Experimental Study and Evaluation of Paper-based Inkjet Electrodes

for ECG Signal Acquisition

Ana Priscila Alves

, Jo

ao Martins

, Hugo Pl

acido da Silva

, Andr

e Lourenc¸o

1,3

, Ana Fred

and Hugo Ferreira

Instituto de Telecomunicac¸

oes, Instituto Superior T

ecnico, Avenida Rovisco Pais, 1, 1049-001 Lisboa, Portugal

Instituto de Biof

ısica e Engenharia Biom

edica, Faculdade de Ci

encias da Universidade de Lisboa,

Alameda da Universidade, 1649-004 Lisbon, Portugal

Instituto Superior de Engenharia de Lisboa, Rua Conselheiro Em

ıdio Navarro, 1, 1959-007 Lisboa, Portugal

Keywords:

Electrodes, Paper, Inkjet, Electrocardiography, Device.

Abstract:

Applications involving biosignals, such as Electrocardiography (ECG), are becoming more pervasive with

the extension towards non-intrusive scenarios helping targeting ambulatory healthcare monitoring, emotion

assessment, among many others. In this study we introduce a new type of silver/silver chloride (Ag/AgCl)

electrodes based on a paper substrate and produced using an inkjet printing technique. This type of electrodes

can increase the potential applications of biosignal acquisition technologies for everyday life use, given that

there are several advantages, such as cost reduction and easier recycling, resultant from the approach explored

in our work. We performed a comparison study to assess the quality of this new electrode type, in which

ECG data was collected with three types of Ag/AgCl electrodes: i) gelled; ii) dry iii) paper-based inkjet

printed. We also compared the performance of each electrode when acquired using a professional-grade gold

standard device, and a low cost platform. Experimental results showed that data acquired using our proposed

inkjet printed electrode is highly correlated with data obtained through conventional electrodes. Moreover, the

electrodes are robust to high-end and low-end data acquisition devices.

1 INTRODUCTION

Pervasive healthcare applications are becoming an in-

valuable tool for regular and non-intrusive monitor-

ing. Biosignals play an important role in this kind

of applications since they give information about the

state of several vital organic tissues. Electrocardio-

graphic (ECG) signals are probably the most well-

known biosignals, and can be found in multiple ap-

plications in the medical and quality of life domains.

It is commonly used to assess the overall cardiac func-

tion, measure the rate and regularity of heartbeats,

and detect the presence of any pathology in the heart.

The classical acquisition methods used in clinical or

research studies typically recur to gelled silver/silver

chloride (Ag/AgCl) electrodes. Given that ECG data

acquisition has become more pervasive and inexpen-

sive, enabling an easy access to continuous monitor-

ing of the cardiac function, new and cheapest solu-

tions have been proposed, with more practical elec-

trodes and acquisition setups (Silva et al., 2011; Silva

et al., 2013).

Paper has several advantages for ECG data acqui-

sition in daily life scenarios; it enables: a) lower pro-

duction costs; b) easier recycling; and c) simpler pro-

duction, especially when considering the possibility

of inkjet printing. When compared to plastic sub-

strates such as polyethylene terephtalate (PET, ≈ 2

cent dm

−2

) and polymide (PI, ≈ 30 cent dm

−2

), pa-

per has signiﬁcantly lower production costs (≈ 0.1

cent dm

−2

). In addition to this, considering the active

disassembly design principles (Chiodo and Ijomah,

2012), paper is a good choice due to its environmen-

tally friendly characteristics. Recently, it has been

considered as a potential substrate for low-cost ﬂex-

ible electronics (Siegel et al., 2010; Leenen et al.,

2009), which motivated us to do research on the pos-

sibility of using paper-based electrodes for biosignals

acquisition. With such an approach and its ready

availability, the electrodes can even be produced by

the user himself or his caregivers.

The deposition of the conductive part of the elec-

trodes to the paper substrate can be made recurring

to photo-lithography, vacuum processes or printing

275

Alves A., Martins J., Plácido da Silva H., Lourenço A., Fred A. and Ferreira H..

Experimental Study and Evaluation of Paper-based Inkjet Electrodes for ECG Signal Acquisition.

DOI: 10.5220/0004720802750281

In Proceedings of the International Conference on Physiological Computing Systems (PhyCS-2014), pages 275-281

ISBN: 978-989-758-006-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

techniques. The use of printing techniques for fab-

ricating electronics has several advantages over labo-

ratory scale and subtractive batch processes (Tobj

ork

and

Osterbacka, 2011); printing is fast, low-cost, and

widely used. In particular digital inkjet printing,

which has been used as a research tool, is facilitating

initial explorations of various aspects of printed elec-

tronics targeting the consumer market (Singh et al.,

2010). The focus of this work was to explore the po-

tential use of paper-based inkjet printed electrodes for

ECG signal acquisition.

The most commonly used type of electrode is the

gelled Ag/AgCl electrode; however, to make an ac-

quisition setup more convenient for everyday use ap-

plications, other alternatives are emerging. Previous

work from our group has started to explore the use

of dry Ag/AgCl electrodes (Silva et al., 2011), which

usually leads to signals with lower signal-to-noise ra-

tio, although still suitable for monitoring or other non-

intrusive applications. Thus, to study the characteris-

tics of the paper-based inkjet printed electrodes, we

perform a comparative study against the most com-

mon alternatives: i) gelled; ii) dry.

The remainder of the paper is organized as fol-

lows: in Section 2 we describe the proposed elec-

trodes, focusing on their production and main char-

acteristics; Sections 3 and 4 present the methodology

applied in the comparison of the different electrode

types and their quantitative evaluation; and ﬁnally, in

Sections 5 and 6 we provide a summary of the exper-

imental results and outline the main conclusions.

2 PAPER-BASED INKJET

PRINTED ELECTRODES

The possibility of printing materials using inkjet tech-

nology brought several advantages to the conven-

tional manufacturing procedures used, such as photo-

lithography, transfer printing, among others. Compar-

ing with those standard techniques for patterning thin

ﬁlms with high precision, some differences stand out.

The appeal of inkjet technology lies in the fact that

it is based on contactless deposition, which implies a

lesser risk of contaminating the material, it is a mask-

less approach that makes an intuitive procedure, and

it is an additive procedure, i.e., it is possible to print

over a previous printed pattern (Singh et al., 2010).

Producing electrodes by inkjet printing enables

the use of thin and ﬂexible substrates that may also be

biocompatible, examples of which are polydimethyl-

siloxane (PDMS) or biocellulose. On the other hand,

low-cost paper-like substrates such as photo paper can

be used as an alternative substrate and several conduc-

tive inks can already be used, such as silver, gold or

conductive polymer (Calvert, 2001)).

We fabricated the electrodes using photo paper as

substrate, due to its ﬂexibility, availability, reduced

thickness (230 µm) and easy maneuverability. To

create the conductive part of the electrode we used

a commercial printable silver ink from SunTronic,

which is composed of silver nanoparticles and has

been shown to provide good electrical conductivity

for electronic applications.

The electrodes devised in the scope of our work

were designed as a ﬂat rectangle shape, with dimen-

sions of 8 cm length, 3 cm width and approximately

1 µm thick. Each electrode has a total of 24 cm

area in contact with the skin. The electrodes were ﬁrst

printed with four silver layers and aftwerwards sub-

jected to heat treatment during 20 minutes at a tem-

perature of 85

◦

C. With this heat treatment, we ob-

tained a silver resistivity of 1.68 ×10

−6

Ω.m .

The second step of the fabrication process was

to produce a layer which enables the transduction

of ionic concentrations measured by electrodes into

electrical potentials. At the skin-electrode interface,

the ionic signal (Cl

ion transports the charge) is trans-

formed into an electric signal. Likewise, in common

silver electrodes this layer is typically made of AgCl

(Clark et al., 2009). The formation of this layer was

achieved by adding Cl

ions, enabling a reaction be-

tween Ag and Cl to produce AgCl. However, due to

the thin layer of silver and the fragility of the photo

paper, the amount and the manner of introducing Cl

ions is important. This process was optimized by us-

ing commercial bleach deposited by an airbrush at a

distance of approximately 30 cm.

The third step in the production of these electrodes

was focused on ensuring a good, long lasting, and

practical contact between the electrodes and the ac-

quisition hardware. To facilitate the connection of ca-

bles and make the electrodes practical for regular use,

we use a metal stud and conductive snap. The snaps

were placed in the back of the printed surface and the

communication to the front was made through a hole

ﬁlled with a conductive silver paste from Agar Scien-

tiﬁc. We estimated that each electrode produced with

the procedure described would cost, approximately,

0.03e.

Figure 1: Electrode leads placement.

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

276

3 METHODOLOGY

We benchmarked the performance of our paper-based

inkjet printed electrodes for ECG data acquisition,

comparing them both to standard pre-gelled Ag/AgCl

electrodes, and to the dry electrodes approach that we

have been recently following (Silva et al., 2011). Ref-

erence data was collected using a BIOPAC biosignal

acquisition unit, which has seen extensive use in the

research domain and is considered to be a gold stan-

dard in biomedical research. However, this system

has restricted operations and experimenting new cus-

tomized solutions can damage the device. As such,

we have used a BITalino acquisition system (Alves

et al., 2013; Guerreiro et al., 2013), which give us a

higher control over the system to try different experi-

mental setups.

This work is aligned with our research towards

off-the-person ECG sensing (Silva et al., 2013), rea-

son for which the ECG signals were acquired in the

palmar region of the left and right hands, as illus-

trated in Figure 1. The electrodes used for data acqui-

sition with the BIOPAC were always the pre-gelled

Ag/AgCl, while with the BITalino we tested the pre-

viously mentioned 3 types of electrodes.

We devised our comparative study in two objec-

tives:

1. comparison of the BITalino performance with a

gold standard acquisition system, the BIOPAC;

2. comparison of electrodes for ECG acquisition.

The BITalino acquisition device adopts the 2-

electrode approach with virtual ground, while the

BIOPAC system is designed to collect data with the

ground electrode. In order to inquire the BIOPAC

performance after removing the ground electrode, we

performed two experiments, with and without the

ground electrode. To evaluate the performance of the

dry, and paper-based inkjet electrodes in the ECG ac-

quisition, we did 2 experiments in which we com-

pared them with the pre-gelled ones. The experiments

are summarized in Table 1.

Table 1: Summary of the experiments.

Experiment

BIOPAC

BITalino

type GND

1 Gel Yes Gel

2 Gel No Gel

3 Gel No Dry

4 Gel No Paper

Each experiment consisted of a 30 seconds record-

ing performed simultaneously with the BIOPAC and

the BITalino; we used a sampling rate of 1000 Hz in

both devices and a 12-bit resolution for the BIOPAC,

whereas the BITalino has a 10-bit resolution. The

BIOPAC raw data was reduced to 10 bits, to be at

the same resolution as the BITalino signals. We

have collected raw ECG data from 20 subjects in a

static standing position, with the electrodes applied

as shown in Figure 1

The data obtained by each device was pre-

processed in three main steps, as represented in Fig-

ure 2. Taking the raw data as input, the baseline

wander was corrected through a two-stage median

ﬁlter, as proposed by (De Chazal et al., 2004), and

the signals were ﬁltered using a Finite Impulse Re-

sponse (FIR) bandpass ﬁlter with a Hamming window

of 300 ms, and cutoff frequencies of 5 − 20Hz. The

ﬁltered signals were normalized to their maximum

and minimum amplitudes, where the original signal

is subtracted of its mean, and divided by its standard

deviation. To prevent any possible electrical interfer-

ence between the devices prone to bias the results and

resulting from a hard wired connection between both

devices, we chose to do the synchronization using the

Figure 2: Block diagram of the pre-processing steps we have performed, to compare the signals acquired from both devices.

The curves plotted in black were acquired using the BIOPAC while the blue ones with the BITalino.

ExperimentalStudyandEvaluationofPaper-basedInkjetElectrodesforECGSignalAcquisition

277

RR time intervals. Given that the comparison of the

ECG data obtained from two independent systems can

only be correctly performed for data expressed in the

same time base, our synchronization method consists

on the following steps:

1. Detection of the QRS complex in each indepen-

dent signal, using the method proposed by (En-

gelse and Zeelenberg, 1979)

2. Let RR

BIOPAC

= {RR

BIOPAC

, ..., RR

BIOPAC

} and

BITalino

= {RR

BITalino

, ..., RR

BITalino

} be a set

of RR time intervals for the n and m heartbeat

waveforms detected respectively in the BIOPAC

and BITalino ECG time series.

3. Construct a matching matrix, M, in which the en-

try M(i, j) corresponds to the absolute value of

the difference between the RR time intervals ex-

tracted from the BIOPAC and BITalino ECG time

series, that is:

M(i, j) = |RR

BITalino

− RR

BIOPAC

| (1)

4. Let #M be the number of items where M(i, j) ≤

5. If #M > Sync

, the synchronization is complete.

Otherwise, go to next step.

6. Consider RR

BITalino

(k) =

{RR

BITalino

, ..., RR

BITalino

}. Repeat steps

3 and 4 for each value of k ∈ {1, ..., m}

and compute each value of M(i, j) =

|RR

BITalino

(k) − RR

BIOPAC

7. Find the k value where #M is higher

8. Synchronize the signals by applying a delay of k

samples to the BITalino signal.

The acquisition was always initiated ﬁrst with the

BITalino, so it has the higher time series. We deﬁned

2 thresholds in the synchronization method, Sync

and RR

. The Sync

value applied was 20, since it is

approximately the minimum number of heartbeats ex-

pected in a 30 seconds ECG signal. The RR

thresh-

old represents the minimum difference of RR time in-

tervals, from different acquisitions, where the R peaks

are considered to match in the time domain. Since

the acquisitions were performed by two different sys-

tems, it is expected a small deviation between the in-

stants where the same R peaks occur. Therefore, we

considered that 5 ms is the maximum value where the

R peaks are considered to occur in the same instant.

Finally, the individual heartbeat waveforms were seg-

mented and scaled between 0 and 1; we consider the

heartbeat waveform to be the [−200; 400]ms interval

around the R peak instant.

4 EVALUATION METRICS

Two metrics were employed for numerical evaluation

purposes, namely the Signal-to-Noise Ratio (SNR)

computed from the data collected with both devices

for each of the 4 experiments, and the Root Mean

Square Error (RMSE) of the cosine distance, to as-

sess the morphological correlation between the heart-

beat waveforms obtained with the BIOPAC and the

BITalino, when using each type of electrodes. For

the SNR calculation, we considered the interest sig-

nal to be concentrated on the 5 − 20 Hz band of its

frequency spectrum, and the remainder as noise. For

each record we calculated the difference between the

SNR obtained from BITalino and BIOPAC acquisi-

tion.

Figure 3 illustrates an example of the frequency

spectrum of ECG data acquired in both devices, for

(a) BIOPAC (b) BITalino

Figure 3: Example of the ECG signal frequency spectrum for data collected with each acquisition device in one of the

recording sessions. The blue region shows the interest spectral band and the remainder the noise.

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

278

one of the test subjects in the experiment 1. The 50

Hz power line interference is visible in both signals;

however, since the BITalino ECG sensor has an ana-

log band pass ﬁlter from 0.5 to 40Hz, the higher fre-

quencies are almost eliminated, contrary to what hap-

pens with the BIOPAC.

For the cosine distance calculation, the synchro-

nized signals were segmented into individual heart-

beat waveforms, and the distance between a given

segment in the BIOPAC time series and the matching

segment in the BITalino time series was calculated.

The cosine distance, D

cos

, between the signals x and

y is given by Equation 2

cos

(x, y) = 1 −

∑

k=1

x[k]y[k]

∑

k=1

x[k]

∑

k=1

y[k]

, (2)

The reason why we have calculated the cosine dis-

tance for each heartbeat, instead of using the entire

signal, is due to the fact that we were only interested

in the ECG waveform shape, which is comprised in

the heartbeat region. To validate the similarity be-

tween the signals acquired from the two devices, we

compute the RMSE, as deﬁned in Equation 3

RMSE(x, y) =

∑

j=1



cos

(x, y)



(3)

5 EXPERIMENTAL RESULTS

The results obtained for each experiment in the 20

subjects are represented in Figure 4.

The box plots display the distribution of the differ-

ence between BITalino and BIOPAC SNR, for each

experiment, across all the subjects. The height of

the box plot indicates the degree of dispersion, the

band inside the box represents the median, and the

Figure 4: Boxplot of the difference between BITalino and

BIOPAC Signal-to-Noise Ratio for each experiment.

bottom and top of the box are the ﬁrst and third quar-

tiles. The smallest SNR difference between devices

was obtained in the experiment 1, where the median

value is lower and the degree of dispersion is reduced.

This was already expected since the presence of the

ground electrode in the BIOPAC device and the use

of gelled electrodes in both systems correspond to the

best case scenario in which the amount of captured

noise is minimal. The higher dispersion obtained was

in the experiment 4, due to higher noise presence in

the signals.

Table 2 summarizes the results obtained for the

signals collected using each device. In all the experi-

ments, the SNR of BITalino was higher than BIOPAC,

which was already expected due to the analogic ﬁlter-

ing occuring in the BITalino ECG sensor.

Table 2: Experimental Results from BIOPAC and BITalino

ECG signals acquisition, in the 4 experiments.

Experiment RMSE

SNR [dB] SNR [dB]

BITalino BIOPAC

1 0.0043 ± 0.0053 −1.02 ± 2.04 −2.04 ± 2.31

2 0.0042 ± 0.0039 −1.19 ± 1.84 −3.35 ± 2.45

3 0.0063 ± 0.0055 −1.62 ± 2.21 −3.69 ± 2.54

4 0.0042 ± 0.0043 −1.87 ± 2.14 −3.80 ± 2.66

The lowest value of SNR with the BITalino de-

vice was obtained in the experiment 4, when using

the paper electrodes, indicating a higher noise pres-

ence. In what concerns the morphological correla-

tion between waveforms, all the experiments have

shown a high similarity between the ECG signals ob-

tained from both devices. The signals acquired have

a good approximation to the well known prototypi-

cal ECG waveform, providing an easy identiﬁcation

of the characteristic P-QRS-T complexes. Figure 5

presents an overlay with all the individual heartbeat

waveforms collected in one of the recording sessions,

showing the median and standard deviation of all the

segments obtained from both devices in the four ex-

periments. As we can see, the waveform morphology

is maintained throughout the experiments and is virtu-

ally indistinguishable between devices and materials.

From the cosine distance results, we have calcu-

lated the Root-Mean-Square Error (RMSE), and the

results are described in Table 2. For all the experi-

ments, we veriﬁed very low RMSE values, indicat-

ing that the signals obtained from all three types of

electrodes retain much of the waveform morphology

when compared to the signals obtained with the gold

standard BIOPAC setup. An interesting ﬁnding is

that the inkjet printed electrodes shows a very good

performance when compared to the other electrodes,

with a RMSE of 0.0042, while with the dry elec-

ExperimentalStudyandEvaluationofPaper-basedInkjetElectrodesforECGSignalAcquisition

279

(a) Exp 1 (b) Exp 2

Figure 5: Segmented heartbeat waveforms from the BITal-

ino (blue) and the BIOPAC (grey); the solid wave represents

the mean, and dashed line the standard deviation.

trodes we obtained the worst results, with a RMSE of

0.0063. Although the signals obtained with the paper-

based electrodes present a lower signal-to-noise ra-

tio, the ECG morphology is maintained, which results

in a similar performance to that found for the case

in which standard clinical-grade pre-gelled Ag/AgCl

electrodes are used. Moreover, the signals acquired

with the BITalino device are highly correlated to those

obtained with the BIOPAC, actually exhibiting lower

noise levels in raw ECG signals.

6 CONCLUSIONS

In this paper we have proposed and evaluated paper-

based inkjet printed electrodes for ECG data acquisi-

tion. We presented the fabrication steps, and bench-

marked our electrodes against standard clinical-grade

pre-gelled Ag/AgCl electrodes, and dry electrodes.

Data acquisition was performed using a BIOPAC sys-

tem, considered to be a gold standard within the

biosignal research community, although due to the

fact that it is a closed system, we have also supported

our analysis on the BITalino, a physiological comput-

ing platform ﬁrst introduced by our team.

The collected data was evaluated using the Signal-

to-Noise Ratio (SNR), and a morphological wave-

form correlation index based on the Root Mean

Square Error (RMSE). Experimental results have

shown that the proposed approach explored in this

work achieves comparable performance when com-

pared with a reference sensor. Our evaluation

has revealed that the heartbeat waveforms measured

through the proposed approach are nearly identical to

those obtained with the gold standard equipment.

This approach opens new possibilities in the ﬁeld

of biosignals, enabling people (e.g. patients and/or

caregivers) to have easier access to consumables in

continuous ambulatory monitoring scenarios. We be-

lieve our approach to have the threefold advantage

of reducing production costs, being easier to recycle,

and being more accessible when compared to conven-

tional approaches.

ACKNOWLEDGEMENTS

This work was partially funded by Fundac¸

ao para

a Ci

encia e Tecnologia (FCT) under the grants

PTDC/EEI-SII/2312/2012, SFRH/BD/65248/2009

and SFRH/PROTEC/49512/2009, whose support the

authors gratefully acknowledge.

REFERENCES

Alves, A. P., Silva, H., Lourenc¸o, A., and Fred, A. (2013).

BITalino: A Biosignal Acquisition System based on

Arduino. In Proceeding of the 6th Conference on

Biomedical Electronics and Devices (BIODEVICES).

Calvert, P. (2001). Inkjet Printing for Materials and De-

vices. Chemistry of Materials, 13(10):3299–3305.

Chiodo, J. and Ijomah, W. (2012). Use of active disassem-

bly technology to improve remanufacturing produc-

tivity: automotive application. International Journal

of Computer Integrated Manufacturing, 0(0):1–11.

Clark, J. W., Neuman, M. R., Olson, W. H., Peura, R. A.,

and Primiano, F. P. (2009). Medical instrumentation:

application and design. John Wiley & Sons, inc,

fourth edition.

De Chazal, P., O’Dwyer, M., and Reilly, R. (2004). Auto-

matic classiﬁcation of heartbeats using ECG morphol-

ogy and heartbeat interval features. IEEE Transac-

tions on Biomedical Engineering, 51(7):1196–1206.

Engelse, W. A. H. and Zeelenberg, C. (1979). A single scan

algorithm for QRS-detection and feature extraction.

Computers in Cardiology, 6:37–42.

Guerreiro, J., Silva, H., Lourenc¸o, A., Martins, R., and Fred,

A. (2013). BITalino:A Multimodal Platform for Phys-

iological Computing. In Proceeding of the 10th In-

ternational Conference on Informatics in Control, Au-

tomation and Robotics (ICINCO).

Leenen, M. A. M., Arning, V., Thiem, H., Steiger, J.,

and Anselmann, R. (2009). Printable electronics:

ﬂexibility for the future. physica status solidi (a),

206(4):588–597.

Siegel, A. C., Phillips, S. T., Dickey, M. D., Lu, N., Suo, Z.,

and Whitesides, G. M. (2010). Foldable printed cir-

cuit boards on paper substrates. Advanced Functional

Materials, 20(1):28–35.

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

280

Silva, H., Carreiras, C., Lourenc¸o, A., and Fred, A. L. N.

(2013). Off-the-person electrocardiography. In In-

ternational Congress on Cardiovascular Technologies

(CARDIOTECHNIX).

Silva, H., Lourenc¸o, A., Lourenc¸o, R., Leite, P., Coutinho,

D., and Fred, A. (2011). Study and evaluation of a sin-

gle differential sensor design based on electro-textile

electrodes for ECG biometrics applications. In Pro-

ceedings of the IEEE Sensors Conference, pages 1764

– 1767.

Singh, M., Haverinen, H. M., Dhagat, P., and Jabbour, G. E.

(2010). Inkjet printing-process and its applications.

Advanced Materials, 22(6):673–685.

Tobj

ork, D. and

Osterbacka, R. (2011). Paper electronics.

Advanced Materials, 23(17):1935–1961.

ExperimentalStudyandEvaluationofPaper-basedInkjetElectrodesforECGSignalAcquisition

281