Cardio: An Edge-enabled Wearable ECG Vest for Ofﬁce Worker’s Heart

Condition Monitoring

Dimitrios Amaxilatis

1 a

, Athanasios Antoniou

and Ioannis Chatzigiannakis

2 b

SparkWorks ITC Ltd, Derbyshire DE11 8HS, U.K.

Department of Computer, Control, and Management Engineering “Antonio Ruberti”, Sapienza University of Rome,

00185 Rome, Italy

Keywords:

Electrocardiogram, Wearable Device, Feature Extraction.

Abstract:

Heart conditions are one of the most common health problems for people aged above 50 years, with the

percentage of people suffering from chronic heart diseases increasing year by year. These problems are more

common in modern western societies, where sedentary life and stressful lifestyles are the norms. People at

these ages are in the ﬁnal steps of their professional careers and need to balance the effect of their work on

their health while staying safe and productive to achieve the best future quality of life for themselves and their

families. In this work, we present a novel wearable ECG Vest that can help them monitor in real-time their

known heart conditions while they work, reducing stress and fear. Its operation is simple enough for the device

to be worn, as a normal jacket without the need to know where exactly to connect electrodes. Its operation is

also controlled with a single button without the need for any further conﬁguration.

1 INTRODUCTION

Health monitoring is an extremely active research

ﬁeld, especially after the recent COVID-19 pandemic

crisis that altered the lives of billions of people on

the planet. Visits to hospitals have been signiﬁcantly

reduced, due to the COVID-19 focused operation

of hospitals or even the people’s fear of coming in

contact with the virus. Physicians, clinics and gov-

ernments have been trying to ﬁnd alternative ways

to provide their patients and citizens with effective

telemedicine and home monitoring and home care so-

lutions, to reduce the stress in hospitals and medi-

cal personnel. Detecting potentially dangerous health

conditions quickly and with minimal effect on the

people’s routines and ensuring them that they are safe

during their everyday interactions is now of utmost

importance to everyone, as societies start to emerge

from prolonged lockdowns and people start to return

to their ”normal” daily routines.

SmartWork is an EU-funded research project that

intends to provide older workers (aged 55+) with ser-

vices that help them stay safe and more productive in

their professional life. Such services can offer them

https://orcid.org/0000-0001-9938-6211

https://orcid.org/0000-0001-8955-9270

and their loved ones reassurances that their health

conditions are not deteriorated by their work. Smart-

Work uses an Internet of Things (IoT) powered un-

obtrusive sensor network to understand and observe

the older worker’s health conditions, their work en-

vironments, and their productivity in real-time. This

information is then used to provide suggestions for

behaviors and habits that have negative effects on

the health conditions of workers and their work ef-

ﬁciency. SmartWork achieves all that by using robust

and proven in real world trials solutions for collecting,

storing and processing of all the needed information

using resource-constrained devices that are integrated

into the ﬁnal system (Chiang and Zhang, 2016).

Cardiovascular diseases, especially after the pan-

demic, are of extreme importance and require imme-

diate care, as the stress and anxiety cause by stay

at home guidelines can have serious negative effects.

Additionally diseases of the circulatory system cause

more than 1.68 million deaths every year (based on

data from 2016) (cde, ), with more than 10 million

patients that needed hospital care in 2018, depicting

the importance of early diagnosis and proper monitor-

ing. Having access to tools and methods to effectively

monitor heart conditions can give doctors the chance

to save thousands of lives every year (O’Connor et al.,

2015). The most common way to detect and iden-

420

Amaxilatis, D., Antoniou, A. and Chatzigiannakis, I.

Cardio: An Edge-enabled Wearable ECG Vest for Ofﬁce Worker’s Heart Condition Monitoring.

DOI: 10.5220/0010724700003063

In Proceedings of the 13th International Joint Conference on Computational Intelligence (IJCCI 2021), pages 420-426

ISBN: 978-989-758-534-0; ISSN: 2184-3236

tify these heart conditions is the use of an Electro-

cardiography monitoring (ECG) device. This device

can monitor the electrical activity of the heart, and

its electrical impulses generated by the polarization

and depolarization of cardiac tissue through properly

placed electrodes. ECG devices range in both sizes

and accuracy from stationary hospital-grade equip-

ment to miniature devices with lower accuracy but

higher portability and comfort for the users.

The Cardio ECG Vest is one of the SmartWork IoT

devices used to gather cardiological-health related in-

formation from ofﬁce workers. It is a miniature ECG

device designed as a wearable vest that can be used

during ofﬁce-related or other activities while collect-

ing ECG data in real-time and with minimal discom-

fort to the ofﬁce workers. It is capable of producing

high quality, hospital grade, 12-channel ECG record-

ing with additional features such as on-board data

processing, real-time notiﬁcations while being com-

fortable to wear and easy to use. This is a huge beneﬁt

in terms of accuracy and data quality, when compared

to other portable ECG devices that rely on a limited

number of leads with limited accuracy, such as the

AliveCor Heart Monitor

, the HeartCheck Pen

or the

Polar H10 Heart Rate Sensor

. The produced electro-

cardiogram (ECG) and the extracted analysis can be

used in decision-support systems to assist physicians

and cardiologists evaluate irregular heart rhythm, po-

tentially diagnose cardiac abnormalities, and predict

critical clinical states (da S. Luz et al., 2016).

A huge problem with existing solutions is the high

noise added to the ECG recording by nearby elec-

tronic devices (e.g., mobile phones, electrical wires,

appliances) or by the muscular movement of the

wearer that can severely affect the quality of the data

collected (Luo et al., 2017). The Cardio ECG Vest is

equipped with specially designed noise reduction cir-

cuits that can help reduce such interference to a min-

imum. It can also operate for long times, at least 8

hours (the full duration of a normal workday) without

the need for recharging. The data produced, are pro-

cessed to a very large degree on the device itself, elim-

inating the need to send unnecessary personal data to

a nearby smartphone or any cloud service.

Transmitting sensor data to cloud services for pro-

cessing and analysis creates several security issues

that need to be addressed as they are directly related to

the privacy of the users (Angeletti et al., 2018; Chatzi-

giannakis et al., 2011) Existing portable solutions are

cloud-centric: all personal data collected are stored

http://www.alivecor.com/

http://www.theheartcheck.com/

https://www.polar.com/en/products/accessories/H10 h

eart rate sensor

on the cloud and, in most cases, users have reduced

control over the data they produce, although certain

legislative actions have given users much more power

over their personal data (e.g., the EU General Data

Protection Regulation, GDPR (Commission, 2018)).

This cloud-focused architecture severely limits the

ability of the user to maintain control of personal data.

Now, more than ever, there is a need for privacy-

preserving applications where users remain always in

control of their sensitive data. (Angeletti et al., 2018;

Angeletti et al., 2017).

The goal of this paper is to present the usage of

this novel miniature ECG device and its wearable vest

design in SmartWork powered ofﬁce environments to

provide workers with reassurance over their chronic

health conditions. This is achieved using lightweight

algorithms for the analysis and interpretation of ECG

sensor data that can be executed in the embedded pro-

cessor of the wearable device.

In this context, the wearable device becomes re-

sponsible for the extraction of features from the col-

lected sensor data and providing actionable alerts

without any dependence on cloud services. It is an

evolution from the traditional cloud-based or ofﬂine

(Holter ECG devices) solutions following the highly

promoted wearable approach. By following this path,

we do not only achieve a much more energy and pro-

cessing power-efﬁcient solution but also manage to

respect users’ privacy.

The rest of the paper is structured as follows: In

Section 2 we present the miniature ECG device and

the Vest used. Section 3 describes some basic points

regarding the ECG data analysis and the steps per-

formed to extract characterizations for the worker’s

heartbeats. In Section 4 and 5 we showcase the ar-

chitecture of our system and the connectivity options

available for integrating the ECG device with a com-

panion smartphone application and the cloud. An

evaluation of the operation and the data analysis of the

Cardio ECG Vest is available in Section 6. Finally, in

Section 7 we present our conclusions and next steps.

2 THE ECG DEVICE

The wearable hardware device consists of a small

main board responsible for the core processing func-

tions, communication with a smart device and ten (10)

ECG sensor pads which are attached via a ribbon ca-

ble to the main board.

The main board features the ultra low power sys-

tem on chip (SoC) nRF52840, with Low Energy Blue-

tooth capabilities, and a number of peripheral mod-

ules, such as an Real-Time Clock, an Inertial Mea-

Cardio: An Edge-enabled Wearable ECG Vest for Ofﬁce Worker’s Heart Condition Monitoring

421

Figure 1: The main board of the Cardio ECG device.

surement Unit, an external Flash Memory module,

a LED, a multi-functional push button and a power

and battery charging circuit. The nRF52840 is built

around a 32-bit ARM Cortex M4F CPU which, as

the name indicates, has a dedicated hardware ﬂoating-

point unit (FPU). It also has a Bluetooth 5, IEEE

802.15.4-2006, 2.4 GHz transceiver, which is back-

wards compatible with the BLE 4 communication

protocol ensuring compatibility with a wide range of

smart devices in the market. A MCP79411 RTC mod-

ule provides timestamps for the ECG sampling ses-

sions and can also be used to set alarms for the de-

vice to wake up at speciﬁc time or intervals in or-

der to further reduce power consumption. The Iner-

tial Measurement Unit is the LSM6DSRXTR module

which features a 3D accelerometer and gyroscope and

can provide 16 bit samples at a conﬁgurable rate and

scale, based on the required sensitivity of the mea-

surements by the application. This module also has

a few embedded functions that can provide interrupts

to the main SoC for events such as signiﬁcant move-

ment, or taps, which can be utilized to detect inac-

tivity periods to put the device to sleep mode, or as

wake-up signals for the device. The (W25Q128JVSIQ

ﬂash memory chip is a NOR Flash memory with a 16

MBs storage capacity, which can store up to about 46

minutes of 12bit ECG samples from 8 analogue leads

sampled at a rate of 500Hz. The memory can also be

used for storing user proﬁles and application settings,

as well as pre-compiled machine learning models for

the classiﬁcation of the sampled data. The ECG sen-

sor pads have electrodes that provide the eight (8) ana-

log input signals to the internal 12bit ADC module of

the nRF52840 SoC. These eight signals are used to

produce the 12-lead ECG of the subject patient.

With the support of Nordic Semiconductor’s pre-

compiled library, SoftDevice S140, which is special-

Figure 2: The Cardio ECG Vest.

ized for the nRF52 SoC series, implementing the

Bluetooth protocol stack and providing an easy to use

API to conﬁgure Bluetooth connectivity, the applica-

tion ﬁrmware can construct a custom high level com-

munication protocol to facilitate the interactions with

a client Bluetooth enabled smart device. Our protocol

exposes a custom Bluetooth Service, and a number of

child BLE characteristics which are used to receive

commands, stream sampled or processed data and

notify for device status changes and special events.

Speciﬁcally, there are characteristics that handle is-

sued commands and their execution status. Such com-

mands include starting and stopping the sampling of

ECG and IMU data, setting and reading the RTC time,

resetting the device, formatting the external mem-

ory, and resetting to factory defaults. The ECG data

are streamed in a dedicated BLE characteristic, with

packet sizes that are set based on whether the device

operates in BLE4 or BLE5 mode. BLE5 mode al-

lows for a high throughput of up to 2Mbps, facili-

tating greater sampling rates. Similarly, the sampled

data from the IMU are streamed from a separate char-

acteristic and independently of the ECG sampling.

Another import BLE characteristic is used to notify

the calculated extracted features on the sampled data,

which can be used to either form training vectors for

our machine learning models or used for the classiﬁ-

cation of the active ECG data stream. Finally, aux-

iliary exposed characteristics include one providing

information on the ﬁrmware and conﬁguration of the

device, an indicator of the device’s charging status,

and one notifying periodically of the current battery

power level.

The whole system is powered by a high capacity

402030 3.7V 500mAh Lithium-ion battery cell. It also

supports battery charging via a micro-USB port and

has an onboard extension port for wireless charging.

SmartWork 2021 - 2nd International Workshop on Smart, Personalized and Age-Friendly Working Environments

422

3 ECG DATA PROCESSING

The analysis of an ECG recording can be performed

by many available methods in an embedded device.

Most of these techniques offer low accuracy lev-

els and result in a high number of alerts (Shah and

Rubin, 2007). Machine learning techniques (Saini

et al., 2013) and deep neural networks (DNN) (Ka-

plan Berkaya et al., 2018) have been tested and

showed that can achieve higher levels of accuracy in

diagnosing heart conditions using appropriate mod-

els and relying on high-performance computing in-

frastructures. To use them in low-power embedded

hardware with limitations in memory and processing

power, proper transformations are needed, using hard-

ware ((Du et al., 2017; Gokhale et al., 2017)) and

algorithmic solutions ((Luo et al., 2017; Qin et al.,

2020; for the Advancement of Medical Instrumenta-

tion, 2013)).

Our processing over the sampled data on the main

ECG device tries to follow the best outcomes from

the above approaches to get the best accuracy with

the lowest penalties on alert numbers and power con-

sumption. Our aim is to produce a dataset of extracted

features from the ECG signals recorded. These fea-

tures can the be used to provide descriptive vectors

on consecutive segments of an actively sampled ECG

that is cross-checked and validated by a separate pro-

cessing module on the client smart device or even on

the ECG device itself.

In the current implementation, we use a single 12

bit lead signals as the input to extract descriptive fea-

tures from. The feature extraction process takes place

over ten (10) second intervals of the input signal, and

employs a basic implementation of the Pan-Tompkins

algorithm (Pan and Tompkins, 1985) for QRS com-

plex detection. The extracted features include:

• the average R-R interval in milliseconds,

• the count of R peaks,

• the average R-amplitude,

• the average S-trough amplitude,

• the average R-S peak-to-peak amplitude and

• a history of the 5 most recently calculated R-R

intervals

The implementation uses variables to adapt for

various sample rates on the input ECG signal and

noisy segments of data. The whole feature vector can

be used to classify this 10-second segment in one of

3 generic groups, normal ECG recording, noisy ECG

recording, marked ECG recording. The marked ECG

recordings are those that display characteristics that

need to be reviewed by the medical practitioner in or-

der to understand if they are actually showing a phys-

iological condition that needs to be further checked.

4 DEVICE TO SMARTPHONE

CONNECTION

The ECG device itself has limited interaction with the

user. It is equipped with an LED light for simple bat-

tery charging and operation indications and a single

button used to power on the device from its deep sleep

state. Starting and stopping ECG recordings is done

using a companion application in the user’s smart-

phone, with which the device is communicating over

Bluetooth-LE.

While the phone is in the vicinity of the ECG de-

vice it can receive in real time both the full ECG

recording, IMU data, results from the on-board fea-

ture extraction, battery levels and notiﬁcations (ECG

or device related). Each one of these sources is a sep-

arate BLE characteristic that the smartphone can sub-

scribe to, in order to receive it form the ECG device

according to the BLE5 speciﬁcation. In order to ﬁt all

these data in the BLE notiﬁcation packets and avoid

hardware related restrictions, we attempted to limit as

much the number of notiﬁcations generated by the de-

vice, increasing the amount of data we send in batches

during each notiﬁcation. For this, each ECG data noti-

ﬁcation contains a total of 19 ECG samples with each

sample consisting of 8 12-bit values, while each IMU

data notiﬁcation contains 19 full samples with each

sample containing 6 values of 2 bytes. In this con-

ﬁguration, we manage to keep a constant ﬂow of data

from the ECG device to the smartphone with minimal

packets lost. For example, in a 3 minute recording

window, the number of packets lost less than 300 out

of a total of 90000 (around 0.003%). The same be-

haviour is observed with the IMU data packets, where

in the same period, around 50 packets are lost out of a

total of 18000 (again around 0.003%). Similar num-

bers are achieved even when only one of the notiﬁca-

tions is enabled, letting us believe that this behavior

is not related to the actual BLE medium but to the

handling of the packets on the smartphone side or the

generation of the packets in the ECG device’s side.

As this was observed using multiple phones with dif-

ferent speciﬁcations, the most probable cause is the

later.

The smartphone application itself does minimal

processing on the received data. It displays the re-

ceived ECG in a simple graph for the user to see that

everything is operating normally, the battery level of

the ECG device, and some of the features provided by

Cardio: An Edge-enabled Wearable ECG Vest for Ofﬁce Worker’s Heart Condition Monitoring

423

the device. The most important task is the storage of

the real-time ECG recording on the phone’s storage so

that it can then be shared with a medical practitioner

as needed. The ﬁle is stored in a properly encrypted

format to avoid any usage without the direct consent

of the worker.

5 SMARTPHONE TO CLOUD

CONNECTION

To provide the data collected on the phone to their

medical practitioner, users have two options. The ﬁrst

one is the most simple, by physically giving their

phone to their doctor to review the data collected.

This is no a very sophisticated approach but simpli-

ﬁes all the actions needed to safeguard personal data,

as no information leaves the user’s phone ever. The

second option is to properly share the recordings with

the medical practitioners via our cloud services. In

this ﬂow, the user uploads the recording to the Car-

dio service and shares it with the doctor’s account.

The doctor by accepting the recording receives the

keys to access it and sends a data access receipt to the

worker/patient. During the whole ﬂow, the data are

encrypted, and the worker has the option to revoke

access at any time.

This whole ﬂow is implemented using Amazon

Web Services

. In more detail we are using Amazon

Cognito & IAM to manage users and their permis-

sions, Amazon S3 to store and share recordings, AWS

SNS for sending notiﬁcations to users, and Amazon

Lambda to implement our serverless API for sharing

and getting access to ECG recordings and user infor-

mation (such as access receipts).

6 EVALUATION

To prove our system’s operation we performed a se-

ries of tests on the ECG devices using both ECG sim-

ulators and in-lab test patients. As soon as the sys-

tem’s operation is proven, we will extend our testing

during the SmartWork’s trials, where workers will use

the ECG vest to perform a number of recordings of

variable intervals to help us understand better (1) the

ease of use of the ECG Vest itself, (2) the ease of use

of the Cardio application interface, (3) the quality of

the data recorded on various body types, (4) the qual-

ity of the feature extraction algorithms with real world

data.

https://aws.amazon.com/

From the performed in-lab tests we performed, we

hereby present some results that depict the key fea-

tures of the operation of our system.

Figure 3: Recorde ECG pulses - L1 channel.

Fig. 3 and Fig. 4 show 5 consecutive recorded

heartbeats recorded from an in-lab test patient dur-

ing a larger recording. These ﬁgures give us a better

understanding of how heartbeats are depicted in dif-

ferent channels and how their PQRST characteristics

look like. Based on this view we are able to extract

the needed characteristics (as we described in Sec-

tion 3), like R-R interval, the R-amplitude or the R

peak count.

Figure 4: Recorde ECG pulses - L2 channel.

For each one of the features, we present here the

calculated values from the onboard algorithm. We

start with the R-R interval, which gives us an esti-

mate of the heart rate of the patient and its variance

during each interval. Fig. 5 shows the calculated R-R

interval from the ECG device (tumbling R-R) and the

calculated value using a sliding window (sliding R-R)

instead of a tumbling one as they are calculated on the

smartphone application. As we see, values using the

tumbling window are more smooth than the sliding

one, as they are not so much affected by noise in the

ECG signal.

SmartWork 2021 - 2nd International Workshop on Smart, Personalized and Age-Friendly Working Environments

424

Figure 5: R-R interval using tumbling and sliding windows

of 10 seconds.

Fig. 6 shows the number of R peaks detected from

our algorithm in each 10 second window the ECG de-

vice detects. These values do help us detect both the

number of heartbeats and the location of these heart-

beats inside the window, helping ﬁnd the exact start

and end locations of each heartbeat, information that

will help us in the future of our research. Using the

R counts we can calculate the heartrate of the wearer

( 10-sec-R-count ×6 = heartrate), so in our test, this

results in a heart rate of around 90 beats per minute.

Figure 6: R count in each 10 second tumbling window.

Finally, Fig 7 depicts the average R peak am-

plitude, S amplitude, and RS amplitude for each of

the tumbling windows. As we see, these values are

quite stable, as expected, during the whole recording

given that this recording is from a test patient with no

chronic heart condition diagnosed.

7 CONCLUSIONS

In this work, we showed how a wearable ECG mon-

itoring device is used in the context of SmartWork to

provide information and feedback on ofﬁce workers

with chronic heart conditions regarding the effects of

their works in their everyday lives and their diagnosed

Figure 7: R, S and RS amplitudes in each 10 second tum-

bling window.

conditions. We presented how the device operates in

order to generate such conclusions based on the ECG

input signals and how the resulting events are trans-

ferred to SmartWork to properly notify the worker’s

doctors. We showcased the resulting data from the

ECG recordings and the results of the analysis done

on the ECG device.

As our next steps, we plan to further increase the

analysis of the ECG traces on the device itself, provid-

ing more accurate detection and heartbeat character-

ization by identifying the possible problems in each

heartbeat, instead of a general alert characterization.

We also plan to test our system with real users, during

the SmartWork’s trials that are currently being exe-

cuted to gather as much user feedback regarding the

usability of our system and the quality of the whole

experience.

ACKNOWLEDGEMENTS

This work has been partially supported by the Smart-

Work project (GA 826343), EU H2020, SC1-DTH-

03-2018 - Adaptive smart working and living envi-

ronments supporting active and healthy ageing.

REFERENCES

Cardiovascular diseases statistics.

Angeletti, F., Chatzigiannakis, I., and Vitaletti, A. (2017).

The role of blockchain and iot in recruiting partic-

ipants for digital clinical trials. In 2017 25th In-

ternational Conference on Software, Telecommunica-

tions and Computer Networks (SoftCOM), pages 1–5.

IEEE.

Angeletti, F., Chatzigiannakis, I., and Vitaletti, A. (2018).

Towards an architecture to guarantee both data privacy

and utility in the ﬁrst phases of digital clinical trials.

Sensors, 18(12).

Cardio: An Edge-enabled Wearable ECG Vest for Ofﬁce Worker’s Heart Condition Monitoring

425

Chatzigiannakis, I., Pyrgelis, A., Spirakis, P. G., and Stama-

tiou, Y. C. (2011). Elliptic curve based zero knowl-

edge proofs and their applicability on resource con-

strained devices. In 2011 IEEE Eighth International

Conference on Mobile Ad-Hoc and Sensor Systems,

pages 715–720.

Chiang, M. and Zhang, T. (2016). Fog and iot: An overview

of research opportunities. IEEE Internet of things

journal, 3(6):854–864.

Commission, E. (2018). 2018 reform of EU data protection

rules. https://ec.europa.eu/commission/sites/beta-

political/ﬁles/data-protection-factsheet-

changes en.pdf.

da S. Luz, E. J., Schwartz, W. R., C

amara-Ch

avez, G., and

Menotti, D. (2016). Ecg-based heartbeat classiﬁcation

for arrhythmia detection: A survey. Computer Meth-

ods and Programs in Biomedicine, 127:144–164.

Du, L., Du, Y., Li, Y., Su, J., Kuan, Y.-C., Liu, C.-C., and

Chang, M.-C. F. (2017). A reconﬁgurable streaming

deep convolutional neural network accelerator for in-

ternet of things. IEEE Transactions on Circuits and

Systems I: Regular Papers, 65(1):198–208.

for the Advancement of Medical Instrumentation, A.

(2013). Testing and reporting performance re-

sults of cardiac rhythm and st segment measure-

ment algorithms. American National Standard 2013,

ANSI/AAMI EC57:2012.

Gokhale, V., Zaidy, A., Chang, A. X. M., and Culurciello,

E. (2017). Snowﬂake: An efﬁcient hardware acceler-

ator for convolutional neural networks. In 2017 IEEE

International Symposium on Circuits and Systems (IS-

CAS), pages 1–4. IEEE.

Kaplan Berkaya, S., Uysal, A. K., Sora Gunal, E., Ergin, S.,

Gunal, S., and Gulmezoglu, M. B. (2018). A survey

on ecg analysis. Biomedical Signal Processing and

Control, 43:216–235.

Luo, J.-H., Wu, J., and Lin, W. (2017). Thinet: A ﬁlter level

pruning method for deep neural network compression.

In Proceedings of the IEEE international conference

on computer vision, pages 5058–5066.

O’Connor, R. E., Al Ali, A. S., Brady, W. J., Ghaem-

maghami, C. A., Menon, V., Welsford, M., and Shus-

ter, M. (2015). Part 9: acute coronary syndromes:

2015 american heart association guidelines update for

cardiopulmonary resuscitation and emergency cardio-

vascular care. Circulation, 132(18 suppl 2):S483–

S500.

Pan, J. and Tompkins, W. J. (1985). A real-time qrs de-

tection algorithm. IEEE Transactions on Biomedical

Engineering, BME-32(3):230–236.

Qin, H., Gong, R., Liu, X., Bai, X., Song, J., and Sebe, N.

(2020). Binary neural networks: A survey. Pattern

Recognition, 105:107281.

Saini, I., Singh, D., and Khosla, A. (2013). Qrs detection

using k-nearest neighbor algorithm (knn) and evalua-

tion on standard ecg databases. Journal of Advanced

Research, 4(4):331 – 344.

Shah, A. P. and Rubin, S. A. (2007). Errors in the com-

puterized electrocardiogram interpretation of cardiac

rhythm. Journal of Electrocardiology, 40(5):385 –

390.

SmartWork 2021 - 2nd International Workshop on Smart, Personalized and Age-Friendly Working Environments

426