1538 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 5, MAY 2014Unobtrusive Sensing and Wearable Devicesfor Health InformaticsYa-Li Zheng, Student Member, IEEE, Xiao-Rong Ding, Carmen Chung Yan Poon, Benny Ping Lai Lo, Heye Zhang,Xiao-Lin Zhou, Guang-Zhong Yang, Fellow, IEEE, Ni Zhao, and Yuan-Ting Zhang*, Fellow, IEEEAbstractThe aging population, prevalence of chronic diseases,and outbreaks of infectious diseases are some of the major chal-lenges of our present-day society. To address these unmet health-careneeds, especiallyfortheearlypredictionandtreatmentofmajor diseases, health informatics, which deals with the acquisi-tion, transmission, processing, storage, retrieval, and use of healthinformation, has emerged as an active area of interdisciplinary re-search. In particular, acquisition of health-related information byunobtrusive sensing and wearable technologies is considered as acornerstone in health informatics. Sensors can be weaved or inte-grated into clothing, accessories, and the living environment, suchthat health information can be acquired seamlessly and pervasivelyin daily living. Sensors can even be designed as stick-on electronictattoos or directlyprintedonto humanskin to enable long-termhealth monitoring. This paper aims to provide an overview of fouremergingunobtrusiveandwearabletechnologies, whicharees-sential totherealizationofpervasivehealthinformationacqui-sition, including:1)unobtrusivesensingmethods, 2)smarttex-tiletechnology, 3) exible-stretchable-printableelectronics, and4)sensorfusion, andthentoidentifysomefuturedirectionsofresearch.IndexTermsBodysensornetwork, exibleandstretchableelectronics, health informatics, sensor fusion, unobtrusive sensing,wearable devices.Manuscript received October 15, 2013; revised January 11, 2014 and Febru-ary 14, 2014; accepted February 18, 2014. Date of publication March 5, 2014;dateofcurrentversionApril17, 2014. Thisworkwassupportedinpartbythe National Basic Research Program 973 (Grant 2010CB732606), the Guang-dong Innovation Research Team Fund for Low-cost Healthcare Technologies inChina, the External Cooperation Program of the Chinese Academy of Sciences(Grant GJHZ1212), the Key Lab for Health Informatics of Chinese Academyof Sciences, andtheNational Natural ScienceFoundationof China(Grant81101120). Asterisk indicates corresponding author.Y.-L. ZhengandX.-R. DingarewiththeDepartment of ElectronicEn-gineering, Joint ResearchCentrefor Biomedical Engineering, TheChineseUniversity of Hong Kong, Hong Kong (e-mail: ylzheng@ee.cuhk.edu.cn; xrd-ing@ee.cuhk.edu.hk).C. C. Y. Poon is with the Department of Surgery, The Chinese University ofHong Kong, Hong Kong (e-mail: cpoon@surgery.cuhk.edu.hk).B. P. L. Lo and G.-Z. Yang are with the Department of Computing, ImperialCollege London, London, SW7 2AZ, U.K. (e-mail: benny.lo@imperial.ac.uk;g.z.yang@imperial.ac.uk).H. Zhang and X.-L. Zhou are with the Institute of Biomedical and Health Engi-neering, Chinese Academy of Sciences (CAS), SIAT, Shenzhen, China, and alsowith the Key Laboratory for Health Informatics of the Chinese Academy of Sci-ences (HICAS), SIAT, Shenzhen 518055, China (e-mail: hy.zhang@siat.ac.cn;xl.zhou@siat.ac.cn).N. Zhao is with the Department of Electronic Engineering, The Chinese Uni-versity of Hong Kong, Hong Kong (e-mail: nzhao@ee.cuhk.edu.hk).Y.-T. ZhangiswiththeDepartmentofElectronicEngineering, JointRe-searchCentreforBiomedicalEngineering, TheChineseUniversityofHongKong, HongKong, andalsowiththe KeyLaboratory forHealthInformaticsof the Chinese Academy of Sciences (HICAS), SIAT, Shenzhen 518055, China(e-mail: ytzhang@ee.cuhk.edu.hk).Color versions of one or more of the gures in this paper are available onlineat http://ieeexplore.ieee.org.Digital Object Identier 10.1109/TBME.2014.2309951I. INTRODUCTIONGLOBALhealthcaresystems arestrugglingwithagingpopulation, prevalenceofchronicdiseases, andtheac-companyingrisingcosts[1].Inresponsetothesechallenges,researchers have beenactivelyseekingfor innovative solu-tions andnewtechnologies that couldimprove the qualityof patient care meanwhile reduce the cost of care throughearly detection/intervention and more effective disease/patientmanagement. It is envisaged that the future healthcare systemshould be preventive, predictive, preemptive, personalized, per-vasive, participatory, patient-centered, and precise, i.e., p-healthsystem.Healthinformatics, whichisanemerginginterdisciplinaryareatoadvancep-health, mainlydealswiththeacquisition,transmission,processing, storage, retrieval, anduseofdiffer-ent typesofhealthandbiomedical information[2]. Thetwomain acquisition technologies of health information are sensingand imaging. This paper focuses only on sensing technologiesand reviews the latest developments in unobtrusive sensing andwearabledevicesfor continuoushealthmonitoring. Lookingback in history, it is not surprised to notice that innovation inthis area is closely coupled with the advancements in electronics.Usingelectrocardiogram(ECG)deviceasanexample,Fig.1illustrates the evolution of sensing technologies, where the coretechnologies of these devices have evolved from water bucketsand bulky vacuum tubes, bench-top, and portable devices withdiscretetransistors, totherecent clothingandsmall gadgetsbased wearable devices with integrated circuits [3]. In the fu-ture, it may evolve into exible and stretchable wearable deviceswith carbon nanotube (CNT)/graphene/organic electronics [4].There is a clear trend that the devices are getting smaller, lighter,and less obtrusive and more comfortable to wear.Although physiological measurement devices have beenwidelyusedinclinicalsettingsformanyyears,someuniquefeaturesof unobtrusiveandwearabledevicesduetothere-cent advances insensing, networkinganddatafusionhavetransformedthewaythattheywereusedin. First, withtheirwireless connectivity together with the widely available inter-netinfrastructure, thedevicescanprovidereal-timeinforma-tionandfacilitatetimelyremoteinterventiontoacuteeventssuchasstroke,epilepsyandheartattack,particularlyinruralor otherwise underserved areas where expert treatment may beunavailable. In addition, for healthy population, unobtrusive andwearable monitoring can provide detailed information regard-ing their health and tness, e.g., via mobile phone or exibledisplays, such that they can closely track their wellbeing, whichwill not only promote active and healthy lifestyle, but also allow0018-9294 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publicationsstandards/publications/rights/index.html for more information.ZHENG et al.: UNOBTRUSIVE SENSING AND WEARABLE DEVICES FOR HEALTH INFORMATICS 1539Fig. 1. Timelines of medical devices for ECG measurement with the evolution of electronic technology.detection of any health risk and facilitate the implementation ofpreventive measures at an earlier stage.Theobjectivesofthispaperaretoprovideanoverviewofunobtrusivesensingandwearablesystemswithparticularfo-cusonemergingtechnologies, andalsotoidentifythemajorchallenges relatedtothis areaof research. Therest of thispaper isorganisedasfollows: SectionII will discussunob-trusivesensingmethodsandsystems.SectionIIIwillpresentthe recent advances andcore technologies of wearable de-vices and will also highlight the latest developments in smarttextile technology and exible-stretchable electronics. SectionIVwill discussdatafusionmethodsfor sensinginformaticsaswell astheimpactsof bigdatainhealthcare. SectionVwill concludethepaperwithpromisingdirectionsforfutureresearch.II. UNOBTRUSIVE SENSING METHODS AND SYSTEMSThe main objective of unobtrusive sensing is to enable con-tinuous monitoring of physical activities and behaviors, as wellas physiologicaland biochemicalparametersduring the dailylifeofthesubject.Themostcommonlymeasuredvitalsignsinclude: ECG, ballistocardiogram(BCG), heart rate, bloodpressure(BP), bloodoxygensaturation(SpO2), core/surfacebodytemperature,posture,andphysicalactivities.Aconcep-tual model of unobtrusive physiological monitoring in a homesettingis showninFig. 2[5]. Unobtrusivesensingcanbeimplemented in two ways: 1) sensors are worn by the subject,e.g., in the form of shoes, eyeglasses, ear-ring, clothing, glovesand watch, or 2) sensors are embedded into the ambient envi-ronment or as smart objects interacting with the subjects, e.g.,a chair [6], [7], car seat [8], mattress [9], mirror [10], steeringwheel [11], mouse [12], toilet seat [13], and bathroomscale [14].Fig. 3 shows some prototypes of the unobtrusive sensing sys-tems developed by different research groups. Information canbecollectedbyasmartphoneandtransmittedwirelesslytoaremotecenterforstorageandanalysis. Inthefollowingsec-tion, we will discuss some unobtrusive sensing methods for theacquisition of vital signs.A. Capacitive Sensing MethodCapacitance-coupled sensing method is commonly used formeasuringbiopotentialssuchasECG, electroencephalogram(EEG) and electromyogram (EMG) [7], [15]. For this method,theskinandtheelectrodeformthetwolayersof acapaci-tor. Without direct contact withthebody, someissues, suchas skininfectionandsignal deterioration, brought about byadhesiveelectrodesinlongtermmonitoringcanbeavoided.SometypicalimplementationsofcapacitiveECGsensingaresummarizedinTable I. Inaddition, capacitive sensingcanalsobeusedforother applicationssuchasrespiratorymea-surement,e.g.,usingacapacitivetextileforcesensorweavedinto clothing [16], or a capacitive electrical eld sensor arrayplacedunderasleepingmattress[17]. Themajorchallengesindesigningthesenoncontact electrodeslieinthehighcon-tactimpedanceduetotheindirectcontactandthecapacitivemismatchcausedbymotionartifacts[15]. It maycauselowsignaltonoiseratioandthusleadtochallengesinthefront-end analog circuit design. The input impedance of the amplierhas to be extremely large (>1 T) to reduce the shunt effectformedbythecapacitorandtheinputimpedance. Moreover,motionartifactsmaybecomemoresignicant innoncontactsensing.Somemethodshaverecentlybeenproposedtoover-1540 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 5, MAY 2014Fig. 2. Illustration of unobtrusive physiological measurements in a home environment [5].Fig. 3. Unobtrusive sensing devices with sensors embedded in daily objects,such as mirror [10], sleeping bed [9], chair [6], steering wheel [11], and toiletseat [13].come these problems to achieve robust measurement in practi-cal situations. For instance, the gradiometric measurement tech-nique introduced by Pang et al. can considerably reduce motionartifacts [24].B. Photoplethysmographic Sensing MethodPhotoplethysmographic (PPG) sensing, which involves a lightsource to emit light into tissue and a photo-detector to collectlight reected from or transmitted through the tissue, has beenwidely used for the measurement of many vital signs, such asSpO2, heart rate, respiration rate, and BP. The signal measuredby this method represents the pulsatile blood volume changes ofperipheralmicrovasculatureinducedbypressurepulsewithineachcardiaccycle.Traditionally,thesensingunitisindirectcontact with skin. Recent research has shown that sensors can beintegrated into daily living accessories or gadgets like ear-ring,gloveandhat, toachieveunobtrusivemeasurement. VariousTABLE IDIFFERENT IMPLEMENTATIONS OF CAPACITIVE ECG SENSORStypes of PPG measuring devices at different sites of the bodyare summarized in Table II.Recently, Jaeet al. [25]proposedanindirect-contact sen-sor for PPG measurement over clothing. A control circuit wasadopted to adaptively adjust the light intensity for various typesofclothings. Ontheotherhand, Pohet al. [26]showedthatZHENG et al.: UNOBTRUSIVE SENSING AND WEARABLE DEVICES FOR HEALTH INFORMATICS 1541TABLE IIPPG MEASURING DEVICES AT DIFFERENT SITES OF THE BODYDevicesLocation of Sensor/Operation modeMeasuredParameters PPG ring [27]Finger/ReflectiveHeart rate and its variability, SpO2Pulsear [28]External ear cartilage/ReflectiveHeart rateForehead mounted sensor [29]Forehead /ReflectiveSpO2e-AR [30]Posterior, inferior and anterior auricular /ReflectiveHeart rateIN-MONIT system [31]Auditory canal /ReflectiveHeart activity and heart rateGlove and hat based PPG sensor [32]Finger and forehead/ReflectiveHeart rate and pulse wave transit timeEar-worn monitor [33]Superior Auricular, superior Tragus and Posterior Auricular /ReflectiveHeart rateHeadset [34]Ear lobe/TransmissiveHeart rateHeartphone [35]Auditory canal /ReflectiveHeart rateMagnetic earring sensor [36]Ear-lobe /ReflectiveHeart rateEyeglasses [37]Nose bridge/ReflectiveHeart rate and pulse transit timeEar-worn PPG sensor [38]Ear lobe /ReflectiveHeart rateSmartphone [39] Finger /Reflective Heart rateheart rateandrespirationratecanbederivedfromPPGthatwasremotelycapturedfromasubjectsfaceusingasimpledigital camera. However, the temporal resolution of the bloodvolume detected by this method is restricted by the sample rateof the camera (up to 30 frames per second), thus affecting itsaccuracy.C. Model-Based Cufess BP MeasurementSphygmomanometer, whichhasbeenusedover acenturyforBPmeasurement, isoperatedbasedonaninatablecuff.Conventional methods such as auscultatory, oscillometric, andvolumeclamparenot suitablefor unobtrusiveBPmeasure-ment. Pulse wave propagation method is a promising techniquefor unobtrusive BP measurement. It is based on the relationshipbetween pulse wave velocity (PWV) and arterial pressure ac-cording to Moens-Korteweg equation. Pulse transit time (PTT),thereciprocalofPWVcanbereadilyderivedfromPPGandECG in an unobtrusive way. Various linear and nonlinear mod-elsthat expressedBPintermsofPTThavebeendevelopedforcufessBPestimation.Thesubject-dependentparametersinBP-PTTmodelshouldbedeterminedrstwhenusingthisapproach for BP estimation. Avery simple way to implement in-dividual calibration is to use the hydrostatic pressure approach,wherethesubjectsarerequiredtoelevatetheirhandstospe-cicheightsabove/belowtheheart level [40]. Atheoreticalrelationship between PTT, BP, and height can be written in (1),asshownat thebottomofthepage, wherebisthesubject-dependent parameter characterizing the artery properties,L isthe distance traveled by the pulse,Phis the hydrostatic pres-suregh,andPiistheinternalpressure.Usingtheproposedmodel, the individualized parameters in BP-PTT model can bedeterminedfromsomesimplemovements. Thismodel-basedcufess BP measurement method can be implemented in a va-riety of platforms for unobtrusive monitoring, such as the bedcushion and chair, as shown in Fig. 3.Although the accuracy of this method has been validated inmanyrecent studies[41], thereareconcernsovertheuseofPTT as a surrogate measurement of BP. It has been recognizedthat the major confounding factors in the present relationship ofBP and PTT are vasomotor tone and pre-ejection period [42].New models should be developed to include these confoundingfactors in order to explore the potential of PTT-based methodfor unobtrusive BP measurement in future.D. Other Unobtrusive Sensing MethodsStrain sensors are commonly used to measure body motionsuch as respiration, heart sound and BCG. Piezoelectric cablesensor, whosesensingelement ispiezoelectricpolymer, hasbeen used for respiration rate monitoring [43]. Flexible and thinsensors, such as piezoresistive fabric sensors [44] and lm-typesensors like polyvinylideneuoride lm(PVDF) and electrome-chanical lm (EMFi) based sensors [45] have been widely usedfor cardiopulmonary applications due to the easiness of beingembedded into clothing or daily objects like chair or bed. Com-parative study has been conducted to evaluate the performancesof the two different strain sensors in the measurement of bodymotion [46]. They have shown that only small differences werefound in heart rate measured by PVDF-based and EMFi-basedsensors especially at supine posture, which was possibly causedby their different sensitivities to different force components.Inductive/impedance plethysmography is another widelyused method for respiratory measurement and has been devel-oped in the forms of clothing and textile belt [47], [48]. Twosinusoidwirecoilslocatedat ribcageandtheabdomenaredriven by a current source that generates high-frequency sinu-soidalcurrent. Themovementofthechestduringrespirationcauses changes of the inductance of the coils and thus modu-latestheamplitudeofthesinusoidalcurrent, fromwhichthePTT =Lb1 + exp(bPi); h = 02Lbgh ln

exp[b(Pi Ph)] + exp(bPh)

exp(bPh)

exp[b(Pi Ph)]+11

; h = 0(1)1542 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 5, MAY 2014respiratorysignal canbedemodulated. Arecent studycom-paredtheperformancesof four different methodsfor wear-able respiration measurement, including inductive plethysmog-raphy, impedance plethysmography, piezoresistive pneumogra-phy, and piezoelectric pneumography piezoelectric, and showedthat piezoelectric pneumography provided the best robustnessto motion artifacts for respiratory rate measurement [43].Optical bers have also been adopted for unobtrusive mon-itoringbyembeddingthemintodailyobjectsor clothes. Incontrast to electronic sensors, they are immune to the electro-magneticinterference.Recently,berBragggrating hasbeenproposed as a vibration sensor for BCG measurement, based onthe fact that the Bragg wavelength is correlated with the gratingperiodinresponsetothebodyvibrationcausedbybreathingandcardiaccontractions[49].A pneumaticcushionbasedonthismethodhasbeendevelopedtomonitorthephysiologicalconditionsofpilotsanddrivers[49].DAngelodevelopedanoptical ber sensor embedded into a shirt for respiratory motiondetection [50].Other remote sensing methods have also been proposedfor unobtrusive physiological measurement, such as frequencymodulated continuous wave Doppler radar for BCG measure-ment [51] and radiometric sensing for body temperature mea-surement [52].III. WEARABLE DEVICESIn this section, an overviewon the state-of-the-art of wearablesystems is presented. Several key enabling technologies for thedevelopment of these wearable devices, such as miniaturization,intelligence, networking, digitalization and standardization,security, unobtrusiveness, personalization, energyefciency,androbustnesswillbediscussed. Sensorembodimentsbasedon textile sensing, exible and stretchable electronics are alsointroduced.A. Overview of Unobtrusive Wearable DevicesA variety of unobtrusive wearable devices have been devel-oped by different research teams as shown in Fig. 4: the watch-type BP device [3], clip-free eyeglasses-based device for heartrate and PTT measurement [37], shoe-mounted system for theassessment of foot and ankle dynamics [53], ECG necklace forlong-termcardiac activity monitoring [54], h-Shirt for heart rateand BP measurement [55], an ear-worn activity and gait mon-itoring device [56], glove-based photonic textiles as wearablepulse oximeter [57], a strain sensor assembled on stocking formotion monitoring [58], and a ring-type device for heart rate andtemperature measurement [59]. These embodiments are capa-ble of providing measurements ubiquitously and unobtrusively.Manyof themalreadyhave a varietyof applications inhealthcareas well as wellness and tness training. For example, postureandactivitymonitoringoftheelderlyandthedisabled,long-term monitoring of patients with chronic diseases like epilepsy,cardiopulmonary diseases, and neurological rehabilitation.Implantable sensors: It is worth noting that sensing devicesshould not be limited to those designed to be worn on the body.There are increasing numbers and varieties of implantable sen-Fig. 4. Unobtrusive wearable devices for various physiological measurementdeveloped by different groups: watch-type BP device [3], PPG sensors mountedoneyeglasses[37], motionassessmentwithsensorsmountedonshoes[53],wireless ECG necklace for ambulatory cardiac monitoring (courtesy of IMEC,Netherlands)[54], h-Shirt forBPandcardiacmeasurements[55], ear-wornactivity recognition sensor [60], glove-type pulse oximeter [57], strain sensorsmounted on stocking for motion monitoring [58], and ring-type device for pulserate and SpO2measurement [59].sors have been introduced, which are designed to be implantedinside the human body. For example, CardioMEMS recently de-veloped a wireless implantable haemodynamic monitoring sys-tem, which can provide long-term measurement of pulmonaryarterial pressure of patients with heart failure [61]. Arecent clin-ical study was conducted to validate the safety and effectivenessof thisdevice, andtheresultsshowedthat theheart-failure-related hospitalization can be signicantly reduced in the groupwith the implant compared to the control group [61]. Anotherexample is the wireless capsule device that can be pinned on theesophageal to measure pH level over a 48-hour period for bet-ter diagnosis of gastroesophageal reux disease [62]. However,the sensor is difcult to be securely attached for the completesensing periods in some subjects, and premature losing of thecapsule has been reported quite often. Other technologies suchas the development of mucosal adhesive patches could be a po-tential solution in this regard. Another important application formucosal adhesive patches is to design themas drug delivery sys-tems. By placing these devices inside the gastrointestinal tractandwithfeedbackmechanismsinstalled, thereleaseofdrugcan be better controlled. This is anticipated to be helpful to thebetter management of many chronic diseases, such as diabetesand hypertension.B. Key Wearable TechnologiesIn addition to unobtrusive sensing methods mentioned in thelastsection,somekeytechnologiesshouldalsobedevelopedtoimplement unobtrusivewearabledevicesforpractical use.ZHENG et al.: UNOBTRUSIVE SENSING AND WEARABLE DEVICES FOR HEALTH INFORMATICS 1543Zhangs research group summarized these key technologies asMINDS (Miniaturization, Intelligence, Networking, Digital-ization, Standardization) [63]. The recent development of thesetechnologies will be elaborated as follows together with someother important issues such as unobtrusiveness, security, energy-efciency, robustness, and personalization.1) MiniaturizationandUnobtrusiveness canenhance thecomfortness of using wearable devices, and thus increasing thecompliance for long-termand continuous monitoring. Thanks tothe rapid development of integrated circuit technologies and mi-croelectromechanical technologies, the size of processing elec-tronics and inertial measurement units (IMUs) has been signi-cantly reduced for wearable applications. For instance, Briganteet al. recently developed a highly compact and lightweight wear-able systemfor motion caption by careful selection of IMUs andoptimizedlayout design[64]. Ahighlyintegratedmicrosys-tem was designed for cardiac electrical and mechanical activitymonitoring, whichassembledthemultisensormodule, signalprocessingelectronics, andpoweringunit intoasingleplat-form with a exible substrate [65]. In addition to the integrateddesign and compact packaging, the development of new mea-suring principle is another way to achieve miniaturization. Theaforementioned PTT-based cufess BP measuring method is apromising substitute for the widely used cuff-based methods inthisregard.Furtherminiaturizationcanbeachievedbyusingother technologies such as inductive powering where the sensorcan be powered partly or completely by RF powering, and thebattery can be removed [66].As mentioned, capacitive coupling sensing is one commonlyusedmethodfor unobtrusive capturingof bioelectrical sig-nals, likeECGandEEG. Textilesensingisanotherunobtru-sive method for continuously monitoring physiological param-eters. A textile-integrated active electrode was presented in [67]for wearable ECG monitoring, which could maintain signal in-tegrity after a ve-cycle washing test. In addition, PPG offersa good way for unobtrusive cardiovascular assessment, such asnoninvasive measurement of BP based on PTT.2) Networking and Security: Networking is an integral partof wearable devices to deliver high-efciency and high-qualityhealthcare services fromthe m-health perspective [68]. The termBSNbody sensor networkwas coined to harness severalallied technologies that underpin the development of pervasivesensing for healthcare, wellbeing, sports, and other applicationsthat require ubiquitous and pervasive monitoring of physi-cal, physiological, and biochemical parameters in any environ-ment and without activity restriction and behavior modication.BSN can be wired (e.g., interconnect with smart fabric) or wire-less (making use of common wireless sensor networks and stan-dards, e.g., BAN, WPAN(IEEE 802.15.6), Bluetooth/BluetoothLow Energy (BLE or Bluetooth Smart), and ZigBee). BSN ispresently a very popular research topic and extensive progresseshave been made in the past decades. Related to networking, tech-nical challenges include user mobility, network security, mul-tiple sensor fusion, optimization of the network topology, andcommunication protocols for energy-efcient transmission [69].In a broader sense, networking for sensing can also includewireless personal area network (WPAN), wireless local area net-work (WLAN) and cellular networks (3G/4G network, GPRS,GSM) or wireless wide area network (WWAN), which are usedtosendall theacquireddatafromwearabledevicestodataservers for storage and postprocessing. The most critical issueinWPAN/WLANtechnologiesistoguaranteethequalityofservice (QoS), which contains latency, transmission power, reli-ability (i.e., error control) and bandwidth reservation. While forhealthcare applications, these issues become more challengingbecause all the specications should be satised simultaneously.Forexample, monitoringECGrequiresadatarateoftenstohundreds of kilobit/s, end-to-end delay within 2 s and bit errorrate of below 104[70]. Under some life-critical situations, thecompromise between reliability and latency constraint of datatransmissionisevenmoreprominent. Effectiveerrorcontrolschemes have been proposed to address this issue, such as thecombineduseoffeed-forwarderrorcontrol(FEC)andblockinterleaving in data-link layer [71]. For bandwidth demandingapplications, such as two-dimensional (2-D) mode of echocar-diograph and ultrasound video transmission, to transmit the datawith minimal loss of diagnostic information and in real time,theprotocolshouldbedesignedtooptimizethetransmissionrate, minimize the latency and maximize the reliability. Variousenhanced protocols on rate control have been proposed for theseapplications, such as a hybrid solution either using retransmis-sion or retransmission combined with FEC techniques depend-ing on the channel conditions [72], adapting the sending rate ofsource encoder according to the mobile link throughput [73].Security is a critical issue of networking, especially for med-ical applications. Without adequatesecurity, thetransmittedinformation is vulnerable to potential threats, such as eavesdrop-ping, tampering, denial of service, which are caused by unau-thorized access. The challenging issues to guarantee security liein three aspects: how to prevent the disclosure of patients data,who have the right to access the system, and how to protect theprivacy of the user [74]. Securing the data transmission betweenwearable sensors should be the rst step. To secure inter-sensorcommunications, themostimportantissueistodesignakeyagreement for encryption. Identity-based symmetric cryptogra-phy and asymmetric key cryptography based on elliptic curvehave been adopted for BSN [75], [76]. Another novel methodisbasedonbiometricstraits.Sincethehumanbodycontainsits own transmission system such as blood circulation system,it is believed that the information from the system itself can beused to secure the information communication between sensors.Meanwhile, since wearable sensors are already there to capturephysiological signals, it is convenient to implement encryptionschemes using physiological signals in the key agreement. Poonetal.rstlyproposedtousetheinterpulseintervals(IPIs)asthe biometric traits to encrypt the symmetric key, as shown inFig. 5 [77]. The results showed that a minimum half total errorrate of 2.58% was achieved when the signals were sampled at1000 Hz and then coded into 128-bit binary sequences. Subse-quently, other similar approaches have been proposed [78]. Forexample, Venkatasubramanian et al. proposed using frequencydomainfeaturesofPPGasthebasisofkeyagreement [79].Furthermore, the complexity of the security scheme should alsobe considered due to the constraints on the energy budget and1544 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 5, MAY 2014Fig. 5. Illustration of using biometrics approach to secure inter-sensors com-munication for wearable health monitoring applications [77].Fig. 6. (a) High-sensitivity photo-transistor based on organic bulk heterojunc-tion; (b) PPG signal detected by this photo-transistor [84].computational power of the wearable sensor. Various potentialsolutions have been developed, such as the data condential anduserauthenticationschemeswithlowcomplexityrealizedbyminimizing the shared keys [80], and mutual authentication andaccess control scheme based on elliptic curve cryptography withthe advantage of energy-efcient [81].3) Energy-Efciency and Digitalization: Energy-efciency isa crucial element of a wearable device that directly affects thedesign and usability of the device, especially for long-termmon-itoring applications. There are mainly three strategies to achieveenergy-efcient design. The rst one is to improve the energyefciency of the existing energy storage technologies, lithium-ionbatteriesandsupercapacitors.Liuetal.designedanovelstructureofZnCo2O4-urchins-on-carbon-bersmatrixtofab-ricateenergystoragedevicethatexhibitsareversiblelithiumstorage capacity of 1180 mAh/g even after 100 cycles [82]. Fuet al. rst adoptedcommercial peninkasanactivematerialfor supercapacitors. With simple fabrication, the exible bersupercapacitors can achieve areal capacitance of 11.919.5 mFcm2and power density of up to 9.07 mW cm2[83]. The sec-ond is energy-aware design of wearable system. For example, ahigh-sensitivity near-infrared photo-transistor was recently de-velopedbased on an organicbulk heterojunctionas shown inFig. 6(a) [84]. The results showed that the responsivity of thisdevice can achieve 105A W1. It has been successfully usedfor PPG measurement as shown in Fig. 6(b) [84]. With the highresponsivity, thisdeviceallowstheuseofalow-powerlightsource,andthusreducingthepowerconsumptionofthesen-sor. Power management approaches, such as dynamic voltagescaling, battery energy optimal techniques and system energymanagement techniques have also been proposed [85]. Last butnot theleast istoscavengeenergyfromtheenvironment orhuman activities, such as the lower-limb motion and vibration,bodyheat andrespiration, whichhavegreat potential tobe-come sustainable power sources for wearable sensors. The MITMedia Lab [86] has realized an unobtrusivedevicethat scav-engedenergyfrom heel-strike oftheuserwith shoe-mountedpiezoelectric transducer. Recently, Leonov [87] has developeda hidden thermoelectric energy harvester of human body heat.Itwasintegratedintoclothingasareliablepoweringsource.For implantable devices, the preferred power solution is wire-lesspowering. Kimetal. havederivedthetheoreticalboundof wireless energy transmit efciency through tissue. By opti-mizing the source density, the peak efciency can achieve 4 105at around 2.6 GHz, which is 4 times larger than that of theconventional near-eld coil-based designs [88].Digitalization is necessary to enable data analysis and storageafter acquiring the analog signals from the body with wearabledevices. Due to the rigorous power constraints of wearable de-vices, thesamplingrateshouldbeminimizedtosavepowerwhilst not compromising the diagnostic accuracy. For example,the sampling rate for ECG should be no less than 125 Hz; other-wise, it will affect the accuracy of measurements [89]. Recently,novel nonuniformly sampling methods have been investigatedto compress the representation of ECG signal in which the rel-evant information is localized in small intervals. One exampleis the integrate-and-re sampler that integrates the input signalagainst an averaging function and the result is comparedto apositive or negative threshold as shown in (2),k=

tk + 1tk + x(t)e(ttk + 1 )dt; k {p, n} . (2)Apulsewillbegeneratedwheneitherofthethresholdsisreached. The original signal thus can be represented by a nonuni-formly spaced pulse train from the output of the integrate andre model [90]. This sampler can be implemented by effectivehardwaretosavepowerandfacilitatesensorminiaturization.However, the tradeoff is that exact reconstruction of the originalsignal is impossible [90]. Nonetheless, some critical informa-tionforaccuratediagnosiscanbepreserved. Without recon-struction, a time-based integrate and re encoding scheme canachieve 93.6%classication rate between normal heartbeats andarrhythmias, which is comparable to other methods [91].4)StandardizationandPersonalization:Standardizationisa crucial component of any commercial exploitation of wear-abledevices, asitensuresthequalityofthedevicesanden-ables interoperability among devices. However, the process ofstandardizationisoftenverycomplicatedandtimeconsum-ing, especially in standardizing healthcare devices. Health in-formationcanbeverydifferent rangingfromphysiological(ECG, bodytemperature, oxygensaturation, etc.) tophysi-cal (posture, activity, etc.). Manufacturers tend to dene theirownproprietaryformatsandcommunicationprotocols.Inter-operabilitythereforebecomesamajorobstacleinintegratingZHENG et al.: UNOBTRUSIVE SENSING AND WEARABLE DEVICES FOR HEALTH INFORMATICS 1545devices intohealthcaresystems. Toresolvethisissue, stan-dardprotocolssuchastheHL7referenceinformationmodelhavebeenproposedfor theintegrationof different sensingmodalities. For instance, LohandLeeproposedtheuseofanOracleHealthcareTransactionBase(HTB)systemastheplatform for integrating different information into a healthcaresystem [92]. ISO/IEEE 11073 (X73), initially intended for theimplementationofPoint-of-Caredevices, isnowextendedtoPersonal HealthDevices(PHD)toaddresstheinteroperabil-ity problem. This standard provides the communication func-tionality for a variety of wearable devices (i.e., pulse oximeter,heart rate monitor, BP monitor, thermometer, etc.) and standard-izes the format of information for plug-and-play interoperabil-ity [93]. A recent study has validated the feasibility to apply thisstandard to wearable and wireless systems in home settings [94].Tomeettheneedsofsomescenarioswithlimitedprocessingabilities, some amendments have been made based on this stan-dard,suchastheadapteddatamodelproposedin[95]andanew method to implement X73PHD in a microcontroller-basedplatform with low-voltage and low-power constraints [96].However, very few standards are available for evaluating theaccuracy of wearable devices, albeit the necessity of standards toguarantee the reliability of these devices. For example, there arethree well-known standards for assessing the accuracy of cuff-based BP measurement devices, i.e., the American Associationfor the Advancement of Medical Instrumentation (AAMI), theBritish Hypertension Society (BHS), and the European Societyof Hypertension (ESH). On the other hand, Zhangs team hasbeen spear-headed a standard for cufess BP devices [97]. Inthis work, a new distribution model of the measuring differencebetween the test and the reference device was proposed. The cu-mulative percentages (CP) of absolute differences between thetestdeviceandthereferencederivedbyintegratingtheprob-abilitydensityfunctionovertherequiredrangeoflimits(L)for normal and general t distribution are shown in (3) and (4),respectively:CPgL(u, d, L)=L

Lp(x|u, d)dx =1d2L

Le( x u )22 d 2dx (3)CPtL(u, d, v, L)=L

Lp(t|u, d, v)dt=d1(v+12)(v2)

(v 2)L

L(1 +(t u)2d2(v 2))v + 12dt(4)where u, d, and v denote mean, standard deviation, and degreeoffreedom,respectively.Itwasfoundthattheagreementbe-tween the evaluation results of 40 cuff-based devices by AAMIandBHSprotocolswasbetterfortheproposedtdistribution(82.5%)thannormal distribution(50%). Theproposederrordistribution model was further validated by the goodness-of-tof 999 datasets obtained from 85 subjects by a PTT-based cuf-ess BP estimation method to the hypothesized distribution. Inaddition, this study also proposed new evaluation scales undergeneral t distribution to assess the accuracy of cufess BP de-vices, including: mean absolute difference (MAD), root meansquare difference (RMSD) and mean absolute percentage dif-ference (MAPD) as shown in (5) and (6) [97]MAD = 2s

v((v + 1)/2)(v/2)(v 1)

1 +u2s2v

v 12+|u|

1 I

vv + (u/s)2 ;v2, 12

(5)RMSD = u2+ d2(6)wheresisthescaleandequalstod

(v 2)/(v> 2).Basedon(3)and(4),amappingcharttorelatetheproposedscales with the AAMI, BHS, and ESH evaluation criteria undert4 distribution was developed.On the other hand, personalization of wearable devices is alsoimportant. Apart from personalizing the design of the devices,it can be referred to personalizing the sensor calibration, diseasedetection, medicine, and treatments. Taking BP as an example:individualized BP calibration procedure is required for the PTTbased unobtrusive BP measurement to ensure the accuracy. Thecalibration can be implemented through body movement suchas hand elevation [40]. On the other hand, from the perspectiveof personalized medicine, since BP can be very diverse amongindividuals, it is necessary to track a persons BP in long-termby wearable devices to establish an individual database for per-sonalizedBPmanagement. Thiscanleadtoamoreprecisediagnosis and treatment.5) Articial IntelligenceandRobustness: Articial intelli-gence enables autonomic functions of wearable devices,suchas sending out alerts and supporting decision making. Articialintelligencemayalsorefertoascontext-awarenessanduser-adaption[98]. VariousclassicationmethodssuchashiddenMarkovmodel,articialneuralnetworks,supportvectorma-chines, random forests, and neuro-fuzzy inference system havebeenextensivelyappliedfor varioushealthcareapplications.The need of individualized training data, the lack of large scalestudies in real-world situations, and the high rate of false alarmsaresomeof themajor obstaclesthat hinder thewidespreadadoption of sensing technologies in clinical applications.Oneofthemajorproblemsforwearablesensorsismotionartifacts. The reduction of motion artifact is still a challengingproblemforthedevelopment ofwearablesystemduetotheoverlapping of its frequency band with the desired signal. Sev-eral methods have been proposed for motion artifacts reduction.For instance, Chung et al. [99] presented a patient-specic adap-tive neuro-fuzzy interference system (ANFIS) motion artifactscancellation for ECG collected by a wearable health shirt. WithANFIS, it was able to remove motion artifacts for ECG in realtime. Adaptive ltering is another commonly used method formotion artifacts removal, and algorithms based on it have beenappliedforvariousapplications. Koetal. [100]proposedanadaptive ltering method based on half cell potential monitor-ing to reduce the artifacts on ECG without additional sensors.Poh et al. [36] embedded an accelerometer in an earring PPG1546 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 5, MAY 2014Fig. 7. Various wearable garments for physiological and activity monitoring.(a) Georgia Tech Wearable Motherboard (Smart Shirt) for the measurement ofECG, heart rate, body temperature, and respiration rate [105]; (b) the EKG Shirtsystemthat used interconnection technology based on embroidery of conductiveyarnfor heart rate[106]; (c) theLifeShirt systemfor themeasurement ofECG, heart rate, posture and activity, respiration parameters, BP (peripheral isneeded), temperature, SpO2 [107]; (d) the ProTEXgarment for the measurementofheartrate, breathingrate, bodytemperature, SpO2, position, activity, andposture[108];(e)theWEALTHYsystemwithknittedintegratedsensorsforthe measurement of ECG, heart rate, respiration, and activity monitoring [44];(f) the VTAMN system for the measurement of heart rate, breathing rate, bodytemperature, and activity [109]; (g) the h-Shirt system for the measurement ofECG, PPG, heart rate, and BP [55].sensor to obtain motion reference for adaptive noise cancella-tion. Although adaptive ltering has shown to be an effectivemethod, it requires extensive processing time, which hinders itsreal-timeapplications. YanandZhang[101]developedaro-bust algorithm, called minimum correlation discrete saturationtransform, which is more time efcient than discrete saturationtransformthat usedadaptivelter. Othermethods, likeinde-pendent component analysis and least mean square based activenoise cancellation, have also been used to remove motion arti-facts.Insummary, themainissuestobeaddressedfortheubiq-uitous use of wearable technologies canbe summarizedasSUPERMINDS(i.e., Security, Unobtrusiveness, Personal-ization, Energyefciency, Robustness, Miniaturization, Intel-ligence, Network, Digitalization, andStandardization). Moreattentions should be paid to these aspects for the future devel-opment of wearable devices.C. Textile Wearable DevicesSmart textile technology is a promising approach for wear-ablehealthmonitoring, sinceit providesaviablesolutiontointegratewearablesensors/actuators,energysources,process-ing, andcommunicationelementswithinthegarment. Somerepresentative garment-based systems have been developed asshown in Fig. 7.Despite extensive progresses been made in wearable devicesfor physiological monitoring, little attention has been paid onthe other health information such as biochemical and autonomicinformation that can provide a more comprehensive picture ofFig. 8. Flexible and stretchable wearable devices for health care applicationsdeveloped by different groups: (a) multifunctional epidermal electronic devicefor electrophysiological, temperature, and strain monitoring [4]; (b) a exiblecapacitive pressure sensor for radial artery pressure pulse measurement [112];(c) e-skin with pressure and thermal sensors [113]; (d) a bandage strain sensorfor breathing monitoring [58]; (e) electrochemical tattoo biosensors for real-timenoninvasive lactate monitoring in human perspiration [116].the subjects health state. At this point, textile technology pro-vides feasible solutions to the unobtrusive measurement of thesenew parameters. For instance, a textile-based sweat rate sensorhas recently been developed, which measures the water-vaporpressure gradient by two humidity sensors at different distancesfromskin[102]. Textile-basedpHsensorwasalsoproposedusing a colormetric approach with pH sensitive dyes and opto-electronics [102]. In addition, Poh et al. developed conductivefabricelectrodesforunobtrusivemeasurement ofelectroder-malactivity(EDA),whichcanbeusedforlong-termassess-ment of autonomic nerve activities during daily activities [103].This textile sensor has been designed into a wrist-worn devicethat can provide ambulatory monitoring of autonomic parame-ters for patients with refractory epilepsy. The strong correlationbetweenEDAamplitudeandpostictal generalizedEEGsup-pression (PGES) duration, an autonomic biomarker of seizureintensity, indicates that EDA can be used as a surrogate mea-surement of PGES for ambulatory monitoring of epilepsy andthus provides opportunity for the identication of sudden deathin epilepsy [104].D. Flexible, Stretchable, and Printable DevicesFlexible electronics, where electronic circuits are manufac-tured or printed onto exible substrates such as paper, cloth fab-rics and directly on human body to provide sensing, powering,and interconnecting functions, have a broad range of biomedicalapplications. They have enabled the development of wearabledevicestobethinner,lighterandexible.Themostcommonmethods to fabricate these exible sensors and other functionalcircuitsonexiblesubstratesinclude: inkjet-printing, screenprinting, transfer printing [4], low temperature deposition, etc.Flexibleelectronicsforhealthmonitoring:Fig. 8(b)to(e)shows some cutting-edge exible wearable systems for healthmonitoring applications. For example, Baos team developed ahigh-sensitivityexiblecapacitivepressuresensorwithpoly-dimethylsiloxane as the dielectric layer [111]. By microstruc-turing the dielectric layer and integrating it in a thin-lm poly-mer transistor, it is able to achieve very high sensitivity whenZHENG et al.: UNOBTRUSIVE SENSING AND WEARABLE DEVICES FOR HEALTH INFORMATICS 1547driving the exible pressure-sensitive transistor in the subthresh-oldregime. Thisdevicehasbeenappliedtomeasureahu-manradial arterypulsewavewithhigh-delityasshowninFig. 8(b) [112]. Other examples include the e-skin with pres-sure and thermal sensors, and the exible strain-gauge sensor forbreathing monitoring as shown in Fig. 8(c) and (d) [58], [113].Somerecentstudieshaveproposedthefabricationofexiblephoto-transistor [114], which can be used to build PPG sensors.Theseadvanceswilleffectivelypavethewayforunobtrusivephysiological measurementsinthefuture. Inadditiontotheapplications in physiological and physical monitoring, exibleelectronic technology also begins to play a key role in biochem-ical sensing. A exible nano-electronic nose was developedby Michael et al. by transferring hundreds of prealigned siliconnanowires onto plastic substrate. It would open up new oppor-tunitiesforwearableorimplantablechemical andbiologicalsensing[115]. AnotherexampleistheelectrochemicaltattoobiosensorrecentlydevelopedbytheresearchteamfromUni-versity of California San Diego, as shown in Fig. 8(e) [116]. Itcan be easily worn on body for lactate monitoring during aerobicexercise. The performance of this tattoo sensor was evaluatedintermsofitsselectivityforlactatemeasurement,theabilityof adhering to epidermal surface, and the robustness when ex-posed to mechanical stretching and bending. The results haveshowed its potential for unobtrusive and continuous monitoringof lactate during exercise [116].Flexiblesensorswithinorganic,organic,andhybridmate-rials:Avarietyoforganicandinorganicmaterialshavebeenadopted for the development of exible sensors. CNT is a verypromising carbon based material for the design of exible de-vices due to its unique mechanical and electrical properties. Aexible lm strain sensor based on CNT has been designed andintegrated into fabrics for human motion monitoring [58]. It canendure strain up to 280% and achieve high durability of 10 000cycles at 150% strain as well as fast response of 14 ms, whichmight be used for breath monitoring for the early detection ofsudden infant death syndrome in sleeping infants. On the otherhand, the printable feature and relative ease of fabrication pro-cess make organic materials suitable for the implementation ofexible wearable devices. Flexible organic sensors employingthin-lm transistor structure for skin temperature measurementand pressure sensing [117], [118] have been proposed.Due to the complementary benets of inorganic and organicmaterials in terms of electrical conductivity, mechanical prop-erties, and low-cost fabrication process, inorganic and organichybridmaterialshavegreat potential forthedevelopment ofhigh-performanceexiblewearabledevices. For example, ahigh-performanceinorganicandorganichybridphotodetec-tor based on P3HT and nanowire was fabricated on a printingpaper, which has high exibility and electrical stability [119].Leesteamfabricatedaexibleorganiclight emittingdiodewith modied graphene anode to achieve extremely high ef-ciency [120]. Xu et al. developed an organic bulk heterojunctionbased near-infrared photo-transistor [84], which can be modi-ed to a exible device in the future. This device showed greatpotential for the design of low power PPG sensor due to its ul-trahigh responsivity. Pang et al. developed a exible and highlysensitive strain-gauge sensor that is based on two interlocked ar-rays of high-aspect-ratio Pt-coated polymeric nanobres [121].This device is featured by its simple fabrication process as wellas the ability of measuring different types of mechanical forceswith high sensitivity and wide dynamic range. The assembleddevicehasbeenusedtomeasurethephysical forceofheartbeat by attaching it directly above the artery of the wrist undernormal and exercise conditions.In addition to the aforementioned advances in the design ofexiblesensors, exibletechnologiesalsoenableotherfunc-tional circuits suchas aninkjet-printedexibleantennaforwireless communication [122], exible batteries with wirelesscharger [123], exible supercapacitors on graphene paper [124]and cloth fabrics [125] for wearable energy storage, etc.Althoughsignicant advanceshavebeenmadeinexibleelectronics, only limited applications in unobtrusive health mon-itoring have been explored until now. Some key challenges stillhinder the application of these systems in real life situations forwearable health monitoring, such as engineering new structuralconstructs from established and new materials to achieve higherexibility, improving the durability of sensors and interconnectsthat are frequently exposed to mechanical load, integrating wire-less data transmission and powering electronics with the exiblesensors, etc.Stretchable electronics for health monitoring: Flexible elec-tronicsalreadyoffersgreat opportunitiesintheapplicationsof wearablehealthmonitoring, but is not enoughtoensurethe conformal integration of the devices with arbitrary curvedsurfaces. Stretchability, which enables the devices bend to ex-tremely small radius and accommodate much higher strain whilemaintaining the electronic performance, is a more challengingcharacteristic and renders wider application possibilities. Somestretchable systems for health monitoring applications have beenproposed. One example is the epidermal electronic system de-veloped by Rogers research group for measuring electrophysi-ologicalsignals(ECGandEMG),temperature,andstrain,asshowninFig. 8(a). The systemconsists of multifunctionalsensors,processingunit,wirelesstransmissionandpowering,which are all mounted on an elastic polymer backing layer. Thepolymer sheet can be dissolved away, leaving only the circuitsadhere to the skin via van der forces [4]. One problem of thisdevice is that it can easily fall off during bathing or exercise.Therefore, they recently developed a new epidermal electronicsystem that has a total thickness of 0.8 m and can be directlyprinted onto the skin without the polymer backing. A commer-cially available spray-on bandage is then adopted as adhesivesand encapsulants [110]. Other stretchable systems for biomedi-cal applications have also been reported, such as the CNT-basedstrain sensor for human motion monitoring [58], electronic skin[113], and other as reviewed in [126].A number of methods have been proposed for the fabricationof stretchable electronics, such as LED, conductors, electrodes,and thin-lm transistors, and can be classied into two groups:exploring innovative structures and engineering new materials.Rogers and coworkers [127] congured the structures into wavyshapessupportedbyelastomericsubstrate. Thestructurecanaccommodateupto20%30%compressiveandtensilestrain1548 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 5, MAY 2014by changing the wave amplitudes and wavelengths. This buck-ling approach has also been adopted to fabricate graphene rib-bonsonstretchableelastomers[128]. Rogersandcoworkersalsoproposedanoncoplanarmeshdesignwithactivedeviceislands connected by thin polymer bridges on elastomeric sub-strates to achieve 140% level of strain [129]. Recently, Whiteet al. [130] demonstrated an ultrathin, exible and stretchablepolymer-based light emitting diode (PLED), which can be fab-ricatedwithverysimpleprocessingmethod.Bypressedontoa prestrained elastomer, the ultrathin PLED formed a randomnetwork of folds when releasing the strain on the elastomer. Theexperiment showedthat theultrathinPLEDcanbestretchedup to 100% tensile strain. Meng et al. [131] designed and fab-ricatedtheassembledbersupercapacitorscomposedoftwointertwined graphene electrodes that showed highly compress-ible(strainof50%)andstretchable(strainof200%)proper-ties, andcanbeeasilywovenintotextilefor wearableap-plications. Newmaterialscanalsoprovidestretchabilityforelectronics. Someyas groupfabricatedacompositelmbyuniformlydispersingsingle-walledCNTs inpolymer matri-ces [132]. Other stretchableelectronics basedonnanocom-positessuchasgoldnanoparticle-polyurethanecompositeasconductor[133], grapheneoxidenanosheet-conductingpoly-mer as electrode [134] have also been proposed recently. Thesenanocomposites achieved both good electronic performance andmechanical robustness. These recent advancements in stretch-able electronics will facilitate the adoption of these technologiesin healthcare applications.To summarize, with the advances in material, textile, powerscavenging, sensingandwirelesscommunications, therehasbeen a rapid increase in the number of new wearable sensingdevices launched in the consumer market for sports, wellbeingand healthcare applications. Considering the growing interestsin these research areas, the number and variety of sensors willcontinue to grow in a very rapid rate. In addition, following thecurrent trend of sensing development, new type of implantablesensors, such as transient and zero-power implants, could soonbecome reality. Standards have been emerged aiming to stan-dardize these devices as well as the wireless communication andthe exchange of information among devices and systems.IV. SENSING INFORMATICS: DATA FUSION AND BIG DATAThe development of sensing technology has largely increasedthe capability of sensor to acquire data, and multiple sensors areexpected to provide different viewpoints of the health status ofthe patients. However, multisensor data fusion is one great chal-lenge because heterogeneous data need to be processed in orderto generate unied and meaningful conclusion for clinical diag-nosis and treatment [135]. The fusion of sensing data with otherhealth data such as imaging, bio-markers and gene sequencingand so on is even more challenging. The denitions of data fu-sion are different in the literature. In [136], data fusion is denedas a multilevel, multifaceted process handling the automatic de-tection, association, correlation, estimation, and combination ofdata and information from several sources". A comprehensivereview and discussion of data fusion denitions are presentedin[137].Weproposethedenitionofdatafusionas:tode-velop efcient methods for automatically or semi-automaticallytranslating the information from multiple sources into a struc-tured representation so that human or automated decision can bemade accurately". Data fusion is denitely a multidisciplinaryresearcharea,whichhasintegratedmanytechniques,suchassignal processing, information theory, statistical estimation andinference, and articial intelligence. In this section, we will dis-cuss multisensor fusion methods and the fusion of sensing datawith other types of health data for clinical decision support. Atthe end of this section, we will provide a critical on the impactof big data in the healthcare area.A. Data FusionSince most of health data are accompanied with a large num-ber of noisy, irrelevant and redundant information, which maygive spurious signals in clinical decision support, it is thereforenecessary to lter the data before fusion. To address this issue,rankedlistsofeventsorattributesclearlyrelevanttoclinicaldecision-making should be created [138]. Temporal reasoningmethod has been suggested for detecting associations betweenclinical entities [139]. More sophisticated methods such as con-textuallters[140],statisticalshrinkagetowardsthenullhy-pothesis of no association [141] were also proposed. How to l-ter information is denitely clinically meaningful, and it wouldbecomemoreandmoreimportantbutchallengingduetotheever-increasing data types and volumes.Multisensor fusion: A number of algorithms have been pro-posed in the literature for multisensor data fusion. Due to hetero-geneous nature of the data that need to be combined, differentdata fusion algorithms have been designed for different appli-cations. These algorithms can utilize different techniques froma wide range of areas, including articial intelligence, patternrecognition, statistical estimation, and other areas. Health infor-maticshasnaturallybenetedfromtheseabundantliterature.Forinstance, thesefusionalgorithmscanbecategorizedintothe following groups:1) Statistical approach: weighted combination, multivariatestatistical analysis and its most state-of-the-art data miningalgorithm [142]. Among all the statistical algorithms, thearithmeticmeanapproachisconsideredasthesimplestimplementationfordatafusion. However, thestatisticalapproach might not be suitable when the data is not ex-changeable or when estimators/classiers have dissimilarperformances [143].2) Probabilistic approach: MaximumlikelihoodmethodsandKalmanlter [144], probabilitytheory[145], evi-dentialreasoningandmorespecicallyevidencetheoryare widely used for multisensor data fusion. Kalman lteris considered as the most popular probabilistic fusion al-gorithm because of its simplicity, ease of implementation,and optimality in a mean-squared error sense. However,Kalman lter is inappropriate for applications whose errorcharacteristics are not readily parameterized.3) Articial intelligence: articial cognition includinggenetic algorithms and neural networks. In manyZHENG et al.: UNOBTRUSIVE SENSING AND WEARABLE DEVICES FOR HEALTH INFORMATICS 1549applications,thelaterapproachservesbothasatooltoderive classiers or estimators and as a fusion frameworkof classiers/estimators [142], [143].Fusion of sensing data with other health data: It is of vitalimportance to integrate sensing data with other health data, in-cluding genetic, medication, laboratory test, imaging data andnarrativereportsthatprovidecontextinformationforclinicaldiagnosis. Electronic health record (EHR) reposits these healthdata in a computer-readable form, which can be used for clinicaldecision support [146]. The heterogeneous contents and hugesize of EHR bring great challenges for the fusion of these data.Bayesiannetwork,neuralnetworks,associationrules,patternrecognitionandlogisticregressionhavebeenusedtoextractknowledgefromEHRforpatientstraticationandpredictivepurposes in the past years. These methods offer signicant ad-vantages to traditional statistical methods because they are ableto identify more complex relationships among variables. Acom-parison has been made among the three different classicationmethods in terms of their performance in predicting coronaryartery disease [147]. It was found that the multilayer perceptronbased neural network method showed the best performance inprediction [147].Although these studies have shown promising results aboutthe effectiveness of these computerized methods to be a com-plementarytool ofthepresent statistical modelsformedicaldecisionsupport, thesemethodsstill sufferfromsomeprob-lems, such as over tting, computationally expensive, and someof them are difcult to be interpreted by experts. Unlike thesecompletely data-driven approaches, fuzzy cognitive map (FCM)is a logical-rule-based method constructed fromhuman expertsknowledge. It models the decision systemas a collection of con-cepts and causal relationships among these concepts using fuzzylogic and neural networks [148]. It, therefore, can be interpretedby experts and provides transparent information to the experts.However, it could be subjective and may not be reliable by solelyrelying on experts knowledge to generate the rules. AnewFCMframework was proposed recently [149], which extracts fuzzyrules from both experts knowledge and data using rule extrac-tion methods. This novel framework for FCM system has beenused in the radiation therapy for prostate cancer.The early prediction of cardiovascular diseases (CVDs) hasbeen a very popular yet challenging research topic, and manyfusionworkshavebeenconductedinthisarea. Forinstance,in an European Commission funded project, euHeart Project, aprobabilistic fusion framework was developed to assimilate dif-ferent health data across scales (e.g., protein level ion channelsux and whole organ deformation) and functions (e.g., mechan-icalcontractionandelectricalactivation), intoapersonalizedmultiphysics cardiac model by minimizing on the discrepancybetween the measurements and the estimating derived from thecomputational model and then to discover new knowledge us-ing the personalized model [150]. By integrating continuous andreal-time sensing data from unobtrusive/wearable devices withthe existing health data, the real-time prediction of acute CVDevents may become possible. Zhangs team has proposed a per-sonalized framework for quantitative assessment of the risk ofacute CVD events based on vulnerable plaque rupturing mech-anism as shown in Fig. 9 [151]. This framework does not onlytaketraditional riskfactors, sensitivebiomarkers, bloodbio-chemistry, vascular morphology, plaque information, and func-tional image information as inputs of the prediction model, butalso gathers physiological information continuously fromunob-trusive devices and body sensor networks as the trigger factorstargeting for real-time risk assessment of acute cardiovascularevents.B. Impact of Big Health DataWiththeincreasingnumberofsensingmodalitiesandlowcost sensing devices becoming more accessible to wider pop-ulations, the amount and variety of health data have increasedrapidly, healthdataarethereforeconsideredasbigdata. Forexample,ifwecollectthehealthdatarelatedtocardiovascu-larsystem, includingECG, PPG, BPwaveform, BCG, EEG,EMG, PTT, cardiac output, SpO2, body temperature, and imag-inginformationfromcomputedtomography, magneticreso-nance imaging, ultrasound imaging and so on, the size of datacollectedfromallthepeopleintheSouthernChina,fromallthe people in the China, and from all over the world, are over4 zettabytes, 40 zettabyes, and over 200 zettabytes per year, re-spectively. As of year 2012, the size of data sets that are feasibleto process in a reasonable amount of time were in the order ofexabytes, which means the size of health data is over one thou-sand times larger than the current limits. Based on the 4 Vdenition of big data by Gartner, i.e., volume, velocity, varietyand value, the characteristics of health data can be summarizedas 6 V: vast, volume, velocity, variety, value, variation. Vastis the core feature of health data, i.e., the volume, velocity, va-riety, value and variation are vast. Volume refers to the size ofthe information. Because health data are collected from over ahuge amount of people simultaneously, the velocity of the col-lection of health data can be extremely high. A variety of healthdata from genetic or molecular level like gene sequences andbiomarkers to system levels such as physiologicalparametersand medical imaging data are now available. If we can build upa systematic analysis framework, we can denitely mine knowl-edge of great value from the health data. Since the biologicalsystem of the human being is dynamic and evolving, the healthdata must be variational.Due to the ongoing developments of network, mobile com-putingandcomputerstorage, thestorageandretrievalofbighealth data have attracted great attentions. With the increasinguse of various kinds of long-term monitoring devices, there isan urgent demand for the intelligent management of big healthdata. A large number of cloud-based storage and retrieval sys-tems are emerging in recent years. Through outsourcing the bighealth data, storage-as-a-service is an emerging solution to alle-viate the burden and high cost of large local data storage. Somenew storage strategies have been proposed, such as storage con-solidationandvirtualizationtooptimizethetradeoffbetweencomputation and storage capacity [152]. Retrieving specic in-formation from the cloud-based health data is also very chal-lenging. Query accuracy and time efciency are two importantmetricstoevaluatetheperformanceof theretrieval system.Extensive efforts have been made on optimizing the query com-plexity and expansion strategies to improve the performance of1550 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 5, MAY 2014Fig. 9. The framework for the quantitative assessment of the risk of acute cardiovascular events. Reproduced from [151].the retrieval systems [153], [154]. Recently, a standards-basedmodel has been proposed to provide a feasible solution for themanagement (i.e., collection, storage, retrieval and sharing) ofpatient-generated personal health data in EHR[155]. This modeldescribed the required components and minimal requirementsfordatacollectionandstorage,aswellasthemethodfortheexchange of health data between patients and EHR providers.Uptodate, anumberofhealthcareapplicationsbasedoncloud computing technology have been reported, such as stor-age and management of EHRs [156], automatic health data col-lection in health care institutions [157], intelligent emergencyhealth care [158], radiotherapy, etc. It will further provide op-portunities in clinical decision support (particularly for chronicdiseases), public health management (e.g., controlling of infec-tious diseases), development of new drugs, etc [159], [160].Althoughmanycloud-basedstorageandretrieval systemshaverecentlybeenproposed,fusiontechniquesforanalyzingbig health data are still in the conceptual stage, we believe thatwith the advances of computing power as well as the increasingavailabilityofmoreandmoreinteroperableelectronichealthrecords in the healthcare ecosystems, the intelligent utilizationof big health data will eventually become feasible in the nearfuture.V. CONCLUSION AND FUTURE DEVELOPMENTIn this paper, an overview of unobtrusive sensing platformseither in wearable form or integrated into environments is pre-sented. Although signicant progresses in developing these sys-temsforhealthcareapplicationshavebeenmadeinthepastdecades, most of them are still in their prototype stages. Issuessuch as user acceptance, reduction of motion artifact, lowpowerdesign, on-node processing, and distributed interference in wire-less networks still need to be addressed to enhance the usabilityand functions of these devices for practical use. Due to the mul-tidisciplinary nature of this research topic, future developmentwill greatly rely on the advances in a number of different areassuchasmaterials,sensing,energyharvesting,electronicsandinformation technologies for data transmission and analysis. Inthefollowing, weput forwardsomepromisingdirectionsonthe development of unobtrusive and wearable devices for futureresearch:1) To develop exible, stretchable and printable devices forunobtrusivephysiological andbiochemical monitoring:Researchonavarietyof semiconductor materials, in-cluding small-molecule organics and polymers, inorganicsemiconducting materials of different nanostructures, likenanotube, nanowire, andnanoribbonsandhybridcom-posite materials, could prosper the design of exible andstretchable sensors withhighoptical, mechanical andelectrical performance. Withthedevelopment of exi-ble, stretchableandprintableelectronics, wearablede-vices would evolve to be multifunctional electronic skinsor skin-attachabledevices, whichareverycomfortabletowear. Meanwhile, other applicationsof exibleandstretchable devices in wearable health monitoring shouldbe explored.2) To develop wearable physiological imaging platforms, es-pecially unobtrusive ones: Physiological information pro-videdbythepresent wearablesystemsaregenerallyinone-dimensional(1-D)format. Theextensionfrom1-Dto 2-D by including spatial information would be desir-able to provide more local information. Compared to theexistingnoninvasiveimagingmodalities(e.g., magneticresonance imaging, computedtomography, ultrasound,etc.), wearablephysiologicalimagingdeviceswouldbemuch faster to achieve high temporal resolution im-ages, which is expected to provide additional clinical andhealthinformationforearlydiagnosisandtreatment ofdiseases.ZHENG et al.: UNOBTRUSIVE SENSING AND WEARABLE DEVICES FOR HEALTH INFORMATICS 15513) To develop wearable devices for disease intervention: Theexisting wearable devices are mainly designed for the con-tinuous monitoring of a persons physiological or physicalstatus. For future development, wearable devices shouldalso play a role in disease intervention through integrationwith actuators that are implanted inside/on the body. Onewell-known example is the wearable articial endocrinepancreas for diabetes management, which is a close-loopsystem formed by a wearable glucose monitor and an im-planted insulin pump. Wearable drug delivery systems forBP management can also be developed in the future.4) To develop systematic data fusion framework: To integratethe multimodal and multiscale big health data from sens-ing, bloodtesting, bio-marker detection, structural andfunctional imagingfor thequantitativeriskassessmentand the early prediction of chronic diseases.It is believed that with these future developments, unobtru-sive and wearable devices could advance health informatics andlead to fundamental changes of how healthcare is provided andreform the underfunded and overstretched healthcare systems.REFERENCES[1] D. E. Bloom, E. T. Caero, E. Jan e-Llopis, S. Abrahams-Gessel,L. R. Bloom, S. Fathima, A. B. Feigl, T. Gaziano, M. Mowa,A. Pandya, K. Prettner, L. Rosenberg, B. Seligman, A. Z. Stein, andC. Weinstein, The Global Economic Burden of Non-Communicable Dis-eases. Geneva, Switzerland: World Economic Forum, Jul. 2011.[2] Y. T. Zhang and C. C. Y. Poon, Editorial note on the processing, storage,transmission,acquisition,andretrieval(P-STAR)ofbio,medical,andhealth information, IEEE Trans. Inf. Technol. Biomed., vol. 14, no. 4,pp. 895896, Jul. 2010.[3] C. C. Y. Poon, Y. M. Wong, and Y.-T. Zhang, M-Health: The develop-mentofcuff-lessandwearablebloodpressuremetersforuseinbodysensornetworks,inProc.IEEE/NLMLifeSci.Syst.Appl.Workshop,Bethesda, MD, 2006, pp. 12.[4] D. H. Kim, N. Lu, R. Ma, Y. S. Kim, R. H. Kim, S. Wang, J. Wu,S. M. Won, H. Tao, A. Islam, K. J. Yu, T.-i. Kim, R. Chowdhury,M. Ying, L. Z. Xu, M. Li, H. J. Chung, H. Keum, M. McCormick, P. Liu,Y. W. Zhang, F. G. Omenetto, Y. G. Huang, T. Coleman, and J. A. Rogers,Epidermal electronics, Science, vol. 333, no. 6044, pp. 838843, Aug.2011.[5] Y. T. Zhang and C. C. Y. Poon, Health informatics: Unobtrusive physi-ological measurement technologies, IEEE J. Biomed. Health Informat.,vol. 17, no. 5, p. 893, Sep. 2013.[6] K. F. Wu, C. H. Chan, and Y. T. Zhang, Contactless and cufess moni-toring of blood pressure on a chair using e-textile materials, in Proc. 3rdIEEE/EMBS Int. Summer School Med. Devices Biosensors, Cambridge,MA, 2006, pp. 98100.[7] H. Baek, G. Chung, K. Kim, and K. Park, A smart health monitoringchair for nonintrusive measurement of biological signals, IEEE Trans.Inf. Technol. Biomed., vol. 16, no. 1, pp. 150158, Jan. 2012.[8] M. Walter, B. Eilebrecht, T. Wartzek, and S. Leonhardt, The smart carseat: Personalized monitoring of vital signs in automotive applications,Pers. Ubiquitous Comput., vol. 15, no. 7, pp. 707715, Oct. 2011.[9] W. B. Gu, C. C. Y. Poon, H. K. Leung, M. Y. Sy, M. Y. M. Wong, andY. T. Zhang, A novel method for the contactless and continuous mea-surement of arterial blood pressure on a sleeping bed, in Proc. Annu.Int. Conf. IEEE Eng. Med. Biol. Soc., 2009, pp. 60846086.[10] M. Z. Poh, D. McDuff, and R. Picard, A medical mirror for non-contacthealth monitoring, in Proc. ACM SIGGRAPH 2011 Emerging Technol.,Vancouver, BC, Canada, 2011, pp. 11.[11] J. Gomez-Clapers and R. Casanella, A fast and easy-to-use ECG acqui-sition and heart rate monitoring system using a wireless steering wheel,IEEE Sens. J., vol. 12, no. 3, pp. 610616, Mar. 2012.[12] Y.Lin,Anaturalcontactsensorparadigmfornonintrusiveandreal-time sensing of biosignals in human-machine interactions, IEEE Sens.J., vol. 11, no. 3, pp. 522529, Mar. 2011.[13] K. Ko Keun, L. Yong Kyu, and P. Kwang-Suk, The electrically non-contacting ECG measurement on the toilet seat using the capacitively-coupled insulated electrodes, in Proc. 26th Annu. Int. Conf. IEEE Eng.Med. Biol. Soc., San Francisco, CA, USA, 2004, pp. 23752378.[14] O. T. Inan, D. Park, L. Giovangrandi, and G. T. A. Kovacs, Noninvasivemeasurement ofphysiological signalsonamodiedhomebathroomscale, IEEE Trans. Biomed. Eng., vol. 59, no. 8, pp. 21372143, Aug.2012.[15] Y. M. Chi, S. R. Deiss, and G. Cauwenberghs, Non-contact low powerEEG/ECG electrode for high density wearable biopotential sensor net-works, in Proc. 6th Int. Workshop Wearable Implantable Body SensorNetw., Berkeley, CA, 2009, pp. 246250.[16] T. Hoffmann, B. Eilebrecht, and S. Leonhardt, Respiratory monitoringsystem on the basis of capacitive textile force sensors, IEEE Sens. J.,vol. 11, no. 5, pp. 11121119, 2011.[17] T. Wartzek, S. Weyer, and S. Leonhardt, A differential capacitive elec-trical eld sensor array for contactless measurement of respiratory rate,Physiol. Meas., vol. 32, no. 10, pp. 15751590, Oct. 2011.[18] P. Chulsung, P. H. Chou, B. Ying, R. Matthews, and A. Hibbs, An ultra-wearable, wireless, low power ECG monitoring system, in Proc. IEEEBiomed. Circuits Syst. Conf., London, U.K., 2006, pp. 241244.[19] A. Aleksandrowicz and S. Leonhardt, Wireless and non-contact ECGmeasurement systemThe Aachen smartchair, Acta Polytechnica,vol. 2, pp. 6871, 2007.[20] Y. Yama, A. Ueno, and Y. Uchikawa, Development of a wireless ca-pacitive sensor for ambulatory ECG monitoring over clothes, in Proc.29th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., Lyon, France, 2007,pp. 57275730.[21] S. Fuhrhop, S. Lamparth, andS. Heuer, Atextileintegratedlong-term ECG monitor with capacitively coupled electrodes, in Proc. IEEEBiomed. Circuits Syst. Conf., Beijing, China, 2009, pp. 2124.[22] Y. M. Chi and G. Cauwenberghs, Wireless non-contact EEG/ECG elec-trodes for body sensor networks, in Proc. Int. Conf. Body Sensor Netw.,2010, Singapore, pp. 297301.[23] E. Nemati, M. J. Deen, and T. Mondal, Awireless wearable ECGsensorfor long-term applications, IEEE Commun. Mag., vol. 50, no. 1, pp. 3643, Jan. 2012.[24] P. GuoChen and M. F. Bocko, Non-contact ECGsensing employing gra-diometer electrodes, IEEE Trans. Biomed. Eng., vol. 60, no. 1, pp. 179183, Jan. 2013.[25] B. H. Jae, C. G. Sung, K. K. Keun, K. J. Soo, andP. K. Suk, Photo-plethysmogram measurement without direct skin-to-sensor contact us-ing an adaptive light source intensity control, IEEE Trans. Inf. Technol.Biomed., vol. 13, no. 6, pp. 10851088, Nov. 2009.[26] M. Z. Poh, D. J. McDuff, andR. W. Picard, Advancementsinnon-contact, multiparameter physiological measurements using a webcam,IEEE Trans. Biomed. Eng., vol. 58, no. 1, pp. 711, Jan. 2011.[27] H. H. Asada, P. Shaltis, A. Reisner, R. Sokwoo, and R. C. Hutchinson,Mobile monitoring with wearable photoplethysmographic biosensors,IEEE Eng. Med. Biol., vol. 22, no. 3, pp. 2840, 2003.[28] P. Celka, C. Verjus, R. Vetter, P. Renevey, andV. Neuman, Motionresistant earphone located infrared based heart rate measurement device,in Proc. 2nd Int. Conf. Biomed. Eng., Innsbruck, Austria, 2004, pp. 582585.[29] Y. Mendelson, R. J. Duckworth, and G. Comtois, Awearable reectancepulse oximeter for remote physiological monitoring, in Proc. 28th Annu.Int. Conf. IEEE Eng. Med. Biol. Soc., New York, NY, 2006, pp. 912915.[30] W. Lei, B. Lo, and G.-Z. Yang, Multichannel reective PPG earpiecesensor with passive motion cancellation, IEEE Trans. Biomed. CircuitsSyst., vol. 1, no. 4, pp. 235241, Dec. 2007.[31] S. Vogel, M. Hulsbusch, T. Hennig, V. Blazek, and S. Leonhardt, In-earvital signs monitoring using a novel microoptic reective sensor, IEEETrans. Inf. Technol. Biomed., vol. 13, no. 6, pp. 882889, Nov. 2009.[32] J. Spigulis, R. Erts, V. Nikiforovs, andE. Kviesis-Kipge, Wearablewireless photoplethysmography sensors, in Proc. Photon. Eur., 2008,pp. 69912O169912O7.[33] J. A. C. Patterson, D. G. McIlwraith, and G.-Z. Yang, A exible, lownoise reective PPGsensor platformfor ear-worn heart rate monitoring,inProc. 6thInt. WorkshopWearableImplantableBodySensorNetw.,Berkeley, CA, 2009, pp. 286291.[34] K. Shin, Y. Kim, S. Bae, K. Park, and S. Kim, A novel headset withatransmissivePPGsensorforheartratemeasurement,inProc.13thIntl. Conf. Biomed. Engineering, vol. 23, C. LimandJ. H. Goh, Eds.Berlin, Germany: Springer, 2009, pp. 519522.1552 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 5, MAY 2014[35] M.-Z. Poh, K. Kim, A. Goessling, N. Swenson, andR. Picard, Car-diovascular monitoringusingearphonesandamobiledevice,IEEEPervasive Comput., vol. 11, no. 4, pp. 1826, Oct.Dec. 2012.[36] M. Z. Poh, N. C. Swenson, and R. W. Picard, Motion-tolerant magneticearringsensorandwirelessearpieceforwearablephotoplethysmogra-phy,IEEETrans.Inf.Technol.Biomed.,vol.14,no.3,pp.786794,May 2010.[37] Y. L. Zheng, B. Leung, S. Sy, Y. T. Zhang, and C. C. Y. Poon, A clip-free eyeglasses-based wearable monitoring device for measuring photo-plethysmograhic signals, in Proc. 34th Annu. Int. Conf. IEEE Eng. Med.Biol. Soc., San Diego, CA, 2012, pp. 50225025.[38] Y. H. Lin, C. F. Lin, H. Z. You et al., A drivers physiological mon-itoringsystembasedonawearablePPGsensor andasmartphone,in Security-Enriched Urban Computing and Smart Grid, vol. 223,R.-S. Chang, T.-H. Kim, andS.-L. Peng, Eds. Berlin, Germany:Springer, 2011, pp. 326335.[39] F.Lamonaca,Y.Kurylyak,D.Grimaldi,andV.Spagnuolo,Reliablepulserateevaluationbysmartphone,inProc. IEEEInt. Symp. Med.Measurements Appl. Proc., Budapest, Hungary, 2012, pp. 14.[40] C. C. Poon, Y.-T. Zhang, andY. Liu, Modelingofpulsetransittimeunder the effects of hydrostatic pressure for cufess blood pressure mea-surements, in Proc. 3rd IEEE/EMBS Int. Summer School Med. DevicesBiosensors, 2006, pp. 6568.[41] C. C. Y. Poon and Y. T. Zhang, Cuff-less and noninvasive measurementsof arterial blood pressure by pulse transit time, in Proc. 27th Annu. Int.Conf. IEEEEng. Med. Biol. Soc., Shanghai, China, 2005, pp. 58775880.[42] R. Payne, C. Symeonides, D. Webb, and S. Maxwell, Pulse transit timemeasuredfromtheECG: Anunreliablemarkerofbeat-to-beat bloodpressure, J. Appl. Phys., vol. 100, no. 1, p. 136, 2006.[43] A. Lanata, E. P. Scilingo, E. Nardini, G. Loriga, R. Paradiso, and D. De-Rossi, Comparativeevaluationofsusceptibilitytomotionartifact indifferent wearable systems for monitoring respiratory rate, IEEE Trans.Inf. Technol. Biomed., vol. 14, no. 2, pp. 378386, Mar. 2010.[44] R. Paradiso, G. Loriga, and N. Taccini, A wearable health care systembased on knitted integrated sensors, IEEE Trans. Inf. Technol. Biomed.,vol. 9, no. 3, pp. 337344, Sep. 2005.[45] S. Rajala and J. Lekkala, Film-type sensor materials PVDF and EMFiin measurement of cardiorespiratory signals: A review, IEEE Sens. J.,vol. 12, no. 3, pp. 439446, Mar. 2012.[46] S. Karki and J. Lekkala, Anewmethod to measure heart rate with EMFiand PVDF materials, J. Med. Eng. Technol., vol. 33, no. 7, pp. 551558,2009.[47] Z. B. Zhang, Y. H. Shen, W. D. Wang, B. Q. Wang, andJ. W. Zheng,Design and implementation of sensing shirt for ambulatory cardiopul-monarymonitoring,J. Med. Biol. Eng., vol.31,no.3,pp.207215,2011.[48] A. Lanata, E. P. Scilingo, and D. De Rossi, A multimodal transducer forcardiopulmonaryactivitymonitoringinemergency,IEEETrans. Inf.Technol. Biomed., vol. 14, no. 3, pp. 817825, May 2010.[49] L. Dziuda, F. W. Skibniewski, M. Krej, and J. Lewandowski, Monitoringrespiration and cardiac activity using ber bragg grating-based sensor,IEEE Trans. Biomed. Eng., vol. 59, no. 7, pp. 19341942, Jul. 2012.[50] L. T. DAngelo, S. Weber, Y. Honda, T. Thiel, F. Narbonneau, andT. C. L uth, Asystemfor respiratorymotiondetectionusingopticalbers embedded into textiles, in Proc. 30th Ann. Int. Conf. IEEE Eng.Med. Biol. Soc., Vancouver, BC, Canada, 2008, pp. 36943697.[51] O. Postolache, P. S. Girao, R. N. Madeira, andG. Postolache, Mi-crowave FMCW Doppler radar implementation for in-house pervasivehealth care system, in Proc. IEEE Int. Workshop Med. MeasurementsAppl.,, Ottawa, ON, Canada, 2010, pp. 4752.[52] Q. Bonds, J. Gerig, T. M. Weller, and P. Herzig, Towards core body tem-perature measurement via close proximity radiometric sensing, IEEESens. J., vol. 12, no. 3, pp. 519526, Mar. 2012.[53] H. M. Schepers, H. F. J. M. Koopman, and P. H. Veltink, Ambulatoryassessment ofankleandfoot dynamics,IEEETrans. Biomed. Eng.,vol. 54, no. 5, pp. 895902, May 2007.[54] J. Penders, J. van de Molengraft, M. Altini, F. Yazicioglu, and C. VanHoof, A low-power wireless ECG necklace for reliable cardiac activitymonitoring on-the-move, in Proc. Int. Conf. IEEE Eng. Med. Biol. Soc.,2011, pp. 14.[55] W. B. Gu, C. C. Y. Poon, M. Y. Sy, H. K. Leung, Y. P. Liang, andY. T. Zhang, A h-shirt-based body sensor network for cufess calibra-tion and estimation of arterial blood pressure, in Proc. 6th Int. WorkshopWearable Implantable Body Sensor Netw., Berkeley, CA, 2009, pp. 151155.[56] B. Lo, G.-Z. Yang, andJ. Pansiot, Bayesiananalysisofsub-plantargroundreactionforcewithBSN,inProc. 6thInt. WorkshopWear-ableImplantableBodySensorNetw., Berkeley, CA, 2009, pp. 133137.[57] M. Rothmaier, B. Selm, S. Spichtig, D. Haensse, and M. Wolf, Photonictextiles for pulse oximetry, Opt. Exp., vol. 16, no. 17, pp. 1297312986,Aug. 2008.[58] T. Yamada, Y. Hayamizu, Y. Yamamoto, Y. Yomogida, A. Izadi-Najafabadi, D. N. Futaba, and K. Hata, A stretchable carbon nanotubestrainsensorforhuman-motiondetection,Nat. Nanotechnol., vol. 6,no. 5, pp. 296301, May 2011.[59] Y. C. Wu, C. S. Chang, Y. Sawaguchi, W. C. Yu, M. J. Chen, J. Y. Lin,S. M. Liu, C. C. Han, and W. Liang, Amobile-phone-based health man-agement system,inHealthManagementDifferent ApproachesandSolutions, K.Smig orski, Ed.: Rijeka, Croatia: InTech, 2011.[60] L. Atallah, B. Lo, R. King, and G.-Z. Yang, Sensor positioning for ac-tivity recognition using wearable accelerometers, IEEE Trans. Biomed.Circuits Syst., vol. 5, no. 4, pp. 320329, Aug. 2011.[61] P. B. Adamson, W. T. Abraham, M. Aaron, J. M. Aranda, Jr.,R. C. Bourge, A. Smith, L. W. Stevenson, J. G. Bauman, and J. S. Yadav,CHAMPION trial rationale and design: The long-term safety and clini-cal efcacy of a wireless pulmonary artery pressure monitoring system,J. Card. Fail., vol. 17, no. 1, pp. 310, 2011.[62] J. E. Pandolno, J. E. Richter, T. Ours, J. M. Guardino, J. Chapman,and P. J. Kahrilas, Ambulatory esophageal pHmonitoringusing awirelesssystem,Am. J. Gastroenterol., vol. 98, no. 4, pp. 740749,2003.[63] C. C. Y. Poon and Y. T. Zhang, Perspectives on high technologies forlow-cost healthcare, IEEE Eng. Med. Biol., vol. 27, no. 5, pp. 4247,Oct. 2008.[64] C. M. N. Brigante, N. Abbate, A. Basile, A. C. Faulisi, andS. Sessa,TowardsminiaturizationofaMEMS-basedwearablemotioncapturesystem, IEEE Trans. Ind. Electron., vol. 58, no. 8, pp. 32343241, Aug.2011.[65] Y. Chuo, M. Marzencki, B. Hung, C. Jaggernauth, K. Tavakolian, P. Lin,and B. Kaminska, Mechanically exible wireless multisensor platformfor human physical activity and vitals monitoring, IEEE Trans. Biomed.Circuits Syst., vol. 4, no. 5, pp. 281294, Oct. 2010.[66] S. Mandal, L. Turicchia, and R. Sarpeshkar, A low-power, battery-freetag for body sensor networks, IEEE Pervasive Comput., vol. 9, no. 1,pp. 7177, Mar. 2010.[67] C. R. Merritt, H. T. Nagle, and E. Grant, Fabric-based active electrodedesign and fabrication for health monitoring clothing, IEEE Trans. Inf.Technol. Biomed., vol. 13, no. 2, pp. 274280, Mar. 2009.[68] R. S. H. Istepanian, E. Jovanov, and Y. T. Zhang, Guest editorial intro-duction to the special section on m-Health: Beyond seamless mobilityand global wireless health-care connectivity, IEEE Trans. Inf. Technol.Biomed., vol. 8, no. 4, pp. 405414, Dec. 2004.[69] G. Z. Yang, Body Sensor Networks. London, U.K.: Springer, 2006.[70] N. Golmie, D. Cypher, andO. R ebala, Performanceanalysisoflowrate wireless technologies for medical applications, Comput. Commun.,vol. 28, no. 10, pp. 12661275, 2005.[71] K. Kang, J. Ryu, J. Hur, and L. Sha, Design and QoS of a wireless sys-tem for real-time remote electrocardiography, IEEE J. Biomed. HealthInformat., vol. 17, no. 3, pp. 745755, May 2013.[72] E. Cavero, A. Alesanco, and J. Garcia, Enhanced protocol forreal-time transmission of echocardiograms over wireless channels,IEEETrans. Biomed. Eng., vol. 59, no. 11, pp. 32123220, Nov.2012.[73] R. S. H. Istepanian, N. Y. Philip, and M. G. Martini, Medical QoS pro-visionbasedonreinforcement learninginultrasoundstreamingover3.5Gwirelesssystems,IEEEJ. Sel. AreasCommun., vol. 27, no. 4,pp. 566574, May 2009.[74] Y. Ren, R. W. N. Pazzi, and A. Boukerche, Monitoring patients via asecure and mobile healthcare system, IEEE Wirel. Commun., vol. 17,no. 1, pp. 5965, Feb. 2010.[75] K. Malhotra, S. Gardner, and R. Patz, Implementation of elliptic-curvecryptographyonmobilehealthcaredevices,inProc.IEEEInt.Conf.Netw., Sens. Control, London, U.K., 2007, pp. 239244.[76] C. C. Tan, H. Wang, S. Zhong, and Q. Li, Body sensor network security:anidentity-basedcryptographyapproach,presentedat the1st ACMConf. Wireless Netw. Security, Alexandria, VA, USA, 2008.[77] C. C. Poon, Y. T. Zhang, andS. D. Bao, Anovelbiometricsmethodto secure wireless body area sensor networks for telemedicine and m-health, IEEE Commun. Mag., vol. 44, no. 4, pp. 7381, Apr. 2006.ZHENG et al.: UNOBTRUSIVE SENSING AND WEARABLE DEVICES FOR HEALTH INFORMATICS 1553[78] K. K. Venkatasubramanian, A. Banerjee, andS. K. S. Gupta, PSKA:Usable and secure key agreement scheme for body area networks, IEEETrans. Inf. Technol. Biomed., vol. 14, no. 1, pp. 6068, Jan. 2010.[79] K. K. Venkatasubramanian, A. Banerjee, and S. K. S. Gupta,Plethysmogram-based secure inter-sensor communication in body areanetworks, in Proc. IEEE Mil. Commun. Conf., San Diego, CA, 2008,pp. 17.[80] P. K. Sahoo, EfcientsecuritymechanismsformHealthapplicationsusing wireless body sensor networks, Sensors, vol. 12, no. 9, pp. 1260612633, 2012.[81] X. H. Le, R. Sankar, M. Khalid, and S. Lee, Public key cryptography-basedsecurityschemeforwirelesssensornetworksinhealthcare,inProc. 4th Int. Conf. Uniquitous Informat. Manage. Commun., 2010, p. 5.[82] B. Liu, X. F. Wang, B. Y. Liu, Q. F. Wang, D. S. Tan, W. F. Song,X. J. Hou, D. Chen, and G. Z. Shen, Advanced rechargeable lithium-ion batteries based on bendable ZnCo2O4-urchins-on-carbon-bers elec-trodes, Nano Res., vol. 6, no. 7, pp. 525534, Jul. 2013.[83] Y. Fu, X. Cai, H. Wu, Z. Lv, S. Hou, M. Peng, X. Yu, and D. Zou, Fibersupercapacitors utilizing pen ink for exible/wearable energy storage,Adv. Mater., vol. 24, no. 42, pp. 57135718, 2012.[84] H. Xu, J. Li, B. H. K. Leung, C. C. Y. Poon, B. S. Ong, Y. Zhang, andN. Zhao, A high-sensitivity near-infrared phototransistor based on or-ganic bulk heterojunction, Nanoscale, vol. 5, no. 23, pp. 1185011855,2013.[85] A. A. Abidi, G. J. Pottie, and W. J. Kaiser, Power-conscious design ofwireless circuits and systems, Proc. IEEE, vol. 88, no. 10, pp. 15281545, Oct. 2000.[86] N. S. Shenck and J. A. Paradiso, Energy scavenging with shoe-mountedpiezoelectrics, IEEE Micro., vol. 21, no. 3, pp. 3042, MayJun. 2001.[87] V. Leonov, Thermoelectric energy harvesting of human body heat forwearablesensors,IEEESens.J.,vol.13,no.6,pp.22842291,Jun.2013.[88] S. Kim, J. S. Ho, andA. S. Y. Poon, Mideldwirelesspoweringofsubwavelength autonomous devices, PhRvL, vol. 110, no. 20, p. 203905,2013.[89] G. P. Pizzuti, S. Cifaldi, and G. Nolfe, Digital sampling rate and ECGanalysis, J. Biomed. Eng., vol. 7, no. 3, pp. 247250, 1985.[90] H. Feichtinger, J. Prncipe, J. Romero, A. Singh Alvarado, andG.Velasco,Approximatereconstructionofbandlimitedfunctionsfortheintegrateandresampler,Adv. Comput. Math., vol. 36, no. 1,pp. 6778, 2012.[91] A. S. Alvarado, C. Lakshminarayan, andJ. C. Principe, Time-basedcompression and classication of heartbeats, IEEETrans. Biomed. Eng.,vol. 59, no. 6, pp. 16411648, Jun. 2012.[92] P. K. K. Loh and L. Allan, Medical informatics system with wirelesssensor network-enabled for hospitals, in Proc. Int. Conf. Intell. Sensors,Sensor Netw. Inf. Process. Conf., 2005, pp. 265270.[93] Health Informatics Personal Health Device Communication, IEEE Stan-dard P11073, 2007[94] J. Yao and S. Warren, Applying the ISO/IEEE 11073 standards to wear-able home health monitoring systems, J. Clin. Monitor. Comp., vol. 19,no. 6, pp. 427436, Dec. 2005.[95] M. Damas, H. Pomares, S. Gonzalez, A. Olivares, and I. Rojas, Ambi-ent assisted living devices interoperability based on OSGi and the X73standard, Telemed. E-Health, vol. 19, no. 1, pp. 5460, 2013.[96] M. Martnez-Espronceda, I. Martnez, L. Serrano, S. Led, J. U. D. Trigo,A. Marzo, J. Escayola, and J. E. Garca, Implementation methodologyforinteroperablepersonalhealthdeviceswithlow-voltagelow-powerconstraints, IEEE Trans. Inf. Technol. Biomed., vol. 15, no. 3, pp. 398408, 2011.[97] I. R. Yan, C. C. Y. Poon, and Y. T. Zhang, Evaluation scale to assess theaccuracyofcuff-lessbloodpressuremeasuringdevices,BloodPressMonit., vol. 14, no. 6, p. 257, 2009.[98] F. Buttussi and L. Chittaro, MOPET: Acontext-aware and user-adaptivewearable system for tness training, Artif. Intell. Med., vol. 42, no. 2,pp. 153163, Feb. 2008.[99] S. L. Koay, T. H. Kwon, and W. Y. Chung, Motion artifacts cancellationbyadaptiveneuro-fuzzyinferencesystemforwearablehealthshirts,Sens. Lett., vol. 9, no. 1, pp. 399403, Feb. 2011.[100] B. H. Ko, T. Lee, C. Choi, Y. H. Kim, G. Park, K. Kang, S. K. Bae,andK. Shin, Motionartifact reductioninelectrocardiogramusingadaptivelteringbasedonhalf cell potential monitoring,inProc.2012Annu. Int. Conf. IEEEEng. Med. Biol. Soc., 2012, pp. 15901593.[101] Y. S. YanandY. T. Zhang, Anefcient motion-resistant methodforwearable pulse oximeter, IEEE Trans. Inf. Technol. Biomed., vol. 12,no. 3, pp. 399405, May 2008.[102] S. Coyle, K. T. Lau,