MEASURING P-WAVE MORPHOLOGICAL VARIABILITY FOR
AF-PRONE PATIENTS IDENTIFICATION
Valeria Villani
,
, Antonio Fasano
, Luca Vollero
, Federica Censi
and Giuseppe Boriani
Universit
`
a Campus Bio-Medico di Roma, Roma, Italy
Department of Technology and Health, Italian National Institute of Health, Roma, Italy
Institute of Cardiology, University of Bologna and Azienda Ospedaliera S. Orsola-Malpighi, Bologna, Italy
Keywords:
ECG, Atrial fibrillation, P-wave variability.
Abstract:
Atrial fibrillation is the most common arrhythmia encountered in clinical practice. Abnormal P-waves have
been observed in patients prone to AF and the analysis of P-waves from surface electrocardiogram has been
extensively used to identify patients prone to atrial arrhythmias. Measuring the temporal variability of P-
waves, i.e., the variation over time of morphological characteristics of single P-waves, may represent a useful
method for characterizing and predicting AF cases. In this paper, we propose a method for the statistical
analysis of P-waves variability. It is based on the evaluation of the empirical distribution function of differences
energy among normalized P-waves. The proposed method seems promising for capturing atrial anomalies and
identifying patients prone to AF.
1 INTRODUCTION
Atrial fibrillation (AF) is the most common arrhyth-
mia encountered in clinical practice (about 4.5 million
people in the European Union): its prevalence is esti-
mated between 0.4% and 1% in the whole population
and increases with age (5% for patients older than 65
years and 8% for those older than 80 years). Although
it is not a lethal disease, AF may increase mortality
up to 2-fold, primarily because of embolic stroke. In-
deed, the lack of coordinated atrial contraction leads
to unusual fluid flow states through the atrium. These
could favor the formation of thrombus at risk to em-
bolize, especially after return to normal sinus rhythm.
When normal cardiac impulse travels through
atrial myocardium, surface electrocardiogram (ECG)
recordings show the P-wave. If atrial depolarization
patterns differ from normal ones, P-waves may ap-
pear prolonged and highly variable. Indeed, slowed
conduction velocity in several atrial regions together
with different cell refractory periods are believed to
be the electrophysiological conditions provoking and
maintaining AF (Platonov et al., 2000).
Abnormal P-waves have been observed in AF-
prone patients and the analysis of P-waves from
surface ECG has been extensively used to identify
patients prone to atrial arrhythmias, especially AF
(Dilaveris et al., 1998; Darbar et al., 2002; Michelucci
et al., 2002; Dilaveris and Gialafos, 2001).
In this context, P-wave analysis is usually per-
formed following both the conventional 12-leads
ECG approach and the three bipolar orthogonal leads
one, used to determine the P-wave vector magnitude
(Clavier et al., 2002; Klein et al., 1995; Jordaens et al.,
1998; Dilaveris and Gialafos, 2002). Given the re-
latively low P-wave amplitude with respect to back-
ground noise, both approaches use signal averaging
techniques to obtain a P-wave template. The analy-
sis of P-wave template turned out to be useful in dis-
criminating patients at risk of developing AF or with
paroxymal AF. A typically considered parameter is
the P-wave duration.
According to literature, the classical approach fol-
lowed to study the relation between P-wave charac-
teristics and AF is the analysis of P-wave templates.
Nevertheless, following an approach similar to that
used for T-waves in the analysis of ventricular repo-
larization (Pueyo et al., 2009), it could be worth in-
vestigating P-waves variability, i.e., the variation over
time of morphological characteristics of P-waves. In-
deed, as shown in Fig. 1 and Fig. 2, variability of P-
waves morphology usually appears to be significantly
higher in AF patients than in healthy controls. Hence,
such an approach seems to be more suitable for in-
vestigating the complicated electrophysiological con-
ditions believed to provoke and maintain AF.
481
Villani V., Fasano A., Vollero L., Censi F. and Boriani G..
MEASURING P-WAVE MORPHOLOGICAL VARIABILITY FOR AF-PRONE PATIENTS IDENTIFICATION.
DOI: 10.5220/0003164204810484
In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2011), pages 481-484
ISBN: 978-989-8425-35-5
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
0 50 100 150
−0.2
−0.15
−0.1
−0.05
0
0.05
0.1
Time [ms]
Amplitude [mV]
Figure 1: Butterfly plot of P-waves belonging to a patient
prone to AF.
0 50 100 150
−0.2
−0.15
−0.1
−0.05
0
0.05
0.1
Time [ms]
Amplitude [mV]
Figure 2: Butterfly plot of P-waves belonging to a healthy
subject.
The aim of this work is to propose an approach to
measure P-waves variability that could be effective in
identifying AF prone patients.
2 METHOD
This section describes the strategy we have develo-
ped in order to measure P-waves variability within an
ECG tracing of cardiac cycles.
2.1 Study Population
The study population consisted of 37 subjects, di-
vided into 2 groups: 21 patients had a persistent
form of atrial fibrillation and underwent electrical car-
dioversion (AF group); 16 subjects had no history of
AF and have been considered as controls in this paper
(control group). The AF group consists of 10 patients,
who experienced at least another documented episode
of AF within 3 months after cardioversion (AF relapse
group), and 11 patients, who did not experience any
documented AF episodes after cardioversion (no-AF
relapse group). ECG was recorded for 5 minutes us-
ing a high resolution mapping system (Biosemi Ac-
tiveTwo, Amsterdam, Netherlands) with a sampling
frequency of 2048Hz, a resolution of 24-bit (about
30nV LSB), and a bandwidth from DC to 400Hz. For
AF patients, ECG was recorded after successful car-
dioversion.
2.2 Pre-processing of P-waves
P-waves were first isolated according to the method
proposed in a previous paper (Censi et al., 2007). Af-
ter the detection of each R-wave, P-waves were ex-
tracted in a 200ms-long window starting from 300 ms
before the R-wave. Then, in order to remove base-
line wander, a beat-by-beat linear piecewise interpo-
lation was performed, selecting fiducial points from
TP and PQ tracks of each beat for linear interpola-
tion. Finally, ectopic atrial signals or P-waves with
excessive noise were excluded by template matching
of each P-wave as described in (Censi et al., 2007).
The extracted P-waves are finite length sampled sig-
nals available in the following as vectors having the
same size. P-waves from ECG II lead were consid-
ered.
2.3 Quantification of the Energy of
P-waves Differences
P-waves selected during the above mentioned pre-
processing phase are analyzed in order to define and
characterize statistical indicators of AF phenomena
based on P-waves morphological variability.
We denote by p
p
p
i
the i-th segmented P-wave of the
ECG tracing under analysis. Let us assume to have
N of such waves each L samples long. In order to
emphasize the morphological differences instead of
absolute differences among waves, a unit-norm nor-
malization is applied to each of them:
ˆ
p
p
p
i
=
p
p
p
i
||p
p
p
i
||
.
Denote by ε
i, j
the energy of the difference bet-
ween
ˆ
p
p
p
i
and
ˆ
p
p
p
j
:
ε
i, j
= ||
ˆ
p
p
p
i
ˆ
p
p
p
j
||
2
.
In the following we consider the squared Eu-
clidean norm:
ε
i, j
=
L
k=1
ˆp
i,k
ˆp
j,k
2
where ˆp
i,k
is the k-th component of the vector
ˆ
p
p
p
i
.
However other norms revealing specific aspects of
waves can be used.
BIOSIGNALS 2011 - International Conference on Bio-inspired Systems and Signal Processing
482
The empirical cumulative distribution function of
the differences energy among normalized P-waves is
hence obtained as:
F
E
(ε) =
2
N (N 1)
N
i< j
1
1
1
{ε
i, j
ε}
where 1
1
1
{·}
denotes the indicator function of the set
within brackets.
Although the empirical cumulative distribution
function is rich of information about waves variabi-
lity, the availability of descriptive parameters is also
of interest. In the following we consider the sample
mean and the sample variance of the differences ener-
gies as indicators:
µ =
2
N (N 1)
N
i< j
ε
i, j
σ
2
=
2
N (N 1) 2
N
i< j
[ε
i, j
µ]
2
3 RESULTS
Before ECG acquisition, AF patients underwent elec-
trical cardioversion. In Fig.s 3, 4 and 5 we report the
empirical cumulative distribution functions pertain-
ing to controls, no-AF relapse group, and AF-relapse
group, respectively. Each curve corresponds to a pa-
tient.
Fig. 3 shows the empirical cumulative distribu-
tion function of P-waves differences energy for pa-
tients belonging to the control group. All the func-
tions exhibit similar behavior, with a high probability
of having low differences among different P-waves.
Conversely, Fig. 4 and Fig. 5 show the empirical cu-
mulative distribution function of P-waves differences
energy for AF patients. In this case the probability
of significant differences among P-waves increases.
This effect is considerably more evident in the case of
AF relapse group. It is worth noting that the distri-
bution functions of AF relapse group are significantly
different from the corresponding functions pertaining
to the control group. Moreover, considering the AF
relapse and the no-AF relapse groups it is possible
to appreciate differences in the distribution functions
that allows same level of discrimination between the
two groups.
The joint analysis of sample mean and sample
variance confirms the difference between the control
group and the AF patients groups. Fig. 6 shows the
log-log plot of the pair sample mean and sample vari-
ance, namely (µ, σ
2
), for each patient. We can distin-
guish two regions: (i) low values region and (ii) high
0 0.1 0.2 0.3 0.4
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Differences energy
Probability
Figure 3: Controls: empirical cumulative distribution func-
tion of P-waves differences energy.
0 0.1 0.2 0.3 0.4
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Differences energy
Probability
Figure 4: AF patients (no-AF relapse group): empirical
cumulative distribution function of P-waves differences en-
ergy.
0 0.1 0.2 0.3 0.4
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Differences energy
Probability
Figure 5: AF patients (AF relapse group): empirical cumu-
lative distribution function of P-waves differences energy.
values region. In the low values region both mean
and variance of the differences energy is small. This
region identifies controls and, hence, patients with-
out AF problems. Conversely, the high values re-
gion identifies AF patients belonging to the AF re-
lapse group. In this figure, control group results to
be well separated from AF-relapse group. No-AF
relapse group turns out to be characterized by inter-
mediate values with respect to other groups. The
MEASURING P-WAVE MORPHOLOGICAL VARIABILITY FOR AF-PRONE PATIENTS IDENTIFICATION
483
Figure 6: Sample mean and sample variance of differences
energy for AF group and control group.
proposed method seems able to discriminate between
healthy and AF patients. In this regard it could be of
help in identifying AF-prone patients.
4 CONCLUSIONS
The availability of methods for measuring P-waves
variability over time represents a useful tool to deeply
understand the mechanisms underlying the atrial elec-
trical substrate, and may help in identifying patients
with substrates predisposing to AF. Indeed, the P-
wave variability is related to the dispersion of atrial
refractory period. In this paper, we propose a method
to measure such variability. It is based on the compu-
tation of the empirical cumulative distribution func-
tion of the differences energy among normalized P-
waves. The proposed method is able to discriminate
between AF patients and control subjects. This fact is
highlighted by the joint analysis of estimated statisti-
cal parameters such as the sample mean and the sam-
ple variance of differences energy. It is worth noting
that the proposed method exhibits some ability even
in discriminating between patients who experienced
AF relapse from patients who did not. In conclusion,
the analysis of the empirical distribution function of
differences energy among normalized P-waves seems
promising for capturing atrial anomalies and identify-
ing patients prone to AF.
REFERENCES
Censi, F., Calcagnini, G., Ricci, C., Ricci, R. P., Santini,
M., Grammatico, A., and Bartolini, P. (2007). P-
wave morphology assessment by a gaussian functions-
based model in atrial fibrillation patients. IEEE Trans
Biomed Eng, 54:663–672.
Clavier, L., Boucher, J. M., Lepage, R., Blanc, J. J., and
Cornily, J. C. (2002). Automatic P-wave analysis
of patients prone to atrial fibrillation. Med Biol Eng
Comput, 40:63–71.
Darbar, D., Jahangir, A., Hammill, S. C., and Gersh, B. J.
(2002). P wave signal-averaged electrocardiography
to identify risk for atrial fibrillation. Pacing Clin Elec-
trophysiol, 25:1447–1453.
Dilaveris, P. E., Gialafos, E. J., Sideris, S. K., Theopis-
tou, A. M., Andrikopoulos, G. K., Kyriakidis, M.,
Gialafos, J. E., and Toutouzas, P. K. (1998). Sim-
ple electrocardiographic markers for the prediction of
paroxysmal idiopathic atrial fibrillation. Am. Heart J.,
135:733–738.
Dilaveris, P. E. and Gialafos, J. E. (2001). P-wave disper-
sion: a novel predictor of paroxysmal atrial fibrilla-
tion. Ann Noninvasive Electrocardiol, 6:159–165.
Dilaveris, P. E. and Gialafos, J. E. (2002). Future concepts
in P wave morphological analyses. Card Electrophys-
iol Rev, 6:221–224.
Jordaens, L., Tavernier, R., Gorgov, N., Kindt, H., Dim-
mer, C., and Clement, D. L. (1998). Signal-averaged
P wave: predictor of atrial fibrillation. J. Cardiovasc.
Electrophysiol., 9:S30–34.
Klein, M., Evans, S. J., Blumberg, S., Cataldo, L., and Bo-
denheimer, M. M. (1995). Use of P-wave-triggered,
P-wave signal-averaged electrocardiogram to predict
atrial fibrillation after coronary artery bypass surgery.
Am. Heart J., 129:895–901.
Michelucci, A., Bagliani, G., Colella, A., Pieragnoli, P.,
Porciani, M. C., Gensini, G., and Padeletti, L. (2002).
P wave assessment: state of the art update. Card Elec-
trophysiol Rev, 6:215–220.
Platonov, P. G., Carlson, J., Ingemansson, M. P., Roijer,
A., Hansson, A., Chireikin, L. V., and Olsson, S. B.
(2000). Detection of inter-atrial conduction defects
with unfiltered signal-averaged P-wave ECG in pa-
tients with lone atrial fibrillation. Europace, 2:32–41.
Pueyo, E., Martinez, J. P., and Laguna, P. (2009). Car-
diac repolarization analysis using the surface electro-
cardiogram. Philos Transact A Math Phys Eng Sci,
367:213–233.
BIOSIGNALS 2011 - International Conference on Bio-inspired Systems and Signal Processing
484