GafChromic EBT film dosimetry with flatbed CCD scanner: A novelbackground correction method and full dose uncertainty analysis

Sigrun Saura�

Department of Physics, The Norwegian University of Science and Technology, N-7491 Trondheim, Norwayand Department of Oncology and Radiotherapy, St. Olavs University Hospital,N-7006 Trondheim, Norway

Jomar FrengenDepartment of Oncology and Radiotherapy, St. Olavs University Hospital, N-7006 Trondheim, Norway

�Received 11 January 2008; revised 8 May 2008; accepted for publication 10 May 2008;published 13 June 2008�

Film dosimetry using radiochromic EBT film in combination with a flatbed charge coupled devicescanner is a useful method both for two-dimensional verification of intensity-modulated radiationtreatment plans and for general quality assurance of treatment planning systems and linear accel-erators. Unfortunately, the response over the scanner area is nonuniform, and when not correctedfor, this results in a systematic error in the measured dose which is both dose and position depen-dent. In this study a novel method for background correction is presented. The method is based onthe subtraction of a correction matrix, a matrix that is based on scans of films that are irradiated tonine dose levels in the range 0.08–2.93 Gy. Because the response of the film is dependent on thefilm’s orientation with respect to the scanner, correction matrices for both landscape oriented andportrait oriented scans were made. In addition to the background correction method, a full doseuncertainty analysis of the film dosimetry procedure was performed. This analysis takes into ac-count the fit uncertainty of the calibration curve, the variation in response for different film sheets,the nonuniformity after background correction, and the noise in the scanned films. The film analysiswas performed for film pieces of size 16�16 cm, all with the same lot number, and all irradiationswere done perpendicular onto the films. The results show that the 2-sigma dose uncertainty at 2 Gyis about 5% and 3.5% for landscape and portrait scans, respectively. The uncertainty graduallyincreases as the dose decreases, but at 1 Gy the 2-sigma dose uncertainty is still as good as 6% and4% for landscape and portrait scans, respectively. The study shows that film dosimetry usingGafChromic® EBT film, an Epson Expression 1680 Professional scanner and a dedicated back-ground correction technique gives precise and accurate results. For the purpose of dosimetricverification, the calculated dose distribution can be compared with the film-measured dose distri-bution using a dose constraint of 4% �relative to the measured dose� for doses between 1 and 3 Gy.At lower doses, the dose constraint must be relaxed. © 2008 American Association of Physicists inMedicine. �DOI: 10.1118/1.2938522�

Key words: film dosimetry, flatbed scanner, background correction, dose uncertainty

I. INTRODUCTION

The Gafchromic® EBT film �International Specialty Prod-ucts, Wayne, NJ� is a promising tool for precise and accuratetwo-dimensional �2D� dosimetry with high spatial resolutionand response within clinical dose levels. The atomic compo-sition is near-tissue equivalent13 and the energy dependencyis very weak,1–3,13 making the dose response insensitive tochanges in field size3,4 and depth.3 In addition the dose re-sponse is independent of the dose rate.3,5 These propertiesmake the EBT film suitable for quality assurance of treat-ment planning systems and linear accelerators in general andverification of intensity modulated radiation therapy in par-ticular. The film is self-developing and insensitive to roomlight, which is an advantage during the quality control.

When scanning the EBT film with a flatbed chargecoupled device �CCD� scanner, several studies have shown anonuniformity effect in the direction of the CCD array.3,6–9

This effect is caused by different light scattering conditions

3094 Med. Phys. 35 „7…, July 2008 0094-2405/2008/35„7…/

due to a finite light source in addition to other scanner-dependent factors.6 A less pronounced nonuniformity effectis also seen in the direction orthogonal to the CCD array.6,7

The magnitude of these nonuniformity effects makes it im-portant to establish a 2D correction method. Earlier studies6,9

have shown that the relative correction needed is dose de-pendent: a greater relative correction is needed when thedose is increased. A correction method that works in thewhole range of relevant dose levels would have simplifiedthe correction procedure. Hence, a novel correction methodwas established, based on the subtraction of a correction ma-trix, and the method is presented and evaluated in this article.

The manufacturer of the EBT film points out the impor-tance of maintaining the same orientation of the films duringscanning. The active component in the EBT film is a needle-like polymer, and the polymers will tend to align parallelwith the coating direction of the film, which is parallel to the

short sides of the EBT film sheets, as illustrated in Fig. 1.

30943094/8/$23.00 © 2008 Am. Assoc. Phys. Med.

3095 S. Saur and J. Frengen: Uncertainty analysis of EBT film dosimetry 3095

The alignment of the active particles results in anisotropiclight scattering during scanning, and thereby an orientationdependent film response. As a standard practice the manufac-turer recommends to scan the films with landscape orienta-tion �the original short dimension of the film parallel to thescanning direction� instead of portrait orientation �the origi-nal long dimension of the film parallel to the scanning direc-tion�, see Fig. 1.

Before a verification of radiotherapy treatment plans canbe done, it is important to establish the uncertainties relatedto the measurement method, which for film dosimetry will beaffected by the scanning and image analysis procedure. Thetotal dose uncertainty for both scan directions, including thefit uncertainty of the calibration curve, the variation in filmresponse for different film sheets, the uniformity within anarea of 16�16 cm after the use of the established correctionmethod, and the noise in the scanned films was determinedexperimentally for films with the same lot number. A com-parison with a more standard correction method, based on anunirradiated film, was also performed.

II. MATERIALS AND METHODS

The dosimetric precision of Gafchromic® EBT film incombination with an Epson Expression 1680 Professionalscanner �Seiko Epson Corporation, Nagano, Japan� wasevaluated. Film pieces of various sizes were irradiated with 6MV photons using an Elekta Synergy linear accelerator �Ele-kta AB, Stockholm, Sweden�. The film pieces were placed ina SolidWater slab phantom at 10 cm depth and with a source-surface-distance of 90 cm. All irradiations were performedperpendicularly to the film plane. Unless otherwise stated, afield size of 10 cm�10 cm at the film plane was used, re-sulting in a so-called standard geometry. When another field

FIG. 1. Illustrations of the scanner-film system with abbreviations used inthe text. �a� The scanner and its coordinate system. �b� The scanner maskwith a cutout of size 16�16 cm. The mask was made of black plastic. Toavoid contact between the film and the scanner’s glass plate, the mask hadsome small pins for the film to rest on. �c� The EBT film �20.3�25.4 cm� and its coating direction. The needle-like active components inthe film are visualized to illustrate that they will align parallel to the coatingdirection of the film. The black dotted square illustrates a 16�16 cm filmpiece. �d� The film response will be dependent on the films’ original orien-tation with respect to the scanner. Scanning the film with landscape orien-tation means that the original short dimension of the film �same as coatingdirection� is parallel to the scanning direction, while scanning the film withportrait orientations means that the original long dimension of the film isparallel to the scanning direction �resulting in coating direction perpendicu-lar to the scanning direction�.

size was used, the established field size factor of the linac

Medical Physics, Vol. 35, No. 7, July 2008

was used to obtain the dose values. A perfectly calibratedlinac was assumed, i.e., no correction for the daily variationof the linac’s output was made. To allow for the postirradia-tion polymerization to be completed, scanning was per-formed at least 6 h after irradiation. All films within the sameexperiment were irradiated at the same time and then storedin light tight envelopes before they were scanned. The sametime interval between irradiation and scanning was main-tained for films within the same experiment.

Since the scanner’s response �i.e., the measurement of thelight transmission� is dependent on the scanning bed positionwith the central part of the transparency reading area of thescanner being most homogenous, a mask was made, cover-ing the entire transparency reading area, see Fig. 1. Thismask had a 16�16 cm cutout, and this cutout was centrallyplaced to achieve the best possible homogeneity. The maskdid also ensure reproducible positioning of the films duringscanning. Since contact between the film and the scanner’sglass plate may produce image artifacts in some instances,the mask had some tags for the film to rest on. The cutout inthe mask corresponds to the scanning area in this analysis.The choice of a film size and a scanning area of 16�16 cm were made because it corresponds to the size of ourverification phantom �OmniPro I’mRT, Scanditronix Well-höfer AB, Uppsala, Sweden�.

II.A. Image processing and analysis

The films were scanned with the software package “Epsonscan,” which was used in professional mode and with allimaging adjustment options turned off, with a resolution of127 dpi �0.2 mm/pixel�. Transmission images were saved in48 bit RGB uncompressed tagged image file format �TIFF�.The scanning area provided images of 801�801 pixels�160.3�160.3 mm�. The images were imported into IDL6.3 �ITT Visual Information Solutions, Boulder, CO� foranalysis. The EBT film has absorption maxima in the redregion of the visible spectrum,13 consequently the responseof the film will be enhanced by measurements with red light.Therefore, the transmission values from the red color chan-nel were extracted, resulting in 16 bits images with pixelvalues in the range 0–65535.

To reduce the scan-to-scan uncertainty, every film wasscanned successively five times, and the average of the lastthree scans was used for further analysis. The first two scanswere not used because they have different response than theother scans due to the warming up effects of the scannerlamp �the lamp is only turned on during scanning�. This pro-cedure is adapted from Paelinck et al.8 and ensures reproduc-ible scanning conditions for all films being scanned. Theaveraging will also reduce noise caused by the scanner andthereby reduce the overall measurement uncertainty. The ef-fect of this procedure is not evaluated in this study, but Devicet al.10 have analyzed the effect of averaging over five con-secutive scans and found a great improvement in the doseuncertainty.

The films were median filtered using a filter width of 5

pixels before the image resolution was reduced from 5

3096 S. Saur and J. Frengen: Uncertainty analysis of EBT film dosimetry 3096

pixels/mm �127 dpi� to 1 pixel/mm �25.4 dpi� using the con-grid function in IDL. A resolution of 1 mm/pixel correspondsto the maximal dose matrix resolution in our department’streatment planning system. No attempt was made to trans-form the pixel values into optical density.

II.B. Background correction

Since the relative correction needed to achieve uniformscan readout increases with dose level,6,9 and the fact thattransmission values decrease with dose, it is apparent that acorrection method based on the subtraction of absolute pixelvalues will be less dependent on dose.

To evaluate the possible dose-dependent nonuniformitywithin the scanning area, ten film pieces, each 3�3 cm,from the same sheet of film, were irradiated to 8, 34, 58, 87,116, 154, 193, 231, and 293 cGy, one was left unexposed.The film pieces were all scanned in 25 regularly spaced po-sitions within the scanning area. A transparent mask fittingthe scanning area was used. This mask had 25 cutouts of 2�2 cm, and the film pieces were placed onto this maskduring scanning to ensure reproducible positioning of thefilm pieces and to avoid contact with the scanner’s glassplate. All film pieces were scanned at all 25 positions, andthe average pixel values from a region of 3�3 mm at each

(a)

(b)

FIG. 2. Point deviations across the central part of the scanning area �lines are�i.e., along the scanner’s x axis� for landscape and portrait scans, respectivescanner’s y axis� for landscape �upper graph� and portrait scans �lower grapthree repeated experiments �different films�.

of these positions were found. The absolute difference in

Medical Physics, Vol. 35, No. 7, July 2008

pixel values from the central film position was found andused to establish a “difference map” for each dose level.

It has been shown that the optical density for EBT filmincreases when the number of scans increases.8 Therefore, incontradistinction to the other experiments in this study, thefilms were scanned only once in each position. Not averag-ing over several scans will influence the precision in themeasurement, but this effect was reduced somewhat by per-forming the experiment three times. That means that all 25positions had three measurements, coming from three differ-ent film sheets �but the same lot number�, at each dose leveland for both scanning orientations.

II.C. Film-scanner characterization

The uncertainties related to film-to-film-variation, nonuni-formity, and noise were experimentally estimated. In addi-tion, the fit uncertainty related to the choice of fitting curvefor calibration was evaluated. The resulting total uncertainty,expressed in standard deviation, in measurements using EBTfilm was found by the following equation:

�total�d� = ��fit2 �d� + �film-film

2 �d� + �uniformity2 �d� + �noise

2 �d� ,

�1�

and was calculated based on experimentally determined es-

c)

d)

d for clarity�. �a� and �b� are plots in the direction parallel to the CCD array� are plots in the direction perpendicular to the CCD array �i.e., along the�d� the symbols are explained. The value at each point is the average from

(

(

addely. �c

h�. In

timators of the different sources of uncertainty.

3097 S. Saur and J. Frengen: Uncertainty analysis of EBT film dosimetry 3097

The relative 2-sigma dose uncertainty �95% confidencelevel� in the measurement of an unknown dose was calcu-lated as

�Drelative95% �d� =

1

d·

2 · ��d���d�

· 100% , �2�

where � is the standard deviation �in pixel values�, d is thedose �in Gy� and � is the dose response �pixel values per Gy�at the corresponding dose level. The dose response wasfound by computing the first derivatives of the fit functionsat the respective dose levels.

II.C.1. Film response and film-to-film variation

Three film sheets with the same lot number were each cutinto ten pieces of size 4�4 cm. The pieces from each filmsheet were irradiated to 8, 34, 58, 87, 116, 154, 193, 231, and293 cGy, one was left unexposed. During scanning, the filmpieces were placed in the central position of the scanning

(a)

(b)

FIG. 3. Average difference matrices for films irradiated to doses between 8and 293 cGy when scanning the films with �a� landscape orientation and �b�portrait orientation. Each gray level denotes an interval of 200 pixel valuescorresponding to the values on the z axis of the graphs. X and y relates to thescanner’s coordinate system and the origin is in the middle of the scanningarea �see Fig. 1�.

area �in this case no background correction was needed�,

Medical Physics, Vol. 35, No. 7, July 2008

both with portrait and landscape orientation. To avoid contactwith the scanner’s glass plate, a transparent mask fitting thescanning area was used. This mask had a central cutout ofabout 3�3 cm. The average pixel values �from three films�from regions of interest �ROIs� consisting of 10�10 pixels�1.0�1.0 cm� were then found, and these values were plot-ted as a function of the delivered dose. For both portrait andlandscape scans three functions were fitted to the averageROI values using Prism 5 �GraphPad Software, Inc., SanDiego, CA�; a one-phase exponential decay function, a two-phase exponential decay function, and a fourth order polyno-mial function. Each fit was based on minimizing the sum ofsquares �SS� of the deviations between the data points andthe fitted curve. To evaluate which fitting function is bestsuitable for calibration purpose, the standard deviation be-tween the observed and curve-fitted values was found bycalculating the root mean square for each fit, which was usedas an estimate for the fitting uncertainty, �fit,

�fit =�SS

df. �3�

Here df is the degrees of freedom, which is equal to thenumber of points �dose levels� on the calibration curve minusthe number of parameters in the fit.

The standard deviation of the average ROI values at eachdose level, �film-film�d�, was calculated and used as an esti-mate of the intralot film-to-film variation �this estimate willalso include possible scan-to-scan variation�.

II.C.2. Uniformity and noise

Nine films of size 16�16 cm and with the same lot num-ber were irradiated to 8, 34, 58, 87, 116, 154, 193, 231, and293 cGy using a field size of 20�20 cm at the film plane.The uniformity of the radiation field was determined usingan ionization chamber �IC70 Farmer chamber, ScanditronixWellhöfer AB, Uppsala, Sweden� by measuring the dose atpoints separated by 4 cm, covering the 16�16 cm scanningarea with 16 measurement points. The chamber measure-ments were made just before the films were irradiated.

After scanning, background correction of these uniformlyirradiated films using the average difference matrices �seeSecs. II B and III A�, denoted d�i , j�, was performed on apixel-by-pixel basis by subtraction, hereafter called absolutecorrection

pirradiatedcorr,absolute�i, j� = pirradiated�i, j� − d�i, j� . �4�

Then the pixel values from 16 different ROIs �each 10�10 mm and with a position corresponding to the cambermeasurements� regularly distributed on the films was found.Any deviation from a nonuniform radiation field as measuredwith the ion chamber was corrected for by adding/subtracting the corresponding change in pixel values beforethe uniformity was evaluated. Uniformity was defined as thestandard deviation, �uniformity�d�, of the average pixel valuesfrom these 16 ROIs and was investigated for each dose level.

For comparison with a more standard background correc-

tion method, the same uniformity analysis was performed for

3098 S. Saur and J. Frengen: Uncertainty analysis of EBT film dosimetry 3098

the films when corrected with correction matrices establishedby using unirradiated films. After scanning, the unirradiatedfilms were median filtered with a filter width of 5 mm andthen divided by the mean transmission value in the centralregion of the film, producing matrices with correction factorsin two dimensions, denoted k�i , j�. The correction of the ir-radiated films was thereby done on a pixel-by-pixel basis bydivision, hereafter called relative correction

pirradiatedcorr,relative�i, j� =

pirradiated�i, j�k�i, j�

. �5�

To estimate the influence of noise, the average standard de-viation, �noise�d�, of the ROI pixel values from the 16 posi-tions on each film were found for each dose level.

III. RESULTS

III.A. Background correction

The signal variation within the scanning area in the direc-tion parallel to the CCD array �defining the x axis of thescanner� and in the direction perpendicular to the CCD array�defining the y axis of the scanner� in terms of deviationfrom the central value is shown in Fig. 2. The systematicchange �off-center deviation� seen along the x axis for por-trait scans is less pronounced for landscape scans. Whenscanning with portrait orientation, the deviation from thecentral point seems to increase somewhat with increasingdose, but this trend is not systematic. No such dose-dependent trend is seen for landscape scans. Along the y axisthere is no dose-dependent trend for any of the two scanorientations. The variation is somewhat greater for landscapescans, but in all, the deviations along the scanner’s y axis aresmaller than along the scanner’s x axis. A correction basedon the subtraction of the mean deviation from the centralvalue seems appropriate for both landscape and portraitscans.

Figure 3 shows the dose-averaged �i.e., average for alldose levels� difference matrices resulting from scanning withlandscape and portrait orientation small pieces of irradiatedfilms at 25 positions regularly spaced on the scanning area

TABLE I. Fit functions for films scanned with landscapthe fitting to nine calibration points, the df, and the reuncertainty, �fit, is also tabulated.

Function Fit

Land

One-phase decay 22 617+29Two-phase decay 19 604+4915·e−0.0

Fourth order polynomial 52471−260.1·d+1.35

Por

One-phase decay 21 307+29Two-phase decay 17 690+6442·e−0.0

Fourth order polynomial 51 095−269.2·d+1.41

�each film scanned once at each position�. The values are

Medical Physics, Vol. 35, No. 7, July 2008

based on three repeated experiments at each dose level. Thenonirradiated film was not included in the average differencematrices because the 0 Gy difference matrices differed some-what from the irradiated ones. Both Figs. 2 and 3 clearlydemonstrate that scanning the films with landscape orienta-tion results in a more uniform scan than scanning with por-trait orientation.

Using bilinear interpolation between the points, these ma-trices were resized to 801�801 pixels and used as absolutecorrection matrices in the uniformity evaluation described inthe preceding section.

III.B. Film-scanner characterization

III.B.1. Film response

Calibration curves from scanning the films with landscapeand portrait orientation are presented in Fig. 4. Three func-tions were fitted to the data points, and Table I summarizesthe fit functions and their corresponding uncertainty for land-scape and portrait scans. The pixel value of unexposed filmwas omitted when fitting a function to the calibration points,since it caused a substantial increase in the fit uncertainty.

portrait orientation, d is the dose in cGy. The SS forg root mean square used as an estimate for the fitting

tion SS df �fit

scans

·e−0.007 531·d 624 390 6 323+28 226·e−0.005 339·d 7599 4 440.0043·d3+5 ·10−6 ·d4 33 866 4 92

cans

·e−0.007 867·d 691 186 6 339+27 197·e−0.005 069·d 17 981 4 67−0.0044·d3+5 ·10−6 ·d4 41 501 4 101

FIG. 4. EBT film response when scanned with landscape and portrait orien-tation at an Epson Expression 1680 Professional scanner. Pixel values aretaken from the red color channel in a 48 bit RGB TIFF file. A two-phase

e andsultin

func

scape

25029 76·d

9·d2−

trait s

18225 14·d

9·d2

exponential decay function is fitted to the data points.

3099 S. Saur and J. Frengen: Uncertainty analysis of EBT film dosimetry 3099

A two-phase exponential decay function showed to havethe best fit to the data points and was therefore used in thedose resolution analysis �Sec. III B 5�.

III.B.2. Film-to-film-variation

The standard deviation between films irradiated to thesame dose appeared to be dependent on dose level; the stan-dard deviation �expressed in pixel values� decreased whenthe dose increased, as shown in Fig. 5. This can be explainedby the fact that the transmission value decreases as the doseincreases. The standard deviations were greater for landscapethan portrait scans.

III.B.3. Uniformity

The uniformity within uniformly irradiated films after ab-solute correction �Eq. �4��, expressed as the standard devia-tion of the average pixel value from 16 regularly positionedpoints on the scanning area, are presented in Fig. 5. Thevalues in Fig. 5 also include possible nonhomogeneitieswithin the films. No dose-dependent trend was found, andthe values for landscape and portrait scans were almost thesame. The same analysis was also performed for scans cor-

(a)

(b)

FIG. 5. Uncertainty estimates for three of the sources of uncertainty in EBTfilm dosimetry, expressed as the standard deviation �in pixel values� for �a�landscape scans and �b� portrait scans. For the evaluation of uniformity, datafor both the absolute and the relative correction method are presented. Thelines connecting the data points are added for clarity.

rected with the relative correction method described by Eq.

Medical Physics, Vol. 35, No. 7, July 2008

�5�, and the values are included in Fig. 5 for comparison. Forlandscape scans, there was almost no difference between thetwo correction methods �Fig. 5�a��, but for portrait scansthere was a great difference �Fig. 5�b��, in favor of the abso-lute correction method. Although not shown in any figure,the difference between these two correction methods corre-sponds to a reduction of the total dose uncertainty for portraitscans from approximately 6% to approximately 4% for dosesbetween 1 and 3 Gy.

III.B.4. Noise

The standard deviations �expressed in transmission val-ues� within ROIs of uniform dose was virtually constant forthe doses investigated, as shown in Fig. 5. The values weresimilar for landscape and portrait scans.

III.B.5. Total uncertainty

How the different uncertainty estimates translate into doseuncertainty by using Eq. �2� is shown in Fig. 6. By addingthe sources of uncertainty according to Eq. �1�, the resultingtotal dose uncertainty was found by using Eq. �2�, and this isalso shown in Fig. 6. The values are based on a two-phase

(a)

(b)

FIG. 6. Total 2-sigma dose uncertainty �95% confidence level� in EBT filmdosimetry for films scanned with �a� landscape orientation and �b� portraitorientation, as well as its components. The absolute method for uniformitycorrection is used. The lines connecting the data points are added for clarity.

exponential decay fit to the calibration data points and using

3100 S. Saur and J. Frengen: Uncertainty analysis of EBT film dosimetry 3100

the absolute correction method. At 2 Gy, the total 2-sigmadose uncertainty is about 5% for landscape scans and 3.5%for portrait scans. At lower doses, the dose uncertainty in-creases, and portrait mode seems to be the better choice forall dose levels investigated. This is caused by a greater film-to-film-variation for landscape scans than for portrait scans�as shown in Figs. 5 and 6�.

IV. DISCUSSION

The absolute nonuniformity correction method was shownto be superior to the relative correction method for portraitscans, but not for landscape scans. The reason for this is thatportrait scans have a much more pronounced and systematicdeviation in the direction of the CCD array than landscapescans. Landscape scans can therefore be properly correctedby using an unirradiated film as a relative correction matrix,while portrait scans demands a more sophisticated correctionmethod to achieve an acceptable level of accuracy. The studyhas shown that when portrait scans are corrected with theabsolute correction method, portrait scans has the lower doseresolution compared to landscape scans.

Although the subtraction matrices to some extent are in-fluenced by the relatively large number of scans �each filmpiece was scanned 25 times to obtain the matrices�, the use-fulness of such a correction was evaluated, which means thatthis source of error is included in the analysis. The effect of25 scans on the transmission value was tested for films irra-diated to 1 and 2 Gy, and was found to decrease by approxi-mately 200 and 300 pixel units from the first to the last scanfor portrait and landscape scans, respectively �data notshown�. This will cause a systematic error in the difference-maps at each dose level, but since care was taken to ensurethat the film pieces irradiated to different dose levels werescanned in different order at the 25 scan positions, this sys-tematic effect will average out for the dose-averaged differ-ence matrices.

The absolute uniformity correction in this article is basedon a dose-averaged subtraction matrix resulting from ninedifferent dose levels. However, the results have shown thatthe deviations from the central transmission value on thescanner are not systematically dependent on dose. Therefore,the subtraction matrix may equally well be based on a single�or a few� dose level�s�, preferably the dose level�s� forwhich one aims to minimize the dose uncertainty.

When using a two-phase exponential decay fit to the cali-bration data, the major source to dose uncertainty is the film-to-film-variation. The film-to-film-variation observed here isin correspondence with the results from Fuss et al.3 The un-certainty originating from nonuniformity �after using abso-lute correction for portrait scans� and noise is low and quitestable for doses between 0.3 and 3 Gy. In all, we found thatthe noise is the minor contributor to the dose uncertainty inEBT film dosimetry when using our scanning and imageanalysis procedures.

Using the method described in this study, the total2-sigma dose uncertainty is about 6% and 4% at 1 Gy and

5% and 3.5% at 2 Gy, for landscape and portrait scans, re-

Medical Physics, Vol. 35, No. 7, July 2008

spectively. The main reason for the better precision for por-trait scans than landscape scans is the lower film-to-filmvariation. The precision of EBT film dosimetry has also beeninvestigated by Devic et al.,10 who found a 1-sigma doseuncertainty below 2% for doses above 0.4 Gy. Taking intoaccount that our study, in contrast to the work by Devic etal.,10 also includes film-to-film-variation and uniformitywithin a 16�16 cm scan area, the results may be consideredto be in agreement, although different scanning and imageanalysis procedures have been used.

This study has shown that accurate dosimetry can beachieved with the EBT film and a flatbed CCD scanner usinga relatively simple procedure. However, since the film-to-film variation is the major source to the dose uncertainty,further reduction of the dose uncertainty may be possible byirradiating multiple films, including the films used for cali-bration, and average the outcome. When the uniformity ofthe scanner is concerned, an improvement of the uniformityfor both scan orientations might be achieved with the use ofthe central area of an A3 document scanner by obtainingmore homogeneous light scattering conditions.

V. CONCLUSION

The study has shown that the nonuniformity effect alongthe detector array in radiochromic film dosimetry using aCCD-based document scanner can be properly corrected forby using one absolute correction matrix irrespective of doselevel. For portrait scans this absolute correction method issuperior to the conventionally recommended method of rela-tive correction using an unirradiated film’s response. Forlandscape scans the nonuniformity was found to be less pro-nounced and the need for correction was not as crucial as forportrait scans. As a consequence, there was negligible differ-ence between the two correction methods for this scanningorientation. Even though the nonuniformity was a minorproblem for landscape scans compared to portrait scans, theoverall dose resolution was superior by the use of portraitscanning with absolute correction due to the difference infilm-to-film variation for the two scanning orientations.Curve fitting using a two-phase exponential decay functionwas found to be superior compared to a one-phase decay or afourth order polynomial function.

With the total 2-sigma dose uncertainty estimated to bewithin 4% for doses between 1 and 3 Gy and reaching aminimum around 2 Gy, the usefulness for absolute dose veri-fication in the clinic is documented. Using the gammamethod11,12 to verify a treatment plan with a fraction dose of2 Gy, a dose constraint of 3.5% �relative to the measureddose� are appropriate at this dose level. However, the doseconstraint should be relaxed according to the actual doseuncertainty at other dose levels.

ACKNOWLEDGMENTS

The authors would like to thank Professor Tore Lindmo at

the Norwegian University of Science and Technology for

3101 S. Saur and J. Frengen: Uncertainty analysis of EBT film dosimetry 3101

reading the manuscript and giving them valuable feedback inthe writing process. The comments of the reviewers are alsoacknowledged.

a�Electronic mail: sigrun.saur@ntnu.no1S.-T. Chiu-Tsao, Y. Ho, and R. Shankar, “Energy dependence of responseof new high sensitivity radiochromic films for megavoltage and kilovolt-age radiation energies,” Med. Phys. 32, 3350–3354 �2005�.

2M. J. Butson, T. Cheung, and P. K. Yu, “Weak energy dependence of EBTgafchromic film dose response in the 50 kVp–10 MVp x-ray range,”Appl. Radiat. Isot. 64, 60–62 �2006�.

3M. Fuss, E. Sturtewagen, and C. De Wagter, “Dosimetric characterizationof GafChromic EBT film and its implication on film dosimetry qualityassurance,” Phys. Med. Biol. 52, 4211–4225 �2007�.

4T. Cheung, M. J. Butson, and P. K. N. Yu, “Independence of calibrationcurves for EBT Gafchromic films of the size of high-energy x-ray fields,”Appl. Radiat. Isot. 64, 1027–1030 �2006�.

5A. Rink, I. A. Vitkin, and D. A. Jaffray, “Characterization and real-timeoptical measurement of the ionizing radiation dose response for a newradiochromic medium,” Med. Phys. 32, 2510–2516 �2005�.

Medical Physics, Vol. 35, No. 7, July 2008

6S. Devic, Y.-Z. Wang, and N. Tomic, “Sensitivity of linear CCD arraybased film scanners used for film dosimetry,” Med. Phys. 33, 3993–3996�2006�.

7B. D. Lynch, J. Kozelka, and M. K. Ranade, “Important considerationsfor radiochromic film dosimetry with flatbed CCD scanners and EBTGAFCHROMIC film,” Med. Phys. 33, 4551–4556 �2006�.

8L. Paelinck, W. De Neve, and C. De Wagter, “Precautions and strategiesin using a commercial flatbed scanner for radiochromic film dosimetry,”Phys. Med. Biol. 52, 231–242 �2007�.

9C. Fiandra, U. Ricardi, and R. Ragona, “Clinical use of EBT model Caf-chromic film in radiotherapy,” Med. Phys. 33, 4314–4319 �2006�.

10S. Devic, J. Seuntjens, and E. Sham, “Precise radiochromic film dosime-try using a flat-bed document scanner,” Med. Phys. 32, 2245–2253�2005�.

11D. A. Low and J. F. Dempsey, “Evaluation of the gamma dose distribu-tion comparison method,” Med. Phys. 30, 2455–2464 �2003�.

12T. Depuydt, A. Van Esch, and D. P. Huyskens, “A quantitative evaluationof IMRT dose distributions: refinement and clinical assesment of thegamma evaluation,” Radiother. Oncol. 62, 309–319 �2002�.

13Product manual at http://www.ispcorp.com/products/dosimetry/content/gafchromic/index.html.