This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2016.2600198, IEEE
Transactions on Biomedical Engineering

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING 1

Simple T wave metrics may better predict early

ischemia as compared to ST segment
Glenn Terje Lines1 , Bernardo Lino de Oliveira1 , Ola Skavhaug2 , Mary M Maleckar1 ,
1
Simula Research Laboratory, Center for Biomedical Computing and Center for Cardiological Innovation, P.O. Box 134, Lysaker 1325,
Norway
2
Expert Analytics AS, 90 Henrik Ibsens Gate, 0255 Oslo, Norway

AbstractThere is pressing clinical need to identify developing
heart attack (infarction) in patients as early as possible. However,
current state-of-the-art tools in clinical practice, underpinned
by evaluation of elevation of the ST-segment of the 12-lead
electrocardiogram, do not identify all patients suffering from
lack of blood flow to the heart muscle (cardiac ischemia),
worsening the risk for further adverse events and patient outcome
overall. In this study, we aimed to explore and compare the
portions of cardiac repolarization in the ECG that best capture
the electrophysiological changes associated with ischemia. We
developed 3D electrophysiological models of the human ventricles
and torso, incorporating biophysically-based membrane kinetics
and realistic activation sequence, to compute simulated ECGs
and their alteration with application of simulated ischemia of
differing severity in diverse regions of the heart. Results suggest
that metrics based on the T-wave in addition to the ST segment
may be more sensitive to detecting ischemia than those using the
ST segment alone. Further research into how such simulation-
aided risk assessment methods may aid workflows in extant
clinical practice, with the ultimate goal of multi-modality clinical
support, is warranted.
Index TermsT-wave, simulated ECG, ischemia detection
I. I NTRODUCTION
The electrocardiogram (ECG) is used to measure the hearts
rate, rhythm, and morphology and remains the single most
important diagnostic tool for identifying and diagnosing acute
coronary syndrome (ACS). Acute blockage of blood flow
(ischemia) can manifest as changes in the appearance of
the portion of the ECG trace known as the ST-segment.
On the basis of this and other ECG findings, along with a
detailed clinical history, imaging of the coronary vessels via
angiography and blood tests to assess biomarkers indicative
of tissue damage, clinicians can determine whether a patient
is having an ST-elevation myocardial infarction (STEMI) or a
non-ST-elevation acute coronary syndrome (NSTEACS) event
[4]. When acute ischemia is confirmed, e.g. via ST-elevation,
an emergency course of action based on available guidelines
can be pursued. However, the presence and degree of ischemic
injury to the heart in NSTEACS is often unclear. Blood
markers may not be measurable until several hours after the
onset of an episode, giving time for any injury to the heart
muscle to deepen.
Corresponding Author: M. Maleckar, Simula Research Laboratory,
mmaleck@simula.no.
Copyright (c) 2016 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending an email to pubs-permissions@ieee.org.
In particular, risk assessment for non-ST-segment elevation
myocardial infarction (NSTEMI) patients, who suffer from is-
chemia severity sufficient to cause tissue damage (despite lack
of ST-segment elevation on ECG recordings) could improve
substantially. This early, specific identification would result in
decreased mortality and improved long-term prognosis.
Thus alternate metrics to permit timely identification of
ischemia not initially confirmed by traditional ECG analysis
is very attractive clinically, as it would allow rapid and
improved risk-assessment for NSTEACS patients as potential
candidates for emergent interventions. Recent clinical work
[1] has identified that dispersion in ventricular depolariza-
tion alters the T-wave of the ECG prior to any ST-segment
deviation and may be a good classifier of NSTEMI events.
Furthermore, others have found increased variability of the T-
wave amplitude with ventricular tachycardia [8], as well as
reduction in the T-wave amplitude and an increase in the QT
variability for infarcted patients [6]. Recent work showed that
the epicardial T-wave may be a more sensitive index of acute
ischemia than epicardial ST segment changes, especially in
the early stages of acute ischemia development [2]. However,
identification and validation of novel ECG metrics to support
NSTEACS diagnostics requires that we probe further to under-
stand the relationship between ischemia and its emergent ECG
biomarkers. In this study, we aimed to further explore the T-
wave portion of ECG repolarization and its ability to capture
the electrophysiological changes associated with developing
ischemia in the human ventricles. Specifically, we will look
at several quantitative T-wave metrics in silico and compare
the sensitivity of these metrics to standard assessment (ST-
elevation) in identifying the presence of developing ischemia.
To achieve this aim, we exploit the flexibility of biophysically-
based simulations and employ 3D human electrophysiological
models of the human heart and torso to compute a simulated
ECG and simulate its alteration with application of simulated
ischemia of differing severities in diverse regions of the heart.
II. M ETHODS
A. Geometrical model
A 3D finite element model of the human ventricles embed-
ded in a torso with ECG electrode lead positions as shown in
Figure 1A was constructed from MR images.
Fiber directions were set using the algorithm of Bayer et
al [3]. The heart volume was delineated into three categories:

0018-9294 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2016.2600198, IEEE
Transactions on Biomedical Engineering

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING 2

epicardium, midmyocardium, and endocardium, as suggested

by [9].

B. Electrophysiological modeling
A RA
LA B
To model healthy, baseline electrophysiology, we employed V1 V2 V3
V4
V5
the ten Tusscher 2006 model for the human ventricular V6

myocyte [10]. To account for regional electrophysiological

differences in the heart, each (epicardium, midmyocardium,
endocardium) was assigned a corresponding cell model as
based on previous parameterisation [9]. Following the work of
Keller, et al [7], we also included base-to-apex (longitudinal)
heterogeneity in IKs , which gives rise to shorter apical APDs. LL

To model ischemic cells, we used the approach of Wilhelms, et

al [11]; parameter sets from the original, regional cell models
were altered to account for the the three main effects observed C
on ischemic cells (hyperkalemia, acidosis and hypoxia). We
note that one cannot expect to reach a steady state orbit
in the phase space for ischemic cells, as these are not in
metabolic equilibrium. Concentrations in the cell models were
not clamped (except Ko as listed in Table I); instead for each
stage of ischemia and each cell type, an initial condition was
computed by cycling the cell for one minute at 80BPM.
Table I shows the altered parameters for each stage of
ischemia studied here. To model this developing ischemia
spatially, a spherical ischemic region of radius r = 0.5, 1.0, 1.5,
Fig. 1. Computing the simulated ECG. A. The simulated 12-lead ECG is
and 2.0cm (4 cases) was applied to one of 17 locations in the computed using the standardized lead positions and torso geometry shown.
left ventricle. The core lay along the endocardium, as shown B. A simulated activation map following pseudo-Purkinje activation for the
in Figure 2, such that regions of varying transmurality and reference case (control, no ischemic region) in the human ventricular geometry
employed in this study. C. A comparison of the 12-lead ECG between the
severity could be investigated (an ischemia region of radius reference case (black), and that of a heart with regional ischemia of stage 1
1cm was approximately 85% transmural). In order to more severity and of size r = 1cm in segment N2 (red).
precisely refer to these 17 sites, we use the cardinal directions
(N,W,S,E) along with a number indicating the distance from
the apex. For example, the dark red circle in the upper right
corner of Figure 3A would be E2.
Realistic activation patterns (as seen in Figure 1B and the
QRS complex in Figure 1C) were achieved using a pseudo-
Purkinje network based on the classic Durrer paper offering
insight into activation sequence in the human ventricle [5].
Ischemic tissue had relatively little effect on the activation
sequence, and as such, we did not rely on a detailed forward
simulation of the early phase of the action potential. This
allowed us carry out the T-wave analysis using a relatively
coarse mesh (148075 nodes and 826442 elements) that was
still able to comfortably capture the wavelength of the repo-
larisation wave.
The monodomain model was used to compute propagation
Fig. 2. A reference system to show the locations of the core of simulated
of electrical activity in tissue. 12-lead ECGs were computed ischemia. A. The human ventricular geometry with 17 different locations lying
as a post-process wherein the membrane potential computed upon the epicardial surface for the core of simulated spherical ischemia in the
in the monodomain simulation were given as input to forward left ventricular wall. Brown, at apex; red, along the septal wall; green, along
the posterior left ventricular wall; yellow, at left ventricular free wall; blue, at
solver for the electrical potential in the torso (result shown in anterior left ventricular wall. B. A color-coded cross schematically illustrates
Figure 1C). the ischemic locations, and is a convenient method to visualize alterations
in ECG-based measurements with applied simulation ischemia of different
size and severity in different regions, as shown in subsequent Figures. The
III. R ESULTS locations are also coded by direction for presentation in Tables: A, apex; E1-4,
along septum, with 4 being the closest to the base; S1-4, along posterior left
For each of the cases with varying ischemic severity, size, ventricular wall; W1-4, along left ventricular free wall; N1-4, along anterior
and location presented above, we calculated the observability left ventricular wall.
of the ischemia in the computed 12-lead ECG by using the
norm:

0018-9294 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2016.2600198, IEEE
Transactions on Biomedical Engineering

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING 3

A L1 norm over cardiac cycle (0-400ms) B L1 norm over ST segment (80-250ms) C L1 norm over T-wave (250-400ms)
stage 1 stage 2 phase 1b stage 1 stage 2 phase 1b stage 1 stage 2 phase 1b

0.5 0.5 0.5

ischemic radius

ischemic radius

ischemic radius
1.0 1.0 1.0

V (mV)
1.5 1.5 1.5

2.0 2.0 2.0

Fig. 3. Examining average ECG metrics for ischemia detection. Results presented are in change in voltage from baseline with respect to a reference ECG
(computed without an imposed simulated ischemia). For example, the darkest blue indicates a very small deviation from the reference ECG, whereas the
darkest red indicates 0.3mV or more in average deviation from the reference 12 lead ECG. (A) L1 norm over the entire cardiac cycles (0 to 400 ms duration).
(B) L1 norm over the ST segment (80-250 ms). (C) L1 norm over the T-wave (250-400 ms).

TABLE I As an example, observe the difference in site N2 for r = 1

PARAMETERS OF THE I SCHEMIC C ELL MODEL cm and stage 1 (Figure 3B and C). There is little information
in the ST-segment (i.e. N2 is blue), but a significant amount of
Parameter Control stage 1 stage 2 phase 1b information in the T-wave (i.e. N2 is cyan). Figure 1C shows
[K + ]o (mmol/l) 5.4 8.7 12.5 15.0
gN a ,gCaL (%) 100 87.5 75 50 the ECG in question; this is indeed the case as evidenced
dVm N a (mV) 0.0 1.7 3.4 3.4 directly on the ECG. Specifically, we see significant deviations
[AT P ]i (mmol/l) 6.8 5.7 4.6 3.8 for the T-wave in leads V3 and V4.
[ADP ]i (mol/l) 15 87.5 99 101.5
kN aCa (%) 100 100 100 20 In order to assess the visibility of ischemia-induced alter-
PN aK (%) 100 100 100 30 ations in the T-wave, we examined the change in a variety of
Vrel (%) 100 100 100 5 quantitative T-wave metrics with respect to reference lead I:
Vmaxup (%) 100 100 100 90
(a) T-wave time-to-peak, (b) T-wave duration, (c) T-wave peak,
and (d) area under the T-wave. The most promising metrics for
identification of ischemic regions of medium size and severity,
(a) and (b), are examined in Figure 4 (results for (c) and (d) not
1 1
12 t1
d(r, e) = |ri (t) ei (t)| dt (1) shown). While both (a) the change in time to T-wave peak, and
12 i=1 t1 t0 t0 (b) the change in T-wave duration are metrics permitting clear
identification of a stage 2, phase 1a ischemic region of radius =
where r is the reference ECG (i.e. the ECG obtained when
1.0cm (Figure 4A), an early ischemic region of limited extent
ischemia is absent from the ventricles), and e is the ECG
(stage 1, phase 1a, radius = 0.5cm) is not so easily identified
obtained when an ischemic region exists.
by the same differential metrics (Figure 4B).
Figure 3 examines the L1 norm (as given in equation (1)
above) over the entire cardiac cycle (i.e. t0 = 0ms and t1 = To further evaluate the potential visibility of ischemia-
400ms), the ST-segment (t0 = 80ms and t1 = 250ms), and the induced alterations in the T-wave for early ischemia of limited
T-wave (t0 = 250ms and t1 = 400ms), in Panels A, B, and C, extent, we again examine the calculated quantitative T-wave
respectively. Immediately, independent of metric, we see that metrics calculated with respect to reference lead I (i.e. (a)
early-stage, small ischemic regions (upper left corner) change T-wave time-to-peak, (b) T-wave duration, (c) T-wave peak,
the ECG relatively little as compared to larger and/or later and (d) area under the T-wave, as shown in Table I). For this
stage ischemic events. Furthermore, ischemic regions closest case (stage 1, phase 1a, r=0.5 cm), Tables IV and V reveal, the
to the chest are the most visible, followed by the ones on the change in the peak of the T-wave and/or the change in the area
LV free wall (2B). under the T-wave may be far better indicators of the existence
In Figure 3B and C we have looked at the deviations for of developing ischemia than time-to-peak, (a), or duration, (b).
only the ST-segment and the T-wave segment, respectively. As evidenced by Figure 5, both change in the peak of
In the chosen L1 norm, more information emerges in the the T-wave, (c), (Figure 5A) and change in the area under
T-wave, particularly for earlier ischemia of lesser extent, as the T-wave, (d) (Figure 5B), may be appropriate measures
compared to the ST-segment, suggesting that that T-wave can for detection of developing ischemic regions of lesser spatial
be additionally probed as a source of information regarding extent and severity as compared to change in time-to-peak, (a),
developing ischemia. or change in T-wave duration, (b) (as seen in Figure 4). ST-

0018-9294 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2016.2600198, IEEE
Transactions on Biomedical Engineering

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING 4

A B
30 30

Time to T-wave peak (ms)

Time to T-wave peak (ms)

0 0

-30 -30

C D
105 105

T-wave duration (ms)

T-wave duration (ms)

0 0

-100 -100

Fig. 4. Examining T-wave metrics for ischemia detection. Change in time to T-wave peak versus lead I and compared to healthy reference (circle exterior to
each cross) for an ischemia at stage 2 of phase 1a and of radius 1cm (A) and that for an ischemia at stage 1 of phase 1a and of radius 0.5cm (B). (C) and
(D) show results for change in T-wave duration for the same severity and size of simulated ischemia, respectively.

Time-to-peak N2 W2 S2 E2
III 1 1 1 1
ventricles, towards blended clinical methods, e.g. combining
V1 2 1 1 0 cardiac ultrasound and simulation techniques to improves risk
V3 -7 2 2 1 assessment and outcomes for NSTEACS patients.
V5 0 0 2 1 To this end, we examined several quantitative T-wave
T-wave duration N2 W2 S2 E2
III 0 2 0 1 metrics and well as ST elevation in silico and compared
V1 -2 2 2 -2 their ability to identify the presence of developing ischemia
V3 16 0 3 0 in detailed 3D human electrophysiological models. Results,
V5 2 -5 -1 1
as highlighted in Figures 3B and 5C, reveal ST segment
T-wave peak N2 W2 S2 E2
III 0.02 -0.02 -0.00 0.01 elevation to be a suboptimal metric for the identification of
V1 0.04 -0.01 -0.01 0.02 early ischemia of smaller spatial extent as compared to several
V3 0.35 -0.01 -0.01 0.02 T-wave metrics (Figure 3C).
V5 0.07 0.04 -0.00 -0.00
Area under T-wave N2 W2 S2 E2
Examining the potential utility of T-wave metrics further
III 0.9 -0.7 0.3 1.5 was thus warranted. As shown in Figure 4 and Figure 5A and
V1 1.6 -0.4 -0.2 1.4 B, these can be seen to be more sensitive to earlier, developing
V3 23.9 -0.0 -0.4 1.5 cardiac ischemia in assessment metrics independent of a basal
V5 4.1 2.3 -0.6 -0.6
ECG (calculated with respect to lead I and compared to healthy
TABLE II reference). Notably, different metrics may be appropriate as
D IFFERENCE AS COMPARED TO NON - ISCHEMIC REFERENCE IN THE
labeled T-wave metric FOR SMALL AND EARLY ISCHEMIC REGIONS IN the ischemia develops; regional ischemia of medium size and
THE LOCATION INDICATED ( PLEASE SEE F IGURE 1A FOR REFERENCE ). in phase 2 of stage 1a was best assessed by change in the
time to T-wave peak and change in T-wave duration (Figure
4A and B), while earlier, smaller ischemic regions were better
identified by change in the T-wave peak and the area under
segment elevation is the least sensitive of all metrics studied
the T-wave (Figure 5A and B).
here, as shown in Figure 5C.
The main challenge in translating present results into real
ECG studies is the fact that we employ deterministic simu-
IV. C ONCLUSION lations to represent a phenomenon that is variable over both
The aim of this paper was to initially investigate the time and heart rate. The sensitivity of the variations in our
feasibility of using ECG T-wave metrics to offer more in- biomarkers might be too low to detect ischemia with an
formation about developing ischemic regions in the human adequate level of confidence during screening in the clinical

0018-9294 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2016.2600198, IEEE
Transactions on Biomedical Engineering
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING
Ischemia phase 1a stage 1, radius = 0.5 cm
A B
1.65
T-wave peak (mV)
-0.15
Fig. 5. Examining two additional quantitative T-wave metrics for early ischemia detection. Change in T-wave peak (mV) versus lead I and compared to
healthy reference (circle exterior to each cross) for an ischemia at stage 1 of phase 1a and of radius 0.5cm (A), change in the area under the T-wave (mV)
for the same severity and size of simulated ischemia (B), and, for comparison, ST elevation under identical ischemic conditions (microV) (C).
scenario. However, in the time scales of the development of
ischemia (minutes), it is possible to measure multiple beats
and average the result, filtering this variability over time. The
variability with respect to heart rate can be quantified, and our
biomarkers could be corrected with respect to heart rate in a
similar fashion as the QTc interval is used to assess arrhythmia
risk. Changes in the variability of our biomarkers themselves
might also be used as additional information, that can not be
captured by the current study.
These results strongly imply the need to have several quan-
titative measures available in clinical practice, prospectively
based on the T-wave of the ECG. Despite acknowledged
challenges associated with T-wave variability in vivo and the
need for further work, the use of temporally-differential T-
wave signals may provide an appropriate baseline for T-wave
metric assessment in the patient admitted with NSTEACS and
in need of triage for potential interventions.
ACKNOWLEDGMENT
The authors gratefully acknowledge support via partnership
in the Center for Cardiological Innovation at Oslo Univer-
sity Hospital and by a Center of Excellence grant from the
Research Council of Norway to the Center for Biomedical
Computing at Simula Research Laboratory.
R EFERENCES
[1] Salah S Al-Zaiti, Clifton W Callaway, Teri M Kozik, Mary G Carey,
and Michele M Pelter. Clinical Utility of Ventricular Repolarization
Dispersion for Real-Time Detection of Non-ST Elevation Myocardial
Infarction in Emergency Departments. Journal of the American Heart
Association, 4(7):112, 2015.
[2] Kedar Kirtikumar Aras. Bioelectric source characterization of acute
myocardial ischemia. Dissertation, University of Utah, USA., 2015.
[3] J. D. Bayer, R. C. Blake, G. Plank, and N. A. Trayanova. A novel rule-
based algorithm for assigning myocardial fiber orientation to computa-
tional heart models. Annals of Biomedical Engineering, 40:22432254,
2012.
[4] Scirica BM. Acute coronary syndrome: emerging tools for diagnosis
and risk assessment. J Am Coll Cardiol, 55:140315, 2010.
[5] D. Durrer, R. Th. Van Dam, G. E. Freud, M. J. Janse, F. L. Meijler,
and R. C. Arzbaecher. Total Excitation of the Isolated Human Heart.
Circulation, 41(6):899912, June 1970.
[6] Baumert M. Hasan MA, Abbott D. Beat-to-beat QT interval variability
and T-wave amplitude in patients with myocardial infarction. Physiol
Meas, 34(9):107583, 2013.
0018-9294 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
5
C
0.65 0.3
Area under T-wave (mV)
ST elevation (V)
-0.1 0.0
[7] David U J Keller, Daniel L Weiss, Olaf Dossel, and Gunnar Seemann.
Influence of I(Ks) heterogeneities on the genesis of the T-wave: a
computational evaluation. IEEE Trans Biomed Eng, 59(2):311322,
February 2012.
[8] Yamamoto M Sano K Harigaya H Okuda K Ozaki Y. Sobue Y,
Watanabe E. Clinical Utility of Ventricular Repolarization Dispersion
for Real-Time Detection of Non-ST Elevation Myocardial Infarction in
Emergency Departments. Europace, 13(11):16128, 2011.
[9] K H W J ten Tusscher. A model for human ventricular tissue. American
journal of physiology. Heart and circulatory physiology, 286(4):H1573
H1589, December 2003.
[10] K. H. W. J. ten Tusscher, R. Hren, and A. V. Panfilov. Organization
of ventricular fibrillation in the human heart. Circ Res, 100:e87e101,
2007.
[11] Mathias Wilhelms, Olaf Dssel, and Gunnar Seemann. In silico investiga-
tion of electrically silent acute cardiac ischemia in the human ventricles.
IEEE Trans Biomed Eng, 58:29612964, 2011.
Glenn Terje Lines He received the M.Sc. and Ph.D. degrees from the
Department of Informatics at the University of Oslo, Oslo, Norway, in 1994
and 1999, respectively. In the doctoral thesis he developed a simulator for the
electrical activity of the heart. His current research interests include numerical
methods for partial differential equations and computer simulation of cardiac
electrophysiology.
Bernardo Lino de Oliveira Dr. Bernardo Lino de Oliveira received his
PhD from the Department of Informatics at the University of Oslo, Oslo,
Norway, in 2014. In his doctoral thesis he developed models and methods for
computational simulations of the electromechanical activity in the heart. His
current research interests include mathematical and computational models to
simulate the electrical, mechanical and metabolic activities in the heart.
Mary M. Maleckar Dr. Mary (Molly) Maleckar received her PhD from the
Department of Biomedical Engineering/Institute for Computational Medicine
at Johns Hopkins University in 2008 and began work at Simula in the
Scientific Computing Department in January 2009, where she is currently
a Senior Research Scientist.