Radiotherapy and Oncology 96 (2010) 172–177

Contents lists available at ScienceDirect

Radiotherapy and Oncology

journal homepage: www.thegreenjournal .com

Prostate IMRT

Evaluation of the ‘dose of the day’ for IMRT prostate cancer patients derivedfrom portal dose measurements and cone-beam CT

Mathilda van Zijtveld, Maarten Dirkx *, Marcel Breuers, Ruud Kuipers, Ben HeijmenDepartment of Radiation Oncology, Erasmus MC-Daniel den Hoed Cancer Center, Rotterdam, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history:Received 3 August 2009Received in revised form 7 May 2010Accepted 17 May 2010Available online 25 June 2010

Keywords:Dose reconstructionIn-vivo dosimetryCone-beam CTEPIDProstate cancerIMRT QA

0167-8140/$ - see front matter � 2010 Elsevier Irelandoi:10.1016/j.radonc.2010.05.015

* Corresponding author. Address: Department of RaMedical Physics, Erasmus MC-Daniel den Hoed Cancer3075 EA Rotterdam, The Netherlands.

E-mail address: m.dirkx@erasmusmc.nl (M. Dirkx)

Purpose: High geometrical and dosimetrical accuracies are required for radiotherapy treatments whereIMRT is applied in combination with narrow treatment margins in order to minimize dose delivery tonormal tissues. As an overall check, we implemented a method for reconstruction of the actually deliv-ered 3D dose distribution to the patient during a treatment fraction, i.e., the ‘dose of the day’. In this arti-cle results on the clinical evaluation of this concept for a group of IMRT prostate cancer patients arepresented.Materials and methods: The actual IMRT fluence maps delivered to a patient were derived from measuredEPID-images acquired during treatment using a previously described iterative method. In addition, thepatient geometry was obtained from in-room acquired cone-beam CT images. For dose calculation, amapping of the Hounsfield Units from the planning CT was applied. With the fluence maps and the mod-ified cone-beam CT the ‘dose of the day’ was calculated. The method was validated using phantom mea-surements and evaluated clinically for 10 prostate cancer patients in 4 or 5 fractions.Results: The phantom measurements showed that the delivered dose could be reconstructed within 3%/3 mm accuracy. For prostate cancer patients, the isocenter dose agreed within �0.4 ± 1.0% (1 SD) with theplanned value, while for on average 98.1% of the pixels within the 50% isodose surface the actually deliv-ered dose agreed within 3% or 3 mm with the planned dose. For most fractions, the dose coverage of theprostate volume was slightly deteriorated which was caused by small prostate rotations and small inac-curacies in fluence delivery. The dose that was delivered to the rectum remained within the constraintsused during planning. However, for two patients a large degrading of the dose delivery was observed intwo fractions. For one patient this was related to changes in rectum filling with respect to the planning CTand for the other to large intra-fraction motion during treatment delivery, resulting in mean underdosag-es of 16% in the prostate volume.Conclusions: A method to accurately assess the ‘dose of the day’ was evaluated for prostate cancerpatients treated with IMRT. To correct for observed dose deviations off-line dose-adaptive strategies willbe developed.

� 2010 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology 96 (2010) 172–177

In recent years, new techniques like Intensity Modulated Radio-therapy (IMRT) and Image Guided Radiotherapy have been intro-duced to tailor dose distributions more tightly to the targetvolume, while simultaneously minimizing the dose to organs atrisk. These techniques have been used to escalate the tumour dosein order to increase the tumour control rate or to reduce normaltissue complications. However, due to the increasing complexityof these techniques, the steeper dose gradients outside the targetvolume and the smaller margins that are often applied to accountfor positioning errors, more elaborate quality assurance (QA)methods are mandatory for safe application.

d Ltd. All rights reserved.

diation Oncology, Division ofCenter, Groene Hilledijk 301,

.

For verification of the treatment delivery, existing QA proce-dures focus both on image guidance to ensure accurate patientset-up and monitor changes in patient geometry during treatment[1–3], and on dosimetric verification of the dose delivery to the pa-tient [4–7]. Ultimately, as the most complete and clinically rele-vant check, these two aspects could be combined into oneprocedure by calculating the actual 3D dose distribution receivedby a patient during a treatment fraction, i.e., the ‘dose of theday’. When performing this procedure daily, the accumulated dosedelivered to the tumour and organs at risk could be derived, allow-ing dose-guided radiotherapy.

In the past years, several methods to use electronic portal imag-ing devices (EPIDs) for verification of dose delivery during patienttreatment have been developed and implemented [4–7]. EPIDswere shown to have clear advantages over more traditional deviceslike TLD, diodes and MOSFETs in terms of high resolution in 2D,

M. van Zijtveld et al. / Radiotherapy and Oncology 96 (2010) 172–177 173

acquisition of digital data in a short measurement time and highmeasurement accuracy [8–10].

In our clinic, IMRT QA is already being performed for over sevenyears using a Theraview NT electronic portal imaging device (Cab-lon Medical Theraview Technology, Leusden, The Netherlands).This CCD-camera based system has a standard fluorescent layerfor conversion of high-energy photons into visible light [8,10].The EPID-based QA includes daily verification of the output, beamflatness and leaf calibration of the linac, as well as patient-specificmeasurements, both pre-treatment (i.e., without patient in thebeam) [11] and more recently also in-vivo (i.e., during patienttreatment) [5]. In contrast with other approaches for EPID dosim-etry, our method does not rely on relative or absolute dose mea-surements with an ionization chamber or detector array, but isfully calibrated using EPID measurements only [5,12].

To allow a more clinically relevant evaluation of 2D in-vivomeasurements, a method was recently developed to reconstructthe actually delivered 3D patient dose from measured portal doseimages and a Hounsfield Units (HU) corrected cone-beam CT(CBCT) [13] acquired during treatment delivery. Other groups havebeen working on a similar approach [14–16], but till now resultson the clinical evaluation of the ‘dose of the day’ for more thanone or a few example patients are scarce. McDermott presenteddata for nine rectum cancer patients treated with a hypofractionat-ed IMRT treatment [14] and van Elmpt compared planned andreconstructed dose distributions for four lung cancer patients (intotal 10 fractions) treated with stereotactic body radiotherapy [16].

In this article, our method to reconstruct the delivered doseduring treatment is applied to a group of 10 prostate cancer pa-tients with implanted markers. Deviations with respect to theplanned treatment are discussed.

Methods and materials

Reconstruction of the actually delivered 3D dose distribution toa patient in a certain treatment fraction consists of two main parts:(1) derivation of the fluence that was delivered for each beamdirection during treatment and (2) definition of the patient geom-etry at the time of treatment using a CBCT scan. For each treatmentbeam the delivered fluence is obtained by comparing the portaldose image (PDI) measured during treatment with the correspond-ing predicted PDI. To eliminate the impact of changes in patientgeometry or absorbing parts of the treatment couch, the Split IMRTField Technique (SIFT) is used [17].

Measurement and prediction of PDIs

To derive a PDI from a measured electronic portal image (EPI)the cross-talk is first removed by deconvolution with position-dependent cross-talk kernels [18]. The pixel values are then con-verted into absolute portal doses by multiplication with a calibra-tion factor. During commissioning of the EPID for dosimetry thisfactor is derived from the on-axis gray-level in an EPI of a symmet-ric 10 � 10 cm2 field delivered with 100 MU.

To predict the PDI behind a patient (or phantom), the portaldose contributions due to transmission of primary radiationthrough the patient and those due to scatter from the patient(S(x, y)) are summed [5]. The primary component is obtained bymultiplying the predicted PDI without the patient in the beam(PDI0(x, y)) with the primary transmission through the patient(Tp(x, y)). Tp(x, y) and S(x, y) are derived using a transmission modeland an equivalent homogeneous phantom (EHP) of the patientgeometry [5,19]. Due to the use of SIFT [17], the EHP may be de-rived from the anatomy as observed in the planning CT scan, evenif it is different from the patient anatomy during treatment deliv-

ery. To predict the portal dose image without the patient in thebeam, PDI0, the incident fluence on the EPID is convolved with along- and short-range kernel to model the impact of head-scatterand the penumbra of measured EPIs, respectively [12].

Use of SIFT to derive the incident fluence

When comparing a measured and a predicted PDI, based on theplanning CT, differences might be observed that are either relatedto differences in the patient geometry or to deviations in the actu-ally delivered fluence maps. To distinguish both sources of differ-ences the SIFT method [19] is applied. For SIFT, each derivedIMRT field is split into a static, low MU field and a residual IMRTfield. PDIs are acquired for both fields and the ratio of both images(PDIratio) is calculated.

Because the impact of (moderate) changes in patient geometrywith respect to the time of the planning CT cancels out in PDIratio,the actually delivered IMRT fluence can be derived by comparingthe measured PDIratio with the predicted PDIratio based on the plan-ning CT. A similar iterative approach to the one we previously appliedfor reconstruction of 3D dose distributions based on pre-treatmentportal dose measurements and the planning CT is used [20]. In thefirst iteration, the fluence that was originally used by the TPS is mul-tiplied with the relative differences between the predicted and mea-sured PDIratio to derive the first estimate of the delivered IMRTfluence. Using this adapted fluence, a new PDIratio is predicted, whichis again compared with the measured PDIratio, resulting in a new up-date of the fluence. This iterative procedure continues until the meanabsolute value of the relative differences between the measured andpredicted PDIratio has been minimized. Generally, this stopping crite-rion is reached within less than 10 iterations. For each iteration, amaximum adaptation of the fluence by 20% is allowed, to reduceoverestimations in the first iterations, which would lead to unneces-sary long convergence times. After convergence, the adapted fluenceis considered equal to the actually delivered IMRT fluence.

Calculation of dose of the day

To calculate the dose of the day, the acquired CBCT scan is used.To get reliable CT values for dose calculation, the CT values of theplanning CT are mapped onto the CBCT [13]. First, the planningCT is automatically aligned to CBCT based on bony anatomy inthe vicinity of the target using a rigid 3D transformation and mu-tual information metric. After registration, the planning CT isresampled onto the grid of the CBCT. HU values of the planningCT scan are then mapped to the CBCT. To preserve the shape ofthe body outline as observed in the CBCT scan, all pixels outsidethis surface are assigned the standard CT value for air(�1000 HU), and all pixels inside the body surface of the CBCT,but outside the body outline of the planning CT scan, get the stan-dard CT value for water (0 HU). Similarly, to preserve the shape ofair cavities observed in the CBCT scan, defined using a fixed thresh-old, all pixels inside air cavities in the CBCT get the value�1000 HU, and pixels inside cavities in the planning CT scan, butoutside cavities in the CBCT, get the value 0 HU.

Reconstruction of the actually delivered dose distribution wasperformed with a research version of our clinically used TPS XIO(Elekta-CMS, version 4.3.3), based on the Hounsfield Units cor-rected CBCT scan and the actually delivered fluence maps, as de-rived from the portal dose measurements. The isocenter of thetreatment plan was set to the center of the CBCT scan.

Validation of dose reconstruction

Our method to reconstruct the actually delivered dose duringtreatment was first validated using phantom measurements. For

174 ‘Dose of the day’ for IMRT prostate cancer patients

a reference 10 � 10 cm2 static field, delivered with 100 MU, errorsin fluence delivery were simulated by manually adding 6 or 3 MUto the central 2 � 2 cm2 part of the field. PDI images were acquiredfor the modified fields delivered to either a solid water slab phan-tom of 20 cm thickness or to a lung phantom consisting of perspexwith cork inserts. Actually delivered fluence maps were derivedfrom the measured PDIs and the PDIs predicted for the original(unmodified) fields, using the iterative method described before.Using the TPS, the dose at 10 cm depth in a solid water phantomwas reconstructed, which was compared to the dose distributionthat was based on the theoretical fluence map of the modified field.

A similar validation was performed for several step-and-shootIMRT fields. Of each field, one segment was removed to simulatea delivery error. The adapted IMRT fields were delivered to the so-lid water slab phantom. For each of the modified fields, the actuallydelivered fluence was reconstructed from the measured PDI andthe predicted PDI of the unmodified IMRT field. Dose distributionsbased on the reconstructed fluence maps and the theoretical flu-ence of the modified IMRT fields were compared at 10 cm depth.

Fig. 1. Results of validation measurements. (A) The gamma evaluation (using 3%dose difference and 3 mm distance-to-agreement as reference values) of observedportal dose differences between the predicted PDI, based on the unmodified IMRTfluence, and the measured PDI for an IMRT field in which one of segments wasmanually removed. (B) The gamma evaluation between the dose distribution at10 cm depth in a solid water phantom that was derived from the reconstructedfluence map, based on the observed portal dose differences observed in (A), and thedose distribution that was derived from the theoretical fluence map of the modifiedIMRT field.

Clinical evaluation of 3D dose reconstruction for prostate cancerpatients

Clinical evaluation of the reconstructed ‘dose of the day’ wasperformed for 10 prostate cancer patients. These patients weretreated in a randomised clinical trial, comparing a standard frac-tionation scheme, delivering a total dose of 78 Gy in fractions of2 Gy (five times a week), with a hypofractionated scheme, deliver-ing a total dose of 64.6 Gy in fractions of 3.4 Gy (three times aweek). The treatment plan consisted of five IMRT fields at gantryangles of 0�, 55�, 110�, 250� and 305�. These gantry angles weresometimes slightly optimized based on the beam’s eye view pro-jection of the planning target volume (PTV). The 10 MV photonbeam of an Elekta Precise, equipped for step-and-shoot IMRT deliv-ery, was used for treatment. At least 99.5% of the PTV was plannedto receive at least 95% of the prescribed dose. The maximum dosesin the PTV and in the normal tissues were 107% and 100% of theprescribed dose, respectively. In addition, 60% of the rectal volumewas allowed to receive no more than 83% of the prescribed tumourdose.

Before each treatment fraction, the position of the prostate wasverified, based on markers implanted in the prostate, using a com-bination of MV and kV imaging [21] and on-line correction for theobserved translation of the center-of-mass of the markers was ap-plied using an automatic table shift. After correction, a lateral kVimage was acquired for verification of the on-line correction inthe sagittal plane. PDIs were acquired for all treatment fields ineach treatment fraction. According to our clinical protocol, a CBCTwas acquired immediately after treatment during the first threefractions and during one fraction in the fourth and the last butone week of treatment. For some patients either the CBCT or thePDI for any treatment field was not acquired successfully in oneof these fractions. As a result, the ‘dose of the day’ could be recon-structed for 4 or 5 fractions per patient.

Due to the limited quality of the CBCT scans, it was not possibleto accurately delineate the prostate volume in those scans. Becausein a previous study it was demonstrated that prostate deformationwith respect to implanted fiducial markers was small (standarddeviation 61 mm) [22], a mapping of the prostate volume, beingcontoured in the planning CT, to the CBCT was applied, based ona rigid registration (using both translations and rotations) of themarkers that were visible in both scans. In addition, the rectumvolume was manually delineated in each CBCT.

The reconstructed fluences, based on the portal dose measure-ments, and the Hounsfield Units corrected CBCTs were importedinto the planning system to reconstruct the actually delivered 3D

dose distribution. The results were compared with the planneddose distribution, by evaluating the isocenter dose, the 3D gammadistribution of the observed dose differences (using 3% of the pre-scribed dose and 3 mm distance-to-agreement as reference crite-ria), and the dose–volume histograms (DVHs) for the prostateand the rectum. For the prostate, the volume that received at least95% of the prescribed fraction dose, and the fraction dose that wasdelivered to 99% and 1% of the volume (i.e., the near-minimum andnear-maximum dose, respectively) were derived. For the rectumthe volume that received 83% of the prescribed fraction dose wasderived. In addition, the equivalent uniform dose (EUD) for rectalbleeding was calculated, considering a = 9 [23].

Results

Validation of dose reconstruction

Within the 10 � 10 cm2 treatment field, with either a 6% or 3%fluence offset in the central 2 � 2 cm2 region of the field, the differ-ences between the reconstructed and forward calculated dose dis-tributions at 10 cm depth in a solid water phantom were less than1%. When repeating this measurement for the lung phantom, theoverall agreement was within 2%.

For one of the IMRT fields, the gamma distribution of observedportal dose differences between the predicted PDI, based on theunmodified IMRT fluence, and the measured PDI for a modifiedIMRT field is shown in Fig. 1A. In a part of the field the measureddose was more than 3% lower than the predicted one, becauseone of the IMRT segments was omitted. Fig. 1B shows the gammadistribution of the observed dose differences in a coronal plane at10 cm depth, using either the fluence that was reconstructed basedon the portal dose differences observed in Fig. 1A, or the theoreticalfluence map of the modified IMRT field. Within the field the meangamma value is 0.33, while only 1.1% of the pixels have a gammalarger than 1. For the other IMRT fields similar results wereachieved, indicating that the delivered dose can be accurately de-rived from measured PDIs.

Clinical evaluation of 3D dose reconstruction for prostate cancerpatients

For a typical patient, results of the 3D gamma analysis on thedifferences between the originally planned and reconstructed dosedistribution in the transversal plane through the isocenter areshown in Fig. 2A. The gamma values were smaller than 1, except

M. van Zijtveld et al. / Radiotherapy and Oncology 96 (2010) 172–177 175

in the build-up region of the beams. Probably, these larger differ-ences are mainly related to the accuracy in definition of the bodycontour in the CBCT scan. For this patient CBCTs were acquiredin four treatment fractions. The DVHs for the prostate and the rec-tum that were derived from the reconstructed plans are shown inFig. 2B and C. For the prostate good agreements with the DVH ofthe clinical plan were observed, except for CBCT 4 in which about3% of the prostate volume was underdosaged. For each of the frac-tions the dose that was delivered to the rectum was lower than inthe original treatment plan.

In general, the reconstructed isocenter dose was in very goodagreement with the planned one. The mean difference, averagedover the 10 patients, was �0.4 ± 1.0% (1 SD), with day-to-day vari-ations per patient of 1.1% on average. For patients 6 and 8 the larg-est deviations (�2.0 ± 0.7% and �1.8 ± 1.1%, respectively) wereobserved. 3D gamma analyses showed that for 98.1 ± 1.7% (1 SD)

Fig. 2. (A) Gamma evaluation of observed dose differences between the plannedand reconstructed dose distribution in the transversal plane through the isocenterfor patient 4. (B) Dose–volume histograms for the prostate derived from theoriginally planned and reconstructed dose distributions for this patient. (C) Dose–volume histograms for the rectum derived from the originally planned andreconstructed dose distributions.

of the points within the 50% isodose surface the planned andreconstructed 3D dose distribution agreed within 3% or 3 mm.The mean day-to-day variation in the number of failed gamma pix-els per patient was 1.2%. Within the 20% isodose surface still96.7 ± 2.1% (1 SD) of the points had gamma values smaller than 1.

The results of the DVH analyses for the prostate are shown inFig. 3. For most treatment fractions, the volume that received atleast 95% of the prescribed dose was reduced, showing that somepart of the prostate volume was underdosed (Fig. 3A). For 8 ofthe patients the mean volume reduction within the 95% isodosesurface remained within 1.5%, but for patients 3 and 7 this volumereduction was 3.8% and 16%, respectively. The near-minimum dosewas reduced by up to 3% on average, except for patient 7, showinga mean reduction in D99 of 14% (Fig. 3B). With respect to the plan-ning CT, a rotation of the prostate in the sagittal plane by morethan 8� was observed in two CBCTs acquired for patient 3. Conse-quently, the prostate was no longer fully covered by the treatmentbeams, yielding some underdosages at the cranial side of the pros-tate. For patient 7, the underdosages in the prostate were causedby large, systematic intra-fraction motion of the prostate, mainlyin the cranial direction. The kV verification image taken directlyafter the on-line correction confirmed that at that stage the pros-tate was positioned within 2 mm accuracy in the sagittal plane.However, the CBCT acquired after treatment showed systematicchanges in patient posture, resulting in a shift of both the prostateand the bony anatomy by more than 5 mm (range 6–25 mm). Aver-aged over 10 patients the mean deviation in the near-maximumdose was 0.4 ± 1.3% (1 SD) only, with day-to-day variations of0.9% on average per patient .

Although large variations in rectal filling were observed be-tween the CBCTs and the planning CTs, the rectal dose remainedwithin the planning constraints for all patients and fractions, ex-cept for two fractions for patient 8. For this patient a larger overlapof the rectum and the high dose region was observed in the CBCTs

Fig. 3. (A) Differences in the prostate volume that receives at least 95% of theprescribed dose (V95) between the planned and reconstructed dose distribution. (B)Differences in the minimum dose in 99% of the prostate volume (D99). The dotsindicate the differences for individual fractions per patients, while the lines indicatethe average results over the 4–5 CBCTs per patient. For patient number 7, not allresults are included in the graph.

176 ‘Dose of the day’ for IMRT prostate cancer patients

compared to the planning CT due to changes in rectal filling. Theresults for the EUD for rectal bleeding are illustrated in Fig. 4. Com-pared to the planned EUDs, the calculated values were on average1.0 ± 4.5% higher.

Discussion

In this work, a method to derive the delivered ‘dose of the day’was described, based on the actually delivered fluence duringtreatment and the patient geometry as derived from a CBCT scan.Application of this method to 10 prostate cancer patients showedan overall good agreement between the planned and reconstructed3D dose distributions. However, for 2 patients large deviationswere found in two fractions that were most likely related to differ-ences in rectal filling with respect to the planning CT (resulting inrotation of the prostate) and intra-fraction motion during treat-ment, respectively.

Because a CBCT acquired at the end of treatment was used forreconstruction of the ‘dose of the day’, the accuracy of this proce-dure may be limited by intra-fraction motion. In the majority offractions (77%), this turned out to be no problem, because the in-tra-fraction motion that could be derived from a comparison ofthe center-of-mass of the prostate markers in the post-SGT verifi-cation image and the CBCT was less than 2 mm (2D vector). How-ever for patient 7, intra-fraction motion by 14 and 25 mm,respectively, was observed in two fractions, resulting in seriousdeviations between the planned and the reconstructed dose distri-butions (Fig. 3). But, most likely, these differences were smaller inreality, because a CBCT that is acquired after treatment is used fordose reconstruction, while the intra-fraction motion took placegradually during treatment, as was visible on the EPIs of the staticfields. To minimize the impact of intra-fraction motion in the treat-ment of prostate cancer patients, an extension of our stereographictargeting procedure, called intra-fraction SGT (or iSGT), was re-cently implemented clinically [24]. During iSGT, the position ofthe implanted markers is monitored at several stages during treat-ment delivery using the EPIs of the static fields. If the observed set-up deviation is larger than some threshold (currently set to 4 mm),a new on-line correction is applied by automatically shifting thecouch. With this method the effective systematic and randomset-up errors for the patient population can be maintained within1 mm. As a result, application of iSGT will also allow a more reli-able reconstruction of the ‘dose of the day’.

With SIFT the actually delivered fluence can be derived frommeasured portal dose images within about 1% accuracy, even whenthe patient geometry has been changed between acquisition of theplanning CT and the time of treatment delivery, or when absorbingparts of the treatment couch or positioning devices are present inthe treatment beam [25]. Also, when no CBCT is acquired during acertain fraction, the delivered fluence can still be verified accu-rately. For reconstruction of the delivered IMRT fluence from thePDIratio, it is considered that the low MU static field was delivered

Fig. 4. EUD for rectal bleeding derived from the planned (lines) and reconstructed(dots) dose distributions.

accurately. In general this is the case, because both the beam out-put (i.e., cGy/MU) and the beam profile of these static fields arereproducible within 0.5% [19]. With SIFT these parameters arenot verified, since they cancel out in PDIratio, but they are checkedon a daily basis during our daily morning QA with the EPID [26]. Analternative approach to SIFT is to directly compare the measuredPDI of the IMRT fields with the corresponding predicted PDI, basedon an EHP of the patient derived from the CBCT [15]. This wouldtake both the actual patient geometry into account and allow forabsolute fluence verification, including a check on the beam outputand the beam profile. The accuracy of this approach is currentlybeing assessed. A possible problem could be intra-fraction changesin patient geometry occurring between the actual dose deliveryand acquisition of the CBCT, for instance due to gas pockets passingthrough the bowels. Because with SIFT the PDIs for the static andmodulated fields are acquired immediately after each other, the in-tra-fraction changes in patient anatomy are generally negligible inthis short time span, yielding a more reliable assessment of theactually delivered fluence. Moreover, derivation of actually deliv-ered fluences from PDIs instead of PDIratio requires accurate model-ling of the treatment couch and immobilisation devices in the PDIprediction.

For the Hounsfield Unit correction of the CBCT scans, simplethresholds were used for definition of air and bone tissue. In con-trast to our previous application for head and neck cancer patients[13], a separate threshold had to be used to discriminate air insidethe patient (mainly in the rectum) and air outside the patient(defining the body contour), due to cupping artefacts in the CBCT,being acquired without bowtie filter [27]. Especially in lateraldirection, where the impact of scatter on the CBCT is largest, theuse of fixed thresholds resulted in systematic deviations in thebody contour of the patient by a few millimeters, being smallerthan the one observed in the planning CT. For different thresholdsused to define the body contour, the impact on the reconstructeddose distribution was evaluated. Except in the build-up region ofthe treatment beams (i.e., the first 1 cm within the body surface)only minor dose differences were observed. So, in this stage, wedid not consider the implementation of more advanced methodsfor edge-detection for determination of the body contour, e.g.,based on local instead of global contrast, to be important.

Other methods for calculation of the actually delivered dose to apatient have been described in the literature [14–16,28]. Apartfrom the use of SIFT, unique in our approach is that the entire cal-ibration to derive measured and predicted portal dose images isbased on EPID measurements only. For the fluence reconstruction,it is not necessary to relate measured PDI pixel values directly todose in the patient. For reconstruction of the delivered 3D patientdose we use our clinical planning system, while others use an inde-pendent dose calculation algorithm [15,16,28]. The advantage ofthis latter approach might be that errors in the dose calculationcan also be detected at once. However, we decided to separate ver-ification of the 3D dose calculation (by recalculating the plan witha second, independent dose algorithm prior to treatment and com-paring the results) from verification of the dose delivery, makingthe distinction of dose differences resulting from errors in fluencedelivery or changes in patient geometry and dose differences dueto calculation errors more straightforward.

In this article, we derived the ‘dose of the day’ for a limitednumber of fractions per patient. For dose-adaptive radiotherapy,a method to sum the dose distributions from different fractionshas to be developed. One approach could be to use deformable reg-istration methods [29] on the acquired CBCT images to map thedose from one fraction onto that of the next. However, for accuratedetermination of the total dose to a specific structure, accuratestructure definition in the cone-beam CT is also required. In ourexperience, the image quality of the cone-beam CT scans does

M. van Zijtveld et al. / Radiotherapy and Oncology 96 (2010) 172–177 177

not always allow for accurate delineation of most structures. Also,the FOV of the scans is limited, such that some structures are not(entirely) visible. These issues are currently being investigatedwith the intention to develop protocols and action levels for off-line dose-adaptive radiotherapy.

Conclusion

A method for calculation of the actually delivered dose to thepatient was developed and validated. Application of the methodto a group of 10 prostate cancer patients showed that generallyDVH parameters remained within the planning constraints. Theisocenter dose was found to be within �0.4 ± 1.0% (1 SD) and thedifference between the planned and delivered dose was within3%/3 mm for 98.1% of the pixels with a dose higher than 50% ofthe prescribed dose. A large degradation of the dose delivery tothe prostate was observed in two patients for two fractions each,resulting in underdosages. For one patient this was mainly relatedto changes in rectal filling and for the other to intra-fraction mo-tion of at least 5 mm. To correct for this kind of deviations, dosereconstructions for several fractions will be added in the near fu-ture for application in off-line dose-guided strategies.

Acknowledgements

This work was financially supported by the Dutch Cancer Soci-ety (Grant DDHK 2004-3107). The authors thank Elekta-CMS forproviding a research version of XIO to perform the dosereconstruction.

References

[1] Fu W, Yang Y, Yue NJ, Heron DE, Huq MS. A cone beam CT-guided online planmodification technique to correct interfractional anatomic changes forprostate cancer IMRT treatment. Phys Med Biol 2009;54:1691–703.

[2] Rijkhorst EJ, van Herk M, Lebesque JV, Sonke JJ. Strategy for online correction ofrotational organ motion for intensity-modulated radiotherapy of prostatecancer. Int J Radiat Oncol Biol Phys 2007;69:1608–17.

[3] Smitsmans MH, de Bois J, Sonke JJ, Betgen A, Zijp LJ, Jaffray DA, et al. Automaticprostate localization on cone-beam CT scans for high precision image-guidedradiotherapy. Int J Radiat Oncol Biol Phys 2005;63:975–84.

[4] Lee C, Menk F, Cadman P, Greer PB. A simple approach for using an amorphoussilicon EPID to verify IMRT planar dose maps. Med Phys 2009;36:984–92.

[5] van Zijtveld M, Dirkx M, Breuers M, de Boer H, Heijmen B. Portal dose imageprediction for in vivo treatment verification completely based on EPIDmeasurements. Med Phys 2009;36:946–52.

[6] van Elmpt W, McDermott L, Nijsten S, Wendling M, Lambin P, Mijnheer B. Aliterature review of electronic portal imaging for radiotherapy dosimetry.Radiother Oncol 2008;88:289–309.

[7] McDermott LN, Wendling M, Sonke JJ, van Herk M, Mijnheer BJ. Replacingpretreatment verification with in vivo EPID dosimetry for prostate IMRT. Int JRadiat Oncol Biol Phys 2007;67:1568–77.

[8] Franken EM, De Boer JCJ, Barnhoorn JC, Heijmen BJM. Characteristics relevantto portal dosimetry of a cooled CCD camera-based EPID. Med Phys2004;31:2549–51.

[9] Louwe RJ, McDermott LN, Sonke JJ, Tielenburg R, Wendling M, van Herk MB,et al. The long-term stability of amorphous silicon flat panel imaging devicesfor dosimetry purposes. Med Phys 2004;31:2989–95.

[10] De Boer JCJ, Heijmen BJM, Pasma KL, Visser AG. Characterisation of a highelbow, fluoroscopic electronic portal imaging device for portal dosimetry. PhysMed Biol 2000;45:197–216.

[11] Van Zijtveld M, Dirkx MLP, De Boer HC, Heijmen BJ. Dosimetric pretreatmentverification of IMRT using an EPID; clinical experience. Radiother Oncol2006;81:168–75.

[12] Franken E, De Boer H, Heijmen B. A closed system for dosimetric IMRTverification with an EPID, eliminating ion chamber calibration measurements.Radiother Oncol 2006;81:S157.

[13] van Zijtveld M, Dirkx M, Heijmen B. Correction of conebeam CT values using aplanning CT for derivation of the ‘‘dose of the day’’. Radiother Oncol2007;85:195–200.

[14] McDermott LN, Wendling M, Nijkamp J, Mans A, Sonke JJ, Mijnheer BJ, et al. 3Din vivo dose verification of entire hypo-fractionated IMRT treatments using anEPID and cone-beam CT. Radiother Oncol 2008;86:35–42.

[15] van Elmpt W, Nijsten S, Petit S, Mijnheer B, Lambin P, Dekker A. 3D in vivodosimetry using megavoltage cone-beam CT and EPID dosimetry. Int J RadiatOncol Biol Phys 2009;73:1580–7.

[16] van Elmpt W, Petit S, de Ruysscher D, Lambin P. Dekker A 3D dose deliveryverification using repeated cone-beam imaging and EPID dosimetry forstereotactic body radiotherapy of non-small cell lung cancer. RadiotherOncol 2010;94:188–94.

[17] Viera SC, Dirkx MLP, Heijmen BJM, De Boer JCJ. SIFT: a method to verify theIMRT fluence delivered during patient treatment using an electronic portalimaging device. Int J Radiat Oncol Biol Phys 2004;60:981–93.

[18] Franken EM, De Boer JCJ, Heijmen BJM. A novel approach to accurate portaldosimetry using CCD-camera based EPIDs. Med Phys 2006;33:888–903.

[19] Pasma KL, Heijmen BJM, Kroonwijk M, Visser AG. Portal dose image (PDI)Prediction for dosimetric treatment verification in radiotherapy. I. Analgorithm for open beams. Med Phys 1998;25:830–40.

[20] van Zijtveld M, Dirkx ML, de Boer HC, Heijmen BJ. 3D dose reconstruction forclinical evaluation of IMRT pretreatment verification with an EPID. RadiotherOncol 2007;82:201–7.

[21] Mutanga TF, de Boer HC, van der Wielen GJ, Wentzler D, Barnhoorn J, IncrocciL, et al. Stereographic targeting in prostate radiotherapy: speed and precisionby daily automatic positioning corrections using kilovoltage/megavoltageimage pairs. Int J Radiat Oncol Biol Phys 2008;71:1074–83.

[22] van der Wielen GJ, Mutanga TF, Incricci L, Kirkels WJ, Vasquez Osorio EM,Hoogeman MS, et al. Deformation of the prostate and seminal vesicles relativeto intraprostatic fiducial markers. Int J Radiat Oncol Biol Phys2008;72:1604–11.

[23] Peeters STH, Hoogeman MS, Heemsbergen WD, et al. Rectal bleeding, fecalincontinence, and high stool frequency after conformal radiotherapy forprostate cancer: normal tissue complication probability modeling. Int J RadiatOncol Biol Phys 2006;66:11–9.

[24] De Boer H, Rajan V, Barnhoorn J, Mutange T, Wentzler D, Heijmen B. Fullyautomated correction for intrafraction prostate motion by continuous imagingof fiducial markers. Radiother Oncol 2008;88:S31.

[25] Vieira SC, Kaatee RSJP, Dirkx MLP, Heijmen BJM. Two-dimensionalmeasurement of photon beam attenuation by the treatment couch andimmobilization devices using an electronic portal imaging device. Med Phys2003;30:2981–7.

[26] Vieira SC, Bolt RA, Dirkx MLP, Visser AG, Heijmen BJM. Fast, daily linacverification for segmented IMRT using electronic portal imaging. RadiotherOncol 2006;80:86–92.

[27] Mail N, Moseley DJ, Siewerdsen JH, Jaffray DA. The influence of bowtiefiltration on cone-beam CT image quality. Med Phys 2009;36:22–32.

[28] van Elmpt W, Nijsten S, Mijnheer B, Dekker A, Lambin P. The next step inpatient-specific QA: 3D dose verification of conformal and intensity-modulated RT based on EPID dosimetry and Monte Carlo dose calculations.Radiother Oncol 2008;86:86–92.

[29] Vásquez Osorio EM, Hoogeman MS, Al-Mamgani A, Teguh DN, Levendag PC,Heijmen BJ. Local anatomic changes in parotid and submandibular glandsduring radiotherapy for oropharynx cancer and correlation with dose, studiedin detail with nonrigid registration. Int J Radiat Oncol Biol Phys2008;70:875–82.