidence intervals]. LOA is limit of
agreement
Table 7 Bland–Altman analysis comparing interobserver variability of relative lesion composition (% calcification) for all lesions
and by lesion classification
Lesion classification Relative lesion composition (% calcification)
Bias Lower LOA Upper LOA
All 1.2 [-0.76, 3.1] -9.8 [-13, -6.5] 12 [8.8, 16]
Noncalcified 0.5 [-2.1, 3.1] -6.7 [-11, -2.2] 7.7 [3.2, 12]
Mixed 1.5 [-2.5, 5.5] -13 [-20, -6.3] 16 [9.3, 23]
Calcified 1.3 [-1.2, 3.9] -5.2 [-9.6, -0.75] 7.8 [3.4, 12]
Note: All interobserver variability measures reported in absolute terms [95% confidence intervals]. LOA is limit of agreement
716 Int J Cardiovasc Imaging (2010) 26:711–720
123
Author's personal copy
similar trend. As expected, noncalcified, mixed, and
noncalcified lesions had significantly different lesion
mean HU with increasing HU, respectively. Calcified
and mixed lesions had substantially more lesion HU
variation than noncalcified lesions, although one may
have expected calcified lesions to be substantially
more homogenous than mixed lesions.
A large interobserver bias and variability for all
lesions was found for lesion volume, maximum
plaque burden, and lesion mean HU; however,
much smaller differences existed for relative lesion
composition. In general, Observer 2 consistently
reported larger lesion volumes and maximum
plaque burdens than Observer 1. In general, no
trend in lesion volume interobserver variability was
evident from Bland–Altman analysis when includ-
ing all lesions (Fig. 2). For all lesion parameters,
larger interobserver bias and variability existed for
calcified and mixed lesions when compared to
noncalcified lesions (Tables 4,5,6,7); however, no
trend with respect to mean lesion volume from
Bland–Altman was noted. Analysis on a subgroup
of volume distribution-matched lesions showed that
interobserver differences in measured lesion volume
were only dependent on relative lesion content, not
mean lesion volume.
The larger interobserver variability in lesion
volume and maximum plaque burden present for
calcified and mixed lesions is most likely attributed to
differences in interpretation of edges where calcium
interfaced with adjacent tissue (e.g., contrast, non-
calcified plaque, myocardium, etc.). For example,
calcium blooming (i.e., partial volume averaging and
beam hardening) may have hampered the delineation
of calcium borders [8].
Moreover, window width and window level were
freely adjustable by both observers, leading to a
potential bias and variability between observers.
Given that observers were also free to edit the
automatic results with manual correction tools may
have had a confounding effect in contributing to a
higher interobserver variability, as expected with
manual and semiautomatic algorithms. Furthermore,
it was noted during the study that the automatic
contouring algorithms were more successful at
delineating lesions with less relative calcified content.
As a result, less manual correction of those noncal-
cified lesions was required, thus contributing to the
lower interobserver variability for those lesions. On
the other hand, differences in relative composition
were less likely to be attributed to manual interaction
due to the use of the automatic Gaussian voxel
classification algorithm.
Fig. 2 Bland–Altman plot showing interobserver bias and
variability (95% limits of agreement) of lesion volume for all
35 lesions included in the study. Notably, there is no trend
relative to lesion size
Fig. 3 Relationship between lesion volume interobserver
variability and relative lesion composition. Interobserver
variability cumulatively includes all lesions with relative
lesion composition less than or equal to a specific value.
Clinically speaking, this cumulative measurement indicates
which lesions, by relative composition, can have their volume
quantified with a particular level interobserver variability.
Notably, the cumulative interobserver variability for noncalci-
fied lesions was less than 28.5%
Int J Cardiovasc Imaging (2010) 26:711–720 717
123
Author's personal copy
Fig. 4 Curved multiplanar
reformatted images from a
patient with an isolated
noncalcified plaque in the
proximal left anterior
descending (LAD) coronary
artery scanned in 2005 with
a 40-channel MDCT
scanner (a) and again in
2009 with a 256-slice
MDCT scanner (b). Plaque
analysis provided an
interesting tool for the
monitoring of statin therapy
and demonstrated a
reduction of noncalcified
lesion volume of greater
than 50%
718 Int J Cardiovasc Imaging (2010) 26:711–720
123
Author's personal copy
In contrast to the mixed and calcified lesions, the
interobserver variability in lesion volume and max-
imum plaque burden for the 10 noncalcified lesions
(B10% calcified content) was much lower. As
shown in Fig. 3, the cumulative interobserver vari-
ability for these noncalcified lesions was less than
28.5%. This interobserver variability of lesion vol-
ume was better or comparable to other studies of
MDCT-based plaque assessment [4,18]. With regard
to the higher interobserver variability associated
with quantifying calcified lesions, we feel that those
measurements are of limited clinical value given the
existence of well validated and readily available
software applications for the quantification of coro-
nary calcium [19].
Since interobserver variability is often the major
contributor to decreased measurement reproducibil-
ity, it is also a primary limitation for longitudinal
assessment. Therefore, the interobserver variability is
the lower bound with which lesion volume differ-
ences can be measured over time. In this way, the
minimum amount of plaque regression or progression
that may be detected, despite interobserver variabil-
ity, for noncalcified lesions would be greater than
28.5%.
Studies have shown good reproducibility of
volumetric IVUS radiofrequency-based analysis of
coronary plaque composition in vivo [20]. IVUS and
angioscopy studies have also indicated that reduc-
tion of lesion volume and relative content changes
(e.g., early loss of yellow color by angioscopy) may
both reduce plaque vulnerability in an additive
manner [21]. Our study, while demonstrating that a
degree of noncalcified lesion volume change may be
quantifiable longitudinally, also indicates that the
relative composition (percent calcification) of non-
calcified lesions is very reproducible, with absolute
interobserver bias and variability below 0.5 and
7.2%, repsectively. Further studies of the Gaussian
mixture model-based method, compared with tradi-
tional HU threshold-based methods, are necessary to
determine the impact of algorithm choice on relative
lesion composition interobserver variability. Not-
withstanding, this high degree of reproducibility in
assessing lesion composition changes over time may
permit studying the effects of statin or other drug
therapy to stabilize potentially vulnerable lesions
(Fig. 4).
Study limitations
We acknowledge the following limitations of our
study. First, selection bias was present, as patients
with only single circumscript LAD lesions were
included and total plaque burden was not assessed
for each patient. Second, the study had a relatively
small sample size and volumetric lesion parameters
derived from semiautomatic MDCT-based plaque
analysis software were not compared to a reference
standard, such as IVUS. Third, image quality was not
formally assessed, although all cases were judged to
be of diagnostic quality by both observers. In addition,
plaque analysis was based on a pre-release version of
software from the vendor. We also did not evaluate
the effect of changing the fixed attenuation thresh-
old of 200 HU on volumetric lesion parameters.
Moreover, manual correction methods were not
harmonized between observers, and when large
discrepancies existed between observers for a given
measurement, no attempt was made at adjudication or
at reaching consensus, and correction time was not
recorded.
Conclusion
While mixed and calcified lesions demonstrated a
high degree of lesion volume interobserver variabil-
ity, noncalcified lesions had a lower degree of lesion
volume interobserver variability. In addition, relative
noncalcified lesion composition had a very low
interobserver variability. Therefore, there may a role
for MDCT in serial noncalcified plaque assessment
with semiautomatic analysis software.
Acknowledgments Grants This study has been exclusively
financed from research funds provided by the state of Baden-
Wu
¨rttemberg, Germany.
References
1. Dey D et al (2006) Computer-aided detection and evalua-
tion of lipid-rich plaque on noncontrast cardiac CT. AJR
Am J Roentgenol 186(6 Suppl 2):S407–S413
2. Shinohara M et al (2008) Atherosclerotic plaque imaging
using phase-contrast X-ray computed tomography. Am J
Physiol Heart Circ Physiol 294(2):H1094–H1100
Int J Cardiovasc Imaging (2010) 26:711–720 719
123
Author's personal copy
3. Brodoefel H et al (2008) Characterization of coronary
atherosclerosis by dual-source computed tomography and
HU-based color mapping: a pilot study. Eur Radiol
18(11):2466–2474
4. Petranovic M et al (2009) Assessment of nonstenotic cor-
onary lesions by 64-slice multidetector computed tomog-
raphy in comparison to intravascular ultrasound:
evaluation of nonculprit coronary lesions. J Cardiovasc
Comput Tomogr 3(1):24–31
5. Achenbach S et al (2004) Detection of calcified and non-
calcified coronary atherosclerotic plaque by contrast-
enhanced, submillimeter multidetector spiral computed
tomography: a segment-based comparison with intravas-
cular ultrasound. Circulation 109(1):14–17
6. Feuchtner G et al (2008) Is there a relation between non-
calcifying coronary plaques and acute coronary syn-
dromes? A retrospective study using multislice computed
tomography. Cardiology 110(4):241–248
7. Hoffmann U et al (2006) Noninvasive assessment of pla-
que morphology and composition in culprit and stable
lesions in acute coronary syndrome and stable lesions in
stable angina by multidetector computed tomography. J
Am Coll Cardiol 47(8):1655–1662
8. Leschka S et al (2005) Accuracy of MSCT coronary
angiography with 64-slice technology: first experience. Eur
Heart J 26(15):1482–1487
9. Glodny B et al (2009) A method for calcium quantification
by means of CT coronary angiography using 64-multide-
tector CT: very high correlation with Agatston and volume
scores. Eur Radiol 19(7):1661–1668
10. Walker MJ et al (2009) New radiation dose saving tech-
nologies for 256-slice cardiac computed tomography
angiography. Int J Cardiovasc Imaging 25(Suppl 2):
189–199
11. Klass O et al (2009) Prospectively gated axial CT coronary
angiography: preliminary experiences with a novel low-
dose technique. Eur Radiol 19(4):829–836
12. Ecabert O et al (2008) Automatic model-based segmenta-
tion of the heart in CT images. IEEE Trans Med Imaging
27(9):1189–1201
13. Lorenz C, von Berg J (2006) A comprehensive shape
model of the heart. Med Image Anal 10(4):657–670
14. Schoenhagen P, Nissen SE (2004) An atlas and manual of
coronary intravascular ultrasound imaging. Panthenon
Publishing, New York
15. Duda RO, Hart PE, Stork DG (2000) Pattern classification,
2nd edn. Wiley-Intersience, New York
16. McCollough C et al (2008) The measurement, reporting,
and management of radiation dose in CT. Tech Rep 96
American Association of Physicists in Medicine. Available
online at: www.aapm.org/pubs/reports/rpt_96.pdf
17. Shrimpton P (2004) Assessment of patient dose in CT.
Tech Rep NRPB-PE/1/2004 National Radiological Pro-
tection Board
18. Leber AW et al (2004) Accuracy of multidetector spiral
computed tomography in identifying and differentiating
the composition of coronary atherosclerotic plaques: a
comparative study with intracoronary ultrasound. J Am
Coll Cardiol 43(7):1241–1247
19. Budoff MJ et al (2009) Coronary calcium predicts events
better with absolute calcium scores than age-sex-race/eth-
nicity percentiles: MESA (Multi-ethnic study of athero-
sclerosis). J Am Coll Cardiol 53(4):345–352
20. Hartmann M et al (2009) Reproducibility of volumetric
intravascular ultrasound radiofrequency-based analysis of
coronary plaque composition in vivo. Int J Cardiovasc
Imaging 25(1):13–23
21. Hirayama A et al (2009) Qualitative and quantitative
changes in coronary plaque associated with atorvastatin
therapy. Circ J 73(4):718–725
720 Int J Cardiovasc Imaging (2010) 26:711–720
123
Author's personal copy