Compensators for dose and scatter managementin cone-beam computed tomography
S. A. GrahamOntario Cancer Institute, Princess Margaret Hospital, Toronto, Ontario, M5G 2M9, Canadaand Department of Medical Biophysics, University of Toronto, Ontario, M5G 2M9, Canada
D. J. MoseleyRadiation Medicine Program, Princess Margaret Hospital, Toronto, Ontario, M5G 2M9, Canadaand Department of Radiation Oncology, University of Toronto, Toronto, Ontario, M5G 2M9, Canada
J. H. SiewerdsenOntario Cancer Institute, Princess Margaret Hospital, Toronto, Ontario, M5G 2M9, Canadaand Department of Medical Biophysics, and Department of Radiation Oncology, University of Toronto,Toronto, Ontario, M5G 2M9, Canada
D. A. JaffrayRadiation Medicine Program, Princess Margaret Hospital, Toronto, Ontario, M5G 2M9, Canada,Department of Radiation Oncology, and Department of Medical Biophysics,University of Toronto, Toronto, Ontario, M5G 2M9, Canada;and Ontario Cancer Institute, Princess Margaret Hospital, Toronto, Ontario, M5G 2M9, Canada
�Received 22 May 2006; revised 19 April 2007; accepted for publication 20 April 2007;published 11 June 2007�
Standard practice in external beam radiation therapy relieson imaging techniques that lack the soft-tissue detectabilitynecessary for precisely and accurately locating soft-tissuetargets. Treatment setup relies on a combination of skinmarks and/or megavoltage radiographs �portal images� thatcan guide treatment based largely on the location of bonyanatomy. Image guidance is an important development formore precise treatment delivery, with numerous techniquesdeveloped to accomplish visualization of soft-tissue struc-tures of interest with the patient in treatment position.1 Arecently developed technology for soft-tissue visualizationand image-guided radiation therapy is kV cone-beam CT�CBCT�.2
CBCT systems employ conventional x-ray tubes andx-ray detectors such as image intensifier systems3,4 or flat-panel detectors.2,5 In this work, recently developed flat-paneldetectors �FPDs� are used to generate two-dimensional radio-graphic projection data that can be processed to form a volu-metric reconstruction with sub-millimeter spatial resolution.
FPDs provide digital images read out at frame rates up to 30
2691 Med. Phys. 34 „7…, July 2007 0094-2405/2007/34„7…/2
frames per second, providing sufficient numbers of projec-tions for reconstruction with a single rotation of the source-detector pair. The use of CBCT for image guidance is gain-ing widespread popularity with a number of manufacturersoffering CBCT imaging platforms integrated with linear ac-celerators. This allows radiography, fluoroscopy, and CBCTimages to be acquired in the treatment position. Although thelarge area of FPDs is advantageous for CBCT acquisitions ofclinically relevant fields of view, the FPDs currently em-ployed in CBCT systems suffer from limited dynamic range.To achieve reasonable signal levels through the midline ofmany patients it is necessary to deliver a high exposure perprojection. It is often the case that the techniques used over-whelm the signal range of the detector at the periphery of thepatient, leading to a loss of information in projections andartifacts in reconstruction due to the truncation of anatomy.
Another issue arising in CBCT imaging is the large quan-tity of scattered radiation generated in the patient and reach-ing the detector, to be hereafter referred to simply as scatter,due to the large projection field sizes ��25�25 cm2� used in
6
patient imaging. In some imaging geometries, the scatter
2692 Graham et al.: Compensators for cone-beam CT 2692
fluence exceeds the primary fluence at the detector. Withsuch a large fraction of the detected fluence arising fromscatter, there can be significant artifacts in the resulting re-constructions. The presence of scatter reduces contrast andalso contributes additional x-ray quantum noise and induceslocalized artifacts such as streaking and reduced attenuationestimation at the center of the object �known as a cuppingartifact�.6 Scattering within the patient also contributes addi-tional dose to the patient, which does not necessarily contrib-ute to corresponding improvements in image quality.
The harmful effects of scatter in cone-beam CT imagingand the rapid deployment of these systems make this an im-portant area of research. The development of robust correc-tion schemes7–11 for removal of scatter in projections is anactive area of research. In addition to postprocessing algo-rithms applied to data to reduce scatter effects it is also pos-sible to use more mechanical methods to reduce scatter. Thiscan be, for example, in the form of larger air gaps betweenthe patient and detector,12–15 or by implementing scatter re-jecting grids.16–18 Each of these approaches has limitations.The use of a grid is known to reduce the primary fluence atthe detector, despite having delivered the dose to the patient,whereas the postprocessing methods are limited to artifactreduction and they cannot address the additional x-ray quan-tum noise induced by the scatter fluence.
While the scatter correction and rejection methods men-tioned above may be capable of reducing the effects of scat-ter in the reconstructed volumes, it would be beneficial to usescatter rejection techniques, which also reduce the dose tothe patient. The use of a smaller longitudinal field-of-view�FOVz� by reducing the collimator opening during imagingwould reduce scattered radiation, but would also limit thetotal volume that could be reconstructed.
The implementation of compensating filters19 �a subset ofwhich is known as bow-tie filters� in CBCT offers an alter-native method of addressing the issue of scatter. The scat-tered radiation in CBCT has been shown in a qualitativefashion to be reduced when using compensators.20,21 Thesecompensating filters can also provide a method of both re-ducing scattered dose within the patient and relaxing the dy-namic range requirements for the FPD.
Compensators have existed in some form since it wasrecognized that the detectability in film radiographs wouldbe improved if a filter were used to deliver a more uniformfluence at the film.22 Filters could be designed to account formore subtle changes in patient anatomy to further increaseimage quality.23 Furthermore, the use of appropriate filters inradiography has been shown to reduce patient dose.24 Com-pensation schemes were implemented early on in computedtomography scanning, with the EMI scanner using a waterbag compensator.25,26 CT scanners later used metal “dodg-ers” made of low atomic number material to closely mimicwater-based compensation.27
The purpose of compensation in CT is both to accommo-date the dynamic range of the detectors and preferentiallyharden the x-ray beam.19–21,27 Patient dose is also reduced
21,28–30
when introducing a compensating filter. Despite scat-
Medical Physics, Vol. 34, No. 7, July 2007
ter not playing as large a role in conventional CT imaging asin CBCT imaging, scatter induced artifacts have been re-ported as being reduced by use of a bow-tie filter.31
For compensator use in multi-row CT or CBCT it is nec-essary to manufacture 2D shaped filters32 to modulate thex-ray fluence across the imager. Filters of this type can beexpected to provide similar benefits as bow-tie filters providein conventional CT, though they may potentially play alarger role in scatter and dose reduction when applied toCBCT imaging.
Although qualitatively it is recognized that compensatorsoffer benefits to CBCT, it is necessary to quantify the mag-nitude of effect obtained under realistic conditions. The in-vestigations are performed assuming a cone-beam CT imag-ing geometry consistent with current radiotherapy image-guidance systems. These investigations will providequantitative evidence of the value of using compensators forthe purpose of improving cone-beam CT image quality, re-ducing the magnitude of x-ray scatter at the detector, andreducing patient dose.
II. MATERIALS AND METHODS
A. Experimental cone-beam CT bench
Investigations of compensating filters for CBCT were per-formed on a bench-top CBCT system �Fig. 1�. The x-raysource was a Rad-94 rotating anode x-ray tube �Varian Medi-cal Systems� in a Varian Sapphire housing with a maximumpotential of 150 kVp, 14° tungsten rhenium molybdenumgraphite target, and 0.4 and 0.8 mm focal spot sizes. Thegeometry of the system was configured to have a source-to-isocenter distance �dSA� of approximately 100 cm, and asource-to-detector distance �dSD� of approximately 155 cm,
FIG. 1. Schematic of the bench-top cone-beam CT system used for investi-gations into the dose and scatter reduction capabilities of compensatingfilters. Parameters that were varied in order to alter the scatter and doseincluded the compensator, z collimation, phantom size, and the air gap be-tween the phantom and detector. The geometry of the system is also sum-marized in Table I.
which is consistent with the geometry of clinical CBCT sys-
2693 Graham et al.: Compensators for cone-beam CT 2693
tems used for image guidance during radiation therapy. Ob-ject rotation during CBCT acquisition was performed with adirect-drive rotation stage �Dynaserv DM1060B�. The detec-tor used was a 14 bit indirect-detection flat-panel detector�PaxScan 4030A, Varian Medical Systems� with a 600 �mthick CsI:Tl scintillating layer. The pixel array was a 2048�1536 �40�30 cm2� amorphous silicon flat panel matrixwith a pixel pitch of 194 �m. This detector provides a25.6�19.2 cm2 maximum FOV at isocenter when using acentered detector. Table I summarizes the relevant param-eters of the bench-top CBCT system.
B. Compensating filters
There are numerous possible designs for compensatorsimplemented on CBCT imaging systems. Compensatorscould potentially be optimized based on a variety of imagingtasks while taking account of the tradeoffs between imagequality and dose to the patient. Patient size, scatter, dose, andexposure dependent detective quantum efficiency �DQE� ofthe detector are examples of quantities that could be consid-ered in the design of modulating filters for cone-beam CT.Simple representative compensators were used in this workto examine the dose and scatter effects of modulating theinput fluence. Although compensators could be manufacturedto have variable modulation across the two-dimensionalcone-beam projection images, the compensators used in thisstudy have constant modulation in the z direction.
The two bow-tie filters constructed for this work werebuilt based on the objective of achieving uniform fluence
TABLE I. Nominal imaging geometry and acquisition/reconstruction param-eters.
Imaging system geometryFocal spot to isocenter distance �dSA� 100 cmNominal focal spot to detector distance �dSD� 155 cmNominal isocenter to detector distance �dAD� 55 cmFocal spot to compensator distance �dcomp� 9.7 cmAxial field of view �FOVx� 25.6 cmMaximum longitudinal field of view �FOVz� 19.2 cmMinimum longitudinal field of view �FOVz� 2 cm
Image acquisitionTube potential 120 kVpTube current 100 mAExposure time
Head phantom 10 msBody phantom 20 ms20.5 cm diameter watercylinder 4 ms
Frame rate 1 s/projectionNumber of exposures 320Nominal beam filtration 4 mm Al, 0.13 mm Cu
Image processing and reconstructionDefect pixel removal 3�3 median filterReconstruction filter HammingMaximum reconstruction volume 25.6�25.6�19.2 cm3
Voxel dimension 0.05�0.05�0.05 cm3
through a cylindrical head phantom with a diameter of 16 cm
Medical Physics, Vol. 34, No. 7, July 2007
measured at 120 kVp. The shape of the filters �Fig. 2� wasthen smoothed using an unweighted sliding average to re-move discontinuous first derivatives that are difficult to ma-chine and that could induce artifacts under conditions of im-perfect geometric calibration. The compensators weremanufactured from 99.9% pure copper plates, as this mate-rial made it possible to machine to the desired shapes andwould allow thin compensators that could be placed betweenthe x-ray tube and collimator assembly at approximately10 cm from the focal spot. At this position it is expected thatvery few x rays scattered from the compensators will reachthe detector because of the large �145 cm� air gap. The bow-tie filters were manufactured from 125�110 mm2 sheets ofCu approximately 2.4 and 1.6 mm in thickness. The plateswere machined to the desired shape down to a copper thick-ness of approximately 0.1 mm at the center �central axis ofthe beam�. The compensators were mounted on 2.5 mm thickacrylic sheets for mechanical support. The chosen thick-nesses and profiles provide compensators with modulationfactors of approximately 8:1 and 4:1, where the modulationfactor was defined as the ratio between the measured detectorsignal at the center of a projection, through the thinnest areaof the compensators, to that at the periphery of a projection�in plateau of compensator profile� where the beam is attenu-ated by 0.1, 1.6, or 2.4 mm of copper for the 1:1, 4:1, and 8:1compensators, respectively. The modulation factor was de-fined only in terms of the measured detector signal in floodfield images where the compensators are in place but noobject is present in the beam. The added filtration used on theCBCT system is commonly 4 mm of aluminum and0.13 mm of copper �i.e., no compensator, referred to here as
FIG. 2. Thickness profile designed for the 8:1 and 4:1 modulation factorcompensators, shown with machining tolerances for the 8:1 filter. Measure-ments of the compensator thickness performed after the completion of ma-chining, shown here only for the 8:1 compensator, indicate that the 4:1compensator corresponded well with the designed thickness of 0.1 mm,while the 8:1 compensator was measured to be 0.14±0.03 mm at the center.This discrepancy in thickness at the center was corrected for in the measure-ments shown in this work. The width of the compensators continues as a flatfunction out to ±62.5 mm, but the axes of this figure have been adjusted toemphasize the shape of the machined section of the copper plates.
the 1:1 compensator�. When placing the 4:1 or 8:1 compen-
2694 Graham et al.: Compensators for cone-beam CT 2694
sator in the beam, the 0.13 mm copper filter was removed toobtain a primary fluence at the center of the beam that isnearly equivalent in all cases.
C. Scatter-to-primary ratio measurements
The SPR at the center of the detector was measured forthe 8:1, 4:1, and 1:1 compensators. The SPR is defined as
SPR =S
P, �1�
where S is the energy integrated signal of the scattered ra-diation measured as the average of a 100�20 pixel2 area onthe FPD, and P is the signal due to the primary radiation.The SPR was measured at the location of the central axis inthe projection images for both an acrylic 16 cm cylindricalhead phantom and the 16 cm head phantom placed inside ofan acrylic annulus with an outside diameter of 32 cm.Acrylic has a Compton scattering contribution to the massattenuation coefficient of approximately 2% to 3% higherthan that of tissue �water� in the relevant energy range �forexample, acrylic has values of 0.173 cm2 g−1 at 60 keV and0.158 cm2 g−1 at 100 keV, while water has values of 0.177and 0.163 cm2 g−1 at the same energies�,33 and we believethis is sufficiently close to demonstrate the effect the com-pensators would have on scatter in patient imaging. The pro-jection images used in the measurement of the SPR were anaveraged set of 30 in-air exposures �flood field images� with
dark fields subtracted. Projections were acquired at 120 kVp
Medical Physics, Vol. 34, No. 7, July 2007
with 1 mAs/projection for the head phantom and2 mAs/projection for the body phantom.
The determination of the value of SPR in the images wasaccomplished by varying the longitudinal field of view�FOVz� from approximately 18.5 cm at isocenter down to theminimum field of view achievable with the collimator as-sembly, which is approximately 2 cm �±0.05 cm�. The axialfield of view �FOVx� was fixed at the maximum detectable�25.6 cm�. The total signal in the projections was assumed tobe the sum of the primary and scattered radiation signals. Byplotting the total signal measured in images against FOVz, acurve could be found which was employed to extrapolate thetotal signal to a FOVz of zero. For the 16 cm phantom thedata were fit to a square root function, while the 32 cm datawere found to be more linear with field size. At the point ofzero FOVz the signal was attributed completely to the pri-mary fluence and was assumed to be constant in all of theprojections. Subtracting the primary signal estimate in allprojections permitted the determination of the scatter fromthe phantom at the center of the projection images. Addi-tional measurements of the SPR were performed with a vari-able air gap �dAG� between the phantom and detector, whilemaintaining the FOVz at 18.5 cm to compare the effects ofthe modulating filters with alternative imaging geometries.
A secondary check of some of the SPR measurements wasalso performed using a beam-blocker method.34 This wasdone in order to compare the SPR values obtained by vary-
FIG. 3. �a� SPR and �b� the corre-sponding SRF value at the center ofthe projection in a 16 cm diameteracrylic phantom �head� and �c� SPRand �d� SRF for a 32 cm diameteracrylic phantom �body� for varyinglongitudinal FOV and the three com-pensators employed in this study. Thevalue of dAD was held at 55 cm for allscans �giving a dAG of 39 cm for thebody phantom and 47 cm for the headphantom�. Measurements are alsoshown for the beam-blocker methodfor the 1:1 �open circle� and 8:1 �opensquare� compensators.
ing the longitudinal FOV with a second method of determin-
2695 Graham et al.: Compensators for cone-beam CT 2695
ing SPR. The lead blocker had approximately an 8 mm widthand 1 cm thickness. Measurements were made at a singleFOVz of 18 cm using the 16 and 32 cm phantoms and the1:1 and 8:1 compensators. A correction accounting for shad-owing of part of the phantom by the blocker was accom-plished by performing measurements with blockers of vary-ing widths �5 mm up to 14 mm� and extrapolating to ablocker size of zero.
In order to quantify the benefits offered by the compen-sating filters, the scatter reduction factor �SRF� was definedas the ratio �for a given phantom, air gap, and FOVz� of theSPR measured with a compensating filter divided by the SPRcorresponding to a flat filter �i.e., 1:1 case�. The SRF showsthe magnitude of scatter reduction offered by the compensa-tors as well as the trends in scatter reduction when alteringFOVz and air gap. An SRF of one would indicate no changein scatter magnitude, less than one would indicate scatterremoval, and greater than one would signify elevated scatter.
D. Dose measurements
Dose was measured in the acrylic 16 cm head and 32 cmbody phantoms exposed at 120 kVp with 1 mAs/projectionfor the head phantom and 2 mAs/projection for the bodyphantom. The total dose was found based on an acquisition
of 320 projections across 360°. Measurements were per-
Medical Physics, Vol. 34, No. 7, July 2007
formed with a 0.6 cc Farmer ionization chamber �NE 2571,S/N 1700� and a Keithley 35040 electrometer �S/N 62968�with a 300 V bias. The charge integrated by the electrometerwas converted to dose using an air kerma calibration factor�Nk� for the specific ionization chamber/electrometer pair de-termined from a standard ionization chamber/electrometerpair calibrated by the National Research Council �Ottawa,Ontario, Canada�. The Nk corresponding to 120 kVp was420 mGy/10−8 C. The dose was measured for variations incompensator modulation, longitudinal field of view �FOVz�,and position of the dosimeter in the phantom. The dose re-duction factor �DRF� was defined similarly to the SRF andwas used to quantify the dose reductions achieved with thecompensators in place.
Slight differences in filter thickness along the central axisfor the three choices of compensator were expected to play asmall role in creating the differences in measured dose. Mea-surements of the copper thickness after machining was com-plete indicated that the 4:1 compensator was 0.10±0.03 mmat the center, while the 8:1 compensator was 0.14±0.03 mmat the center. To provide the most accurate comparison of the1:1, 4:1, and 8:1 compensators the differences caused by thediffering primary fluence along the central axis were cor-rected out of the measured doses. This was performed by
FIG. 4. Variation in SPR for differentspacing between the phantoms andflat-panel detector. The �a� SPR and�b� SRF results are shown for the headphantom along with the SPR and SRFshown respectively in �c� and �d� forthe body phantom.
measuring the detector signal under narrow beam geometry
2696 Graham et al.: Compensators for cone-beam CT 2696
�2�2 cm2 field at isocenter� with each of the compensatorsin place and no object in the beam. Ideally, the measuredsignal through each of the compensators would be identical.The measured signal through the 1:1 filter was compared tothe signal through the center of the 4:1 and 8:1 compensatorsand deviations from the 1:1 signal were used to correct themeasured doses. These measurements showed that the trans-mission through the center of the 4:1 compensator matched�within 2%� that of the 1:1 filter. The transmission throughthe 8:1 filter was found to be approximately 7% lower than
FIG. 5. SPR measured as a function of modulation factor for the 16 cm�head� and 32 cm �body� phantoms.
Medical Physics, Vol. 34, No. 7, July 2007
that of the 1:1 filter. Appropriate adjustments were made tothe doses measured with the 8:1 filter to account for thissystematic discrepancy.
E. Reconstructed images
The influence of the compensators on CBCT image qual-ity was assessed using a simple cylindrical water phantomfor examination of cupping artifacts and noise. The acrylicwall of the phantom had an outside diameter of 20.5 cm anda thickness of 0.5 cm. The height of the phantom was ap-proximately 28 cm. Projection images were acquired forFOVz values of 2 and 18.5 cm at 120 kVp and0.4 mAs/projection with 320 projections being acquiredthrough 360°. The raw projection data acquired at 2048 pix-els by 1536 pixels were down-sampled to 512�384, gainand offset corrected using floods taken with the proper com-pensator in place, and median filtered to remove defectivepixels from the images. Volumes of 25.6�25.6�1.3 cm3
�0.05�0.05�0.05 cm3 voxels� were reconstructed using theFeldkamp algorithm35 for each case �FOVz=2 cm, 18.5 cm;compensation: 1:1, 4:1, 8:1� so that comparisons could bemade in terms of apparent cupping, noise, and anycompensator-induced artifacts. A summary of the reconstruc-tion parameters is given in Table I.
Images of the phantom were also performed with twoliver-equivalent tissue inserts �GAMMEX rmi, �e
w=1.07�suspended in the phantom for the purpose of contrast-to-noise ratio �CNR� measurements. The CNR was defined as
FIG. 6. Dose measured in the 16 cmacrylic head phantom at �a� the centerand �c� the periphery of the phantom.The compensators reduce the dose inboth locations, as demonstrated by theDRF for the �b� center and �d� periph-eral doses.
2697 Graham et al.: Compensators for cone-beam CT 2697
the difference between the mean attenuation value in the in-sert and in the surrounding water �both measured in a 100voxel ROI�, divided by the standard deviation of the attenu-ation value in the water. The images acquired for the CNRmeasurements were measured at 4 mAs/projection, and anadditional 3 mm of copper filtration added to the beam. Thiswas necessary to reduce beam-hardening artifacts present inimages. All other parameters for the CNR measurementswere the same as those given above.
III. RESULTS
A. Scatter-to-primary ratio measurements
The SPR at the center of each phantom was found todecrease as the longitudinal field of view was reduced, andthe modulation factor was increased �Fig. 3�. The SPR in thehead phantom reached a value greater than 100% for thehighest FOVz for the 1:1 modulation factor filter. With the4:1 and 8:1 compensators the magnitude of the scatter signaldoes not reach the same level as the primary signal for eventhe highest field of view. The SRF showed a constant 20%reduction in scatter for all FOVs when applying the 4:1 com-pensator, and a greater than 40% reduction for the 8:1 com-pensator. Beam-blocker measurements agreed well with theextrapolation measurements giving SPR values of 1.02±0.03for the 1:1 compensator and 0.54±0.03 for the 8:1 compen-sator at a FOVz of 18 cm, compared with 1.01±0.03 and0.60±0.02 for the 1:1 and 8:1 compensators, respectively, at
18.5 cm using the extrapolation method.
Medical Physics, Vol. 34, No. 7, July 2007
For the body phantom similar effects were noticed,though the levels of scatter were much higher. The scattersignal equalled that of the primary signal when applying flatfiltration even at relatively small FOVz with an SPR of 100%at a longitudinal field of view of approximately 3.7 cm. Thesame SPR when using the 4:1 and 8:1 compensators could beachieved with a FOVz of 6.5 and 9.8 cm, respectively, dem-onstrating the large gains that can be made through theimplementation of compensation schemes in cone-beam CT.The reduction of scatter was more pronounced for the bodyphantom than the head phantom, with SRF values of ap-proximately 0.6 and 0.4 for the 4:1 and 8:1 compensators,respectively. Beam-blocker measurements gave values of4.4±0.4 and 2.1±0.2 for the 1:1 and 8:1 compensators.
With FOVz fixed, compensating filters reduced the SPRregardless of the air gap �Fig. 4� for the range of geometriesexamined. Although the primary signal in each pixel de-creased as the detector moved away from the phantom andx-ray tube, the scatter fell off more quickly, resulting in adecreasing SPR. This is consistent with the well-documentedrole of air gaps in reducing scatter at the detector.13,15 Whilethe smallest air gap caused a large value of SPR when usinga flat filter, the application of the 8:1 compensator reducedthe SPR such that the SPR at the smallest air gap for the 8:1compensator was less than that at the highest air gap for the1:1 case for both the head and body phantoms. SRF valuesremained close to the values found at the nominal distance
FIG. 7. Dose measured in the 32 cmacrylic body phantom at �a� the centerof the phantom and �c� the peripheryof the phantom, shown with the DRFfor �b� the center of the phantom and�d� the periphery of the phantom.
between the phantom and the detector �dAD=55 cm�. There
2698 Graham et al.: Compensators for cone-beam CT 2698
was a slight decrease in SRF as the air gap was increased,though the slope was not significantly different from zero.
Figure 5 summarizes the SPR as a function of the modu-lation factor. The SPR results for an 18.5 cm longitudinalfield of view at isocenter and a dSD of 155 cm at the center ofthe two phantoms clearly show a substantial decrease in SPRas the modulation factor of the compensator increases. Theeffect is larger in the body phantom where the SPR is muchgreater.
B. Dose measurements
The dose at the center and periphery of the 16 cm acrylichead phantom was shown to vary with FOVz and compensa-tion �Fig. 6�. At the center of the phantom the primary inten-sity of the beam was approximately constant, so the reduc-tion in dose when increasing the compensator modulationwas due to the reduction in scatter. The dose at the center ofthe phantom for the 8:1 compensator was approximately75% of the dose measured at the center of the phantom whenusing a flat filter. The compensators further reduced the doseat the periphery of the phantom, since both the primary andscatter were reduced. When using the 8:1 compensator in-stead of the flat filter the dose was reduced to approximately55% of the original value at the periphery. The reduction indose measured at the periphery of the phantom when em-ploying compensating filters came largely from the reductionin primary x rays. The DRF appears constant across all
fields-of-view for both positions in the phantom, except for a
Medical Physics, Vol. 34, No. 7, July 2007
suspicious drop in the DRF when the dose was measuredwith the lowest FOVz at the periphery of the phantom.
Similar trends in dose reduction were also seen whenmeasuring dose in the body phantom while varying the com-pensation scheme and FOVz �Fig. 7�. The dose decreased toabout 60% and 40% of the value measured with the 1:1modulation filter at the center and periphery, respectively,when applying the 8:1 filter. There are large reductions indose at the periphery of the phantom because the compensa-tors used in this study were designed according to the fluencethrough the 16 cm acrylic head phantom and could not beplaced closer to the x-ray tube to adjust the magnification ofthe compensator shape. Thus the fluence pattern applied tothe body phantom reduced the primary fluence to a largerdegree than a filter that was designed to accommodate theexact shape of this larger phantom.
Despite this large reduction in dose to the periphery of thephantom, it was seen that, unlike the case of the head phan-tom, the dose at the periphery of the phantom for the 8:1compensator was larger than the dose at the center. The doseas a function of distance from the center of the phantom forthe different compensators was measured in the head andbody phantoms �Fig. 8�. In the head phantom the dose in-creased towards the outside of the phantom in the 1:1 case,was nearly constant for the 4:1 case, and was reduced for the8:1 case. In the body phantom, the dose increased whenmoving away from the center of the phantom up to a point
FIG. 8. Dose as a function of the dis-tance from the center of the phantomin both the �a� head and �c� body phan-tom. The DRF is shown for the �b�head and �d� body phantoms.
where the dose began to drop. This drop is attributed to the
2699 Graham et al.: Compensators for cone-beam CT 2699
edge of the phantom moving outside the field-of-view at ap-proximately 12.8 cm from the center of the phantom. Thepeak dose was found at different distances from the center ofthe phantom because of the changes in dose due to scatter.The DRF shows, as expected, that the dose reducing capa-bilities of the compensators increased when moving awayfrom the center of the phantom—this is a combination of thereduced primary and the reduction in primary-induced scatterfluence within the phantom.
Figure 9 shows the achievable decrease in dose at thecenter and periphery of the head and body phantoms with thecompensators used in this study. Similar trends were seen inboth phantoms when varying the modulation of the compen-sator. The decrease in dose measured at the center of thephantom was due only to the decreased scatter dose arisingfrom the modified primary fluence pattern offered by thecompensators. The decrease in dose shown at the peripheryof the phantoms was due to a combination of the decrease inscatter, along with the decrease in primary fluence caused by
FIG. 9. For both the head and body phantoms the dose is shown to vary withvarying modulation factors. The dose shown in �a� is measured at the centerof the phantoms where the primary intensity of the beam is approximatelythe same for all modulations. In �b� the dose is shown at a depth of 1 cm inthe two phantoms where both the scatter and primary fluences are altered bythe compensators during a CBCT scan.
the shape of the filter.
Medical Physics, Vol. 34, No. 7, July 2007
C. Image reconstructions
Reconstructions were performed on projection images ofa cylindrical water phantom for a FOVz of 2 cm �low scatter,Fig. 10� and 18.5 cm �high scatter, Fig. 11�. In the low scat-ter case the 1:1 compensator gives an image with a veryuniform reconstructed attenuation value. The 4:1 image issimilar in signal uniformity to the 1:1 case, though somereduction in the attenuation value is seen at the edges of thephantom. Large discrepancies are seen with the 8:1 compen-sator image where a considerable “cap” is seen in the imagewith a peak in reconstructed attenuation values at the centerof the image. The cause of the nonuniformity in the recon-struction is hypothesized to be the change in the energy spec-trum of the x-ray beam when passing through the compen-sators. The shape of the compensators results in a spectrumwith a higher mean energy as the beam moves away from thecentral axis. The harder x-ray spectrum directed at the out-side of the cylinder, caused by the thicker copper on thecompensator, results in a larger percentage of the x rays pass-ing through the cylinder resulting in a perceived reduction inx-ray attenuation at the periphery of the cylinder. A signal-to-noise �SNR� analysis of the images was performed in 10voxel by 10 voxel regions near the center and edges of re-constructed images. This analysis demonstrates that the com-pensators provide a more uniform pattern of noise across theimages, with SNR values at the edges of the phantom of72±4 for the 1:1 case, 54±5 for the 4:1 compensator, and41±4 for the 8:1, and SNR values near the center of thephantom of 33±3, 36±5, and 37±5 for the 1:1, 4:1, and 8:1compensators, respectively.
Increasing the FOVz up to 18.5 cm significantly increasesthe percentage of scatter present in the system. In the imageacquired with the 1:1 compensator a severe cupping artifactcaused by the increased scatter is now seen. The reduction inscatter afforded by the 4:1 compensator provides a substan-tial reduction in the cupping due to scatter. An additionalbenefit is that the 4:1 image is acquired using an estimated15% to 20% decrease in dose compared to the 1:1 case,though the SNRs for these two images are not significantlydifferent �72±4 at the edge and 42±3 at the center of the 1:1image, compared to 71±5 at the edge and 48±4 at the centerof the 4:1 reconstruction�. In the 8:1 case the change in thespectrum across the beam continues to cause an artifact inreconstruction. In fact, the capping artifact in the 8:1 imagehas increased in the presence of higher amounts of scatteredx rays. This effect is hypothesized to be due to the change inthe distribution of scattered x rays relative to primary x raysacross the phantom in the presence of the 8:1 compensator.The SNR at the center of the 8:1 image is comparable to the1:1 and 4:1 images with a value of 42±6. At the edge of the8:1 reconstruction the SNR dropped to 42±4.
With the large errors caused by beam hardening in the 8:1images it is difficult to assess any improvements in imagequality afforded by the compensating filters. To reduce theseerrors images were acquired with an additional 3 mm of cop-per filtration to considerably harden the x-ray beam. Al-
though this may not be a realistic approach for everyday
2700 Graham et al.: Compensators for cone-beam CT 2700
imaging because of the heating of the x-ray tube, it doesprovide some idea of the benefits of compensation. Figure 12shows images of the phantom with liver inserts in place forthe 1:1 and 8:1 compensators, as well as profiles across theimages. It is evident from these profiles that the cuppingartifact due to scatter is greatly reduced by the compensator.With these imaging conditions the degree of cupping, givenby6
tcup =�edge − �center
�edge� 100 % , �2�
was found to be 12.5% in 1:1 images and −1% in 8:1 images�negative because of the presence of beam hardening cappingartifacts� acquired with a FOVz of 18.5 cm. With a 2 cmFOVz the cupping was 3% in the 1:1 case and −0.6% in the8:1 case. Measurements of the CNR when imaging with a2 cm FOVz give values at the center of the phantom of2.6±0.2 and 2.4±0.2 for the 1:1 and 8:1 compensators, re-spectively, and at the edge of the phantom of 3.1±0.2 for the1:1 compensator and 2.7±0.2 for the 8:1 compensator. So,for small longitudinal FOVs compensators do not offer anyimprovements in image quality. If the FOV is increased,there is a larger amount of scatter, and the CNR in the 1:1case drops to 1.9±0.2 at the center of the phantom and2.5±0.1 at the periphery of the phantom. The CNR for the8:1 compensator was measured as 2.4±0.1 at the center and
2.3±0.2 at the periphery. The 8:1 compensator does not im-
Medical Physics, Vol. 34, No. 7, July 2007
prove the CNR at the edge of the phantom compared to the1:1 compensator when imaging under high scatter condi-tions, but this is expected since the 8:1 compensator greatlyreduces the primary radiation at the periphery of the phan-tom. The improvement in CNR from the 8:1 compensator isseen at the center of the phantom where the CNR increasedfrom 1.9 up to 2.4 when compared to the 1:1 filter.
IV. DISCUSSION
The influence of modulating filters on both dose and scat-ter in cone-beam CT was evaluated. The application of com-pensating filters for all choices of phantom, air gap, and fieldof view resulted in decreased scatter and dose, with the com-pensators removing over half the scatter and reducing thedose at the center of the phantom by nearly a factor of 2 insome cases. One known limitation in this work is the pres-ence of extra-focal radiation, which cannot be separated fromx rays scattered from the imaged objects. Potential futureinvestigations will look to Monte Carlo methods for separat-ing these effects and verifying the trends seen for the DRFand SRF.
As scatter is likely the largest factor in the reduction ofimage quality in CBCT many approaches are being investi-gated to address it. This management may take the form ofalgorithms applied after imaging, or as physical modifica-
FIG. 10. Reconstructions of the cylin-drical water phantom with a FOVz ofapproximately 2 cm for the 1:1, 4:1,and 8:1 compensating filters.
tions to the imaging system that reduce the amount of scat-
2701 Graham et al.: Compensators for cone-beam CT 2701
tered radiation reaching the detector. Scatter correction algo-rithms could rely on Monte Carlo simulation methods orestimations of the scatter based on experimentally acquiredscatter fluences.36 Some of the physical modifications thatcould be made to a cone-beam CT imaging system for thepurpose of reducing scatter include adjusting the air-gap be-tween the patient and detector, reducing the longitudinalfield-of-view employed during imaging to only image theclinically relevant anatomy, utilizing anti-scatter grids to re-ject scatter, and, as evidenced by this work, application ofcompensating filters. This work gives indication of the mag-nitude of scatter reduction that can be accomplished withsimple compensators, which could be combined with othermethods of scatter reduction to achieve more quantitativelyaccurate and artifact-free reconstructions. An added benefitof compensators is that, unlike most of the methods for re-moving the effects of scatter in CBCT imaging, compensa-tors also have the ability to modulate the primary and scatterdose delivered during image acquisition.
Although the compensators are shaped to account for theshape of a head phantom, and not for the body phantom, thedecrease in image quality associated with the decrease inprimary fluence at the periphery of the phantom would notbe a problem if, for example, the anatomy at the center of thepatient volume is the most clinically relevant. Similar to re-
gion of interest imaging, this is one example of how com-
Medical Physics, Vol. 34, No. 7, July 2007
pensators could be employed while taking into account thevarious factors that they are capable of influencing.
Images acquired of a cylindrical water phantom haveshown that for compensators with considerable modulationfactors there is a possibility of introducing considerable arti-facts into the images due to the spectral hardening of thex-ray beam. Although the compensators may reduce scatter-induced artifacts, the distortion of the x-ray energy spectrumacross the FOV introduced by the compensators may bemore of a problem than the cupping from scatter. Beam hard-ening corrections have been investigated for CBCTsystems,37,38 though beam hardening corrections generallycorrect spectral changes caused only by the patient, and notby compensating filters that are not present in the recon-structed images. Implementation of beam hardening correc-tions for cone-beam CT that can account for the large spec-tral changes introduced by compensators with considerablemodulation factors may be necessary in order to use com-pensators with larger modulation factors. Further investiga-tion into the selection of materials and compensation meth-ods that minimize the spectral perturbations caused by thecompensator is required. Despite the fact that the compensa-tors employed here induce artifacts, there is still great poten-tial for the use of compensators to reduce both the scatter atthe detector and the patient dose. Compensators with lower
FIG. 11. Reconstructions of the cylin-drical water phantom with a FOVz ofapproximately 18.5 cm for the 1:1,4:1, and 8:1 compensating filters.
modulation factors than those discussed in this work are a
2702 Graham et al.: Compensators for cone-beam CT 2702
possible solution leading to a compromise between beam-hardening artifacts and the removal of scattered x rays fromthe system.
Cone-beam CT is used in a number of clinical systems,including linear accelerator and C-arm based CBCT systems.While the performance of these systems is sufficient to bevalued, image quality is still recognized as an area requiringimprovement. The results of this work demonstrate that com-pensators, if applied carefully, can play a role in reducing
FIG. 12. Reconstructions of the cylindrical water phantom with a FOVz ofapproximately 18.5 cm for the 1:1 and 8:1 compensators. Liver equivalentsoft-tissue inserts have been placed at two locations inside the phantom toallow measurement of the contrast-to-noise ratio.
patient dose and increasing CNR through reduction of scat-
Medical Physics, Vol. 34, No. 7, July 2007
tered x rays. The challenge and next logical step of this workis to seek the appropriate compromises that still satisfy theclinical requirements.
V. CONCLUSIONS
In this study, two representative compensators have beenused to demonstrate some of the capabilities of modulatingfilters in cone-beam CT. Large reductions in scatter and pa-tient dose could be made, as well as increased uniformity inreconstructed images, a decrease in cupping artifacts in re-constructed volumes, and an increase in CNR at the centerphantoms imaged under high-scatter conditions. All of theseresults are expected to be similar to those that could beachieved with compensators of different shape or material.One complicating factor that requires more investigation isthe introduction of beam-hardening artifacts into recon-structed images. One possibility for the design of a compen-sator would be to optimize the compensator such that it in-troduced the least amount of spectral hardening artifacts.
An approach that would potentially be more beneficial topatients would be that the attenuation profile of compensat-ing filters could be optimized to account for numerous prop-erties of an imaging system. Depending on the desired im-aging task, various aspects of the cone-beam CT imagingsystem, such as the magnitude of scatter, dose, and distribu-tion of primary fluence, can be influenced by compensatingfilters. Further investigations into the possibility of creatingoptimal fluence patterns with compensators to be utilizedduring CBCT imaging continue to be an active area of re-search.
ACKNOWLEDGMENTS
The authors wish to acknowledge the radiation physicsmachine shop at Princess Margaret Hospital for constructionof the compensators, and Bronwyn Hyland for assistance inediting the manuscript. This project was supported by theOntario Graduate Scholarship program, the National Insti-tutes of Health/National Institutes of Aging �R21/R33-AG19381�, and the National Institutes of Health/National In-stitute of Biomedical Imaging and Bioengineering�8R01EB002470-04�.
1“Management of target localization uncertainties in external-beamtherapy,” Seminars in Radiation Oncology, edited by G. S. Mageras, Vol.15, pp. 133–216 �2005�.
2D. A. Jaffray, J. H. Siewerdsen, J. W. Wong, and A. A. Martinez, “Flat-panel cone-beam computed tomography for image-guided radiationtherapy,” Int. J. Radiat. Oncol., Biol., Phys. 53, 1337–1349 �2002�.
3R. Fahrig, A. J. Fox, S. Lownie, and D. W. Holdsworth, “Use of a C-armsystem to generate true three-dimensional computed rotational angio-grams: Preliminary in vitro and in vivo results,” AJNR Am. J. Neurora-diol. 18, 1507–1514 �1997�.
4R. Fahrig and D. W. Holdsworth, “Three-dimensional computed tomog-raphic reconstruction using a C-arm mounted XRII: Image-based correc-tion of gantry motion nonidealities,” Med. Phys. 27, 30–38 �2000�.
5D. A. Jaffray and J. H. Siewerdsen, “Cone-beam computed tomographywith a flat-panel imager: Initial performance characterization,” Med.Phys. 27, 1311–1323 �2000�.
6J. H. Siewerdsen and D. A. Jaffray, “Cone-beam computed tomographywith a flat-panel imager: Magnitude and effects of x-ray scatter,” Med.
2703 Graham et al.: Compensators for cone-beam CT 2703
7J. M. Boone and J. A. Seibert, “An analytical model of the scatteredradiation distribution in diagnostic radiology,” Med. Phys. 15, 721–725�1988�.
8J. M. Boone and J. A. Seibert, “Monte Carlo simulation of the scatteredradiation distribution in diagnostic radiology,” Med. Phys. 15, 713–720�1988�.
9L. Spies, M. Ebert, B. A. Groh, B. M. Hesse, and T. Bortfeld, “Correctionof scatter in megavoltage cone-beam CT,” Phys. Med. Biol. 46, 821–833�2001�.
10R. Ning, X. Tang, and D. Conover, “X-ray scatter correction algorithm forcone beam CT imaging,” Med. Phys. 31, 1195–1202 �2004�.
11J. H. Siewerdsen, M. J. Daly, B. Bakhtiar, D. J. Moseley, S. Richard, H.Keller, and D. A. Jaffray, “A simple, direct method for x-ray scatterestimation and correction in digital radiography and cone-beam CT,”Med. Phys. 33, 187–197 �2006�.
12J. Persliden and G. A. Carlsson, “Scatter rejection by air gaps in diagnos-tic radiology. Calculations using a Monte Carlo collision density methodand consideration of molecular interference in coherent scattering,” Phys.Med. Biol. 42, 155–175 �1997�.
13J. H. Siewerdsen and D. A. Jaffray, “Optimization of x-ray imaging ge-ometry �with specific application to flat-panel cone-beam computed to-mography�,” Med. Phys. 27, 1903–1914 �2000�.
14J. M. Boone, K. K. Lindfors, V. N. Cooper III, and J. A. Seibert, “Scatter/primary in mammography: Comprehensive results,” Med. Phys. 27,2408–2416 �2000�.
15U. Neitzel, “Grids or air gaps for scatter reduction in digital radiography:A model calculation,” Med. Phys. 19, 475–481 �1992�.
16T. M. Bernhardt, U. Rapp-Bernhardt, T. Hausmann, G. Reichel, U. W.Krause, and W. Doehring, “Digital selenium radiography: Anti-scattergrid for chest radiography in a clinical study,” Br. J. Radiol. 73, 963–968�2000�.
17A. Malusek, M. Sanbourg, and G. A. Carlsson, “Simulation of scatter incone beam CT—effects on projection quality,” Proc. SPIE 5030, 740–751 �2003�.
18J. H. Siewerdsen, D. J. Moseley, B. Bakhtiar, S. Richard, and D. A.Jaffray, “The influence of antiscatter grids on soft-tissue detectability incone-beam computed tomography with flat-panel detectors,” Med. Phys.31, 3506–3520 �2004�.
20R. Ning, University of Rochester, “Apparatus and method for x-ray scat-ter reduction and correction for fan beam CT and cone beam volume CT,”U.S. Patent No. 6,618,466�2003�.
21R. Ning, X. Tang, D. Conover, and R. Yu, “Flat panel detector-based conebeam computed tomography with a circle-plus-two-arcs data acquisitionorbit: preliminary phantom study,” Med. Phys. 30, 1694–1705 �2003�.
22C. D. Smith, “X-ray filter,” U.S. Patent No. 2,216,326 �1940�.
Medical Physics, Vol. 34, No. 7, July 2007
23P. Edholm, N. B. Jacobson, and Medinova A B, “Exposure compensatingdevice for radiographic apparatus,” U.S. Patent No. 3,755,672 �1973�.
24T. Katsuda, M. Okazaki, and C. Kuroda, “Using compensating filters toreduce radiation dose,” Radiol. Technol. 68, 18–22 �1996�.
25G. N. Hounsfield, EMI Limited, “Body portion support for use with pen-etrating radiation examination apparatus,” U.S. Patent No. 3,867,634�1975�.
26R. A. Brooks and G. Di Chiro, “Beam hardening in x-ray reconstructivetomography,” Phys. Med. Biol. 21, 390–398 �1976�.
27H. H. Barrett and W. Swindell, Radiological Imaging: The Theory ofImage Formation, Detection, and Processing �Academic, New York,1981�.
28R. G. Walters and R. W. Carlson, Technicare Corporation, “Computerizedtomographic scanner with shaped radiation filter,” U.S. Patent No.4,288,695 �1981�.
29J. E. Tkaczyk, Y. Du, D. Walter, X. Wu, J. Li, and T. Toth, “Simulation ofCT dose and contrast-to-noise as a function of bowtie shape,” Proc. SPIE5368, 403–410 �2004�.
30S. Itoh, S. Koyama, M. Ikeda, M. Ozaki, A. Sawaki, S. Iwano, and T.Ishigaki, “Further reduction of radiation dose in helical CT for lung can-cer screening using small tube current and a newly designed filter,” J.Thorac. Imaging 16, 81–88 �2001�.
31G. H. Glover, “Compton scatter effects in CT reconstructions,” Med.Phys. 9, 860–867 �1982�.
32J. Hsieh, GE Medical Systems Global Technology Co., LLC, “Methodand apparatus for optimizing dosage to scan subject,” U.S. Patent No.6,647,095 �2003�.
33M. J. Berger, J. H. Hubbell, S. M. Seltzer, J. Chang, J. S. Coursey, R.Sukumar, and D. S. Zucker, XCOM: Photon Cross Section Database(version 1.3) �National Institute of Standards and Technology, Gaithers-burg, MD, 2005�. �Available online: http://physics.nist.gov/xcom �13 Feb-ruary 2007��.
34P. C. Johns and M. Yaffe, “Scattered radiation in fan beam imaging sys-tems,” Med. Phys. 9, 231–239 �1982�.
35L. A. Feldkamp, L. C. Davis, and J. W. Kress, “Practical cone-beamalgorithm,” J. Opt. Soc. Am. A 1, 612–619 �1984�.
36G. Jarry, S. A. Graham, D. J. Moseley, D. A. Jaffray, J. H. Siewerdsenand F. Verhaegen, “Characterization of scattered radiation in kV CBCTimages using Monte Carlo simulations,” Med. Phys. 33, 4320–4329�2006�.
37J. Hsieh, R. C. Molthen, C. A. Dawson, and R. H. Johnson, “An iterativeapproach to the beam hardening correction in cone beam CT,” Med. Phys.27, 23–29 �2000�.
38L. Zhang, H. Gao, S. Li, Z. Chen, and Y. Xing, “Cupping artifacts analy-sis and correction for a FPD-based cone-beam CT,” Proc. SPIE 6065,282–291 �2006�.
