Quantitative Assessment of the Infarct Transmurality Using Delayed
Contrast Enhanced Magnetic Resonance Images: Description and Validation
N Kachenoura1, A Redheuil1,2, R El-Berbari1, C Ruiz Dominguez1,
A Herment1, E Mousseaux1,2, F Frouin1
1
INSERM, U678, Laboratory of functional imaging, Paris, France
2
Cardiovascular Radiology Department, HEGP, Paris, France
nuclear methods particularly in the detection of sub
endocardial necrosis (6). The delayed enhancement image
is acquired 10 to 20 minutes after the injection of the
contrast agent. At this steady-state, non viable regions
are characterized by hyper-enhanced signal (7).
Myocardial functional recovery after an acute coronary
event is inversely correlated with the extent of non viable
myocardium (8). Therefore, it is crucial to measure
precisely and with high reproducibility infarct
transmurality.
In clinical routine, delayed enhancement images are
interpreted visually (9). To reduce the subjectivity of
visual assessment, a number of quantitative approaches
have been developed. The first methods use thresholds
which are based on statistical criteria such as the mean
gray level and its standard deviation in the myocardium.
However, the threshold value depends on the study and
can vary from 1 to 6*SD (7, 10, 11). The second
semiautomatic approach defines a threshold value as a
mean value of two regions, the first one being delineated
in the remote myocardium and the second in the infracted
tissue. This threshold combined with the gray level on the
centerline chords provides the infarct transmurality.
Finally a method based on full-width at half maximum
(FWHM) criteria was developed, the user clicks in the
hyperenhanced region, and a multi pass growing
algorithm is used to delineate the infarcted area (12). For
all these methods, in addition to the delineation of
myocardial borders, the operator has to delineate a region
within a remote and or infarcted tissue which increases
variability (13).
In this paper, a semi-automatic method based on the
unsupervised algorithm of the fuzzy c-means clustering
was developed to quantify myocardial infarct extent and
to provide segmental delayed enhancement scores.
Abstract
The extent and degree of myocardial injury after an
ischemic event are strong predictors of patient’s outcome.
After acute infarction, delayed contrast enhancement
magnetic resonance imaging allows clinicians to
distinguish between viable and non-viable myocardium
and can delineate with high precision the infarcted tissue.
The aim of this study is to provide a quantitative method
based on the fuzzy c-means clustering algorithm to assess
the location and the extent of the infarcted area.
Segmental infarct trasmurality was visually assessed on a
5-point scale for the 288 segments. The agreement
between visual and quantitative analyses was good: 84%
of the segments were categorized similarly by quantitative
and visual analyses.
1.
Introduction
Noninvasive assessment of myocardial infarct size is
important in the follow-up of patients with coronary
artery disease because of its known prognostic value (1).
After myocardial infarction it allows the evaluation of
myocardial viability, which has an important impact in
patient management. One of the main challenges remains
to differentiate reversible from nonreversible myocardial
injury.
Several methods have been developed for the
measurement of infarct size, such as contrast
echocardiography (2) which, however, is not able to
delineate specifically nonreversible injury. Another
method validated and accepted in clinical routine is based
on nuclear imaging techniques; it provides semi
quantification of the infarct size (3).
Several studies show the relevance of magnetic
resonance imaging in the assessment of myocardial
viability which can be assessed using contraction MR
studies under pharmacological stress (4) or more directly
thanks to delayed enhancement images which illustrate
with high accuracy even small infarcts (5). Thanks to its
precision, this new method has proven its superiority over
0276−6547/05 $20.00 © 2005 IEEE
25
2.
Methods
2.1.
step
Image acquisition and preprocessing
Computers in Cardiology 2005;32:25−28.
computed and called enhancement index :
MR acquisition for 15 patients having myocardial
infarction and 3 normal subjects were acquired at
European Hospital Georges Pompidou (HEGP) according
to a standard clinical protocol. Images were acquired
using a clinical GEMS Signa CV/i ™ 1.5 T magnet. For
each patient, three short axis slices were selected (apical,
mid-ventricular and basal).
Myocardial borders
were outlined manually by an expert. Then myocardial
segments were defined using two manually placed points
(the centre of the left ventricle and the superior
intersection between the two ventricles) according to the
recommended multimodality 17-segments model (14).
Segmental infarct size was visually assessed by an expert
on a 5-point scale:
0: No infarction,
1: Transmurality ≤25%,
2: Transmurality 26% to 50%,
3: Transmurality 51% to 75%,
4: Transmurality 76% to 100%.
2.2.
U ( j , k ) = mean (u ( p, E ))
p∈( S j ∩ Lk )
2.3.
Segmental scoring
For each patient enhancement indexes U(j,k) (j=1 to
48, k=1 to 4) are stored and then a threshold value (TH) is
used to give for each sub segment a transmurality extent
according to a decision tree (Figure 2). This decision tree
takes into account the fact that, within the framework of
the myocardial infarction, epicardial attack can’t occur
without endocardial attack.
yes
U(j,1) < TH
yes
Image processing
no
U(j, 2) < TH
yes
Inside the region delimited by the epicardial contour,
including the myocardium and the cavity, the fuzzy kmeans algorithm (number of clusters= 2, parameter of
fuzziness =2) was applied. Two classes were estimated:
the first class referred to us (E) contains enhanced pixels;
it is represented by both the cavity and the infarcted
tissue. The second class contains none enhanced pixels
(NE). For each pixel p, the membership probability to the
class E u(p,E) is stored.
These probabilities are restricted to the myocardium.
Each segment is then divided into 4 layers using
successively the centerline method (Figure 1). These
layers are called L1, L2, L3, L4 from the endocardial
border to the epicardial border.
no
no
U(j,3) < TH
yes
0
1
2
3
U(j,4) < TH
no
4
Figure 2: scoring process
For each standard segment three scores are allocated.
To take the final decision we use the following analysis:
- If the three sub segments have the same score this
one defines the global score of the segment.
- If the three sub segments have three different scores
then the highest score defines the global score of the
segment.
- If two sub segments present the same score and the
third sub segment a different score then there are two
possibilities:
- 1) if the difference between the two scores is
equal to one then the score of the two sub
segments defines the global score of the
segment.
- 2) if the difference between the two scores is
higher than one the highest score defines the
global score of the segment.
The last step of this analysis has two main targets: the
first one is to avoid artifacts by ignoring slight and
isolated differences. The second target is to design a
pessimistic method, which is important to avoid serious
L4
L3
L2
L1
Figure 1: Four myocardial layers
Thereafter, each segment is subdivided in three sub
segments, resulting in 48 sub segments per patient (12
sub segments for the apical slice and 18 for both midventricular and basal slices).
For each sub segment Sj (j=1:48) and for each layer Lk
(k=1:4) a mean membership probability U(j, k) is
26
proposed. It shows the relevance of using the fuzzy cmeans algorithm.
This approach has some limits: first of all, contours of
the myocardium were traced manually. This operation is
time consuming and it also introduces variability. Indeed,
if the endocardial contour particularly is badly placed,
pixels belonging to the cavity risk to be placed in the
myocardium and to be classified like delayed
enhancement because of their strong intensity.
Taking into account the difficulty of positioning
endocardial contour on the delayed enhancement images
(because of low contrast between the cavity and the
infarcted myocardium), the use of the contraction data
may help in contour tracing and increase its quality.
Secondly, the developed method is sensitive to the
artefacts in the image (1 segment was scored 0 by the
expert and 4 by the quantitative method).
In addition, 10 % of the segments are composite (part
pathological part healthy myocardium): this problem is
inherent to the 17 segment model. These segments were
taken into account in the developed method by dividing
each segment into 3 sub segments. This operation
improves the classification of all composite segments
except for three segments. Two of them have a part which
is scored 4 while the other part is scored less than 4.
Given that the developed method is pessimistic (Tab1)
these segments are scored 4 while scored 2 by the expert.
The last segment has a very small located infarct scored 2
by the expert and 0 by our method. To have a better
classification for this segment, we have to divide it into 4
sub segments. In this case the number of pixels within
each region will decrease and the effect of noisy pixels
would increase. In spite of the problems inherent in
acquisitions and in myocardial contours tracing, the
proposed method provides very encouraging results.
However, the evaluation must be extended to a larger
number of patients, to show its value versus a strictly
visual approach.
In this study, the goal was to estimate the transmural
extent of myocardial infarcts on 2D images but the
developed method can also provide infarct size. Its
application to 3D data would allow estimating infarct
volume which is a strong prognostic factor for the patient
and a valuable tool in patient medical management.
errors when this method is used to help in diagnosis.
3.
Results
Classifying on both cavity and myocardium is useful
to locate accurately myocardial infarct, but also to affect
to the healthy myocardium low membership probabilities
to the E class (figure 3).
Figure 3: maps of the membership probability for the
class (E) on a healthy subject (first line) and on infarcted
myocardium (second line).
The segments of the database were classified thanks to
a chosen threshold value (TH=37). A variation of 10%
around TH has no serious effect on the final diagnosis.
Thus, a head-to-head comparison was performed
between the visual classification, done by the expert and
the semi-automatic analysis.
Visual assessment
Fuzzy
c-means
0
1
2
3
4
0
20
7
8
1
1
2
3
4
3
5
9
1
4
7
1
2
6
8
8
3
16
Tab.1: Contingency table
Global agreement shows 84% of segments categorized
similarly by visual and quantitative analyses.
4.
References
Discussion and conclusion
[1] Gersh BJ, Anderson JL. Thrombolysis and myocardial
salvage. Results of clinical trials and the animal paradigm-paradoxic or predictable? Circulation 1993;88(1):296-306.
[2] Kaul S. Assessing the myocardium after attempted
reperfusion:
should
we
bother?
Circulation
In this paper a semi-automatic method for quantifying
the infarct extent on delayed enhancement images is
27
1998;98(7):625-7.
[3] Gibbons RJ, Miller TD, Christian TF. Infarct size measured
by single photon emission computed tomographic imaging
with (99m)Tc-sestamibi: A measure of the efficacy of
therapy in acute myocardial infarction. Circulation
2000;101(1):101-8.
[4] Kim RJ, Hillenbrand HB, Judd RM. Evaluation of
myocardial viability by MRI. Herz 2000;25(4):417-30.
[5] Wagner A, Mahrholdt H, Holly TA, Elliott MD, Regenfus
M, Parker M, et al. Contrast-enhanced MRI and routine
single photon emission computed tomography (SPECT)
perfusion imaging for detection of subendocardial
myocardial infarcts: an imaging study. Lancet
2003;361(9355):374-9.
[6] Kuhl HP, Beek AM, van der Weerdt AP, Hofman MB,
Visser CA, Lammertsma AA, et al. Myocardial viability in
chronic ischemic heart disease: comparison of contrastenhanced magnetic resonance imaging with (18)Ffluorodeoxyglucose positron emission tomography. J Am
Coll Cardiol 2003;41(8):1341-8.
[7] Gerber BL, Garot J, Bluemke DA, Wu KC, Lima JA.
Accuracy of contrast-enhanced magnetic resonance
imaging in predicting improvement of regional myocardial
function in patients after acute myocardial infarction.
Circulation 2002;106(9):1083-9.
[8] Kim RJ, Wu E, Rafael A, Chen EL, Parker MA, Simonetti
O, et al. The use of contrast-enhanced magnetic resonance
imaging to identify reversible myocardial dysfunction. N
Engl J Med 2000;343(20):1445-53.
[9] Wu E, Judd RM, Vargas JD, Klocke FJ, Bonow RO, Kim
RJ. Visualisation of presence, location, and transmural
extent of healed Q-wave and non-Q-wave myocardial
infarction. Lancet 2001;357(9249):21-8.
[10] Kuhl HP, Papavasiliu TS, Beek AM, Hofman MB, Heusen
NS, van Rossum AC. Myocardial viability: rapid
assessment with delayed contrast-enhanced MR imaging
with three-dimensional inversion-recovery prepared pulse
sequence. Radiology 2004;230(2):576-82.
[11] Fieno DS, Kim RJ, Chen EL, Lomasney JW, Klocke FJ,
Judd RM. Contrast-enhanced magnetic resonance imaging
of myocardium at risk: distinction between reversible and
irreversible injury throughout infarct healing. J Am Coll
Cardiol 2000;36(6):1985-91.
[12] Amado LC, Gerber BL, Gupta SN, Rettmann DW, Szarf G,
Schock R, et al. Accurate and objective infarct sizing by
contrast-enhanced magnetic resonance imaging in a canine
myocardial infarction model. J Am Coll Cardiol
2004;44(12):2383-9.
[13] Schuijf JD, Kaandorp TA, Lamb HJ, van der Geest RJ,
Viergever EP, van der Wall EE, et al. Quantification of
myocardial infarct size and transmurality by contrastenhanced magnetic resonance imaging in men. Am J
Cardiol 2004;94(3):284-8.
[14] Cerqueira MD, Weissman NJ, Dilsizian V, Jacobs AK,
Kaul S, Laskey WK, et al. Standardized myocardial
segmentation and nomenclature for tomographic imaging
of the heart: a statement for healthcare professionals from
the Cardiac Imaging Committee of the Council on Clinical
Cardiology of the American Heart Association. J Nucl
Cardiol 2002;9(2):240-5.
Address for correspondence
Nadjia Kachenoura
INSERM U678, 91 bvd de l’Hôpital, F-75634 Paris cedex 13
France
E-mail address kacheno@imed.jussieu.fr
28