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.2019.2913913, IEEE
Transactions on Biomedical Engineering
DESIGN AND EVALUATION OF A DIAPHRAGM FOR ELECTROCARDIOGRAPHY IN ELECTRONIC STETHOSCOPES
Design and Evaluation of a Diaphragm for
Electrocardiography in Electronic Stethoscopes
Mónica Martins, Pedro Gomes, Cristina Oliveira
Miguel Coimbra, IEEE Senior Member and Hugo Plácido da Silva, IEEE Senior Member
Abstract—Combining Phonocardiography (PCG) and Electro-
cardiography (ECG) data has been recognized within the state-
of-art as of added value for enhanced cardiovascular assessment.
However, multiple aspects of ECG data acquisition in a stetho-
scope form factor remain unstudied, and existing devices typically
enforce a substantial change into routine clinical auscultation
procedures, with predictably low technology acceptance. As
such, in this paper we present a novel approach to ECG data
acquisition throughout the five main cardiac auscultation points,
and that intends to be incorporated in a commonly used electronic
stethoscope. Therefore, it enables analysis and acquisition of
both PCG and ECG signals in a single pass. We describe
the development, experimental evaluation and comparison of
the ECG signals obtained using our proposed approach and a
gold standard medical device, through metrics that allow the
evaluation of morphological similarities. Results point to a high
correlation between the two evaluated setups, thus supporting
the idea of meaningfully collecting ECG data along medical
auscultation points with the proposed form factor. Moreover,
this work has led us to conclude that for the studied population,
signals acquired on focuses F1, F2, and F3 are usually highly
correlated with leads V1 and V2 of the standard ECG medical
recording procedure.
Index Terms—Cardiovascular diseases, Electrocardiogram,
Phonocardiogram, Electronic stethoscope, Systems integration.
I. I NTRODUCTION
ccording to the World Health Organization (WHO),
A cardiovascular diseases (CVD) continue to be the leading
cause of death in the world [1]. Besides general prevention
measures that should be accounted for, regular heart monitor-
ing is crucial both in chronic patients (to monitor the progress
of their condition) and healthy subjects (for early detection of
potential problems).
Auscultation plays a major role in CVD’s monitoring. Given
that auscultation relies on the human ear, which may not
be sufficient for an accurate evaluation of the patient [2],
phonocardiography appeared as a complement to this exam.
Using electronic stethoscopes [3], a detailed recording of
the heart sounds can be performed, allowing even automated
classification [4]. Regardless, one of the more established
techniques for CVD diagnostics is electrocardiography, by
means of which a multi-lead assessment of the electrical
activity of the heart is performed. While the former performs
This work is funded by FCT/MEC through national funds
and when applicable co-funded by FEDER PT2020 partnership
agreement under the project UID/EEA/50008/2013 SmartHeart.
All authors are with the IT - Instituto de Telecomunicações,
Portugal (e-mail: monica.cm.martins@gmail.com; pedro.tiago@lx.it.pt;
anacristina.oliveira90@gmail.com; {miguel.coimbra, hugo.silva}@lx.it.pt)
0018-9294 (c) 2018 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.
1
a mechanical assessment, the later performs an electrical
assessment, thus being complementary.
With auscultation being a routine exam, performed in most
typical appointments with a health professional, there are
multiple reasons that make the combination of phonocardio-
gram (PCG) and electrocardiogram (ECG) signals a natural
step. Although the combination of both modalities has been
explored in the state-of-the-art, in this paper we present a
design and evaluation of a diaphragm for ECG data acquisition
in an electronic stethoscope, addressing a number of practical
aspects associated with this technique (as further detailed in
Section III).
The remainder of the paper is organized as follows: Section
II presents the motivation for this work, together with a
brief overview of the PCG and ECG; Section III provides a
review of the state-of-the-art in the field; Section IV details
the implementation of our device; Section V explains the
methodology used for experimental evaluation; Section VI
summarizes the results; finally, Section VII provides an outline
of the main findings.
II. BACKGROUND
A. Electrocardiogram (ECG)
An electrocardiogram is a standardized medical examination
that allows the evaluation of the heart condition, through the
study of its electrical activity. It is widely used due to the
high benefit/cost ratio, and also for being an easily deployable
noninvasive test. Its output consists of voltage variations as
a function of time, in result of the successive myocardium
depolarization and repolarization events.
As the heart is formed by thousands of cells, the depo-
larization process in each one will occur at different times.
Therefore, an ECG is a relative and cumulative magnitude of
all cells in depolarization. Regardless, an ECG can be obtained
with different configurations and electrode locations, being
that these will provide different perspectives of the electrical
activity of the heart. The prototypical ECG heartbeat waveform
is shown in Figure 1 (considering a healthy medical case), and
the typically studied leads are Lead I, Lead II, and Lead III,
these being limb leads; aVR (augmented vector right), aVF
(augmented vector foot), and aVL (augmented vector left),
the augmented leads; and V1 to V6, the precordial leads, as
shown in Figure 2 (right) [5].
An ECG analysis typically uses more than one lead, since
it provides different perspectives on the cardiac activity, also
increasing the probability of a good acquisition and a correct
diagnosis.
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.2019.2913913, IEEE
Transactions on Biomedical Engineering
DESIGN AND EVALUATION OF A DIAPHRAGM FOR ELECTROCARDIOGRAPHY IN ELECTRONIC STETHOSCOPES
Fig. 1. Cardiac depolarization process and prototypical waveforms produced
by the different components of the electrical system of the heart (extracted
from [5]).
Fig. 2. Main PCG cardiac auscultation points (left), and representation of the
ECG precordial leads location (right).
B. Phonocardiogram (PCG)
On the other hand, phonocardiography allows the represen-
tation of the heart sounds obtained when performing auscul-
tation. Its outcome, the phonocardiogram (PCG), records the
cardiac sounds resulting from the mechanical response to the
electrical activity of the heart [6], and is usually performed
across the five main points depicted in Figure 2 (left).
It is used as a complementary exam to auscultation [2],
since the latter is difficult to master [6]. The fact that it
does not allow the digital recording of any information, also
stands out as a major limitation in the auscultation process.
Furthermore, specialized healthcare professionals are required,
in order to perform a standard medical evaluation, hence the
importance of using electronic stethoscopes, which typically
combine auscultation and phonocardiography [3].
In order to allow a more complete medical evaluation
and circumvent the lack of diagnosis equipment in some
medical facilities, the PCG can be a compelling approach,
especially considering that it is an extension to the standard
auscultation procedures used in general medical practice. But
then again, this exam also has its drawbacks. The heart sound
has both high and low frequency signals, as well as low
amplitude, making it necessary for the stethoscope to have a
highly selective sensitivity. Also, the kind of data acquisition
made with a stethoscope is highly affected by external noise,
which can mostly interfere with the final diagnosis. Finally,
the interpretation itself is higly subjective, depending on the
healthcare professional and his/her experience, as well as
hearing ability [3].
0018-9294 (c) 2018 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.
2
C. Motivation
The combination of both phonocardiogram (PCG) and elec-
trocardiogram (ECG) signals aims to, on one hand, enhance
PCG heart sounds segmentation capabilities based on ECG
fiducials, but also to provide richer electromechanical under-
standing of the heart in any type of scenario, further improving
the detection of cardiac diseases. The idea is to create a way
to more easily detect a cardiopathy, as an extension of the
auscultation exams already performed in a general practice by
a healthcare professional.
It is also studied if, and how, rotations of the stethocope
on the patient’s chest affect the resulting ECG signal. The
importance of such practice relies on the fact that an elec-
trocardiogram is only taken under medical prescription, while
if performed as an extension of a widespread practice like
auscultation, it has the potential to be much more pervasive.
When combined, the PCG and ECG medical exams can also
improve the early detection and diagnostics of cardiopathies,
as already described in previous studies [7], [8], [9], [10].
III. S TATE - OF - THE - ART
Several devices that ally both exams have already emerged.
However, the majority of them have shown limitations, pre-
venting their standardization in medical practice. Examples
of such devices are the DUO from Eko or the CardioSleeve
from Rijuven, now commercially available [11], [12]. DUO
is a device that combines an electronic stethoscope and an
electrocardiograph (1-lead ECG) [13]. It can be used with the
Eko app, allowing the physician to see and review the patient’s
exam, as well as to share it with other professionals, for a
second opinion. This device provides real-time waveforms,
saving and tracking possibility, as well as live streaming for
use in telemedicine, and it can be used wirelessly (Bluetooth
is incorporated) to transmit data to smartphones, tablets or
desktops [11], [13].
CardioSleeve is another stethoscope that provides ECG,
digital auscultation and instant analysis [14]. It records four
auscultation locations, using dry electrodes, and records ECG
leads 1-3, transmiting the data over Bluetooth to a tablet or
smartphone. Some caveats of this device are its weight (too
heavy), and the mobile application used to see the medical
acquisition (not fully optimized and functional).
More work has been done in this area, giving rise to
results such as those found in [15] and [16], which focus
on developing healthcare devices that record ECG and PCG
signals. Some earlier work on PCG and ECG analysis was
developed, being able to create a digital phonocardiogram
that allows simultaneous ECG acquisition, combining both
sound and image, visualizing auscultation and enabling the
user to distinguish the location of heart sounds by ear [7]. In
a more theoretical way, some studies have been more focused
on the understanding of the cardiac cycle, by performing
simultaneous capture of signals from multi-site auscultation
with the recording of an ECG, and graphic display [8],
[9]. Also, a few projects have arised targeting the field of
telemedicine [10], [17].
In conclusion, several studies have found and proved that
the underpinnings of this work, combining ECG and PCG
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.2019.2913913, IEEE
Transactions on Biomedical Engineering
DESIGN AND EVALUATION OF A DIAPHRAGM FOR ELECTROCARDIOGRAPHY IN ELECTRONIC STETHOSCOPES 3

TABLE I
C USTOM DIAPHRAGMS EVALUATION CRITERIA .

Specification Classification of hearing ability

1 2 3 4 5
S1/S2
Nothing Sometimes Slightly Audible Pretty Well
Breathing
Noise Only Noise Mostly Noise A Bit of Both Mostly Sound Only Sound
Not Good Gives Some Allows Identification Allows Identification Allows Identification
General
Information After a While by a Professional by a Medical Student

Fig. 3. Line-up of all six tested diaphragm models.

TABLE II
C USTOM DIAPHRAGMS EVALUATION SUMMARY.

measurements, provides a more reliable and fast diagnosis [7], Specification Model
187 188 189 190 191 192
[8], [9], [10]. Regardless, existing devices are often cumber-
S1/S2 3.83(3) 4.167 4 3 3.667 2.667
some and disruptive to common practices in what concerns Breathing 1.83(3) 1.667 1.83(3) 1.167 1 1
their form factor when compared to standard auscultation Noise 3.33(3) 4 4 2.667 3 3
stethoscopes, and aspects such as ECG validity or influence External noise 1.83(3) 1 1.83(3) 1.667 1.167 3.83(3)
Noise vs. heart sounds 3.50 3.33(3) 3.167 3 3.50 2.33(3)
of the rotation of the stethoscope head are poorly studied.
General 3.83(3) 4 4 2.667 3.83(3) 2.167
Previous work by our team in the field of PCG has identified
the 3M Littmann 3200 electronic stethoscope [18], [19] as a
support tool with a good compromise between form factor us-
and a version with a solid filled center in the position where the
ability and technical specifications. However, this device does
acoustic sensor is mounted on the stethoscope, both produced
not have ECG data acquisition capabilities, the integration of
in thermoplastic polyurethane (TPU) (items 189-192 in Fig-
which has been done in the scope of this work.
ure 3), resembling the original silicone material of the original
Overall, in order to effectively capitalize on the routine
diaphragm (item 187 in Figure 3). In the version with a solid
practice that is auscultation using a conventional stethoscope
filled center, different thicknesses of the center were tested,
form factor (e.g. the 3M Littmann 3200) as a way of obtaining
considering the space needed to accommodate the sensor in
ECG data as a collateral to the procedure, several challenges
the diaphragm (item 191 in Figure 3) or just the electrodes
are posed. In this work we address such challenges in an end-
(item 190 in Figure 3). As a benchmark for the materials, we
to-end approach, namely by studying the diaphragm mem-
also tested versions with the exact geometry of the original
brane design to accomodate additional components (e.g. the
diaphragm printed in TPU (item 189 in Figure 3) and standard
electrodes), devising a sensor design suitable for ECG data
polylactic acid (PLA), a more rigid material commonly used
acquisition using the small footprint of the diaphragm, and
for desktop 3D printing (item 188 in Figure 3).
evaluating how the ECG signal acquired from the auscultation
points relates with the standard 12-lead ECG leads. We also The diaphragms were tested by three evaluators (two pro-
assess how the rotation of the stethoscope head with respect to fessionals and one amateur) who listened to several heart
the electrical axis of the heart affects the heartbeat waveform sounds recorded using a randomly chosen diaphragm, out of
morphology. the six already mentioned. This process was done without the
evaluators knowing what diaphragm they were using. A total
of 12 cases were studied, and multiple criteria were scored
IV. P ROPOSED A PPROACH
by the evaluators on a five-point scale. Table I describes the
A. Diaphragm criteria groups, as well as the underlying meaning of the
An important requirement for ECG data acquisition is the classification assigned to each one.
existence of an electrical interface between the sensor and the From the General evaluation values, it can be acknowledged
subject’s body, which requires direct contact with the skin and that the best classified diaphragms were the rigid PLA (trans-
can be affected by natural barriers (e.g. androgenic hair). This parent) model (188) and the flexible TPU most similar to
is not a problem on standard electronic stethoscopes, which the original model (189). Heart sounds are best heard, but
typically use silicone membranes for comfort, hygiene and the noise created by the device is amplified as well. When
providing adequate acoustic propagation capabilities. How- considering the Noise vs. Heart Sounds criteria, model 191
ever, given that PCG measures the acoustic propagation of performed better. For the rest of the diaphragms, they not only
the cardiac sounds, any barrier between the sensing element reduce the noise, but also muffle the sound. Likewise, one can
and the body surface will attenuate the signal of interest. As also verify that the 191 model showed a similar overall perfor-
such, significant attention has been given to the diaphragm mance comparing to the original version (187). Considering
design and evaluation. Several versions of the diaphragm were its classification and depth (which allowed the integration of
considered (Figure 3), all produced using standard desktop 3D some electronic components), the 191 model was chosen. To
printing, targeting the study different materials and geometries enable the interface between the ECG sensor and the subject’s
capable of accommodating some of the electronic components body, two protruding electrodes were integrated in the base of
directly in the diaphragm. the diaphragm (the two electrodes are at different potentials).
We tested three main groups of variants, namely a version Our final model, is not only true to the original diaphragm of
with a deep hollow cavity in the center (oddball hypothesis), the 3M Littmann 3200 electronic stethoscope, but also allowed

0018-9294 (c) 2018 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.2019.2913913, IEEE
DESIGN AND EVALUATION OF A DIAPHRAGM FOR ELECTROCARDIOGRAPHY IN ELECTRONIC STETHOSCOPES
TABLE III
E STIMATED GAINS FOR FOCUSES F1
Focuses Max. Peak Ĝ
F1 0.052 mV 17307.69
F2 0.0767 mV 11743.03
F3 0.1287 mV 6993.00
F4 0.1485 mV 6060.06
F5 0.1393 mV 6460.9
for the integration of the ECG electrodes without degradation
of the signals of interest.
B. Analog Front End
Our sensor design builds upon the BITalino (r)evolution
ECG sensor1 , which has been primarily dimensioned for the
standard ECG lead placement where the inter-electrode dis-
tance is considerable. In the scope of the work, the electrodes
are directly placed on the diaphragm footprint (inter-electrode
distance of approximately 2.5 cm), leading to a low signal
magnitude. ECG values from an Einthoven setup (e.g.) would
usually have an amplitude of approximately 0.11 mV, while
ECG values measured with the electrodes on the diaphragm
only reached around 0.04 mV.
As such, we studied the dynamics of the ECG collected at
the cardiac auscultation focus, as a way to estimate the gain,
targeting the modification of the analog front end of the sensor
with a sensitivity more adapted to the stethoscope diaphragm
setup.
VCC
Rmax × G =
2
VCC
Ĝ =
2 × Rmax
Based on experimental data, calculations were made in order
to estimate the most suitable gain of the analog front end
for each auscultation focus. F1 to F3 remark to the farthest
points from the heart, and presented low amplitude signals
with the standard sensor configuration, while F4 & F5 were
usually of much higher amplitude than the latter, given the
heart proximity.
The sensor gain was estimated taking into account that the
ECG signal is centred in V2cc , and that its amplitude must
be multiplied by a certain gain G, in order to achieve the
intended value (Equation 1). Table I summarises the result
of this analysis, showing the most appropriate gain for each
focus. The main conclusion is that, ideally, the sensor should
have an adjustable gain in order to have the best performance,
given the disparity of amplification gains required in-between
focuses.
In the scope of our work, however, we used a fixed gain
compatible with the dynamic range of the signals obtain across
focuses. Considering previous calculations, and to prevent
saturation of most restrictive signal (F4), 5600 was used as
a conservative gain value for the proposed setup.
1 http://bitalino.com/datasheets/REVOLUTION ECG Sensor Datasheet.pdf
0018-9294 (c) 2018 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.
Transactions on Biomedical Engineering
4
A variant of the ECG sensor was therefore implemented,
TO F5 through which electrocardiogram signal amplitudes compati-
ble with the adopted inter-electrode distance can be measured.
This is further confirmed by the experimental component of
our work (see Section VI). It is important to highlight that the
sensor was used in a virtual ground configuration, to minimize
the number of contact points with the body.
While standard circuits require a grounding lead for im-
proved common mode rejection and improved signal quality, in
our design we use a virtual ground circuit, and instrumentation
amplifiers with high input impedance and high common mode
rejection ratio (CMRR). This enables a suitable signal to be
obtained from the differential leads without the need for a
grounding lead [20], [21].
C. Data Acquisition Setup
Given that the analog-to-digital conversion and communica-
tion components are out of the scope of this work, for data ac-
quisition from the analog front end described in Section VI.B
we used a variant of the BITalino (r)evolution platform and
supporting tools. In particular, the micro controller (MCU),
power management (PWR) and Bluetooth (BT) hardware
blocks were used, together with the OpenSignals (r)evolution
software [26]. These enable sampling at up to 1 kHz with
10-bit resolution, and have been shown to have adequate
performance when compared with gold standard devices [20].
Overall, this configuration will be hereinafter referred to as
SmartHeart.
The ECG being an electrical signal in nature, the positioning
of the electrodes (in particular the rotation angle) of the
(1) stethoscope head with respect to the electrical axis of the heart
will influence the morphology of the resulting measured signal.
To study this effect, a second configuration for data acquisition
(2)
system was taken into account, in order to evaluate whether
(and how) the positioning angle during the auscultation exam
influenced the resulting signal. For such purpose, the analog
front end described in Section VI.B was used together with
a BITalino R-IoT device2 , due to the fact that it integrates
an onboard inertial measurement unit (IMU) and implements
the attitude estimation algorithm by Sebastian Madgwick [22].
This device enables data acquisition at up to 200 Hz with 12-
bit resolution, and will be hereinafter designated as R-IoT.
It is important to highlight that both the SmartHeart and the
R-IoT configurations have uniquely advantageous properties
for our work, hence the need for the two. On one hand
the SmartHeart configuration provides high time resolution
and suitable amplitude resolution, but it lacks information
regarding the rotation angle of the device. On the other
hand, the R-IoT configuration allows the measurement of the
diaphragms angle in space, with respect to the earth’s magnetic
pole, but it provides a much lower time resolution. Given that
the ECG data analysis is only dependent on the diaphragm, the
data acquisition setups were integrated in a stethoscope form
factor simulator, in which the diaphragm base integrated the
electrodes and worked as the interface with the subjects body,
2 http://bitalino.com/en/r-iot-kit
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.2019.2913913, IEEE
Transactions on Biomedical Engineering
DESIGN AND EVALUATION OF A DIAPHRAGM FOR ELECTROCARDIOGRAPHY IN ELECTRONIC STETHOSCOPES
Fig. 4. R-IoT stethoscope form factor simulator. Left: Top view of the device
showing the battery, acquisition and communication electronics inside; Center:
Side view of the device showing the grip used to apply the simulator to the
device; Right: Bottom view of the device showing the electrodes configuration.
with the interior of the diaphragm housing the electronics
(Figure 4).
Two simulators were created, one with the SmartHeart
hardware base and another with the R-IoT. Both have a
lithium-polymer (LiPo) battery, and a pair of gold alloy elec-
trodes, which are assembled on a thermoplastic polyurethane
(TPU) diaphragm (which, as described in Section VI.A,
has shown suitable performance when compared with the
default diaphragm). For our proposed approach, the typical
Ag/AgCl electrodes could not used, because their physical
characteristics and geometry do not allow proper moulding
and adaptation to the specific mechanical characteristics of the
3M Littmann 3200 stethoscope diaphragm form factor, without
interfering with the usability of the device and performance of
the PCG signal acquisition component [19]. Gold enabled us to
overcome these limitations, and has been shown to have good
performance when used for measuring biopotentials, including
electrocardiography signals [23], [24], meeting the standards
for our particular application, namely high conductivity, easy
to mould into the shapes found suitable for our proposed
approach, and possibility to obtain in a miniaturized size,
this being an important feature for the integration in the
diaphragm form factor. Furthermore, gold is considered a
reference material for some biosignals measurements.
V. E XPERIMENTAL M ETHODOLOGY
A. Data Acquisition on Subjects
Our proposed setup was tested against a medical-grade gold
standard, with the goal of evaluating its performance. Whilst
the standard 12 leads ECG and electrode placements used in
regular practice are fully characterized in literature, the setup
devised in the scope of this work has several properties that
make it significantly different from both the standard 12 lead
ECG and from other approaches found in the state of the art.
In particular, the electrode material and geometry, inter-
electrode spacing, and the lack of a ground/reference lead are
all non-standard approaches, motivating the study of the valid-
ity of the collected data. For the experimental data acquisition,
only healthy subjects were evaluated, under the guidance of
a cardiopneumologist, and the adopted gold standard device
was the MAC 800 [25], from General Electric (GE).
The overall setup consisted of a GE MAC 800, SmartHeart
and R-IoT stethoscope simulators. With the gold standard all
twelve leads were acquired, while with the simulators ECG
signals were collected in all five auscultation points, one at a
0018-9294 (c) 2018 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
time, with positioning done according to personal preference
of the experimenter. With the R-IoT, 90◦ angle variations
between each position, and for each focus, were considered;
in particular, 90◦ rotation clockwise: 0◦ ; vertical after another
90◦ rotation: -90; horizontal after a final 90◦ rotation: -180◦ .
A total of 15 volunteers (5 of which women) with an average
of 26 ± 4.46 year old were enrolled in the experiment, from
which 60 ECG signal traces were considered (10 for the GE
MAC 800, 10 for the SmartHeart simulator and 40 for the R-
IoT simulator, i.e. one per rotation). The recordings for some
of the subjects were discarded, due to setup conditions and
specific physiological and anatomical features that prevented
the attainment of a good quality acquisition to allow a fair
comparison between systems. Human subjects approval was
granted in accordance with the standard procedures followed
by our institutional review board.
B. Post-Processing
Filtering, R-peak detection within each ECG segment, and
heartbeat waveform segmentation were performed using the
BioSPPy toolbox3 [26]. Within this toolbox, the module
ecg.py contains a set of functions for feature extraction and
segmentation, which were used to support the post-processing
component of the work. In terms of filtering, we used a band-
pass finite impulse response (FIR) with order corresponding
to 30% of the sampling rate, which has been identified as a
suitable approach in previous work by our group. For R-peak
detection, the algorithm following the approach by [27] was
used.
For the heartbeat waveforms extraction (herein also desig-
nated templates), which are a segment of the signal in time,
corresponding to a heart beat cycle, a criteria that selects 0.2 s
before the R peak and 0.4 s after is used, taking into account
the known physiological time intervals for a heart rate at rest.
VI. R ESULTS
In terms of data analysis, we based our study on mor-
phological waveform analysis. For that purpose, the average
heartbeat waveform was obtained for each of the twelve
leads acquired with the gold standard, and for each of the
five auscultations points of both the SmartHeart and R-IoT
simulators. The goal was to determine which (if any), and how,
leads of the standard ECG have a morphological resemblance
with the data acquired with the non-standard procedures and
sensor placements followed in our approach. Two studies were
performed, namely analysis of the correlation with the standard
placement and the assessment of the effect of rotation in the
waveform morphology. A standard ECG waveform collected
in the main auscultation points is represented in Figure 5, to
provide a visual reference to the obtained morphologies.
A. Correlation with Standard Placement
After processing all data from the SmartHeart device,
values of the coefficient of determination (R2 ), for each of
the 12 leads, and for each focus, were used to assess the
3 https://github.com/PIA-Group/BioSPPy
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.2019.2913913, IEEE
Transactions on Biomedical Engineering
DESIGN AND EVALUATION OF A DIAPHRAGM FOR ELECTROCARDIOGRAPHY IN ELECTRONIC STETHOSCOPES 6

subjects. From the numerical results, F1 has lead V2 as the

most similar with R2 = 0.7607 ± 0.1401, as well as F2 and
F3 with R2 = 0.8646 ± 0.0504 and R2 = 0.8421 ± 0.0520,
respectively, F4 has lead V5 as the most similar with R2 =
0.7490 ± 0.0928, and F5 has lead aVF as most similar with
R2 = 0.7158 ± 0.2383. These results have been determined
based on the highest R2 values for each focus.
As shown by the numerical results, focuses F1 to F3 were
readily identified with leads V2. This could be explained by the
electrodes’ positioning factor involved. The first three focuses
are located above the heart, while focuses F4 and F5 are below
the heart, and closer to it. Similarly, leads V1, V2 and even
V3 are acquired at the middle of the chest, also above the
heart.

B. Evaluation of the Effects of Rotation

In addition to the SmartHeart prototype, R-IoT data was
also processed and analysed. The same set of metrics was
adopted: R2 calculation, this time for each of the 12 leads,
5 focuses and 4 rotations (vertical: 90◦ ; horizontal after 90◦
rotation clockwise: 0◦ ; vertical after another 90◦ rotation: -
Fig. 5. Filtered ECG signals acquired for focuses F1 to F5 in a randomly
90◦ ; horizontal after a final 90◦ rotation: -180◦ ). Figure 7
selected case study, normalized by the highest amplitude across focuses. illustrates the results for all case studies, for focus F5. A
morphological analysis was also carried, in order to predict
which lead resembles more to each position of this focus. This
morphological similarity between the heartbeat waveforms. type of reasoning was applied throughout all signals, and all
Multiple analysis were performed, in order to understand the focuses. In the end, this information can be cross linked with
behaviour and correlation between auscultation focuses and SmartHeart results for the same volunteer, in order to verify if
gold standard leads’ signals. R-peaks were detected, and the the same correlation between focus and lead occurs for both
heartbeat waveforms segmented as previously described, to devices.
get a representative description of the morphological pattern
in each focus.
We also present a plot with both MAC 800 and SmartHeart
ECG signals. This was done for each individual lead (I, II,
III, aVR, aVL, aVF, V1, V2, V3, V4, V5 & V6) and for each
focus (F1, F2, F3, F4 & F5). Mean waves were calculated, as
well as the SmartHeart’s data standard deviation (an example
of the signal waveforms for one focus analysis is depicted as
the dotted line in Figure 6).
The purpose of evaluating statistical metrics was the iden-
tification of the gold standard lead that presents the highest
morphological resemblance to each focus signal. In Figure 6,
for example, even visually, a resemblance between leads V1
& V2 (MAC 800) and the SmartHeart signal is apparent.
Numerical analysis was performed to study these relations
in depth. From the adopted metrics, it is possible to identify
a corresponding lead according to the computed values. To
proceed this study, both MAC 800 and SmartHeart heartbeat Fig. 7. Overlapped heartbeat waveforms for focus F5 with the simulator
waveforms were analyzed for morphological resemblances. placed at different rotations.
This analysis was crossed with the electrodes positioning, for
each lead and focus. Taking focus F5 as example, for this case study, numerical
Results point to similarities between focuses F1 to F5 and results mostly pointed to lead V6 as most similar, for the four
one or more leads, according to the numerical results obtained. rotations in general. One could come to a conclusion that in
With this, the intent was to look for similarities between focus F5, ECG signals are numerically more similar to lead V6
empirical data and numerical data. In order to eliminate for a 90◦ and -90◦ position with R2 = 0.9578 ± 0.0107 and
the potential effect of rotation as an additional variable, the R2 = 0.9160 ± 0.0132, respectively, to lead aVL for -180◦
stethoscope simulator was applied in the same position across with R2 = 0.9194 ± 0.0306, and lead V1 for a 0◦ rotation

0018-9294 (c) 2018 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.2019.2913913, IEEE
Transactions on Biomedical Engineering
DESIGN AND EVALUATION OF A DIAPHRAGM FOR ELECTROCARDIOGRAPHY IN ELECTRONIC STETHOSCOPES 7

Fig. 6. Representation of MAC 800 mean waveforms and their templates (in red) overlapped with the SmartHeart heartbeat waveforms and their templates,
as well as standard deviation for focus F3 (in blue).

angle with R2 = 0.4373 ± 0.3921, these being the leads with
the highest coefficients of correlation.
Another important aspect of this analysis is the influence
that rotation angles have in the polarity of the ECG signal. This
was the main purpose of using the R-IoT for different angle
variations throughout the five medical auscultation points. As
expected, inversion of the signal was readily observed when
the device changed positions.
C. Discussion
Our work uses an analog front end for bipolar differential
ECG sensing capable of being integrated in the footprint of a
stethoscope head, without degradation of the PCG signal. In
such setup, for a given focus, the rotation of the stethoscope
head (and consequently the differential leads) with respect
to the electrical axis of the heart will measure the cardiac
depolarization wavefronts at different angles.
As originally set forth by Einthoven, the positioning of
the electrodes with respect to the electrical axis of the heart
will capture different perspectives of its electrical conduction
system. For example, Lead I provides a RA-LA lateral view at
0◦ angle, Lead II provides a RA-LL inferior view at 60◦ angle,
and so on. In theory, the electrodes may be placed distally or
proximally to the source without affecting the recorded output,
due to the dynamics of progression of the depolarization
wavefronts (although the measured voltage potential difference
will depend on the distance between electrodes).
In our approach, the gain and filtering characteristics are
such that even with a 2.5 cm inter-electrode distance it is
possible to measure the depolarization wavefront locally and
without the need for an external earth ground. By rotating the
stethoscope head, the differential leads are placed in different
alignments with respect to the electrical axis of the heart.
Within the experimental component of the work we sought to
evaluate the relation between these different alignments and
those obtained with the standard lead placement. Our research
question for this should be seen as “Can these stethoscope
placements used for auscultation be used as opportunistic
captures of valid ECG lead surrogates?”.
VII. C ONCLUSION
Our work resulted in a novel form factor for ECG acqui-
sition capable of being integrated in a 3M Littmann 3200
electronic stethoscope, which has been benchmarked against a
gold standard device. Synchronization between ECG and PCG

0018-9294 (c) 2018 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.2019.2913913, IEEE
Transactions on Biomedical Engineering
DESIGN AND EVALUATION OF A DIAPHRAGM FOR ELECTROCARDIOGRAPHY IN ELECTRONIC STETHOSCOPES
signals, for segmentation and identification of heart sounds,
can be achieved, increasing the accuracy and reliability of the
medical diagnosis.
Results suggest that ECG signals acquired along the five
main cardiac auscultation points (focuses F1, F2, F3, F4 &
F5) are reliable for preliminary diagnosis of heart diseases.
From the evaluated population, a pattern is found for different
groups of focuses (F1, F2 & F3, and F4 & F5). Focuses more
distant from the heart, i.e. F1, F2 & F3, tend to correlate more
to leads V1 and V2 (also the most distant from the heart,
considering the 12-Lead ECG). On the other hand, focuses
directly around the heart, i.e. F4 & F5, are usually highly
correlated with leads V4, V5 and V6 (also closest to the heart
in a 12-Lead ECG).
The waveforms obtained in these focuses are different
from those of a 12-Lead ECG, due to the distinct electrode
placement, but, according to our results, can be considered
similar enough to assess irregularities of the heart’s electrical
activity. Further numerical analysis needs to be performed to
support this evidence in the scope of future work. Similarly, a
wider population, involving healthy and pathological subjects,
and more angle variations should be considered.
These experiments can evolve our study to assess heart-
beat waveform morphology when the evaluated subject has
a heart disease, and also acknowledge if those alterations are
significant and suffice for an accurate preliminary diagnosis. In
addition, both ECG and PCG could be acquired at the same
time using the developed prototype, as a way to test ECG-
based PCG segmentation.
ACKNOWLEDGMENT
This work is funded by FCT/MEC through national funds
and when applicable co-funded by FEDER PT2020 part-
nership agreement under the project UID/EEA/50008/2013
SmartHeart.
R EFERENCES
[1] WHO, “World Health Organization — Cardiovascular Diseases
(CVDs),” [Online] http://www.who.int/mediacentre/factsheets/fs317/en/,
last accessed on 26-04-2019.
[2] A. Moukadem et al., “Phonocardiogram signal processing module for
auto-diagnosis and telemedicine applications,” in eHealth and Remote
Monitoring, A. H. E. Hassani, Ed. IntechOpen, 2012.
[3] S. Leng et al., “The electronic stethoscope,” BioMedical Engineering
OnLine, vol. 14, no. 66, 2015.
[4] S. Meek and F. Morris, “ABC of clinical electrocardiography. introduc-
tion. I-Leads, rate, rhythm, and cardiac axis.” The BMJ, vol. 324, no.
7334, pp. 415–418, 2002.
[5] J. Malmivuo and R. Plonsey, Bioelectromagnetism: Principles and
Applications of Bioelectric and Biomagnetic Fields, 1st ed. Oxford
University Press, 1995.
[6] A. Castro et al., “Comparison between users of a new methodology for
heart sound auscultation,” in Proc. of the IEEE Annual Int’l Conf. of
the Engineering in Medicine and Biology Society (EMBC), 2016, pp.
5388–5391.
[7] C. Dan et al., “Playing and acquiring heart sounds and electrocardiogram
simultaneously based on LabVIEW,” in Proc. of the World Automation
Congress, 2008, pp. 1–4.
[8] M. Bravo-Zanoguera et al., “Simultaneous capture and display of
electrocardiogram and multi-site phonocardiogram,” in Proc. of the Pan
American Health Care Exchanges (PAHCE) Conf., 2009, pp. 26–28.
[9] J. Alvarez, “ECG-PCG simultáneo (SCG): Un viejo-nuevo método para
diagnóstico del corazón,” Universidad de Los Andes, Tech. Rep., August
2000.
0018-9294 (c) 2018 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.
8
[10] N. Al-Qazzaz et al., “Simulation recording of an ECG, PCG, and PPG
for feature extractions,” Al-Khwarizmi Engineering Journal, vol. 10,
no. 4, pp. 81–91, 2014.
[11] DUO, “Eko DUO stethoscope + EKG,” [Online]
https://ekodevices.com/duo/, last accessed on 26-04-2019.
[12] Rijuven, “CardioSleeve,” [Online] https://www.2mel.nl/project/rijuven-
cardiosleeve-signal-processing-algorithm-development/, last accessed on
26-04-2019.
[13] C. Landgraf et al., “Enhanced wireless communication for medical
devices,” [Online] https://www.google.com/patents/US9736625, last ac-
cessed on 26-04-2019.
[14] R. Kapoor, “Mobile front-end system for comprehensive cardiac diag-
nosis,” [Online] https://www.google.com/patents/US20150065814, last
accessed on 26-04-2019.
[15] S. Nayak et al., “Development of a simultaneous acquisition system for
ECG, PCG and body temperature signals,” in Proc. of the IEEE Annual
India Conf. (INDICON), 2015, pp. 1–6.
[16] S. Ummadisetty et al., “A portable system for simultaneous acquisition
of ECG and PCG in real time,” in Proc. of the IEEE Int’l Conf. on
Power, Control and Embedded Systems (ICPCES), 2014, pp. 1–4.
[17] M. Jimeno et al., “Wireless ECG and PCG portable telemedicine kit
for rural areas of colombia,” Investigacion e Innovación en Ingenierias,
vol. 2, no. 2, pp. 1–9, 2014.
[18] J. Oliveira et al., “A multi-spot exploration of the topological structures
of the reconstructed phase-space for the detection of cardiac murmurs,”
in Proc. of the IEEE Annual Int’l Conf. of the Engineering in Medicine
and Biology Society (EMBC), 2015, pp. 4194–4197.
[19] “3M Littmann manual: Electronic stethoscope model 3200,”
[Online] https://multimedia.3m.com/mws/media/594115O/3m-littmann-
electronic-stethoscope-model-3200-user-manual.pdf, last accessed on
26-04-2019.
[20] H. Silva et al., “Study and evaluation of a single differential sensor
design based on electro-textile electrodes for ECG biometrics applica-
tions,” in Proc. of the IEEE Sensors Int’l Conf., 2011, pp. 1764–1767.
[21] J. Guerreiro et al., “BITalino: A multimodal platform for physiological
computing,” in Proc. of the Int’l Conf. on Informatics in Control,
Automation and Robotics (ICINCO), 2013, pp. 500–506.
[22] S. Madgwick, “An efficient orientation filter for inertial and iner-
tial/magnetic sensor arrays,” x-io Technologies Limited, Tech. Rep.,
2010.
[23] M. Neuman, “Biopotential electrodes,” in The Biomedical Engineering
Handbook, 2nd ed., J. Bronzino, Ed. CRC Press, 2000, ch. 48.
[24] A. Koklu et al., “Rough gold electrodes for decreasing impedance at
the electrolyte/electrode interface,” Electrochimica Acta, vol. 205, pp.
215–225, 2016.
[25] “GE healtcare, MAC 800 resting ECG - technical specifications,”
[Online] http://www3.gehealthcare.in/en/products/categories/diagnostic-
ecg/resting/mac-800, last accessed on 26-04-2019.
[26] H. Silva et al., “BIT: Biosignal Igniter Toolkit,” Computer Methods and
Programs in Biomedicine, vol. 115, no. 1, pp. 20–32, 2014.
[27] P. Hamilton, “Open source ECG analysis,” in Proc. Computers in
Cardiology (CinC), 2002, pp. 101–104.