A vectorial approach for evaluation of depolarization changes during acute myocardial ischemia

Abstract

In the present study we evaluated the upslope (I_US), downslope (I_DS) and terminal slope (I_TS) of the QRS complex in both standard and derived ECG leads obtained from spatial QRS loops, either by the vectorcardiogram (VCG) or by principal component analysis (PCA), in 79 patients undergoing prolonged, elective percutaneous coronary intervention (PCI). For each patient, the slope indices I_US, I_DS and I_TS were evaluated in the PCI recording as well as in a control recording acquired before the PCI procedure, and relative factors of change during PCI were calculated. We showed that IUS and IDS computed over VCG and PCA leads present higher sensitivity to the ischemia-induced changes than the same indices evaluated over the standard 12-lead ECG. Mean relative factors of change were 10.5 and 12.4 for I_US and I_DS in PCA, and 7.87 and 13.7, respectively, in VCG, representing an increase in sensitivity of up to 103% for I_US and 46% for I_DS compared to measurements obtained in lead V3. We conclude that evaluation of slope indices in leads derived from QRS loops significantly increases their potential value for detection of acute myocardial ischemia.

Full text

Page 1

A Vectorial Approach for Evaluation of Depolarization Changes during Acute
Myocardial Ischemia
D Romero1,2, M Ringborn3, P Laguna1,2, O Pahlm4, E Pueyo1,2
1Communications Technology Group, I3A, University of Zaragoza, Spain
2CIBER de Bioingenieria, Biomateriales y Nanomedicina (CIBER-BBN), Zaragoza, Spain
3Department of Cardiology, University of Lund, Sweden
4Department of Clinical Physiology, University of Lund, Sweden
Abstract
In the present study we evaluated the upslope (IUS),
downslope (IDS) and terminal slope (ITS) of the QRS com-
plex in both standard and derived ECG leads obtained
from spatial QRS loops, either by the vectorcardiogram
(VCG) or by principal component analysis (PCA), in
79 patients undergoing prolonged, elective percutaneous
coronary intervention (PCI). For each patient, the slope
indices IUS, IDSand ITSwere evaluated in the PCI record-
ing as well as in a control recording acquired before the
PCI procedure, and relative factors of change during PCI
were calculated. We showed that IUSand IDScomputed
over VCG and PCA leads present higher sensitivity to the
ischemia-induced changes than the same indices evaluated
over the standard 12-lead ECG. Mean relative factors of
change were 10.5 and 12.4 for IUSand IDSin PCA, and
7.87 and 13.7, respectively, in VCG, representing an in-
crease in sensitivity of up to 103% for IUSand 46% for IDS
compared to measurements obtained in lead V3. We con-
clude that evaluation of slope indices in leads derived from
QRS loops significantly increases their potential value for
detection of acute myocardial ischemia.
1.Introduction
Early diagnosis of patients with acute myocardial is-
chemia is essential to optimize treatment and hence clin-
ical outcome. In addition to patient history and clinical ex-
amination, the standard 12-lead electrocardiogram (ECG)
is the most important tool in the acute evaluation, both
in the pre-hospital setting and in the emergency room.
By convention, changes in the repolarization phase (ST-
T) are most widely used to detect acute myocardial is-
chemia. Changes also occur in the depolarization phase
(the QRS complex) of the ECG during acute ischemia that
could add information beyond the ST-T analysis. How-
ever, these changes have historically been more difficult to
characterize and have not come into clinical practice. In
a recent study a more reliable method for characterizing
changes in the QRS complex due to both amplitude and
duration changes have been proposed by measuring the up-
slope and down-slope of the R wave during myocardial is-
chemia induced by elective PCI [1]. That method was sub-
sequently improved in [2] by incorporating a normaliza-
tion procedure that attenuates the variations of the slopes
indices at control, thereby increasing their sensitivity to
ischemia-induced changes. In this study we propose the
evaluation of the slope indices over new derived leads ob-
tained by projection of 3D vector-based ECG loops com-
puted from the vectorcardiogram (VCG) or principal com-
ponent analysis (PCA) as in [3] and [4].
Our aim was to evaluate and compare the performance
of the QRS slope indices in monitoring the QRS changes
along the dynamic ECG recordings during PCI-induced is-
chemia on the standard 12-lead ECG and on leads derived
from the spatial QRS loop.
2. Methods
2.1. Population
The study population comprised 79 patients taken from
the STAFF III dataset, which contains patients admitted
to the Charleston Area Medical Center in West Virginia,
USA, for prolonged, elective PCI due to stable angina pec-
toris [5]. All patients met the following inclusion criteria:
no clinical or ECG evidence of an acute or recent myocar-
dial infarction, no intraventricular conduction delay with
QRS duration ≥ 120 ms (including LBBB, RBBB), no
pacemaker rhythm, low voltage, atrial fibrillation/flutter or
any ventricular rhythm at inclusion or during the PCI. Also
patients undergoing an emergency procedure, or who pre-
sented signal loss during acquisition were not considered.
All ECGs were recorded using equipment provided by
Siemens-Elema AB, Solna, Sweden. Nine standard leads
(V1-V6, I, II and III) were recorded and digitized at a sam-
ISSN 0276−6574
265
Computing in Cardiology 2010;37:265−268.

Page 2

0

0.5 1.01.52.0
−2
−1
0
1
2

0

0.51.01.52.0
−2
−1
0
1
2

−2
−1
0
1
2
−1
0
1
−1.0
−0.5
0
0.5
1.0
−2
−1
0
1
2
−1
0
1
−1.0
−0.5
0
0.5
1.0
a)
b)
x(n)
y(n)
z(n)
x(n)
y(n)
z(n)
w1(n)
w2(n)
w3(n)
w1(n) (mV)
w2(n) (mV)
w3(n) (mV)
Amplitude (mV)
Amplitude (mV)
time (s)
time (s)
Figure 1. a) Orthogonal X, Y and Z leads derived from Dower Inverse Matrix in a time segment, and their corresponding
loops; b) Transformed orthogonal ECG leads obtained from the PCA technique, and their corresponding loops.
pling rate of 1 kHz with an amplitude resolution of 0.6
µV. The three augmented leads aVL, -aVR and aVF were
then generated from the limb leads to yield the complete
standard 12-lead ECG. For each patient a control record-
ing was acquired continuously for 5 min, at rest in supine
position prior to the PCI procedure. Another continuous
ECG was acquired during the PCI, starting before balloon
inflation and ending after deflation. The duration of the
occlusion ranged from 1 min 30 s to 7 min 17 s (mean 4
min 26 s). The occlusion sites of the PCI procedures were:
left anterior descending coronary artery (LAD) in 25 pa-
tients, right coronary artery (RCA) in 38 patients, and left
circumflex artery (LCX) in 16 patients.
2.2. Leads derived from the QRS loops
a) QRS loop from the VCG: From the standard 12-lead
ECG l1(n),...,l12(n) it is possible to generate the three
orthogonal leads x(n), y(n), and z(n) by applying the
Dower Inverse Matrix over leads V1-V6, I and II [6].
These orthogonal leads can be represented in a 3D space
so that one can observe the variations of the electrical heart
vector (VCG), given by vVCG(n) = [x(n),y(n),z(n)]T.
During depolarization, the dominant direction u of the
QRS loop (QRSVCG) points to the QRS loop tip, called the
mean electrical axis. Thus, determining the main direction
of the QRSVCGloop, a new lead was obtained by project-
ing the loop onto that vector. For this analysis we first
searched for the main direction u by maximizing the fol-
lowing equation:
u = [ux,uy,uz]T= [x(n0),y(n0),z(n0)]T,
with
n0= argmax
n
[x2(n) + y2(n) + z2(n)]
(1)
where n spans over the samples of the running beat from
10 ms before QRS onset location (nON) to 130 ms after
nON. Then the new projected lead g(n) was calculated by
projecting the points of the QRSVCGloop onto the u axis:
gPCA(n) =vT
PCA(n) u
?u?
.
(2)
b) QRS loop using PCA: One way to implement PCA
is by applying singular value decomposition (SVD) on the
standard 12-lead ECG to generate a new lead system that
concentrates the most energy of the signal in a small set
of leads [7]. Specifically, the SVD was applied over leads
V1-V6, I and II to obtain 8 transformed leads wk(n), k =
1,2,...,8, by using the following transformation:
w(n) = UTl(n)
(3)
where the vector l(n) = [l1(n),l2(n),...,l8(n)]Tcon-
tains the original leads (only V1-V6, I and II) and U
is the matrix containing the right singular vectors of a
training set obtained from L= [l1,...,l8], with lk =
[lk(1),lk(2),...,lk(N)]T, and N the number of samples
in the recording. The new lead system given by the or-
thogonal transformed leads w1(n), w2(n) and w3(n) (the
first 3 of the 8 transformed leads wk(n) which mostly con-
centrate the energy of the original leads) was subsequently
used to represent the QRS loop in a different way, called
the QRSPCAloop. Analogously to the process described
in section 2.2, the same methodology was applied here
to compute a new lead gPCA(n) by projecting the QRSPCA
loop onto its dominant direction using equations (1) and
(2). The only difference with respect to 2.2 is that vVCG(n)
was replaced with vPCA(n), defined as:
vPCA(n) = [w1(n),w2(n),w3(n)]T.
(4)
266

Page 3

Examples of the two approaches described above (VCG
and PCA) are presented in Fig. 1.
2.3.Preprocessing
All signals involved in the study were preprocessed as
follows: (1) QRS detection, (2) normal beats selection
according to [8], (3) baseline drift attenuation via cubic
spline interpolation, (4) delineation using a wavelet-based
technique [9], and (5) ECG normalization [2].
2.4.Depolarization indices
Once the two leads derived from the spatial QRS loops,
gVCG(n) and gPCA(n) were obtained, the next step was to
compute the QRS slope indices (defined below) over all
the standard leads as well as on those two derived leads.
The methodology employed to compute the slope indices
is described in detail in [10]:
• IUS: the upward slope of the R wave.
• IDS: the downward slope of the R wave.
• ITS: the upward slope of the S wave (in leads V1-V3).
The ITS index corresponding to the upslope of the S
wave was only computed for leads V1-V3, where the S
wave is usually more pronounced.
The relative change during the PCI procedure was
tracked by the parameter RI, for each index I, and was
referred to the normal variation of I measured at resting
state [5]. RIevaluated at time tj, with tjtaken in incre-
ments of 10 s from the occlusion start (t = 0), was de-
fined as the ratio between the change observed during PCI
evaluated up to time tj, denoted by ∆I(tj), and the nor-
malfluctuationsofI observed duringthecontrolrecording
prior to the PCI, defined by the standard deviation (SD) of
I, denoted by σI: RI(tj) = ∆I(tj)/σI.
3.
3.1.
Results
Depolarizationchangesduringischemia
evaluated in standard ECG leads
The relative changes RIof the QRS slopes measured
during PCI were computed and averaged over patients for
the standard 12-lead ECG. The performances of the three
QRS slopes (IUS, IDSand ITS) were analyzed for the precor-
dial leads V1-V3, where we found that the two last slopes
within the QRS complex (i.e. IDSand ITS) present similar
behaviors along time, but not so for IUS. Figure 2 shows
the averaged relative factor of change (¯RI) for the three
slopes during 5 min of coronary occlusion in leads V2 and
V3. Itisclearthatthereisastrongrelationshipbetweenthe
slopes. In all the other leads, where the ITSindex was not
evaluated, the IDSslope presented higher sensitivity to the
ischemia-induced changes than IUS, with maximum values
reached in leads V3 and V5. In lead V3, the maximum av-
eraged factors of change (¯RI) of IDSand IUSwere found to
be 9.31 and 5.11, respectively. In lead V5, the maximum
¯RIwere 8.06 and 6.01.
0

011223344550
0
5
10
15

0

0112233445 50
0
5
10
15

20
40
60
80
100
20
40
60
80
100
79
73
62
57
57
62
73
79
IUS
IDS
ITS
IUS
IDS
ITS
V2
V3
¯RI(t)
¯RI(t)
patients (%)
patients (%)
time (min)
time (min)
Figure 2. Averaged relative changes RIof the three QRS
slopes (IUS, IDSand ITS) in leads V2 and V3. The black
lines represent the percent of patients that remain under
occlusion at each time instant.
3.2. Depolarizationchangesduringischemia
evaluated in QRS loop-derived leads
To corroborate whether the new approaches based on
the QRS loop provide QRS-slope measurements that per-
form better than those directly measured on the standard
12-lead ECG, we compared the relative slope changes av-
eraged over the whole population in lead V3 and the two
leads obtained by projection of the QRS loops (gVCGand
gPCA). As can be observed in Fig. 3, the methods based on
the QRS loop were superior.
Regarding IUS(see Fig. 3-a), the PCA-derived lead was
more sensitive to the ischemia-induced relative changes
(RI) than the VCG-derived lead, with their maximum¯RI
values being 10.5 (103% higher than in V3) and 7.87 (54%
higher than in V3), respectively. In the case of IDS(Fig.
3-b) the two loop methods showed very similar behavior,
with maximum relative factors of 12.4 and 13.7 for PCA-
and VCG-derived leads, which were 36% and 47% higher,
respectively, compared to lead V3. Despite the fact that
the maximum absolute change for IDSin the VCG-derived
lead was slightly inferior to that of the PCA-derived lead,
its variation in the control was substantially smaller, thus
explaining the slightly superior relative factor of change
found for IDSin the VCG-derived lead.
267

Page 4

012345
0
5
10
15


012345
0
5
10
15


V3: 2.05 ± 1.68
PCA: 3.79 ± 6.59
VCG: 1.63 ± 2.51
V3: 3.53 ± 4.53
PCA: 4.45 ± 7.82
VCG: 1.81 ± 2.92
time (min)time (min)
¯ RIUS(t)
¯ RIDS(t)
a)
b)
Figure 3. Evolution of the averaged relative factor of change¯RIfor IUS(a) and IDS(b) during PCI. Mean ± SD of σIover
patients in the control recordings are displayed on top of each graph.
4.Discussions and conclusion
In this study we measured the slopes of the QRS com-
plex (IUS, IDSand ITS) and assessed their performances for
evaluation of myocardial ischemia induced by coronary
occlusion during prolonged PCI. We proposed an improve-
ment for the quantification of QRS slope changes with the
purpose of providing more sensitive estimates of the oc-
currence of significant changes in the depolarization phase
during ischemia. In a previous work we introduced a dy-
namical ECG normalization to avoid very low-frequency
oscillations that directly influence the variability of the es-
timated slopes, specially at baseline recordings, and that
normalization led to an increase in the ratio RI, represen-
tative of the relative changes in the slope indices during
PCI [2]. In the present study the QRS slopes were also
evaluated in leads derived from the QRS loops obtained by
two different ways: from VCG and from PCA. The results
obtained using these methods far exceeded those obtained
in the standard 12-lead ECG system, reaching up to 103%
improvement for IUSand 46% for IDSmeasured in RIwith
respect to lead V3. That superiority is justified by the fact
that the slopes measured from the QRS loop show higher
absolute changes during the induced ischemia due to the
fact that the loop-derived leads result from the projection
onto a dominant vector with maximized amplitude, either
generated from the VCG loop or the PCA loop.
Results based on the QRS loop approaches seem to be
more sensitive to the induced ischemia than evaluation of
the QRS slopes from the standard leads. QRS slope anal-
ysis could act as a robust method of depolarization evalua-
tion in addition to repolarization changes in risk stratifica-
tion during monitoring of patients with acute ischemia.
Acknowledgements
This work was supported by a personal grant to Daniel
Romero from SCH bank and UZ, by project TEC2007-
68076-C02-02/TCM from MCyT and FEDER, Spain, and
by project PI144/2009 from Gobierno de Arag´ on, Spain.
References
[1] Pueyo E, S¨ ornmo L, Laguna P. QRS slopes for detection
and characterization of myocardial ischemia. IEEE Trans
Biomed Eng 2008;55(2):468-477.
[2] Romero D, Esther P, Ringborn M, Laguna P. QRS slopes for
ischemia monitoring in PCI recordings. In WC 2009. Mu-
nich: IFMBE Proceedings 2010;25(4):1695-1698.
[3] Correa R, Laciar E, Arini P, Jan´ e R. Analysis of QRS loop
changes in the beat-to-beat Vectocardiogram of ischemic
patients undergoing PTCA. In EMBC’2009. Minneapolis,
USA: IEEE Eng Med Biol Soc 2009;1750–1753.
[4] Correa R, Arini P, Laciar E, Laguna P, Jan´ e R. Study of
morphological parameters of QRS loop using singular value
decomposition during ischemia induced by coronary angio-
plasty. In Computers in Cardiology 2009. Park City, USA:
IEEE Computer Society Press 2009;36:693-696.
[5] Garc´ ıa J, Wagner G, S¨ ornmo L et al. Temporal evolution
of traditional versus transformed ECG-based indexes in pa-
tientswithinducedmyocardial ischemia.JElectrocardiology
2000;33(1):37-47.
[6] Edenbrandt L, Pahlm O. Vectorcardiogram synthesized from
a 12- lead ECG: superiority of the inverse Dower matrix. J
Electrocardiology 1988;21(4):361-367.
[7] Castells F, Laguna P, Sornmo L, Bollmann A, and Roig
J. Principal component analysis in ECG signal processing.
EURASIP J. Adv. Signal Process 2007.
[8] Moody GB, Mark RG. Development and evaluation of a 2-
lead ECG analysis program. In Computers in Cardiology
1982. IEEE Computer Society Press, 1982;39–44.
[9] Mart´ ınez JP, Almeida R, Olmos S et al. A wavelet-based
ECG delineator: evaluation on standard databases. IEEE
Trans Biomed Eng 2004;51(4):570-581.
[10] Romero D, Ringborn M, Laguna P, Pahlm O, Pueyo
E. Depolarization changes during acute myocardial is-
chemia by evaluation of QRS slopes. Standard lead
and vectorial approach. IEEE Trans Biomed Eng 2010,
doi:10.1109/TBME.2010.2076385.
Address for correspondence:
Daniel Romero P´ erez
I3A, Mariano Esquillor s/n / 50018 / Zaragoza / Spain
e-mail: daromero@unizar.es
268