Cone beam CT for QA of synthetic CT in MRI only for prostate patients

Abstract Purpose Magnetic resonance imaging (MRI)‐only radiotherapy is performed without computed tomography (CT). A synthetic CT (sCT) is used for treatment planning. The aim of this study was to develop a clinically feasible quality assurance (QA) procedure for sCT using the kV‐cone beam CT (CBCT), in an MRI‐only workflow for prostate cancer patients. Material and method Three criteria were addressed; stability in Hounsfield Units (HUs), deviations in HUs between the CT and CBCT, and validation of the QA procedure. For the two first criteria, weekly phantom measurements were performed. For the third criteria, sCT, CT, and CBCT for ten patients were used. Treatment plans were created based on the sCT (MriPlannerTM). CT and CBCT images were registered to the sCT. The treatment plan was copied to the CT and CBCT and recalculated. Dose–volume histogram (DVH) metrics were used to evaluate dosimetric differences between the sCT plan and the recalculated CT and CBCT plans. HU distributions in sCT, CT, and CBCT were compared. Well‐defined errors were introduced in the sCT for one patient to evaluate efficacy of the QA procedure. Results The kV‐CBCT system was stable in HU over time (standard deviation <40 HU). Variation in HUs between CT and CBCT was <60 HU. The differences between sCT–CT and sCT–CBCT dose distributions were below or equal to 1.0%. The highest mean dose difference for the CT and CBCT dose distribution was 0.6%. No statistically significant difference was found between total mean dose deviations from recalculated CT and CBCT plans, except for femoral head. Comparing HU distributions, the CBCT appeared to be similar to the CT. All introduced errors were identified by the proposed QA procedure, except all tissue compartments assigned as water. Conclusion The results in this study shows that CBCT can be used as a clinically feasible QA procedure for MRI‐only radiotherapy of prostate cancer patients.


| INTRODUCTION
The traditional radiotherapy workflow for prostate cancer is built upon two imaging modalities, that is, computed tomography (CT) and magnetic resonance imaging (MRI). The CT data set provides Hounsfield Units (HUs), which the treatment planning system converts to relative electron densities (RED) needed for the absorbed dose calculations.
MRI is preferred as a complementary imaging modality due to the superior soft tissue contrast, and therefore enables a more accurate target delineation. 1 A coregistration of the two data sets is used to create a foundation for target delineation. Although this is a common procedure, uncertainties associated with the MR-CT registration has been estimated to 2 mm for prostate cancer patients. 2 Interest has been directed toward a workflow using MRI as the only imaging modality for treatment simulationoften denoted as an MRI-only workflow. An MRI-only workflow will not only avoid registration prior target delineation, but also simplify the workflow, reduce the workload and therefore become more cost effective. 3 However, to enable treatment planning from an MRI data set the need for RED required for absorbed dose calculation must be considered.
There are many different methods available for converting MRI data to create a so-called synthetic CT (sCT). [4][5][6][7][8][9][10][11][12] An sCT is an image that looks very similar to a CT image and can be used by the treatment planning system for absorbed dose calculations as well as provide digital reconstructed radiographs (DRR). Recently, different methods for generating sCT data set were reviewed. 13 At present there are two vendors that provide sCT data sets for the male pelvis region (MRCAT TM and MriPlanner TM ). 11,12 In a recent multicenter, multivendor study (MR-only prostate external radiotherapy (MR-OPERA)), the MriPlanner TM was validated for the male pelvis. 14 Dose-volume histogram (DVH) comparisons between sCT and CT dose distributions resulted in mean dose deviations that were below 0.3% for all DVH metrics and therefore negligible compared with other uncertainties in radiotherapy. No major errors were found in any of the steps of the sCT image conversion.
Although the method is validated, there might be conditions where the sCT data generation is inadequate due to other circumstances, such as operator noncompliance with the MR protocol or unexpected anatomic structures. Therefore, an independent quality assurance (QA) procedure of the sCT data would be preferred before, or in connection with, the first treatment fraction. [13][14][15] Recently, the cone beam CT (CBCT), which is performed on the treatment machine, was suggested as a way to perform patient specific QA for radiotherapy of the brain. 16 Assuming that the CBCT data set gives similar attenuation information as a CT, the information can be used to validate the sCT, with respect to HUs. The brain is a rather fixed anatomical geometry, while we expect the male abdomen to be less stable between the planning session and the treatment. Therefore, we need to study if CBCT can be used for QA of sCT data for prostate cancer patients. In addition, it would be of interest to go one step further and verify the sCT data with respect to absorbed dose using CBCT data.
The aim of this study was to develop a clinically feasible QA procedure for MRI-only radiotherapy of prostate cancer patients using the CBCT to detect errors in a synthetic CT.

| METHODS
In this work, an MRI only workflow using a sCT generated from a standard T2-weighted MRI generated by MriPlanner TM (Spectronic Medical AB, Sweden) was considered. The workflow is based on the following steps; first an MRI encompassing the outer body contour, large field of view (FOV) MRI, is acquired along with sequences dedicated for delineation and fiducial marker identification. A sCT is generated based on the large FOV MRI through an automatic generation process directly connected to the MR scanner and treatment planning system. The sCT is used in the same manner as a conventional CT throughout the treatment planning process. The MRI is used for delineation of target and organs at risk (OAR) and fiducial marker identification, without the need for image registration toward the sCT. A treatment plan is created and dose is calculated using the sCT. The patients are scheduled for the first treatment fraction approximately 2 weeks after the MRI. In our suggested workflow a CBCT is acquired at the first treatment fraction and the QA procedure can be performed before the second treatment fraction. The patients are positioned using gold fiducials markers and the CBCT is solely to be used for QA purposes.
In order to use the CBCT data as a QA tool for sCT data the following criteria need to be addressed: 1. Stability in HU over time for the CBCT system used. To establish that the CBCT system does not drift over time, introducing systematic errors if not corrected for when dose calculating.
2. Deviation in HU between CT and the CBCT system used. To investigate the need of using specific HU to RED calibration curves.
3. Validation of the QA procedure using a relevant patient data set.
To investigate the feasibility for dose recalculation on CBCT data.

2.A | Phantom measurements
Addressing criteria 1), periodic measurements were carried out to evaluate the stability of the system used for kV-CBCT imaging. The kV-CBCT system used was Varians On-Board Imager TM TrueBeam TM .
Measurements were performed weekly on one treatment unit for 8 weeks using the CIRS Model 062M electron density phantom (Computerized Imaging Research Systems Inc., Norfolk, VA, USA).
The mean HU of the phantom inserts were evaluated in the PALMÉR ET AL.
| 45 registration module Eclipse Image Registration (Varian Medical Systems v. 13.6, Palo Alto, CA, USA). Addressing criteria 2), one phantom measurement was performed on the CT scanner Siemens Somatom Definition AS+. The mean HU of the phantom inserts were compared between the CT and CBCT units used in this study.

2.B | Patients
The patients in this paper were from an ongoing study (MR-only prostate radiotherapy excluding CT (MR-PROTECT)), which studies a radiotherapy workflow incorporating all mentioned advantages of using an MRI-only routine. The study has been approved by the

2.C | Imaging
The patients underwent MR and CT imaging prior to their treatment, and a CBCT image was acquired at the first treatment fraction.
CBCT was solely acquired for QA purposes and the patients were positioned daily using gold fiducial markers during the whole course of treatment including the first fraction. The conversion software MriPlanner TM was used for sCT generation. 12 In the MR-PROTECT study, the CT data set was primarily acquired to enable checks and validation of the MRI-only workflow. The primary use of the CT data in this presenting study was to validate the dose calculation ability in CBCT data. Images of sCT, CT, and CBCT are given for one patient (Fig. 1).
The MR scanner and field strength used was GE Discovery, 750 w 3.0 T with a flat table top. Patient fixation included knee, ankle support, and reference tattoos to enable a reproducible patient positioning. The receiver coil used was a 16-channel GEM anterior array placed on a stiff coil bridge, ensuring no deformation of the outer patient body contour. The sequence used for the sCT generation has previously been described. 14,17 The sCT conversion software required a T2-weighted MRI data set with a FOV enclosing the complete patient contour. The cranio-caudal extent of the MRI was set to include the target area (i.e., the prostate) and the OARs, including rectum and bladder.
The CT was acquired directly after the MRI using equivalent patient fixation, using the same CT scanner used for phantom measurement. The CBCT data set was acquired at the first treatment fraction, approximately 2 weeks after CT and MR imaging, using the same kV-CBCT imaging device as for the phantom measurements.
Equivalent patient fixation was used as for both MR and CT imaging.
The reference tattoos made at the MR were used to reproduce the patient position during both CT and CBCT imaging. Parameter settings for MR, CT, and CBCT imaging are given in Table 1.

2.D | Treatment plan recalculation
The original treatment plans in this study were created based on the sCT data. Target and OARs were contoured manually using the MRI.
CT and CBCT data were automatically matched rigidly, excluding rotation, based on bony anatomy toward the sCT in Eclipse. Target volumes and OARs were transferred from the sCT data set to the CT and CBCT data set. The body contour structure was separately generated for each data set. According to the clinical protocol, any connected areas of air in the rectum exceeding 4 cm in diameter in any direction is not acceptable for treatment planning. Therefore, if such areas were present in the CT and/or CBCT data set, these were  A statistical analysis was carried out to investigate if there was a significant difference between the total mean dose deviations for the DVH metrics, resulting from recalculations on CT and CBCT data. An independent two-tail, two-sample t-test, assuming equal variance, with a 5% significance level was performed using MATLAB R2017a (MathWorks, Inc., Natick, MA, USA) and the resulting confidential intervals (CI) were evaluated.
In two patients, representing the best and worst case with respect to CT/CBCT PTV D mean difference, comparison of the HU distributions in sCT, CT, and CBCT were carried out. To enable a fair comparison of HU distributions, all data sets for these patients were recalculated to have the same resolution and slice thickness.

2.E | Simulation of sCT data with errors
In order to assess the capability of the method to detect sCT/CBCT mismatch, errors were intentionally introduced in the sCT data set.
This was performed on one patient having the smallest dose difference in PTV D mean between the original sCT plan and the recalculated CBCT plan (−0.03%). The following errors were introduced: all tissue compartments assigned as water, the complete bone structure assigned as cortical bone, uniformly (2 mm) enlarged bone structures, and changes in patient body contour dimension of ±10 mm (Fig 2).
The extended bone structure was assigned cortical bone and extended patient body contour to water. For each new sCT data set impaired with an error, a treatment plan was created and optimized according to local clinical protocol. The new calculated plan was transferred onto the original sCT and CBCT data sets and absorbed dose was recalculated. This resulted in three treatment plans based on three different images for each introduced error. PTV D mean was used to compare dose differences between the plan based on sCT with error and the recalculated sCT and CBCT plans.

| RESULTS
Phantom measurements showed that the kV-CBCT system was stable in HU over time. The standard deviation (SD) ranged from 5.9 to 40 HU for materials with 0.2-1.7 RED. There was less than 60 HU variation between CBCT and CT data.
For all patients, a successful dose recalculation was performed based on the CBCT data sets using the standard clinical HU to RED conversion. Recalculations based on CT and CBCT data showed similar deviations from the sCT data calculation. The results are described as dose differences in percent of prescribed dose for the clinical protocol DVH metrics, comparing CT and CBCT dose distribution to the sCT dose distribution for ten patients (Fig 3). The baseline of zero represents no difference between the DVH metrics compared.
There were only small deviations for all DVH metrics between the sCT and CT dose distributions. Median doses were within 0.9% deviation for all metrics. The deviations ranged from −2.1% to 3.4%.
An outlier was found for the DVH metric left femoral head D 2% and had a value of 3.4%. Rectum D 30% was the DVH metric that showed the largest overall deviation with a range of −0.5% to 3.0%.
Generally, there was a larger median deviation in the investigated DVH metrics for the CBCT versus sCT dose distribution compared to CBCT versus CT. Two median doses, CTV min and rectum D 10% , were below the -1% deviation. The deviations ranged from −2.2% to 0.6%, a slightly smaller range than seen for the CT comparison. An outlier for the CBCT data set could be seen for the DVH metric rectum D 30% with a value of -1.1%. The outlier seen for the femoral head in the CT comparison was not present in the CBCT comparison. Table 2 shows the total mean absorbed dose difference in percent of prescribed dose for the clinical protocol DVH metrics for ten patients and the corresponding SD. Mean dose differences between sCT and CT dose distributions were below or equal to 0.8% for all DVH metrics. The respective mean dose differences between sCT and CBCT dose distributions were below or equal to 1.0%. There was a significant difference between CT and CBCT in the right femoral head D 2% . All other investigated DVH metrics showed no significant difference between recalculated CT and CBCT plans.  In general, the HU histograms overlap for all data sets. A local overlap could clearly be seen in the muscle and fat region, representing the peaks, in Fig. 4. The CBCT distribution seemed to have a more level appearance compared to the other distributions. Table 3  Our method relies on the consistency in HU of the specific CBCT system. and rectum D 30% . No significant difference was found for 10 out of T A B L E 2 Mean dose deviation between sCT and CT and between sCT and CBCT for ten patients. T A B L E 3 PTV D mean difference between error sCT and recalculated sCT and CBCT plans for one patient. DVH appearance and PTV D mean was found, but was reflected in the point dose of the femoral head. In the CBCT data set, the cavity was depicted in the images but was not present in the dose comparison. We hypothesize that this is due to the higher scattering in the CBCT data set, resulting in a smoothing effect of the cavity in the CBCT data set. The rectum DVH metrics had a larger overall range of dose deviations, which is a phenomenon that has been observed previously. 14 This could be explained by the presence of air cavities in the rectum, which are not generated in the synthetic CT, for the reasons outlined above.
In a previous publication validating dosimetric accuracy and clinical robustness of MriPlanner TM , an overshoot was found between CT and sCT comparisons, 14   As an attempt to test the QA procedure, we chose to introduce simple errors in our sCT data. The result from this simulation showed that our suggested method could detect our introduced errors except when all tissue was assigned as water. The recalculation on the original sCT and CBCT data resulted in similar results indicating that the CBCT QA tool could accurately detect the introduced error. Setting all tissues to water does not introduce large dosimetric errors in the pelvis area, which is in accordance with our result. 23 The results in our study indicate that fairly large errors must be introduced in order to reach absorbed dose deviations above 1%. The simulation of sCT data with errors performed in this study aimed at demonstrating the feasibility of using CBCT data as a QA tool in an MRI-only workflow. In order to make a full evaluation of the sensitivity of the QA tool, a more thorough simulation is needed, and recommendations on action levels developed. The action levels will depend on the impact the error will have on the delivered dose to the patient and would probably be site-specific. Careful evaluation of the potential errors that could be introduced in a sCT and their significance must be considered, although this was outside the scope of this study.  3 Focusing on dosimetric accuracy, it could be seen that the dosimetric deviation between CBCT and sCT data was smaller than the estimated level of overall dose uncertainty (5%-10%) for external beam radiotherapy. 24 The proposed QA procedure would benefit from further investigations of the type and magnitude of errors that can be detected.
This would require measurements on a larger population and a wider span of introduced errors in the sCT. Future work could include development of a QA procedure for an MRI-only workflow that does not involve dose recalculations. Although our results show that this is feasible, and in our clinic very straight forward to implement, a more general method applicable to any CBCT vendor and clinic would be beneficial. This could include the comparison of histograms and correlations to recalculated doses such as those produced in this study. With such a database, incoming sCT and CBCT data sets could be compared within the database and a theoretical expected dose deviation achieved.
Prior to the introduction of CBCT as a QA tool of MRI only in the clinic, attention should be paid to the need for using specific calibration curves and consistency in HU of the CBCT system. Therefore, an investigation of stability over time and deviation in HU between CT and CBCT system before applying any QA method based on CBCT data is needed. If several different systems are used for CBCT imaging, the variation between systems and time dependence should also be studied prior to any use of suggested method.
Using patients with corresponding CT examinations for validation of the CBCT QA is also recommended. This procedure needs to be repeated for different anatomical regions.

| CONCLUSION S
The CBCT and CT systems fulfilled the criteria enabling CBCT data to be used as a QA tool for sCT. Dose calculations based on sCT and CBCT data sets were found to have an absorbed dose difference within clinically acceptable criteria. This shows that dose plan recalculation on CBCT data set can be used as a feasible QA procedure for detection of gross errors in MRI-only radiotherapy of prostate cancer patients. The suggested method has been demonstrated to successfully detect introduced errors in a sCT.

ACKNOWLEDG MENTS
The authors would like to acknowledge the clinical staff for their excellent help throughout the study. A special thanks to Lee Ambolt for his kind contribution reading the manuscript. This work was sup-

CONFLI CT OF INTEREST
No conflicts of interest.