Stability of daily rectal movement and effectiveness of replanning protocols for sparing rectal doses based on the daily CT images during proton treatment for prostate cancer

Abstract Purpose To evaluate the optimal period of replanning to spare the rectal dose by investigating daily rectal movements during computed tomography (CT) image‐guided proton therapy for prostate cancer. Materials and methods To evaluate the optimum reference period for replanning, we analyzed 1483 sets of daily CT (dCT) images acquired from 40 prostate cancer patients and measured the daily rectal movement along the anterior–posterior direction based on the simulator CT (sCT) images and dCT images. We calculated daily dose distributions based on initial plans on the sCT images and replans on the dCT images for 13 representative patients, and evaluated daily dose volume histograms (DVHs) for the prostate, seminal vesicles, and rectum. Results The rectal anterior side on the dCT images around the seminal vesicles largely deviated toward the anterior side relative to the position on the reference sCT images, but the deviation decreased by referring to the dCT images and became nearly zero when we referred to the dCT images after 10‐day treatment. The daily DVH values for the prostate showed good dose coverage. For six patients showing rectal movement toward the anterior side, the daily rectal DVH (V77%) showed a 3.0 ± 1.7 cc excess from the initial plan and this excess was correlated with 9.9 ± 6.8 mm rectal movement. To identify the patients (37.5% in total) for whom the replanning on the 10th‐day and 20th‐day CT images reduced the V77% excess to 0.4 ± 1.5 cc and −0.2 ± 1.3 cc, respectively, we evaluated the accumulated mean doses with a 1.2 cc criterion. Conclusion Our data demonstrate that the daily movement of the rectal anterior side tends to move toward the anterior side, which results in a rectal overdose, and the mean of the movement gradually decreases with the passage of days. In such cases, replanning with the reference CT after 10 days is effective to spare the rectal dose.


| INTRODUCTION
Proton therapy is now regarded as a common application for prostate cancer, and the numbers of patients undergoing proton therapy have continued to increase due to this therapy's benefit of the sharp distal dose fall-off beyond the Bragg Peak, which spares organs at risk (OARs). The use of proton therapy is thus expected to reduce the rates of gastrointestinal (GI) and genitourinary (GU) acute and late toxicities compared to the dose distribution provided by conventional photon radiotherapy. 1 In Japan, patients with prostate cancer undergo fractionated proton radiotherapy with a prescribed dose of 74-78 Gy in 37-39 fractions and 70 Gy in 28 fractions as unified protocols. Clinical trials have been conducted to test the effectiveness of hypofractionated proton therapy using a delivered dose of 63 Gy in 21 fractions with the goal of increasing the treatment throughput and achieving higher tumor control. Since the treatment period still lasts 4-8 weeks, highly precise image guidance for the daily placement of radiation with respect to the target is a key technique to take advantage of the physical selectivity of proton therapy.
This has motivated particle therapy vendors to provide 3D volumetric image guidance as a standard function, enabled by cone beam computed tomography (CBCT) or in-room CT imaging in the current therapy facilities. 2 With these setups, in principle, one can identify the target and OAR locations on daily CT images and validate the effect of daily changes of patient anatomies on the dose distributions. This could not be done with the two-dimensional (2D) conventional kV x-ray image guidance used in most particle therapy facilities in the past.
With the use of opposed lateral beams as is now done widely for proton prostate cancer therapy, the movements of pelvic organs along the beam axis matter for the target dose coverage, and the movements along directions lateral to the beam axis matter for the dose to OARs and thus for the GI and GU acute and late toxicities. The stability of the pelvic organs' daily positions and the stability of the shape of the anterior rectal wall placed at nearby targets over the superior-inferior (SI) direction are important factors in the control of the daily dose to the rectum. In this context, we previously investigated the effects of daily organ motion on prostate treatment for ten patients by using daily CT images acquired throughout the patients' in-room CT image-guided proton treatment. 3 The results demonstrated that daily movements of the anterior rectal wall along the anterior-posterior (AP) direction from the referenced simulator CT images tended to increase from the inferior side to the superior side, due to the daily rectal deformation. The resulting averaged positional deviation was 5 mm at the inferior side and 15 mm at the superior side around the seminal vesicles (SVs). Thus, if we take these averaged positional deviations as a margin for the creation of the planning organ at risk volume (PRV) of the rectum, the resulting rectal dose value does not meet the dose constraint, for example, V 60Gy < 18%, which is used as the one of our rectal dose constrains 4 and was reported originally by Nagoya city university. 5 In principle, uniform irradiation to the target volume with the same dose formation as that used in the original planning is not sufficient to maintain the daily dose under the planning conditions for a deformable OAR, for example, the rectum. One of the best approaches to this issue is the use of online adaptive planning in combination with CT image guidance, and simulation studies using daily CT images acquired during the treatment have thus been conducted. [6][7][8] However, further research is needed to examine the contouring and quality assurance of online planning, and this must be done for clinical practice. A replanning protocol based on the statistical knowledge of daily anatomical movements over treatment periods may provide much simpler and more realistic clinical applications compared to the online adaptation of radiation formation to daily random changes. To address this concern, a method of plan selection based on library data sets was proposed for dealing with the daily positional changes of patients' pelvic anatomies. 9 In the present study, we performed novel analyses to evaluate the replanning protocol for rectal dose sparing based on daily CT images acquired for image guidance prior to proton beam irradiation.
In our clinical experience with proton prostate treatment using inroom CT image guidance, the daily positions of the prostate and the anterior rectal wall identified on online daily CT (dCT) images tended to show differences from the positions in the simulator CT (sCT) images acquired at approx. 10 days pretreatment. We created a replan by referring to the new CT images for particular patients if the dose validation showed its necessity.
We thus studied the stability of the daily movements of the anterior side of the rectal wall by referring to dCT images with four times higher values compared to our previous findings. 3 In this work, first we evaluated the mean, the random errors, and the systematic errors of daily movement of the rectal anterior wall along the SI direction based on the same analyses method as previous work. 3 As new approach in this evaluation, the dCT images acquired from the first treatment day to the 20th treatment day were used sequentially as the reference images in addition to sCT images, and then we evaluated the optimal period of replanning by using dCT images as a reference. Second, we examined the impact of these movements on the daily proton dose, and a clinical application was simulated in order to evaluate the effectiveness of the replanning protocol on daily dose parameters.

2.A | Patient data, immobilization, and CT image acquisition
This study was approved by the Institutional Review Boards of our hospitals. We analyzed the dCT images and sCT images that had been acquired for 40 patients undergoing prostate cancer treatment. 4 All of the patients gave written informed consent to participate in this study. Among them, 27 patients diagnosed with low-risk prostate cancer were treated with a delivered dose of 74 Gy in 37 fractions, and the other 13 patients diagnosed with intermediate-tohigh-risk cancer were treated with 78 Gy in 39 fractions. For all patients, the patient positioning was done using in-room CT-image guidance without the use of implanted prostate fiducial markers or a rectal balloon. As described previously, 4 all patients were scanned in the supine position with a suction-type fixed bag (RSF-19Gl; Engineering System Co., Nagano, Japan). The sCT images and MRI images were acquired approx. 10 days before the patient's first treatment. The dCT images were acquired by a self-propelled CT scanner on rails (Aquilion LB, Canon Medical Systems, Tochigi, Japan) installed in one of the gantry rooms at the Fukui Prefectural Hospital Proton Therapy Center. The daily volume of the bladder was monitored by ultrasound scans to ensure that it was similar to the volume of the sCT images, and patients drank water when the bladder was not sufficiently distended. The defecation and exhaust gases are managed in order to maintain the condition of the rectum as that observed on the sCT images. The dCT images were acquired with a tube current of 150 mA and potential of 120 kV, while the sCT scans were acquired with a current of 480 mA, and reconstructed with a slice thickness of 2 mm and a transversal pixel size of 1.07 × 1.07 mm 2 , which were the same conditions as for the sCT scans.

2.B | Data processing and image registration procedure
We analyzed a total of 1483 sets of CT images of the 40 patients.
We created regions of interest (ROIs) for the prostate, SVs, and rectum on the sCT image and dCT images. First the ROIs of the sCT images were created by experienced radiation oncologists with the support of MRI images. Second the dCT images were registered into the sCT images by referring bony structure using a rigid image registration (MIM Maestro ver. 6.8, MIM Software, Cleveland, OH, USA), and then all ROIs were produced on the dCT images performing a deformed image registration of ROIs of the sCT images using MIM Software. Finally radiation oncologists and an experienced medical physicist carried out the manual correction and validation of ROIs on all sets of dCT images.
Then, to evaluate daily movements of the patients' anatomy in respect to the reference CT images, all dCT images with ROIs were registered to the reference CT images by simulating prostate-rectum boundary (PRB) registration as the same procedure described  4 In contrast to the approach used in the previous study, 3 we sequentially referred the sCT images and the dCT images acquired between the first day and the 20th day of the patient's treatment for the reference CT images in order to evaluate the optimum reference period for replanning. All procedures described above and those performed later with the reference CT images were repeated sequentially.

2.C | Evaluation of the interfractional movement
for the anterior side of the rectum

2.D | Mean, random errors, and systematic errors
We evaluated the mean, the random errors, and the systematic errors of the daily positional deviations of the anterior side of the rectum as a function of the scaled Z coordinate with respect to the reference. In the following analysis, as noted before, the reference images were sequentially scanned from the sCT images to the dCT images acquired from the first treatment day to the 20th treatment day, denoted by r = 0, 1-20, respectively. For each patient denoted by i, the mean value (m r i ) and standard deviation (σ r i ) were evaluated as a function of the scaled Z coordinate from the measurements among responsible fractions by , where P d,r are the daily and referenced positions of the anterior rectum along the AP direction, respectively, N denotes the total number of responsible fractions, and n denotes the total fractions. The mean value (m r ), random error (σ r ), and systematic error (Σ r ) were obtained by the mean of m r i , the root mean square of σ r i , and the standard deviation of m r i , respectively, among all patients. We also estimated the average positional deviations using the formula 2.5Σ + 0.7σ proposed by Van Herk et al. to provide the margin values with 95% confidence. 10 We also obtained the mean value, random error, and systematic error of the positional correction of the prostate along the AP and SI directions with respect to the reference number as described above to see how these values changed relative to the reference number.

2.E | Planning simulation and daily proton dose calculation
The initial treatment plan and replans were created using a proton treatment planning system (PTPS) (XiO ® -N; Elekta Corp., Stockholm, Sweden) based on a passive scattering method used in our proton therapy facility (Hitachi, Tokyo). The planning procedure and parameter settings were essentially the same as those used for the planning in our current prostate cancer treatment as described previously. 4 Briefly stated, a clinical target volume (CTV) was created by experienced radiation oncologists. The CTV for low-risk prostate cancer consists of the prostate only or the prostate and a part of the proximal area of the SVs, whereas most of the SVs were included additionally for the patients with higher risk cancer. A planning target volume (PTV) was created by expanding the CTV by 6 mm with the exception of the use of a 5-mm margin exclusively at the posterior side of the CTV. The beam isocenter was set to the geometrical center of the PTV. The rectal volume for the DVH evaluation along the slice direction was determined by the size of the CTV with the expansion of 10 mm along the SI direction, and the averaged rectal volume used for the DVH evaluation was 41.4 AE 9.1 cc. The representative rectal dose constraint of V 60Gy < 18% was maintained by tuning a collimator margin to the PTV around the posterior side, and the averaged V 60Gy in cubic centimeter was 7.3 AE 2.5 cc. We aimed to maintain V 95% = 100% for cases in which the CTV is expanded by 3 mm, where V D represents the relative volume receiving at least a specified absolute or relative dose, D. However most of the planning particularly for higher risk cancer did not meet V 95% = 100% for the expanded CTV as well as the CTV to keep the rectal dose condition due to the inclusion of the SVs in the CTV. In this case, we gave priority to maintaining the dose coverage of V 95% = 100% for the prostate and the rectal dose condition of V 60Gy < 18% while we deteriorated the dose coverage of the SVs. The replanning with the dCT images was created using the same procedure as used initially for the sCT images.
Based on the above parameters determined in the initial plan and replans, we performed daily dose calculations using all of the dCT images registered on the reference images with respect to the bony structure by applying the daily isocenter corrections along the AP and SI directions evaluated by the PRB procedure, and a delivered dose on each of the dCT images was set to be 2 Gy. The detail workflow for the daily dose calculations was described in our previous work. 4  by our PTPS was validated in several phantoms made of water, bone, and lung equivalent materials (Kyoto Kagaku Co., Ltd., Kyoto, Japan) as well as with the CT images used for the prostate cancer treatment. Overall agreement was achieved between the two calculations, but the distance of the lateral penumbra in the Axion4S software was observed to be approx. 0.5 mm smaller than the distance in our PTPS system; this resulted in values that were a few percent different in the dose volume histogram (DVH) of the rectum from the DVH of the PTPS. The dose distributions of the initial plans as well as those of the replans were therefore recalculated in the Axion4S software and used for later analyses.
The daily DVHs for the prostate and rectum were calculated for 2 Gy irradiation, and the daily changes in the value of V 95% were evaluated for the prostate and the SVs; the daily changes in the value of V 77% in cubic centimeter were evaluated for the rectum.
For the rectal V 77% evaluation, the region of the rectal volume on the dCT images along the slice direction was maintained to the one on the sCT images. Seventy-seven percent of the total delivered dose of 78 Gy corresponds to 60 Gy. For the SVs, we evaluated the daily value of V 95% using full SVs volume also for the lower risk cancer to see their deviation. We evaluated the mean, standard deviation, and range of daily DVH values of each organ for each patient over the entire treatment period.
In order to make primary decisions for the replanning according to the daily dose parameters, we also calculated the accumulated mean dose values using the following formula ∑ Nm i¼0 V i =N m , where i denotes the treatment day running from the first day up to the monitoring day, V i denotes the daily dose parameters, and N m denotes the total number of accumulated days. We evaluated the overdose criteria for the rectum for the accumulated mean value to identify problematic patients during the early period of the treatment. We also evaluated the correlation between daily rectal movement and rectal dose parameters to provide the empirical function for the dose estimation based on the online measurement of the daily rectal movement. Table 1

| DISCUSSION
In general, the initial plan of radiotherapy treatment at our hospitals in Japan is created based on the sCT images acquired at approx. conventional kV x-ray imaging, the use of CT image guidance could help address this problem by monitoring daily changes of the rectal shape and its dose parameters.
We propose an accumulated mean dose method to follow the daily dose changes. This method can be used to make a primary judgment regarding replanning on an online basis during the early period of each patient's treatment with CT image guidance. In particular, this method used during the treatment can be useful to monitor daily dose parameters for patients whose daily random variation of the anterior rectal wall is rather large and results in a rectal overdose. Our data showed that the accumulated mean dose up to 10th day of treatment and the mean dose over the whole treatment showed the linear correlation ρ = 0.97 with the residual error of 0.6 cc (RMS). Assuming Gaussian distribution on the residual, if we consider 2.5 cc as a tolerance level for the mean of the rectal overdose during whole treatment, the probability to identify these treatment on the 10th day by setting 1.2 cc tolerance threshold on the accumulated mean dose was estimated to be 97.7% due to more than two sigma separation, that is, 2.5 cc > 1.2 cc + 2 Â 0.6 cc.
Nevertheless, since the tolerance value was estimated using our limited data, we recommend that the value need to be validated before clinical application.
In combination with this method, a dose verification system dedicated to CT image guidance is desired to achieve more precise approaches to the monitoring and validation of the changes in daily dose parameters with considerations of the online image registration just prior to the patients' treatment. For this system, precise automatic contouring and deformed image registration using deep neutral F I G . 6. The differences (red squares) in the daily dose parameters from the initial planned values for V 95% of the prostate (left panels) and the seminal vesicles (middle panels) and V 77% of the rectum (right panels) throughout the entire treatment period for representative Patients #2, #7, #8, #10, and #13. Patient numbers are indicated in the left panels. Solid lines: the accumulated mean values from the first day up to the online day (indicated on the horizontal axis) for the dose parameter difference. For Patients #8, #10, and #13 categorized as the third group, the differences in the daily dose parameters from the replanning using daily CT images acquired on the 10th day (red triangles) and 20th day (red circles) are also shown. The corresponding accumulated mean values from the first day up to the online day based on the replanning using CT images on the 10th and 20th day are also shown by long dashed lines and dotted lines, respectively.
networks may be necessary in order to accumulate daily dose distributions on the reference images. 11 To our knowledge, such a system is not yet clinically available at the proton therapy facilities equipped with CT image guidance, and we are currently researching this issue.
As an empirical and alternative approach, we also provided linear regression functions based on our data sets to estimate the daily dose deviation of the rectal V 77% with the knowledge of the daily shift of the anterior rectum. Although the precision for the estimation of the rectal dose parameters is limited at 1.5 cc, this approach does not require dose calculations or anatomical contouring, and thus it can be easily incorporated into clinical workflows.
The beam scanning method has been widely adopted as a threedimensional irradiation technique in particle therapy facilities, and in principle, online adaptive replanning is possible based on this scanning method. However, technical problems remain to be resolved, that is, automatic segmentations, first planning optimization, and planning quality assurance. We propose (a) the use of an optimal period for replanning to spare the rectal dose by analyzing the pelvic anatomical movement on daily CT images acquired throughout prostate treatment, and (b) an approach to identify problematic patients for the daily rectal dose during the first period of the patient's treatment. The results of our analyses demonstrate that the replanning protocol conducted during the treatment after 10 days have passed was effective to control the target dose coverage as well as the rectal dose, and the proposed protocol is clinically applicable for prostate treatment using the scanning method as well as the passive method. However, if the daily rectal dose variation before the replanning period is rather large, higher doses may accumulate in the rectum during this period for particular patients. In such cases, stronger dose constraints of the rectum compared to the conventional protocol in replanning must be set to compensate for the accumulated rectal dose over the whole treatment, and for some patients it will be necessary to compensate for the maintenance of the rectal dose constraint due to the dose coverage of the target.
In order to avoid the above situations, dose distributions on daily CT images should be validated more frequently during the first period of the treatment, and the frequency of the replanning needs to be increased. Since the changes in the mean values of the prostate and the anterior rectal wall between the simulator CT images and the first daily CT images were observed to be the most significant, the replanning based on daily CT images during the first week of treatment can be considered to be effective for this issue and is a clinically realistic approach, particularly for treatment using a beam scanning system with in-room CT image guidance.
The applications of the rotational correction (pitch angle) over the sagittal plan in addition to two transversal corrections performed in the PRB registration for the soft tissue alignment may improve the prostate dose coverage and also reduce the daily positional deviation of the rectal anterior side around the SV region as well as SVs, which may result in sparing the daily rectal dose further. The daily random error of the prostate rotation over the sagittal plane was found to be significantly large in order of 7-10 degree. 12 Our data showed the random error of 2-9 degree. Generally the rotational movement of the patient couch over the sagittal plane is limited mechanically, and the limitation is within 2 degree in the case of our proton facility. Therefore this issue has to be considered to develop the registration method taking the prostate rotational variations into account. Moreover such registration method for the soft tissue alignment can be realized in the recent proton therapy facility with the 6 degree-of-freedom robotic couch in combination with CT-image guidance, 1 and we will study the effectiveness of the proposed replanning protocol based on the PRB registration method including rotational angle corrections in the future.
Our data showed that the daily dose coverage of the SVs varied among patients, and the replanning protocols based on the rectal movement could not improve the maintenance of the daily parameters as planned. This was because the daily movement of the tip portion of the SVs located at each of left and right side of the prostate was found to be larger than the movement at the center of the prostate, 3 and may have no correlation with the daily movement of the anterior rectal wall. In addition, the field margin around the tip portion of the SVs used for opposed lateral proton beams was not sufficient to keep the dose coverage due to the rectal dose contain.
Adaptive therapy by optimizing the aperture shape of the multileaf collimator or the proton range adjustment for anterior-oblique beams may overcome this issue in combination with in-room CT image guidance. 13 The use of hydrogel rectal spacers for proton prostate treatment is considered to be effective to reduce the rectal dose, resulting in lower toxicity by making space between the rectum and prostate.
The use of this technique can be expected to increase, but the stability of implantation as well as the absorption rate of a biodegradable compound over the treatment period must be studied by using several imaging modalities to adjust treatment plans appropriately. 14,15 In particular, the implantation stability around the SV F I G . 8. Daily dose parameters of V 95% for the prostate and seminal vesicles, and the parameters of V 77% for the rectum evaluated based on the replanning using daily CT images acquired on the 10th day (left) and 20th day (right) for Patients #8-#13 in the third category. Blue circles: the initial planned values. Red squares: mean values. Error bars: the standard deviation. Rectangles: minimum-maximum range among the daily treatments.
region is questionable, since the daily movement of the rectum around this region was found to be significant herein. However, this topic is beyond the scope of this paper and our database, and it will be examined elsewhere.
The limitations of this work are as follows. Since our findings were based on image datasets acquired only at our single facility for a limited number of patients, the stability of the rectal shape at other proton facilities might show characteristics that differ from our datasets even if the same immobilization method and pretreatment are used. We thus recommend that the position of the rectum as well as the prostate during treatment be monitored with CT image guidance, and the influence of anatomy changes on daily dose parameters should be investigated by using an approach that is similar to what we have described before the clinical application of the protocol that we have described.

| CONCLUSION
We analyzed 1483 sets of daily CT images acquired throughout the proton therapy for 40 patients, which is a fourfold greater number of image sets than used in our previous study. The daily rectal movement was measured along the AP direction with respect to the reference CT images (including daily acquired images) by simulating the PRB registration. The results of our analyses demonstrated that the daily movement of the rectal anterior side around the seminal vesicle region tends to move toward the anterior side compared to that in the simulator CT images, resulting in a higher rectal dose in daily treatment, and the mean of the daily movement gradually decreases as the days pass. Our results also demonstrated that the accumulated mean rectal dose during the treatment can be useful to make a primary online decision regarding replanning for a patient with large daily variation of the anterior rectal wall movement that results in a rectal overdose. For such cases, we observed that replanning with the reference CT image after 10 days have passed was effective to spare the rectal dose.

ACKNOWLEDG MENTS
We thank all of the staff of the Proton Therapy Center at Fukui Prefectural Hospital for their expert support. This work was supported by a grant from the Japan Society for the Promotion of Science (JSPS) KAKENHI, no. JP17K09079.

CONFLI CT OF INTEREST
None of the authors have any financial interest or personal relationships with other persons or organizations that could inappropriately influence our work.