Review of photoacoustic flow imaging: its current state and its promises

Review

article

Review

of

photoacoustic

flow

imaging:

its

current

state

and

its

promises

P.J.

van

den

Berg

*

,

K.

Daoudi,

W.

Steenbergen

BiomedicalPhotonicImagingGroup,MIRAInstituteforBiomedicalTechnologyandTechnicalMedicine,UniversityofTwente,POBox217,7500AEEnschede,

TheNetherlands

Contents

1. Introduction... 89

1.1. Currentmodalities... 90

1.2. ThecaseforPAflowimaging ... 90

2. Photoacousticflowimagingmethods... 91

2.1. Dopplershift... 91

2.1.1. Continuous-wavephotoacousticDoppler... 91

2.1.2. Structured-illuminationphotoacousticDoppler... 91

2.2. Densitytracking... 93

2.2.1. Cross-correlationflowimaginginthetimedomain ... 93

2.2.2. Cross-correlationflowimaginginthespatialdomain... 93

2.3. Transittime... 94

2.3.1. Transit-timeflowimagingofsingle-particles ... 94

2.3.2. Transit-timeflowimagingofparticleensembles... 94

2.4. Amplitudeencoding... 96

2.4.1. Photoconversionandreplenishment ... 96

2.4.2. Heatencodingandconvection... 96

2.5. Discussionandconclusion... 98

Acknowledgements... 98

References... 98

1. Introduction

Bloodflowinarteries,veinsandsmallercapillariesisanimportant aspectinthediagnosisofawiderangeofpathologiesanddiseases.

ARTICLE INFO

Articlehistory:

Received18May2015

Receivedinrevisedform24July2015

Accepted2August2015

Availableonline13August2015

Keywords: Photoacoustic/optoacousticimaging Flowimaging Flowgraphy/flowmetry Doppler Perfusion Functionalimaging ABSTRACT

Flowimagingisanimportantmethodforquantificationinmanymedicalimagingmodalities,with applications ranging from estimating wall shear rate to detecting angiogenesis. Modalities like ultrasoundandopticalcoherencetomographybothofferflowimagingcapabilities,butsufferfromlow contrasttoredbloodcellsandaresensitivetoclutterartefacts.Photoacousticimaging(PAI)isarelatively newfield,witharecentinterestinflowimaging.TherecententhusiasmforPAflowimagingisduetoits intrinsiccontrasttohaemoglobin,whichoffersanewspinonexistingmethodsofflowimaging,and someuniqueapproachesinaddition.Thisreviewarticlewilldelveintotheresearchonphotoacoustic flowimaging,explaintheprinciplesbehindthemanytechniques andcommentontheirindividual advantagesanddisadvantages.

ß2015TheAuthors.PublishedbyElsevierGmbH.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

* Correspondingauthor.

E-mailaddress:p.j.vandenberg@utwente.nl(P.J.vandenBerg).

ContentslistsavailableatScienceDirect

Photoacoustics

j ou rna l h ome p a ge : w ww . e l se v i e r. co m/ l oc a te / p a cs

http://dx.doi.org/10.1016/j.pacs.2015.08.001

2213-5979/ß2015TheAuthors.PublishedbyElsevierGmbH.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/

Bloodflowcanbeestimatedwitheitherflowimagingorperfusion imaging,andbothprovidedistinctandvaluableinformation.Flow imagingistheprocessofmappingwherefunctionalvascularityis,or whattheflowprofileiswithinanarteryorvein.Thelattercanbeused forestimating,forexample,thewallshearrate[1]ordetectingwhere turbulentbloodflowoccurs[2,3].Imagingvascularityisimportantto revealforexampleangiogenesis[4],theprocessduringwhichrapid growth of new vasculature occurs. The resulting vasculature, irregularlyand haphazardly shaped,is the pathological result of forinstancetumourgrowth[5]andinflammatorydiseasessuchas rheumatoidarthritis[6].

Perfusionimagingisrelatedtoflowimaging,butprovidesamore globalpictureofvascularity,thatdescribeshowmuchbloodreaches organs,musclesorskinovertime[7].Theamountofskinperfusion determinesforinstance thechanceof a burnhealing[8], andin cerebralischemia,malperfusionofpartsofthebrainleadspotentially tostroke[9].Theperfusioncanbevisualisedastheintegrationofflow speedoverthetotalcrosssectionalareaofthefeedingvasculature

[10].Perfusionandtheamountofflowimagedarethereforeclosely related in some clinical applications, however, computing the perfusionfromimagedvascularityisoftenchallenging[11].

Photoacousticimaginghasthepotentialtodoboth:perfusion imagingandflowimaging.Inthisreviewwewillonlyfocusonthe latter.Awidevarietyofmethodsinvestigatedbydifferentresearch groups,inspiredbyotherimagingmodalities,inadditiontoafew approachesthatutiliseuniqueaspectsofphotoacousticimaging, makethereviewaworthwhileinvestigation.Wewillfirstbriefly discuss existing imaging modalities that are capable of both perfusionimagingandflowimagingbeforedescribingthedifferent photoacousticapproachesdevelopedsofar.

1.1. Currentmodalities

Magneticresonanceimaging(MRI)isawidelyusedmodalityfor imagingperfusionandflow,usingarangeofmethods[12,13].For instance,imagingflowcanbeperformedusingphasecontrastMRI

[14], while perfusion imaging can be done with arterial spin labelling[15]ordynamiccontrastenhancedimaging[16].MRIcan provide blood flow information of the whole body, with high sensitivity and resolution [17], which makes it unique in that aspect.Moreover,itcanbecombinedwithblood-oxygenation-level dependent(BOLD)MRI[18].However,MRIis veryexpensivein both the initialinvestment and the upkeep, and the extensive pulsingschemesinMRIalsomakeimagingaslowprocess[19].

Dynamic contrastenhanced computedtomography (CT)is a moreaffordablemodalityforperfusionimaging,unlikeMRI.Itcan alsobecombinedwithpositronemissiontomography (PET)for quantificationofe.g.glucoseconsumption forestimationof the completemetabolicactivity[20].LikeMRI,CTisalsocapableof imagingcompleteorgansoreventhewholebody,butwithworse resolutionanditreliesonionizingradiation,makingitunpractical formonitoring[19].

Ultrasound(US)imagingisanotherwidelyusedtechniquefor both flow and perfusion imaging. Over several decades, the

ultrasound community has developed many techniques. For

instance,flowimagingcanbeperformedusingcontinuouswave excitationinspectralDopplerUS[21],byimagingphasechangeof reflectedUSpulsesincolourflowimaging[22]orbytransverse speckle tracking [23]. Perfusion imaging is performed using dynamic contrast enhanced US, for example by using a flash-replenishmenttechnique[24].USimagingisveryscalablethrough theinverserelationshipofpenetrationdepthandresolution,which can be tuned using the ultrasound emission frequency and bandwidth[25].Furthermore,USimagingisbothaffordableand portable[26].Whileimagingdepthislargerthanopticalcoherence tomography,theresolutionispooreranditsuffersfromclutter[27].

Severalmodalities arebeingdevelopedtoestimateflowand perfusion using visible or infrared light, which makes these techniques both harmless and affordable. For example, laser Dopplerperfusionimaging(LDPI)andlaserspeckleimaging(LSI) aretwomodalitiesbasedontrackingdiffuselyreflectedlightover timeandcanbeusedforperfusionimagingoftheskin[28],with the advantages of portability, low cost and real-time imaging

[29,30].Ontheotherhand,LDPIandLSIarelimitedtosuperficial imaging and low resolution [28] and do not feature depth resolution.While LDPIhastheobjectiveunderlyingprinciple of velocity related Doppler shifts, quantifying perfusion in an absolutemannerremainschallenginginbothLDPIandLSIbecause of the unknown optical properties of the tissue [31]. Another opticaltechnique,orthogonalpolarizationspectralimaging(OPS), can be used for high-resolution imaging of micro-vasculature

[32].Resolution is highand penetrationdepth is fair, but flow quantificationisverychallenging.

Opticalcoherence tomography(OCT)is anoptical technique which can be used for flow imaging as a function of depth

[33].Flowcanbequantified,forinstance,bytrackingthephase changeofthereflectedlightovertime,orbycomputingthespeckle variance[34].OCThashighresolution,isdepthresolvedandcanbe portable, but suffers from limited penetration depth and is hinderedbyclutter[35].

In this regardPAI is comparable tofluorescencemicroscopy approaches like confocal microscopy [36] or the two-photon variant [37]. In both microscopy techniques flow imaging is performedbylaser scanningalongtheflowdirection.However, confocal and two-photon require fluorescent markers that are susceptible to photobleaching; are only useable in superficial applications; and thelaser scanningmakes it only suitable for imaging a few blood vessels at a time [38]. As we will see, photoacousticscanbeusedtoovercometheseproblemsinflow imaging.Photoacousticflowimaging,usesendogenouscontrast, andallowsapproachesnotlimitedtospecifictargetingofblood vessels.

1.2. ThecaseforPAflowimaging

Photoacousticimaging(PAI)isanopticalmodalitythatrelieson light pulses to generateultrasound at locations ofhigh optical absorption[39]. Nanosecondlight pulses aredirected onto the skin, where they diffuse through tissue, down to several centimetresfornear-infrared light.Thelightis locallyabsorbed by tissue chromophores, and is converted into heat, causing a pressurebuild-up.Thisbuild-upisreleasedintheformofpressure waves:soundwavesverysimilartothoseemittedinpulse-echo ultrasound.

Themaintissuechromophoresinthevisibleandnearinfrared wavelengths(NIR)arehaemoglobinandmelanin(<1000nm);at NIRwavelengths(>930nm)lipidsalsoexhibitabsorptionpeaks thatcanbeutilized[40].PAIcanthereforebeused spectroscopi-callyand,usingspectralunmixing,therelativeconcentrationsof chromophorescanbeextracted[41].Inthisway,theoxygenation ofbloodvesselscanbedeterminedbyexploitingthe oxygenation-dependentabsorptionspectrumofhaemoglobin[42,43].

Likeultrasoundimaging,PAIisrelativelylowcost,canbeused in portable devices [44], and the resolution can be increased relativelyeasilythoughatthecostofimagingdepth.PAIisusedin microscopy,tomographyandlineararraysystems,withvarying resolutionsandimagingdepths[45].Animportantdistinctionin systems can be made, namely between acoustic and optical resolutionsetups.Inopticalresolutionmode,laserlightisfocussed onto a sample; in acoustic resolution mode, generated sound wavesarefocussedindetectionwithalargenumericalaperture– either physically orin computedreconstruction. Thefocal size

determinesthelateralresolution.Theopticalfocusthereforegives ahigherresolution,butatthecostofpenetrationdepth:optical scatteringwillbroadentheopticalfocusconsiderably.Itshouldbe notedthatinbothcasestheaxialresolution–alongtheultrasound transduceraxis–remainsgovernedbytheacousticbandwidthof thetransducer.

Thepromiseofusingphotoacousticsforimagingflowisdueto its reliance on absorption for contrast, as opposed to

back-scattered waves in US and OCT. Moreover, because of the

relativelylowconcentrationofhaemoglobinandother chromo-phores in tissue, there is a high contrast in PAI between vasculature andthe surroundingtissue [46]. Inaddition, PAI– unlikeultrasound–isrelativelyspecklefree[47,48].Thelackof thesetwopropertiesisthemainreasonitischallengingtodetect flowinsmallbloodvesselsandnearvesselwallswithOCTandUS flowimaging[49–51].

In the following review, we present a summary of the

research performed on photoacoustic flow imaging. We

describethe similarities betweensome flow imaging methods using PA with other modalities, and conclude with the key advantagesanduniquefeaturesofPAflowimaging.Thereview is partly written as a tutorial on the various flow imaging techniques, but it will also go into detail on performance characteristics like the minimum and maximum measureable velocities.

2. Photoacousticflowimagingmethods 2.1. Dopplershift

2.1.1. Continuous-wavephotoacousticDoppler

Fangetal.proposedatechniquewheretheDopplershiftofaPA modulation was estimated [52], similar to what is done with spectral Doppler US. The authors used a diode laser that was intensitymodulatedwithafrequency f0:

IðtÞ¼I0

2½1þcosð2

p

f0tÞ

withtthetimewithinoneacquisition,I0andIðtÞthemaximum andtime-variant lightintensityrespectively.A modulatedlaser willgenerateaPAresponsethatisalsooscillatingwithfrequency f0(seeFig.1).Ifexcitedredbloodcells(RBCs)areflowingat

v

flow, thentheparticleswillemitaPAresponsethatisDopplershifted approximatedby:

D

f¼ f0

v

flow

v

s cos

u

with

u

theanglebetween theflowdirectionand thedetector’s viewingline,and

v

sthespeedofsound. Theyimplementedthe technique on an acoustic resolution setup with a 2.5MHz focussed mono-element transducer and a lock-in amplifier for detection; themodulationwasmatchedtothis frequency.Fang etal.measuredadistributionofDopplershifts–notasinglevalue– due tospectralbroadening,causedby theopening angleofthe ultrasonic transducer and the presence of a distribution of velocitiesinthetubing.

The authors demonstrated thetechnique on flowing carbon particles,whichservedasamodelforRBCs.Intheirfirstpaperthey measured velocitiesof flowingcarbon particlesthrough tubing, whichrangedfrom0.055to8.8mm/s.Withtheirtechnique,the minimummeasureablevelocitywasfundamentallylimitedbythe frequencyresolution(=1/acquisitiontime)ofthesystem,whereas themaximummeasureablevelocitywaslimitedbythesystem’s SNR:forgreaterflowvelocitiesthedistributionofDopplershifts broadened,andthereforetheamplitudedecreased.

Inasecondstudytheyusedamorerealisticphantomwhere theyincludedalsoscattering(

m

0s¼0:4=cm,pathlength3cm)and increasedthedensityofcarbon particlesfrom15% to40%(v/v)

[53].Themaximummeasureablevelocitydroppedto1mm/sdue to the scattering, which reduced the SNR; but measurements seemed unaffectedbytheincreasedparticle concentration. The authorsindicatedbloodmeasurementsweresuccessful,butonly inwater,withoutanyscattering.

Sheinfeldetal.followeduponthisworkandimprovedtheaxial resolutionusingshortburstsofmodulatedlight[54,55].Theyused rectangularburstswithintensity

IðtÞ¼I0 2½1þcosð2

p

f0tÞ XN n¼1 rect tnTrep Ton  

where Trep and Tonare therepetitioninterval and cyclelength respectively,ofburstnumbern.

Theyusedanacousticresolutionsetupwitha1MHztransducer todetecttheDopplershiftfromthePAresponsesofflowingcarbon particles(intheircaseina10%(v/v)suspension).Theyapplieda timegatetothetimetracesandestimatedtheDopplershiftina 3

m

s window for a range of positions. The size of time gate determined the axial resolution, which therefore was 4.5mm. Their1sacquisitionlengthwaslongenoughtodetectmultiple bursts,sothetimegatewasrepeatedeveryTrep¼70

m

stoincrease thespectralresolution.

Theauthorsmeasuredvelocitiesfrom3.5to203mm/s;intheir casethemaximumspeedwaslimitedbythesyringepump,the minimum–asbefore–bythespectralresolution.

ForbothSheinfeldetal.andFangetal.,thelateralresolution was determined by the ultrasound transducer’s beam width. Speculatingonthemaximumpenetrationdepth;itisexpectedto be fairly limited. In fact, Fang et al., found imaging through scattering(

m

0s¼0:4=cmwitha3cmpathlength)challengingin spiteofusinghighlyabsorbingcarbonparticlesandmuchlower opticalscatteringthanintissue(

m

0s8=cm).Foranoverviewof thetechnique’skeyattributes,seeTable1attheendofthereview article. In this table, the flow imaging techniques yet to be discussedwillbelistedalso.

2.1.2. Structured-illuminationphotoacousticDoppler

Zhanget al.implementedphotoacousticmodulationinthe spatialdomaininsteadoftimedomainbycreatingillumination fringes with a nanosecondpulsed laser [56]. The modulation frequency corresponds to f0¼

v

s=d,with a fringe pitch of d;

and the Doppler shift of the PA modulation as before:

D

f¼ f0

vflow

vs cos

u

Fig.1.Incontinuous-wavephotoacousticDoppler,sinusoidalmodulationofthe

excitationlightsourcecausesaphotoacousticresponse,whichisDopplershiftedif

Sincetheirdetectiongeometrywasparalleltothemodulation– ascanbeseeninFig.2–theDoppleranglewas

u

¼0_.

They estimated not only theDoppler shift, but also related quantities,namelythephaseshift,thechangeinarrivaltimeand timecompressionofthephotoacousticresponse.Intheir experi-mentstheyusedaninksolutionwithflowspeedsfrom20mm/sto

1400mm/s. When the authors added optical scattering

(

m

0s¼6:2=cm) they measured up to 600mm/s. Zhang et al. foundthetimecompressiontogivethebestaccuracycomparedto the Doppler shift (Root-mean-square error, RMSE of 57mm/s

instead of 89mm/s), especially with optical scattering

(RMSE=66mm/sinsteadof120mm/s).Notethattime compres-sionisinverselyrelatedtobandwidthbroadening:itseemslikely thattheirflowsetupshowsawidedistributionofflowspeeds,and thereforehighamountofbroadening.

The minimum and maximum measureable velocities were

determined,asbefore,bythefrequencyresolutionandtheSNR respectively.TheauthorsachievedamuchhigherSNRbyusinga pulsed laser–compared to CW modulation–which explains the highmaximummeasureablevelocity.

Optical scattering decreases the modulation depth and therefore decreases the SNR of the Doppler shift. This makes spatialmodulationmoresuitedforflowestimationwith superfi-cial applications like photoacoustic microscopy. The lateral resolutionwouldthenbedeterminedbythelengthofthefringe pattern;theresolutionindepthwouldbeverylimitedbecausethe ultrasoundtransducerisusedtodetermineflowcomponentalong thetransducertime-axis.

Yao et al. presented a method using a microscopy setup with spatial light modulation to estimate the transverse flow component–whichisperpendiculartothetransducer’ssymmetry

axis–instead of the axial component [57]. Modulation was performedusingalaserfringepattern:

IðxÞ/1þacos 2

p

x d

 

with a the modulation depth and x the transverse axis. They recorded pnTrep,themaximumofeachphotoacousticresponse sampled at each excitation pulse n, also called the ‘slow-time response’.It istermed such todistinguish it from the fast-time responsethatis theresult ofanindividuallaserpulse,ofwhich Zhang et al computed the Doppler shift. Yao et al.’s fast-time responsewasarepresentationoftheparticledensitydistributionthat flowed through the fringe pattern, and the amplitude of any photoacousticresponsewasproportionaltothenumberofparticles withinonefringe.Thisdensitydistribution‘fingerprint’is determin-istic and changes slowly over time. As the particle distribution flowedfromfringetofringe,anamplitudemodulationwascreated intheslow-timeresponse(seeFig.3),whichcouldbeobservedas side bandsat 2

pv

x=d. They assumed the particles to haveno correlationlengthandthereforeaflatspatialfrequencyspectrum.

Yao et al.first demonstrated themethodby flowing bovine bloodthroughtubingandforvarious(<10

m

m)particlesizeswith flowspeedsrangingfrom5to20mm/s.Theauthorsalsoshowed the radial flow profile of the flow in the tubing. They also demonstrated their method in the superficial vasculature of a mouseear,wheretheymeasuredaflowspeedofapproximately 1.9mm/s.

Thetheoreticalminimummeasureablespeedinthesystemwas 0.01mm/s, limited by the total acquisition time of 1 s. Their maximummeasureablespeedwas25mm/s,limitedintheircase bythecombination offringespacing andlaser pulserepetition frequency(PRF),sincethedensitydistributionrequiresadequate

Table1

Summaryoftheattributesofphotoacousticflowimagingtechniques.*indicatesalternativeresolution(acoustic/optical)mightbeusable,buthasnotbeendemonstratedyet.

Method Calibrationrequired? Meas.quantity Flowaxis Resolutionlimit

Lateral Axial

Dopplershift

Continuous-wave(Sec.2.1.1) Fouriertransform No Velocityspectrum Axial Acoustic* #Cycles

Structured-illumination(Sec.2.1.2) No Either #Fringes Acoustic

Densitytracking

Time-domain(Sec.2.2.1) Crosscorrelation No Velocity Axial Acousticoroptical Acoustic

Spatialdomain(Sec.2.2.2) No Lateral Distancebetweenfoci Acoustic

Transittime

Single-particles(Sec.2.3.1) FWHM Yes Speed Lateral Optical* Acoustic

Particleensembles(Sec.2.3.2) Autocorrelationorbandwidth Yes Velocity Lateral Optical* Acoustic

Amplitudeencoding

Contrastbased(Sec.2.4.1) Contrastinflow Yes Speed Any Acoustic* Acoustic

Heatencoding(Sec.2.4.2) Heatedspot No Velocity Any Heatingspot Acoustic

Fig.2.Usingspatiallasermodulationinstructured-illuminationphotoacoustic

DopplercausesamodulationofthePAresponsesimilarlytothetime-domain

version,andthemodulationislikewiseDopplershiftedunderflow.

Fig.3.Inthetransversestructured-illuminationmethod,thePAresponseasthe

cellsorparticlesmovethroughthefringesgetsmodulatedwithafrequencybased

samplingintime.Thelateralresolutionwaslimitedbythesizeof the acoustic focus, about 71

m

m and the axial resolution at 15

m

mbythebandwidthoftheultrasoundtransducer.

Note that both spatial-modulation techniquesonly measure onelateralflowcomponentatatime,andthatnodirectionalityis included;thelattermightbepossiblebyscanningthesystemback andforth.Also,asecondscancouldbemadetodeterminetheother lateralflowcomponent.

2.2. Densitytracking

2.2.1. Cross-correlationflowimaginginthetimedomain

Brunkerand Beardusedtwopulsedlaserstoexciteparticles withasmalltimedelayinbetween,suchthatthetime-of-flight could be extracted [58]. They use the fact that the density distributionofparticles orRBCs canalsobesampledalong the transducer axis (the ‘fast-time response’). If the transducer is alignedalong theparticle flow then thedistributionwill move

toward or away from the transducer–depending on the flow

direction(seeFig.4).Thisalsoholdswhenthetransducerisaligned at an angle

u

ð6¼90_{Þ to the flow: the particles within the} measurement volume will shift along the detection-time axis. Duringthetwolaserpulsesthedensitydistributionshiftedalong thedetectiontimeby

D

tinthefast-timeresponse.Thistimeshift couldbeestimatedusingacross-correlation,andtheaxialvelocity couldthenbeestimatedwith:

v

axial¼

v

s

D

t Tpcosð

u

Þ

withTpthetime between thetwo laserpulses. Themethodis similarto how ultrasonic colour flow imaging works (CFI, also termed‘ColourDoppler’),wherethebackscatteringamplitudeof ultrasound is linked to the particle density. In CFI the flow estimationisoftenperformedwithatwo-dimensional autocorre-lation,sincetheimagingisataconstantPRF(thereaderisreferred to[22]formoreinformation).

Brunker and Beard demonstrated the method on a rotating Perspexdiskwithanacetatesheetimprintedwitharandomdotted pattern.Therotationofthesedotswouldsimulatetheflowofred bloodcells.Theauthorsinitiallyusedtwoindividuallytriggered 10Hzpulsedlaserstointerrogatetherandomdottedpatternon therotatingdisk.Theymeasured velocitiesupto1m/s, witha systematicerrorof1-3%andarandomerrorof0.02-0.05m/s.

Brunkeretal.laterusedadouble-pulselaser,whichprovided 2pulsesper10Hz,andappliedthemethodsuccessfullytoflowing particlesandwholeblood[59].Here,theyusedtimegatingonthe sampleddensitydistributiontoestimatethemaximumvelocityin thetubing.Withwholebloodthemaximummeasureablevelocity was20mm/sandtherandomerror5mm/s.

The minimum measureable velocity was limited by the

precisionofthetimeshiftestimation.Whilethisestimationcan be performed sub-sample–interpolating or fitting the cross-correlationwithaGaussianfunction–itisfundamentallylimited by theSNR.In addition,thedegreeofcorrelationbetween two subsequentPAresponsesandthetransducerbandwidthalsoaffect theprecision.Thefundamentallowestpossiblerandomerrorin delayestimationisdescribedinultrasoundcolourflowimagingby theCramerRaolowerbound[60]:

s

Dt ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3 2f3 0

p

2T B 3 þ12B   1

r

2 1þ 1 SNR2  2 1 ! v u u u t

with T thewindow length of the time gate,

r

the correlation coefficient of the two photoacoustic responses, f0 the centre frequencyandBthefractionalbandwidth.

The maximum measureable velocity in their rotating disk experimentswaslimitedbythesizeofthetransducerfocusandTp, becausethedotshad tobein thedetectionareafor both laser pulses. On blood, their measurementsat higher flow velocities werelimitedbydecorrelationofthedensitydistributionbetween thetwolaserpulses.Intheoryalsoaliasingmightplayarole,but thiswasnotobservedwiththerotatingdisk,anditisexpectedat highervelocitiesthantheauthorscouldmeasurewithblood.The axialresolutionwasdeterminedbythesizeofthetimegate;the lateralresolutionbythesizeofthetransducerfocus.

Theirsetupwouldbelimitedinpenetrationdepthto<1mm duetotheuseof532nmexcitation(limitedlightpenetrationin tissue and blood), and the use of a 30MHz transducer (high ultrasound attenuation in tissue). Nevertheless, there is no fundamental reason which prevents the technique from being used in a tomography or linear array system. This would also enabletheestimation oftheangle

u

,which inits currentform wouldrequirescanningthemono-elementtransducer.

Yao et al. implemented a similar method using an optical resolution photoacoustic microscope(OR-PAM) [61]. Instead of computingthecross-correlationoftwoconsecutivephotoacoustic timetraces,theyestimatedthephaseshift

Df

todeterminethe axialflowvelocity:

v

axial¼ 1 2

p

v

s

Df

f0

Theyvalidatedthetechniqueonbloodflowintubing,where theycouldmeasureflowvelocitiesfrom0.1mm/sto8mm/s.Their theoreticalmaximummeasureablevelocitywas20mm/s,limited bythewidthoftheiropticalfocus.

Theycombinedtheaxialvelocityestimationwithatransverse methodbasedonbandwidthbroadening(seesection2.3.2).Inthis waytheycouldestimatethetotalflowvelocityanddeterminethe Dopplerangle;thenscanningtheopticalfocusallowedthemto performflowmappingasshowninFig.5.

Song et al. applied this method with a laser scanning photoacoustic microscope (LS-PAM) [62]. In LS-PAM, an unfo-cussedultrasoundtransduceriskeptstatic,whiletheopticalfocus isscannedwithinthefieldofviewofthetransducer.Fromascan linealongtubingtheyusedthedepthinformationtoestimatethe flowangle

u

.

2.2.2. Cross-correlationflowimaginginthespatialdomain Liangetal.implementedcross-correlationflowimaginginthe spatial-domainbyprojectinganimageofadigitalmicro-mirror device(DMD)ontoasampleusinganobjective[63].TheDMDwas programmedtoprovidephotoacousticexcitationattwolocations, a certain distanced apart. By switching the DMDpattern, the

Fig.4.Intime-domaincross-correlationflowimaging,thecells’orparticles’density

distributionisinterrogatedwithtwoconsecutivelaserpulses.ThePAresponseof

excitation locations could be alternated, thereby sampling the densitydistributionsateitherlocationseparately(seeFig.6).The

time-shift

D

T between the two slow-time responses over

successivelaserpulsescouldbeusedtoestimatethetransverse flowvelocity:

v

trans¼ d

D

Tsinð

u

Þ

Theydemonstratedthetechniqueusingvariousparticlesizes (5,10,15

m

m)andtheyshowedtheflowmeasuringperformanceis fairly independent of these particle sizes. Measureable flow velocitieswerefrom1.13to13.20mm/swhen300

m

mofchicken breastwasplacedontopofthetubing.Thesystematicerrorintheir latter experiments was 0.29mm/s and their random error 0.19mm/s.Thesamegroupdemonstratedthetechnique with blood-flowinadifferentpublication[64].Theyshowed measure-able velocities of bovine blood flowing through tubing from 0.45mm/sto18mm/s.Severalexamplesofflowestimationin-vivo withmouseearvasculaturewerealsoshown.

In theory the maximum measureable velocity was limited throughj

v

maxj¼d=Trep,butthislimitwasnotreached.Itseems likelythattheirmeasurementswerelimitedbydecorrelationof the density distribution, comparable to what Brunker et al. observed. As was the case with other methods, the minimum measureablevelocitywaslimitedbytheaccuracyoftheerrorsin estimatingthetime-shift

D

T.Implementedasa‘point’evaluation, themeasurementsdidnottakelong(100ms),butwhenscannedin 2Dthetechniquewouldperhapsbeontheslowside.

Theaxial resolutionwaslimitedasbefore bythetransducer bandwidth;the lateralresolution along theflow direction was determined by the distance between the excitation spots; the lateralresolutionacrosstheflowdirectionwasdeterminedbythe spotsize.

Wangetal.developedarelatedtechnique[65].TheyusedPA microscopywithrepeatinglinescanningalongavesselinamouse eartoimageindividualRBCsinthevessel.Byrepeatedlyscanning

thevesseltheycouldimagethemovementofthesecellsovertime andcouldusethistoquantifytheflowvelocity.Theycombinedthe flowvelocitywiththeoxygenationtomodeltherateofoxygen release.ThetechniquewasalsousedbyYaoetal.todeterminethe effectofsensoryinputonbloodflowinmousebrains[66]. 2.3. Transittime

2.3.1. Transit-timeflowimagingofsingle-particles

FangandWangusedaPAM,butinopticalresolutionmodeto estimatethetransittimeofparticlesflowingthroughtheoptical focus[67].Thistransittime,whencombinedwiththesizeofthe focusprovidedanestimationforthelateralflowvelocityascanbe seeninFig.7.Sincetheslow-timeresponseastheparticletransits thefocuswasaconvolutionoftheopticalintensityprofileandthe particlesize,itneededcompensationforthesizeoftheparticles. Theyapproximatedtheresponsebya Gaussianfunctionwitha waistequaltothesumofthefocalandparticlesizes,andfittedthe measureddatausingthisapproximation.

Theyshoweda parabolic flow profileof particles flowingin tubing,andmeasuredspeedsrangingfrom0.35mm/sto1mm/ s.Notethatonlythemagnitudeoftheflowcouldbedetermined, not its directionality in the lateral plane. The minimum and maximummeasureableflowspeedswerenotinvestigatedbythe authors,butthemaximumwaslikelydeterminedbytheoptical focussizeandthePRF:theparticlemustbeinthefocusduringat least3laserpulses.Theminimumwaslikelylimitedbythelength ofthetotalacquisition.Theaxialresolutionwas,asbefore,mainly determined by the transducer bandwidth, whereas the lateral resolutionwasdeterminedsimplybytheopticalfocus.

Sarimallaogluetal.appliedthetechniquetoestimatetheflow speedofmelanomacellsinjectedintomice[68].Theyalsousedit in-vitrotodeterminewhethergoldnanorods(GNRs)wereboundto breastcancercells.TheynoticedaslowerspeedwhenGNRswere bound,arguingthatunboundGNRswouldflowinthecentreofthe tubingandtheseheaviercellswouldnot.Besidesinvestigatingthis transittime technique,Sarimallaogluet al. alsouseda method similar tothose in section 2.2.2,instead using three beams to recordthetime it takesmelanomacells toflowbetween these beams.

The method employed by both authors is limited to the detectionofsingleparticlesorsinglecells:inanapplicationon whole blood it is likely difficultto distinguishindividualRBCs. Nevertheless,itmightfindanapplicationinthedetectionofsingle contrastparticlesorcirculatingtumourcellsinwholeblood. 2.3.2. Transit-timeflowimagingofparticleensembles

Yao&WangandChenetal.proposedasolutiontomakethe single-particlemethodtomoresuitable foruseon wholeblood

[69–72].Whatholdsforsingleparticlesalsoholdsforadensity

Fig.5.In-vivoexampleofflowmappingusingascannedphotoacousticmicroscope.

Adaptedwithpermissionfrom[61].Copyright(2012)bySagePublications.

Fig.6.Principlebehindspatial-domaincross-correlationflowimaging[63,64].The

densitydistributionisinterrogatedattwolocationsoverwithalternatinglaser

pulses.Thetime-shiftbetweenthetwoslow-timeresponsesisdeterminedbythe

flowvelocity.

Fig.7.Transit-timeflowimagingprincipleforsingleparticles[67,68]:thetime

duringwhichacelloraparticleisvisibleisdeterminedbythespeedatwhichit

distributionofparticlesmovingthroughthefocus.Thefasterthe distribution moves through the optical focus, the sharper its features(the‘peaks’)become(seeFig.8).Whileitisineffectiveto fiteverysinglepeakwithaGaussianfunction,itisrelativelyeasyto computeeithertheautocorrelationorthebandwidthofthetotal photoacousticacquisition:bothdeterminetheaveragepeakwidth. For faster flowtheautocorrelation coefficientbecomes smaller, whereasthebandwidthincreases,sincethebandwidthisaFourier transformpairwiththeautocorrelationfunction.

Starting with Yao and Wang, they used the slow-time

bandwidth broadening for transverse flow velocity imaging

[72].Asparticlesflowquickerthroughthefocus,thetransittime becomesshorterandthebandwidth,whichisitsinverse,broadens by: Bd f0

v

trans

v

s w Rsinð

u

Þ

wherewistheopticalfocuswidth,Rthefocallengthand f0the centrefrequency oftheUS transducer.Theauthorsfoundtheir inspirationforthistechniqueinOCT[73].YaoandWangapplied the technique using an optical resolution PAM. They scanned the PAM back and forth, causing a change in broadening: an increase when moving against the flow and vice-versa. They

used this change to determine the flow velocity via

v

trans¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

v

2

tþþ

v

2t2

v

2m q

,where

v

tþand

v

tarespeedsmeasured inbothdirectionswhilescanning,and

v

misthemotorspeed.This scanning approach can be viewed as a combination of laser scanningconfocal[36]andOCT[73].

Theydemonstrated thetechnique using particles (1 6

m

m) flowing through tubing, withmeasureable flow velocities from 0.1mm/supto2.5mm/s.Theyalsomappedtheparabolicflow profileinthetubing.Yaoetal.,inadifferentarticle,subsequently appliedthetechniqueonblood-flowin-vitroandin-vivo[71]. In-vitro,theyshowedmeasureableflowvelocitiesupto5mm/s,and used this to calibrate their in-vivo measurements.They finally appliedtheir methodtoblood flowonvasculature inachicken

embryo (see Fig. 9), and were able tomeasure theflow-speed variationscausedbytheheartbeating.

The theoretical maximum measureableflow velocity of this techniquewaslimitedbytheopticalfocusandthePRFofthelaser illumination,at about7.4mm/s.Theminimum,indicated tobe about0.1mm/s,waslimitedbySNR,andpresumablyacquisition time.

Chenetal.usedtheautocorrelationGð

t

Þ¼

d

PðtÞ

d

Pðtþ

t

Þ=PðtÞ2 to estimate the flow speed [69]. The autocorrelation function, whenassumingaGaussianopticalfocus,couldbeapproximated by: Gð

t

Þ Gð0Þ¼exp 

t

t

0  2!

with

t

0¼w=

v

trans.Instead ofscanningthesystem, theylimited

themselvestoestimatingflow ata fewlocations, andtherefore werenotabletodeterminetheflowdirection.Theydemonstrated the methodin their first article [69]using particles flowing at speedsfrom14

m

m/sto200

m

m/s.Intheirsecondarticle[70]they appliedthe methodon chickenembryo vasculature,measuring flowspeedsatindividuallocations56-77

m

m/s–basedonan in-vitrocalibrationusingparticles.

Ning et al. used theautocorrelation approach toinvestigate bloodflowintheearofanudemouse[74].Usingthediameterfrom thePAimage,theywereabletodeterminethevolumetricflowrate ofthevasculature,andtheinflowandoutflowatbifurcations.They calibratedtheirflowestimationandverifiedthecomputationof thevolumetricflowratebothin-vitro.

Althoughnotinvestigatedbytheauthorsofeitherapproach, universal quantification of flow speed or velocity will remain difficult. The reason: quantification relies on knowledgeof the opticalfocusandpresumablyalsocellsizes.Althoughbothcanbe either measured independentlyor otherwise calibrated for,the opticalfocusmayincreasethroughscattering,andRBCscanform clustersatsmallflowspeeds.

Tayetal.proposedatechniquetoreducetheeffectthatoptical scattering has on the optical focus [75]. They implemented wavefront shaping using the photoacoustic amplitude as a feedbackmechanism.Wavefrontshapingcanbeusedto compen-sateforscatteringbyiterativelyguessingtheverycomplex phase-and-amplitudetransformationsthattheinputbeamexperiences. The ‘goodness’ of the guesswork was observed here using the photoacousticamplitude,whichthereforeservedasthefeedback. This resulted in a bright optical spot at the location of the ultrasoundfocus.Tayetal.implementedtheiterativeprocesswith aDMDprojectingaseriesofHadamardpatternsonthesample.The optimalpatterncouldbefoundbymultiplyingthephotoacoustic responseofthem-thHadamardpattern(vectorIm)bytheinverse Hadamardmatrix.

Inthisway,Tayetal.couldmeasuretheflowspeedofparticles behind a ground diffusorfrom0.4mm/s upto 2mm/s. The random error of their measurements was–at 0.5mm/s–quite high,whichtheyattributedtoparticlesizevariations.Theaxialand

Fig.8.Transit-timeprincipleforparticleensembles[69–72].Ascellstransitthe

focuswithagreaterspeed,featuresofthedensitydistributionwillshowupwith

sharperpeaks.

lateral resolutions were governed here by the ultrasound transducer.

TheoptimizationoftheDMDpatterntook2hours,makingtheir setupunsuitablefortissueimagingastheauthorsnote.Tospeed uptheprocess,afasterDMDandlaserwouldberequiredtobeat the tissue dynamics, which are on a millisecond-timescale. A currentlimitationtotheirmethodistheinabilitytousewavefront shaping to focus on a spot inside and not behind a scattering medium. In theory the autocorrelation/bandwidth broadening technique likely also works based on just the acoustic focus, assumingthewidthandnumericalapertureoftheUStransducer providesufficientSNRwithouttheopticalfocus.

2.4. Amplitudeencoding

2.4.1. Photoconversionandreplenishment

Lietal.estimatedflowusingashapetransitionofgoldnanorods

[76]. Exciting the rods at their peak-absorption wavelength converts them to nanospheres, which have a different peak-absorptionwavelength.Therefore,pulsing light on a volumeof nanorodseffectivelyphoto-destroystherods.Afteralightpulse, newnanorodswillflowintothevolume;arecoverythatisrelated totheflowspeed(seeFig.10).Overtime,repeatedpulsingwill bringthephotoacousticamplitudePðkÞdowntoaconstantlevel

Pconst.Thedropinamplitudecouldbeapproximatedby:

PðkÞ¼Pconstþð1þPconstÞrkexp u k

PRF

 

;

Pconst¼ 1expðu=PRFÞ

1rexpðu=PRFÞ

withkthek-thlaserpulse,uaparameterproportionaltotheflow speed,andraconstantrelatedtothelaserpower.

Theyestimatedtheflowspeedofasuspensionofgoldnanorods flowingthroughtubing.Theymeasuredspeedsfrom0.35mm/sup to2.83mm/swithasystematicerrorof0.2mm/sandrandom errorof30%themeasuredspeed.

Weietal.proposedanalternatemethod[77],usingoneinitial high-energylaserpulsetoconvertalargefractionofthenanorods, andsubsequentlyusinglower-energylaserpulsestotrackthe in-flowofnewnanorods,ascanalsobeseeninFig.11.Therecoveryof thephotoacousticamplitudecouldthenbeapproximatedby:

PðtÞ ¼1cexp

lv

flowt

 

;

with c and

l

constants. This method is very similar to the ultrasoundcontrastflash-replenishmenttypemethods,where a high-intensity burst of ultrasound is used to destruct micro-bubbles.

The authors measured flow speeds of a GNR solution from 0.35to2.83mm/s asbefore. Theycompared thephotoacoustic flash-replenishment method with the constant-energy method fromLietal.,butfoundtheconstant-energymethodtoprovidethe betterperformance.Areasonforthismightbe,theauthorswrite, their limited measurement time that made the measurements inaccurate.

Liao et al.implemented theconstant-energy methodwitha lineararray,providingB-modeflowimaging[78].Theyalsotested theperformanceunder5mmofchickenbreast,andcouldmeasure flowspeedsfrom0.125to2mm/swithanaverageerrorof30%. However,theyhadtorecalibratetheirmeasurementsbecausethe beamwidthhadchangedduetoopticalscattering.Theaxialand lateralresolutionsweredeterminedbytheROIselectioninthe B-modeimageofthetubing,0.6mmand1.2mmrespectively.The theoreticallowerlimitsoftheseresolutionsweredeterminedby thepointspreadfunctionofthesystem.Thetimeresolutionwas determinedbythetotalmeasurementtime,5-20sdependingon thesettings.

Although not investigated by the authors, the maximum measureable speed–when there is only a relatively small drop inPAamplitude–waslikelydeterminedbythelaserfluenceatthe GNRsandthemeasurementtime.Alargerfluencewouldcausea largerdropinamplitudeandalongermeasurementtimewould providea moreaccuratedetermination of Pconst. The minimum measureablespeedlikelydependedonwhetherslowflowcouldbe distinguishedfromnoflow,whichwouldbemainlydeterminedby theSNR.NotethattheauthorsdidnotmixtheGNRswithblood, whichmightdeterioratethecontrastfurther.

2.4.2. Heatencodingandconvection

SheinfeldandEyalproposedtheuseofphotothermal modula-tionforamplitudeencoding[79].Theyusedtwosinemodulated laser diodes of different frequency to create an amplitude modulated photoacoustic signal: the heating of the medium increasesitsGrueneisenparameterandcausesthephotoacoustic responsetoincrease(seeFig.12).Modulatingtheheatingprocess causedanamplitudemodulationofthephotoacousticsignal,of whichthemodulationfrequencyresponsewasgivenby: Pð

v

PTÞ/

teff 1þj

v

PTteff

with

v

PT¼2

p

fPT thephotothermal modulationfrequency and teff¼ t1conductionþtconvection1

 1

acombinationoftimeconstantsof conduction and flow-dependent convection. Sheinfeld and Eyal usedaconstant0.5MHzphotoacousticmodulationandvariedthe PTmodulation

v

PT from0.1Hzto20Hz.Theyfittedtheobserved Pð

v

PTÞ with theabove function todetermine teff. The authors

Fig.10.Inthephotoconversionandreplenishmenttechniquewithconstantenergy

[76],ateverylaserpulse,afractionofthegoldnanorodsisconvertedintospheres.

Aftereverylaserpulse,therodsreplenishpartly,untilanequilibriumforms.

Fig. 11. Flow measurement based on photoconversion using an a flash/

replenishmentapproach[77].Asingleintenselaserpulseconvertsmostofthe

nanorods, after which low intensity pulses are used to observe the rods’

obtainedthecontributionofconductiontoteff bymeasuringthe responseforzeroflow,butfoundthecontributiontoberelatively small.Theycalibratedtherelationshipbetweenteff andtheflow speed.

TheauthorsappliedthePTmodulationonsheepbloodflowing throughtubing,heatingthebloodlessthan28C.Theymeasured flowsinarangeof1to21mm/s.Theirmaximumerrorwas0.3mm/ s,which,whenignoringthecontributionofconductionincreased slightlyto0.75mm/s.Althoughtheauthorsdidnotinvestigatethe speedlimits,theminimummeasureablespeedwaslikely deter-minedsimplybywhetherflowandnoflowcanbedistinguished, which might be dependent on whether the conduction can be

modelled accurately. The maximum measureable speed was

probably linked to the maximum modulation frequency, since for highflow speeds themodulation responsechanges only at higherfrequencies.Theaxial resolutionwaslikelyverylimited, whereasthelateralresolutionwasprobablydeterminedbythesize

of the heating beam. As with photoconversion, deep flow

measurements with this technique are likely very challenging becauseopticalscatteringchangesthesizeoftheheatingbeam.

Wanget al.proposed touseultrasound toheat theflowing medium[80]tosolvetheopticalscatteringissueswiththeabove methods.Theyheatedduringaperiodof300ms,causinga4to78C temperatureincrease,andmeasuredthedecreaseinphotoacoustic amplitudeastheheatedsectiontravelledoutoftheultrasound detectionarea.Theyfoundthisdecreasecouldbeapproximated by:

PðtÞ¼c1þc2elt

withc1andc2constantsand

l

¼

l

v

flow

 

aflow-dependentdecay constant.TheyimplementedthemethodonanAR-PAMsystemto quantify bovine blood flowing through tubing under 1.5mm chickenbreast.Aftercalibrationtheycouldmeasureflowspeeds from2.97mm/supto41mm/s.

Wangetal.,inaseparatepaper,showed ultrasound-heating-basedPA flowmetryusing a tomography setup[81]. Insteadof monitoring at a single point the photoacoustic response after heating,theyusedthetomographysetuptoreconstructa2Dimage of thetubing, in which they couldtrack themovement of the heated spot (see Fig. 13). The setup was based on a HIFU transducer,whichheatedthemediumwitha7.5MHzwaveform, which was modulated with a 0.0625Hz sine to generate the heatingincycles.Theamountheatedwasapproximately18C.The velocity could then be calculated from the rate at which the heatingtravelsalongthevesselatacertainradialposition.This techniquehassomeparallelswitharterialspinlabellingfromMRI

[15],althoughthespinlabellingisusedforperfusionimaging,and not for estimation of flow velocities because it allows longer labellingtimesbutnotaveryrapidimagingrate.

Figure14

Wangetal.measuredbloodflowintubing–withoutscattering– withvelocitiesfrom0.24mm/sto11.8mm/sandaRMSEof2.7%. They alsoshoweda measurementunder a 5mm thickchicken breastlayerwithasingleflowvelocity,whichtheymeasuredat 1.60.2mm/s.Toachievethistheyaveragedoveradditionalheating cycles(35insteadof7).

Wangetal.’sminimummeasureablevelocitywasdetermined bywhethertheflowwasfastenoughtoensurethattheheated regionwasnotyetsmoothedbythermalconduction,andthatthe heated part would not be visible above the noise level. Their maximummeasureableflowvelocitywasdeterminedbythesize ofthefieldofviewandthe2Dimagingrate,sincetheheatedspot needstobewithinthefieldofviewinatleastafewimages.

Anadvantageofthistechniqueisitscapabilityofimagingblood flowatdiffuseopticaldepths,whichismadepossibleduetothe use of ultrasound instead of light for heating. Therefore, the maximumimagingdepthwilllikelydependonlyontheSNRofthe basephotoacousticresponseandon theacousticattenuationby tissueoftheHIFU cycles.The resolutioninthis case waslikely optimal along the transducer axis and across the flow profile, however,alongtheflowdirectionitwaslimitedbythedistancethe heatedregionwasrequiredtomove.

Zhangetal.alsoappliedthistechnique,but withopticalCW heatinginsteadofHIFU[82].Theyimplementedthetechniqueon anOR-PAMsystemandscannedthemicroscopefor2Dimaging instead of relying on a tomography system. Their analysis is identical to that of Wang et al. [81]. They could measure comparable in-vitro velocities (0.23 to 11mm/s with a RMSE

Fig.12.Flowmeasurementusingphotothermalmodulation[79].Alow-frequency

heatinglasermodulatestheGueneisencoefficient,whichcanbeobservedusinga

modulatedPAlaser.

Fig.13.FlowmeasurementbasedonconvectionofanUS-generatedtemperature

increase[81].High-frequencyfocussedultrasoundisemittedinburststoheatapart

oftheflowingmedium.Theflowofthisspotcanbetrackedusingaphotoacoustic

tomographysystem.

Fig. 14. Example of in-vitro heat encoding. Adapted with permission from

of2%).Finally, theyalsosuccessfullyappliedtheirtechnique to bloodflowinamouseear.

2.5. Discussionandconclusion

Inthisreviewweinvestigatedalltheapproachesthathavebeen developedforphotoacousticbloodflowimaging.Table1givesan overviewofthekeyelementsofeachmethod.Ouroverviewshows thatphotoacousticflowimagingcanovercomelimitationsofother modalities such as ultrasound and optical techniques. Photo-acousticsallowsadirect translationofflowimaging techniques fromultrasoundpulse-echo,likethetime-domain cross-correla-tion,or bandwidth broadening fromOCT. There are also some similaritiesbetweenheatencodinginPAandarterialspinlabelling inMRI.Butsomeapproachesareuniqueforphotoacoustics,like thespatial-domainDopplerflowimaging.Allinall,awiderangeof flowimagingmethodsisavailable,ofwhichmanywarrantfurther investigation.

Asdescribedintheabovereview,notallmethodsareequally advancedinitsdevelopment,andmanyhavebeenappliedonly in-vitrowithmoderateresults.Methodsinthemostadvancedstageof development are the implementations using PA microscopy, havingoftenbeendemonstrated in-vivosuchasinamouseear or a chicken embryo. PAM allows high-frequency ultrasound detectionandfocussedopticalexcitationforanoptimalSNRand interrogation of the densely packed red blood cells in blood– thoughatthecostofpenetrationdepth.

Acousticresolutionsetupsalsoshowpromise,andanumberof techniques,suchasthetomographyimplementedheatencoding, have been shown to work under optical-scattering conditions. Signaltrackingmethods,suchasthetime-domain cross-correla-tion,intheoryarealsopromisingfortomographysetups,asthey arebasedonlyontheuseofapulsedlaser.Themethodsdorequire someoptimizationofdetectionfrequencyandbandwidth howev-er,foritseemsthatthesameprinciplethatmakesPAIspeckle-free at5MHz,willalsosmoothdensityvariationsinredbloodcells. On theother hand,at higher frequencies it appearsPAI might surpassultrasoundinthevisibilityofRBCs[47,48].

Uptonow,PAflowimagingseemsnotyetappliedclinicallyor pre-clinically.Althoughitwillnotlikelycompetewithmodalities likeMRIorCT,itmightfindapplicationsinareaswhereUS, two-photonmicroscopyorOCTflowimagingareactive.Inthefuture,

photoacoustic tomography and microscopy may provide flow

imagingcapabilities–exploitingitshighcontrasttohaemoglobin– toapplications ranging fromrheumatoid arthritis diagnosis, to burn-woundassessment.

Conflictofinterest

Theauthorsdeclarethattherearenoconflictsofinterest. Acknowledgement

This research was funded by the European Community’s

Seventh Framework Programme (FP7/2007-2013) under grant

agreementn8318067. References

[1]R.S.Reneman,T.Arts,A.P.G.Hoeks,Wallshearstress-animportant determinantofendothelialcellfunctionandstructure-inthearterialsystem invivo,JVascRes43(3)(2006)251–269.

[2]M.Kadbi,M.J.Negahdar,J.W.Cha,M.Traughber,P.Martin,M.F.Stoddard, etal.,4DUTEFlow:APhase-ContrastMRITechniqueforAssessmentand VisualizationofStenoticFlows,MagnResonMed73(3)(2015)939–950.

[3]D.N.Ku,Bloodflowinarteries,AnnuRevFluidMech29(1997)399–434.

[4]J.Folkman,AngiogenesisinCancer,Vascular,RheumatoidandOtherDisease, NatMed1(1)(1995)27–31.

[5]D.Hanahan,R.A.Weinberg,HallmarksofCancer:TheNextGeneration,Cell 144(5)(2011)646–674.

[6]E.Choy,Understandingthedynamics:pathwaysinvolvedinthepathogenesis ofrheumatoidarthritis,Rheumatology51(2012)V3–V11.

[7]P.Vaupel,F.Kallinowski,P.Okunieff,Blood-Flow,OxygenandNutrientSupply, andMetabolicMicroenvironmentofHuman-Tumors-aReview,CancerRes49 (23)(1989)6449–6465.

[8]A.D.Jaskille,J.W.Shupp,M.H.Jordan,J.C.Jeng,CriticalReviewofBurnDepth AssessmentTechniques:PartI,HistoricalReview.JBurnCareRes30(6) (2009)937–947.

[9]U.Dirnagl,C.Iadecola,M.A.Moskowitz,Pathobiologyofischaemicstroke:an integratedview,TrendsNeurosci22(9)(1999)391–397.

[10]J.D.Briers,LaserDoppler,speckleandrelatedtechniquesforbloodperfusion mappingandimaging,PhysiolMeas22(4)(2001)R35–R66.

[11]P.V.Pandharipande,G.A.Krinsky,H.Rusinek,V.S.Lee,Perfusionimagingofthe liver:Currentchallengesandfuturegoals,Radiology234(3)(2005)661–673.

[12]T.Barrett,M.Brechbiel,M.Bernardo,P.L.Choyke,MRIoftumorangiogenesis,J MagnResonImaging26(2)(2007)235–249.

[13]P.D.Gatehouse,J.Keegan,L.A.Crowe,S.Masood,R.H.Mohiaddin,K.F.Kreitner, etal.,Applicationsofphase-contrastflowandvelocityimagingin

cardiovascularMRI,EurRadiol15(10)(2005)2172–2184.

[14]M.Markl,A.Frydrychowicz,S.Kozerke,M.Hope,O.Wieben,4DflowMRI,J MagnResonImaging36(5)(2012)1015–1036.

[15]J.A.Detre,J.J.Wang,Z.Wang,H.Y.Rao,Arterialspin-labeledperfusionMRI inbasicandclinicalneuroscience,CurrOpinNeurol22(4)(2009)348– 355.

[16]A.R.Padhani,Dynamiccontrast-enhancedMRIinclinicaloncology:Current statusandfuturedirections,JMagnResonImaging16(4)(2002)407–422.

[17]G.P.Schmidt,M.F.Reiser,A.Baur-Melnyk,Whole-bodyMRIforthestaging andfollow-upofpatientswithmetastasis,EurJRadiol70(3)(2009)393–400.

[18]M.Notohamiprodjo,M.F.Reiser,S.P.Sourbron,Diffusionandperfusionofthe kidney,EurJRadiol76(3)(2010)337–347.

[19]C.S.Kidwell,A.W.Hsia,Imagingofthebrainandcerebralvasculaturein patientswithsuspectedstroke:AdvantagesanddisadvantagesofCTandMRI, CurrNeurolNeurosci6(1)(2006)9–16.

[20]W.vanElmpt,M.Das,M.Hullner,H.Sharifi,C.M.L.Zegers,B.Reymen,etal., CharacterizationoftumorheterogeneityusingdynamiccontrastenhancedCT andFDG-PETinnon-smallcelllungcancer,RadiotherOncol109(1)(2013) 65–70.

[21]M.P.Spencer,J.M.Reid,QuantitationofCarotidStenosiswith Continuous-Wave(C-W)DopplerUltrasound,Stroke10(3)(1979)326–330.

[22]T.Loupas,J.T.Powers,R.W.Gill,AnAxialVelocityEstimatorforUltrasound Blood-FlowImaging,BasedonaFullEvaluationoftheDopplerEquationby Meansofa2-DimensionalAutocorrelationApproach,IeeeTUltrasonFerr42 (4)(1995)672–688.

[23]A.Swillens,P.Segers,H.Torp,L.Lovstakken,Two-DimensionalBloodVelocity EstimationWithUltrasound:SpeckleTrackingVersusCrossed-BeamVector DopplerBasedonFlowSimulationsinaCarotidBifurcationModel,IeeeT UltrasonFerr57(2)(2010)327–339.

[24]C.Greis,Ultrasoundcontrastagentsasmarkersofvascularityand microcirculation,ClinHemorheolMicro43(1–2)(2009)1–9.

[25]F.S.Foster,C.J.Pavlin,K.A.Harasiewicz,D.A.Christopher,D.H.Turnbull, Advancesinultrasoundbiomicroscopy,UltrasoundMedBiol26(1)(2000)1– 27.

[26]K.W.Ferrara,C.R.B.Merritt,P.N.Burns,F.S.Foster,R.F.Mattrey,S.A.Wickline, EvaluationoftumorangiogenesiswithUS:Imaging,Doppler,andcontrast agents,AcadRadiol7(10)(2000)824–839.

[27]S.K.Alam,K.J.Parker,Implementationissuesinultrasonicflowimaging, UltrasoundMedBiol29(4)(2003)517–528.

[28]M.Draijer,E.Hondebrink,T.vanLeeuwen,W.Steenbergen,Reviewoflaser specklecontrasttechniquesforvisualizingtissueperfusion,LaserMedSci24 (4)(2009)639–651.

[29]N.Hecht,J.Woitzik,J.P.Dreier,P.Vajkoczy,Intraoperativemonitoringof cerebralbloodflowbylaserspecklecontrastanalysis,NeurosurgFocus27(4) (2009).

[30]S.C.Sandker,E.Hondebrink,J.G.Grandjean,W.Steenbergen,Laserspeckle contrastanalysisforquantifyingtheAllentest:Afeasibilitystudy,LaserSurg Med46(3)(2014)186–192.

[31]A.B.Parthasarathy,W.J.Tom,A.Gopal,X.J.Zhang,A.K.Dunn,Robustflow measurementwithmulti-exposurespeckleimaging,OptExpress16(3)(2008) 1975–1989.

[32]W.Groner,J.W.Winkelman,A.G.Harris,C.Ince,G.J.Bouma,K.Messmer,etal., Orthogonalpolarizationspectralimaging:Anewmethodforstudyofthe microcirculation,NatMed5(10)(1999)1209–1213.

[33]R.A.Leitgeb,R.M.Werkmeister,C.Blatter,L.Schmetterer,DopplerOptical CoherenceTomography,ProgRetinEyeRes41(2014)26–43.

[34]G.J.Liu,A.J.Lin,B.J.Tromberg,Z.P.Chen,AcomparisonofDoppleroptical coherencetomographymethods,BiomedOptExpress3(10)(2012)2669– 2680.

[35]H.W.Ren,X.D.Li,ClutterrejectionfiltersforopticalDopplertomography,Opt Express14(13)(2006)6103–6112.

[36]A.Villringer,A.Them,U.Lindauer,K.Einhaupl,U.Dirnagl,CapillaryPerfusion oftheRat-BrainCortex-anin-VivoConfocalMicroscopyStudy,CircRes75 (1)(1994)55–62.

[37]D.Kleinfeld,P.P.Mitra,F.Helmchen,W.Denk,Fluctuationsand stimulus-inducedchangesinbloodflowobservedinindividualcapillariesinlayers

2through4ofratneocortex,PNatlAcadSciUSA95(26)(1998)15741– 15746.

[38]A.Devor,S.Sakadzic,V.J.Srinivasan,M.A.Yaseen,K.Nizar,P.A.Saisan,etal., Frontiersinopticalimagingofcerebralbloodflowandmetabolism,Journalof cerebralbloodflowandmetabolism:officialjournaloftheInternational SocietyofCerebralBloodFlowandMetabolism32(7)(2012)1259–1276.

[39]P.Beard,Biomedicalphotoacousticimaging,Interfacefocus1(4)(2011)602– 631.

[40]K.Jansen,M.Wu,A.F.vanderSteen,G.vanSoest,Lipiddetectionin atherosclerotichumancoronariesbyspectroscopicintravascularphotoacoustic imaging,OptExpress21(18)(2013)21472–21484.

[41]J.Glatz,N.C.Deliolanis,A.Buehler,D.Razansky,V.Ntziachristos,Blindsource unmixinginmulti-spectraloptoacoustictomography,OptExpress19(4) (2011).

[42]R.O.Esenaliev,I.V.Larina,K.V.Larin,D.J.Deyo,M.Motamedi,D.S.Prough, Optoacoustictechniquefornoninvasivemonitoringofbloodoxygenation:a feasibilitystudy,ApplOptics41(22)(2002)4722–4731.

[43]S.Gottschalk,T.F.Fehm,X.L.Dean-Ben,D.Razansky,Noninvasivereal-time visualizationofmultiplecerebralhemodynamicparametersinwholemouse brainsusingfive-dimensionaloptoacoustictomography,JCerebrBloodFMet 35(4)(2015)531–535.

[44]K.Daoudi,P.J.VanDenBerg,O.Rabot,A.Kohl,S.Tisserand,P.Brands,etal., Handheldprobeintegratinglaserdiodeandultrasoundtransducerarrayfor ultrasound/photoacousticdualmodalityimaging,Opt.Express22(21)(2014) 26365–26374.

[45]L.H.V.Wang,S.Hu,PhotoacousticTomography:InVivoImagingfrom OrganellestoOrgans,Science335(6075)(2012)1458–1462.

[46]E.M.Strohm,E.S.Berndl,M.C.Kolios,Probingredbloodcellmorphologyusing high-frequencyphotoacoustics,Biophysicaljournal105(1)(2013)59–67.

[47]Z.J.Guo,L.Li,L.H.V.Wang,Onthespeckle-freenatureofphotoacoustic tomography,MedPhys36(9)(2009)4084–4088.

[48]Z.J.Guo,Z.Xu,L.H.V.Wang,Dependenceofphotoacousticspeckleson boundaryroughness,JBiomedOpt17(4)(2012).

[49]C.B.Burckhardt,SpeckleinUltrasoundB-ModeScans,IeeeTSonUltrason25 (1)(1978)1–6.

[50]Y.J.Yu,S.T.Acton,Specklereducinganisotropicdiffusion,IeeeTImageProcess 11(11)(2002)1260–1270.

[51]J.M.Schmitt,S.H.Xiang,K.M.Yung,Speckleinopticalcoherencetomography,J BiomedOpt4(1)(1999)95–105.

[52]H.Fang,K.Maslov,L.Wang,PhotoacousticDopplerEffectfromFlowingSmall Light-AbsorbingParticles,Phys.Rev.Lett.99(18)(2007)184501.

[53]H.Fang,K.Maslov,L.V.Wang,PhotoacousticDopplerflowmeasurementin opticallyscatteringmedia,AppliedPhysicsLetters91(26)(2007)264103.

[54]A.Sheinfeld,S.Gilead,A.Eyal,Simultaneousspatialandspectralmappingof flowusingphotoacousticDopplermeasurement,JBiomedOpt15(6)(2010).

[55]A.Sheinfeld,S.Gilead,A.Eyal,PhotoacousticDopplermeasurementofflow usingtoneburstexcitation,OptExpress18(5)(2010)4212–4221.

[56]R.Zhang,J.Yao,K.I.Maslov,L.V.Wang,Structured-illuminationphotoacoustic Dopplerflowmetryofaxialflowinhomogeneousscatteringmedia,ApplPhys Lett103(9)(2013)94101.

[57]J.J.Yao,R.C.Gilson,K.I.Maslov,L.D.Wang,L.H.V.Wang,Calibration-free structured-illuminationphotoacousticflowgraphyoftransverseflowin scatteringmedia,JBiomedOpt19(4)(2014).

[58]J.Brunker,P.Beard,PulsedphotoacousticDopplerflowmetryusing time-domaincross-correlation:accuracy,resolutionandscalability,J.Acoust.Soc. Am.132(3)(2012)1780–1791.

[59]J.Brunker,P.Beard,AcousticresolutionphotoacousticDopplerflowmetry: practicalconsiderationsforobtainingaccuratemeasurementsofbloodflow, Proc.ofSPIE8943(2014)89431K.

[60]W.F.Walker,G.E.Trahey,AFundamentalLimitonDelayEstimationUsing PartiallyCorrelatedSpeckleSignals,IeeeTUltrasonFerr42(2)(1995)301– 308.

[61]J.J.Yao,K.I.Maslov,L.H.V.Wang,InvivoPhotoacousticTomographyofTotal BloodFlowandPotentialImagingofCancerAngiogenesisand

Hypermetabolism,TechnolCancerResT11(4)(2012)301–307.

[62]W.Song,W.Liu,H.F.Zhang,Laser-scanningDopplerphotoacousticmicroscopy basedontemporalcorrelation,AppliedPhysicsLetters102(20)(2013).

[63]J.Y.Liang,Y.Zhou,K.I.Maslov,L.H.V.Wang,Cross-correlation-basedtransverse flowmeasurementsusingopticalresolutionphotoacousticmicroscopywitha digitalmicromirrordevice,JBiomedOpt18(9)(2013).

[64]Y.Zhou,J.Y.Liang,K.I.Maslov,L.H.V.Wang,Calibration-freeinvivotransverse bloodflowmetrybasedoncrosscorrelationofslowtimeprofilesfrom photoacousticmicroscopy,OptLett38(19)(2013)3882–3885.

[65]L.D.Wang,K.Maslov,L.H.V.Wang,Single-celllabel-freephotoacoustic flowoxigraphyinvivo,PNatlAcadSciUSA110(15)(2013)5759–5764.

[66]J.J.Yao,L.D.Wang,J.M.Yang,K.I.Maslov,T.T.W.Wong,L.Li,etal.,High-speed label-freefunctionalphotoacousticmicroscopyofmousebraininaction,Nat Methods12(5)(2015)407-+.

[67]H.Fang,L.H.V.Wang,M-modephotoacousticparticleflowimaging,OptLett 34(5)(2009)671–673.

[68]M.Sarimollaoglu,D.A.Nedosekin,Y.Simanovsky,E.I.Galanzha,V.P.Zharov,In vivophotoacoustictime-of-flightvelocitymeasurementofsinglecellsand nanoparticles,OptLett36(20)(2011)4086–4088.

[69]S.L.Chen,T.Ling,S.W.Huang,H.W.Baac,L.J.Guo,Photoacousticcorrelation spectroscopyanditsapplicationtolow-speedflowmeasurement,OptLett35 (8)(2010)1200–1202.

[70]S.L.Chen,Z.X.Xie,P.L.Carson,X.D.Wang,L.J.Guo,Invivoflowspeed measurementofcapillariesbyphotoacousticcorrelationspectroscopy,Opt Lett36(20)(2011)4017–4019.

[71]J.Yao,K.I.Maslov,Y.Shi,L.A.Taber,L.V.Wang,Invivophotoacousticimaging oftransversebloodflowbyusingDopplerbroadeningofbandwidth,Opt.Lett. 35(9)(2010).

[72]J.Yao,L.V.Wang,TransverseflowimagingbasedonphotoacousticDoppler bandwidthbroadening,JBiomedOpt15(2)(2010)021304.

[73]H.W.Ren,K.M.Brecke,Z.H.Ding,Y.H.Zhao,J.S.Nelson,Z.P.Chen,Imagingand quantifyingtransverseflowvelocitywiththeDopplerbandwidthina phase-resolvedfunctionalopticalcoherencetomography,OptLett27(6)(2002)409– 411.

[74]B.Ning,M.J.Kennedy,A.J.Dixon,N.D.Sun,R.Cao,B.T.Soetikno,etal., Simultaneousphotoacousticmicroscopyofmicrovascularanatomy,oxygen saturation,andbloodflow,OptLett40(6)(2015)910–913.

[75]J.W.Tay,J.Y.Liang,L.V.Wang,Amplitude-maskedphotoacousticwavefront shapingandapplicationinflowmetry,OptLett39(19)(2014)5499–5502.

[76]P.C.Li,S.W.Huang,C.W.Wei,Y.C.Chiou,C.D.Chen,C.R.C.Wang,

Photoacousticflowmeasurementsbyuseoflaser-inducedshapetransitionsof goldnanorods,OptLett30(24)(2005)3341–3343.

[77]C.W.Wei,S.W.Huang,C.R.C.Wang,P.C.Li,Photoacousticflowmeasurements basedonwash-inanalysisofgoldnanorods.IEEETransactionsonUltrasonics, Ferroelectrics,andFrequencyControl54(6)(2007)1131–1141.

[78]C.K.Liao,S.W.Huang,C.W.Wei,P.C.Li,Nanorod-basedflowestimationusing ahigh-frame-ratephotoacousticimagingsystem,JBiomedOpt12(6)(2007).

[79]A.Sheinfeld,A.Eyal,Photoacousticthermaldiffusionflowmetry,BiomedOpt Express3(4)(2012)800–813.

[80]L.D.Wang,J.J.Yao,K.I.Maslov,W.X.Xing,L.H.V.Wang,Ultrasound-heated photoacousticflowmetry,JBiomedOpt18(11)(2013).

[81]L.Wang,J.Xia,J.Yao,K.I.Maslov,L.V.Wang,UltrasonicallyEncoded PhotoacousticFlowgraphyinBiologicalTissue,PhysicalReviewLetters111 (20)(2013)204301.

[82]R.Y.Zhang,L.D.Wang,J.J.Yao,C.H.Yeh,L.H.V.Wang,Invivoopticallyencoded photoacousticflowgraphy,OptLett39(13)(2014)3814–3817.

PimvandenBergisaPhDresearcherattheUniversity

of Twente, the Netherlands. He is working on the

EuropeanprojectFullphase,whichaimstodevelopan

affordableandportableultrasound/photoacoustic(US/

PA) system for early disease detection. His main

researchinterestsareflowimagingusing

photoacous-tics and the application of US/PA imaging for the

assessmentofrheumatoidarthritis.Beforestartinghis

PhD, Pim did his master studies on Optics and

Biophysics,buildingaSTORMsuperresolution

micro-scope and used it for characterization of protein

aggregation in Parkinson’s disease. Interests also

includehighschoolsciencepromotion,having

partici-patedinamediapusharoundphotoacousticimagingforthepopularizationof

appliedsciences.

KhalidDaoudi received his PhDdegree in Applied

Optics fromuniversity Pierre et MarieCurie,Paris,

Franceforhisworkontransientopto-elastography.He

iscurrentlyworkingasPostdoctoralresearcheratthe

Institute for Biomedical Technology and Technical

Medicine, BMPI group, university of Twente in

Netherlands.Hisresearchfocusesonopticalandhybrid

acousticalandopticalimagingmethods:

photoacous-tics,acousto-optics,lighttissueinteractionand

ultra-sound.

Wiendelt Steenbergen obtained a masterdegree in

aerospaceengineeringattheUniversityofTechnology

inDelft.In1995heobtained aPhDdegreeinfluid

dynamicsattheEindhovenUniversityofTechnology.In

1995hejoinedtheUniversityofTwente,Enschede(the

Netherlands)asapostdoc.In2000hewasappointed

assistantprofessorinbiomedicalopticsin2000and

broadenedhisscopetolowcoherenceinterferometry

andphotoacousticandacousto-opticimaging.In

2007-2008hewasvisitingresearcherintheKrotoInstituteof

theUniversityofSheffield.In2008hewasappointed

associate professor, and in 2010 he became full

professor and group leader of the newly formed

BiomedicalPhotonicImaginggroupoftheUniversity