Adaptive radiation therapy strategies in the treatment of prostate cancer patients using hypofractionated VMAT

Abstract Purpose To perform a comprehensive evaluation of eight adaptive radiation therapy strategies in the treatment of prostate cancer patients who underwent hypofractionated volumetric modulated arc therapy (VMAT) treatment. Material and methods The retrospective study included 20 prostate cancer patients treated with 40 Gy total dose over five fractions (8 Gy/fraction) using VMAT. Daily cone beam computed tomography images were acquired before the delivery of every fraction and then, with the application of deformable image registration used for the estimation of daily dose, contouring and plan re‐optimization. Dosimetric benefits of the various ART strategies were quantified by the comparison of dose and dose‐volume metrics derived from treatment planning objectives for original treatment plan and adapted plans with the consideration of target volumes (PTV and CTV) as well as critical structures (bladder, rectum, left, and right femoral heads). Results Percentage difference (ΔD) between planning objectives and delivered dose in the D99% > 4000cGy (CTV) metric was −3.9% for the non‐ART plan and 2.1% to 4.1% for ART plans. For D99% > 3800cGy and Dmax < 4280cGy (PTV), ΔD was −11.2% and −6.5% for the non‐ART plan as well as −3.9% to −1.6% and −0.2% to 1.8% for ART plans, respectively. For D15% < 3200 cGy and D20% < 2800 cGy (bladder), ΔD was −62.4% and −68.8% for the non‐ART plan as well as −60.0% to −57.4% and −67.0% to −64.0% for ART plans. For D15% < 3200 cGy and D20% < 2800 cGy (rectum), ΔD was −11.4% and −8.15% for non‐ART plan as well as −14.9% to −9.0% and −11.8% to −5.1% for ART plans. Conclusions Daily on‐line adaptation approaches were the most advantageous, although strategies adapting every other fraction were also impactful while reducing relative workload as well. Offline treatment adaptations were shown to be less beneficial due to increased dose delivered to bladder and rectum compared toother ART strategies.


INTRODUCTION
Radiation therapy is a major treatment option for patients with prostate cancer. However, variations in the patients' anatomy during radiotherapy treatment can present challenges. Numerous studies show that the variable location of the prostate, bladder filling, and pockets of gas often present in the rectum might significantly compromise dose coverage of the target structure and increase the dose delivered to critical organs. [1][2][3][4][5] Hypo-fractionated radiotherapy, delivering fewer, higher fraction doses increases the dosimetric impact of anatomic variability compared to conventional fractionation schemes. For prostate cancer patients, moderate hypofractionated (70 Gy in 28 fractions, 2.5 Gy/fraction) intensity modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) were proven to keep early normal tissue toxicity at acceptable levels. 6,7 More aggressive hypofractionation delivering 33.5-37.5 Gy in five fractions has also been shown to achieve acceptable toxicity and quality of life, 9,10 even for high risk and very high risk (including node-positive) prostate cancer patients.
The safe and effective delivery of high radiation doses in hypofractionated schemes requires a high level of precision, but inter-and intra-fractional patient anatomical variation is present and known to compromise dosimetric aspects of the treatment. 8,9 Adaptive radiation therapy (ART) strategies, in particular online ART, have the ability to account for systematic anatomic changes of prostate swelling as well as random anatomic changes such as inter-and intra-fraction bladder and rectal filling, in addition to independent movement and deformation of multiple targets. 8,10 The necessity and the benefits of ART application in stereotactic body radiation therapy (SBRT) prostate treatments have been shown in other recent studies. [9][10][11][12] It should also be noted that existing image guided radiation therapy (IGRT) techniques, although allowing for prostate motion management, have some limitations. For example, they might require an invasive procedure carrying the risk of bleeding, infection, and discomfort for the patient (radiopaque intraprostatic fiducial markers method). Another prostate IGRT example is a technique that utilizes inserted electromagnetic transponders. In this case, patient eligibility criteria are very strict as only patients without hip prosthesis, metal implants, peacemaker, or other electromagnetic devices are eligible, as well as relatively thinner patients due to the maximum range of the beacon detection by a required external array (reading) device. 13 Adaptive radiation therapy is currently an active area of research, and there are still many novel ART approaches that have not been explored yet but could make a significant contribution to the field. The current study focuses on a comprehensive evaluation of several ART methods that have not been explored for the prostate VMAT hypofractionation schemes examined here. The purpose of this research was to retrospectively investigate eight adaptive radiation therapy strategies (including both online and offline scenarios) for hypofractionated VMAT treatments based on imaging and treatment plan data of 20 prostate cancer patients with the application of deformable image registration (DIR). The online and offline adaptations considered were compared to the non-ART (not adapted) delivery scenario.

Patient data
The imaging and treatment planning data for twenty prostate cancer patients, with an average age of 77 (±7) years were retrospectively used for this study. The study was approved by the local research ethics board (University of Manitoba). All 20 patients had previously received a 40 Gy/5 fraction treatment regimen and were treated at Can-cerCare Manitoba. One pre-treatment, planning computed tomography (pCT) imaging scan, and five sets of on-treatment cone beam CT (CBCT) imaging scans were obtained for each patient. CBCT images were acquired during each treatment fraction right before radiation delivery to ensure proper patient positioning. Anatomic structures considered in the dosimetric analysis included the clinical target volume (CTV), the planning target volume (PTV = CTV+0.5 cm margin) as well as organs-at-risk (OARs) -bladder, rectum, and femoral heads. An experienced radiation oncologist segmented these structures on both CT and CBCT imaging data sets. These structures were also used for plan adaptation and optimization purposes.
The pCT images, at 512 × 512 pixels, were obtained with a spatial resolution of 1.17 mm × 1.17 mm per pixel and 3.0 mm slice thickness (total of ∼210 slices) on a Philips Brilliance Big Bore CT scanner. The CBCT images, at 384 × 384 pixels, were obtained with a spatial resolution of 1.17 mm × 1.17 mm per pixel and 2.5 mm slice thickness (total of 64 slices) using an OBI Cone-Beam CT unit (Varian Medical Systems, Palo Alto, CA).

Dose delivery and treatment planning
The treatment was delivered using VMAT with two full arcs. Every patient was treated with a full bladder and empty rectum as per local clinical protocol. The intent of the radiation therapy was curative for all patients.  Figure 6b for CTV and PTV as well as in Figure 7b for Bladder and Rectum. Femoral heads (right -RT; left -LT) were not included in the plan evaluation as the radiation doses for all plans were significantly below the planning objective thresholds. Abbreviations: CTV, clinical target volume; PTV, planning target volume.
Treatment plans were created in the external beam planning module of the eclipse treatment planning system, version 13.6 (Varian Medical System, Palo Alto, CA, USA). Dose and dose-volume objectives for the radiation treatment are summarized in Table 1. The beam energy and the maximum dose rate for both arcs were 6MV and 1000MU/min (SRS mode). All the non-ART plans were normalized to the dose received by 95% of the PTV volume. Specifically, the dose was determined based on the 95% of the PTV volume on the dose volume histogram of the original treatment plan. The non-ART plans were not normalized.

DIR
Daily CBCT images were acquired before the delivery of every fraction. Planning CT images were then registered to CBCT data sets using a Bspline-based 14 automated DIR algorithm available in Velocity AI, version 3.2 (Varian Medical System, Palo Alto, CA, USA). Deformed pCT images ('dCT') were then used for daily dose estimation with respect to the treatment plan (adapted and non-adapted) that was delivered to the patient during a particular fraction. Since our study was retrospective, the plan delivery was simulated (i.e.,it was not actually delivered to the patient). Contours that allowed for the estimation of the dose delivered to the considered anatomical structures were delineated by an experienced physician on CBCT and then propagated to daily dCT image scans. To adapt to the current patient anatomy, and as needed for some of the strategies studied here including adaption while accounting for the dose delivered in the previous fraction(s), the subsequent fraction treatment plans were re-optimized simulating offline or/and online scenarios.

Plan optimization
For the optimization of VMAT plans in this study, the progressive resolution optimizer, 15 version 10.0.28 was used, while for dose calculation, the AAA algorithm (v.10.0.28) was utilized with a 0.25 cm calculation grid resolution. The total dose delivered to the patient after performing a given plan adaptation was estimated by mapping daily doses back to the reference (planning CT) image using an inverted deformation vector field obtained through DIR and then by performing dose accumulation using the Velocity AI software. The objectives were consistent and unmodified throughout the optimization process relative to the Planned plans.

Adaptive radiation therapy strategies
For this study, online, offline, and dose feedback (DF) approaches were examined by simulation using the available daily anatomical CBCT data set. Online plan adaptations were simulated to occur immediately before a dose delivery while offline modifications were simulated to occur between fractions n and n+1. DF adaptation was simulated as an offline strategy and utilized the dose delivered in the previous fraction to guide a plan adaptation (re-optimization) for the next fraction. Overall, eight adaptive radiation therapy strategies were simulated for all patients as described below.
1. DF 2-4 -A combination of the non-ART treatment plan and DF adaptation. In the 1st, 3rd, and 5th fraction, the non-ART plan was delivered. In the 2nd and 4th fraction, the non-ART plan was reoptimized based on the dose delivered during the previous fraction. The reasoning behind performing reoptimization only during the 2nd and 4th fraction is as follows: The second fraction is the first fraction that can use the feedback from the dose delivered in the previous (1st) fraction. During the third fraction, the non-ART treatment plan was delivered because the dose delivered during the 2nd fraction accounted for dose discrepancies resulting from dose delivery during the first fraction, and thus up to the 3rd fraction the plan was assumed to be delivered optimally. Therefore, the time-consuming DF adaptation was not used during the 3rd fraction. To examine the performance of DF applied to every fraction (for example with the availability of in vivo patient dosimetry), we have also tested a continuous DF adaptation (Cont.+DF approach). The dose delivered in the previous fraction was incorporated in the optimization process for the current fraction using the "dose-based"plan optimization module of eclipse. Before the start of the optimization process, the dose delivered in the previous fraction was mapped to the patient's anatomy of the current fraction using DIR. Once the optimization was initiated, the optimizer compensated for regions of lower than or higher than intended dose by delivering a higher/lower dose to those regions, so that the total accumulated dose delivered during the previous and the current fraction would meet the treatment plan objectives. The application of the DF in the subsequent plan adaptation scenarios was performed in the same manner. 2. Offline -based on the offline adaptation of the non-ART treatment plan. In the 1st fraction, the non-ART plan was delivered. To deliver the dose during the 2nd, 3rd, 4th, and 5th fraction, the treatment plan was adapted by plan re-optimization using the dCT from the previous fraction (obtained based on the previous fraction's CBCT). As it implies, in this adaptation scenario, only changes in the patients' anatomy detected on the daily CBCT image data set relative to the planning CT images were accounted for. The dose delivered during the previous fraction was not considered as described in approach (i). 3. Offline + DF -based on the combination of offline adaptation of the non-ART plan and DF adaptation.
In the 1st fraction, the non-ART plan was delivered. In the 2nd fraction, the treatment plan was re-optimized using the daily image data sets (daily dCT) and the dose delivered during the previous fraction (DF adaptation). In the 3rd fraction, the dose was delivered using an offline adapted plan (based on the previous fraction image data sets). In the 4th and 5th fractions, DF and offline adaptation were applied, respectively. 4. Online -based on the daily online adaptation of the non-ART plan. In all fractions, before the daily dose was delivered, the treatment plan was re-optimized according to the patient's anatomy just before treatment delivery.

5.
Online + DF -based on the combination of daily online adaptation and DF adaptation. In the 1st, 3rd, and 5th fraction, online adaptation was performed, while in the 2nd and 4th, fraction DF adaptation was performed. 6. Cont. + DF -based on the combination of continuous DF adaptation. In the 1st fraction, the non-ART plan was delivered. During the remaining, 2nd-5th fractions, the treatment plan was adapted using the patient's daily anatomy. The plan re-optimization was performed based on the total dose delivered over all the previous fractions. 7. Online 1-3-5 -based on the daily online adaptations and non-ART treatment plan. In the 1st, 3rd, and 5th fraction, the treatment plan was adapted using an online approach, while in the 2nd, and 4th fraction, the non-ART treatment plan was delivered. 8. Offline+Online -based on the combination of online and offline adaptations. In the 1st, 3rd, and 5th fraction, the treatment plan was adapted using an online approach while in the 2nd, and 4th fraction using offline plan adaptation.
The purpose behind creating various adaptive radiation therapy strategies was to find an optimal solution for treatment plan adaptation which minimizes the negative impact of changes in the patient's anatomy on the accuracy of the delivered dose,while also minimizing the time it would take to perform such adaptations in a clinical environment. For example, the rationale behind examining the Online 1-3-5 strategy, where online adjustments of the treatment plan were performed every second fraction instead of every fraction, was to decrease the total time of adaptation relative to a full Online strategy, where online adjustments of the treatment plan were performed every fraction. Another example is a DF 2-4 strategy (where DF = DF). Incorporating a DF step allows accounting for potentially inaccurate dose delivery in the previous fraction and the change in the patient's anatomy because optimization involving the DF step is performed on the daily imaging data. Although the implementation of those steps will increase total treatment time, it is expected that the improved accuracy in the dose delivery will justify the additional workload.
In this study, we define the reference non-ART plan as the one that was created based on the pCT data and was delivered at every treatment fraction without modifications (dose delivered was calculated based on the CBCT and mapped back to the reference pCT image). This is an estimate of what dose is actually delivered by the conventional non-ART approach.
Planned delivery reflects the intended (ie. prescribed) dosimetry of the treatment plan,as approved by the radiation oncologist. TA B L E 2 Quantitative metrics used for evaluation of adaptive radiation therapy strategies. Apart from all listed metrics, maximum dose (D max ), mean dose (D mean ), and minimum dose (D min ) for each structure were determined as well

Bladder & Femur-LT & CTV & PTV Rectum Femur-RT
D v% -minimum dose delivered to the "hottest"v% of the volume,D1cc -minimum absolute dose for the "hottest" 1cm 3 of the volume, V d% -a volume that received d% or more of the prescription dose expressed as a percentage of volume, HI -homogeneity index calculated as a ratio (D 2% -D 98% )/D 50% , 24 CI -conformity index calculated as a ratio V 95% /Volume of PTV. 25 Abbreviations: CTV, clinical target volume; PTV, planning target volume.

Evaluation of adaptive radiation therapy strategies
The dosimetric effectiveness of adaptive radiation therapy strategies was evaluated using a variety of dose and dose-volume metrics for target and OARs as specified in Table 2. Metrics were selected based on the relevant literature to ensure general applicability. [16][17][18][19][20][21][22][23] The values of each metric were associated with anatomic structures, ART strategies, and the reference plan. Where applicable, the evaluation metrics for ART plans were presented relative to the reference plan as well to quantify the dosimetric benefit of applying plan adaptations compared to the situation where the adaptations were not incorporated in the treatment process.
The calculation of the percentage of plans that passed the treatment planning criteria for CTV, PTV, and OAR structures was reported as well.Due to their clinical relevance, the percentage deviations from the treatment planning objectives were also reported. Importantly, our conclusions with respect to superiority and inferiority of particular ART strategies relative to other dose delivery approaches were mostly driven by analysis of treatment planning objectives. Specifically, the larger the passing rate, the more clinically feasible we considered a given ART approach. For CTV and PTV, the smaller the absolute percentage deviations from the planning objectives were considered better. For OARs, the smaller percentage deviations were considered better. The large num-ber of other dose-volume metrics ( Table 2) that we have included in our study provide a more comprehensive view on the dosimetry of all considered plans but were not considered in the exact treatment planning objectives used to derive the plans (these are included in Table 1).
The statistical significance of the results comparing the non-ART plan to all the other plans was determined using paired t-tests, using the p-value associated with a 95% confidence level. The time efficiency of the best performing ART method was also reported. The time required for each ART strategy was estimated with the consideration of plan optimization, dose calculation, and DIR procedures.

Qualitative assessment of image registration
In order to ensure that the image registration did not introduce any major errors in terms of the patient's anatomy deformations, a qualitative (i.e., visual) evaluation of deformed images and deformable vector fields was performed. In particular, the deformed images were compared to the target images by using image overlays, checkboard filters, dynamic magnifying window focusing on soft tissue and bone tissue alignment as well as external body contour, for every registration. The analysis of the vector fields included the inspection of the deformed grid that reflected the magnitudes and directions in the field. The inspection was conducted using tools available in the commercial image registration software Velocity AI (as specified in the Section 2.3.1). Figure 1 shows the percentage differences relative to the non-ART (i.e., reference, not adapted) plan in D max , D mean, and D min for CTV, PTV (Figure 1a) as well as in D max and D mean for OARs (Figure 1b). D min for OARs was calculated as well but due to their limited applicability were not included in the results. Appendix (Table  A1) contains the detailed tabular data for Figure 1 including D min for OAR and standard deviations for all metrics.

Maximum, minimum, and mean doses
All the metrics in Figure 1a indicate that values of D max , D mean , and D min in the case of CTV and PTV for adapted plans were closer to the original (Planned) plans compared to the unadapted reference plan. Overall, in terms of dose metrics reported in Figure 1a, continuous DF adaptation outperformed other ART strategies and had a performance close to online adaptations. F I G U R E 1 Comparison of D max and D mean , and D min metrics between the Planned plan as well as non-adaptive radiation therapy (ART) and adapted plans for (a) target structures and (b) for OARs. Bars represent percentage differences (averaged over 20 patients) in particular metrics relative to the Planned treatment. The measure of 0% on the y-axis is the reference point reflecting the value of a given metric for the Planned plan (i.e., reference plan).
The Online and Online 1-3-5 plans scored as well as Planned plans in terms of the maximum dose delivered to PTV. The most apparent difference between adapted plans and the dose delivered by the non-adapted plan was reflected in the value of the minimum dose to PTV. In particular, the unmodified plan resulted in the delivery of approximately 17% lower D min to PTV compared to the planned minimum dose. Most adapted plans significantly improved the delivered dose in terms of this evaluation metric. Overall, it can be noticed that non-ART plans delivered a lower radiation dose to both CTV and PTV in terms of D max , D mean , and D min . Figure 1b for the OARs shows that D max for both bladder and rectum in the non-ART plan differed from the planned dose by around 5% (decrease). However, adapted plans were able to closely match the Planned plans decreasing the difference in D max to around 1%-2%. In contrast, the mean dose for bladder and rectum showed larger variations for the adapted plans relative to the non-ART plan. In the case of the bladder, the majority of adaptations increased the D mean by less than 5%. Only Cont.+DF plans escalated the mean dose by around 6%.Online and Online 1-3-5 plans were able to slightly reduce the D mean for the bladder relative to the planned dose. The rectum received approximately 5% lower mean dose upon delivery of the non-ART plan compared to the planned dose. As can be seen, Cont.+DF, Offline+DF, and online strategies were able to decrease that difference to roughly 0.5%-2%. The dose to the femoral heads was spared the most through the application of Online and Online 1-3-5. As for the right femoral head, those two online adaptation techniques delivered maximum doses significantly smaller even compared to the planned dose.

F I G U R E 2
Comparison of dose-volume metrics between the Planned plan as well as non-adaptive radiation therapy (ART) and adapted plans for the clinical target volume (CTV) and planning target volume (PTV). Bars represent percentage differences (averaged over 20 patients) in particular metrics relative to the Planned treatment. The measure of 0% on the y-axis is the reference point reflecting the value of a given metric for the Planned plan.

F I G U R E 3
Comparison of homogeneity index (HI) and conformity index (CI) between the Planned plan as well as non-adaptive radiation therapy (ART) and all adapted plans for clinical target volume (CTV) and planning target volume (PTV). Bars represent percentage differences (averaged over 20 patients) in the particular metrics relative to the Planned treatment. The measure of 0% on the y-axis is the reference point reflecting the value of a given metric for the Planned plan.
However, the Cont.+DF adaptation resulted in higher than intended dose to both left and right femoral heads -a 6% and over 10% increase in the mean dose, respectively.

Dose-volume metrics
3.2.1 Target structures Figure 2 shows the relative values of dose-volume metrics that were calculated within the evaluation of adaptive radiation therapy strategies for the CTV and PTV. Overall, the dose delivered with the various adaptive strategies was consistently closer to the planned dose compared to the non-ART approach. In that regard, Offline+Online and Offline plans demonstrated the lowest while Cont.+DF and online adaptations demonstrated the highest dosimetric performance for both target structures. The PTV benefited from ART more than CTV as shown by D 95% , D 98% , and D 99% , metric. Notably the values of these three metrics for non-ART plans were approximately 4% lower for CTV and as much as 11% lower for PTV compared to Planned plans. Tabular data for Figure 2 with standard deviations are included in the Appendix (Table A2).
When it comes to homogeneity index (HI) for CTV (Figure 3), the majority of adapted plans improved the homogeneity (i.e., a lower HI indicates a higher homogeneity level) of the dose distribution relative not only to F I G U R E 4 Comparison of dose-volume metrics between the Planned plan as well as non-adaptive radiation therapy (ART) and all adapted plans for bladder and rectum. Bars represent percentage differences (averaged over 20 patients) in particular metrics relative to the Planned treatment. The measure of 0% on the y-axis is the reference point reflecting the value of a given metric for the Planned plan. The charts (a) and (b) were separated for better visualization of the results due to the large differences in y-axis values between V 80% , V 95%, and the rest of the metrics. the non-ART plan but also to the Planned standard treatment.Online and Online 1-3-5 delivered the highest benefit. For PTV, HI was also improved among all adapted plans. Specifically, online and DF plans performed similarly and outperformed offline strategies by around 20%-40%. Compared to the non-ART dose delivery, CI for the planning target volume was approximately 10% higher when adapted plans were utilized. Online ART, in particular, very closely matched the Planned treatment. According to most metrics presented in Figure 3, Offline and Offline+Online adaptations were not meaningfully beneficial to the radiation treatment dosimetry.

3.2.2
Organs at risk Figure 4 illustrates the dose-volume metrics for the bladder and rectum.In the case of the bladder,D 1% ,D 1cc ,and D 2% did not differ significantly from the planned dose for adapted plans but were lower by approximately 7% for non-ART plans. Only Cont.+DF, and a few Offline adaptations, showed slightly higher values compared to the Planned delivery. Larger variations in magnitude were observed in V 15% , V 20% , and V 50% metrics. The most desirable results were obtained through Online and Online 1-3-5 strategies. Cont.+DF adaptive plans showed poor performance with large volumes receiving 15%, 20%, and 50% of the prescription dose as can be seen in Figure 4a. The same observation can be made in Figure 4b showing V 80% and V 95% metrics for the bladder. Considering results for the rectum, most ART strategies were able to closely match Planned values of D 1% , D 1cc , and D 2% (Figure 4a) as well as values of V 80% and V 95% (Cont.+DF, Online+DF, Online and Offline+DF in Figure 4b). For V 15% and V 20% , the majority of adapted plans delivered nearly the same results F I G U R E 5 Comparison of dose-volume metrics between the Planned plan as well as non-adaptive radiation therapy (ART) and all adapted plans for left and right femoral heads. Bars represent percentage differences (averaged over 20 patients) in particular metrics relative to the Planned treatment. The measure of 0% on the y-axis is the reference point reflecting the value of a given metric for the Planned plan. The lack of expected symmetry in the dose delivered by adapted plans to both femoral heads is explained in the Discussion section. except for Cont.+DF for which approximately 2% volume increase was noted for 15% and 20% dose prescription levels. Compared to adapted, the non-ART plans were closer to the Planned plans for those two volume metrics. V 50% for Planned treatment was approximately the same as for DF 2-4 and Online+DF adaptations. The Cont.+DF approach resulted in V 50% being around 7% lower compared to the Planned delivery. Figure 5 demonstrates that for the left femoral head only the Online 1-3-5 plans were able to keep the D 1% , D 1cc , D 2%, and D 5% at the level close to the Planned treatment. All the other plan modifications, except for non-ART resulted in doses higher than the Planned plans by around 4%-11%. In contrast, the majority of adapted plans (except for Cont.+DF and DF 2-4), especially Online and Online 1-3-5, deliver a lower D 1% , D 1cc , D 2%, and D 5% by up to 5% to the right femoral head compared to Planned dose delivery.

3.3.1
Target structures Figure 6a demonstrates the percentage of patients for whom a given treatment plan met the treatment planning objectives specified in Table 1 (Section 2.2) for CTV and PTV structures. For CTV, all plans had at least a 90% passing rate except for non-ART plans for which no patients passed CTV or PTV planning objectives. One hundred percent of patients passed the CTV criteria in the case of DF 4-2 and three Online adaptations. For the PTV, these same four strategies were able to achieve a 60%-80% passing rate for the D max objective, while other ART approaches received 50% and lower rates.
The D 99% criterion for PTV was very hard to reach even for well-performing Online adaptations. The maximum passing rate was achieved by Online ART and was equal to slightly above 20%. Figure 6b details the dose difference, ΔD, between the considered criteria and the value achieved by the plan. For the CTV, nearly all the plans were able to meet the treatment planning objective, and ΔD is positive ranging from 1%-4%. The dose difference in D max for the PTV was approximately 1% for investigated ART approaches. It is noted that even though the passing rate for offline adaptations was lower in comparison to the rest of the ART strategies, the ΔD is, on average, positive for Offline+DF, Offline+Online, and Offline plans. This clearly shows that several patients delivered higher doses to the CTV so that it was able to cause the increase in the dose averaged over all 20 patients. The ΔD for D 99% objective for PTV ranged from around −4% in the case of Offline plans to approximately −2% for Online adaptations. Compared to all the ART approaches the non-ART plans differs significantly from the Planned plans by −4% to −11% depending on the planning criteria.
In summary, the implementation of the majority of the ART strategies improved the overall passing rate and ΔD for most of the plans, especially for daily online adaptations compared to the delivery of an unchanged non-ART treatment plans.

Organs at risk
The passing rate presented in Figure 7a is equal to 100% across all the plans for the bladder in the case of both plan objectives. Consistently, Figure 7b shows that ΔD for the bladder is significantly (around 60%) below F I G U R E 6 (a) Passing rates for clinical target volume (CTV) and planning target volume (PTV) for three treatment planning objectives. (b) Dose differences between the planning objectives and the dose delivered by the specific treatment plan.
tolerance doses (D 15% and D 20% ). In the case of the rectum, the passing rate for three Offline adaptations and the non-ART plan was in the range of 60%-85% for the D 20% metric and from 80% to 100% for the D 15% metric. For Online plans, the analogous range was from 80% to 95% for D 20% and from 95% to 100% for D 15% . ΔD was negative for all the plans and had nearly the same magnitude (of 10%) for most of the adaptations for the D 15% and D 20% criteria. The absence of femoral heads in this analysis is addressed in the discussion section.

Statistical significance
As mentioned in the Materials and Methods section, the statistical significance of the results was calculated using paired, two-tailed t-tests. The analysis was based on the comparison of dose and dose-volume metrics for target and OARs structures between the non-ART plan and the other treatment strategies for all 20 patients.The relevant p-values with a 95% confidence level are presented in Table 3 for D max , D min , and D mean as well as in Appendix (Table A3) for the remaining metrics. Both tables also summarize the total number of metrics for which the test determined the statistical significance at the level of p < 0.05. Table 3 shows that the majority of the results were statistically significant for the non-ART treatment. Among ART plans the lowest p-value was observed for Cont.+DF and offline adaptations, while the highest were observed for the remaining plan modification strategies. The results for PTV and CTV indicate a similar level of statistical significance. When considering organs at risk, the results for the rectum demonstrated a statistical significance similar to that of the bladder with the exception of minimum dose for which results corresponding to bladder were more significant compared to the rectum.

Time efficiency
The time required to manually complete key steps in the ART loop in the clinic was: (i) plan optimization ∼4 min 30 s.; (ii) dose calculation ∼1 min 30 s.; (iii) DIR ∼1 min. 50 s, for a total time of around 7 min 50 s per online plan adaptation for a single fraction. The limitation of our time estimation is that it did not include CBCT image acquisition, data processing, and possible verification step that may be required for a newly adapted treatment plan.

Qualitative assessment of image registration
Thorough visual inspection of image registration results did not reveal any major, non-physical image deformations that could negatively impact the dosimetric results of ART strategies explored in this study. The alignment of the soft and bone tissues as well as external body contours that were inspected with overlays, checkerboards filter and dynamic magnifying window confirmed a high quality of image registrations. The deformation vector field was smooth without folding distortions, indicating that only realistic deformations of the patient anatomy occurred during the registration process.

DISCUSSION
Offline and Offline+Online plan adaptations resulted in the highest delivered dose to CTV and PTV compared to other ART strategies which was demonstrated by the majority of dose-volume metrics ( Figure 2).As explained in the Material and Methods section, Offline adaptation relies on the patient's anatomy from the previous fraction to modify the treatment plan that will be delivered in the next fraction. The possible issue in that approach is that if the magnitude of interfractional changes in the patient's anatomy is significant, the offline adapted TA B L E 3 The results of paired, two-tailed t-test for D max , D min , and D mean. The fields for which 0.01 < p ≤ 0.05 are highlighted in green (i.e., significant), while those with p ≤ 0.01 were highlighted in red (i.e., strongly significant)  Figure 2 also demonstrates that relative to CTV the dose coverage of the PTV is more sensitive to anatomical changes. The D 98% and D 99% for that structure were on average 5% and 6% lower for the unmodified plan. The importance of proper PTV coverage highlights the significance of ART application, as it has proven its ability to improve and maintain PTV coverage. Overall, both the dose and dose-volume metrics shown in Figures 1  and 2 show that non-ART plans delivered consistently lower doses to CTV and PTV compared to the Planned and all the ART plans.
Regarding the HI index (Figure 3), although we found that for CTV the daily plan re-optimizations were generating satisfactory dose distributions, their combination in the dose accumulation step resulted in an even higher level of homogeneity for the majority of patients. Despite the promising results in the HI index for CTV, the accumulated dose distribution needs to be carefully examined in clinical practice. This is because the dose accumulation can result in the appearance of hot and cold spots depending on the spatial relation between daily dose distributions. This effect is most likely responsible for the presence of relatively high doses delivered to the bladder by the Cont.+DF adaptation as seen in Figure 4a (D 1% , D 1cc , and D 2% ) and Figure 4b (V 80% and V 95% ). It is also worth mentioning that for Online adaptations, a clinically acceptable value of HI was not associated with an increased dose in OARs. This is contrary to the findings of Banaei et al. who conducted a study based on 15 prostate cancer patients that were delivered IMRT treatment. 27 Researchers reported inverse exponential relationships between the OAR sparing and HI, which might be the case due to the differences in dose delivery techniques. Banaei et al. used nine static IMRT beams, while this study utilized VMAT techniques with two arcs. Chow et al. showed that in the case of prostate cancer, VMAT compared to IMRT provides more desirable PTV coverage in terms of both HI (0.09 vs. 0.12) and CI index (0.94 vs. 0.89). 28 It should also be noted that based on the data presented in the Figure 4, the non-ART plans delivered lower radiation doses to both bladder and rectum compared to all other delivery approaches. Although it can be seen as a desirable outcome, the non-ART plans are not optimal because as mentioned in the previous paragraph, both CTV and PTV coverage significantly differ from the planned dose.
It is also important to notice that as shown in Figure 5, the dose delivered by adapted plans to the right femoral head was noticeably lower than for the left femoral head. It is not very clear why the dose sparing is not approximately symmetrical for these two structures. The authors suspect that because the dose delivered to femoral heads was significantly lower than planning objective threshold, the observation of asymmetry is a form of systematic noise and an indication of limited priority given to those structures by the optimizer during plan re-optimization. The direction of the first arc may also contribute to this, as the second arc (in the opposite direction) typically has a lesser dose impact than the first arc (i.e., the second arc "fine tunes" the dose). In the early adoption of VMAT in clinical practice, Hardcastle et al. 29 reported such asymmetry but admitted that the reason was also not clear, further adding that gantry rotation direction did not affect the asymmetry of the dose distribution. Also, Tran et al. 23 indicated that the left femoral head received a higher dose than the right femoral head however due to the fact that dose-volume objectives were met, this observation was not discussed. Figure 6a clearly shows the advantage of ART application in that the passing rate for D 99% in CTV was around 100% for the Online, and Online+DF adaptations compared to 0% for the unaltered non-ART plan. The 100% passing rate for the bladder for all plans, as shown in Figure 7a was anticipated because VMAT plans have been proven to be able to decrease the dose delivered to that organ very effectively (large ΔD∼60% on Figure 7b) in prostate cancer radiotherapy. 30 Limiting the dose to the rectum by the application of various ART strategies was more challenging (small ΔD∼10% in Figure 7b,passing rate as low as 65% for D 20% in Figure 7a) due to the position, size, and the increased daily movement of the rectum. Also, the Online approach resulted in the highest passing rate for rectum planning objectives.The analysis of treatment planning objective passing rates and dose deviations ΔD shows that Online and Online 1-3-5 strategies are very promising for ART adoption in the clinical environment. It can be noticed that the DF approaches resulted in the lowest absolute values of ΔD (Figure 6b) for target structures; however, compared to online strategies they are significantly more resource intensive thus may not be an optimal choice in practical ART applications.
It is also interesting to observe that the passing rate for PTV objective (D99% > 3800cGy, Figure 6a) was very low. In particular non-ART, Online 1-3-5, DF 2-4, Offline+DF, Offline+Online, Offline all had zero or close to zero passing rates. Figure 6b shows why this might be the case. We can see that even the original (Planned) plans barely meet that criteria (ΔD is almost equal to zero; 0.02%). This means that any, even the smallest random error related to any aspect of treatment delivery, patient positioning etc. could easily invalidate this particular objective. In our case, the source of this error could be very small random contouring variability. Only the most resource intensive strategies (Cont.+DF, Online+DF, and Online) were able to slightly mitigate this discrepancy.
One of the study limitations is that the comprehensive quantitative evaluation of the DIR algorithm as well as the spatial relationship between the registration error and its impact on the accuracy of dose estimation was beyond the scope of this work. However, it is expected to be a relatively small effect compared to the impact of the various adaptation strategies. 31 Table 3 shows that the statistical significance of the results is, in general, higher for Online 1-3-5 plans compared to Online ART. Although overall the Online adaptation delivered a dosimetric performance slightly closer to the initially intended plan, the Online 1-3-5 strategy is 40% faster due to fewer adaptations required. The trade-off between the time efficiency and the dosimetric results presented can be useful for both busy clinics and centers with larger time allocation per treatment plan.
As mentioned in the introduction, ART is an active area of research; however, the number of papers reporting the comprehensive evaluation of various adaptation scenarios is limited. Often authors study just a single or few approaches. 32-36 Therefore, we believe that our comprehensive approach brings value in evaluating ART focused on the two full arcs hypofractionated VMAT treatments for prostate cancer patients considered in this study. The evaluation of a variety of ART strategies will help to easier identify an ART approach that is best suited for individual clinics.

CONCLUSIONS
The aim of this research was to investigate and quantify the dosimetric benefits resulting from the application of several different adaptive radiation therapy strategies for hypofractionated VMAT treatments for prostate cancer patients. The findings of our work quantify these improvements and indicate that performing daily online adaptations, every fraction or every second fraction improves the dosimetric outcomes of delivered radiotherapy treatment compared to the plan that was created solely based on the pre-treatment planning CT scan and was then delivered without accounting for interfractional changes in the patient's anatomy. The strategy of adapting every second fraction achieves nearly the same dosimetric benefit to the patient but with significantly reduced resources used and may represent the most clinically attractive strategy examined here for significantly hypofractionated prostate cancer patients.

AC K N OW L E D G M E N T S
The authors would like to acknowledge the financial support of the Natural Sciences and Engineering Research Council of Canada (discovery grant) as well as the University of Manitoba.

C O N F L I C T O F I N T E R E S T
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.

TA B L E A 3
The results of paired, two-tailed, t-test for D max , D min , and D mean . The fields for which 0.01 < p ≤ 0.05 were highlighted in green (i.e., significant), while those with p ≤ 0.01 were highlighted in red (i.e., very significant) Abbreviations: ART, adaptive radiation therapy; CTV, clinical target volume; PTV, planning target volume.