Seminal vesicle inter- and intra-fraction motion during radiotherapy for prostate cancer: A review

E.M.C

One of the common treatment modalities for prostate cancer (PCa) is external-beam radiotherapy [1]. Considering the relatively low alpha/beta ratio for PCa [2,3], hypofractionation could yield higher tumour control rates with acceptable genitourinary and gastrointestinal toxicity rates [4]. Dose-escalation has shown improved treatment outcomes [5] and the use of modern imageguidance techniques, like fiducial markers, have lowered the margin needed around the prostate and thereby lowered side effects [6,7]. Furthermore, multiple randomized trials on low and favourable intermediate risk PCa reported a non-inferiority regarding tumour control and toxicity rates of moderate hypofractionation [8,9] and ultra-hypofractionation [10] compared to conventional fractionation schemes. Here, ultra-hypofractionation is defined as a dose per fraction of 5 Gray (Gy) or more.
A next logical step would be the use of ultra-hypofractionation for high-risk patients, but this is challenging as the entire seminal vesicles (SV) are normally included in the target volume [11]. The SV belong to the male reproduction system and are about 3-5 cm long and 1 cm in diameter [12], however their exact shape and size can differ substantially. The SV are attached bilaterally to the prostate on the cranioposterior side and they lie superior to the rectum, inferior to the fundus of the bladder and posterior to the prostate [12]. The motion of the SV, similarly to the prostate, is caused by changes in bladder and rectal filling status. The SV can show tumour involvement [11], the probability of which can be predicted with the use of nomograms [13,14]. Recently, the addition of MRI imaging to these clinical prediction tools was shown to increase the robustness of these models [15][16][17].
Due to their inter-and intra-fraction motion, the SV require a relatively large planning target volume (PTV)-margin [18][19][20][21], which in combination with a high fraction dose could result in unacceptable dose to the organs at risk and thereby higher toxicity rates. A number of papers have recently been published showing the feasibility of ultra-hypofractionation (5 fractions of 7 Gy or 7.25 Gy) in small groups of patients including the SV in the clinical target volume (CTV) using different treatment modalities [22][23][24].
To safely introduce ultra-hypofractionation for high-risk PCa patients, strategies to optimize PTV-margins around the SV are tainties that contribute to a PTV-margin is crucial in this process. The last review on this topic was published in 2001 [25]; since then several articles have been published with methodologies that are more in line with the current technological advancements in PCa treatment. Therefore, this article critically reviews all relevant existing literature since 2001 on the inter-and intra-fraction motion of the SV during external-beam radiation of PCa with the aim of improving the understanding of these treatment uncertainties, which is needed to design adaptive treatment strategies for PTV-margin reduction.

Search strategy
In collaboration with the Erasmus MC Medical Library, Embase, Medline, Web of science Core collection, Cochrane CENTRAL register of trials and Google scholar were searched for relevant publications. This search was first performed on the 7th of February 2020 and last updated on the 18th of January 2021. There were no restrictions regarding date of publication or language in the initial search. See Appendix A for the detailed search queries.

In-and exclusion process
These searches yielded 646 unduplicated results. All articles before 2001 were excluded as the last review on this subject dates from 2001 [25] and the image guidance for prostate treatments has changed significantly since then. Using Endnote (version X9 build 12062), these results were screened on title/abstract and full text afterwards. This was done by VB with MM as second reviewer.
Publications that were not written in the English language, as well as publications without a specific record of SV motion, deformation, volume changes and/or PTV-margins were excluded. Publications with abstracts referring to quantitative values for motion, deformation, volume changes and/or margins of the prostate and the SV were eligible for full text screening. This yielded 170 publications.
Translations, rotations, deformations, volume changes and/or margins of the entire SV had to be reported for inclusion in the final review. Studies in which the prostate and the SV are combined in a single CTV or PTV and analysed as such were excluded, as well as studies which only incorporated part of the SV. After screening done by VB and MM, one article on volume changes was added outside of this search. In total 22 publications were included in this review

Data extraction
The general data extracted from the publications, if provided, were the author and year of publication, number of patients, number of scans (planning and repeat scans), average patient age, fractionation scheme and tumour stage. The extracted data regarding the SV motion were image modality, specific inter-or intrafraction motion, reference point of motion, type of image registration used, rectal and bladder preparation, motion surrogate used (e.g. centre of mass (COM) of the SV), and finally the SV displacement in the form of the mean, the standard deviation (SD) and the systematic and random errors, in mm or degrees, along the 3 principal axes. If present the, anisotropic, PTV margins were also extracted.
Hence, we have used the same formalism to calculate PTV margins, if not reported, from datasets. To compare publications reporting systematic errors to those reporting means of motion, for the latter the standard deviation of the group mean was used as the systematic error. A limitation for the application of this margin recipe to SV is the lack of inclusion of rotations and deformations. The margins stated in this review are based on conventional fractionation schemes and will have to be adjusted when hypofractionation is used. Another limitation is that this recipe is only valid for conventional fractionation schemes, with a need to increase the margin when hypofractionation is used. Therefore, the margins in this review only indicate a lower limit. Table 1 summarizes the error parameters used in this study.

Results
The number of publications regarding inter-fraction motion, volume changes and/or margins exceed those reporting intrafraction motion, volume changes and/or margins by 19 to 4. One article describes both. The number of patients ranges between 9 and 90 and the number of scans used for data collection ranges between 21 and 771. Multiple image modalities (CBCT, CT or MRI) have been used as well as multiple points of reference (bony anatomy, prostate or first image in series) to which the motion was measured. A range of protocols to control bladder and rectal filling, such as the use of laxatives [28,30,45,47] or instructions to drink a certain amount of water before treatment [28][29][30]32,34,38,47], have been employed in the studies. Table 2 summarizes the 22 articles included in this review. Fig. 1a shows the mean values for inter-fraction translation, M inter , of the SV. 7 out of 9 articles used a prostate match (matched on fiducials or COM of the prostate) to obtain these values. The means were derived from relative values, i.e. negative and positive directions of translation. As expected from unbiased studies, M inter is below or around 1 mm, with the exception of two articles that reported values up to À3.3 mm [35,40]. Fig. 1b shows the mean intra-fraction translation of the SV, M intra . The reported intrafraction translation depends strongly on the reference point and shows values of À1.5 up to 7 mm. Only two publications [35,44] reported the intra-fraction translation, both relative to the prostate, with values ranging from À0.4-1.2 mm.
The systematic error for inter-fraction translation, P inter , is shown in Fig. 1c. These systematic errors vary from 1 to 7 mm with only 2 publications reporting values above 4 mm. Higher values for systematic errors were reported in the anteroposterior (AP) direction, 1.7-7.3 mm, and craniocaudal (CC) direction, 1.3-4.5 mm, compared to the left-right (LR) direction, 1.0-2.0 mm with one outlier of 3.6 mm [27]. The systematic errors obtained from a match on bony anatomy [30,42] appear to be larger, all show values >3 mm, than the systematic errors obtained from a prostate match of which 4 out of 6 publications show values <3 mm. Fig. 1d shows the systematic errors of intra-fraction translation, P intra . The range shown, 1.6-4.1 mm, is smaller to that of P inter , 1.4-7.3 mm. However, these datasets use different reference points: P inter is reported relative to the prostate and other reference points, whereas P intra is only reported relative to other reference points than the prostate. Besides translations, rotations also have an impact on treatment uncertainty. Two studies were identified in which SV rotationsrelative to the prostate -were analysed. First, van der Burgt et al. [46] reported on inter-fraction rotations of the whole SV, after a prostate match. Three groups of 30 patients, with each 8 CBCTs, were divided by level of SV invasion: none, minimal (<5 mm) and extensive (>5 mm). Means, systematic (R rotation ) and random (r rotation ) rotations were given around the LR, CC and AP axis. The means of the rotations in the LR-axis for the minimal and extensive group, 2.0°and 2.3°respectively, and the CC rotation for the extensive group, 1.0°, were significantly different from 0. The systematic and random errors of the LR rotations were found to be higher, ranging from 5.0°À6.7°, compared to the rotations in the CC and AP-axes, ranging from 1.8°-2.4°and 1.6°-2.7°respectively. Two rotations were significantly lower in the extensive group compared to the no invasion group: R rotation in AP-axis (1.6°vs 2.3°respectively) and r rotation in LR-axis (5.2°vs 6.3°). Secondly, de Boer et al. [28] analysed inter-fraction rotations around the LR-axis for 20 patients with repeat CBCTs. They found a mean rotation around the LR-axis of À0.4°, a R rotation of 7.2°and a r rotation of 6.4°. These rotations were significantly correlated (p < 0.001) with prostate translations in the CC and AP direction and with prostate rotations around the LR and AP axes.
Apart from translations and rotations, deformations are also considered a source of uncertainty in the treatment of SV. Deformations of the SV were discussed in 5 publications of which one reported intra-fraction deformation and one reported both intraand inter-fraction deformation. The deformations were measured after a prostate match in all cases. Sheng et al. [44] described intra-fraction deformations of 15 patients with 5 pairs of CBCT (before and after treatment). Mean edge-to-edge distance in millimetres (with 95% data range) for Left, Right, Cranial, Caudal, Anterior and Posterior border were reported to be <1.1 mm. Li et al. [35] reported on both intra-and inter-fraction deformation. Similarly, all intra-fraction deformations were reported to be <1.1 mm. In contrast, the inter-fraction deformations showed values up to 2.8 mm (caudal border) and À2.9 mm (posterior border). Interfraction deformations were studied by Hollander et al. [33] in 10 patients with weekly verification scans (66 scans in total). They found mean edge-to-edge displacements < 0.6 mm of all borders, except for the anterior border with a deformation of 2.4 mm (-3.9-8.8 mm). Van der Wielen et al. [47] reported inter-fraction deformations for 21 patients with 3 repeat CT scans. Standard deviations along local surface normals for lateral SV, SV tip, Anterior SV and Posterior SV were 1.7 mm, 2.3 mm, 2.4 mm and 2.6 mm respectively. Lastly, Mayyas et al. [39] studied 10 patients with 20 CBCTs and looked at percentage of CBCTs in which deformation vector fields exceeded 3, 5 or 10 mm. For both 3 and 5 mm poste- Table 1 Definitions of used error parameters; x = individual measurements; n = number of measurements per patient; N = number of patients in the study; m p = mean per patient; SD p = standard deviation of the patient mean.

Error parameter Definition
Patient mean (lÞ  The anisotropic PTV-margins, including both the margins reported in the publications and the margins calculated by us using the van Herk formula, are shown in Fig. 1f. PTV-margins based on systematic and random errors are reported to be around 8 mm, a value widely used in clinical practice for the SV [19,20]. Larger values (>9 mm margins) were found in 2 out of 8 articles with reported PTV-margins up to 10.5 mm [38] and 14.9 mm [42]. When the systematic and random errors are measured relative to the bony anatomy, the PTV-margins are larger than those based on a prostate match. Due to the absence of published random errors, no margins correcting purely for intra-fraction motion can be/are reported. Fig. 2 shows the different proposed PTV margins, maximum stated values isotropically applied, in a typical prostate patient case.
To visualize the effect of preparation protocols on the motion of the SV, in the form of the inter-fraction mean, systematic error, random error and PTV-margin were plotted for rectal and bladder preparation protocol (figs. C.1 and C.2, Appendix C). From these figures, no apparent trend between rectal and bladder preparation and magnitude or direction of motion was observed. Similarly, the effect of rectal and bladder preparation on intra-fraction motion is inconclusive.

Discussion
This review focuses on understanding the inter-and intrafraction motion of the SV during external-beam radiation therapy of PCa and the PTV-margins needed to correct for this motion. This is required to devise safe PTV volume reduction strategies to enable the ultra-hypofractionated treatment of high risk PCa. The literature reported in this review show an extensive variety in methods used for obtaining and reporting motion, making a secondary analysis or generating average values not possible.

Inter-vs intra-fraction translations
The mean inter-fraction translation, M inter , of 7 out of 9 publications is below 1.5 mm, suggesting a limited group mean error. Two publications report means above 2 mm, both of which report on a small set of 10 patients [35,40]. For the intra-fraction translation 2 of the 3 available publications report means, M intra , up to 1.5 mm.
Regarding study [32] reporting means of up to 7 mm, it remains unclear whether relative or only absolute values of translation were reported, as well as which reference point was used. Overall, the values of M inter and M intra are comparable and in the order of 1 mm, which would be expected from unbiased data. However, especially on intra-fraction translation, the number of publications are still limited with only four studies. The systematic inter-fraction error, P inter , shows values between 1 and 3.5 mm. Note that the P inter values reported by Chin et al. [27] and Frank et al. [30] were derived from the SD of the M inter that was given. There are only two reports that discuss the systematic errors of intra-fraction translation, P intra [29,32]. Both articles report motion of 1.5-4 mm. These values for P intra were derived by us by using the SD of the group mean. Comparing Seminal vesicle inter-and intra-fraction motion during radiotherapy for prostate cancer: A review anisotropic PTV-margins of 2-5 mm and 4-10 mm after a prostate vs a bone match respectively. Similarly, Meijer et al. [18] reported an isotropic margin of 13 mm when matched on bony anatomy compared to 8 mm when matched on the prostate. This indicates that there is at least some level of correlation between the interfraction motion of the prostate and of the SV. Similar to the prostate, the motion of the SV is caused by changes in rectal and bladder distention. However, the reported levels of correlation between prostate and SV motion vary. All publications note that the SV can move semi-independently from the prostate and the amplitude of motion is larger. Smitsmans et al. [45] reported that as much as 42% of the AP SV inter-fraction translation was correlated to the LR prostate gland rotation. Similarly, Liang et al. [36], showed that inter-fraction translation of the prostate and the SV in the AP direction was correlated (R 2 of 0.7), both driven by rectum and bladder changes. No correlations were found for the other directions. De Boer et al. [28] show an inverse correlation between the LR rotations of the SV and the prostate LR rotation. A large interpatient variety in correlation of intra-fraction SV and prostate translation was shown by Gill et al. [32]. The reported Pearson correlation coefficients, R, ranged from À0.23 to 0.82, with the 7 out of 10 patients showing no linear correlation trend. Consequently, imaged-guided strategies that only focus on the prostatic gland will not fully compensate for SV motion.
Gill et al. [32] showed that there appears to be a plateau in SV displacement that was reached 10 min after starting radiation delivery. No significant increase in displacement was seen after this time. It is unclear how this corresponds exactly to on-table time. De Muinck Keizer et al. [49] report that the extent of intrafraction motion of the prostate is reached after 30 min on-table time.

Rotations, deformations and volume changes
Means of inter-fraction rotations were discussed by two publications [28,46], most of which <1°, as expected in unbiased data, with only three rotations in one publication [46] significantly different than 0. R rotation and r rotation around the LR-axis were also reported in both articles and range between 5.0°and 7.2°. Van der Burgt et al. [46] reported that these rotations around the LRaxis were larger than the rotations around the CC and AP-axes with max rotations of 2.4°and 2.7°respectively. This is in line with Hoogeman et al. [50] who described that prostate+SV rotations were largest in the LR-axis with R rotation and r rotation of 3.6°and 5.1°respectively. These rotations were significantly correlated to differences in rectal volume (p < 0.0001) [50].
Mean deformations after prostate match were mentioned by five publications [33,35,39,44,47]. All five reported the highest deformations in the anterior, caudal and posterior borders. This is somewhat in line with the largest translations being in AP and CC axis and the largest rotations being around the LR-axis and can be explained by rectal and bladder volume changes as well [47]. Mean intra-fraction deformations, <1.1 mm [44], appeared to be smaller than the mean inter-fraction deformation, <3 mm [33,47]. This was also described by Li et al. who reported both [35]. Mayyas et al. appear to report higher deformations than the previously mentioned articles (1% >10 mm) [39]. However, this can be explained by the fact that only Mayyas et al. [39] did not use means to report their deformations. Important to note here is that all deformations mentioned are measured after prostate matching. No residual deformations after correction for SV translation were described. The publications reporting on the magnitude of volume changes show different results, which can be, partially, explained by the different experimental methods. Where Miralbell et al. [41] used the planning scan as reference, Bairstow et al. [26] used the mean SV volume as a reference. The latter also only reported on two extreme cases from their population, where Miralbell et al. [41] reported on all 9 patients. Liu et al. [37] reported a study in which 28 patients with at least 15 follow-up CT-scans were analysed. The volume, compared to the planning scan, decreased significantly in 3 cases and increased significantly in one case. In contrast, Frank et al. [31] found no significant volume changes in 15 patients with repeated CT-on-rails images. As these varying results suggest, no consensus regarding the extent of these volume changes has been reached and further research is needed to clarify the geometrical and clinical effect of these volume changes [26].

PTV-margins to account for SV motion
The three studies reporting PTV-margins [34,40,42] all used the van Herk margin recipe (Eq. 1) [48] except for one article which used an alternate version for a 2D dose distribution: 2.15 R + 0.7 r [45]. This review compares these reported PTV-margins, with margins we calculated from reported systematic (R) and random errors (r) using the same van Herk margin recipe [48]. One publication reports both systematic and random errors and a PTVmargin. The margins recalculated by us are very similar to those reported [42] (i.e. 4.4 vs 4.4 in LR, 10.0 vs 9.9 in CC and 7.4 vs 7.5 in AP respectively).
Most publications included in this review report anisotropic PTV-margins for the SV of approximately 8 mm (see Fig. 2). This value is also used in multiple studies looking at the effect of margins on target coverage. Meijer et al. [18] showed that an isotropic PTV-margin of 3 mm for the prostate and 8 mm for the SV ensures 95% CTV-coverage for 90% of the patients using a prostate fiducial match. Mutanga et al. [19] reported that an isotropic 8 mm expansion for the SV resulted in a clinically acceptable coverage. Thörnqvist et al. [21] found that an isotropic PTV-margin of 7 mm resulted in 95% coverage of the target volume for 18/19 patients. Stenmark et al. [20] looked at the coverage for the proximal 1 cm as well as the entire SV. For 95% geometrical coverage of the CTV for 90% of the patients 5 mm and 8 mm isotropic margins were required when treating the partial SV and the full SV respectively.
Two publications reported SV margins > 9 mm, one of which used a bony anatomy match to register the SV motion [42]. Using a prostate match, Mak et al. [38] reported larger margins, i.e. 10 mm in the CC-direction, possibly limited by the 5 mm CT slice thickness in this direction.
Sheng et al. [44] reported a 5 mm isotropic margin around the SV to ensure a 95% coverage in 90% of the fractions. However, this margin assumes intra-fraction motion tracking of the prostate. The literature on intra-fraction motion of the SV is still too limited to extract a PTV-margin based on intra-fraction motion alone.
Translation consistently appears to be the smallest in the LR direction ( Fig. 1a-f). This offers opportunities of anisotropic PTVmargins. Smitsmans et al. [45] reported margins for the SV of 4.6 mm and 7.6 mm for the LR-and AP-direction respectively, not taking into account deformation and rotation. In addition, the rotations of the SV are largest around the LR axis, which will mostly contribute to motion in CC and AP direction. Most dosimetric studies report isotropic PTV-margins in the order of 8 mm [18][19][20][21]. Margin reduction in the LR-direction might have a limited clinical impact, considering most toxicity comes from the bladder and rectum that lie inferior and superior to the SV.

Influencing factors
Bladder and especially rectal volume changes are known to play a significant role in prostate inter-and intra-fraction motion [25]. For the SV similar patterns of correlations between rectal and bladder filling and SV motion have been observed [27,29,30,38,51]. However, fig. C.1, in Appendix C, shows that different efforts to control the rectal filling status do not have a clear effect on the amplitude of inter-and intra-fraction motion. Similarly to rectal filling status, no apparent trend is visible in the amplitude of SV motion, with respect to bladder preparation ( Fig. C.2, in Appendix C). The absence of a correlation between rectal and bladder filling protocols and the amplitude of systematic and random errors in our study can be, at least partially, explained by their mixed success rate to effectively control the filling status, as shown by a review on this topic [52].
Only one study reported on the effect of tumour invasion on SV motion. Van der Burgt et al. [53] compared the differences in interfraction motion between patient groups with different levels of tumour invasion in the SV. The random displacements in the group with extensive invasion were statistically significantly lower than those of the minimal and the no invasion group. However, this reduction was small and the SV motion remained considerable.

Limitations
There are several limitations in the van Herk margin recipe as in equation 1 that are relevant in applying the margin formula to SV. First of all, only translations are taken into account. Rotations, deformations, and volume changes all contributing to errors in the treatment of SV, are ignored. Studies that do include rotations show that rotational errors can cause a loss in tumour control probability [54], especially for non-spherical targets [28]. An example can be found in de Boer et al. [28] who state a margin of 11.6 mm including rotational errors of the SV and 8.2 mm when correcting for them. Including rotations will lead to anisotropic and location specific margins as the margin will be dependent on the distance to the rotation axes, generally assumed to lie near the apex of the prostate [55,56]. As the correlation between the prostate and SV rotations is limited, there is a residual deformation of the SV in the order of 2-3 mm SD that needs to be taken into account [30]. Hence, deformations, rotations, and volume changes that are not fully corrected for before the start of treatment lead to an increased PTV margin to ensure CTV coverage and the van Herk recipe will only give a lower limit of the margin required. Another limitation is that the van Herk margin is valid for conventional fractionation. To translate the results from the referenced publications to a hypofractionated treatment scheme, the margin will have to be increased. In a treatment consisting of only a few fractions, the average random error might deviate from zero, resulting in an additional systematic error [55,57,58]. As an indication, the PTV margin will have to be increased from 8 mm to 8.5 and 9.2 mm respectively for a 5 and 2 fraction treatment, based on a calculation using equal systematic and random errors.

Possibilities for margin reduction
With conventional image guided radiotherapy (IGRT) [59] PTVmargin reduction for the SV has been achieved, but remains with 8 mm substantial [18][19][20]. Further margin reduction with IGRT might be difficult to achieve and therefore ultrahypofractionation for patients with a target volume including the SV remains challenging.

Correcting for inter-fraction motion
Inter-fraction motion can be corrected off-and online by adaptive radiation therapy (ART). ART for prostate has been extensively studied and reported [60][61][62][63][64][65][66]. However, only a limited amount of publications on ART for the SV exist. Xia et al. reported on a library-of-plans approach [67] whereas De Boer et al. [28] used a hybrid registration technique, prostate markers followed by a soft-tissue registration of the SV. Both showed promise in possible margin reduction around the SV. However, most recent research regarding prostate ART and margin reduction still focusses on prostate only and is fuelled by the developments of MR-guided radiation treatment systems [68,69] Correcting for intra-fraction motion In contrast to inter-fraction motion, intra-fraction motion is more complex to take into account. A straightforward solution is to minimize fraction duration as the displacement increases with time [29,32,35]. Intra-fraction motion correction of the prostate has been demonstrated using Calypso 4D tracking [70], real-time tracking using the CyberKnife [71], a library of plans [72], and soft tissue gating using the MRidian [69]. However, the challenge remains how to apply intra-fraction motion management for adjacent targets, in this case the prostate and the SV, that move semicorrelated. Beam-per-beam online replanning with all its challenges could pose a solution [68,73].

Conclusion
This extensive literature review shows that the inter-and intrafraction motion of the SV is substantial and largely uncorrelated with prostate motion. Main factors influencing the prostate and SV motion are differences in rectal and bladder filling. Strategies to control rectum and bladder filling status, and thereby reduce treatment uncertainties, appear to lack effectiveness. When calculating PTV-margins for the SV, translations, rotations and deformations need to be taken into account as they can be substantial, even after an initial match on the prostate. To reduce PTV margins around the SV, their inter-and intra-fraction motion needs to be adequately accounted for. Further research is required to quantify the safety and feasibility of PTV-margin reduction for the SV, in particular in context of ultra-hypofractionation for high risk prostate cancers, which will be subject of further studies in our institute.

Support
This work was in part funded by a research grant of Accuray Inc., Sunnyvale, USA.

Conflicts of interest
This work was in part funded by a research grant of Accuray Inc., Sunnyvale, USA. Erasmus MC Cancer Institute also has a research collaboration with Elekta AB, Stockholm, Sweden and Varian Medical Systems, Inc., Palo Alto, USA.