Beam generation and planar imaging at energies below 2.40 MeV withcarbon and aluminum linear accelerator targets

David ParsonsDepartment of Physics and Atmospheric Science, Dalhousie University, 5820 University Avenue, Halifax,Nova Scotia, B3H 1V7, Canada

James L. RobarDepartment of Radiation Oncology, Dalhousie University, 5820 University Avenue, Halifax, Nova Scotia,B3H 1V7, Canada

(Received 20 February 2012; revised 6 June 2012; accepted for publication 7 June 2012; published 5July 2012)

Purpose: Recent work has demonstrated improvement of image quality with low-Z linear acceleratortargets and energies as low as 3.5 MV. In this paper, the authors lower the incident electron beamenergy between 1.90 and 2.35 MeV and assess the improvement of megavoltage planar image qualitywith the use of carbon and aluminum linear accelerator targets.Methods: The bending magnet shunt current was adjusted in a Varian linear accelerator to allowselection of mean electron energy between 1.90 and 2.35 MeV. Linac set points were altered toincrease beam current to allow experimental imaging in a practical time frame. Electron energy wasdetermined through comparison of measured and Monte Carlo modeled depth dose curves. Planarimage CNR and spatial resolution measurements were performed to quantify the improvement ofimage quality. Magnitudes of improvement are explained with reference to Monte Carlo generatedenergy spectra.Results: After modifications to the linac, beam current was increased by a factor greater thanfour and incident electron energy was determined to have an adjustable range from 1.90 MeV to2.35 MeV. CNR of cortical bone was increased by a factor ranging from 6.2 to 7.4 and 3.7 to 4.3for thin and thick phantoms, respectively, compared to a 6 MV therapeutic beam for both aluminumand carbon targets. Spatial resolution was degraded slightly, with a relative change of 3% and 10% at0.20 lp/mm and 0.40 lp/mm, respectively, when reducing energy from 2.35 to 1.90 MV. The percent-age of diagnostic x-rays for the beams examined here, ranges from 46% to 54%.Conclusion: It is possible to produce a large fraction of diagnostic energy x-rays by lowering thebeam energy below 2.35 MV. By lowering the beam energy to 1.90 MV or 2.35 MV, CNR improvesby factors ranging from 3.7 to 7.4 compared to a 6 MV therapy beam, with only a slight degradation ofspatial resolution when lowering the energy from 2.35 MV to 1.90 MV. © 2012 American Associationof Physicists in Medicine. [http://dx.doi.org/10.1118/1.4730503]

Key words: low-Z target, carbon, aluminum, planar imaging

I. INTRODUCTION

In the past, megavoltage portal imaging has been used toalign a patient’s position for radiation therapy. These imageshave inherently poor contrast due primarily to the Comptondominant interaction of photons in the megavoltage energyrange. The introduction of a diagnostic quality x-ray tubeand detector, orthogonal to the therapy beam line1 haslargely replaced portal imaging as a means for image guidedradiation therapy. While the use of this technology for planarimaging and cone beam computed tomography (kV CBCT)has greatly improved available image quality, it also involvesgreater cost, increased quality assurance, and does not allowfor beam’s eye viewing.

Several groups have shown that the removal of the highatomic number (Z) target and flattening filter from the therapybeam line and the introduction of a low-Z target can greatlyenhance image quality compared to standard megavoltagetherapy beams.2–6 Most recently, Faddegon et al.7 used a1.32 cm thick carbon target placed in the high-energy electron

scattering port of a high-energy linac with 4.2 MeV electronsincident on the target. This work showed that soft tissue CNRwas improved up to a factor of three in CBCT images com-pared to a 7.0 MV treatment beam line. Similarly, Robertset al.8 used a 2 cm thick carbon target (ρ = 1.89 g/cm3)placed in the high-energy collimator port of high-energylinac operating in 4 MeV electron mode with the primaryand secondary scatter foils removed. This work showed thatthe megavoltage planar image contrast of dense bone inwater was improved by a factor of 4.62 for thin phantoms(5.8 cm thick) and 1.3 for thicker phantoms (25.8 cm).Spatial resolution was improved when compared to the 6 MVtherapy beam. Orton and Robar9 modeled 7.0 MeV electronsincident on an aluminum and beryllium low-Z targets in ahigh-energy linac, with a QC3 phantom and 25-layer aS500detector, demonstrating a contrast enhancement factor rang-ing from 1.6 to 2.8. Sawkey et al.10 showed a 25% increase incontrast when decreasing the electron energy from 6 MeV to4 MeV on a diamond target in a high-energy linac, with no

4568 Med. Phys. 39 (7), July 2012 © 2012 Am. Assoc. Phys. Med. 45680094-2405/2012/39(7)/4568/11/$30.00

4569 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4569

appreciable change in spatial resolution. Similarly Robaret al.11 showed CNR increases by factors ranging from of 1.2to 1.7 for 7.0 MeV incident on an Al target compared to a6 MV therapeutic beam, and when incident electron energywas decreased to 3.5 MeV, this factor increased, ranging from2.7 to 3.8. Connell and Robar12 showed that spatial resolutionis improved with certain combinations of low-Z materialand thickness compared to a 6 MV therapy beam. A slightdecrease in spatial resolution was observed when decreasingthe incident electron from 7.0 to 4.5 MeV (Ref. 12).

It is clear that by lowering the incident electron energy incombination with a low-Z target, for a given imaging dose,CNR can be greatly enhanced. To date, the lowest energy ex-amined with a low-Z target beam has been 3.5 MeV (Ref. 11);here we address the issues of (i) the amount by which the inci-dent electron energy can be lowered in a standard high energylinac, and (ii) the resultant gains in planar image CNR. Wedemonstrate practical limitations with regard to the reductionof beam current with lowered energy,13 and report on depthdose, imaging CNR, and energy spectral characteristics, forbeams generated with carbon and aluminum targets, with in-cident electron beams with energies below 2.4 MeV.

II. MATERIALS AND METHODS

II.A. Beam generation and low-Z targets

The 6 MV and low-Z target beams were generated us-ing a Varian 2100EX linear accelerator (Varian Medical, Inc.,Palo Alto, CA). Two separate low-Z targets were investigated(i) 6.7 mm thick aluminum (ρ = 2.69 g/cm3) and (ii) 7.6 mmthick carbon (ρ = 1.88 g/cm3). The aluminum target was orig-inally designed and used by Robar et al.11 previously for a3.5 MV beam. The thickness of the carbon target was calcu-lated to coincide with the continuous slowing down approx-imation range for a 2.5 MeV electron. While an increase inlow-Z target thickness slightly degrades resolution,12 a “full-thickness” target alleviates the need for a polystyrene filter toabsorb transmitted electrons and increases the production oflower energy photons.2 As described previously,9, 11, 12 whenused experimentally, the carousel is operated in manual modeand the appropriate target is rotated into the beam line viaa rotary switch. The linac is operated in electron mode andequipped with a dedicated 4 MeV program board used forresearch purposes only. The bending magnet was adjustedto tune the beam energy down from the nominal operatingpoint of 4.5 MeV to the lower energies used. Complete de-sign and installation of similar targets used in this study hasbeen published recently9, 12 and will not be explained in fullhere. Briefly, targets were mounted to the beam side of thecarousel and vertically offset 0.9 cm from the beryllium exitwindow of the primary collimation vacuum. By keeping thetarget as close as possible to the beryllium exit window, theamount of electron scattering in air is minimized.8, 9, 12

II.B. Beam tuning to maximize current

Adjustment of the bending magnet current can be used toselect the mean energy of electrons exiting the bending mag-

net, with electrons outside the mean energy being stoppedwithin the linac head. Mean energy selection is accomplishedby an energy slit within the bending magnet that allows elec-tron energy selection within 3% of the nominal energy. Whenusing this technique, the number of electrons within the se-lected energy window decreases when the selected energy dif-fers from the nominal energy for the selected mode. This isdue to the irregular electron spectra exiting the waveguide.Although the electron energy spectrum exiting the waveguideis often approximated as Gaussian,14–16 in fact it can be highlyasymmetric,14, 15 with a sharp decrease in electron populationaway from the nominal energy. An example of this can be seenin the work by Wei et al.,14 which gives electron spectra re-constructed from depth dose curves. The results demonstratethat low-energy electrons, e.g., ranging from 1 to 3 MeV, arepresent in a 6 MeV spectrum; however, the relative fluenceof these electrons is approximately one hundredth of that atthe nominal energy. As a result, any lowering of the incidentelectron energy results in a considerably lower beam currentexiting the bending magnet and, for our application, signifi-cantly protracted image acquisition times. A solution, withoutsignificant alteration to the waveguide, is to increase the totalcurrent exiting the waveguide. To observe slight changes inbeam current, a 4 MeV tantalum electron scattering foil wasinserted into the beam line via a rotatory switch controllingthe carousel. The linac was operated in 4 MeV electron modeat a MU rate of 1000 MU/min with the dose servos turned off.In maximizing beam current, we optimized the combinationof two parameters: gun high voltage (HV) and grid voltage,over the full ranges of each parameter recommended by themanufacturer for same operation. Course setting of both pa-rameters is done inside the gun deck of the linear accelerator,while the grid voltage may also be fine-tuned from outside theroom with the beam on, using the “gun-I” potentiometer. Theprocess of maximizing beam current thus involved (i) settingthe first gun HV value in the range, (ii) setting the first gridvoltage coarse value on the gun deck, (iii) fine-tuning the gridvoltage with the beam on, and (iv) recording the maximumachievable dose rate. For the same gun HV setting, the nextcoarse grid voltage value was then set on the gun deck, andfine-tuned. After exploring the full allowable range of gridvoltage, the next gun HV setting was chosen on the gun deck.This was repeated to produce a two dimensional matrix ofmeasured dose rate values, from which the optimal combina-tion of the two parameters was chosen. Finally, for this com-bination, the solenoid and buncher cavity steering set pointswere tuned on the program board to further tune beam current.We reiterate that, to avoid potential damage, the examinedrange did not exceed the manufactures recommendations. Fi-nally, once the final tuning was complete, the scattering foilwas replaced by either aluminum or carbon targets for x-rayproduction.

II.C. Electron energy determination

Over the low electron energy range produced here, the re-lationship between bending magnet and electron energy wasnot known precisely based on manufacturer specifications.

Medical Physics, Vol. 39, No. 7, July 2012

4570 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4570

To determine incident electron energy, the 4 MeV tantalumelectron scatting foil described previously was rotated into thebeam line. Percent depth dose (PDD) curves were measuredin a 50 × 50 × 50 cm3 scanning water tank (Scandtronix,Uppsala, Sweden) with a p-type silicon electron field diode(EFD, Scandtronix, Uppsala, Sweden) at a source to surfacedistance (SSD) of 100 cm and with a 10 × 10 cm2 electron ap-plicator. Measurements were taken at 0.2 mm intervals. Thesecurves were then matched to Monte Carlo PDD curves ofknown electron energy. BEAMnrc (Ref. 17) and DOSXYZnrc(Ref. 18) were used to simulate the linac and water phantom.All simulations were run in accordance with the physicalsetup using validated models provided by Orton and Robar.9

Parallel beams of mono-energetic electrons were used witha spot size of 0.1 cm. Global electron (ECUT) and photon(PCUT) cut-off energies of 0.521 MeV and of 0.010 MeV,respectively, were used. The incident electron energy wasvaried in 0.025 MeV increments between 1.700 MeV and2.500 MeV and plotted against the measured data in MATLAB

(Mathworks, Natick, MA).

II.D. Photon measurements

Given that low-Z target photon beams have not been pro-duced previously using the low electron energies consideredhere, it was of interest to examine depth dose characteristics.Photon depth dose measurements were acquired using a 50× 50 × 50 cm3 water tank with 0.015 cm3 (PTW N31014,Freiburg, Germany) and 0.125 cm3 (PTW N31010, Freiburg,Germany) cylindrical ion chambers. For all measurements,an SSD of 100 cm and a field size of 10 × 10 cm2 were used.The 0.015 cm3 ion chamber was used for measurementsshallower than 1.0 cm below the surface to provide resolutionin the buildup region, while the 0.125 cm3 ion chamber wasused for improved signal-to-noise characteristics at greaterdepths. These two sets of data were concatenated at a depthof 1.0 cm and the resultant curve was normalized to the max-imum dose. Photon depth dose measurements were acquiredfor both carbon and aluminum targets, for the minimum andmaximum energies examined in this investigation, which,using the method described above, were determined to be1.90 MV and 2.35 MV.

II.E. Photon Monte Carlo simulations

In order to study changes in photon fluence with energyand target material, BEAMnrc (Ref. 17) was used to simulatethe linac beam-line. The linac was modeled in accordancewith the physical setup using validated models provided byOrton and Robar.9 1.1 × 108 incident electron histories wererun for the aluminum and carbon targets, with energies of 1.90and 2.35 MeV. Linac collimation was set to a field size of 10× 10 cm2 (specified at isocenter). Selective bremsstrahlungsplitting was used, and as recommended by Rogers et al.,19

an effective splitting field size of 20 × 20 cm2 was used at anSSD of 100 cm with minimum and maximum bremsstrahlungsplitting numbers of 100 and 1000, respectively. PCUT andECUT were set to 0.010 MeV and 0.700 MeV, respectively.

For confirmation of the modeled spectra, percent depth doseswere run for the aluminum target at 1.90 MV and 2.35 MVin DOSXYZnrc (Ref. 18) and compared to correspondingmeasured results. To determine the spectral distributions,BEAMdp (Ref. 20) was used to analyze the phase space data.To determine the source of the photons and their resultantspectral contribution, the LATCH feature was utilized toidentify the origin of generated photons. In addition to sim-ulating the two homogeneous carbon or aluminum targets, acarbon target with various thickness of copper (on the exitside) was modeled to examine the requirements for filtrationof very low energy (e.g., <25 keV) photons that wouldcontribute to superficial patient dose but not image formation.

II.F. Imaging system

An aS1000/IAS3 imaging system (Varian Medical Sys-tems, Inc.) was used for all planar imaging. The system wasset up as a stand-alone configuration separate from the clini-cal imaging system on the treatment unit. The aS1000 panelhas an active area of 30 × 40 cm2 and includes a 1.0 mm Cubuildup plate, a Gd2O2S:Tb scintillating phosphor layer and a1024 × 768 array of photodiodes switched by thin-film tran-sistors deposited on a glass substrate. The setup for imagingwas as follows: The gantry and couch were rotated to 90◦. Theimaging panel was placed in a stand located on the couch withfine adjustments screws and leveled to ensure orthogonalityof the panel and the beam axis. For all imaging, a source-to-detector distance (SDD) of 140 cm was used. The number ofmonitor units (MU) per exposure was varied to set the doseto the phantom, where the IAS3 allows cMU control of thisparameter. Prior to imaging, flood field and dark field cali-brations were performed where the former has the effect ofcorrecting for the forward peaked nature of the low-Z target,flattening filter free beam.

II.G. Image quality: Contrast to noise ratio

To observe changes in CNR with target material andenergy, a contrast phantom was constructed for planarimaging. The phantom, shown in Fig. 1, consisted of an18 × 18 × 3 cm3 block of polystyrene with six circular holesof 2.5 cm diameter for tissue equivalent materials (Gammex,Middleton, WI). To measure dose, a channel was added alongthe phantom midline to accommodate a 0.125 cm3 cylindricalion chamber (PTW N31010, Freiburg, Germany). The cham-ber was calibrated for the low-Z target beams by measuringthe dose per MU in solid water (Gammex, Middleton, WI)with thermoluminescent dosimeters (TLDs) (TLD-800,Saint-Gobain Crystals, Hiram, OH) at an SSD of 95 cmdepth of 5 cm with a 10 × 10 cm2 (defined at isocenter).Manganese doped lithium tetraborate (Li2B4O3:Mn) TLDswere used due to their minimal energy dependence21 (0.9 at30 keV/60Co). The TLDs were then removed and replacedby the ionization chamber in the same location, to obtain aconversion from collected charge to cGy. This was done forthe 2.35 MV carbon and aluminum target beams. Duringimaging the phantom was oriented as shown in Fig. 1, such

Medical Physics, Vol. 39, No. 7, July 2012

4571 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4571

FIG. 1. Left figure shows a schematic drawing of the contrast phantom. Shown are tissue equivalent inserts with electron densities relative to water. Right figureshows the experimental setup for planar imaging.

that the circular cross sections of material regions faced thebeam and the phantom was centered on the linac isocenter.Images were acquired with dose values at the phantom centerranging from approximately 0.01 to 0.16 cGy. This low doserange was chosen to match a typical anticipated dose perprojection used in CBCT, given that this is a likely applicationof the low-Z target beam.11 CNR was calculated as

CNR = |Pmaterial − Ppolystyrene|σ 2

material + σ 2polystyrene

, (1)

where Pmaterial is the average pixel value within the insert,Ppolystyrene is the average value of the surrounding polystyrene,σ material is the average noise within the material and σ polystyrene

is the average noise in the surrounding polystyrene. Pixelvalues in the surrounding polystyrene were measured withintwo concentric circles with radii of 1.8 and 2.1 cm (measuredat isocenter). Error bars were found by calculating the meanand standard deviation of CNR measured in multiple images.While the 3 cm thick phantom in Fig. 1 allows sensitivityin measuring changes of CNR with beam parameters, it isnot representative of typical patient separation. Thus, theexperiment was repeated with the phantom padded by 5 cmof solid water on the source side and 7 cm on the detectorside, for a total phantom thickness of 15 cm.

II.H. Image quality: Spatial resolution

Images of the QC3 phantom22 were taken at photon ener-gies of 1.90, 2.15, and 2.35 MV/carbon and a 6 MV thera-peutic beam. These images were analyzed according to themethod outlined by Rajapakshe et al.,22 where the relativemodulation transfer function (RMTF) is defined as

RMTF = M(f )

M(f1), (2)

where M( f ) is the output modulation of the line pair in a re-gion of interest and M( f1) is the output modulation for thelowest frequency line pair region. As suggested by Droege

and Morin,23 the output modulation is obtained by using therelationship between the signal amplitude and its variance isgiven by

M2(f ) = σ 2m(f ) − σ 2(f ), (3)

where σ m2( f ) is the measured total variance within a region of

interest and σ 2( f ) is the variance due to random noise withinthe region of interest. The variance due to random noise ascalculated by Rajapakshe et al.22 is given by

σ 2(f ) = σ 2sub

2, (4)

where σ 2( f ) represents the variance within the region of in-terest of two subtracted images. Error bars were found by cal-culating the mean and standard deviation of RMTF measuredin multiple image sets.

II.I. Qualitative planar imaging

Planar imaging of a sheep head was done to observe thequalitative effects on imaging with changes in target compo-sition and energy. For these images the CNR phantom wasreplaced with an adult sheep head, all other setup parameterswere unchanged. An approximate imaging dose of 0.14 cGywas delivered to the center of the head, located on isocen-ter. Images were acquired at 2.35 MV with both carbon andaluminum target beams and a 6 MV therapeutic beam. The re-sultant images were analyzed in MATLAB and were comparedwith identical gray level window settings.

III. RESULTS AND DISCUSSION

III.A. Beam current maximization

After adjusting the bending magnet shunt current onlywith the scattering foil in place, the electron MU rate wasapproximately 600 MU/min at 2.35 MeV. As shown inFig. 2, upon altering the voltage of the electron gun and gridto the optimal setting the MU rate increased to approximately

Medical Physics, Vol. 39, No. 7, July 2012

4572 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4572

FIG. 2. MU rate in relation to electron energy for both electrons and photons. The upper plot shows observed MU rate with tantalum scattering foil in place.The lower plot shows observed MU rate with aluminum and carbon targets in place. Error bars are generated from the observed variation in MU rates.

2800 MU/min. The MU rate gradually decreases as the inci-dent electron energy is decreased to 1.90 MeV, below which asharp drop was observed from approximately 1400 MU/minto 20 MU/min. Due to this, it does not appear possible to oper-ate at a lower energy with this current setup. With the carbonand aluminum targets in place, a MU rate was observed rang-ing from 5–9 MU/min and 3–6 MU/min, respectively, beforeadjustments this rate was 0–1 MU/min. With these improve-ments in beam current, it is now possible to acquire multiplelow-Z target images per second at doses typical of CBCT pro-jections. It should be noted that the monitor chamber was notcalibrated and that these are only relative MU rate measure-ments. However, with the carbon target in place, a dose rate of0.7 cGy/min was measured for a 10 × 10 cm2 field at a depthof 5 cm and an SSD of 95 cm was measured with a 0.125 cm3

cylindrical ion chamber (PTW N31010, Freiburg, Germany).

III.B. Incident electron energy

Figure 3 shows the experimentally measured and MonteCarlo modeled electron PDD curves for energies rangingfrom 1.90 MeV to 2.35 MeV. The measured and modeleddata show excellent agreement in the dose build-up andfall-off region of the PDD, with a slight variation wherethe central axis depth dose curve meets the bremsstrahlungbackground. The depth of the 90% PDD (R90) ranges from2.9 mm to 4.3 mm for 1.90 MeV and 2.35 MeV, respectively.Beyond the buildup region, the dose quickly decreases withR50 ranging from 4.2 mm to 6.1 mm for 1.90 MeV and2.35 MeV, respectively, and the maximum range is less than9.0 mm for the 2.35 MeV electron beam. To our knowledge,these are the lowest energy beams used in conjunction with alow-Z target produced with a megavoltage Clinac.8, 10, 11

FIG. 3. Experimental (curves) and Monte Carlo (points) electron percent depth dose curves ranging from 1.90 MeV to 2.35 MeV.

Medical Physics, Vol. 39, No. 7, July 2012

4573 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4573

FIG. 4. Percent depth doses for both aluminum and carbon targets are shown as well as corresponding Monte Carlo modeled depth dose curves for the aluminumtarget.

III.C. Photon beam characteristics

Figure 4 shows measured percent depth dose curves nor-malized to maximum dose for 1.90 and 2.35 MeV electronbeams incident on the aluminum and carbon targets. Alsoshown are modeled depth dose curves for the aluminum tar-get. Because of the low dose rates produced, relatively fewdepths were measured compared to the number of modeleddepths. As expected, as energy was lowered the dose at agiven depth was decreased. At 2.35 MV, we observe dmax at0.6 ± 0.1 cm and 0.4 ± 0.1 cm for the aluminum and carbontargets, respectively. No distinguishable shift in dmax towardthe surface is observed when decreasing the incident electronenergy from 2.35 MV to 1.90 MV. For comparison, Faddegonet al.7 and Sawkey et al.10 reported a dmax between 0.9 cm and1.1 cm for 4.2 MV/carbon beams, respectively. Compared tothe 4.2 MV/carbon beam generated by Faddegon et al.,7 the2.35 MV/carbon beam is significantly less penetrating witha change in dmax of approximately 0.5 cm, toward the surfaceand a decrease in the percent dose at 10 cm of 25%. Comparedto the 4 MV/aluminum investigated by Orton and Robar,9 thepercentage of photons within the diagnostic energy range wasenhanced by approximately 9% and 13% when decreasing theincident electron energy to 2.35 and 1.90 MeV, respectively,on an aluminum target. At a depth of 10 cm, a decrease inpercent dose of 20% and 22% was observed when changingthe target from aluminum to carbon at 2.35 MV and 1.90 MV,respectively. These observations indicate that the aluminumtarget produces a harder beam compared to the carbon target.

This is confirmed when examining the spectral distributionsshown in Fig. 5, which shows that the mode of the spectrumis shifted from 20 keV to 50 keV when replacing aluminumwith carbon, and significantly more photons exist below40 keV (25% for carbon compared to 6% with aluminum).This is caused by the reduced hardening of the beam throughphotoelectric absorption by the lower-Z target material.

Figure 6 illustrates the sources of photon generation withinthe linac. Approximately 12%–17% of photons are generatedwithin the beryllium exit window of the primary collimator,with 87%–83% being generated in the aluminum and car-bon targets, respectively, at 2.35 MeV. The amount of photonsgenerated in the exit window was relatively minor comparedto that report by Roberts et al.,8 in which approximately 71%of the beam was generated in the nickel exit window and 28%in the carbon target.

The considerable population of very low energy photonsfrom the carbon target may increase patient dose without con-tributing to image formation. Figure 6 shows 2.35 MV/carbonspectra for various thicknesses of copper on the exit side ofthe target. The percentage of photons below 25 keV can bereduced by half with as little as 25 μm of copper.

III.D. Quantitative image quality

Figure 7 shows planar CNR images of the cortical boneinsert with 6 MV therapeutic, and 2.35 MV carbon andaluminum beams at 0.01, 0.05, and 0.10 cGy for a thin (3 cm)

Medical Physics, Vol. 39, No. 7, July 2012

4574 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4574

FIG. 5. Monte Carlo generated spectral distributions for carbon and aluminum targets over the diagnostic energy range (25–150 keV), for the maximum andminimum energies used.

and a thick (15 cm) phantom. These images show that CNRis improved with increased dose and a decreased phantomthickness and also highlight the difference in image qualitybetween a 6 MV therapeutic beam and the low-Z targetbeams. Figure 7 also shows nonuniformity in the centralregion of the detector when imaged with the 6 MV thera-peutic beam, as previously reported by Orton and Robar.9

Figure 8 shows the corresponding thin phantom CNR resultsfor cortical bone, inner bone, brain, and breast inserts as afunction of dose at the center of the phantom. At approxi-mately 0.05 cGy, CNR values between a 6 MV therapeuticbeam and the low-Z target beams were increased by a factorranging from 6.2 to 7.4, from 8.5 to 9.7, from 6.7 to 7.5, andfrom 2.2 to 2.7 for cortical bone, inner bone, brain, and breast,respectively. It is notable that even for this thin phantom, theCNR values for the 6 MV beam do not exceed 1.0, exceptfor cortical bone. Figure 9 shows the corresponding thickphantom CNR results. At approximately 0.05 cGy, CNRvalues between a 6 MV therapeutic beam and the low-Ztarget beams were increased by a factor ranging from 3.7to 4.3, from 5.0 to 6.0, and from 7.2 to 10.0 for corticalbone, inner bone, and brain, respectively, with no significantimprovement for breast material for the thick phantom. Nomeasureable improvement in CNR was seen with reductionof energy from 2.35 MV to 1.90 MV, or between targetmaterials. Ideally an increase in CNR should be observedfor the lowest energy and atomic number material.2, 6, 9, 11

However, Fig. 5 illustrates that the relative increase in photon

fluence for carbon is primarily for the spectral componentbelow approximately 50 keV. In this very low energy range,a significant proportion of photons will be absorbed withinthe phantom or within the copper layer in detector, therebynot contributing to imaging data.5 Previous measurements ofCNR (Refs. 2, 4–6, 8, and 11) with low-Z target beams haveinvolved a variety of materials and phantom geometries, andtherefore it is in general difficult to compare directly to CNRvalues reported in the work of others. However, extrapolatingthe reduction of measured CNR with the five-fold increasein phantom thickness, we would expect that the gains inplanar CNR reported here are improved upon the 2.7–3.8reported by Robar et al.11 for a 22.5 cm thick water phantom.Although this direct comparison is difficult, it is intuitivethat increasing the proportion of photons in the diagnosticenergy range from 37% reported by Orton and Robar9 fora 4.0 MV/aluminum to approximately 50% should increasethe CNR characteristics. It is clear from work reported hereand the reports by Ostapiak et al.,4 Roberts et al.,8 andRobar et al.11 that image quality is degraded with increasingphantom thickness and therefore is not ideally suited to largethicknesses such as those seen in the pelvic region.

Figure 10 shows RMTF curves of the QC3 phantom, forthe carbon target at 1.90, 2.15, and 2.35 MV and for the 6 MVtherapeutic beam. Measured spatial resolution is consistentwith Connell and Robar,12 Roberts et al.,8 and Sawkey et al.10

with a spatial frequency of 0.40 lp/mm being distinguishable.Slight degradation of the spatial resolution is apparent with

Medical Physics, Vol. 39, No. 7, July 2012

4575 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4575

FIG. 6. Left figure shows Monte Carlo generated spectral distributions for the beryllium exit window and the carbon and aluminum targets, at 2.35 MV. Rightfigure shows Monte Carlo generated spectral distributions at 2.35 MV for the carbon target with various thickness of copper (on the exit side) and the percentageof photons with energy less than 25 keV.

the low energy carbon target beams, compared to 6 MV.The frequency where RMTF equals 0.5, or f50, decreases byapproximately 0.1 lp/mm, from 0.42 to 0.32 lp/mm. Withbeams of 4.5 MV/tungsten and 7.0 MV/beryllium, Connell

and Robar12 showed a relative reduction of f50 by 10.4%–15.5% compared to a 6 MV therapeutic beam. As reportedby Roberts et al.8 and Connell and Robar,12 the loss of spatialresolution results from larger angle electron scatter in the

FIG. 7. Planar images of the cortical bone insert at 2.35 MV with both aluminum and carbon targets and 6 MV therapeutic beam at 0.01, 0.05, and 0.10 cGyfor both thin (3 cm) and thick (15 cm) phantoms.

Medical Physics, Vol. 39, No. 7, July 2012

4576 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4576

FIG. 8. Contrast-to-noise ratio results of the thin (3 cm) phantom for the four materials used at 1.90, 2.15, and 2.35 MV with both aluminum and carbon targets.For comparison, a 6 MV therapy beam is also shown.

target when decreasing incident electron energy. In our data,small decreases of RMTF (by 3% and 10% at 0.20 lp/mmand 0.40 lp/mm, respectively) are observed with a reductionof beam quality from 2.35 to 1.90 MV. These are minor

changes in RMTF compared to that observed by Connelland Robar;12 however, the work by this group involved acomparatively larger reduction of beam quality from 7 to3.5 MV.

FIG. 9. Contrast-to-noise ratio results of the thick (15 cm) phantom for the four materials used at 1.90, 2.15, and 2.35 MV with both aluminum and carbontargets. For comparison, a 6 MV therapy beam is also shown.

Medical Physics, Vol. 39, No. 7, July 2012

4577 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4577

FIG. 10. Relative modulation transfer function for 1.90, 2.15, and 2.35 MV/carbon beams and a 6 MV therapeutic beam.

FIG. 11. Planar sagittal images of a sheep head at 2.35 MV with both carbonand aluminum targets and a 6 MV therapeutic beam.

III.E. Qualitative image quality

Figure 11 shows planar sagittal images of a sheep head at2.35 MV with both carbon and aluminium targets and a 6 MVtherapeutic beam. Relatively little variation in image qualityis observed between the aluminium and carbon targets, whichis consistent with the observations above. Images with the2.35 MV beams show significant enhancement in imagequality compared to a 6 MV therapeutic beam in both boneand soft tissues. Roberts et al.8 showed similar gains in planarimage quality between their low-Z target beam line and a6 MV therapeutic beam in planar images of a head and neckphantom.

In summary, given (i) that the beam current is maximizedat 2.35 MV, thus shortening imaging times, (ii) that there isno measurable improvement in CNR compared to the lowestachievable energy, and (iii) that there is not significant conse-quence with regard to variation of spatial resolution, 2.35 MVappears to be the optimal selection for low-Z target imagingover the range explored here.

IV. CONCLUSIONS

In this work, we have shown that it is possible to decreasethe incident electron energy produced by a Varian 2100EXlinac to values ranging from 1.90 MeV to 2.35 MeV. Inelectron mode, beam current can be increased from approx-imately 600 MU/min to 2800 MU/min by altering set pointsof the electron gun, grid, solenoid, and bunching magnets,thereby deceasing image acquisition times. While it ispossible to lower the incident electron energy further, below

Medical Physics, Vol. 39, No. 7, July 2012

4578 D. Parsons and J. L. Robar: Beam generation below 2.40 MeV with low-z targets 4578

1.90 MeV, the beam current falls sharply to an impracticalvalue. Lowering the incident electron energy and increasingthe beam current has optimized the photon beams for imag-ing, yielding approximately 50% of generated photons withinthe diagnostic energy domain. This allowed an approximateincrease in CNR of cortical bone ranging from 6.2 to 7.4 andfrom 3.7 to 4.3 for a thin and a thick phantom, respectively,compared to 6 MV therapy beam for both aluminum andcarbon targets. Slight relative decreases of 3% and 10% at0.20 lp/mm and 0.40 lp/mm, respectively, were observedin spatial resolution when lowering from 2.35 MV to1.90 MV. Planar images of a sheep head showed that2.35 MV carbon and aluminum beams produced enhance-ments to image quality compared to a 6 MV therapeuticbeam, in both bone and soft tissues. Based on the observationsand constraints encountered in this study, we suggest that2.35 MV with a carbon target should be the normal operatingenergy for further low-Z imaging with this linac. Removalof the copper layer within the detector7, 9 and detectionwith a higher-efficiency receptor8 should allow for furtherincreases in image quality without further alteration of thebeam.

ACKNOWLEDGMENTS

The authors are grateful for support provided by VarianMedical Systems for the funding of this project and the Nat-ural Sciences and Engineering Research Council of Canadafor their additional financial support. The authors would liketo thank Robert Moran for his vital electronic and technicalsupport, Ian Porter for fabrication of apparatus, Dr. RobinKelly and Dr. Mammo Yewondwossen for their knowledgeand teaching of techniques utilized in this project.

1D. A. Jaffray, J. H. Siewerdsen, J. W. Wong, and A. A. Martinez, “Flat-panel cone-beam computed tomography for image-guided radiation ther-apy,” Int. J. Radiat. Oncol., Biol., Phys. 53, 1337–1349 (2002).

2D. M. Galbraith, “Low-energy imaging with high-energy bremsstrahlungbeams,” Med. Phys. 16, 734–746 (1989).

3D. W. Mah, D. M. Galbraith, and J. A. Rawlinson, “Low-energy imagingwith high-energy bremsstrahlung beams: Analysis and scatter reduction,”Med. Phys. 16, 653–665 (1993).

4O. Z. Ostapiak, P. F. O’Brien, and B. A. Faddegon, “Megavoltage imagingwith low Z targets: Implementation and characterization of an investiga-tional system,” Med. Phys. 25, 1910–1918 (1998).

5A. Tsechanski, A. F. Bielajew, S. Faermann, and Y. Krutman, “A thin targetapproach for portal imaging in medical accelerators,” Phys. Med. Biol. 43,2221–2236 (1998).

6S. Flampouri, P. M. Evans, F. Verhaegen, A. E. Nahum, E. Spezi, andM. Partridge, “Optimization of accelerator target and detector for portalimaging using Monte Carlo simulation and experiment,” Phys. Med. Biol.47, 3331–3349 (2002).

7B. A. Faddegon, V. Wu, J. Pouliot, B. Gangadharan, and A. Bani-Hashemi,“Low dose megavoltage cone beam computed tomography with an unflat-tened 4 MV beam from a carbon target,” Med. Phys. 35, 5777–5786 (2008).

8D. A. Roberts, V. N. Hansen, A. C. Niven, M. G. Thompson, J. Seco, andP. M. Evans, “A low Z linac and flat panel imager: Comparison with theconventional imaging approach,” Phys. Med. Biol. 53, 6305–6319 (2008).

9E. J. Orton and J. L. Robar, “Megavoltage image contrast with low-atomicnumber target materials and amorphous silicon electronic portal imagers,”Phys. Med. Biol. 54, 1275–1289 (2009).

10D. Sawkey, M. Lu, O. Morin, M. Aubin, S. S. Yom, A. R. Gottschalk,A. Bani-Hashemi, and B. A. Faddegon, “A diamond target for megavoltagecone-beam CT,” Med. Phys. 37, 1246–1253 (2010).

11J. L. Robar, T. Connell, W. Huang, and R. G. Kelly, “Megavoltage planarand cone-beam imaging with low-Z targets: Dependence of image qual-ity improvement on beam energy and patient separation,” Med. Phys. 36,3955–3963 (2009).

12T. Connell and J. L. Robar, “Low-Z target optimization for spatial res-olution improvement in megavoltage imaging,” Med. Phys. 37, 124–131(2010).

13S. Zhang, P. Liengsawangwong, P. Lindsay, K. Prado, T. Sun, R.Steadham, X. Wang, M. Salehpour, and M. Gillin, “Clinical implementa-tion of electron energy changes of varian linear accelerators,” J. Appl. Clin.Med. Phys. 10, 177–187 (2009).

14J. Wei, G. A. Sandison, and A. V. Chvetsov, “Reconstruction of electronspectra from depth doses with adaptive regularization,” Med. Phys. 33,354–359 (2006).

15J. O. Deasy, P. R. Almond, and M. T. McEllistrem, “Measured electronenergy and angular distributions from clinical accelerators,” Med. Phys.23, 675–684 (1996).

16B. W. Wessels, B. R. Paliwal, M. J. Parrott, and M. C. Choi, “Characteriza-tion of Clinac-18 electron-beam energy using a magnetic analysis method,”Med. Phys. 6, 45–48 (1979).

17D. W. Rogers, B. A. Faddegon, G. X. Ding, C. M. Ma, J. We, andT. R. Mackie, “BEAM: A Monte Carlo code to simulate radiotherapy treat-ment units,” Med. Phys. 22, 503–524 (1995).

18I. Kawrakow, “Accurate condensed history Monte Carlo simulation of elec-tron transport. I. EGSnrc, the new EGS4 version,” Med. Phys. 27, 485–498(2000).

19D. W. O. Rogers, B. Walters, and I. Kawrakow, BEAMnrc Users ManualNRCC Report PIRS-0509(A)revL (NRCC, Ottawa, Canada, 2011).

20C. M. Ma and D. W. O. Rogers, BEAMdp Users Manual NRCC ReportPIRS-0509(C)revA (NRCC, Ottawa, Canada, 2009).

21F. H. Attix, Introduction to Radiological Physics and Radiation Dosimetry(Wiley, 1986).

22R. Rajapakshe, K. Luchka, and S. Shalev, “A quality control test for elec-tronic portal imaging devices,” Med. Phys. 23, 1237–1244 (1996).

23R. T. Droege and R. L. Morin, “A practical method to measure the MTF ofCT scanners,” Med. Phys. 9, 758–760 (1982).

Medical Physics, Vol. 39, No. 7, July 2012