Reliability of the prediction of the location of the culprit lesion from the ECG in totally occluded arteries in case of single vessel disease

Abstract

This paper deals with the reliability of algorithms predicting the location of the culprit lesion in a totally occluded coronary artery from a 12 lead ECG recording. Four commercial ECG analysis programs and two specific algorithms were evaluated. Five hundred well documented cases were used for testing. Evaluation of the results reveals that all algorithms perform suboptimal.

Full text

Page 1

Reliability of the Prediction of the Location of
the Culprit Lesion from the ECG in Totally
Occluded Arteries in Case of Single Vessel
Disease

WA Dijk1, AC Maan2, NHJJ van der Putten3, ET van der Velde2, CA Swenne2,
R Hoekema4, WRM Dassen5, JP Busman1

1University Medical Center Groningen, Groningen, The Netherlands
2Leiden University Medical Center , Leiden, The Netherlands
3Erasmus Medical Center, Rotterdam, The Netherlands
4University Medical Center St Radboud, Nijmegen, The Netherlands
5Maastricht University Medical Center, Maastricht, The Netherlands
Abstract
This paper deals with the reliability of algorithms
predicting the location of the culprit lesion in a totally
occluded coronary artery from a 12 lead ECG recording.
Four commercial ECG analysis programs and two
specific algorithms were evaluated. Five hundred well
documented cases were used for testing. Evaluation of
the results reveals that all algorithms perform
suboptimal.

1. Introduction
Invasive treatment through Percutaneous Coronary
Intervention (PCI) of patients experiencing symptoms of
acute myocardial ischemia has become widely accepted.
In case of ST-elevated infarction (STEMI), minimising
the time delay between onset of the symptoms and the
first balloon inflation of the culprit lesion is of utmost
importance for the patient’s
electrocardiogram (ECG) recorded by the alarmed
general practitioner or ambulance nurse is considered the
single most important test for diagnosing myocardial
infarction. Reflecting the electrical activity of the heart,
the ECG is but an indirect indicator of the presence of a
complete arterial occlusion. The consensus reached by
the American Heart Association about nomenclature and
location of the 17 myocardial segments assigns an area of
the myocardium to specific coronary artery territories [1].
Also the coronary artery anatomy is described in 16
segments by the American Heart Association [2]. In this
study we investigated the reliability of the prediction of
the location of the culprit lesion from the ECG by
commercially available ECG analysis programs handling
outcome. The
the indirect approach and 2 recently proposed algorithms
pinpointing directly to the occluded segment.
2. Methods
In the Netherlands the departments of Cardiology of
the 8 University Medical Centers cooperate in the
Interuniversity Cardiology Institute in the Netherlands
(ICIN). In these centers 4 different commercially
available ECG analysis programs are utilized (Table 1).

Table 1. Commercially available ECG analysis
programs
Company
Dräger
General Electric
Mortara
Philips

The participating centers contributed to a database
containing 508 well documented PCI procedures
concerning patients suffering from a single vessel disease
with Thrombolysis In Myocardial Infarction (TIMI) flow
grade equal 0 and a digitally available standard ECG
recording within a timeframe of maximal 2 hours
previous to the first balloon inflation. In Table 2 the
overall distribution of occluded segments is displayed.
The right coronary artery (RCA, segment 1,2,3,4,16) was
involved in 41% of the cases, the left descending artery
(LAD, segments 6/10) in 42% and the circumflex artery
(LCX, segments 11/15) in 17%. This distribution agrees
well with other publications.
The diagnoses generated by the ECG analysis
programs are grouped according to the affected artery in
Table 3. The relation between the diagnosed injured area
ECG system
Megacare
Muse
Escribe
Tracemaster
ISSN 0276−6574
701
Computers in Cardiology 2009;36:701−704.

Page 2

of the myocardium and the vessel concerned is defined
by the above mentioned consensus by the American
Heart Association. The left main stem has been discarded
for the obvious reason that the chance that a patient with
TIMI flow grade 0 reaches the hospital in time is
negligible.

Table 2. Distribution of occluded segments.

Segment
1.Proximal RCA
2.Mid RCA
3.Distal RCA
4.Right posterior descending artery
5.Left main stem
6.Proximal LAD artery
7.Mid-LAD artery
8.Distal LAD artery
9.First diagonal branch
10.Second diagonal branch
11.Proximal LCX
12.Obtuse marginal branch
13.Mid-LCX
14.Ramus posterolateralis from LCX
15.Ramus postero decendens from LCX
16.Ramus posterolateral from RCA

Three specific algorithms identifying the culprit artery
and the location of the occlusion were published by Fiol
et. al. [3] and Tierala et. al. [4]. The latter paper deals
only with the prediction of the RCA and LCX as culprit
artery. For the identification of the LAD (algorithm 1) we
used the results of the Fiol algorithm which states that,
essentially, any ST elevation in the precordial leads that
exceeds ST elevation in the inferior leads is decisive for
the LAD as the culprit artery.
The ECGs from the participating centers were
transformed into one intermediate format: 10 s of the 8
independent leads I, II, V1-V6 with a sample frequency
of 500 Hz as comma separated ASCII values. These CSV
ASCII files were converted to a Megacare-specific
format (Siemens Data Format, .sdf) and imported into the
Megacare database. For each ECG a new measurement
matrix and the corresponding interpretation was
generated based on the built-in GRI interpretation
program by MacFarlane et. al. [5]. For each ECG the
calculated ST-segment amplitude of each of the 12 leads
was exported. These values were then used as input for
the algorithms, programmed in the Matlab environment
according to figures 1 and 2.
Since the description of the original algorithms did not
explicitly define ST-elevation, we defined elevation as an
amplitude > 200 µV in V1 and > 100 µV in all other
#
74
87
39
9
0
112
86
6
8
1
33
19
28
4
0
1
leads, measured 60 ms after the J-point.
ST-depression was defined as < -100 µV in all leads.
Isoelectricity was defined as absence of both elevation
and depression.


Figure 1. Algorithm 2 according to Fiol et. al. [3] to
decide for RCA or LCX as culprit artery.

Figure 2. Algorithm 3 according to Tierala et. al. [3] to
decide for RCA or LCX as culprit artery.
702

Page 3

Table 3. Diagnoses generated by ECG analysis
programs versus segment coronary artery

Defined
territories [1]
RCA
region
Inferoseptal (3,9)

LAD
region
Anteroseptal (2,8,14,17)

LCX
region
Anterolateral (6,12,16)

No infarction diagnosis


3. Results
In table 3 the results of the ECG analysis programs are
displayed. The columns represent the artery segments and
the rows the infarcted myocardial area. The rows are
grouped according to the segments that are assigned to
the 3 main vessels [1]. The bottom row of the table shows
the number of diagnoses not containing any reference to
an infarction. The anterolateral segments, denoted by the
AHA committee as a LCX territory, does not show any
preference to any coronary artery. The same observation
can be made for the inferior region, where the
contribution of occluded LCX segments even doubles
that of its own “home” territory.
Table 4 shows the results of the algorithms by Fiol and
Tierala. Columns 3, 4, 5 and 6 relate to the Fiol
algorithm, columns 7, 8 and 9 relate to the Tierala
algorithm (which did not include a specific LAD
algorithm). Columns 6 and 9 display the number of
segments that were not classified as infarction. Both
algorithms lower the misclassifications of the LCX
contribution in the inferior segments, but no improvement
can be seen for the anterolateral region.
In table 5 the sensitivity and specificity of the
algorithms are displayed. The described algorithms 1,2
and 3 perform slightly better than the average of the
commercial products, except the sensitivity for the LCX.

4. Discussion and conclusions
The data originates from patients with
angiographically confirmed single totally occluded
segments. Therefore the resulting diagnosis from the
different algorithms can be analysed in a straightforward
manner. It is remarkable that all algorithms generate a
significant number of non infarction related diagnoses: 9,
20 and 24% respectively. In this study we used the






Table 4. Results of algorithms 1, 2 and 3


commonly accepted definition of ST-elevation and ST-
depression, resulting in a substantial number of non-
classifiable cases for algorithms 1, 2 and 3. As the
electrocardiogram is the single most important test for
diagnosing acute myocardial infarction, this high
percentage of false negatives hampers patient care
Myocardial area | RCA | LAD | LCX
1 2 3 4 16
Inferior (4,10,15) 31 42 13 6
16 9 10

Anterior (1,7,13) 7 2 1


Inferolateral (5,11) 1 3
14 21 8

5 10 7 3
6 7
4
8 9 10 11 12 13 14 15
19 4 21
1

3 1 5


1
4 8 5

6 4










2









1

3








59 44
13 16


35 22

5

1








4 2



2 1 2
S
E
G
M
E
N
T
R
C
A

Alg
2
L
A
D

Alg
1
L
C
X

Alg
2
U
n
d

Alg
2
R
C
A

Alg
3
L
C
X

Alg
3
U
n
d

Alg
3
RCA
region
1
51 10 1 12 50 14
2
64 10 1 12 53 1 23
3
27 1 11 25 13
4
7 2 7 2
16
1 1
LAD
region
6
98 14 14
7
72 1 13 14
8
1 4 1 1 1
9
3 5 5
10
1
LCX
region
11
7 9 4 13 7 2 15
12
3 3 1 12 2 1 13
13
9 7 4 8 9 3 9
14
1 2 2 1 2
15


Total 171 220 12 105 156 7 125
703

Page 4

significantly. Further investigations should take place to
optimise the definition of ST-deviation.
According to expectation, the accuracy of diagnosing
the LAD as culprit artery exceeded that of diagnosing the
RCA and LCX as the artery with the culprit lesion.

Table 5. Sensitivity and specificity of the algorithms

Algorithm Vessel
ECGsystem RCA
ECGsystem LAD
ECGsystem LCX

2 RCA
1 LAD
2 LCX

3 RCA
3 LCX

The sensitivity and specificity of the algorithms (table
5) does not identify a superior algorithm. Due to the large
percentage of non classification, the sensitivity and
specificity of the algorithms is rather low, with a
minimum for the LCX. The numbers indicate that, under
the described conditions, substitution of the acute single
vessel disease algorithms in the commercial systems by
the specifically designed ones is not advisable.

Sensitivity Specificity
60.9
63.8
17.8
81.8
94.5
73.3

71.4
83.6
10.6
93.0
85.8
82.7

64.8
7.1
93.3
79.4











References
[1] Cerqueira MD, Weissman NJ, Dilsizian V, Jacobs AK,
Kaul S, Laskey WL, Pennell DJ, Rumberger JA, Ryan T,
Verani MS. 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
2002;105;539-542.
[2] Austen WG, Edwards JE, Frye RL, Gensini GG, Gott VL,
Griffith LSC, McGoon DC: Murphy ML, Roe B. A
reporting system on patients evaluated for coronary artery
disease. Report of the Ad Hoc Committee for Grading of
Coronary Artery Disease. Circulation 1975;51;959-969.
[3] Fiol M, Cygankiewicz MD, Guindo J, Flotats A, Bayes
Genis A., Carreras F., Zareba W., Bayes de Luna A.
Evolving Myocardial Infarction with ST Elevation: Ups
and Downs of ST in Different Leads Identifies the Culprit
Artery and Location of the Occlusion. Ann Noninvasive
Electrocardiol 2004;Vol. 9, No. 2; 180-186.
[4] Tierala I, Nikus KC, Sclarovsky S, Syvanne M, Eskola M.
Predicting the culprit artery in acute ST-elevation
myocardial infarction and introducing a new algorithm to
predict infarct-related artery in inferior ST-elevation
myocardial infarction: correlation with coronary anatomy
in the HAAMU Trial. Journal of Electrocardiology; 42
(2009); 120-127.
[5] MacFarlane PW, Lawrie TDV. Diagnostic criteria. In:
Comprehensive Electrocardiology (1989), Vol. 3; 1527 –
1551. Pergamon Press Inc. Oxford.

Address for correspondence

Arnold Dijk
Thoraxcenter
University Medical Center Groningen
Hanzeplein 1
9713 EZ Groningen
The Netherlands
W.A.Dijk@thorax.umcg.nl

Assiciation. Circulation


704