Seediscussions,stats,andauthorprofilesforthispublicationat:http://www.researchgate.net/publication/265016066

Heartmonitoringsystems-Areview

ARTICLEinCOMPUTERSINBIOLOGYANDMEDICINE·NOVEMBER2014

ImpactFactor:1.48·DOI:10.1016/j.compbiomed.2014.08.014

CITATION

1

DOWNLOADS

24

VIEWS

63

2AUTHORS,INCLUDING:

PuneetJain

IndianInstituteofTechnologyJodhpur

4PUBLICATIONS6CITATIONS

SEEPROFILE

Availablefrom:PuneetJain

Retrievedon:20June2015

Heart monitoring systems—A review$

Puneet Kumar Jain n, Anil Kumar TiwariCenter of Excellence in Information and Communication Technology, Indian Institute of Technology Jodhpur, Rajasthan, India

a r t i c l e i n f o

Article history:Received 15 April 2014Accepted 12 August 2014

Keywords:Heart monitoring systemCardiovascular diseasesCardiographyElectrocardiographyPhonocardiographyPhotoplethysmographySeismocardiography

a b s t r a c t

To diagnose health status of the heart, heart monitoring systems use heart signals produced during eachcardiac cycle. Many types of signals are acquired to analyze heart functionality and hence several heartmonitoring systems such as phonocardiography, electrocardiography, photoplethysmography andseismocardiography are used in practice. Recently, focus on the at-home monitoring of the heart isincreasing for long term monitoring, which minimizes risks associated with the patients diagnosed withcardiovascular diseases. It leads to increasing research interest in portable systems having features suchas signal transmission capability, unobtrusiveness, and low power consumption. In this paper we intendto provide a detailed review of recent advancements of such heart monitoring systems. We introduce theheart monitoring system in five modules: (1) body sensors, (2) signal conditioning, (3) analog to digitalconverter (ADC) and compression, (4) wireless transmission, and (5) analysis and classification. In eachmodule, we provide a brief introduction about the function of the module, recent developments, andtheir limitation and challenges.

& 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Worldwide, the number of patients of cardiovascular diseases(CVD) is huge [1]. Mortality caused by CVD in 2008 was 17.3million which represents 30% of global deaths. In the U.S. alone,2200 persons lose their life due to CVD each day [2]. According toAmerican Heart Association (AHA) report, the total cost of CVDand stroke in the U.S. for 2008 is estimated to be 298 billion dollar[1]. 80% of the total mortality caused by CVD occur in low andmiddle-income countries.

These figures indicate need of systems that should be (1) sensi-tive to detect CVD at early stage, (2) capable of continuousmonitoring, (3) light weight for portability, (4) cost effective. Lackof early stage detection and hence delay in medication causesheart diseases to extent at a level where it is difficult to cure [3].Persons diagnosed with CVD need continuous monitoring ofhealth status of their heart as they are at a higher risk to theirlives as compared to the normal persons. According to the HeartAssociation, people diagnosed with CVD have 4–6 times highermortality than normal one [4]. Portability of such systems makes ithighly useful for elderly patients as this minimizes visits to clinicsor hospitals. A cost effective system will emphasis the use of heartmonitoring systems in low and middle income countries. Proper

diagnosis reduces the mortality caused due to CVD which ensueseconomic up-lift of the country [5].

Due to the problems mentioned above, a lot of work has beendone in development of a diagnostic efficient system [6–10].Keeping in view that heart monitoring systems would be used indifferent socio-economic conditions, rural-urban population, anddeficiency of availability of cardiac experts [11], recent research isfocused towards system features such as cost effective, portability,easy diagnose process, and signal transmission capability. In viewof these developments, we propose a review of various work donein this area.

Portable heart monitoring systems are used in two manners, asshown in (Fig. 1), one is on-site and other is off-site. In on-sitemonitoring, the acquired heart signal is processed on the patientsite, without transmitting it to the remote site. While in off-site,the acquired heart signal is transmitted to a remote site usinga wireless module. On-site heart monitoring system has advan-tages in the case of low latency feedback is required or wirelessaccess is not accessible. Furthermore, it eliminates data transmis-sion and hence eliminates the radio power consumption. However,the on-site monitoring has limitation that it has only a set ofgeneral diagnosis steps and thus unable to perform a detaileddiagnosis. On the other hand, in off-site monitoring, diagnosis isperformed at remote location with high computation capableprocessors and supports input from a cardiologist. This makes itsuitable for accurate and detailed diagnosis. It is attractive becauseof higher processing capability and less power restrictions on suchremote computation. Off-site monitoring also reduces the falsealarm rate and thus reduces visits to clinics or hospitals. In view of

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/cbm

Computers in Biology and Medicine

http://dx.doi.org/10.1016/j.compbiomed.2014.08.0140010-4825/& 2014 Elsevier Ltd. All rights reserved.

☆This paper was not presented at any IFAC meeting.n Corresponding author. Tel.: þ91 9252903393.E-mail addresses: puneetjain@iitj.ac.in (P.K. Jain), akt@iitj.ac.in (A.K. Tiwari).

Computers in Biology and Medicine 54 (2014) 1–13

these advantages of off-site monitoring, this paper is intended toprovide a detailed review of recent research in the off-sitemonitoring system.

A typical off-site heart monitoring system consists of fivemodules as shown in (Fig. 2). The system with first four modules– body sensor, signal conditioning, ADC and compression, andwireless module is situated at the patient site. While the fifthmodule that is analysis and classification module is situated atremote site which can be any computational device with highcomputational ability.

Heart monitoring system uses signals produced by heart todiagnose its health status. It extracts diagnostic features from theacquired signal which caries information of heart functionalitysuch as re-polarization, depolarization, and valve movements.Analysis of these features leads to specific health status of heartsuch as normal, arrhythmic, myocardial infarction, regurgitation,and stenosis. However, extraction of diagnostic features from theheart signal is challenging due to its non-stationary nature and thepresence of noises such as muscles movement noise and environ-ment noise in the signal. In this paper, we reviewed recentdevelopments in the area of heart monitoring systems which areportable and have good diagnostic efficiency.

Rest of this paper is organized as follow. Physiology of heartand cardiac cycle in Section 2 and brief introduction aboutimportant cardiovascular diseases in Section 3 are given for read-er's simplicity. Section 4 describes the recent developments in thebody sensors. While different approaches of signal conditioninghave been reviewed in Section 5. In Section 6, analog to digitalconversion and compression techniques are presented. Section 7discusses various wireless transmission technologies and Section 8gives comprehensive review of noise removal algorithms, analysisand classification techniques for heart signals. Conclusion andpotential research area have been presented in Section 10 followedby important references.

2. Physiology of heart and cardiac cycle

Since understanding of various components of heart monitor-ing systems needs knowledge of heart functioning, a relevantphysiology of heart is described in this section.

Heart is a prominent organ of human body. It suppliesreplenish oxygen to each part of the body and removes waste ofeach cell. Physiologically, heart comprises of four chambers namedas left and right ventricles and left and right atrium as shown inFig. 3. There are two atrioventricular valves namely tricuspid valveand mitral valve. As can be seen in Fig. 3, tricuspid valve separatesright atrium and right ventricle while mitral valve separates leftatrium and left ventricle. Aortic valve and pulmonary valve jointlycalled as semilunar valves separate left and right ventricles fromaorta and pulmonary artery respectively. At rest, each cell of theheart muscle has a negative charge, called the membrane poten-tial. Due to rapid change of membrane potential towards zero, dueto influx of positive cations (Naþ and Caþþ), an electricalimpulse is generated at sinoatrial node. From sinoatrial node, theimpulse spreads over both the atrium and both the ventricles.Presence of the impulse causes the contraction of atrium andventricles sequentially.

Contraction of both atrium pushes the blood into respectiveventricles. Then the impulse spreads all over left and rightventricles which causes the contraction of both the ventricles.This contraction results to closing of both atrioventricular valvesand opening of both semilunar valves. During this contractionphase, oxygenated blood from left ventricle flows into the body

Fig. 1. Heart monitoring system.

Fig. 2. Off-site heart monitoring system.

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–132

through aorta while deoxygenated blood from right ventricle flowsinto the lungs through pulmonary artery for oxygenation. Sincebody and lungs receive blood due to contraction process, pressurein body lungs becomes higher than pressure in atrium. Due to thispressure difference, now, blood flows into left and right atriumfrom lungs and body respectively. This process completes acardiac cycle.

3. Important cardiovascular diseases

Heart beats 100,000 times and pumps 2000 gallon blood ina day [12]. Heart function gets affected due to many factors such aspsychosocial stress, smoking, excessive use of alcohol, malnutri-tion, lack of physical activity, and congenital diseases [13]. Thesefactors may affect electrical activity of heart, structure of heart,and arteries. Due to these defects, different heart diseases occur.Dysfunction of electrical conduction system causes diseases suchas sinus arrhythmia, atrial fluttering, atrial fibrillation, ventricularfibrillation, atrioventricular block, and bundle branch block.Defects in structure of the heart cause regurgitation, stenosis,enlargement of chambers, ventricular septum defect, etc. Defectsin arteries cause hypertension, stroke, myocardial infarction, etc.

According to American Heart Association report [2], followingare the most common heart diseases. Myocardial infarction (heartattack), which occurs due to blockage in coronary arteries whichsupply blood to the heart cells. Insufficiency of blood supplycauses dying of the heart cells and consequentially heart muscleslose pumping capacity. Hypertensive heart diseases occur whenthe blood pressure in arteries is much higher than the normal.High pressure causes stiffness of arteries, consequentially bloodflow gets affected. Arrhythmia is fast and irregular beating of theheart. Congenital heart diseases are those which are present at thetime of birth. Heart failure is a condition which indicates inabilityof the heart to pump the blood. Mortality caused by myocardialinfarction was 21.6% of all mortality due to CVD in U.S., while 9.2%by hypertensive heart disease, 5.7% by arrhythmia, 8.0% by con-genital heart diseases, and 6.9% by heart failure was accounted [2].Since heart problems need diagnostic system, the following sec-tion discusses about body sensors used to acquire heart signal forthis purpose.

4. Body sensors

Different body sensors acquire heart signals in different formssuch as electrical signal, acoustic signal, seismic signal, and optical

signal. Fig. 4 shows signals of one cardiac cycle acquired usingelectrodes, stethoscope, accelerometer, and diode asdescribed below.

4.1. Electrodes (electrocardiography)

Initial electrocardiography (ECG) was based on string galvan-ometer and was invented by Willem Einthoven in 1903. Asdiscussed in Section 2, an electrical impulse originates at sinoatrialnode and then travels through atria and ventricles. ECG measuresthe electrical activity of heart using electrodes placed on both sideof the heart. The measured signal consists of different wavesnamed as P, Q, R, S, T and U as shown in Fig. 4(a). P waverepresents atria contraction while Q, R and S waves (called as QRScomplex) reflect contraction of both left and right ventricles. Twave represents relaxation of ventricles and U wave is caused bythe relaxation of inter-ventricular septum. Thus duration andamplitude of these waves provide significant information fordiagnosis of health status of heart.

Extraction of duration and amplitude of some of the waves (P, Tand U) is difficult due to very weak amplitude, typically in therange of 100–300 μV [14]. ECG signal lies in frequency band of 1–250 Hz, where flicker noise is dominant and common-modeinterference from the main power line is likely to interfere with

Fig. 4. Signals of one cardiac cycle: (a) electrocardiography signal, (b) phonocardio-graphy signal, (c) seismocardiography signal, and (d) photoplethysmography signal.

Fig. 3. Anatomy of the human heart.

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–13 3

ECG signal [15]. To overcome these problems signal conditioning,discussed in Section 5, is essential.

Major problem with ECG is to establish good electrical con-ductivity between skin and electrodes. Some of the ECG sensors,widely used in practice, can be classified into following threecategories.

4.1.1. Wet sensorsIn these type of sensors, Ag–AgCl electrodes are attached to the

skin using gel which provide a conducting medium for chargetransfer between the electrodes and the body [6]. These sensorsprovide good signal quality, but it is inconvenient in terms of longterm wear-ability due to use of gel which creates irritation andetching problem [6]. Moreover, For signal acquisition, attachmentof electrodes to different points on the body restricts patient'smobility. The acquired signal quality may deteriorate due to sweat[16] and due to gel dehydration [17].

4.1.2. Dry sensorsThese sensors use a metal plate direct placed on the skin

without the use of gel. Thus the problem of irritation and etchingcaused by gel has been eliminated [18]. Although, it still hasa direct contact with the skin. Dry sensors are robust to environ-ment noises and sweat noise but more vulnerable to motion noisecompare to wet sensors. Quality of the signal acquired using thesesensors depends on the composition of the materials and the sizeof the electrode [17]. Increasing size of the electrode gives bettercapacitance and consequentially good signal quality but itdecreases the patient's convenience.

4.1.3. Capacitive coupled sensorsCapacitive coupled (CC) sensors avoid direct contact with the

skin that minimizes patients' inconvenience as stated in theprevious subsections, (a) and (b). A thin layer of insulator is placedbetween the body and metal-plate sensing electrode [6]. Theelectrode, together with the skin and insulator, forms a capaci-tance that conveys the ECG signal from the body to the sensor.Sensitivity of such sensors increases with the value of capacitance,which can be increased by increasing electrode area, by reducingthickness of insulator, and by using insulator with high dielectricconstant. General expression of capacitance is given as follows:

C ¼ ε0Ad

ð1Þ

where ε0 is the dielectric constant, A is the electrode area, and d isthe thickness of insulator. CC sensors have been developed onchair [19,20], on bed [21] and textiles [22]. Development of sensorson chair, bed and textiles supports continuous heart monitoringeven when working in office and sleeping. CC sensors are highlysensitive to motion noise as in case of dry sensors. This is because,a movement of electrode changes the coupling capacitance andconsequentially the acquired signal [6].

A comparative study between three types of electrodes ispresented in (Table 1). For clinical use, where simplicity ofoperation, less processing time, and good signal quality arepreferred, wet sensors are suitable. Additionally, the availability,relative cheapness, and disposability of wet electrodes overcomehygiene concerns. While, the dry and capacitive electrodes areconvenient in use, and consistent in performance. These featuresmake these sensors suitable for long term, and unsupervisedmonitoring. However, the performance of these types of electrodesdepends on the electrode geometry. Furthermore, these electrodesrequire proper shielding and settling time to perform comparableto, or better than wet electrode. Researchers in the past have madenumerous attempts to overcome these problems [16].

4.2. Stethoscope (phonocardiography)

Stethoscope was invented by René Laennec in 1816. It isbasically a transducer which converts vibration signal into acousticsignal. A phonocardiogram (PCG) is a plot of acoustic signal,acquired by stethoscope. Stethoscope makes PCG a highly porta-ble, low cost, and non-invasive cardiography technique [10]. PCGsignal, as shown in Fig. 4(b), consists of two classical heart soundsknown as S1 (Lub) and S2 (Dub). S1 is generated due to closing ofthe tricuspid and mitral valves. It is composed of energy in 40–65 Hz frequency band and 130–200 ms time duration [10]. S2 isgenerated due to closing of aortic and pulmonary valves and lieswithin 45–65 Hz frequency band and 100–150 ms time duration[10]. Period between S1 and S2 is called as systole, while S2–S1phase is called as diastole. There are two more components in PCGsignal called S3 and S4 and they rarely occur. Murmurs areadditional sounds that lie within a frequency band of 100–500 Hz [10]. They indicate diseases in heart such as aortic stenosis,pulmonary stenosis, mitral regurgitation, and mitral stenosis.Stenosis and regurgitation are valvular diseases caused by stiffnessof valves and improper closing of valves respectively. Stenosisrestricts proper blood flow, while regurgitation causes blood toflow in opposite direction to the normal [10]. PCG has highpotential to detect these valvular diseases and important cardio-vascular diseases except myocardial infarction and congenitalheart diseases. However it indicates the abnormality caused bymyocardial infarction and congenital heart diseases. PCG signalmay get contaminated by noises occurring due to breathing,movement of stethoscope while recording, ambient sources, etc.As a result of these noise problems, many filtering techniques havebeen developed to minimize affect of noise on PCG signal. Detaileddiscussion about the same is given in Section 8.

4.3. Accelerometer (seismocardiography)

Seismocardiography (SCG) is another non-invasive techniquebecause it works using accelerometer, a non-invasive device.It measures mechanical vibrations which are generated by heartmovement and transmitted to the chest wall [23]. As shown inFig. 4(c), SCG contains waves corresponding to atrial contraction(ATC), mitral valve closing (MC), aortic valve opening (AO), point ofmaximal acceleration in the aorta (MA), aortic valve closure (AC),mitral valve opening (MO), and rapid filling of left ventricle (RF)[24,25]. Shapes of these waves give significant information abouthealth status of heart. SCG is convenient to patients, as there is noneed of multiple electrode contacts as in ECG. However, obtaininga clean SCG signal from an accelerometer is difficult, because ofinterference due to breathing.

4.4. Diode (photoplethysmography)

In photoplethysmography (PPG) fluorescent body parts such asearlobe and finger are illuminated with lights of different wave-lengths emitted from light emitting diodes (LEDs). Then intensityof transmitted or reflected light is measured by photo-diode [26].The measured intensity varies in time with the heart beat becauseblood vessels expand and contract with each heartbeat. The PPGwaveform is composed of an ac component and a quasi-dccomponent [27], as shown in Fig. 4(d). The ac component isassociated with heart beat and has fundamental frequency typi-cally around 1 Hz. The quasi-dc component, superimposed with accomponent, relates to the respiration system. As the ac componentof the PPG signal is in synchronization with heart beat, informa-tion about heart rate can be extracted from it. PPG uses LEDs andphoto-diode which makes it low cost, non-invasive, easy to useand portable system. Since it operates optically, it is not

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–134

intrinsically susceptible to capacitive coupling interference as inECG [28]. However, photo-diodes are sensitive to natural andartificial light sources. PPG based heart monitoring systems areunobtrusive as size and weight of the device of PPG is low. PPG isgenerally used in direct contact with the patient's skin as in thecase of other systems (ECG, PCG, SCG). But in the case of neonates,skin-damage, direct contact to skin is not feasible and henceHuelsbusch and Blazek [29] proposed a remote PPG (rPPG) thatcan acquire PPG signal without the contact with skin. The mainconcern with rPPG is its sensitivity to the subject motion.

Portability of the system is induced by small size and lowweight sensor, used by the system. ECG, PCG, SCG and PPG areportable as they satisfy these conditions. Portable diagnosticsystem is useful in many scenarios. Most use of the portablesystem is for long term monitoring, which is required for acutesymptom detection and early diagnosis of the heart diseases.However the acceptance of a portable system depends on thediagnostic efficiency, performance of the system in differentenvironmental conditions, comfort to the user, easy to operate,and cost.

ECG signal contains information about the electrical activity ofthe heart. Thus provides better insight on the issues related toelectrical conduction abnormality of the heart. While the PCGsignal acquires acoustic sounds produced by the heart valves(mechanical action) and thus useful in diagnosis of the valvulardiseases. Due to different source of producing these signals (ECGand PCG), the existence of a problem (e.g. structural abnormalities)in PCG signal does not imply the existence of the same problem inECG signal and vice versa. SCG signal is also produced by themechanical action (acceleration of the heart), measures compres-sion waves. Acceleration is a second derivation of the displace-ment and that is why SCG signal contains more informationcompared to the PCG signal. PPG signal provides only limitedinformation about the heart. It measures the blood variation in theblood vessels. Although, with the combination of ECG or PCG it hasbeen used for the calculation of pulse transit time (PTT), which isan important diagnostic parameter in the case of obstructive sleepapnea detection and blood pressure measurement.

For clinical use, all methods are suitable as signal to noise ratio(SNR) remains high. At home, in the presence of the environ-mental and motion noise, robustness of the sensor to the noises isa major issue. In case of ECG, as discussed in (Table 1), wet sensorsare robust to noise while dry and capacitive sensors are vulnerableto noise. PCG is more vulnerable to patient's motion noise andenvironmental noise compared to the ECG. On the other hand, SCGis robust to the motion noise and environmental noise.

PPG is the most suitable technique in terms of portability andused widely for heart rate calculation. Problem with the ECG is thatit uses a gel which causes etching problem, and reduces comfort ofthe patient and hence reduces acceptability for long term monitor-ing. PCG has advantages over ECG in terms of comfort of the patientand easy to operate. SCG is superior to the PCG in terms of comfortbecause of the low weight (o3 g) accelerometers.

5. Signal conditioning

Heart signals, acquired by different body sensors, often getcontaminated by noise components such as flicker noise,common-mode interference, power-line interference, and baselinewandering [15]. Also, amplitude of the acquired signal is typicallylow. A signal conditioning module typically consists of algorithmfor noise minimization and amplifier to amplify low amplitudesignals. This module operates on signals in analog domain. Powerconsumption of this module used to be low so as to support longterm operability of heart monitoring systems. For the samepurpose, Rieger [30] proposed a variable gain circuit consistingof a continuous-time input stage using lateral bipolar transistors.Spinelli et al. [31] proposed a driven right leg circuit to reducecommon-mode interference. Gomez-Clapers and Casanella [18]used dual ground configuration to reduce the noise caused bypower line interference and base line wandering. Since most of theheart monitoring systems are digital in nature and need commu-nication for remote monitoring, the following section discusesabout analog to digital conversation (ADC) and compressionalgorithms.

6. Analog to digital conversion and compression

Heart signals (analog) are converted into digital signals for itsprocessing on digital computers. This is done by sampling the heartsignal and quantizing the sampled values. This process is done onDigital Signal Processor (DSP) called Analog to Digital Converter (ADC).Selection of the DSP depends on the desired sampling rate, number ofbits to be used for quantisation, operating frequency and powerconsumption. As described in Table 2, MSP based processors havelower power consumption compared to the PIC based processors,while the PIC based processors have higher operating frequencycompared to the MSP based processors. Both types of processorsprovide multiple working and idle modes according to the requiredcomputational power, to optimize the power consumption. DSP with

Table 1Three types of electrodes.

Characteristics Wet electrodes Dry electrodes Capacitive electrodes

Signal acquisition AG/AGcl electrodes, uses electrolyte Benign metal (stainless steel), noelectrolyte

Metal or semiconductor, no directcontact

Signal quality Low contact impedance ensues goodsignal quality

Depends on electrode geometry Depends on electrode geometry

Consistency Gel dehydrate with time,which reduces quality of the signal

After settling time good performancedue to reduction in skin/electrodeinterface

After settling time good performancedue to reduction in skin/electrodeinterface

Convenience Use of electrolyte cause etchingproblem, removal of gel is unpleasantand time consuming, and toxicologicalconcern

Direct contact with skin, which maycause irritation

No direct contact, fabricated in cloth,chair, bed, that increases convenienceto user

Size Lightweight Heavy due to shielding Bulky due to required circuitry forbuffers and extra cables for power

Noise vulnerable Moving charge sensitivity Movement artifact Motion and environment artifact,electric field problem

cost Lower cost Expensive Expensive

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–13 5

low power consumption supports long term monitoring of the heart.To optimize the power consumption, Bachmann et al. [32] proposed aDSP with the capability to perform in different power modes accord-ing to the required accuracy and available computational power.Power consumptionwas optimized at different abstraction layers fromapplication optimization and mapping to system.

Conventional sampling techniques, sample signals at or aboveNyquist rate, which ensues perfect reconstruction of the signal.Nyquist rate is twice the maximum frequency component presentin the signal to be sampled. Typically, heart signal components arebelow 1 kHz frequency and hence, as per Nyquist rate, 2 K samplesper second are sufficient to avoid aliasing error. However, even 2 Ksamples per second sampling rate generates a huge number ofsamples, if the heart is monitored for a long time. Consequentially,the power requirement of DSP increases as the number of samplesto be processed is huge. In spite of it, compressed sensing (CS)enables sub-Nyquist sampling of signals.

Compressed sensing (CS) is a data acquisition approach thatrequires only a few incoherent measurements to compress signalsthat are sparse in some domain [38]. Let [α] be an original inputvector of dimension N�1 and [Ψ] is the N�N sampling basis orsparsifying matrix containing orthonormal basis (such as a waveletbasis) [ψ1;ψ1;…;ψn]. Then [X], sparse in [Ψ] domain with lengthN can be found as

½X� ¼ ½Ψ �½α� ð2ÞThen output compressed vector is defined as

½Y � ¼ ½ϕ�½X� ð3Þwhere [ϕ] is theM �Nmeasurement or sensing matrix. So, we getan output vector [Y] with length M, where MoN. CS captures Mmeasurements from N samples using random linear projections.Now, as the lower number of measurements were taken than theoriginal signal, non-linear optimization techniques are used toreconstruct the original signal [38]. Reconstruction of the signalcan be achieved as

MinJ bX J1 subject to ½Y� ¼ ½ϕ�½bX � ð4ÞPerfect reconstruction of the signal depends on the incoherencebetween [Ψ] and [ϕ] matrices. Thus, random matrices can be usedas a measurement matrix because random matrices are, with highprobability, highly incoherent with any fixed basis [Ψ]. CS is

considered to be non-adaptive because measurement matrix [ϕ]remains constant.

Several measurement matrix design considerations and recon-struction algorithms have been presented in [39] and found thatusing Bernoulli measurement matrix, compression ratio of 16 isachievable. Mamaghanian et al. [40] compared CS and the DWT-based compression algorithms and found that CS was inferior toDWT-based algorithm in compression performance. Despite of it,CS-based compression outperforms in terms of energy efficiencydue to its lower complexity and reduced CPU execution time.

After digitization of analog signals, digital signals are com-pressed to reduce amount of data. The basic purpose of datacompression is to represent the original signal with a smallernumber of bits than that is needed for the original signal. Thecompression is typically achieved by removing redundancy fromthe signal to be compressed. Since power requirement of wirelessmodule directly depends on the amount of data to be transmitted,one of the major advantages with compression is a reduction inpower requirement by wireless module. However, there is a loss ofinformation, in general, when signal is reconstructed from thecompressed data. A proper balance is maintained with compres-sion ratio and the requirement that the information of diagnosticimportance is preserved.

Heart monitoring systems have additional requirement forcompression algorithms to be computationally efficient to supportlong term monitoring. Various compression algorithms for heartsignals have been reported in the literature. Wavelet transform[41,42], Walsh transform [43], Hermite function [44] and discretecosine transform [45] based compression algorithms first decom-pose the signal into coefficients, by projecting the signal onto basisfunctions of transforms. Then compression is achieved by retain-ing only a small number of coefficients which typically preservesessential information of the heart signal. As stated before, compu-tational complexity is crucial for the compression algorithm asthese have to be implemented on the patient side. Computationalcomplexity of DWT is O(N), DCT is O(NlogN), Walsh transform is O(NlogN), and Hermite function is O(Nlog2N). Thus DWT has lowestcomputational complexity. However, compression performanceof the Hermite function based method is better than DCT andDWT based basic compression algorithms [44]. The performanceof these algorithms depends on the following parameters:(a) choice of basis function, (b) decomposition level, and (c) the

Table 2Digital signal processors.

Processor No. of bits Operational frequency (MHz) Power consumption Used in Characteristics

PIC24FJ64GA 10 32 � Run mode: 650 μA at 2.0 V, 1 MHz [6] � Low operating voltage range� Idle mode: 150 μA at 2.0 V � On-the-fly clock switching� Sleep mode: 0.1 μA at 2.0 V

PIC18f2423 12 40 � Run mode: 330 μA at 2.0 V, 1 MHz [33] � Multiple idle and run modes� Idle mode: 5:8 μA at 2.0 V � Nano watt technology� Sleep mode: 0:1 μA at 2.0 V � On-the-fly clock switching

PIC16F877 8 20 � Run mode: 600 μA at 3 V, 4 MHz [34] High performance RISC CPU� Idle mode: 20 μA at 3 V, 32 kHz� Sleep mode: 1 μA

MSP430f2274 10 16 � Run mode: 270 μA at 1 MHz, 2.2 V [18,35] � Ultra-fast wake-up� Idle mode: 0:7 μA � Ultra-low power� Sleep mode: 0:1 μA � RISC mixed-signal microprocessors

MSP430F1611 12 8 � Run mode: 330 μA at 1 MHz, 2.2 V [36] � Ultra-fast wake-up from stand-by� Idle mode: 1:1 μA � Five power saving modes� Sleep mode: 0:2 μA

MSP430F2410 12 16 � Run mode: 270 μA at 1 MHz, 2.2 V [37] � Ultra-fast wake-up�] Idle mode: 0:3 μA � Ultra-low power� Sleep mode: 0:1 μA

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–136

percentage of retained energy (number of coefficients). In theseapproaches, trade-off between the percentage of retained energyand compression ratio is crucial. Increment in the percentage ofretained energy reduces compression ratio and enhances recon-structed signal quality and vice versa. Retained coefficients arecompressed using conventional compression algorithms such aszero-removal [41], Huffman coding [41,42], dead zone quantiza-tion [45], and run length coding [42,46]. While, Sharma et al. [47]applied multi-scale principal component analysis (MSPCA) onwavelet transform coefficients and then MSPCA coefficients areuniformly quantized and encoded by Huffman coding. All theabove algorithms compressed the entire frame with same com-pression ratio. On the other hand, researchers have been proposedapproaches to use different compression ratio for different block ofsignals [43,46,48,49]. Different statistical parameters have beencalculated to identify significance of the segment such as Wanget al. [48] calculated kurtosis, Kim et al. [49] calculated meandeviation (MD), Ma et al. [50] calculated wavelet coefficientenergy. In [43], significance of the segment is calculated basedon the energy of the Walsh coefficients. While, Rajoub [46] appliedDWT and then divided the coefficients into three groups based onmagnitude of coefficients and then applied different thresholdingfor each group.

Researchers have also proposed compression algorithms thatpreserve features of heart signal (rather than preserving thewaveform) [51,52]. In [51] Alvarado et al. proposed a compressionalgorithm based on integrator and fire sampler. Similarly, Kimet al. [52] proposed an algorithm based on curvature points, whichcalculated the important information from the signal.

Compression algorithms which require low computation aresuitable for long term heart monitoring. DWT based compressionalgorithms have lower computational complexity than otheralgorithms and thus have been used extensively. On the otherhand, feature preserving compression algorithms have high com-pression ratio. They are suitable for heart signals because diagnosiscan be performed based on these features. However, selectionprocess of the diagnostic features from the heart signal is complexprocess. Furthermore, the performance of these types of algorithmdeteriorated in the presence of noises.

7. Wireless module

Digitized and compressed heart signals are transmitted toremote site. In off-site monitoring, analysis and classification ofthe heart signals are performed at the remote site. Transmitterconsists of wireless module which helps to transmit heart signalsto remote site. Low power consumption, convenient connectionprocess, and low latency are some important features of wirelessmodules that promote wide acceptance of heart monitoring

systems. In the literature, various wireless communication tech-niques and protocols have been proposed for transmission pur-pose (Table 3). Bluetooth 4.0 [37] wireless system supports24 Mbps data rate, has working range up to 100 m, and consumeslow power. Bluetooth devices with these features are suitable to beintegrated with heart monitoring systems. But Bluetooth wirelesssystems require initial connection setup that has to be donemanually. Patient's intervention is not desirable in a heart mon-itoring system as it reduces convenience. To overcome thisproblem an approach was proposed by Morak et al. [37] usingradio-frequency identification (RFID) and near field communica-tion (NFC). In this approach, the connection establishes by bring-ing two NFC enabled devices closer and using RFID information ofboth devices. The drawback of this approach is that it requirespermanent activation of Bluetooth which results in extra powerconsumption. Moreover, NFC can support data rate up to 424 kbpsonly. Since data rate is lower than Bluetooth 4.0, it takes long timeto transmit data. A good review of the state-of-the art technologiesfor wireless network was presented by [32].

Keeping in view the desired features of wireless module,various protocols have been proposed [6,36]. Chen et al. [36]proposed a reliable protocol based on any-cast routing algorithm.This algorithm automatically selects nearest hop (sink), in case offailure in original path, instead of rebuilding the path from thesource node. Thus, it provides a reliable communication as well asreducing traffic overhead and transmission latency. However,selection of the hop process increases the complexity of routingalgorithm and the complexity increases power consumption. Tooptimize power consumption, Nemati et al. [6] proposed an ANTprotocol. The ANT protocol was used as a low-data-rate wirelessmodule to reduce the power consumption and size of the sensor.ANT is an adaptive isochronous ad hoc wireless protocol based onmaster slave model. It consumes from 1 mA to 6.3 mA current andsupports many topologies such as peer-to-peer, star, tree, andmesh. SimpliciTI is also a low power radio frequency (RF) networkprotocol used in heart monitoring systems [18,35]. SimpliciTI wasdesigned by Texas Instruments for easy implementation anddeployment on RF platforms. It is low data rate and low dutycycle protocol and supports star and peer-to-peer networktopology.

Ma et al. [50] proposed an unequal-error protection approachfor heart signals to reduce transmission distortion and to reducepower consumption of wireless transmission. In this approachmore protection is provided to the segment of heart signal whichcontains diagnostic important features compared to the othersegments. Results showed that nearly 40% of transmission energycan be saved compared to the equal error protection.

In a different approach, Atakan et al. [53] introduced the conceptof a body area network (BAN) with molecular communication,where the messenger molecule is used as a communication carrier

Table 3Wireless modules.

Module Power consumption Size (mm) Transmission range (m) Manufacturer Used in

Receive Transmission

CC2420 (zigbee) 18.8 17.4 7�7 70 Texas instruments [54]Bluescense (blutooth) 33 37�21 Corscience [55]nRF24E1 (Eco-wireless) 22 10 13�11 10 UC Irvine [56]ANT-AP2 17 15 20�20 30 Dynastream Innovations [6]cc2500 (zigbee) 13.3 21.2 4�4 30 Texas instruments [18,35]UZ2400 (zigbee) 18 22 6�6 Uniband Electronics Corp. [36]Zebra (zigbee) 16�33 10–500 senTec Elektronik [57]BlueNiceCom-4 (bluetooth class-2) 65 27�16 20 AMBER wireless [37]Xbee (Emosense) 50 45 24�27 30–90 Digi International Inc. [58]

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–13 7

from a sender to receiver. However, the communications at themolecular scale are subject to numerous problems, some similar tothe ones faced on a larger scale in existing wireless networks.

8. Analysis and classification

Analysis and classification module performs automaticmachine diagnosis that enhances diagnostic accuracy. It is veryhelpful in the present scenario, where the number of cardiologistsis low when compared to the number of cardiac patients [5].Typically, analysis and classification is performed in two steps:noise suppression, and analysis and classification. In noise sup-pression step, noises are suppressed from heart signals. In the nextstep, the heart signals are classified in normal and different CVDs.

8.1. Noise suppression

Noise suppression from heart signals is essential as its presencemay lead to imprecise or inaccurate classification of the signals.For noise suppression, classical filters such as Gaussian filter,Chebyshev filter, Butterworth filter, and Weiner filter have beenused extensively. Because, heart signals lie in the 20–500 Hzfrequency band and these filters are able to suppress noise inthe selected frequency band (below 20 Hz and above 500 Hz). Butnoises, which overlap with spectral contents of heart signals, arenot easy to suppress from the signals. Hence, sophisticated filtershave been proposed in the literature for suppression of these typesof noises (in-band noise) [59,60]. Filters have been developedbased on wavelet transform [59]. In these filters, signals aretransformed into wavelet coefficients, as discussed in Section 6.Then noise suppression is achieved by discarding the coefficientswhich are correlated to noises, by applying threshold. Although,wavelet based filters are able to suppress the in-band noise, butthe threshold value plays a crucial role in this approach. If thethreshold is selected high, signal information will be lost, whilesmall value will not have a significant effect on the signal. Toobtain optimal denoising parameter for DWT based denoising,Messer et al. [61] performed experiments and found that level5 for the signal decomposition and soft thresholding with rigrsurethreshold selection rule gives the best result.

Almasi et al. [60] introduced model-based Bayesian denoisingframework which combined the extended Kalman filter anddynamic model of the heart signal. Results demonstrate that theproposed method has the superiority over wavelet based denos-ing. However, the requirement of a model of the heart signal limitsthe use of this framework.

Researchers have proposed many filtering approaches whichanalyze diversity between characteristics of heart signal compo-nents and characteristics of noises [7,15,62]. Lee et al. [7] used firstorder-intrinsic mode function (F-IMF) to minimize motion noisefrom the heart signals. F-IMF of the clean signal has periodicpatterns, whereas noise contaminated signal has highly varyingirregular dynamics with lower magnitudes. Thus, noisy segmentcan be classified from the clean heart signal. Liu et al. [15] removedthe noises from heart signal components based on the character-istic of wavelet coefficient that the signal coefficients with largemagnitude at a finer scale will also be large in magnitude atcoarser scales. However, for coefficients which are caused bynoises, magnitude will decay rapidly along the scales. Manikandanand Soman [62] calculated lag-1 auto-correlation coefficients,which give positive values for heart signal components andnegative values for spurious noise.

Quasi-cyclostationary nature of heart signals also has beenconsidered to filter noise from the signals [63,64]. Quasi-cyclostationary means that the morphology of the heart signals

does not change abruptly from a cardiac cycle to consecutivecardiac cycle. Thus, noise suppression can be achieved by correlat-ing the consecutive cycles of the signal because noise components,in general, are uncorrelated. However, quasi-cyclostationary nat-ure of the heart signal may not be fulfilled due to variation inwaveform, presence of murmurs, and variation in the timing of theheart sound components. Furthermore, the performance of thisapproach depends on the segmentation of cycle.

Adaptive noise cancellation (ANC) techniques are also foundsuitable for heart signals as they can detect dynamic variation inthe signal [65]. Least mean square (LMS) is an ANC techniquewhich calculates filter coefficients that relate to producing theleast mean squares of error signal (difference between the desiredsignal and the filtered signal). Estimation of filter coefficientsrequires high computation. To reduce the computation of LMSalgorithms, various variations in LMS algorithms have beenproposed in the literature and reported in [65]. To further reducethe computational complexity, the author proposed [65] sign anderror non-linear sign based LMS.

Respiratory system also affects heart signals significantly. Toovercome this problem, Chen et al. [66] proposed a zero crossingmethod. It calculates the time interval between two consecutiveupward and downward points (IBI) in the signal. Then the inverseof IBI gives the frequency of breathing signal, which can beremoved by notch filtering.

Noise often appears in parts of the heart signal recordings.In some part, noise affects severely to the heart signal while inothers it affects mild. In the case of severe contamination, thatpart of the signal can be eliminated from diagnostic considera-tion while in the case of mild contamination, noise suppressionalgorithm can be used. This approach will improve diagnosisefficiency as well as optimize complexity of denoising algo-rithms. This approach will be also helpful in home care systemsfor alarming to the user for the bad signal quality. Thus, it is ofinterest to obtain a signal quality index to find out a subse-quence with better signal quality with respect to the rest of thecycle. Li et al. [67] proposed an optimum heart sound selectionscheme based on cycle frequency spectral density. In thisapproach, the quality of the heart sound signal depends on theperiodicity of the heart signal. In [68], the quality index wascalculated using the Cepstral distance between homogeneouscardiac sounds. In this algorithm, first, the heart signal wassegmented into separate cardiac cycle using wavelet basedapproach. After segmentation, Mel frequency Cepstral coeffi-cient (MFCC) was calculated for each cycle. Finally, the recipro-cal of distance between MFCC coefficients of consecutive cyclegives the quality score. The performance of this algorithmdepends on the segmentation of the heart signal into cardiaccycle. Naseri [69] described an approach to identify the level ofnoise in the heart signal cycle. In this approach, first, the signalis segmented into separate heart cycle. Then, cycles are clus-tered into a finite number of groups based on geometricalparameter and spectral content. Next, median of these clustersis correlated to the test cycle features. Finally, by applying athreshold, the cycle is prescribed as clean or noisy. Although,requirement of a test cycle features limits the use of thisapproach.

8.2. Analysis and classification

Analysis and classification of heart signals are challenging tasksdue to non-stationary nature of the heart signals. Moreover, timeto time varying morphology of heart signals from intra- and inter-patient needs sophisticated classification algorithms. Classificationof heart signals is performed by analyzing diagnostic featurespresent in the signal as follows.

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–138

8.2.1. ElectrocardiographyECG signal consists of different waves, as discussed in Section 4

(A). Each wave is associated with particular functionality of theheart. Analysis of the shape of these waves leads to diagnosis ofimportant cardiovascular diseases which includes MI, hyperten-sive heart diseases, arrhythmia, CHD. The impact of CVDs can beseen on the waves in ECG signal. Myocardial infarction causes STelevation or depression depending on the severity of the infarc-tion. Location of the infarction can be identified by analyzing ECGsignals of different leads. In the case of hypertension, QRS voltageincreases due to both thickening of wall (pressure overload) anddilatation of chamber (volume overload) of the left ventricle. TheRR interval is critical in the diagnosis of many arrhythmia such aspremature ventricular contractions, left and right bundled branchblocks, and paced beats [70].

Classification of ECG signals is performed by analyzing shape ofthe waves presents in the signal. Parameters of the shape of thewaves act as features for classification algorithms. Computationalrequirement of classification algorithms depends directly on thenumber of the features used and accuracy of classification dependson quality of the features. Thus, feature selection plays a promi-nent role in the classification of ECG signals. In the literature, manyapproaches have been proposed to select optimal features. Bashiret al. [70] calculated QRS, P and T waves morphological parametersas features to detect different arrhythmia. Then a parameter scorewas calculated for an adaptive selection of feature subset forparticular arrhythmia. Accordingly, there will be a different featureset for each arrhythmia, which enhances the accuracy, and at thesame time reduces the computational burden. While, Llamedo andMartinez [71] calculated interval features and morphologicalfeatures for classification of arrhythmia. Interval features werecalculated from R peaks, and morphological features were calcu-lated from three sources, R–R interval, 2-D vectocardiogram loopand DWT of the ECG signal. Then outliers form the feature setwere removed based on Kurtosis coefficients. Mar et al. [72]applied sequential forward floating search algorithm with a newcriterion function index. Drawback of the proposed method is thatin many cases the subset with highest criterion value has a verylarge number of features. Kamath [73] selected mean of Teagerenergy operator (TEO) in the time domain and frequency domainas features set. Key characteristic of the TEO is that it modelsenergy of the source that generated signal rather than the energyof the signal itself. Hence, any deviations in the regular rhythmicactivity of the heart get reflected in the TEO. Most of the abovealgorithms face the same challenge, requirement of a largenumber of the feature set. Large number of feature set is requiredfor diagnosis of the different types of diseases, but it results inlarge computational complexity. Another challenge is due tovariation in morphological descriptors of the heart signalwith time.

Since mathematical operators work in the time domain, theseare computationally efficient and hence consume low power.Mathematical morphological operators have been used [74] toextract structural information of the ECG signals. However, com-putational requirement increases as increment in order of theoperators. To optimize computation requirement, Zhang and Bae[8] proposed 1 dilation and 1 erosion based morphology operatorsets. However, effectiveness of these algorithms depends on theselection of three structural components of the operator, shape,length, and amplitude.

T wave delineation is crucial as prolongation of T wave to endof the T wave is associated with ventricular pre-arrhythmicity andsudden cardiac death. Noriega et al. [75] analyzed respirationeffect on T wave. Atrial fibrillation (AF) is associated with anincreased risk of cardiovascular and coronary artery disease,hypertension, etc. AF is typically diagnosed by analyzing irregular

RR intervals. Huang et al. [76] proposed an algorithm to classify AFby analyzing RR interval.

8.2.2. PhonocardiographyAs discussed in Section 4.2, PCG signal consists of four sound

components. Characteristics (intensity, frequency, and duration) ofthese sound components change due to the presence of CVDs.Additional murmur sounds may also be present in PCG signal dueto the presence of CVDs. Although PCG can indicate abnormalitiescaused by important CVDs, it is used extensively for diagnosis ofvalvular diseases as sound components are produced by thevalvular activity [10]. Valvular diseases cause systolic and diastolicmurmurs. Aortic stenosis (AS), pulmonary stenosis (PS), and mitralregurgitation (MR) cause systolic murmurs. On the other hand,diastolic murmurs occur due to aortic regurgitation (AR), mitralstenosis (MS) etc. Systolic murmurs, AS lies in the frequency bandof 120–250 Hz, PS lies in 200–250 Hz, and MR lies in 300–400 Hz.MR can be classified from other two systolic murmurs (PS and AS),as with wider duration and higher frequency band. S1 alsobecomes quieter than normal, in case of MR. Whereas PS causeslonger duration between aortic and pulmonary components of theS2, called as split S2. Splitting of S2 also may occur due to atrialseptal defect and right bundle branch block. Diastolic murmurs liein 100–250 Hz. MS causes mid-diastolic murmur with louder S1and causes a high frequency opening snap of 90–130 ms after theaortic component of S2. While AR is relatively louder than otherdiastolic murmurs. But more severe AR causes a lower intensitymurmur with longer duration. Two other sound components, S3and S4, in PCG signal rarely occur and may indicate abnormalities.Presence of S3 in a child is normal, while in adults, it representsdiastolic overload or cardiomyopathy [77]. S4 occurs just afteratrial contraction and it may be due to ventricular hypertrophy orpulmonary arterial hypertension [77]. Analysis of the soundcomponents and the murmurs leads towards to the classificationof the PCG signals.

Heart sound classification algorithms first partition the PCGsignal into S1, S2, systole, and diastole intervals, by emphasizingthem. To emphasize the heart signal components, envelop basedparameters such as Hilbert transform, Shannon energy, cardiacsound characteristic waveform (CSCW) and time–frequencydomain analysis such as STFT and wavelet have been presentedin the literature. After segmentation of components, classificationof PCG signal is performed by analyzing the characteristics of thesecomponents.

Envelope extraction based classification algorithms are able todetect fundamental heart sound components and to classify thesignal as normal or abnormal. Choi and Jiang [78] compared threeenvelope extraction algorithms, normalized average Shannonenergy; envelope information of Hilbert transform, and the CSCW.As shown in the results, CSCW gives a more uniform representa-tion of the fundamental components. The main challenge for theenvelope extraction based algorithm is the selection of the thresh-old value. A higher value of threshold missed the S1 and S2, whilethe lower value of threshold detects spurious components andinaccurate S1 and S2. To resolve this problem, Atbi et al. [79]proposed a two step thresholding scheme. In the first step thresh-old is selected to detect S1 and S2 and in the next step, to detectmurmurs. Envelop extraction based algorithms are computation-ally low complex. However, the performance of these algorithmsdepends on the morphology of the PCG signal. Furthermore, itbecomes difficult to detect S1 and S2, where murmurs are mergedwith them.

Frequency domain transformation techniques such as Fouriertransform, discrete cosine transform, and auto-regressive basedspectral analysis techniques provide frequency characteristics of

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–13 9

PCG signal components. However, time–frequency domain analy-sis is more suitable for PCG signal analysis due to the diagnosticsignificance of timing and frequency of PCG components. Time–frequency analysis of PCG signal has been done using short timeFourier transform (STFT) [10,80], wavelet transform [80,81]. Bou-tana et al. [10] classified murmurs from PCG sound components byanalyzing the Renyi marginal entropy of STFT coefficients. Renyimarginal entropy remains high for murmurs and low for soundcomponents. While, author [80] implement the PCG analysisalgorithm using STFT and wavelet on digital signal processingboard. In [81], first heart sound signal is segmented into intervalsassociated with cardiac cycle. Then intervals were groupedtogether based on similarities between their STFT coefficients. InSTFT, a trade off between time resolution and frequency resolutionarises. Increment in size of time window increases the frequencyresolution, but reduces time resolution and vice versa. Thus,selection of the optimal size of time window is crucial. For time–frequency analysis, wavelet offers a better compromise in terms ofresolution. Its main difference with STFT is that the size of thewindow is not constant. It varies in inverse proportion tofrequency in such a way that good time and poor frequencyresolution obtain at high frequencies while good frequency andpoor time resolution obtain at low frequencies.

PCG signals have been classified using artificial intelligencealgorithms such as Hidden Markov model [9] and neural network[82,83]. Extracted features from PCG signals using time andfrequency analysis tools such as wavelet are used as feature pointsfor these artificial intelligence techniques [82,83]. Use of themachine learning algorithms reduce tedious envelop analysisand its disadvantage in case of murmurs can be avoided, but atthe cost of having to prepare the training dataset. To prepare thetraining dataset for PCG signal, Ahlstrom et al. [84] proposed afeature subset selection algorithm from features of differentdomain, including Shannon energy, wavelet, fractal dimension,and recurrent quantification analysis.

PCG signal modelling is required to generate test data toanalyze efficacy of the developed algorithm. Modelling of PCGsignals has been done using exponential damped sinusoidal model[85], matching pursuit method [86]. These methods providecomplete parameterization of the signal, but require a largenumber of components. Whereas linear chirp signal modelling isnot suitable for PCG signal because components of PCG signal donot have a linear relationship with time. To achieve betteraccuracy, Xu et al. [87] proposed non-linear chirp signal modellingof the heart sound components.

8.2.3. SeismocardiographySeismocardiography (SCG) measures mechanical vibrations

produced by heart during each cardiac cycle. As discussed inSection 4.3, SCG signal is composed of many waves. Each wave isassociated with a particular event of the cardiac cycle. Thus,analysis of these waves provides diagnostic information relatedto the health of the heart. In [88], author studied the relationbetween the cardiac event position in SCG with ultrasound signaland showed SCG as an accurate indicator for cardiac events. ThusSCG signal can be used to detect cardiac cycle boundary, heart rate[89], heart rate variability [90]. SCG signals have been also used toobtain systolic blood pressure (SBP) [24]. It was shown that SBPhas the correlation with starting point of the SCG signal in thex-axis to the midpoint of the z-axis. However, SCG has been usedfor the heart monitoring purpose, but its sensitivity to motionnoise imposes limitation on its wide use.

Characterization of the relation between SCG signal and ECGsignal provides significant information related to heart function-ality. Wick et al. [88] analyzed relation between the R wave of ECG

and the AC wave of SCG signal. The R-AC period varies across twoindividual and also for the same person at different heart rates.This study strongly suggested the cardiac events also vary in thesame manner. Tavakolian et al. [91] analyzed period between the Rwave of the ECG and the AO wave of SCG to analyze the myocardialcontractility. This period is called as pre-ejection period (PEP),increment in the PEP indicates reduction in contract-ability ofmyocardial. Moreover, SCG has potential benefits over ECG such asbetter specificity and sensitivity for detection of coronary arterydiseases [92].

8.2.4. Photoplethysmography

PPG signal contains sufficient parameters to measure heartrate, arterial oxygen saturation, and information related torespiratory system [27,93,94]. As discussed in Section 4.4, PPGmeasures variation in intensity of light, reflected or transmitted,induced by variation in the amount of blood in blood vessels.Respiration information can be extracted using three vitalparameters: PPG amplitude, variation in SpO2, and respiratorysinus arrhythmia [27]. Now a day, pulse oximeters (variant of PPG)are being used extensively for heart monitoring [27,93–95]. Itmeasures multiple PPG signals at different wavelengths viz., red(660 nm) and infrared (940 nm). Pulse oximeters have been used forsleep apnea detection [27], pulse wave velocity calculation [93],hypoxia detection [94], and heart rate turbulence analysis [95]. Pulseoximeters have been developed as an in-ear sensor for cardiovascularmonitoring [27,94]. This setup of sensor could offer three importantadvantages: (1) comfortable to wear and hence, suitable for long-term monitoring, (2) the tight-fitting could reduce interference frommotion artifacts, and (3) robustness to conditions such as tempera-ture or skin perfusion.

However, PPG signals get contaminated primarily due toambient light, motion artifacts and other physiological process.To extract information from the contaminated PPG signals, Mad-hav et al. [96] proposed a multi scale principal component analysisbased algorithm. In this algorithm, noise suppression from the PPGsignals was achieved using wavelet decomposition and recon-struction. Selection of coefficients to reconstruct relatively cleansignal was done based on two measures, energy contribution level(ECL) and Kurtosis. After reconstruction of the clean signal,principal component (PC) analysis was performed to extractinformation about the respiratory system. Li and Warren [93]developed a sensor circuit in which photodetecters are radiallydistributed around the LED to increase the sensing area. This set-up improved the signal quality without filtering algorithm.Whereas, Stuban and Niwayama [97] analyzed optimal cornerfrequency of low pass filter for PPG signal. Setting the cornerfrequency to the fundamental frequency of the PPG signal resultedin decreased noise, and consequently, decreased standard devia-tion. de Haan and Jeanne [98] analyzed robustness of the chromi-nance based algorithms to separate motion induced distortionfrom rPPG signals in case of modest and vigorous motion.

9. Use of mobile

The latest generation of mobile phones (smartphones) isincreasingly used for health monitoring, due to their powerfulon-board computing capability, large memory, large screens andopen operating systems that encourage application development.Technical features of mobile phone including text messaging,camera, internet access, inbuilt sensors, make it an appropriateplatform for improving health care service [99]. Wireless technol-ogies, including GPRS, GSM, 3GSatelite, Wireless, Lan networks,have been used for wireless transmission of the heart signal [100].

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–1310

Mobile phones are also suitable platform to develop a heartmonitoring system because of (1) the widespread adoption ofphones, (2) peoples tendency to carry their phones with themeverywhere, and (3) context awareness features [99]. Furthermore,visible representation of the health status of the patient on mobileencourages to be attentive for health promoting behaviour.

Mobile phones are using for long term heart monitoring at homefor, both manner, on-site monitoring and off-site monitoring. Homemonitoring supports to reduce the rates of admission to hospital forchronic heart failure, improve the quality, reduce excessive traveltime and reduce the cost [101–103]. While long term monitoringsupports in early identification of deteriorations in patient conditionand symptom control, and timely intervention of a medical team[99,104]. Smartphones based software application can help cliniciansin identifying acute symptoms, decreasing unnecessary tests, tounderstand principles of disease diagnosis, and communicationfacility among clinicians [99,102]. In addition, mobile applicationshave been used for remote coaching, reminder to patient forappointments and health related information, public health research,primary care, emergency care, health information for self, drugreference, medical training, to encourage for primary care check-up,etc. Mobile phone based health monitoring system have beendiscussed in [102]. Some of them for heart monitoring are as follows:(1) cardiomobile is comprised of a heart and activity monitor, singlelead ECG, GPS receiver, and programmed smartphone. The smart-phone sends ECG rate, walking speed, heart rate, elapsed distance,and patient location to a secure server for real-time monitoring bya qualified exercise scientist. (2) Pulmonary rehabilitation is anapplication for chronic obstructive pulmonary disease (COPD) reha-bilitation and self-management, developed for smartphones. (3) mVi-sum is a specialized application for cardiology communications thatmonitor ECG data, alarm the user in abnormal case, and transmitdata to clinician. (4) iCPR is a cardiopulmonary resuscitation (CPR)training application. This application measures the chest compres-sion rate and gives audiovisual feedback, improving the performanceof chest compression by helping the user to achieve the correct chestcompression rate. Another smart-phone based health monitoringsystem, BioSign, is reviewed in [104]. BioSign system alerts thepatient in case of abnormality. It represents the health status of thepatient as patient status index (PSI), which is calculated based on fivevital parameters: heart rate, breathing rate, blood pressure, arterialoxygen saturation, and skin temperature.

To enhance the user acceptability, Scully et al. [105] proposeda reflection photoplethysmography based on imaging by mobilephones. In this approach, the palmer side of the left index fingerwas placed over the camera lens of mobile with its flash turned on.Then variation of intensity in captured video indicates the heartbeat. This approach does not require any extra hardware. Howeversensitivity of device get affects due to motion and pressurevariation of the finger. Another approach proposed by Poh et. al[106] integrates reflective photo-diode into earphone which areunobtrusive, and low in size and weight. Then acquired signal wassent to mobile phone for monitoring purpose.

Major challenges for smartphone-based health care systeminclude cost, network bandwidth, battery power efficiency, smallscreen size, computer viruses, etc. [102]. Other issue is that it mustbe seamless and autonomous in its operation. It would bebeneficial for those who are non-familiar with the technology.

10. Conclusion

In this paper we have provided a detailed review of recentadvancements for portable heart monitoring systems. We consid-ered ECG, PCG, PPG, and SCG based systems and portability of suchsystems are feasible due to use of light weight and small size body

sensors. ECG has diagnostic superiority of important CVDs ascompared to other portable systems. However, it has limitationsin long term monitoring of heart due to the requirement of skincontact of electrodes and the use of gel. PCG also containssufficient diagnostic features, but it is vulnerable to motion noise.Although, PCG has advantage over ECG in terms of easy to operate.This feature makes PCG useful in scenarios where the number ofcardiac experts is low. PPG has found its wider use for long termmonitoring as it is more comfortable in terms of wearability thanother systems. But it has limited diagnostic features related toimportant CVDs as it acquires signals far from the heart. SCG hasnot been used for important cardiovascular diseases. However, itprovides some additional diagnostic parameters than those obtainfrom PCG. Its higher sensitivity to motion noise than PCG and PPGlimits its wider use as portable system.

Recently, focus on at-home monitoring of heart is increasing forlong term monitoring, which is advantageous in cases where heartabnormalities are detected and suspicious to have CVDs. It leads toincreasing research in development of portable systems havingfeatures of low power consumption, signal transmission capability,and unobtrusiveness. Computationally efficient algorithms aredeveloped that ensures low power consumption, Signal transmis-sion capability ensures remote monitoring, and unobtrusiveness ofthe system helps in its use for long term monitoring. Powerconsumption and user friendly connection process are main issuesof signal transmission module. Advanced compression algorithmsfor heart signals will reduce power consumption of this module.To increase the unobtrusiveness, in case of ECG and PPG, research-ers have proposed sensors that do not require skin contact. Use ofthe cellular phones for monitoring purpose eliminates extra hard-ware possession and hence increases user convenience. Nowadays,cellular phones are equipped with considerable computationalpower and hence can play a crucial role for heart monitoring.

Conflict of interest statement

None declared.

References

[1] W.H. Organization, Global Status Report on Noncommunicable Diseases2010. ⟨http://www.who.int/nmh/publications/ncd_report2010/en/⟩, 2010(Online, accessed 19.07.13).

[2] A.H. Association, Heart Disease and Stroke Statistic 2012 Update. ⟨circ.ahajournals.org/content/125/1/e2⟩, 2012 (Online, accessed 19.07.13).

[3] Center for Disease Control and Prevention, State Specific Mortality FromSudden Cardiac Death, ⟨www.cdc.gov/heartdisease/facts.htm⟩, 2002 (Online,accessed 19.07.13).

[4] J. Choi, J. Park, J. Chung, B. Min, An intelligent remote monitoring system forartificial heart, IEEE Trans. Inf. Technol. Biomed. 9 (December) (2005)564–573.

[5] WHO, Global Health Observatory, ⟨http://www.who.int/gho/health_financing/en/index.html⟩, 2011 (Online, accessed 19.07.13).

[6] E. Nemati, M. Deen, T. Mondal, A wireless wearable ecg sensor for long-termapplications, IEEE Commun. Mag. 50 (January) (2012) 36–43.

[7] J. Lee, D. McManus, S. Merchant, K. Chon, Automatic motion and noiseartifact detection in holter ecg data using empirical mode decomposition andstatistical approaches, IEEE Trans. Biomed. Eng. 59 (June) (2012) 1499–1506.

[8] C. Zhang, T.-W. Bae, Vlsi friendly ecg qrs complex detector for body sensornetworks, IEEE J. Emerg. Sel. Top. Circuits Syst. 2 (March) (2012) 52–59.

[9] C. Kwak, O.W. Kwon, Cardiac disorder classification by heart sound signalsusing murmur likelihood and hidden Markov model state likelihood, IETSignal Process. 6 (4) (2012) 326–334.

[10] D. Boutana, M. Benidir, B. Barkat, Segmentation and identification of somepathological phonocardiogram signals using time–frequency analysis, IETSignal Process. 5 (September) (2011) 527–537.

[11] Deloitte, Cardiovascular Diseases in India Challenges and Way Ahead, ⟨www.deloitte.com/in⟩, 2011 (Online, accessed 19.07.13).

[12] V.K. Nelso Fausto, Abul Abbas, Robbins and Cotran Pathologic Basis ofDisease. Saunders, 2004.

[13] M. Clinic, Heart Disease, ⟨www.mayoclinic.com/health/heart-disease/DS01120/DSECTION=causes⟩, 2013 (Online; accessed 19.07.13).

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–13 11

[14] J.G. Webster, Medical Instrum.: Appl. Des., Wiley, New York, 1995.[15] X. Liu, Y. Zheng, M. Phyu, F. Endru, V. Navaneethan, B. Zhao, An ultra-low

power ecg acquisition and monitoring asic system for wban applications,IEEE J. Emerg. Sel. Top. Circuits Syst. 2 (March) (2012) 60–70.

[16] L. Searle, A. Kirkup, Direct comparison of wet, dry and insulating bioelectricrecording electrodes, Physiol. Meas. 21 (May) (2000) 271–283.

[17] H.-C. Jung, J.-H. Moon, D.-H. Baek, J.-H. Lee, Y.-Y. Choi, J.-S. Hong, S.-H. Lee,Cnt/pdms composite flexible dry electrodesfor long-term ecg monitoring,IEEE Trans. Biomed. Eng. 59 (May) (2012) 1472–1479.

[18] J. Gomez-Clapers, R. Casanella, A fast and easy-to-use ecg acquisition andheart rate monitoring system using a wireless steering wheel, IEEE Sens. J. 12(March) (2012) 610–616.

[19] L.S.A. Aleksandrowicz, Wireless and non-contact ecg measurement systemthe aachen smartchair, Acta Polytech. 2 (2007) 6871.

[20] H.J. Baek, G.S. Chung, K.K. Kim, K.S. Park, A smart health monitoring chair fornonintrusive measurement of biological signals, IEEE Trans. Inf. Technol.Biomed. 16 (January) (2012) 150–158.

[21] Y.G. Lim, K.K. Kim, K.-S. Park, Ecg recording on a bed during sleep withoutdirect skin-contact, IEEE Trans. Biomed. Eng. 54 (4) (2007) 718–725.

[22] G. Cho, K. Jeong, M.J. Paik, Y. Kwun, M. Sung, Performance evaluation oftextile-based electrodes and motion sensors for smart clothing, IEEE Sens.J. 11 (December) (2011) 3183–3193.

[23] J. Zanetti, D. Salerno, Seismocardiography: a technique for recording pre-cordial acceleration, in: Proceedings of the Fourth Annual IEEE Symposiumon Computer-Based Medical Systems, 1991, pp. 4–9.

[24] M. Imtiaz, R. Shrestha, T. Dhillon, K. Yousuf, B. Saeed, A. Dinh, K. Wahid,Correlation between seismocardiogram and systolic blood pressure, in: 26thAnnual IEEE Canadian Conference on Electrical and Computer Engineering(CCECE), 2013, pp. 1–4.

[25] M. Di Rienzo, E. Vaini, P. Castiglioni, G. Merati, P. Meriggi, G. Parati, A. Faini,F. Rizzo, Wearable seismocardiography: towards a beat-by-beat assessmentof cardiac mechanics in ambulant subjects, Autonom. Neurosci. 178 (1–2)(2013) 50–59. http://dx.doi.org/10.1016/j.autneu.2013.04.005.

[26] R. Krishnan, B. Natarajan, S. Warren, Two-stage approach for detection andreduction of motion artifacts in photoplethysmographic data, IEEE Trans.Biomed. Eng. 57 (8) (2010) 1867–1876.

[27] B. Venema, J. Schiefer, V. Blazek, N. Blanik, S. Leonhardt, Evaluatinginnovative in-ear pulse oximetry for unobtrusive cardiovascular and pul-monary monitoring during sleep, IEEE J. Transl. Eng. Health Med. 1 (2013)1–8.

[28] K. Sweeney, T. Ward, S. McLoone, Artifact removal in physiological signalsx2014;practices and possibilities, IEEE Trans. Inf. Technol. Biomed. 16 (May)(2012) 488–500.

[29] M. Huelsbusch, V. Blazek, Contactless mapping of rhythmical phenomena intissue perfusion using ppgi, Proc. SPIE 4683 (2002) 110–117.

[30] R. Rieger, Variable-gain, low-noise amplification for sampling front ends,IEEE Trans. Biomed. Circuits Syst. 5 (June) (2011) 253–261.

[31] E. Spinelli, N. Martinez, M. Mayosky, A transconductance driven-right-legcircuit, IEEE Trans. Biomed. Eng. 46 (12) (1999) 1466–1470.

[32] C. Bachmann, M. Ashouei, V. Pop, M. Vidojkovic, H. Groot, B. Gyselinckx,Low-power wireless sensor nodes for ubiquitous long-term biomedicalsignal monitoring, IEEE Commun. Mag. 50 (January) (2012) 20–27.

[33] H. Tang, T. Li, T. Qiu, Cardiac cycle detection for heart sound signal based oninstantaneous cycle frequency, in: 2011 4th International Conference onBiomedical Engineering and Informatics (BMEI), vol. 2, 2011, pp. 676–679.

[34] D.-H. Shih, H.-S. Chiang, B. Lin, S.-B. Lin, An embedded mobile ecg reasoningsystem for elderly patients, IEEE Trans. Inf. Technol. Biomed. 14 (May) (2010)854–865.

[35] R. Dilmaghani, H. Bobarshad, M. Ghavami, S. Choobkar, C. Wolfe, Wirelesssensor networks for monitoring physiological signals of multiple patients,IEEE Trans. Biomed. Circuits Syst. 5 (August) (2011) 347–356.

[36] S.-K. Chen, T. Kao, C.-T. Chan, C.-N. Huang, C.-Y. Chiang, C.-Y. Lai, T.-H. Tung,P.-C. Wang, A reliable transmission protocol for zigbee-based wirelesspatient monitoring, IEEE Trans. Inf. Technol. Biomed. 16 (January) (2012)6–16.

[37] J. Morak, H. Kumpusch, D. Hayn, R. Modre-Osprian, G. Schreier, Design andevaluation of a telemonitoring concept based on nfc-enabled mobile phonesand sensor devices, IEEE Trans. Inf. Technol. Biomed. 16 (January) (2012)17–23.

[38] E. Candes, M. Wakin, An introduction to compressive sampling, IEEE SignalProcess. Mag. 25 (March) (2008) 21–30.

[39] A. Dixon, E. Allstot, D. Gangopadhyay, D. Allstot, Compressed sensing systemconsiderations for ecg and emg wireless biosensors, IEEE Trans. Biomed.Circuits Syst. 6 (April) (2012) 156–166.

[40] H. Mamaghanian, N. Khaled, D. Atienza, P. Vandergheynst, Compressedsensing for real-time energy-efficient ecg compression on wireless bodysensor nodes, IEEE Trans. Biomed. Eng. 58 (September) (2011) 2456–2466.

[41] W.-C. Kao, W.-H. Chen, C.-K. Yu, C.-M. Hong, S.-Y. Lin, Portable real-timehomecare system design with digital camera platform, IEEE Trans. Consum.Electron. 51 (4) (2005) 1035–1041.

[42] J. Martinez-Alajarin, R. Ruiz-Merino, Wavelet and wavelet packet compres-sion of phonocardiograms, Electron. Lett. 40 (August) (2004) 1040–1041.

[43] A. Kadrolkar, R. Gao, R. Yan, W. Gong, Variable-word-length coding forenergy-aware signal transmission, IEEE Trans. Instrum. Meas. 61 (April)(2012) 850–864.

[44] A. Sandryhaila, S. Saba, M. Puschel, J. Kovacevic, Efficient compression of qrscomplexes using hermite expansion, IEEE Trans. Signal Process. 60 (Febru-ary) (2012) 947–955.

[45] A. Bendifallah, R. Benzid, M. Boulemden, Improved ecg compression methodusing discrete cosine transform, Electron. Lett. 47 (January) (2011) 87–89.

[46] B. Rajoub, An efficient coding algorithm for the compression of ecg signalsusing the wavelet transform, IEEE Trans. Biomed. Eng. 49 (4) (2002)355–362.

[47] L. Sharma, S. Dandapat, A. Mahanta, Multichannel ecg data compressionbased on multiscale principal component analysis, IEEE Trans. Inf. Technol.Biomed. 16 (July) (2012) 730–736.

[48] H. Wang, D. Peng, W. Wang, H. Sharif, H. hwa Chen, A. Khoynezhad,Resource-aware secure ecg healthcare monitoring through body sensornetworks, IEEE Wirel. Commun. 17 (February) (2010) 12–19.

[49] H. Kim, Y. Kim, H.-J. Yoo, A low cost quadratic level ecg compressionalgorithm and its hardware optimization for body sensor network system,in: 30th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBS 2008), 2008, pp. 5490–5493.

[50] T. Ma, P. Shrestha, M. Hempel, D. Peng, H. Sharif, H.-H. Chen, Assurance ofenergy efficiency and data security for ecg transmission in basns, IEEE Trans.Biomed. Eng. 59 (April) (2012) 1041–1048.

[51] A. Alvarado, C. Lakshminarayan, J. Principe, Time-based compression andclassification of heartbeats, IEEE Trans. Biomed. Eng. 59 (June) (2012)1641–1648.

[52] T.-H. Kim, S.-Y. Kim, J.-H. Kim, B.-J. Yun, K.-H. Park, Curvature based ecgsignal compression for effective communication on wpan, J. Commun. Netw.14 (February) (2012) 21–26.

[53] B. Atakan, O. Akan, S. Balasubramaniam, Body area nanonetworks withmolecular communications in nanomedicine, IEEE Commun. Mag. 50 (Jan-uary) (2012) 28–34.

[54] S.H. Lee, S.M. Jung, C.K. Lee, K.S. Jeong, G. Cho, S.K. Yoo, Wearable ecgmonitoring system using conductive fabrics and active electrodes, Hum.-Comput. Interact. Ambient, Ubiquitous Intell. Interact. 5612 (2009) 778–783.

[55] G. Gargiulo, P. Bifulco, M. Cesarelli, M. Ruffo, M. Romano, R.A. Calvo, C. Jin,A. Schaik, An ultra-high input impedance ecg amplifier for long-termmonitoring of athletes, Med. Devices (Auckl) 3 (July) (2010) 1–9.

[56] C. Park, P. Chou, Y. Bai, R. Matthews, A. Hibbs, An ultra-wearable, wireless,low power ecg monitoring system, in: Biomedical Circuits and SystemsConference (BioCAS 2006), IEEE, November 2006, pp. 241–244.

[57] S.L.A. Aleksandrowicz, Wireless and non-contact ecg measurement systemthe aachen smartchair, Acta Polytech. 47 (4) (2007).

[58] B. Massot, N. Baltenneck, C. Gehin, A. Dittmar, E. McAdams, Emosense: anambulatory device for the assessment of ans activity x2014; application inthe objective evaluation of stress with the blind, IEEE Sens. J. 12 (March)(2012) 543–551.

[59] S.M. Debbal, F. Bereksi-Reguig, Filtering and classification of phonocardio-gram signals using wavelet transform, J. Med. Eng. Technol. 32 (February)(2008) 53–65.

[60] A. Almasi, M.B. Shamsollahi, L. Senhadji, Bayesian denoising framework ofphonocardiogram based on a new dynamical model, IRBM 34 (3) (2013)214–225.

[61] S.R. Messer, J. Agzarian, D. Abbott, Optimal wavelet denoising for phono-cardiograms, Microelectron. J. 32 (12) (2001) 931–941.

[62] M. Sabarimalai Manikandan, K. Soman, Robust heart sound activity detectionin noisy environments, Electron. Lett. 46 (2010) 1100–1102.

[63] H. Tang, T. Li, Y. Park, T. Qiu, Separation of heart sound signal from noise injoint cycle frequency x2013;time x2013; frequency domains based on fuzzydetection, IEEE Trans. Biomed. Eng. 57 (October) (2010) 2438–2447.

[64] H. Tang, T. Li, T. Qiu, Noise and disturbance reduction for heart sounds incycle-frequency domain based on nonlinear time scaling, IEEE Trans.Biomed. Eng. 57 (2) (2010) 325–333.

[65] M. Rahman, R. Shaik, D. Reddy, Efficient and simplified adaptive noisecancelers for ecg sensor based remote health monitoring, IEEE Sens. J. 12(March) (2012) 566–573.

[66] S. Chen, M. Bolic, V. Groza, H. Dajani, I. Batkin, S. Rajan, Extraction ofbreathing signal and suppression of its effects in oscillometric bloodpressure measurement, IEEE Trans. Instrum. Meas. 60 (May) (2011)1741–1750.

[67] T. Li, T. Qiu, H. Tang, Optimum heart sound signal selection based on thecyclostationary property, Comput. Biol. Med. 43 (6) (2013) 607–612.

[68] F. Beritelli, A. Spadaccini, Heart sounds quality analysis for automatic cardiacbiometry applications, in: First IEEE International Workshop on InformationForensics and Security (WIFS 2009), December 2009, pp. 61–65.

[69] H. Naseri, M. Homaeinezhad, H. Pourkhajeh, Noise/spike detection inphonocardiogram signal as a cyclic random process with non-stationaryperiod interval, Comput. Biol. Med. 43 (9) (2013) 1205–1213.

[70] M. Bashir, D.G. Lee, M. Li, J.-W. Bae, H. Shon, M.C. Cho, K.H. Ryu, Triggerlearning and ecg parameter customization for remote cardiac clinical careinformation system, IEEE Trans. Inf. Technol. Biomed. 16 (July) (2012)561–571.

[71] M. Llamedo, J. Martinez, Heartbeat classification using feature selectiondriven by database generalization criteria, IEEE Trans. Biomed. Eng. 58 (3)(2011) 616–625.

[72] T. Mar, S. Zaunseder, J. Martnez, M. Llamedo, R. Poll, Optimization of ecgclassification by means of feature selection, IEEE Trans. Biomed. Eng. 58 (8)(2011) 2168–2177.

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–1312

[73] C. Kamath, Ecg beat classification using features extracted from teagerenergy functions in time and frequency domains, IET Signal Process. 5 (6)(2011) 575–581.

[74] F. Zhang, Y. Lian, Qrs detection based on multiscale mathematical morphol-ogy for wearable ecg devices in body area networks, IEEE Trans. Biomed.Circuits Syst. 3 (August) (2009) 220–228.

[75] M. Noriega, J. Martinez, P. Laguna, R. Bailon, R. Almeida, Respiration effect onwavelet-based ecg t-wave end delineation strategies, IEEE Trans. Biomed.Eng. 59 (7) (2012) 1818–1828.

[76] C. Huang, S. Ye, H. Chen, D. Li, F. He, Y. Tu, A novel method for detection of thetransition between atrial fibrillation and sinus rhythm, IEEE Trans. Biomed.Eng. 58 (4) (2011) 1113–1119.

[77] F. Beritelli, S. Serrano, Biometric identification based on frequency analysis ofcardiac sounds, IEEE Trans. Inf. Forens. Secur. 2 (September) (2007) 596–604.

[78] S. Choi, Z. Jiang, Comparison of envelope extraction algorithms for cardiacsound signal segmentation, Expert Syst. Appl. 34 (2) (2008) 1056–1069.

[79] A. Atbi, S.M. Debbal, F. Meziani, A. Meziane, Separation of heart sounds andheart murmurs by hilbert transform envelogram, J. Med. Eng. Technol. 37 (6)(2013) 375–387 (doi: 23875931).

[80] D. Balasubramaniam, D. Nedumaran, Efficient computation of phonocardio-graphic signal analysis in digital signal processor based system, Int.J. Comput. Theory Eng. 2 (4) (2010) 660–664.

[81] Z. Syed, D. Leeds, D. Curtis, F. Nesta, R. Levine, J. Guttag, A framework for theanalysis of acoustical cardiac signals, IEEE Trans. Biomed. Eng. 54 (4) (2007)651–662.

[82] T.R. Reed, N.E. Reed, P. Fritzson, Heart sound analysis for symptom detectionand computer-aided diagnosis, Simul. Model. Pract. Theory 12 (2) (2004)129–146 (Advances in modelling and simulation in biology and medicine).

[83] D. BarschdorfF, U. Femmer, E. Trowitzsch, Automatic phonocardiogram signalanalysis in infants based on wavelet transforms and artificial neural net-works, in: Computers in Cardiology, 1995, pp. 753–756.

[84] C. Ahlstrom, P. Hult, P. Rask, J.-E. Karlsson, E. Nylander, U. Dahlstrm, P. Ask,Feature extraction for systolic heart murmur classification, Ann. Biomed.Eng. 34 (11) (2006) 1666–1677.

[85] H. Sava, P. Grant, J.T.E. McDonnell, Spectral characterization and classificationof Carpentier–Edwards heart valves implanted in the aortic position, IEEETrans. Biomed. Eng. 43 (10) (1996) 1046–1048.

[86] X. Zhang, L. Durand, L. Senhadji, H. Lee, J.-L. Coatrieux, Analysis-synthesis ofthe phonocardiogram based on the matching pursuit method, IEEE Trans.Biomed. Eng. 45 (August) (1998) 962–971.

[87] J. Xu, L. Durand, P. Pibarot, Nonlinear transient chirp signal modeling of theaortic and pulmonary components of the second heart sound, IEEE Trans.Biomed. Eng. 47 (October) (2000) 1328–1335.

[88] C. Wick, J.-J. Su, J. McClellan, O. Brand, P. Bhatti, A. Buice, A. Stillman, X. Tang,S. Tridandapani, A system for seismocardiography-based identification ofquiescent heart phases: implications for cardiac imaging, IEEE Trans. Inf.Technol. Biomed. 16 (5) (2012).

[89] H. Nguyen, J. Zhang, Y.-H. Nam, Timing detection and seismocardiographywaveform extraction, in: 2012 Annual International Conference of the IEEEEngineering in Medicine and Biology Society (EMBC), 2012, pp. 3553–3556.

[90] J. Ramos-Castro, J. Moreno, H. Miranda-Vidal, M. Garcia-Gonzalez,M. Fernandez-Chimeno, G. Rodas, L. Capdevila, Heart rate variability analysisusing a seismocardiogram signal, in: 2012 Annual International Conferenceof the IEEE Engineering in Medicine and Biology Society (EMBC), 2012,pp. 5642–5645.

[91] K. Tavakolian, G. Portacio, N. Tamddondoust, G. Jahns, B. Ngai, G. Dumont, A.Blaber, Myocardial contractility: a seismocardiography approach, in: 2012Annual International Conference of the IEEE Engineering in Medicine andBiology Society (EMBC), 2012, pp. 3801–3804.

[92] R.A. Wilson, V.S. Bamrah, J. L. Jr., M. Schwaiger, J. Morganroth, Diagnosticaccuracy of seismocardiography compared with electrocardiography for the

anatomic and physiologic diagnosis of coronary artery disease duringexercise testing, Am. J. Cardiol. 71 (7) (1993) 536–545.

[93] K. Li, S. Warren, A wireless reflectance pulse oximeter with digital baselinecontrol for unfiltered photoplethysmograms, IEEE Trans. Biomed. CircuitsSyst. 6 (3) (2012) 269–278.

[94] B. Venema, N. Blanik, V. Blazek, H. Gehring, A. Opp, S. Leonhardt, Advances inreflective oxygen saturation monitoring with a novel in-ear sensor system:results of a human hypoxia study, IEEE Trans. Biomed. Eng. 59 (7) (2012)2003–2010.

[95] E. Gil, P. Laguna, J. Martinez Cortes, O. Barquero-Perez, A. Garcia-Alberola,L. Sornmo, Heart rate turbulence analysis based on photoplethysmography,IEEE Trans. Biomed. Eng. PP(99) (2013) 1–1.

[96] K. Madhav, M. Ram, E. Krishna, N. Komalla, K. Reddy, Robust extraction ofrespiratory activity from ppg signals using modified mspca, IEEE Trans.Instrum. Meas. 62 (5) (2013) 1094–1106.

[97] N. Stuban, M. Niwayama, Optimal filter bandwidth for pulse oximetry, Rev.Sci. Instrum. 83 (10) (2012), pp. 104708–104708-5.

[98] G. de Haan, V. Jeanne, Robust pulse-rate from chrominance-based rppg, IEEETrans. Biomed. Eng. PP(99) (2013) 1–1.

[99] P. Klasnja, W. Pratt, Healthcare in the pocket: mapping the space of mobile-phone health interventions, J. Biomed. Inf. 45 (1) (2012) 184–198.

[100] E. Kyriacou, C. Pattichis, M. Pattichis, An overview of recent health caresupport systems for eemergency and mhealth applications, in: AnnualInternational Conference of the IEEE Engineering in Medicine and BiologySociety (EMBC 2009), September 2009, pp. 1246–1249.

[101] A. Martnez, E. Everss, J.L. Rojo-lvarez, D.P. Figal, A. Garca-Alberola,A systematic review of the literature on home monitoring for patients withheart failure, J. Telemed. Telecare 12 (5) (2006) 234–241.

[102] A.S. Mosa, I. Yoo, L. Sheets, A systematic review of healthcare applications forsmartphones, BMC Med. Inf. Decis. Mak. 12 (1) (2012) 67.

[103] A.G. Ekeland, A. Bowes, S. Flottorp, Effectiveness of telemedicine: a systema-tic review of reviews, Int. J. Med. Inf. 79 (11) (2010) 736–771.

[104] L. Tarassenko, A. Hann, D. Young, Integrated monitoring and analysis for earlywarning of patient deterioration, Br. J. Anaesth. 97 (1) (2006) 64–68.

[105] C. Scully, J. Lee, J. Meyer, A. Gorbach, D. Granquist-Fraser, Y. Mendelson,K. Chon, Physiological parameter monitoring from optical recordings with amobile phone, IEEE Trans. Biomed. Eng. 59 (February) (2012) 303–306.

[106] M.-Z. Poh, K. Kim, A. Goessling, N. Swenson, R. Picard, Cardiovascularmonitoring using earphones and a mobile device, IEEE Pervasive Comput.11 (4) (2012) 18–26.

Puneet Kumar Jain was born in Rajasthan, India, in 1988. He received his bachelorof technology degree in computer science and engineering from the University ofRajasthan in 2009 and master of technology degree in information system fromDelhi Technological University in 2011. He is currently a research scholar at Indianinstitute of technology Jodhpur. His research interest includes image processing,biometric system and biomedical signal processing.

Anil Kumar Tiwari received the master of technology degree in 2001 from theIndian institute of technology Kharagpur and the Ph.D. degree in electronics andcommunication from the Indian institute of technology Kharagpur in 2005. He iscurrently an assistant professor at Indian institute of technology Jodhpur. Hisresearch interest are in Image Processing, Video Processing and Signal Processingapplication in Bio-Medical.

P.K. Jain, A.K. Tiwari / Computers in Biology and Medicine 54 (2014) 1–13 13