Wearable Sensor Node for Cardiac Ischemia Detection

Piotr Augustyniak

AGH University of Science and Technology, Krakow, Poland

Keywords: Wearable Sensor Network, Intelligent Sensors, Ischemia.

Abstract: Detection of cardiac ischemia based on early repolarization ECG markers has been widely recognized since

three decades. It is also employed as a safety marker in exercise testing and cardiac rehabilitation. It

assumes a standardized load applied to a patient in laboratory conditions, which diminishes patients’

responsiveness and lowers the medical outcome statistics. A remedy consists in detecting ischemia markers

during daily living activities in context of instantaneous physical load. Patients more willingly participate in

diagnostics or rehabilitation, but the procedure requires specialized sensors and processors of ECG

dedicated to use in domestic conditions. To this point we designed a small and lightweight autonomous two-

lead ECG sensor node including heart rate and ST-segment processing algorithm and secure Bluetooth Low

Energy connectivity to a supervising Wearable Sensor Network. In laboratory exercise tests with 50

ischemic patients the sensor issued alerts well coinciding with the output of a standard 12-lead system (with

2 fp and 1 fn cases). Moreover, in a two-lead setup the electrodes can be randomly applied to the skin in

location best corresponding the ischemic region. Low power-oriented design and low transmission duty

cycle result in continuous operation of the sensor for over a month with one coin battery.

1 INTRODUCTION

Recent developments of electronic and data

communication technologies enable to build smart

wearable devices for various medical purposes. This

allows to shift several diagnostic or therapeutic

procedures from hospitals to homes of patients

increasing their comfort and responsiveness. Such

intelligent sensors are organized in wireless sensor

networks (WSN) applied directly on the human body

(therefore called body sensor networks, BSN) or in

the living premises creating active ecosystems for

assisted living (e.g. Hao and Foster, 2008;

Figueiredo et al., 2010; Sahoo et al., 2017). This

idea has been applied far beyond the traditional

areas of medical interest and helps children, elderly,

danger-exposed professionals etc. to supervise their

performance.

Coronary arteries provided the oxygenated blood

to the heart muscle and cells of the conduction

system, thus deficient oxygenation may result in

arrhythmia, conduction blocks, ectopic beats or

acute infarct and partial necrosis of the heart muscle.

Two more facts should be noted here: (1) although

ground-truth ischemia markers require biochemical

blood analysis, its influence to the early

repolarization stage allow for reliable detection

based on ST segment in the electrocardiogram

(O’Gara et al., 2013) and (2) in most cases ischemia

is localized in heart muscle regions depending on the

topology of inefficient coronary vessels. Therefore

in a regular 12-lead exercise system, in one person

the ST-segment changes dominate in V1, V2 and in

another person in V5, V6 precordial leads depending

on localization of atherosclerosis (Menown et al.,

2000).

The need for reliable instrumentation for a daily

life condition exercise test or cardiac rehabilitation

motivated us to design an autonomous ischemia

sensor cooperating as a node of WSN with

accelerometers, posture and pressure sensors

estimating the physical load. The network is

supervised by a wearable server accompanying the

patient during the outdoor activity or by a stand-

alone hub embedded to the infrastructure of the

patient premise (e.g. dormitory). An example of

universal wearable biosignal recorder with a

Bluetooth-extensible WSN was presented in our

previous paper (Augustyniak, 2016), but the

ischemia detector is expected to work correctly with

any device (e.g. a smartphone) with complying

communication protocol.

Augustyniak, P.

Wearable Sensor Node for Cardiac Ischemia Detection.

DOI: 10.5220/0006544400970104

In Proceedings of the 11th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2018) - Volume 4: BIOSIGNALS, pages 97-104

ISBN: 978-989-758-279-0

Usually ischemia alerts are defined as ST-

segment amplitude crossing of arbitrarily given

thresholds, and therefore criticized for their

dependence on subject fatness. Due to the poorer

propagation of heart-related electrical phenomena to

the body surface, obese people in general show

lower ECG amplitudes and the ischemia signs are

underrepresented in amplitude measurements of ST-

segment. The other concern is the adjustment of the

ST-segment length with heart rate changes. In

exercise and daily living ECG recordings the heart

rate varies from 50 to 180 in healthy adults, so

various correction methods are implemented in the

interpretive software. Bazett formula (Bazett, 1920)

is one of most frequently referred, but some

cardiologist believe it is not equally suitable for all

subjects. Accordingly to the paradigm of

personalized medicine our ischemia sensor allows

for parametrization of several on-board signal

processing factors (e.g. threshold values or

correction method) by a supervising device through

the wireless communication channel.

2 IMPLEMENTATION DETAILS

The prototype wearable sensor node for ischemia

detection was designed and build for exploration of

new diagnostic purposes and usage scenarios it may

offer for cardiology. This paper presents the

prototype, while industrial research are continued on

the target circuitry, processing platform and form

factor. The prototype was designed as an

autonomous device identified by a supervising

device and working as a node of sensor network.

This device has not been presented in this paper, we

simply assume that it uses a compatible Bluetooth

Low Energy (BLE) protocol, is able to identify,

initialize and query the node and collects the

received data packets. The supervising device may

be wearable or stationary and may equally manage

other sensors within the network.

The design of prototype ischemia detector

needed will be presented in subsequent sections in

three main engineering aspects: the recording and

processing hardware, the processing algorithm and

the communication module.

2.1 Recording and Processing

Hardware

We based on experience gathered in our previous

project on wearable wireless heart rate monitor for

continuous long-term variability studies

(Augustyniak, 2011). The prototype wearable

monitoring device (fig. 1) is designed as lightweight

(7.3 g) and reduced in size (below 1 sq inch). With

these physical properties the device It is based on

the ARM7 family processor (Atmel

AT91SAM7X256), running at 18 MHz due to very

low power consumption (0.5 mW). It is powered by

a replaceable 3.0 V Li-Manganese Dioxide coin

battery (CR2032, 220 mAh, diameter 20 mm, weight

2.5 g) allowing for 940 hours (i.e. 39 days) of

continuous work. A single-channel low-power

(0.2 mW) analog front-end was custom-built and

verified for compliance with a typical ECG

measurement specification (full range ±12 mV,

noise 1V RMS, k

= 85V/V). Further reduction of

the footprint and the power consumption down to

0.2 W is possible with the use of custom-designed

analog or mixed-signal chip manufactured in 0.18 -

0.35 m CMOS technology (Liu et al., 2010; Tseng

et al., 2012; Teng et al., 2014). These options are

considered for further development of the prototype

to a commercial product. Besides the battery life, the

limitation of long time electrophysiological

recordings is the stability of skin-electrode contact,

which currently allows for two weeks continuous

data capture (Weder et al., 2015). Alternative

approaches use capacitive coupling for ECG

recording (Ueno et al, 2007), but for maintaining

low frequency throughput required for ST-segment

diagnostics large electrode size is required.

The ECG is acquired with the processor built-in

sigma/delta analog to digital converter at a sampling

frequency of 1000 sps and 16 bits resolution. a short

buffer of memory stores approximately 4 seconds of

signal in case of the algorithm requires reprocessing

the raw signal (e.g. search back procedure when no

heart beat detected for more than 2 s.). In a service

mode, the prototype enables a direct digital input

data stream and yields the raw digital ECG signal at

internal connector pins. Otherwise, the recorded

ECG is abandoned once the ST-segment calculations

are completed.

We tried to minimize the size of the prototype to

test its properties as a device worn at everyday living

activities. Twin printed circuit boards with analog

and digital circuitry include the battery and the BLE

antenna. One electrode snap is directly embedded

into the board and, besides the electric contact, used

to hold the device attached to the skin, while the

other is mounted at the end of a 12 cm wire (fig. 1).

BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing

Figure 1: Circuitry of a prototype wearable sensor for

ischemia detection.

2.2 Processing Algorithm

The ECG processing algorithm works in real time

regime. A regular delay between the QRS

occurrence and the availability of heart beat

description is 1.6 s and can occasionally be extended

to 3.6 s in case of search back for a missed heartbeat.

The ECG waveform needs a specific processing

to yield the annotated heart beats positions and ST-

segment description (fig. 2). Single-lead

electrocardiogram is sufficient to provide reliable

temporal markers and therough information about

the origin for each heartbeat. Since ectopic beats

propagate through alternative conduction pathway,

only sinoatrial beats are considered for detection of

ST-segment-based ischemia markers. The

processing algorithm consists of QRS detection,

morphology detection and ST-segment processing

procedures presented in subsequent sections.

Figure 2: Block diagram of on-board processing software

for ischemia detection.

2.2.1 Heartbeat Detection

Although the ECG is acquired with 1ksps, a

subsampled (decimated) version of 250 sps is used

for heart beat detection in real time. This procedure

first calculates a detection function with the use of

signal filtering and mathematical transformations

favouring the features common for the QRS

complex accordingly to a modified Pan-Tompkins

algorithm (Pan and Tompkins, 1985). Although long

filter buffers result in significant delay in QRS

detection, we prefer this classical approach for its

very low computational complexity.

Next, an adaptive threshold is applied to

determine the rough position of each QRS section.

In case no QRS is detected in more than 2 s (i.e. 500

samples) a search-back procedure is employed to

scan again the detection function with a lower

threshold value. The search back helps finding small

supraventricular beats following a pair or series of

high ectopic beats.

The precise localization of the R wave peak is

further refined to 1 ms with the use of five points-

based parabola fitting (Augustyniak, 1999). Since

main QRS components fall far below 125 Hz, this

approach is more robust to the noise and determines

the QRS more precisely than a direct Pan-Tompkins

algorithm applied to a 1ksps ECG time series.

Additionally, the average difference between the

parabola samples and corresponding ECG samples is

taken as a measure of recorded noise. Excess of a

programmed noise threshold activates the noise flag

and deactivates other ST-segment-related alerts as

unreliable.

2.2.2 Morphology Detection

The on-board ECG morphology detection is

restricted to the set of features necessary for

supraventricular beats identification (de Chazal et

al., 2004). The classification procedure distinguishes

normal sinus beats (N) from arrhythmic beats (V,

others, artefacts) and, following the power economy,

is organized in two stages.

First, the rhythm stability is assessed by beat-to-

beat comparison of two features: the difference of

signal sections isolated in the ±100 ms (i.e. 200

samples) vicinity of consecutive R wave peak and

the difference of RR-interval. Next, if both values

fall below the respective thresholds, the beat origin

attribute is copied from the precedent beat. Large

differences in beat shapes or RR intervals indicate

the possible occurrence of other morphologies that

need to be excluded from ST-segment analysis. The

value of threshold is initially set to 10% but may be

modified in sensor initialization stage.

Only for beats with over threshold difference,

signal-based geometrical features selected as the

Wearable Sensor Node for Cardiac Ischemia Detection

most discriminative for atrial and ventricular beats

(the surface to perimeter ratio and the count of

samples of over-threshold speed) are calculated and

contribute to the final decision about the beat origin

attribute (Augustyniak, 1997). The morphology

detector is tuned to restrictively eliminate abnormal

heartbeats (prefers false negative cases). Erroneous

inclusion of a single ectopic beat leads to false ST-

segment measurements and probable false ischemia

alerts. Erroneous exclusion of a true sinus beat, if

occurs occasionally, is delaying but not otherwise

influencing the ischemia detection.

2.2.3 ST-segment Processing

The ST segment is identified as following the R

wave by 60 ms (J-point) and lasting for 80 ms at a

heart rate of 60 beats per minute (Jensen et al.,

2005). As the duration of repolarization phase varies

with the hearth rhythm change, several mathematical

formulas (known as Bazett, Fredericia, Framingham

and Hodges) have been proposed for correction of

QT interval length (Funck-Brentano and Jaillon,

1993). Since the ST-segment is a part of QT interval,

the correction is also applicable for determining ST-

segment border points at which amplitudes are

measured and taken as ischemia markers. All above

correction formulas were implemented in the on-

board software of ischemia sensor. Two

configuration bytes are used for correction

management: one defines the R-to-J distance (fig. 3)

and the other selects the length correction method.

Figure 3: Definitions of baseline (BL) and ST-segment

beginning (J) points for amplitude measurement used in

ischemia markers detection.

Two variables are calculated on ST-segment: the

elevation/depression and the slope. Flat,

downsloping, or depressed ST segment may indicate

coronary ischemia. ST elevation may indicate

transmural myocardial infarction, while ST

depression may be associated with subendocardial

myocardial infarction or other serious heart failures.

The ST elevation/depression is measured as

difference of average values of ECG signal samples

between ST-segment border points and average

values of ECG signal samples on baseline preceding

the QRS complex. The positive value of difference

is called ‘elevation’, while the negative is called

‘depression’.

The ST-segment slope is measured as the slope

coefficient of the straight line best fitted to ECG

signal samples between ST-segment border points

(fig. 4). The slope can be either positive (upwards)

or negative (downwards).

Figure 4: Definitions of ischemia markers: a) elevation/

depression, b) slope.

Although the adopted definitions originate from

medical recommendations, we cannot implement a

single borderline value between normal and

abnormal results as it is common in 12-lead exercise

ECG systems (usually 100 µV for ST elevation). In

a single-lead ischemia sensor with not defined

electrode positions, the threshold values for ST-

segment elevation and slopes are then programmed

at the configuration stage. Depending on electrode

position the elevation may turn to depression and the

slope direction may change (Jiang et al., 2009).

2.3 Communication Module

The ischemia sensor is dedicated to work as a slave

device in a BSN organized with Bluetooth Low

Energy (4.0) technology using either a peer-to-peer

connection, either a star topology. The supervising

device initializes, controls and reads all sensors in

the network accordingly to the programmed

schedule or to the occurrence of ST events.

The communication begins with short data

packets allowing to identify and initialize the sensor

and establish a short time secure channel to transmit

the configuration and long term key. The

initialization procedure is performed repeatedly for

verification of the sensor status (e.g. battery level),

time synchronization or key update. Connection

setup and data transfer are completed within 3 ms,

allowing the supervising device to check and

BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing

100

synchronize the ischemia sensor in short time and at

a cost of very low energy.

In a typical exercise test ST-segment properties

change in response to the physical load thus the

output data are transmitted in synchronous packets at

the end of programmed reporting interval (1 – 256 s)

or asynchronously at each heartbeat interval. The

continuous mode or immediate reporting is not

economical, but exercise test are usually performed

in laboratory condition and last for 30 minutes.

In a typical rehabilitation or surveillance

scenario, a heartbeat interval (RR) should be kept

within a specified range. Excess of the RR range

borders or ST elevation/depression measure is

detected by on-board software and triggers

occasional asynchronous reporting. In this mode

connection duty cycle is very low improving data

security and power saving. This is particularly useful

for everyday living conditions.

The configuration packet has a length of 32 bytes

consists of:

• two configuration bytes for heartbeat and

morphology detectors,

• two configuration bytes for ST-segment

alignment and length correction,

• four bytes of HR and ST thresholds setup,

• two bytes for reporting mode setup,

• twenty bytes of long term security key.

• two spare bytes for further sensor

configuration.

The report packet has a length of 8 bytes and

consists of:

• two bytes of sensor status report,

• four bytes of diagnostic data: RR-interval,

ST-segment elevation and slope,

• one byte of actual ST segment length,

• one byte of threshold excess flags.

The BLE uses modern communication

technology and security-oriented transmission

protocols. The security of low-energy transmission

within the BSN is based on AES-CCM encryption

algorithm with exclusive long time keys passed from

supervising device to each remote sensor via

individual temporary secure channels. A short term

key for the temporary channel is a string of 6

numeric digits generated uniquely for each pairing

(Padgette et al., 2017). Since the sensor

configuration session occurs randomly and lasts for

only few milliseconds, the man-in-the-middle

attacks have little chance to success.

3 EXPERIMENTAL

VALIDATION

3.1 Testing Conditions

The prototype was tested in three laboratory

conditions (referred to by their numbers):

1. with arbitrary waveform generator replaying the

pre-recorded real exercise ECG of known HR

and ST-segment results,

2. with cardiac patients undergoing a regular stress

test diagnosis for possible cardiac ischemia as

accompanying device,

3. with three human volunteers performing

everyday living activities including physical

training (jogging) or work (gardening).

In all experiments the sensor was supervised by a

smartphone with purpose-built application in peer-

to-peer mode. In experiment 2 the smartphone clock

was synchronized with the exercise test system.

Sensor-originated medical measurements were

stored in the smartphone memory (microSD card)

for further analysis.

Experiment 1 was performed with 30 half-hour

records made with commercial stress test system (by

Aspel) during a Bruce protocol treadmill test (Bruce,

1974). Recorded full disclosure 12-lead

electrocardiograms were first replayed digitally to

validate the correctness of sensor on-board ECG

interpretation procedures (fig. 2). Once the

validation was successful, we replayed the signals

again through the digital to analog converter and fed

to the bipolar inputs of the sensor in three various

combinations: V1-V3, V4-V6 and V2-V5.

Consequently, we had 90 half hour test signals with

reference.

Experiment 2 was performed during 20 regular

half-hour stress test diagnoses in patients suspected

for possible cardiac ischemia. Diagnostic records

were made with commercial stress test system

during a Bruce protocol treadmill test. The

electrodes of the tested prototype were placed in two

irregular locations orthogonal to the chest leads line:

symmetrically 5 cm over and under V2 and

symmetrically 5 cm over and under V5 lead. In this

experiment we tested the difference of ischemia

detection between the proposed wearable sensor and

the commercial exercise system.

Experiment 3 was performed in three young

volunteers wearing the sensor for 72h each during

everyday living activities including physical training

(jogging) or work (gardening). The recorded time

series of HR and ST measurements were compared

Wearable Sensor Node for Cardiac Ischemia Detection

101

to the activity record manually made by the subject.

The alert threshold HR was set to 110 bpm, and

amplitude to 50 µV in order to detect exercise

induced ischemia markers in healthy people.

3.2 Testing Results

Values of medical measurements HR and ST

calculated by the sensor in Experiment 1 were

compared to the annotations made by medical staff

on a source 12-lead exercise record. The setup used

raw ECG signals pre-recorded with 12-lead system,

thus it simulated electrode placement in exact

positions of V1-V3, V4-V6 and V2-V5 respectively.

Consequently in each configuration, the results from

the sensor were referred to the respective ECG

channels (tab. 1).

Table 1: Results of Experiment 1 – comparing HR and ST

calculated by the sensor to the annotations made by

medical staff on a source 12-lead exercise record.

Parameter

Value

Amplitude measurement accuracy

1.5 µV (5 LSB)

Average HR difference

0.7 bpm

Standard deviation of HR

1.7 bpm

Average difference of ST

elevation/depression

5.0 µV (16 LSB)

Standard deviation of difference of ST

elevation/depression

8.2 µV (27 LSB)

Average difference of ST slope

coefficient

0,03 µV/s

Standard deviation of ST slope

coefficient

0,12 µV/s

In experiment 2 the sensor was used

simultaneously to commercial exercise test system

during 20 regular half-hour stress test diagnoses. In

eight out of the 20 patients cardiac ischemia was

diagnosed based on the outcome of the commercial

system. The sensor electrodes were positioned in

alternative location to the 12-lead system, thus direct

comparison of amplitudes were not possible.

Consequently, with ST alert threshold set to 100 µV,

we investigated false detection cases and temporal

difference of ischemia detection between the sensor

and the commercial system (tab. 2).

Table 2: False detection and temporal difference of

ischemia detection analysis.

Parameter

Value

Threshold excess time difference

8 s

Standard deviation of threshold excess

time

17 s

False positive ischemia detection

False negative ischemia detection

True positive ischemia detection

True negative ischemia detection

In experiment 3 the sensor was used in three

healthy volunteers with no previous ischemia record

to detect the physical exercise events of two types

(jogging and gardening) in their everyday life. As

the exercise manifests itself in slight deviations of

ST-segment measurements, we programmed lower

values of alert thresholds (110 bpm for HR and

50 µV for ST elevation). The sensor electrodes were

positioned at V2-V5 location. The sensor was paired

with a smartphone and the subjects were asked to

always recharge the battery at night (i.e. where no

activity is performed) in order to capture all

exercise-related ST events. During a continuous 72

hour session the subjects were asked to act

accordingly to their regular activity with taking

notes on beginning and ending time (tab. 3).

Table 3: Statistics of physical activity detected by the

sensor and reported by the volunteers.

Parameter

Value

Jogging events

Average jogging events duration

41 (± 17) min

Jogging start to HR threshold excess time

(average ± STD)

45 (± 18) s

Jogging start to ST elevation threshold

excess time (average ± STD)

170 (± 51) s

Gardening events (digging)

Average gardening events duration

3.3 (± 4.1) min

Gardening start to HR threshold excess

time (average ± STD)

27 (± 37) s

Gardening start to ST elevation threshold

excess time (average ± STD)

71 (± 54) s

BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing

102

4 DISCUSSION

A small and lightweight intelligent sensor for

ischemia detection has been proposed, designed and

prototyped.

First tests with pre-recorded signals helped to

evaluate the correctness of on-board software

implementation and accuracy of performed

measurements. A detailed review of errors leads to

the conclusion that amplitude measurements

accuracy is of order of 5 LSB (1.5 µV), and values

of ST segment amplitude-related errors greater than

that are due to occasional false positive or false

negative morphology detection. Nevertheless, in

comparison with regular 12-lead exercise test

systems the measurement accuracy of the ischemia

sensor is competitive. The accuracy of ST slope

coefficient cannot be quantitatively compared, since

no known exercise test system provide respective

technical data. Most systems simply state the slope

is going upward (positive coefficient) or downward

(negative coefficient), what is sufficient for medical

purposes.

In Experiment 2 the sensor was tested

simultaneously to a commercial exercise test system

with several patients, in some of which cardiac

ischemia was diagnosed. Although the wearable

sensor used alternative lead positions, the test

proved acceptable detection accuracy and reasonable

simultaneity of detection. The true accuracy of ECG

exercise stress testing for ischemia detection is only

of about 75% and all ambiguous cases are subject to

further diagnostics based on coronary catheterization

or magnetic resonance imaging. Since no such

verification was made in our case, the false detection

cases may be interpreted in reference to a

commercial stress test system, and not in reference

to a ground-truth medical diagnosis of cardiac

ischemia.

Third experiment showed that the cardiac

ischemia sensor is a truly wearable device, even if

the prototype is too large and cumbersome for

unobtrusive operation in daily living conditions. The

accurate detection of physical load with the heart

rate is not surprising, but, what is more relevant in

our case, can also be reliably done with change of

ST parameters. What is most important in our case,

the ST segment changes may be analysed in context

of the heart rate, what enabled detection of ECG

ischemia markers during random physical activity

taken by the subject. Moreover, during this test the

sensor worked in asynchronous reporting mode,

where the excess of the RR or ST

elevation/depression range borders is detected by

on-board software and triggers occasional reporting.

Average power consumption with twelve short

communication sessions a day was estimated to

0.8 mW, what grants a month of autonomous

operation without battery replacement.

The presented study has several limitations: only

one prototype was build, simplified algorithms were

used for heart beats detection and morphology

classification, the number of patients was small, and

the number of those with ischemia was even smaller.

Therefore, we consider our work as a proof of

concept and are looking forward to build an

industrial prototype and test it in clinics with true

ischemia reference. This would reveal one of the

expected advantage of the sensor which consists in

unrestricted electrode positions.

The performance of proposed wearable system is

comparable to a standard 12-lead exercise test

system. Thanks to reduced weight and size and

prolonged autonomy its application area extends to

everyday activity detection and risk stratification in

home monitoring. With a wearable ischemia sensor,

the cardiologists may extend the scope of their

cardiac surveillance get better responsiveness from

their patients and optimize the monitoring

accordingly to patients’ medical history (e.g.

localization of the infarct).

ACKNOWLEDGEMENTS

This work is supported by the National Centre for

Research and Development (NCBiR) under Grant

No. STRATEGMED/269043/20/NCBR/2017

REFERENCES

Augustyniak, P., 1997. The use of shape factors for heart

beats classification in Holter recordings. In Computers

in Medicine Conf. pp. 47-52.

Augustyniak, P., 1999. Recovering the precise heart rate

from sparsely sampled electrocardiograms. In

Computers in Medicine Conf, pp. 59-65.

Augustyniak, P., 2011. Wearable wireless heart rate

monitor for continuous long-term variability studies.

Journal of Electrocardiology, vol. 44 no. 2 pp. 195–

200.

Augustyniak, P., 2016. Remotely Programmable

Architecture of a Multi-Purpose Physiological

Recorder”, Microprocessors and Microsystems vol. 46

pp. 55-66 DOI: 10.1016/j.micpro.2016.07.007.

Bazett, J. C., 1920. An analysis of time relation of

electrocardiograms. Heart vol. 7, pp. 353–367.

Wearable Sensor Node for Cardiac Ischemia Detection

103

Bruce, R. A., 1974. Methods of exercise testing: Step test,

bicycle, treadmill, isometrics. Am. J. Cardiol. vol. 33,

iss. 6, pp. 715-720.

de Chazal, P. D., O’Dwyer, M., Reilly, R. B., 2004.

Automatic classification of heartbeats using ECG

morphology and heartbeat interval features. IEEE

Trans. on Biomedical Engineering, vol. 51, pp. 1196-

1206.

Figueiredo, C. P., Becher, K., Hoffmann, K. P., Mendes,

P.M., 2010. Low power wireless acquisition module

for wearable health monitoring systems. In 32nd IEEE

EMBS Conf. 2010, pp. 704-707.

Funck-Brentano, C., Jaillon, P., 1993. Rate-corrected QT

interval: techniques and limitations. Am J Cardiol.

72(6):17B-22B.

Hao, Y., Foster. R., 2008. Wireless body sensor networks

for health monitoring applications. Physiological

Measurement, vol. 29, R27.

Jensen, B. T., Abildstrom, S. Z., Larroude, C. E., Agner,

E., Torp-Pedersen, C., Nyvad, O., Ottesen, M.,

Wachtell, K., Kanters, J. K., 2005. QT dynamics in

risk stratification after myocardial infarction. Heart

Rhythm. vol. 2 no. 4, pp. 357-64.

DOI:10.1016/j.hrthm.2004.12.028.

Jiang, Y., Qian, C., Hanna, R., Farina D., Doessel, O.,

2009. Optimization of electrode positions of a

wearable ECG monitoring system for efficient and

effective detection of acute myocardial infarction, In

36th Annual Computers in Cardiology Conference

(CinC), pp. 293-296.

Liu, X., Zheng, Y. J., Phyu, M. W., Zhao, B., Je, M.,

Yuan, X. J., 2010. A miniature on-chip multi-

functional ECG signal processor with 30 μW ultra-low

power consumption. In 32nd IEEE EMBS Conf 2010,

pp. 2577-2580.

Menown, I. B., Mackenzie, G., Adgey, A. A., 2000.

Optimizing the initial 12-lead electrocardiographic

diagnosis of acute myocardial infarction. European

Heart Journal, vol. 21, no. 4, pp. 275-83.

Moyer, V. A., 2012. Screening for coronary heart disease

with electrocardiography: US Preventive Services

Task Force recommendation statement. Annals of

Internal Medicine, vol. 157, pp. 512–518.

O’Gara, P. T. et al., 2013. 2013 ACCF/AHA Guideline for

the Management of ST-Elevation Myocardial

Infarction. Circulation. vol.127, pp. e362-e425.

Padgette, J., Scarfone, K., Chen, L., 2017. 2017 Security

Guide to Bluetooth - Recommendations of the

National Institute of Standards and Technology,

Special Publication 800-121 Revision 2, [Online]

http://nvlpubs.nist.gov/nistpubs/SpecialPublications/

NIST.SP.800-121r2.pdf (accessed 30.06.17).

Pan, J., Tompkins, W. J., 1985. A real-time QRS detection

algorithm. IEEE Transactions on Biomedical

Engineering, vol. 32, no. 3, pp. 230-236.

Ueno, A., Akabane, Y., Kato, T., Hoshino, H., Kataoka,

S., Ishiyama Y., 2007. Capacitive Sensing of

electrocardiographic Potential Through Cloth From

the Dorsal Surface of the Body in a Supine Position: A

Preliminary Study. IEEE Transactions on Biomedical

Engineering, vol. 54, no. 4, pp. 759-766.

Walsh, J. A., Topol, E. J., Steinhubl, S. R., 2014. Novel

Wireless Devices for Cardiac Monitoring. Circulation,

vol. 130, no. 7, pp. 573–581.

http://doi.org/10.1161/CIRCULATIONAHA.114.0090

24.

Weder, M., Hegemann, D., Amberg, M., Hess, M., Boesel,

L. F., Abächerli, R., Rossi, R. M., 2015. Embroidered

Electrode with Silver/Titanium Coating for Long-

Term ECG Monitoring. Sensors (Basel, Switzerland),

vol. 15, no. 1, pp. 1750–1759.

http://doi.org/10.3390/s150101750.

Sahoo, P. K., Thakkar, H. K., Lee, M.-Y., 2017. A Cardiac

Early Warning System with Multi Channel SCG and

ECG Monitoring for Mobile Health. Sensors, vol. 17,

711; doi:10.3390/s17040711.

Teng, S-L., Rieger, R., Lin, Y-B., 2014. Programmable

ExG Biopotential Front-End IC for Wearable

Applications, IEEE Transactions on Biomedical

Circuits and Systems. vol. 8 no. 4, pp. 543-551.

Tseng, Y., Ho, Y., Kao, S., Su, C., 2012. A 0.09 W Low

Power Front-End Biopotential Amplifier for Biosignal

Recording, IEEE Transactions on Biomedical Circuits

and Systems. vol. 6, no. 5, pp. 508-516.

BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing

104