Accurate IMRT fluence verification for prostate cancer patients using ‘in-vivo’ measured EPID images and in-room acquired kilovoltage cone-beam CT scans

With the introduction of advanced treatment techniques like intensity modulated radiotherapy (IMRT) and volumetric modulated arc therapy (VMAT), the high-dose volume can generally be better conformed to the target volume, while achieving steep dose gradients towards surrounding normal tissues. As a result, the accuracy of patient positioning and dose delivery becomes more critical. To verify delivered treatment fields, dosimetric pre-treatment verification is performed prior to the first treatment fraction using films, 2D detector arrays or Electronic Portal Imaging Devices (EPIDs) in many institutes.

EPIDs have shown to be suitable for dose verification, both pre-treatment and in-vivo because of their high resolution, acquisition of digital data in 2D in a short acquisition time, and high measurement accuracy [1‐10]. In our institute, for over eight years, pre-treatment EPID dosimetry is performed for all IMRT patients to compare fluence maps delivered by the linac with TPS prescriptions. For this purpose, prior to the first treatment fraction, a full treatment including gantry rotations is executed in the absence of the patient and treatment couch, while acquiring for each beam an EPID image. These images are converted into Portal Dose Images (PDIs) [10] and compared with predictions, based on the prescribed leaf sequence and monitor units (MU). For several patients, major errors were captured prior to the start of treatment [2]. Similar errors, like a malfunctioning leaf or unintended changes of treatment parameters in the record and verify system have also occurred during patient treatment, after the pre-treatment verification was successfully performed. So, in spite of the high overall reproducibility of delivered IMRT fluence maps (within 1% (1 SD), as assessed from repeated EPID measurements) [11], pre-treatment verification does not guarantee correct fluence delivery to the patient during the course of treatment. For this purpose, in-vivo dosimetry may be performed. When in-vivo PDI predictions are based on the pCT of the patient, comparison with measured PDIs may result in an unreliable estimate of the fluence delivered by the treatment unit, because of changes in patient positioning and/or patient anatomy (e.g., due to deformation, changing air gaps or weight loss). To be able to accurately derive delivered fluences irrespective of changes in patient anatomy, the Split IMRT Field Technique (SIFT) [12] was introduced in our institute. For this technique, each IMRT field is split into a low MU static field and a residual IMRT field. From the ratio of the PDIs of these fields, both being acquired in a short time interval, the fluence of the IMRT field can be derived within about 1% accuracy. The SIFT method has several limitations: (i) deviations in beam output (cGy/MU) and in open beam profile cannot be detected, because they cancel out in the ratio image, (ii) in commercially available treatment planning systems (TPS), splitting of an IMRT field into a static and residual modulated field is often not possible, and (iii) the technique is only applicable to IMRT; 3D conformal radiotherapy treatments or rotational treatments, like Rapid Arc or VMAT, cannot be verified in a similar way.

CBCTs are increasingly being used to verify and correct target set-up. In this study, we investigated the accuracy of in-vivo fluence verification by analyses based on PDIs derived from measured EPID images and predictions calculated using a CBCT acquired in the same treatment fraction. It was hypothesized that an in-room acquired CBCT would better represent the patient’s anatomy during acquisition of the EPID images than the pCT would do, allowing for an increased PDI prediction accuracy and resulting in an enhanced sensitivity for detecting fluence maps deviations. Therefore, for evaluation purposes, PDIs were also predicted using the patients’ pCTs. It was investigated whether combining in-vivo measurements performed in the first 1–5 fractions could enhance agreement between measurements and predictions, comparable to pre-treatment verification.

As far as we know, this is the first article describing the use of kilovoltage CBCT for prediction of in-vivo measured PDIs to verify delivered IMRT fluence map with an EPID. As expected, the agreement between measured PDIs and CBCT-based predicted PDIs was slightly better in phantom experiments than for patients. This is partly attributed to intra-fraction patient motion. For some patient fractions, the acquired CBCT did not accurately represent the patient’s anatomy for all treatment beams and hence EPID acquisitions. Fraction 3 in Figure 1 shows an example. At the start of treatment, the agreement between measured and CBCT-based predicted PDIs was clearly less good than for beams acquired near the end of the fraction, i.e., shortly before the CBCT was acquired. Apparently, the acquired CBCT did not accurately represent the patient anatomy at the start of treatment, likely due to intra-fraction anatomy changes. It should be noted that the applied on-line set-up correction procedure to account for observed inter- and intra-fraction prostate displacements, as derived from imaged implanted fiducials, could not (fully) account for these effects.

Figure 3 shows that also for in-vivo dosimetric analyses based on CBCT in one single fraction, gamma deviations between measurements and predictions are low, which allows detection of large, clinically relevant errors immediately after the first fraction. Combination of in-vivo results in multiple fractions reduces the impact of intra-fraction motion, and averages out random variations in fluence delivery that do not occur in all fractions. On the other hand, reproducible errors in fluence delivery will not diminish. Consequently, by combining multiple fractions, the sensitivity for detection of systematic errors in fluence delivery (i.e., occurring in all fractions) will increase. This approach was initially suggested by McDermott et al. [21]. For a group of 75 prostate cancer patients, they concluded that by combining in-vivo dosimetry results from 3–5 treatment fractions using ‘low’ γ images, the results were in close agreement with pre-treatment results. In this study, we also investigated the use of these low γ images, both for CBCT- and pCT-based predictions. For the pCT-based approach we found that 4–5 fractions had to be combined to approach pre-treatment results. When using CBCT, already combination of only the first 2 fractions yielded results in very close agreement with pre-treatment analyses. Obviously, even with some intra-fraction motion, CBCT scans do much better represent patient anatomy during the fractions than the pCT does and consequently a more accurate fluence verification was possible than with pCT scans.

Over three years ago a computer-controlled system is running in our department that daily checks whether treatment parameters in our record-and-verify system correspond to the parameters of the plan that was prescribed and approved by the physician [22]. With this system we also check pre-treatment whether a treatment plan is correctly transferred from the planning system to the record-and-verify system. In addition, a redundant monitor unit calculation is performed prior to start of treatment for all our patients. To correct for inter- and intra-fraction motion in prostate cancer patients, on-line set-up verification is performed before and during each treatment fraction [19, 20]. On top of these individualized QA procedures, the IMRT performance of our linacs is checked daily with EPID measurements during the morning checks [23, 24]. During the last eight years, we performed pre-treatment verification for all IMRT patients. For less than 0.3% of the patients, clinically relevant errors were detected. Because of the large number of IMRT patients, performing pretreatment verification for all of them is a workload intensive practice, involving measurements during evenings. Moreover, pre-treatment verification inherently checks profiles that are not necessarily equal to those delivered to the patient.

The results obtained in this work gave us confidence that for prostate patients treated with a large number of fractions, IMRT fluence delivery verification with pre-treatment EPID dosimetry may be replaced by in-vivo EPID dosimetry, when combined with the QA steps mentioned before (in particular the pre-treatment check on data transfer). This has the advantage that fluence delivery verification is performed while the patient is being treated with a plan that has been checked pre-treatment for data transfer errors. In practice, we verify the first 3 fractions, and also analyze the combined fractions, as described in this paper. A relevant side effect is that measurements in evenings and weekends can largely be reduced. Of course, our method can also be used as a supplement to pre-treatment verification, adding fluence verification during patient treatment. In principal, this can be done in each fraction.

Due to the high accuracy, in-vivo verification based on CBCT allows for detection of clinically significant errors with minimal user interaction after the first treatment fraction, as is evident from the gamma results and high percentage of accepted images (95.8%) according to our automatic decision scheme. In case there are doubts on the accuracy of the fluence delivery, a pre-treatment measurement will be performed before the next fraction. If this measurement confirms relevant errors in fluence delivery, replanning should be performed before the next treatment fraction is given. If, on the other hand, the pre-treatment verification does not indicate any clinically relevant deviations, in-vivo measurements will be repeated during the next fractions to verify whether the PDI differences might be related to systematic deviations in patient anatomy. Also in that case replanning should be considered. For fractionated treatments with a large number of fractions we consider this approach acceptable, since corrections to compensate for delivery errors can still be performed in the remaining treatment fractions. McDermott et al. [21] suggested a similar approach.

Several groups have investigated the use of CBCT scans to reconstruct delivered patient doses based on measured EPID images (‘dose of the day’) [25, 26]. In the work of McDermott et al. [25] measured EPID images were back-projected to multiple planes in a (kilovoltage) CBCT, assuming a water-equivalent electron density. Van Elmpt et al. [26, 27] acquired megavoltage CBCT scans with calibrated electron densities to derive the entrance energy fluence for each treatment beam by back-projection of measured PDIs. These fluences were then used for reconstruction of the dose distribution in the CBCT using a Monte Carlo dose algorithm. Our group applied the SIFT method to derive the entrance fluence from measured EPID images and then performed a dose reconstruction using kilovoltage CBCT scans with corrected Hounsfields units [28]. As replacement of the SIFT method, we can now also apply the entrance fluences, as derived from the CBCT-based in-vivo measurements, to reconstruct the delivered patient dose.

The results of this study demonstrate for prostate cancer patients that high accuracy in-vivo verification of the delivered 3D patient dose based on CBCT might be complicated by intra-fraction anatomy variations. Likely, this can be alleviated by the combining analyses performed in 2D for many fractions. Investigations on the combination of several fractions in back projected EPID measured fluencies to estimate delivered 3D patients dose distributions, is a topic for further research that is important for so-called dose-guided radiotherapy.

In this paper we have demonstrated that in-room CBCT and in-vivo measured EPID images can be used for high accuracy in-vivo IMRT fluence verification for prostate cancer patients that are known for intra-fraction patient anatomy changes [20]. For other sites, the impact of such effects on the accuracy of CBCT-based fluence verification has to be investigated. The developed technology may also be applied for fluence verification of VMAT, stereotactic treatment, and 3DCRT. Accuracy assessment for these applications is a topic for further study.

In-room acquired CBCT scans can be used for high accuracy IMRT fluence verification for prostate cancer, based on in-vivo measured EPID images. Combination of γ results for the first 2 fractions can largely compensate for small accuracy reductions with respect to pre-treatment verification, related to intra-fraction motion and anatomy changes. Compared to fluence verification based on the pCT-scan, accuracy of verification using in-room CBCT scans is clearly enhanced.

In this study new tools to verify the accuracy of IMRT delivery using an EPID were developed and evaluated. This work did not affect patient’s treatment. Digital information, already routinely acquired for patient-specific QA, was used, i.e., for patients no additional investigations or measurements were required. For these reasons, according to the Ethics Committee of our hospital, no ethical approval was required for this study. Providing that all patient-related information was anonymized prior to presentation and publication, informed consent from the patients was not needed either.