(kV and MV) radiotherapy imager

13 Pages·2015·1.77 MB·English

Theoretical investigation of the design and performance of a dual energy
(kV and MV) radiotherapy imager
LangechuanLiu,LarryE.Antonuk,a)YoucefEl-Mohri,QihuaZhao,andHaoJiang
DepartmentofRadiationOncology,UniversityofMichigan,AnnArbor,Michigan48109
(Received31October2014;revised19February2015;acceptedforpublication22February2015;
published31March2015)
Purpose: In modern radiotherapy treatment rooms, megavoltage (MV) portal imaging and kilo-
voltage (kV) cone-beam CT (CBCT) imaging are performed using various active matrix flat-panel
imager (AMFPI) designs. To expand the clinical utility of MV and kV imaging, MV AMFPIs
incorporating thick, segmented scintillators and, separately, kV imaging using a beam’s eye view
geometry have been investigated by a number of groups. Motivated by these previous studies, it is
of interest to explore to what extent it is possible to preserve the benefits of kV and MV imaging
using a single AMFPI design, given the considerably different x ray energy spectra used for kV
and MV imaging. In this paper, considerations for the design of such a dual energy imager are
explored through examination of the performance of a variety of hypothetical AMFPIs based on
xrayconvertersemployingsegmentedscintillators.
Methods: Contrast, noise, and contrast-to-noise ratio performances were characterized through
simulationmodelingofCBCTimaging,whilemodulationtransferfunction,Swankfactor,andsignal
performance were characterized through simulation modeling of planar imaging. The simulations
were based on a previously reported hybrid modeling technique (accounting for both radiation and
optical effects), augmented through modeling of electronic additive noise. All designs employed
BGOscintillatormaterialwiththicknessesrangingfrom0.25to4cmandelement-to-elementpitches
rangingfrom0.508to1.016mm.AseriesofstudieswereperformedunderbothkVandMVimaging
conditions to determine the most advantageous imager configuration (involving front or rear x ray
illumination and use of a mirror or black reflector), converter design (pitch and thickness), and
operatingmode(pitch-binningcombination).
Results:Undertheassumptionsofthepresentstudy,themostadvantageousimagerdesignwasfound
toemployrearilluminationoftheconverterincombinationwithablackreflector,incorporateaBGO
converterwitha0.508mmpitchanda2cmthickness,andoperateatfullresolutionforkVimaging
and2×2binningmodeforMVimaging.Suchadualenergyimagerdesignshouldprovidesofttissue
visualization at low, clinically practical doses under MV conditions, while helping to preserve the
highspatialresolutionandhighcontrastofferedbykVimaging.
Conclusions: The authors’ theoretical investigation suggests that a dual energy imager capable of
largely preserving the desirable characteristics of both kV and MV imaging is feasible. Such an
imager, when coupled to a dual energy radiation source, could facilitate simplification of current
treatment room imaging systems (as well as their associated quality assurance), and facilitate more
preciseintegrationofkVandMVimaginginformationbyvirtueofreducedgeometricuncertainties.
C 2015AmericanAssociationofPhysicistsinMedicine.[http://dx.doi.org/10.1118/1.4915120]
Keywords:dualenergyimager,megavoltagecone-beamCT,kilovoltagecone-beamCT,flat-panel
imager, active matrix flat panel imager, hybrid modeling, Monte Carlo simulation, segmented
crystallinescintillators
1. INTRODUCTION transistor(TFT).1PortalimagingisperformedusingtheMV
treatmentbeamandanAMFPIemployingaconvertertaking
Inexternalbeamradiotherapy,thegoalofmaximizingradia- theformofathickphosphorscreenandametalplate(e.g.,an
tiondosetothetumor,whileminimizingdosetosurrounding ∼133 mg cm−2 Gd O S:Tb screen and an ∼1 mm copper
2 2
healthytissues,isassistedthroughroutineuseofmegavoltage plate). kV CBCT imaging is performed using a dedicated,
(MV)portalimagingand/orkilovoltage(kV)cone-beamCT independentdiagnosticxraysourceandanAMFPIemploying
(CBCT) imaging in the treatment room. In each case, the aconvertertypicallyconsistingofan∼600µmCsI:Tlscreen.2
imaging detector is typically based on an indirect detection The kV source and imager are mounted to the gantry of the
active matrix flat panel imager (AMFPI) which consists of treatment machine, positioned at a 90◦ rotational offset with
some form of overlying x ray converter coupled to a two- respecttothecentralaxisofthetreatmentbeam.
dimensional(2D)pixelatedarray,eachpixelofwhichcontains Inthecontextofthesewidelypracticedimagingtechniques,
an amorphous silicon (a-Si:H) photodiode and a thin-film it is interesting to note that a number of investigators have
2072 Med.Phys.42(4),April2015 0094-2405/2015/42(4)/2072/13/$30.00 ©2015Am.Assoc.Phys.Med. 2072
2073 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2073
examined the possibility of expanding the clinically useful image acquisitions has been explored with a conventional,
information that can be obtained from imaging with the orthogonallymountedkVsystem.)34
treatment beam or a kilovoltage x ray source. For example, Given the considerably different x ray energy spectra for
the same combination of a treatment beam and a MV kV and MV imaging, it is of interest to determine to what
AMFPI used for portal imaging can also be employed to extent it is possible to preserve the benefits of kV and
perform volumetric imaging (i.e., MV CBCT) which offers MV imaging using a single imaging detector. In this paper,
several distinct advantages compared to kV CBCT. These considerationsforthedesignofsuchadualenergyimagerare
include the reduction of streak artifacts caused by metal exploredthroughexaminationoftheperformanceofavariety
implants3andthereadyapplicabilityofMVCTnumbersfor ofhypotheticalAMFPIsincorporatingxrayconvertersbased
treatmentplanningdosecalculations.4However,conventional on the segmented scintillator approach. The performance of
MV AMFPIs suffer from very low x ray quantum efficiency suchimagersischaracterizedthroughsimulationmodelingin
(QE) and, as a result, very low detective quantum efficiency terms of contrast, noise, and CNR for a contrast phantom
(∼2% and 1% at 6 MV, respectively),5 necessitating the use using volumetric (CBCT) imaging, as well as modulation
ofrelativelyhighimagingdoses(e.g.,∼50–200cGy)inorder transferfunction(MTF),Swankfactor,andsignalusingplanar
to achieve soft tissue visualization using MV CBCT.6,7 To imaging.
addressthischallenge,variousapproachesforincreasingthe
QE of MV imagers have been investigated or implemented,
such as xenon gas ion chambers arranged in a fan-beam 2. METHODS
geometry,8andthick,segmentedscintillatingcrystalsarranged
2.A. Converterdesignsexaminedinthestudy
in the form of a linear array9–12 or a 2D matrix.13–24 In
particular, the approach involving a 2D matrix of segmented Eachconverterdesignconsistsofa2Dmatrixofidentical
scintillator elements (based on CsI:Tl, Bi Ge O [BGO], or elementscomprisingrectangularcuboid-shaped,scintillating
4 3 12
Lu Y SiO [LYSO]) has been explored both theoretically crystalsopticallyisolatedby0.05mmthick,polystyreneseptal
1.8 0.2 5
and empirically.17–24 Based on this approach, one prototype walls.24(Thus,asinRef.24,theelementsineachconverterare
employing ∼1.13 cm thick BGO crystals has demonstrated notfocusedtowardtheradiationsource.)Alldesignsemploy
DQE values as high as ∼20%—representing an ∼20-fold BGOmaterialwiththicknessesrangingfrom0.25to4cm,and
increasecomparedwiththatofconventionalMVAMFPIs.19 element-to-element pitches (referred to as converter pitches)
Such a large increase in DQE has been shown to enable the ranging from 0.508 to 1.016 mm. BGO was chosen due to
acquisitionofhighqualityMVCBCTimagesatatotaldose the promising performance exhibited by previous prototypes
as low as ∼4 cGy (Ref. 21)—an amount comparable to that basedonthismaterial,whichoffersdesirablepropertiessuch
required to obtain a single portal image from a conventional ashighelectrondensity,highrefractiveindex,andhighoptical
MVAMFPI.1 transparency.19,21,23Thelowerlimitforthicknesscorresponds
Expansion of the clinical utility of kV CBCT can be toapointwheretheQEforMVimagingis∼15%,representing
achievedthroughuseofabeam’seyeview(BEV)geometry a substantial improvement compared to that of conventional
inwhichthediagnosticsourceispositionedsoastoprovide MV AMFPIs, whereas the 4 cm upper limit (for which the
thesameradiationfieldofview(FOV)asthatofthetreatment QE is ∼80%) corresponds to a point beyond which the rate
source. Such BEV kV imaging has been investigated using of improvement in QE as a function of thickness diminishes
differentapproaches:throughtheintegrationofanadditional rapidly.Thelowerlimitforpitchcorrespondstoapointbelow
kVsourceinthetreatmenthead,25–27andthroughmodification which the scintillator fill factor drops sharply (due to the
of the treatment beam line so as to increase the low energy volume occupied by the fixed septal wall thickness), while
component of the beam.28–32 The latter approach has been theupperlimitroughlycorrespondstoapointbeyondwhich
explored through the use of a low-Z target,28–30 a modified the advantageous spatial resolution offered by kV imaging
linear accelerator waveguide,31 or reduction of the electron wouldbeseverelycompromised.Notethatthe1mmcopper
beamenergy.32ComparedwiththecurrentkVCBCTimaging platecommonlyemployedasabuild-uplayerintheconverter
whichisperformedwiththeaforementionedrotationaloffset of conventional portal imagers was eliminated to avoid the
ofthekVsourcerelativetotheMVsource,BEVkVimaging detrimentaleffectoffilteringoflowenergyxraysthatprovide
would eliminate the geometric uncertainties associated with highcontrastinkVimaging.(TheperformanceofMVimaging
thatoffsetandtheneedforadditionalqualityassuranceeffort systems operated in the absence of an overlying metal plate
to ensure the coincidence of the isocenters of the kV and hasbeenpreviouslyexamined.)35,36Alsonotethatareflector
MV radiation fields, while preserving the superior contrast withnegligibleradiationattenuationwasintroducedtoprovide
ofimagescomparedtothatobtainedusingtheMVtreatment desirableopticalpropertiesandwasmodeledinthesimulation
beam.BEVkVimagingcouldalsoenhancetheeffectiveness withazeromassattenuationcoefficientfortraversingxrays.
of tumor tracking33 in a near real-time mode. Furthermore, Twoextremecasesforthereflectorwereexamined:onewith
the coincidence of the kV and MV FOVs would facilitate 100%absorptivityand0%reflectivity(referredtoas“black”),
reconstruction of images with the complementary strengths and the other with 0% absorptivity and 100% reflectivity
of kV and MV imaging, such as superior contrast-to-noise (referredtoas“mirror”).
ratio (CNR) and reduced metal artifacts, respectively. (Note TheAMFPIarraywasmodeledasan∼1mmthickbarium-
that CBCT imaging based on a combination of kV and MV dopedglassplaterepresentingthearraysubstrate,37whilethe
MedicalPhysics,Vol.42,No.4,April2015
2074 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2074
F.1. Radiationdoseprofilescorrespondingto100kVp(solidline)and6
MV(dashedline)beamspectraalongthedepthofan∼1cmthickBGOcon-
verter,obtainedusingthesameMonteCarlosimulationtechniquesdescribed
inSec.2.B.1.Thedosevaluesforeachspectrumhavebeennormalizedto F. 2. Schematic, cutaway diagrams of imagers based on converters em-
unityattheirrespectivemaxima.Notethat,whereasthekVprofileexhibits ployingasegmentedscintillatorillustrating(a)frontand(b)rearillumination
a sharp decrease with depth, the MV profile follows the familiar pattern configurations. The transparent bell-shaped regions superimposed on the
associatedwiththedepth-dosedistributionforatreatmentbeam(Ref.39). diagramssignifytheapproximatelateralspreadofopticalphotonsreaching
thearrayunderkVimagingconditions.Thearrowsappearinginthesere-
gionscorrespondtoexamplesofpossibletrajectoriesofopticallightphotons
reachingthearray.
∼1 µm thick pixel circuitry fabricated on the substrate was
ignored due to its negligible effect on radiation attenuation.
Thearraywascoupledtothesideoftheconverteroppositeto array traverse much shorter distances before being detected,
thatwherethereflectorresides.Inthemodel,avalueof58% therebyreducinglateralopticalspreadandpreservingspatial
wasusedfortheopticalcouplingefficiencyofthephotodiode38 resolution of the imager. In the case of MV imaging, there
for light emitted by BGO. Throughout the study, the pixel isaparallel,thoughmuchreduced,benefit—duetothemore
pitchofthearraywasassumedtobethesameastheconverter uniformdistributionofabsorbedenergyacrossthescintillator,
pitch. asillustratedinFig.1.
Inadditiontothetworeflectortypes,thepositioningofthe Foreachconverterofacertainthicknessandpitch,atotal
array and the reflector relative to the x ray source was also offourconfigurationswereexplored:frontilluminationwitha
varied.Inthispaper,thesideoftheconverterfacingthexray blackreflector(referredtoas“front-black”),frontillumination
sourceisreferredtoastheentrancesurfacewhiletheotherside withamirrorreflector(“front-mirror”),rearilluminationwith
isreferredtoastheexitsurface.UnderkVimagingconditions, a black reflector (“rear-black”), and rear illumination with a
dose drops quickly with depth into the converter from the mirrorreflector(“rear-mirror”).
entrancesurface,withmostoftheradiationstoppinginthefirst
∼2mmwhile,foranMVbeam,doseisdepositedmoreevenly
2.B. Simulationframework
throughout the thickness of the converter—as illustrated in
Fig.1.Thissignificantdifferenceindosedepositionrequires TocharacterizetheMTFandCNRoftheimager,aMonte
careful design of the converter to preserve the respective Carlo-basedhybridmodelingtechniquereportedinaprevious
benefitsofbothkVandMVimaging.Sinceenergydeposition study was employed.24 In addition, the effect of electronic
for kV photons is concentrated near the entrance surface, a additive noise was introduced through an analytical circuit
conventionalfrontilluminationconfiguration,wherethearray noisemodel.Aflowchartillustratingthemajorimplementation
iscoupledtotheexitsurfaceandthereflectortotheentrance steps of this coupled framework is shown in Fig. 3 and is
surface as illustrated in Fig. 2(a), would lead to significant describedbelow.
lateralspreadofthoseopticalphotonsreachingthearray.This
lateralspread,whichisaconsequenceofimperfectionsinthe
2.B.1. Hybridmodel
optical isolation between neighboring scintillator elements,
results in degradation of spatial resolution—an effect which Radiation transport of x rays and optical transport of op-
increasesforprogressivelythickerconverterdesigns. ticalphotonsgeneratedinsidethescintillatorweresimulated
Onepossiblemeanstoreducethislossofspatialresolution usingahybridmodelingtechniquewhichdecouplesradiation
is through the use of a rear illumination configuration40,41 and optical transport simulations and condenses the more
in which the array is coupled to the entrance surface and computationally expensive optical simulation part into a
the reflector to the exit surface of the converter, as shown in singleopticalsimulationperconverterdesign.Thetechnique,
Fig.2(b).Inthisconfiguration,forkVimaging,themajority diagrammatically summarized on the left of Fig. 3, entails
of the optical photons generated in the direction of the a sequential process where radiation images (obtained from
MedicalPhysics,Vol.42,No.4,April2015
2075 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2075
The optical transport simulations were performed using
the 4 package.46 The details and validation of the
optical model, as well as the values for the associated
optical parameters (which were obtained from a prototype
segmentedscintillator),appearinRef.24.
BoththeradiationandopticaltransportMonteCarlosimu-
lationswereperformedona64-bitLinuxclusterwith∼1000
processorcores(4.0GHzAMDFXseries).Thestudyrequired
atotalof∼3.04×106CPUhours,withalargemajorityofthe
timespentontheradiationtransportsimulations.
2.B.2. Electronicadditivenoisemodel
F.3. Flowchartillustratingthesimulationframeworkconsistingofhybrid
An analytical noise model was employed to account for
modeling(boxontheleft)andelectronicadditivenoisemodeling(boxonthe
the effect of electronic additive noise. The symbols in the
right)usedinthestudy.
following equations and the values of related parameters
are summarized in Table I. The model takes into account
the two dominant components of additive noise, reset noise
radiationtransportsimulations)arecorrectedwithanoptical
associated with the thermal noise of the TFTs in the array
gain distribution and optical point spread function (PSF)
pixels (σ ) and noise of the external preamplifier
(bothobtainedfromopticaltransportsimulations)toaccount TFT-thermal
electronics (σ ). The TFT reset thermal noise can be
foropticalSwanknoiseandopticalblur,respectively.24 amp
calculatedfrom47
Monte Carlo radiation transport simulations were per-
formed using the EGSnrc code,42 with the geometries of σ = 12k TC (cid:0)e−(cid:1), (1)
the converters and the contrast phantom modeled using the TFT-thermal q B pd
EGSnrcC++classlibrary(egspp)43andwiththeegsppuser whereqistheelectroncharge,k istheBoltzmannConstant,
B
code (as well as the input file that defines the geometry) T is the room temperature in Kelvin, and C is the photo-
pd
modifiedasnecessary.Theparametersettingsandalgorithm diodecapacitanceofeacharraypixel.Followingtheparallel-
options in the code that were employed in the study follow plate capacitor model, the photodiode capacitance can be
those used in previous theoretical investigations.19,24 Both calculatedfrom48
kV and MV simulations employed a point source located
A ηa2
130cmawayfromtheentrancesurfaceoftheconverter.The C =ε ε pd=ε ε pix, (2)
kV spectrum corresponds to that of the 100 kVp “standard pd 0 Si d 0 Si d
head” protocol of a Varian On-Board Imager (OBI) and was where ε is the vacuum permittivity, ε is the relative static
0 Si
described using the TASMIP model,44 with the tube voltage permittivity of silicon, A is the area of the photodiode in
pd
setto100kVandtheintrinsicfiltrationsetto1.45mmalumi- each pixel, η is the optical fill factor of the array pixels,
num. A 6 MV spectrum corresponding to that of a Varian a is the pixel pitch of the array, and d is the thickness of
pix
radiotherapy linear accelerator was adopted for MV simula- the photodiode. The array fill factor is assumed to be 100%
tions.45 [i.e., η=1.0 in Eq. (2)]—given the relatively large pixel
TI. Symbols,definitions,andvaluesofthefixedparametersusedintheelectronicadditivenoisemodel.Note
thatthevalueusedforphotodiodethickness,d,correspondstothatofamodernarraydesign(M13)(Ref.49).
Symbols Definitions Values
σTFT-thermal TFTresetthermalnoise
σamp Preamplifiernoise
σtotal Totaladditivenoise
Cdata Datalinecapacitance 115pF
Cpd Photodiodecapacitance
Apd Photodiodearea
apix Arraypixelpitch
d a-Silayerthicknessinphotodiode 1.50µm
εSi Relativestaticpermittivityofsilicon 12
ε0 Vacuumpermittivity 8.85×10−12F/m
η Arrayopticalfillfactor 1.0
q Electroncharge 1.60×10−19C/e
kB Boltzmannconstant 1.38×10−23m2kg/(s2K)
T Roomtemperature 295K
MedicalPhysics,Vol.42,No.4,April2015
2076 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2076
pitchesconsideredinthestudy.49Thepreamplifiernoisewas profiles, a weighted central slice CT dose index (CTDI )
W
estimatedusinganexpressionbasedonthecharacteristicsof was used as a surrogate for imaging dose for kV and MV
ahighperformancepreamplifier,50 CBCTinordertofacilitatecomparisonsbetweenthekVand
(cid:0) (cid:1) MVCBCTimagedosesusedinthestudy.CTDI isdefined
σ =15C +285 e− . (3) W
amp data as51
Inthisequation,thevalueofthedatalinecapacitance(C )
data
was conservatively estimated using the capacitance per unit CTDI =1D +2D , (5)
lengthofanarraywithalargepixelpitch,47scaledtothedata W 3 C 3 P
linelengthoftypicalMVAMFPIs(i.e.,40cm).
Thetotaladditivenoise,σtotal,wascalculatedusing where the doses at two landmark locations, the center (DC)
and the periphery (D , defined at 1 cm inside the phantom
 (cid:0) (cid:1) P
σ = σ2 +σ2 e− . (4) surface), are summed over all projection angles in the
total TFT-thermal amp
tomographicscan.Basedonthisdefinition,thefluencevalues
Foreachpitchexaminedinthestudy,aGaussiandistribution usedinkVandMVCBCTsimulationscorrespondtoCTDI
W
with zero mean and variance of σtotal was formed. For each valuesof∼0.91and∼3.0cGy,respectively.
pixel of the optically adjusted images obtained from the For the optical transport simulation, the segmented scin-
hybridmodelingtechnique,theadditivenoisewasintroduced tillator in each converter design took the form of a matrix
through random sampling according to that distribution, of 101×101 elements. For each design, 10000 simulation
resulting in additive noise calibrated images, as indicated in runs, each consisting of 10000 optical photon histories,
Fig.3. were performed. All photons were generated in the central
element of the matrix, following the 3D radiation energy
2.C. DeterminationofCBCTimagingmetrics depositionprofileofthecorrespondingconverterdesign.The
optical gain distribution and optical PSF obtained from this
FortheradiationtransportsimulationofCBCTimaging,a simulation were applied to the simulated radiation images
contrast phantom with dimensions, composition, and inserts to yield optically adjusted images, which accounted for the
similar to those of a phantom used in a previous empirical effect of both optical Swank noise and optical blur.24 The
study21 was simulated under both kV and MV imaging optically adjusted images were then corrected for additive
conditions. The phantom consists of an 11.4 cm diameter, noise of the corresponding design to generate the final
solid water cylinder with three 2.8 cm diameter, cylindrical calibratedimages.
tissue-equivalent inserts, all with a common length of 6 cm. AFeldkamp-basedalgorithmemployingarampfilterwas
Consistentwithearlierstudies,21,24thecenterofthephantom used to reconstruct the volumetric images corresponding to
was positioned 124.2 cm from the source—thereby leaving the contrast phantom from a combination of the calibrated
a 1 mm gap between the converter and the bottom of the projectionimagesandtheaverageofthe180calibratedflood
phantom. The inserts correspond to breast, solid water, and images.Thevoxelpitchandsingleslicethicknessusedinthe
brain with electron densities of 0.954, 0.988, and 1.049 reconstructionwerechosentobeequaltotheconverterpitch
relative to water, respectively.24 All converter designs have foreachdesign.Asuitablenumberofthereconstructedslices
a detection area of ∼70×140 mm2, resulting in matrix werebinnedtogenerateaslicethicknessof∼5mm,followed
formatsof141×281,95×189,and71×141fortheexamined by a cupping artifact correction to remove the background
pitches of 0.508, 0.762, and 1.016 mm, respectively. This trendcausedbybeamhardening.21
converter area was chosen so as to be sufficiently large to TheCBCTperformanceofvariousconverterdesignswas
allowimagingofthephantom.Foragivenconverterdesign, characterized in terms of contrast (Contrast), noise (Noise),
imager configuration, and imaging condition, 180 projection and CNR of the tissue-equivalent inserts relative to the
radiation images were obtained by scanning the phantom water-equivalent background in the reconstructed phantom
tomographically at 2◦ angular increment over 360◦. In the images.18 Contrast for a given insert was calculated in
simulations, the x ray fluence incident on the phantom per Hounsfieldunits(HU)usingtheequation
tomographic scan corresponded to 1.14×108 and 4.32×107
xrays/mm2at130cmfromthesourceforkVandMVCBCT,
S −S
respectively. The kV fluence provides a dose equivalent to Contrast= obj water×1000(HU), (6)
S
the 145 mAs standard head protocol for the Varian OBI water
kV CBCT imaging system, while the MV fluence provides
where S and S represent the mean signal in the insert
a dose equivalent to that used in a previous empirical MV obj water
and water-equivalent background, respectively. Each signal
CBCT study (corresponding to a minimum of one beam
wastakenfromacircularregionwithan∼14.2mmdiameter
pulseperprojection).21Inaddition,foreachconverterdesign,
thatexcludedtheedgesoftheinsertsandthephantom.Noise
imager configuration, and imaging condition, a set of 180
wascalculatedusingtheequation
radiation flood images was obtained in the absence of the
phantom, each using the same fluence as that used for the
σ
individual projection images of the phantom. Due to the Noise= obj ×1000(HU), (7)
distinctivelydifferentcharacteristicsofthekVandMVdose Swater
MedicalPhysics,Vol.42,No.4,April2015
2077 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2077
where σ represents the standard deviation of the signal in Figure 4 shows the kV MTF results for the different
obj
theinsert.Finally,CNRwascalculatedfrom imager configurations. As expected, the rear illumination
results, shown in Figs. 4(b) and 4(d), exhibit systematically
S −S higher MTF compared to that for the corresponding front
CNR= obj water. (8)
σ illumination cases shown in Figs. 4(a) and 4(c). This is
obj
a consequence of reduced lateral optical spread for rear
illumination,asschematicallyillustratedinFig.2.Notethat,
2.D. DeterminationofimagerMTF
for all but the rear-mirror configuration, MTF improves
Thespatialresolutionforeachconverterdesignwaschar- with decreasing converter thickness due to reduced scatter
acterizedintermsofthepresampledMTFusingtheangledslit of primary x rays and lateral spread of optical photons.
technique.52 The radiation and optical transport simulations However, for the rear-mirror configuration [Fig. 4(b)], the
of MTF followed the steps described in a previous study,24 MTF generally degrades with decreasing thickness. This
generating an optically adjusted slit image for each design. reversed behavior originates from the increasing fraction of
Additive noise was subsequently included to yield a final optical photons that are reflected by the mirror reflector and
calibrated slit image. This image was used to determine an reach the array as thickness decreases. Such photons tend to
oversampled line spread function, the 1D Fourier transform undergo more interactions with the septal walls, leading to
ofwhichyieldedtheMTF. more lateral spread which, in turn, results in deterioration
of spatial resolution. The absence of such photons for
the rear-black configuration results in systematically higher
2.E. DeterminationofSwankfactorandsignal
MTF compared to that for the rear-mirror configuration. In
addition, for the rear-black configuration, the MTF curves
The steps used to determine Swank factor and signal
are closely clustered and relatively insensitive to changes in
(accounting for contributions of both radiation and optical
effects) largely followed those reported in Ref. 20. (Swank converter thickness, allowing for the possibility of using a
greater thickness to achieve higher detection efficiency in
factor is a metric with values ranging from 0 to 1, where
higher values correspond to lower Swank noise.53) For each MV imaging without substantial degradation of kV spatial
resolution.
imagerconfigurationandconverterdesign,radiationtransport
As is apparent in Fig. 5(a), under kV conditions, the rear
simulation was performed using the EGSnrc code package,
illumination configurations (open symbols) provide better
yielding a phase space file containing information for each
Swankfactorthantheirfrontilluminationcounterparts(solid
energy deposition event within the converter for each inter-
symbols). In addition, for a given converter thickness, rear-
actingxray.Usingthisphasespacefile,simulationofoptical
transportwassubsequentlyperformedusingthe4code blackandrear-mirrorexhibitnearlyidenticalSwankfactors.
The rear illumination configurations also provide generally
package to tally the number of optical photons detected for
higher signal than their front illumination counterparts, as
eachxrayintheformofapulseheightdistribution.TheSwank
factorI andsignalSwerecalculatedusingtheequations53 shown in Fig. 5(b). For all but the rear-black configuration,
signal decreases with thickness due to the increased number
ofabsorbedphotonsalongalongermeanopticalpathlength.
M2
I= 1 , (9) However,fortherear-blackconfiguration,signalisrelatively
M M
0 2 independent of converter thickness—largely as a result of
M
S= 1, (10) the negligible contribution from optical photons generated
M0 in material beyond ∼0.25 cm. It is interesting to note that,
for configurations with the same type (i.e., mirror or black)
where Mi is the ith order moment of the pulse height but different positioning of reflector (i.e., front and rear),
distribution,P(x),obtainedfrom
signalvaluesgenerallyconvergeatsmallerthicknesses—due
 to similarities in the mean optical path lengths for front and
Mi= kxikP(xk). (11) rear illuminations. This similarity results from a relatively
moreuniformenergydeposition,andthereforemoreuniform
generationofopticalphotons,throughouttheconverterthick-
3. RESULTS ness for thinner converters. Conversely, for configurations
with the same reflector positioning but different reflector
3.A. Comparisonofimagerconfigurations
types, signal values converge at larger thicknesses due
A comparison of the relative merits of the four imager to the diminished importance of reflector type, resulting
configurations described in Sec. 2.A was performed through from reduced numbers of photons reaching the reflector.
characterization of their MTF, Swank factor, and signal For example, for the two rear illumination configurations,
performance under kV and MV imaging conditions. Fig- although rear-black provides lower signal values than rear-
ures 4–7 show results for the full range of converter thick- mirror at smaller thicknesses, this signal difference quickly
nesses (0.25–4.0 cm) at a selected pitch of 1.016 mm—the diminishes as thickness increases, resulting in a relative
samepitchasthatofaprototypeBGOsegmentedscintillator differenceofonly∼2%at2cmandanegligibledifferenceat
reportedinapreviousempiricalstudy.21 4cm.Sinceagreaterthicknessisadvantageousforimproved
MedicalPhysics,Vol.42,No.4,April2015
2078 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2078
F.4. MTFresultsunderkVimagingconditionsfor1.016mmpitchimagersemployingvariousconverterthicknessesforthe(a)front-mirror,(b)rear-mirror,
(c)front-black,and(d)rear-blackconfigurations.
quantumefficiencyforMVimaging,theadoptionofablack The results of the above analysis of MTF, Swank fac-
reflector is not expected to significantly constrain signal tor, and signal suggest that the rear-black configuration
comparedtothemirrorreflector. represents the most favorable design choice for a converter
Figure 6 shows MV MTF results for the various imager intended for kV and MV operation. This conclusion applies
configurations. In all cases, MTF performance is seen to throughout the range of converter pitches considered in the
improve with decreasing converter thickness. In addition, study,theresultsforwhich(thoughnotshown)exhibittrends
for a given thickness, the rear-black configuration is seen similarto thoseat1.016 mm.For thatreason,only therear-
to provide equivalent or higher MTF compared with the blackconfigurationisconsideredinSecs.3.B–3.D.
other three configurations. Figure 7 shows that, as in the kV
case, the rear illumination configurations provide generally
3.B. Comparisonofconverterdesigns
better MV Swank factor and signal performance than their
front illumination counterparts. In Fig. 7(b), while signal CNR performance for CBCT imaging under kV and MV
increases toward an asymptotic limit as thickness increases imaging conditions as a function of converter thickness and
forrearillumination,largelyduetoasymptoticallyincreasing pitch is shown in Fig. 8. For a given thickness and imaging
QE, for front illumination signal initially increases and then condition, CNR consistently increases with pitch as a result
decreases due to the competing effects of increasing QE ofreducednoiseduetomorexrayquantabeingdetectedby
and decreasing optical collection efficiency. Note that the largercross-sectionalareasofscintillatorelements.However,
signal results in Fig. 7(b) exhibit the same general trends for a given pitch, the kV and MV behaviors of CNR as a
of convergence observed in the kV case. Considering the function of thickness are distinctly different. In the case of
tworeflectortypesforrearillumination,althoughrear-black kV and for a given pitch, CNR is relatively insensitive to
provides lower Swank factor and signal than rear-mirror, changesinthickness[asseeninFig.8(a)].Thisinsensitivity
the differences are small for thick converters, with relative is a consequence of the roughly constant level of QE under
differencesofonly∼4%and8%forSwankfactorandsignal, kV imaging conditions (ranging from ∼94% to 99% for the
respectively,atathicknessof2cm,andnegligibledifferences thicknesses examined) as well as of the relative indepen-
at4cm. dence of MTF and Swank factor to changes in thickness (as
MedicalPhysics,Vol.42,No.4,April2015
2079 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2079
F.5. Kilovoltageresultsfor(a)Swankfactorand(b)signal(definedastheaveragenumberofdetectedopticalphotonsperinteractingxray)asafunctionof
converterthicknessfor1.016mmpitchimagers.
describedinSec.3.A).Bycomparison,inthecaseofMVand increasesby∼32%from1to2cm,theincreaseisonly∼9%
for a given pitch, the value of CNR increases with thickness from2to4cm.
[as seen in Fig. 8(b)] due to the significant increase in QE It is clear that the identification of a single segmented
from ∼15% to 80% for the thicknesses examined. Note that scintillatorconverterdesignwhichexhibitsperformancethat
the increase in CNR exhibits an asymptotic behavior due to fulfills the needs of both kV and MV imaging is hindered
diminishing improvement in QE from increased converter by an inherent incompatibility. Specifically, MV imaging
thickness. For example, at a pitch of 0.508 mm, while CNR favors thicker converter designs with larger pitch so as to
F.6. MTFresultsunderMVimagingconditionsfor1.016mmpitchimagersemployingvariousconverterthicknessesforthe(a)front-mirror,(b)rear-mirror,
(c)front-black,and(d)rear-blackconfigurations.
MedicalPhysics,Vol.42,No.4,April2015
2080 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2080
F.7. MVresultsfor(a)Swankfactorand(b)signalasafunctionofconverterthicknessfor1.016mmpitchimagers.
achieve higher CNR performance,24 whereas kV imaging no binning), the corresponding sampling pitches are 0.508,
favors smaller pitch to maintain higher MTF performance, 0.762,and1.016mm—whicharedenotedas508 ,762 ,
1×1 1×1
while being relatively insensitive to the choice of thickness. and1016 ,respectively.Forapitchof0.508mmwith2×2
1×1
For that reason, a converter thickness of ∼2 cm would be binning, the corresponding sampling pitch is 1.016 mm and
a favorable choice given the rapidly diminishing returns on is denoted as 508 . Values for TFT reset thermal noise,
2×2
MVCNRbeyondthisthickness,coupledwithconsiderations preamplifier noise, and total additive noise used in the CNR
of increasing material cost and difficulty of manufacture calculations for the different pitch-binning combinations are
for thicker scintillators. Moreover, within the bounds of the summarizedinTableII.
current study, the preferred converter pitches for kV and Figure 9 shows the MV CNR performance for designs
MV imaging are 0.508 and 1.016 mm, respectively. This incorporating the various pitch-binning combinations. For
difference could be reconciled either through selection of the three combinations with no binning, CNR increases
a common intermediate pitch (e.g., 0.762 mm) or through with converter pitch for a given thickness and increases
selectionofthesmallerpitchincombinationwiththeuseof with converter thickness for a given pitch—in both cases
binning for MV (but not for kV) operation, as explored in due to increased numbers of detected x ray quanta per
Sec.3.C. scintillator element. For all converter thicknesses, 508
2×2
exhibits systematically lower (∼15%–21%) CNR compared
to that of 1016 , despite having identical sampling pitch.
3.C. Comparisonofpitch-binningcombinations 1×1
Thisisprimarilyaconsequenceoftwocontributingfactors—
Inthissection,additivenoise,MVCNR,andMTFperfor- increased Swank noise due to a larger aspect ratio for
manceareexaminedforthevariouscombinationsofconverter the scintillator elements20 and reduced QE due to the
pitches and binning modes. For pitches of 0.508, 0.762, and displacement of BGO crystal by more septal wall material.
1.016mmwith1×1binning(i.e.,fullresolutionreadoutwith However,508 exhibitssystematicallyhigher(∼21%–31%)
2×2
F.8. AbsolutevalueofCNRasafunctionofconverterthicknessandpitchforaphantominsertwitharelativeelectrondensityof0.954.Theresultswere
obtainedusing imagerdesignswiththe rear-black configurationandpitches rangingfrom0.508to 1.016mmunder (a)kVconditionswitha CTDIW of
∼0.91cGyand(b)MVconditionswithaCTDIWof∼3.0cGy.
MedicalPhysics,Vol.42,No.4,April2015
2081 Liuetal.:Dualenergy(kVandMV)radiotherapyimager 2081
TII. EstimatesoftheTFTresetthermalnoiseσTFT-thermal,preamplifiernoiseσamp,andthetotaladditive
noiseσtotalofthevariouspitch-binningcombinationscalculatedusingEqs.(1),(3),and(4).Notethatvaluesof
σampfor5082×2assumeapostacquisition(i.e.,purelydigital)binning.
Pitch-binningcombination 5081×1 7621×1 10161×1 5082×2
Converterpitch(mm) 0.508 0.762 1.016 0.508
Samplingpitch(mm) 0.508 0.762 1.016 1.016
σTFT-thermal(e−[rms]) 2408 3611 4816 4816
σamp(e−[rms]) 2010 2010 2010 4020
σtotal(e−[rms]) 3137 4133 5219 6273
CNR compared to 762 —largely as a result of increased (i.e.,508 ).AnexampleoftheCBCTperformanceofsuch
1×1 2×2
numbersofsampledxrayquanta. animagerisprovidedinSec.3.D.
Figure10showstheMVMTFperformancefor2cmthick
converters corresponding to various pitch-binning combina-
3.D. Performanceoftheproposeddualenergyimager
tions. As a result of binning, the MTF for 508 falls off
2×2
more rapidly with spatial frequency compared to the MTF ReconstructedCBCTimagesofthecontrastphantomob-
for 508 , resulting in a relative difference of ∼25% at tained from imagers incorporating various converter designs
1×1
a frequency of 0.49 mm−1 (the Nyquist frequency for a areshowninFig.11.Figures11(a)and11(b)showsimulated
sampling pitch of 1.016 mm). However, despite having the MVandkVCBCTimagesacquiredfromtheproposedBGO-
same sampling pitch, the MTF for 508 is systematically based dual energy imager (referred to as the BGO imager)
2×2
higher than that for 1016 almost up to the Nyquist corresponding to pitch-binning combinations of 508 and
1×1 2×2
frequency, due to more constrained optical lateral spread 508 ,respectively.ComparedwiththeMVimage(obtained
1×1
as a result of more septal wall material in the path of at ∼3.0 cGy), the kV image exhibits superior soft-tissue
optical photons for the former combination. Note that visualization by virtue of better CNR performance, despite
762 and 508 demonstrate somewhat comparable MTF, the use of a much smaller dose of ∼0.91 cGy—a result of
1×1 2×2
with the former and latter combinations exhibiting slightly highercontrastandlowernoiseunderkVconditions.Itisof
better performance at spatial frequencies above and below interest to note that, for the insert in the lower left of each
∼0.26mm−1,respectively. image, the gray scale relative to the background is reversed
In light of the various findings reported above, and un- betweenthekVandMVimages.Thisislikelyaresultofthe
der the assumptions of the current study, our results sug- specific chemical composition of that insert, combined with
gest that the most advantageous design for a dual energy the difference in the dominant x ray interaction mechanisms
imager based on BGO would incorporate a converter with a and their behaviors under kV and MV imaging conditions:
0.508mmpitchand2cmthickness,operatedusingtherear- the photoelectric effect under kV conditions, which scales
illumination configuration and coupled to a black reflector. with the fourth power of electron density; and the Compton
TheimagershouldbeoperatedatfullresolutionforkVimag- effect under MV conditions, which scales linearly with
ing (i.e., 508 ) and 2×2 binning mode for MV imaging electron density. For purposes of comparison, Fig. 11(c)
1×1
F.9. AbsolutevalueofCNRunderMVimagingconditionsasafunction
ofconverterthicknessforaphantominsertwitharelativeelectrondensityof
0.954.Theseresultswereobtainedusingconverterscorrespondingtovarious F.10. MVMTFresultsfor2cmthickconverterscorrespondingtovarious
pitch-binningcombinations. pitch-binningcombinations.
MedicalPhysics,Vol.42,No.4,April2015