Privacy-Preserving Electrocardiogram Monitoring for Intelligent Arrhythmia Detection

21 Pages·2017·3.62 MB·English

sensors
Article
Privacy-Preserving Electrocardiogram Monitoring for
†
Intelligent Arrhythmia Detection
JunggabSon1,JuyoungPark2,HeekuckOh3,MdZakirulAlamBhuiyan4,JunbeomHur5,*
andKyungtaeKang3,*
1 DepartmentofComputerScience,KennesawStateUniversity,Marietta,GA30060,USA;[email protected]
2 SustainableManagementStrategy,KoreaExpresswayCorporation,Gimcheon39660,Korea;
[email protected]
3 DepartmentofComputerScienceandEngineering,HanyangUniversity,Ansan15588,Korea;
[email protected]
4 DepartmentofComputerandInformationSciences,FordhamUniversity,Bronx,NY10458,USA;
[email protected]
5 DepartmentofComputerScienceandEngineering,KoreaUniversity,Seoul02841,Korea
* Correspondence:[email protected](J.H.);[email protected](K.K.);Tel.:[phone](K.K.)
† ThispaperisanextendedversionofourpaperpublishedinPark,J.;Lee,K.;Kang,K.Intelligent
ElectrocardiogramMonitoringSystemforEarlyArrhythmiaDetection.InProceedingsoftheIEEE
InternationalConferenceonAdvancedInformationNetworkingandApplications(AINA2014),Victoria,
BC,Canada,13–16June2016.
Received:11April2017;Accepted:7June2017;Published:12June2017
Abstract: Long-term electrocardiogram (ECG) monitoring, as a representative application
of cyber-physical systems, facilitates the early detection of arrhythmia. A considerable number
of previous studies has explored monitoring techniques and the automated analysis of sensing
data. However,ensuringpatientprivacyorconfidentialityhasnotbeenaprimaryconcerninECG
monitoring. First,weproposeanintelligentheartmonitoringsystem,whichinvolvesapatient-worn
ECG sensor (e.g., a smartphone) and a remote monitoring station, as well as a decision support
serverthatinterconnectsthesecomponents. Thedecisionsupportserveranalyzestheheartactivity,
usingthePan–Tompkinsalgorithmtodetectheartbeatsandadecisiontreetoclassifythem. Our
system protects sensing data and user privacy, which is an essential attribute of dependability,
by adopting signal scrambling and anonymous identity schemes. We also employ a public key
cryptosystemtoenablesecurecommunicationbetweentheentities. Simulationsusingdatafrom
theMIT-BIHarrhythmiadatabasedemonstratethatoursystemachievesa95.74%successratein
heartbeat detection and almost a 96.63% accuracy in heartbeat classification, while successfully
preservingprivacyandsecuringcommunicationsamongtheinvolvedentities.
Keywords:bodysensornetworks;biomedicalcomputing;electrocardiography;arrhythmiadetection;
communicationsystemsecurity;privacyofpatients
1. Introduction
Cyberphysicalsystemshaveemergedasapromisingparadigmforenrichingtheinteractions
between physical and cybernetic components. Recent advances in sensing technology and smart
devices,whicharethemostimportantdevicesfacilitatingcyber-physicalsystems,havedrastically
alteredtheshapeofthecurrenthealthcareenvironment,whilepresentingnumerousopportunities
andchallengesinpatientmonitoringandassistance. Thisnovelparadigmenablespatientstomonitor
their physical conditions using smart devices [1], which has been particularly useful for chronic
diseasesthatcanbecomelifethreatening,suchashighbloodpressure,hypernatremiaandvarious
Sensors2017,17,1360;doi:10.3390/s17061360 www.mdpi.com/journal/sensors
Sensors2017,17,1360 2of21
heartdiseasesthatarecommonglobally[2]. Bymonitoringachronicdisease,apatientcandealwith
thistransformationatanearlystage. Inparticular, monitoringarrhythmiasisofhighimportance,
becausetheyareanextremelycommoninitialsymptomofcardiacarrestormyocardialinfarction.
Arrhythmiacanbedetectedbyanalyzinganelectrocardiogram(ECG),whichmeasurestherate
andregularityofheartbeatsandischaracterizedbyfivepeaksandvalleyslabeledP,Q,R,SandT
(seeFigure1). TheamplitudeanddurationoftheP-Q-R-S-Twaveprovidesinformationregardingthe
heartdisease. ThepositionanddistanceofthePRintervalandsegment,theSTintervalandsegment
andtheQTintervalandQRScomplexcanbeusedinadiagnosis[3].
RR interval
QRS complex
R R
PR ST
segment segment
T
P P
Q Q
S PR interval S
QT interval
Figure1.StructureofECG.
SeveralsuccessfulECGmonitoringsystemshavebeendeveloped[2–6]. Thesesystemspossess
manyadvantages,suchasallowingforambulatorypatientcare,resultinginsaferandmoreaffordable
healthcare.However,ECGsareprimarilyinterpretedbymedicalexperts,andmostpatientsthemselves
cannotobtaintheinformationfirsthand. Thus,considerableattentionhasrecentlybeendevotedtothe
computerizedautomaticanalysisofheartactivity,whichincludesboththedetectionoftheheartbeat
inanelectrocardiogramandtheclassificationofitstype. Recentadvancesinportabledevices,such
assmartphonesandtabletPCs,alsomakeitpossibleforuserstomoreeasilyself-monitortheirECG
statusandclassificationresults[4].
In orderto exploitthis potential, weproposea noveland intelligentECG monitoringsystem
(I-ECG)thatanalyzesandinterpretsheartactivitybyautomaticallydetectingandclassifyingheartbeats
based on the Pan–Tompkins [7] and C4.5 algorithms [8–10]. Although this was developed to be
effective,simpletouse,sustainableandreliable,itcannotbedirectlyappliedinreal-worldapplications,
owing to the following security and privacy problems. First, individuals close to the patients can
illegallyobtaintheECGsignalduringthecommunicationbetweenthesensorsandthesmartphone.
Second,adecisionsupportserver(DSS)canmanipulatecollectedECGsignals,aswellasthestatus
ofaparticular user. The health information that can be obtained through these security flaws is
sensitivepersonaldata,andthus,itcouldbeabusedformonetarypurposes. Inaddition,theprivacy
ofpatientscouldbeinvadedbytheabuseofthisinformation.
Itshouldbenotedthatmanyarticlestreattheprivacyofpatientsasaprimaryconcernintheusage
of personal healthcare systems [11–13], and some of these assert that patient privacy should be
preservedatalltimesbylaw[12]. Therefore,healthcaresystemsmustbedesignedwithconsideration
giventothesecurity,inordertoprovidejustifiablytrustedservices. Simultaneously,thesystemshould
bedesignedtosafeguardtheprivacyofpatients.
Contributions: Themainobjectiveofthisstudyistodevelopaprivacy-preservingI-ECGthat
detectsarrhythmiaintheearlystage. Thecontributionscanbesummarizedasfollows:
• We outline the guidelines for an intelligent ECG monitoring system and service model,
andproposeanalgorithmforautomaticdetectionandclassificationofarrhythmia,whichrealize
intelligentmonitoringasadecisionsupportsystem.
Sensors2017,17,1360 3of21
• AdditionalcomputationsperformedbythebodysensorstoenableencryptionofECGdatato
protectitfromattackersgenerateheatonthesensors,whichcanresultindiscomfortforthepatient.
Toaddressthisproblem,wehavedevelopedanefficientandeffectivesolutionforencryption
ofECGdatainthebodysensors. Thissolutionselectsthecipherwithminimaloverheadamong
theciphersconsideredascandidatesforourenvironment. Inaddition,oursolutionprotectsthe
patients’privacyagainstattackers,includingprotectionofdataattheDSS,aswellasmonitoring
stationsbyadoptingchangeablepseudonyms.
• WesimulatedourdetectionandclassificationalgorithmfordecisionsupportusingtheMIT-BIH
arrhythmiadatabase[14],andthesimulationresultsdemonstratethattheDSSachievedanoverall
accuracyof95.74%,withasensitivityof97.21%andaspecificityof94.26%forheartbeatdetection.
TheDSSalsoachievedanoverallaccuracyof96.63%,withasensitivityof95.44%andaspecificity
of97.82%forheartbeatclassificationincludingprivacypreservationmeasures.
• We demonstrate that our scheme exhibits a high efficiency compared with conventional
cryptographicalgorithmsandanenhancedrobustnessagainstinsideattackerswhohaveaccess
totheDSS,aswellasagainstoutsideattackerswhohavetheabilitytoeavesdropondatafrom
wirelesscommunications.
The remainder of this paper is organized as follows. In Section 2, we briefly review related
workregardingECGmonitoringsystemsandapplications,withthefocusonsecurityandprivacy.
In Section 3, we describe the system models, threat models and security objectives. We describe
our primary contribution, which is the privacy-preserving ECG monitoring system, in Section 4.
In Section 5, we outline our system implementation. In Section 6, weevaluate and simulate our
proposedsystem. Finally,weprovideourconclusionsinSection7.
2. RelatedWork
Since the1970s, researchers worldwidehave been developingdiagnostic systemsthat enable
patientstomakeECGrecordingsathomeandtransferdatatoacardiologist[2–6]. Researchregarding
remote ECG monitoring systems continues today, with smartphones being employed. Essentially,
remoteECGmonitoringsystems[5,15,16]requireonlyasimplearchitecturebetweenthesensordevice
andasmartphone. ThesystemextractsECGsignalsthroughaleadcable,transmitsthesignaldata
viaBluetoothandthenprocessesanddisplaystheECGwaveformonapersonalcomputer[15]or
smartphone [5,16]. The system displays the ECG waveform on the device for remote monitoring.
However,itdoesnottransmittheECGwaveformtoamedicalexpert. Therefore,thepatienteffectively
receivesnoreal-timemedicalservice.
Systemsonahealthcareserverexpandthefundamentalremotemonitoringsystembyoffering
anoptiontostoreandaccessECGdata. WiththeincreasingpopularityofInternetaccessthrough
mobile phones, these systems provide an ideal platform between remote monitoring systems and
patients. TodisplaytheECG,manystudieshaveemployedawebcomponentthroughaPC[17,18],
single-chipmicrocomputer[19]orsmartphone[2,18,20].
Oneproposeddesignforaremotemonitoring“tele-medicine”systemandwebserverconsists
oftheclientside,ageneralpacketradioservice(GPRS)modemandtheserverside[18]. Theclient
side could be interpreted as the combination of the ECG, collection equipment, user interface
andmicroprocessor. TheGPRSmodem,whichisusedtotransmittheECGsignals,providesalarge
geographical mobility coverage range. The server side is divided into the back and front ends.
TheserverfrontendisusedtodisplaytheECGsignalonthewebforpatientsanddoctors,andtheback
end is designed to receive data from the GPRS modem and store it in a database. However, such
designsareinadequateformedicalexperts,becausetheydonotprovidetheinformationrequiredto
analyzetheECGsignals.
Intherecentpast,manystudieshaveappliedQRSdetectionalgorithmstohealthcareservers.
SuchsystemscanrecordECGsignalsonawebserverandfacilitatetheiranalysis. Oneprototypeuses
amobilephoneasagatewayfortransmittingmeasuredECGdatabacktothemedicalcloudusing
Sensors2017,17,1360 4of21
3GmobiletelecommunicationsorWiFi[6]. ThesystemcanalsocalculatethebeatsoftheheartasRR
intervals,whichisthetimebetweenconsecutiveR-waves,onthemobilephone. Anothersystemoffers
notonlyQRSdetection,butalsoapriority-basedalarmmessagingservice[4].
Althoughtheseapproachesfacilitatetheconvenientmeasurementandanalysisofbodystatus,
theystillencounterproblemswithsecurityandprivacy. Asanumberofhealthcareapplicationsdeal
withphysicalinformation,whichconstitutessensitivepersonalinformation,dataleakageandinvasions
ofprivacyinhealthmonitoringsystemsaresignificantissues[11–13]. In2012,Maetal. proposed
asimple,buteffectivesecuritysolutionforECGsignalsbasedonanECGcompressionalgorithm[21].
Regardingsecurityandprivacychallengesinmobilehealthcare,Luproposedanefficientuser-centric
privacyaccessschemebasedonattribute-basedaccesscontrolandanovelprivacy-preservingscalar
productcomputation[22]. Thisschemeemploysabodysensornodethatcanmonitorvarioustypes
ofhealthinformation,andithassufficientresourcestoapplyawidely-employedsymmetricencryption
schemesuchasAES.Ontheotherhand,inthiswork,wefocusonabodysensingdevicethatonly
monitorsanECG,andthus,asmallersensingdevicecanbeemployed,whichismorecomfortable
forpatients. Becausepatientswearthesensingdevicefortheentireday, itisanadvantagetouse
a smaller device. Regarding this aspect, the scheme described above based on a symmetric key
andhashfunctionsrequiresmorecircuitsandlargerdevicestohandletherequiredcryptographic
functions. Therefore, itremainsachallengetodesignasecureandprivacy-preservingschemefor
resource-constrainedremotehealthcaremonitoringsystemsthatutilizebodysensorsofareducedsize.
More recently, there have been attempts to deal with the use of ECG for identity recognition
and biometric authentication [23,24]. For instance, Peter et al. [23] described the design and
implementationstepsrequiredtorealizeanECG-basedauthenticationsysteminbodyareasensor
networks andutilized ECG features for this purpose. Tan et al. [24] focused rather on enhancing
theeffectivenessandrobustnessofabiometricrecognitionsystemusingacombinationofrandom
forest and wavelet distance measure classifiers. The ECG application investigated in this study
is slightly different and has different design objectives. It is more oriented towards the secure
transmissionofECGdatatoremoteserversforanalysisandtotheapplicationofmachinelearningto
arrhythmiarecognition. Enhancingtheextenttowhicharrhythmiacanbeaccuratelydetectedisone
oftheeventualobjectivesofourstudy,andtheenergyisnotourprimaryconcern,becausedecisions
aremadeonaremoteserver. Regardingnetworks,weareconsiderablyinterestedinsecuredelivery
overwide-areanetworks,ratherthandatatransmissionoverbody-areanetworks.
3. SystemModels,ThreatModelsandSecurityObjectives
Thissectiondescribesthesystemmodels,threatmodelsandsecurityobjectivesofoursystem.
3.1. SystemModel
WeareparticularlyinterestedinrecordingandaccumulatingECGdataforeachpatientoveralong
periodoftime,performingsupervisedlearningbasedonthekeyECGfeaturesandintelligentheartbeat
classificationforearlyandautomaticdetectionofarrhythmia(weusedaclassifierbasedonadecision
tree,whichisconstructedusingtheC4.5learningalgorithm). Thecrucialdesignconsiderationisthe
level of accuracy with which arrhythmia can be recognized. To achieve the desired accuracy, it is
imperativethatalloftheacquiredECGdatashouldbestoredinthepermanentstorageoftheDSS(i.e.,
database)andnotonmobiledeviceswithlimitedstoragespace.
Oursystemcomprisesfourcomponents: arevocationauthority,anECGsensingentity,adecision
supportserver(DSS)andamonitoringstation, asdepictedinFigure2. Thepatientisabletolead
anormallifewhilethesensorcontinuouslyacquiresECGdataandsendsthemtoaBluetooth-enabled
smartphone. The smartphone relays the data to the DSS, which associates incoming data with
the records for the patient in a database and analyzes them. The heartbeat is detected using
thePan–Tompkinsalgorithm,whichhasbeenshowntobeaneffectiveQRSdetectionscheme[25],and
thenclassifiedusingadecisiontree. Ifthesystemidentifiescongestiveheartfailureoranirregular
Sensors2017,17,1360 5of21
heartbeat,thenanalarmissentovertheInternetconnectiontoamonitoringstation,wheremedical
personnelcancarryoutappropriateactions.
Sensor
Patient Individual(1) Individual(2) Individual(n)
(A) ECG sensing entity (2) ECG transmission (C) Decision support server
(A.1) User Interface (C.1) User interface
(A.2) Signal measurement (C.2) Signal decryption
(A.3) Signal encryption (C.3) Heartbeat analysis
(A.4) Data delivery (C.4) Database
(4) Care service
(3) Alarm
(1) Registration and
obtaining AID
Expert
(B) Revocation authority (5) User revocation (D) Monitoring station
(B.1) Issues AID (D.1) Check status
(B.2) Revokes user
Figure2.Overviewoftheproposedsystemarchitecture.AID,anonymousID.
3.1.1. RevocationAuthority
We assume the existence of a revocation authority (RA) to hide a user’s identity, while being
able to make it available during an emergency. The RA has three roles in the proposed scheme.
First,itmanageseachuser’sidentityusingarelatedanonymousID(AID).Second,itrevealsauser’s
identityandcontactinformationinanemergency,followingarequestbythemonitoringstation. Third,
itissuescertificationforeachentity’spublickey. AstheRAplaysanimportantroleinuseridentity
managementandpublickeyconfirmation,wetreattheRAasatrustedthirdparty. Agovernment
agencycouldsetupandmanagetheRAforpublicwelfarepurposes.
3.1.2. ECGSensingEntity
The smartphone of a patient collects ECG signals every day. Figure 3 illustrates the process
ofsignalmeasurementanddelivery. Asensornodeisattachedtothepatient’sbody. Thesignalfrom
thesensornodeisconvertedtoadigitalvalueusingananalog-to-digital(ADC)converterandthen
senttoamicrocontrollerunit(MCU)throughaserialperipheralinterface,whichisusedprimarilyto
communicatebetweenchips. ThesignalattheMCUissenttoaBluetoothmoduleusingauniversal
asynchronousreceiverandtransmitter. SubjectscanviewtheirECGthroughagraphicalinterface
designedfortheirmobiledevice. Althoughitispossibletoperformself-monitoring,theECGsignal
shouldalsobetransmittedtoaDSS,becauseatypicallaymancannotindependentlyinterpretthesignal.
Accordingly,thesignalonthesmartdeviceistransmittedoverawirelessnetwork.
Micro-
Bluetooth
Sensor node controller
Antenna
SPI unit UART
Figure3.ECGsignalmeasurementanddelivery.
Sensors2017,17,1360 6of21
3.1.3. DecisionSupportServer
The DSS analyzes multiple aspects of the ECG. As a convenience for medical experts, it acts
asamonitoringstationfordecisionsupportbydetectingandclassifyingtheheartbeat. Heartbeat
detectioninanECGprimarilydependsontheQRScomplex. However, QRSdetectionbyitselfis
notsufficientforheartbeatdetection,whichmustprecedetherecognitionoffeaturesfordetecting
arrhythmia. OneofthesignificantcontributionsofthisstudyisthemethodofdetectingtheP-wave
using the QRS complex, which is in turn detected by the Pan–Tompkins algorithm [7] (consisting
ofaband-passfilter,adifferentiatorandanintegratoroveramovingwindow). Eventually,theuse
oftheP-wavewiththeQRScomplexleadstotheaccuratedetectionofheartbeatsrelatedtoarrhythmia.
3.1.4. MonitoringStation
Themonitoringstation(MS)providesagraphicaluserinterface. Therefore,boththepatientand
theexpertoperatorscanremotelycheckthesignal. Thepatient,expertsandafewotherswhohave
accessrightscanviewtheECG,heartrateandpatientprofilethroughtheweboramobiledevice
application.Moreover,expertsusingtheMScanalsoviewanddeterminethesignalconditionsofmany
patients through the web application. If an expert identifies a dangerous condition, thentheMS
immediatelycontactsmedicalservices.
3.2. ThreatModel
Inthescopeofthiswork,weconsidernormalusers,includingservers,aspotentialadversaries.
Adversaries have limited capabilities, in that they can only access publicly-available information,
including information from wireless communications. This is no more information than can be
accessedbynormalusersoftheECGmonitoringsystem. Thus,weconsiderthatadversariesdonot
havetheabilitytodistinguishtheoriginatorofdataeavesdroppedfromawirelesscommunication.
Inthecaseofservers, weassumeanhonest-but-curiousmodelthataccuratelyfollowsaprovided
protocol, but may attempt to obtain information from the communication session or stored data.
Inaddition,wedonotconsiderthattheadversariescaneasilycompromiseotherentitiesinorderto
obtaintheidentityofusersorothervaluableinformation. Basedontheseconditions,thefollowing
attackscanbecarriedoutbyadversaries.
• Eavesdropping: Essentially, we assume that an attacker has the ability of eavesdropping on
data during transmissions. Wireless communications, including Bluetooth, which is used to
transmitthesensingdata,arevulnerable,becausethetransmitteddatacanbeleakedbythemajor
functionhookingschemeofthewindowskerneldriver[26,27]. Thus, anattackerclosetothe
sensingsystemcaneavesdroponthesensingdataandtheidentityofauserviatheweakpointof
oursystem.
• Leaking: WetreattheDSSandserviceprovideraspotentialattackers. TheECGanalysisresults
andidentityofanindividualusercouldbeleakedbytheserviceprovidersorsystemmanagers
oftheDSSandMS,respectively. Weonlyfocusoncaseswherehealthinformationisleakedalong
withtheidentityoftherelevantuser.
• Tracking: After obtaining data (raw ECG data or an ECG analysis) by eavesdropping or
leaking, an attacker could attempt to determine the relation between the data and the user.
Inaddition, theattackercouldattempttotraceaparticularuserbydeterminingtherelations
betweenpseudonyms.
Thecollectedsensingdatafordetectingarrhythmiasispersonalinformationfortheuser,andit
couldbeusedbyanattackerinvariousways. Onepossiblescenarioissellingthecollecteddatafor
profit. Pharmaceuticalcompaniesorclinicscouldusesuchdatafortargetedmarketing. Inaddition,
thepatientmaynotwanthis/herillnesstobeknowntoothers. Suchtypesofattacksareonlypossible
iftheattackercouldobtainboththehealthinformationandtheidentityoftheuser.
Sensors2017,17,1360 7of21
3.3. SecurityObjectives
Dependingonthesystemandthreatmodelsdescribedearlier,wedefinetwosecurityobjectives.
• Communication security: An attacker close to a user cannot obtain sensing signals from
the wireless communication between the MCU and the smart device, nor between the smart
deviceandtheDSS.
• Privacy preservation: An attacker cannot establish a link between a particular user and their
sensing signal or analysis results from the DSS. In the case of an emergency, the personal
informationofausercanbedisclosed. Duringthisstep,anattackercanobtainonlyalimited
amountofdatafromtheserver.Inotherwords,fromacollectionofpersonalinformationprovided
andapseudonym,theattackercannotdistinguishtheECGinformationofaparticularuser.
Asimplemethodofpreservinguserprivacyistoemployapseudonymasananonymousidentity
for communication, because this can easily hide the relation between data and a user’s identity.
However,anattackercaneasilyobtaintheidentityofacertainuserthroughlong-termobservation.
Many previous schemes have mentioned the traceability of a single pseudonym[28–31]. In our
system model, the attacker can obtain contact information for the pseudonym in an emergency.
Allofthepatient’s information from before and after the emergency will be leaked if a single
pseudonym-basedapproachisapplied,andthus,thesecuritymodelhastodealwithforwardand
backwardprivacy. Todealwithanemergency,auser’sidentityshouldberevealedtoreceivemedical
service. Otherwise, an inside attacker can determine the relation between the user’s identity and
thesensingsignal. Theforwardprivacyindicatesthattheattackercannotdeterminearelationbetween
sensingsignalsafteranemergency. Thebackwardprivacyindicatesthattheattackercannotdetermine
arelationbetweensensingsignalsbeforetheemergency.
4. SystemArchitectureforPrivacy-PreservingIntelligentECGMonitoring
In this section, we propose an intelligent ECG monitoring system incorporating privacy
preservation. The proposed scheme comprises two stages. First, we design a secure sensing
signalencryptionschemeusingaconventionalpublickeycryptosystem,toprotectthesensingdata
transferredfromthesensortotheMSviaasmartdevice.Second,wepresentanAIDschemetopreserve
theuserprivacyduringthearrhythmiarecognitionprocess. Thisalsohidestheuserinformationfrom
expertsattheMS,butthiscanberevealedinanemergency. Beforedescribingtheproposedscheme,
wepresentthenotationsusedinthispaperinTable1.
Table1.Notations.
Notation Description
k Asymmetricsessionkey
(pk,sk) Public,privatekeypair
p(cid:96) Apseudonym
PS Asetofpseudonyms,PS={p1,p2,...,p(cid:96)}
a(cid:96) AnanonymousID
AID AsetofanonymousIDs,AID ={aj}1≤j≤(cid:96)
H(·) Acryptographichashfunction
PRG(·) Apseudo-randomnumbergenerator
E {} Anencryptionfunctionusingkeyk
k
S {} Asignaturefunctionusingkeyk
k
KS Akeystream
ES AnencryptedECGsignalusingkeyk
4.1. Setup
Each user generates and uses a key and a pseudonym for each session to ensure privacy.
Onesessionisthedurationofapseudonymandthecorrespondingkey. Bychangingthesefrequently,
Sensors2017,17,1360 8of21
thesystemcanprovidestrongerprivacyprotection. Thelengthofeachsessionisflexible. Ifauser
wantsahigherlevelofprivacy,he/shecanadjusttomakethedurationshorter. Theuseralsocan
adjusttomakethedurationlongerinthecasethattheyrequireefficientoperation.
WhenauserfirstestablishesaconnectionbetweentheMCUofabodysensornetworkandasmart
device,theMCUandthesmartdeviceshareasymmetricsessionkey{k }todealwiththelimited
s
storageoftheMCU.
The RA, DSS, MS and the smart device of a user use a public key cryptosystem to
establish secure communications. The public/private key pairs for the entities are (pk ,sk ),
RA RA
(pk ,sk ),(pk ,sk )and(pk ,sk ),respectively.
DSS DSS MS MS U U
4.2. RevocationAuthority
Eachuserhasaunique ID,andtheuserregistersthisalongwithhis/hercontactinformationto
dealwithemergencies. TogenerateanAID,theuserencryptstheir IDusingapublickeyoftheRA
andsendsthistotheRA.Atthispoint,weemployasetofsessionpseudonymssuchthateachuser
usesonepseudonymforonlyonesession,todealwithforwardandbackwardprivacyinthesingle
pseudonymapproach[31].
The RA generates a set of unique pseudonyms PS = {p1,p2,...,p(cid:96)}, and computes the AID
AID = {aj}1≤j≤(cid:96) withitssignatureas:
a = E {H(ID||p )},S {a },1≤ j ≤ (cid:96)
j pkU j skRA j
where(cid:96)isanaturalnumbergreaterthanone,whichisdefinedbytheuser. Toaddressthetrade-off
betweenprivacyandefficiency,wheretheprivacyisenhancedbychangingthepseudonymsmore
frequently,weallowtheusertodefinethefrequencyand (cid:96) whileusingtheapplication. However,
eachsessionshouldbelongerthan30minutestoensureanaccurateanalysisresult.
TheRAstores(ID,A )pairs,encryptstheseusingtheuser’spublickeyandthensendsthemto
ID
theuser’ssmartdevice. Atthispoint,oursystemonlyusesthepublickeytoencryptthe(ID,A )
ID
pairs. Therefore,anattackercannotobtaintheA usedforanactualcommunicationbetweentheuser
ID
and the DSS. Thus, using the user’s public key does not affect the privacy of the system. Finally,
theusersendssensingsignalstotheDSS,preservingprivacybyusingoneoftheAIDs. Astheuser
cannotreceiveserviceswhenanincorrectAIDisused,alegalAIDthatisgeneratedbytheRAshould
beentered. Afterreceivingthe AID,theusergeneratesasetofkeys{ki}a≤i≤(cid:96).
Incaseofanemergency,theMSsendstheAIDtotheRA.TheRAthendeterminesthepatient
informationfromtheAIDlistandperformsthestepofconnectingthepatientwithadoctor.Ourscheme
employstheRAasanadditionalentitytodecentralizesecretinformationandincreasetheeffectiveness
ofprivacypreservation.
4.3. ECGSensing
Totransferthesensingsignalsecurely,weuseastreamcipherthatconsistsofapseudo-random
generator (PRG) and a bitwise exclusive OR operator [32]. A sensor installed on the human body
typically consists of a resource-constrained and battery-powered device. Moreover, considerable
computations result in the production of more heat, which can be problematic in terms of patient
safety. Furthermore, a body sensor requires a powerful processor and a large capacity battery to
applycryptographytoitsvariousfunctions,whichwouldincreasethesizeofthebodysensingsystem
andinterferewithitsday-to-dayusage. Accordingly,weproposeasimpleandsecuremethodthat
minimizestheencryptionoverheadandprotectssensingdataduringtransfer.
The MCU of the user generates the keystream KS using the PRG such that KS = PRG(k ).
s s s
Then,itcomputestheexclusiveORoperationfortheECGsignalSS asES = KS ⊕SS,whereES
i s s i s
denotestheencryptedECGsignal. Afterreceivingthis,thesmartdeviceoftheuserperformsthesame
Sensors2017,17,1360 9of21
operationastheMCUtodecrypttheECGsignal,andtheusercanmonitortheirownECGonselected
smartdevices.
Tosendthesensingsignalsafely,theusergeneratesakeystreamKS usingthePRGsuchthat
i
KS = PRG(k )andencryptsSS usingKS as ES = KS ⊕SS. Subsequently,theuserselectsan a
i i i i i i i i
randomly and sends the received signal stream to the DSS to detect arrhythmias using the ECG.
The first time the user employs a choice of a, they send a,S {a } and E {k } for validity.
i i skRA i pkDSS i
Subsequently,theusersendsthesignalstreamtotheDSSasa ||ES.
i i
Forprivacyreasons,weuseasetofpseudonymsandasetofkeys. Ifauseronlyusesasingle
pseudonym or key, then the DSS can easily trace that user. When a user’s identity is revealed in
anemergency,theDSScanobtainallofthestoredECGsignalsandanalysisresults,aswellasfuture
informationfortheuser. Therefore,usersshouldperiodicallychangetheirpseudonymsandkeysto
minimizeleakeddataafteranemergency. Afterareceivedpseudonymisexhausted,theuserrequests
anovelsetofpseudonymsfromtheRA.
TheDSSdetectsarrhythmiasusingcontinuousECGstreamingfromusers,andthepseudonyms
transformthatstreamintoadiscontinuoussignal. Thus,weshouldconsidertheeffectofpseudonyms
onthedetectionprocess. Consequently,wedescribethesimulationresultsofourschemeinSection6,
basedon30-minuteofECGsignals,andshowthatwecanensure96.63%accuracyifthepseudonym
changingperiodislongerthanhalfanhour.
4.4. AnalysisofanECGforArrhythmiaDetection
The DSS should first decrypt the signal stream to analyze the ECG. To achieve this, the DSS
decryptsthekeysignalusingitsprivatekeyandobtains k . Byobtaining k ,theDSScangenerate
i i
thesameKS astheMCUanddecrypttheECGsignalasSS = KS ⊕ES. UsingSS,theDSSanalyzes
i i i i i
themultipleaspectsoftheECG.
4.4.1. HeartbeatDetectionandFeatureExtraction
First,heartbeatsaredetectedusingthedeterminedQRScomplexesandP-waves. Figure4depicts
theprocessofthePan–Tompkinsalgorithm,andFigure5[25]illustratestheresultsoffeatureextraction,
withthestep-by-stepoutputofthealgorithmperformedonRecord200intheMIT-BIHarrhythmia
database.Figure5ashowstheoriginalECGsignal.Theoriginalsignalisnormalizedbythemeanvalue,
asshowninFigure5b. Theband-passfilteriscreatedbycombiningalow-passfilterwithahigh-pass
filter. This reduces noise such as muscle noise, 60-Hz interference, baseline wander and T-wave
interferenceintheECGsignal. Thedifferentialequationforthelow-passfilteris:
y(nT ) = 2y(nT −T )−y(nT −2T )+x(nT )
s s s s s s
−x(nT −6T )+x(nT −12T ), (1)
s s s s
whereT denotesthesamplingperiod,xistheamplitudeofthen-thECGsampleandyistheamplitude
s
afterfiltering. Thedifferenceequationforthehigh-passfilteris:
y(nT ) = 32x(nT −16T )
s s s
−[y(nT −T )+x(nT )−x(nT −32T )]. (2)
s s s s s
Moving-
Band-pass
Derivative Squaring Window
filter
integration
Figure4.ProcessofthePan–Tompkinsalgorithm.
Sensors2017,17,1360 10of21
Figure5.Resultsoffeatureextractionwithstep-by-stepoutputofthePan–Tompkinsalgorithm.
TheDSSsetsthelow-passfilterwithacutofffrequencyof11Hzandthehigh-passfilterwith
acutofffrequencyof5Hz,asshowninFigure5c,d.Afterbeingfiltered,theECGsignalisdifferentiated
toprovideslopeinformationusingthefollowingdifferentialequation:
1
y(nT ) = [−x(nT −2T )−2x(nT −T )
s s s s s
8T
s
+2x(nT +T )+x(nT +2T )], (3)
s s s s
Equation(3)approximatestheidealderivativeoffrequenciesupto30Hz,andFigure5epresents
the results of the derivative. This is then squared point by point, making all of the data points
in the processed signal positive and emphasizing the higher frequencies, as shown in Figure 5f.
Thedifferentialequationforthissquaringis:
y(nT ) = [x(nT )]2. (4)
s s
Integrating the moving window provides waveform feature information, which is added to
theslopeoftheR-wave. Thisisachievedusingthefollowingdifferentialequation:
1
y(nT ) = [x(nT −(N −1)T )
s s s s
N
s
+x(nT −(N −2)T )+···+x(nT )], (5)
s s s s