10 August 2022
Electrodermal activity in bipolar patients during affective elicitation / Greco Alberto; Valenza Gaetano;
Lanata Antonio; Rota Giuseppina; Scilingo Enzo Pasquale. - In: IEEE JOURNAL OF BIOMEDICAL AND
HEALTH INFORMATICS. - ISSN 2168-2194. - ELETTRONICO. - 18(2014), pp. 1865-1873.
[10.1109/JBHI.2014.2300940]
Original Citation:
Electrodermal activity in bipolar patients during affective elicitation
Published version:
10.1109/JBHI.2014.2300940
Terms of use:
Publisher copyright claim:
(Article begins on next page)
La pubblicazione è resa disponibile sotto le norme e i termini della licenza di deposito, secondo quanto
stabilito dalla Policy per l'accesso aperto dell'Università degli Studi di Firenze
(https://www.sba.unifi.it/upload/policy-oa-2016-1.pdf)
Availability:
This version is available at: 2158/1192138 since: 2021-06-11T13:00:18Z
Questa è la Versione finale referata (Post print/Accepted manuscript) della seguente pubblicazione:
FLORE
Repository istituzionale dell'Università degli Studi
di Firenze
Open Access
DOI:
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, VOL. 18, NO. 6, NOVEMBER 2014 1865
Electrodermal Activity in Bipolar Patients during
Affective Elicitation
Alberto Greco, Student Member, IEEE, Gaetano Valenza, Member, IEEE, Antonio Lanata, Member, IEEE,
Giuseppina Rota, and Enzo Pasquale Scilingo, Member, IEEE
Abstract—Bipolar patients are characterized by a pathological
unpredictable behavior, resulting in fluctuations between states
of depression and episodes of mania or hypomania. In the cur-
rent clinical practice, the psychiatric diagnosis is made through
clinician-administered rating scales and questionnaires, disregard-
ing the potential contribution provided by physiological signs. The
aim of this paper is toinvestigate how changes in theautonomicner-
vous system activity can be correlated with clinical mood swings.
More specifically, a group of ten bipolar patients underwent an
emotional elicitation protocol to investigate the autonomic ner-
vous system dynamics, through the electrodermal activity (EDA),
among different mood states. In addition, a control group of ten
healthy subjects were recruited and underwent the same protocol.
Physiological signals were analyzed by applying the deconvolutive
method to reconstruct EDA tonic and phasic components, from
which several significant features were extracted to quantify the
sympathetic activation. Experimental results performed on both
the healthy subjects and the bipolar patients supported the hy-
pothesis of a relationship between autonomic dysfunctions and
pathological mood states.
Index Terms—Bipolar disorder, deconvolutive analysis, electro-
dermal activity (EDA), mood recognition.
I. INTRODUCTION
B
IPOLAR disorder is a chronic illness involving millions
of people in Europe and in the United States (see the epi-
demiological study in [1]). Patients experience mood swings
whose symptoms can be associated to one of the following
psychophysiological states: depressive, maniac, mixed, and eu-
thymic. During depressive episodes, patients feel sad and, some-
times, desperate. Other neurovegetative symptoms including
loss of appetite and sleep are also present. Depressed patients
might also experience thoughts of ruin, guilt, or death including
suicidal thoughts that might lead to suicide attempts. During
manic episodes, patients are hyperactive, and often experience
Manuscript received July 26, 2013; revised November 5, 2013; accepted Jan-
uary 10, 2014. Date of publication February 4, 2014; date of current version
November 3, 2014. This work was partially supported by the EU Commission
under Contract ICT-247777 Psyche and Project No. 601165 WEARHAP.
A. Greco, G. Valenza, A. Lanata, and E. P. Scilingo are with the Depart-
ment of Information Engineering and Research Center “E. Piaggio”, Faculty of
Engineering, University of Pisa, Pisa 16-56122, Italy (e-mail: alberto.
greco@centropiaggio.unipi.it; g.valenza@ieee.org; a.lanata@centropiaggio.
unipi.it; e.scilingo@centropiaggio.unipi.it).
G. Rota is with the Department of Surgical, Medical, Molecular, and Critical
Area Pathology, Section of Psychology, University of Pisa, Pisa 67-56100, Italy
(e-mail: g.rota@med.unipi.it).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JBHI.2014.2300940
a reduction of the need to sleep. Mixed states are characterized
by both depressive and hyperactivity symptoms. In the intervals
between these episodes, patients typically experience periods of
relatively good emotional balance (labeled as euthymia). More-
over, mood swings are also usually accompanied by anxiety,
which is associated with bipolar disorder either as a symptom
of the bipolar disorder itself or as a separate pathological con-
dition [2].
In spite of the great impact on the population and healthcare
costs, current clinical practice still relies only on the physician
expertise, rating scales and questionnaires, such as the Bauer
Internal Mood Scale, the Hamilton Scale for Depression, and
the Young Mania scale [3]. Physiological parameters (e.g., bi-
ological markers, physiological signals, etc.) are not taken into
account for diagnosis or follow-up purposes [4]–[6]. As a matter
of fact, there is the need of more objective parameters for the
diagnosis of mental disorders. Mental disorders are long-term
illnesses and may remain undetected for years before they are
properly diagnosed and put under treatment. Moreover, patients
are extremely heterogeneous with respect to the phenomenology
and severity of symptoms, number, and duration of episodes, as
well as the time interval between them. Finally, other disorders
may also be present (i.e., comorbidity).
Previous research has shown a link between the auto-
nomic nervous system (ANS) dysfunctions and bipolar disor-
der [7]–[12]. Specifically, studies on sleep [13], voice analy-
sis [14], and circadian heart rate rhythms [15], [16] showed to be
sensitive to changes in the clinical state, suggesting that these pa-
rameters may be considered as markers of clinical change. More-
over, it is known that electrodermal hypoactivity is present dur-
ing depression in both unipolar and bipolar patients [17], [18].
This condition is stable over time, and does not appear to depend
on experimental conditions or stimulus characteristics [19]. In a
recent study, we demonstrated that a single variable approach is
not a reliable method for characterizing mood swings in bipolar
patients while using heart rate variability and respiration activity
series [10], [12]. Nevertheless, a complete and comprehensive
ANS characterization should also rely on other physiological
signals that are strictly related to the sympathetic nerve activity
such as the electrodermal activity (EDA). In the present study,
we investigated EDA dynamics in bipolar patients during an
emotional stimulation paradigm. Since the changes on EDA are
directly related to the sympathetic activity [20], EDA analy-
sis could serve as an effective ANS marker for characterizing
different mood states.
The stimulation protocol proposed in this paper is based
on displaying pictures selected from the international affective
2168-2194 © 2014 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.
Authorized licensed use limited to: Universita degli Studi di Firenze. Downloaded on June 11,2021 at 10:45:08 UTC from IEEE Xplore. Restrictions apply.
1866 IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, VOL. 18, NO. 6, NOVEMBER 2014
picture system (IAPS) [21]) and pictures from the thematic ap-
perception test (TAT) [22]. Theywere presented to the patients in
order to elicit emotional reactions. The IAPS database is widely
used for studies that assess emotional processing (e.g., see pre-
vious studies in [23]–[27]), and it is comprised of hundreds of
pictures which associated a specific emotional rating in terms of
arousal and valence. Arousal refers to the physiological activa-
tion that is elicited by an emotionally salient image resulting in a
subjective state of calmness or excitement. Valence refers to the
experience of pleasantness or unpleasantness induced by view-
ing the image. The TAT is a projective psychological test that
is supposed to reveal repressed aspects of personality. Such an
experimental protocol was administered to ten bipolar patients
as well as ten healthy subjects. Concerning the methodology
of signal processing, we used a deconvolutive approach [28] in
order to retain consecutive sympathetic responses which can be
overlapped whenever the interstimulus interval is shorter than
the previous one.
Similarly to our previous investigation [10], [12], the present
study was carried out in the frame of the European project
PSYCHE, which stands for personalized monitoring systems
for care in mental health. Within this project, a personalized,
pervasive, cost-effective, and multiparametric monitoring sys-
tem based on textile platforms and portable sensing devices
was devised for the long-term and short-term analysis of mood
disorders [10], [12], [14], [29].
II. M
ATERIALS AND METHODS
A. Patient Recruitment and Experimental Protocol
Ten patients affected by bipolar disorder I or II were selected
for this study. None of them had suicidal tendencies, delusions,
or hallucinations. Patients were admitted to the psychiatric unit
of the hospital and periodically screened through a psychiatric
interview. Before each acquisition, a mood label among “eu-
thymic”, “depressed”, “maniac”, and “mixed-state” was associ-
ated to each patient/acquisition. As a control group, a group of
healthy subjects were enrolled and participate to the study. In
particular, ten healthy subjects (five females, age ranged from
20 to 32), i.e., not suffering from both cardiovascular and evi-
dent mental pathologies, were asked to fill out the Patient Health
Questionnaire
TM
. All participants showed scores lower than 5.
Such a cut-off value was chosen in order to avoid the presence
of either middle or severe personality disorders [30].
An ad hoc affective elicitation experimental was administered
to both the healthy and bipolar patients group. In particular,
such an experimental protocol, graphically shown in Fig. 1, was
structured as follows:
1) 5 min at rest with closed eyes;
2) 5 min at rest with open eyes;
3) 6-min slideshow of IAPS pictures with high arousal and
negative valence;
4) up to 4 min of pictures gathered from TAT.
As described previously, the protocol is split into two ses-
sions: rest and emotional elicitation. The latter session is di-
vided, in turn, into two stages, both of which are intended to
elicit a variation of the ANS response. Specifically, IAPS pic-
Fig. 1. Block scheme of the experimental protocol.
TABLE I
C
LINICAL EVALUATIONS OF THE PATIE N TS
tures lasted for 2 s presenting negative emotional contents (high
arousal and negative valence). The same IAPS pictures were
presented to all patients and healthy subjects and nobody was
asked to score the elicited level of arousal and valence. The
images were chosen according to the following characteristics:
arousal score >6.7; valence <4.5. Afterwards, patients were
invited to tell a story based on the input coming from the TAT
pictures. However, in order to avoid biased results related to the
IAPS and TAT sequential order, the IAPS–TAT and TAT–IAPS
session orders were randomly interchanged. The hypothesis of
this study is that the ANS differentially reacts to such emo-
tional stimuli upon different pathological mood states. During
the whole duration of the protocol, the EDA signal was acquired
using the BIOPAC MP150 system with a sampling frequency
of 1000 Hz. EDA sensors were placed on the distal phalanx of
the second and third finger of the nondominant hand, imposing
a dc voltage of 0.5 V. The protocol was run for a follow-up
periodofupto75 days. Patients repeated the protocol at each
mood change, whereas the healthy subjects repeated the exper-
iment twice within two weeks in order to investigate possible
differences in the EDA pattern between repeated acquisitions
during no pathological mood states and swing. Of note, seven
patients (i.e., Pz01, Pz02, Pz04, Pz07, Pz08, Pz09, and Pz10)
were acquired twice, whereas Pz03, Pz05, and Pz06 carried out
a single acquisition. Details are shown in Table I.
B. Methodology of Signal Processing
The EDA decomposition process consisted in three different
steps: a preprocessing phase, in which the signal was filtered to
reduce the noise, a deconvolution process in order to obtain the
phasic and tonic driver, and an optimization stage to improve the
estimation of the parameters of the impulse response function.
Authorized licensed use limited to: Universita degli Studi di Firenze. Downloaded on June 11,2021 at 10:45:08 UTC from IEEE Xplore. Restrictions apply.
GRECO et al.: ELECTRODERMAL ACTIVITY IN BIPOLAR PATIENTS DURING AFFECTIVE ELICITATION 1867
Fig. 2. Electrodermal acquisition and decomposition process. The EDA is filtered to reduce the noise and then decomposed in tonic and phasic componentsby
means of a deconvolution with an IRF called the Bateman function.
The decomposition process was performed by means of Ledalab
3.2.2. software package for MATLAB [30].
1) Preprocessing: In the preprocessing stage, the detection
of movement artifacts was carried out by visual inspection.
Artifact-free signals exclusively were taken into account for
further analysis. In order to limit the frequency bandwidth of
the EDA signal, it was filtered with a low-pass zero-phase for-
ward and reverse digital filter [31], [32] with a cutoff frequency
of 2 Hz, having Buttworth approximation.
2) EDA Deconvolution Analysis: EDA is produced by
changes in the skin conductivity as major effect of the sweat
glands activity. Specifically, sweat is released to the sweat duct,
passes to the stratum corneum, and finally is brought out of the
skin. Accordingly, the dynamics of the variation of concentra-
tion of sweat in the stratum corneum can be represented by a
two-compartment pharmacokinetic model in which the sweat
concentration is assumed to change only by diffusion [33], [34].
The first compartment represents the sweat duct and the second
compartment the stratum corneum. Due to the two compart-
ments being different in dimension (i.e., the stratum corneum is
much larger than the sweat duct), the diffusion can be consid-
ered as a one-way diffusion. Solving the two coupled first-order
differential equations of each compartment, the solution is the
impulse response function IRF(t) which is also known as the
Bateman function [35]:
IRF(t)=(e
−
t
τ
1
− e
−
t
τ
2
) · u(t). (1)
The Bateman function is characterized by a steep onset and a
slow recovery. The steepness of onset and recovery is deter-
mined by the time constants τ
1
and τ
2
.
EDA can be divided into tonic (SCL: Skin Conductance
Level) and phasic components (SCR: Skin Conductance Re-
sponse). The tonic electrodermal component represents the
baseline level of the signal, whereas the phasic component in-
dicates a direct response to a specific stimulus. However, there
are often phasic parts of EDA which cannot be related to any
specific stimulus, and hence, they are called spontaneous or non-
specific SCRs [20]. Sometimes, when the time interval between
two consecutive stimuli is shorter than the recovery period of
SCR, the stimuli responses in the SCR are overlapped. In this
case, the typical shape of the SCR is lost and this could be one
of the main issues for the extraction of the correct information
from the electrodermal signal. In order to overcome this issue,
the EDA signal process is modeled as a convolution process
between the SudoMotor nerve activity (SMNA), as part of the
sympathetic nervous system, and IRF [28] under the hypothesis
that EDA is controlled by SMNA resulting in a sequence of dis-
tinct impulses which regulate the eccrine sweat glands dynamics
(see Fig. 2).
Fig. 3. Example of EDA signal and related components during euthymic state,
extracted through deconvolutive method of analysis. On the top panel, the black
signal representing the raw EDA signal along with the DRIVER
tonic
(red) are
shown. On the lower panel, the DRIVER
phasic
is shown. Rest phases lasted for
the first 600 s. Afterwards, IAPS and TAT emotional stimulation is performed.
Formally, it is possible to write:
EDA = SMNA ⊗IRF (2)
where SMNA =(DRIVER
tonic
+ DRIVER
phasic
).In(2),
SMNA is unknown and it is evaluated by deconvolving the EDA
signal with the IRF. In order to decompose the obtained SMNA
signal into the DRIVER
tonic
and DRIVER
phasic
components,
several algorithmic steps have been processed. A smoothing
Gauss window of 200 ms is applied to SMNA, followed by a
peak detection algorithm in order to find the peaks over a thresh-
old of 0.2 μS. All the points below the threshold were interpo-
lated with a cubic spline fitting method giving the DRIVER
tonic
.
More details can be found in [28]. Finally, the DRIVER
phasic
component, instead, is computed by subtracting the previously
estimated DRIVER
tonic
from the SMNA (see Fig. 3), under the
hypothesis that the tonic activity is observed in the absence of
any phasic activity [20].
Of note, the DRIVER
phasic
signal should have a zero baseline
intermitted by distinct peaks overcoming the issue of having
overlapped SCRs.
3) Optimization: Starting from fixed values, the parameter
set of the IRF (i.e., τ
1
and τ
2
) was optimized according to criteria
evaluating the quality of the model, through the minimization of
a specific cost function given by the sum of the number of points
of the DRIVER
phasic
component that have a negative value and
the number of points above a predefined threshold (equal to
5% of the maximum of DRIVER
phasic
). This procedure aims at
Authorized licensed use limited to: Universita degli Studi di Firenze. Downloaded on June 11,2021 at 10:45:08 UTC from IEEE Xplore. Restrictions apply.
1868 IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, VOL. 18, NO. 6, NOVEMBER 2014
TABLE II
L
IST OF THE FEATURES EXTRACTED FROM THE EDA PHASIC AND TONIC
COMPONENTS
having a signal with a zero baseline peaks as distinguishable as
possible. More details can be again found in [28].
C. Feature Extraction
Features were extracted from the DRIVER
tonic
and
DRIVER
phasic
signals also studying the different effects of
the IAPS and TAT elicitation. Features extracted from the
DRIVER
phasic
signal were calculated into nonoverlapped time
windows of 5 s, according to the knowledge that SCRs arise
within 5 s after the stimulus onset [36], [37]. Nonoverlapped
time windows are justified by the fact that the deconvolution
algorithm misses overlapped responses lasting 4 s [28]. There-
fore, despite the fact that IAPS stimuli were presented each 2 s, a
sort of refractory period at least equal to 4 s is assumed for com-
putational reasons. Features extracted from the DRIVER
tonic
component, instead, were calculated within nonoverlapped time
windows of 20 s, the upper cutoff frequency of the tonic compo-
nent being about 0.05 Hz [38]. Afterwards, features belonging
either to IAPS or TAT elicitation were grouped accordingly. In
Table II, the features set is summarized along with the corre-
sponding description. Each feature was normalized by subtract-
ing its correspondent value at rest.
Statistical analysis: Both the IAPS-related features and the
TAT-related features extracted from several acquisitions were
compared by using statistical analysis. The statistical inference
analysis was performed by means of nonparametric tests due
to the nonGaussianity of the samples (p<0.05 given by the
Kolmogor–Smirnov test with null hypothesis of Gaussian dis-
tributed samples). For each of the seven subjects (i.e., Pz01,
Pz02, Pz04, Pz07, Pz08, Pz09, and Pz10) who performed the
experiment twice, an intrasubject statistical analysis was per-
formed. Each pair of acquisitions was compared by using a
Wilcoxon test for paired data [39]. Moreover, an inter-subject
analysis was performed in order to compare the acquisitions
associated to the same mood label. In this case, different mood
states (i.e., depression, mixed-state, and euthymia) were com-
pared by means of a Kruskal–Wallis test to evaluate whether
they statistically belonged to the same population. In case of
rejection of the null hypothesis, the Mann–Whitney test for un-
paired data [40] with a Bonferroni adjustment for every pair was
carried out.
III. E
XPERIMENTAL RESULTS
In this section, the experimental results performed on both
groups of healthy subjects and bipolar patients are shown in
(a)
(b)
Fig. 4. Pz01’s statistical analysis for IAPS elicitation. Results of Pz01’s area
under the curve (AUC) of (a) DRIVER
phasic
and (b) DRIVER
tonic
features.
detail. Further statistical analyses pointing out differences be-
tween the IAPS and TAT sessions, for each EDA feature and
for each acquisition, as well as results on intra and intersubject
evaluations are given below.
Of note, the time constants τ
1
and τ
2
were independently
estimated for each patient and for each healthy subject. Here,
we report the following statistics calculated among all the 17
EDA series gathered from the ten bipolar patients: Median{on
τ
1
=0.81,onτ
2
=2.49}, Median Absolute Deviation{on τ
1
=
0.16,onτ
2
=0.79},Min{ on τ
1
=0.49,onτ
2
=1.54}, and
Max{ on τ
1
=1.24,onτ
2
=3.83}.
A. Study on Bipolar Patients
A summary of the clinical evaluations of the patients recruited
for this study, expressed as mood label, is shown in Table I.
For each acquisition, we first performed a statistical analysis
to test the null hypothesis of having no significant difference be-
tween the two affective elicitation sessions (i.e., IAPS and TAT
sessions). As the samples were comprised of several values for
each IAPS and TAT session (each feature value was computed
within a sliding window), and a perfect temporal match between
each sample cannot be ensured, the Mann–Whitney tests were
used to compute the p-values. For each acquisition, we found
significant differences (p<0.03) for all of the considered EDA
features but the STD-Tonic.
1) IAPS Stimulation: Wilcoxon test for paired data was ap-
plied on patients with two acquisitions, i.e., Pz01, Pz02, Pz04,
Pz07, Pz08, Pz09, and Pz10. Statistical analysis results show
that all the phasic features resulted to be statistically different
for all subjects. Patients Pz02 and Pz04 showed a nonsignificant
tonic features set between the two acquisitions. More in detail,
patients Pz01, Pz07, Pz08, Pz09, and Pz10 exhibited significant
increase in the mean value, in the area under the curve and in the
maximum value of both DRIVER
phasic
and DRIVER
tonic
com-
ponents during second acquisition (see an example in Fig. 4).
Pz02 showed no statistical difference in tonic features, but an
Authorized licensed use limited to: Universita degli Studi di Firenze. Downloaded on June 11,2021 at 10:45:08 UTC from IEEE Xplore. Restrictions apply.
