Spatial and dosimetric evaluation of residual distortions of prostate and seminal vesicle bed after image-guided de ﬁ nitive and postoperative radiotherapy of prostate cancer with endorectal balloon

Purpose: To quantify daily residual deviations from the planned geometry after image-guided prostate radiotherapy with endorectal balloon and to evaluate their effect on the delivered dose distribution. Methods: Daily kV-CBCT imaging was used for online setup-correction in six degrees of freedom (6-dof) for 24 patients receiving de ﬁ nitive (12 RT def patients) or postoperative (12 RT postop patients) radiotherapy with endorectal balloon (overall 739 CBCTs). Residual deviations were evaluated using several spatial and dosimetric variables, including: (a) posterior Hausdorff distance HD post ( = maximum distance between planned and daily CTV contour), (b) point P worst with largest HD post over all fractions, (c) equivalent uniform dose using a cell survival model (EUD SF ) and the generalized EUD concept (gEUD a with parameter a = − 7 and a = − 20). EUD values were determined for planned (EUD planSF ), daily (EUD indSF ), and delivered dose distributions (EUD accumSF ) for plans with 6 mm ( = clinical plans) and 2 mm CTV-to-PTV margin. Time series analyses of interfractional spatial and dosimetric deviations were con-ducted. Results: Large HD post values ≥ 12.5 mm ( ≥ 15 mm) were observed in 20 / 739 (5 / 739) fractions distributed across 7 (3) patients. Points P worst were predominantly located at the posterior CTV boundary in the seminal vesicle region (16 determined per fraction stays within 95% of prescribed dose. Common PTV margin calculations are overly conservative because after online correction of translational and rotational errors only residual deformations need to be included. These results provide guidelines regarding online navigation, margin optimization, and treatment adaptation strategies.

This was associated with a lower incidence of late rectal toxicity. 3,4 In addition, an endorectal balloon reduced intrafraction prostate motion. [5][6][7] On the other hand, the use of endorectal balloons may affect the geometrical accuracy of treatment delivery. Studies using balloon-type endorectal MRI coils demonstrated that endorectal balloons can cause deformations of the prostate. 8,9 Moreover, analyses of the interfraction variability in endorectal balloon placement relative to the prostate or bony landmarks reported substantial deviations in shape and position from the nominal geometry. [10][11][12] The intra-and interindividual dosimetric consequences of such residual deformations and repositioning inaccuracies on the delivered dose distributions in prostate radiotherapy with endorectal balloon are subject of this paper.
We have analyzed a cohort of prostate cancer patients who

2.A.2 | Treatment planning
Computer tomography (CT) imaging for treatment planning was acquired in supine position with emptied rectum and half-filled urinary bladder (approx. 200 ml). All patients were imaged and treated with an endorectal balloon fabricated in-house in two sizes (small: 10 cm length, 3.6 cm diameter, 23 patients; large: 13.5 cm length, 4.6 cm diameter, 1 patient). Prior to CT imaging and each fraction, the endorectal balloon was covered with anesthesia gel, manually inserted and inflated to a prescribed fill volume of 75 ml (small size) or 125 ml air (large size) using a 50 ml syringe. The endorectal balloon was then gently retracted towards the anal canal. The PTV was calculated as CTV expansion by 6 mm posteriorly and 8 mm in anterior and lateral direction. Treatment delivery and daily image   guidance   Treatments were delivered at a Novalis TrueBeam linac (Varian Med-ical Systems, Palo Alto, CA, USA; BrainLAB AG, Feldkirchen, Germany) equipped with a 6-dof Perfect Pitch couch top. Prior to each fraction, a low-dose CBCT was recorded with acquisition parameters optimized to balance image quality and dose with preponderance of high contrast structures, i.e. bones and the air-filled endorectal balloon (X-ray tube current: 25 mA, voltage: 125 kV, exposure: 335 mAs, exposure time: 13.425 s, full arc, CTDI value: 4.5 mGy, 2 mm slice thickness, 512 × 512 image matrix). A rigid 6-dof registration between planning CT and CBCT was performed automatically using a rectangular region of interest comprising the anterior half of the endorectal balloon posteriorly, the symphysis anteriorly, the periprostatic tissue and the obturator internus muscle laterally, the seminal vesicles superiorly, and the penile bulb inferiorly ("online match").

2.A.3 |
The obtained 6-dof correction vector was used for online adjustment of the patient position using the treatment couch.

2.B | Data analysis
To assess residual setup errors not correctable by rigid registration, deformable image registrations were performed offline between the planning CT (i.e., reference image) and each CBCT (i.e., target image).
The deformable image registration software implemented in Eclipse uses a modified, accelerated demons algorithm. 13,14 Each deformable registration was based on the rigid registration of the online match.
To assess the daily variation in patient anatomy, the resulting registration vector field was used after visual inspection to propagate the planned CTV contour CTV plan from the planning CT to each of the n CBCTs performed per patient, yielding contours CTV CBCTi (i = 1,. . ., n). The CTV CBCTi contours were then rigidly copied back to the planning CT using the online match for further analysis.

2.B.1 | Spatial evaluations
For every patient, the union CTV acc of CTV plan and all CTV CBCTi was calculated. To evaluate how far the daily CTV CBCTi contours projected beyond CTV plan , a series of evaluation PTVs was generated with isotropic margins of 2 to ≥20 mm (step width 1 mm) around CTV plan (PTV xmm ), and shells of 1 mm width were derived. To geometrically assess the dislocation of the daily CTV contour, the absolute volume of each CTV CBCTi outside of PTV xmm (x = 2,. . .,≥20 mm) was determined. From these data, we derived for all patients and every fraction the Hausdorff distance HD iso (superscript iso: isotropic), i.e. the largest of all the distances from a point on CTV plan to the closest point on CTV CBCTi , as a measure of the maximum shift of the daily CTV with respect to CTV plan . Because the posterior CTV margin is of particular interest regarding rectal toxicity, we separately evaluated the dislocation of the daily CTV contour towards the rec- PTV margins were calculated using the recipe proposed by van Herk 15 : To ensure a minimum dose to the CTV of 95% for 90% of the patients, a CTV-to-PTV margin of 2.5ÁΣ+1.64Áσ'-1.64Áσ p is required, where Σ denotes the total standard deviation (SD) of preparation (systematic) errors, σ' the total SD of execution (random) errors combined with the penumbra width, and σ p the SD describing the penumbra width, and σ' 2 = σ+σ p 2 .

2.B.2 | Dosimetric evaluations
Because the workflow for accumulation of full dose distributions over a treatment series is not supported by Eclipse, the dose delivered to selected points of interest per fraction was determined to enable dose summation at these points. For that purpose, the frac-

2.C | Statistical analysis
Statistical analyses were performed in SAS (version 14.1, SAS Institute, Cary, NC, USA) and SPSS Statistics (version 22, IBM, Armonk, NY, USA). All statistical tests and procedures used are specified together with the results. P-values were two-sided, and P < 0.05 was considered statistically significant.  Table S1).

3.A | Size of planned and accumulated volumes
PTV size, bladder and endorectal balloon volumes were not different in the two patient groups.  Table 1. Additional parameters characterizing the distributions including 50% and 95% quantiles are provided in Table S2.    Table 1. Table S4 provides Table S5. Generally not only the maximum values but also the mean posterior shifts were larger at point P worst than at the other points of interest in the respective patient, except for patient 7, points P 3 and P 5 .

3.D | Time series
The time series of posterior shifts at point P worst over all fractions are shown in Fig. 7(a)  Very similar results were derived for the time series of HD post (Fig. S5), with the exception that no autocorrelation was found for patient 110. Specifically, nonstationary or random walk could be T A B L E 2 Absolute EUD SF -and gEUD-values for a = −20 and a = −7 of the original and accumulated dose distributions and resulting EUD differences for clinical treatment plans (6 mm PTV-margin) and plans with 2 mm PTV-margin.   Our analysis of interfraction time series data, based on both spatial and dosimetric variables, could exclude nonstationary or random walk and, instead, revealed a white noise characteristic. This finding is in contrast to the intrafraction prostate motion which has been reported to be a random walk. [29][30][31] In general, no autocorrelation, i.e. no similarity between observations as a function of the time lag between them, was observed. Since the time series are stationary, the expectation value is constant over time. However, the white noise amplitude differs significantly from patient to patient. On repeat planning CTs. 25,26,37,38 Only in two studies, an endorectal balloon was used. 36,37 The majority of the studies analyzed patients treated with definitive RT. One study evaluated patients receiving adjuvant RT of the prostate bed after prostatectomy. 32 While a 3-4 mm posterior margin was considered as too small in some studies, 25,32,33,35 it was found adequate in others. 26 compensate residual variations with a daily correction technique. 33 However, none of the above studies investigated residual deviations after full 6-dof IGRT with online correction of both translations and rotations. In contrast to other studies, we evaluated not only the accumulated dose distribution, but also individual treatment fractions to assess their dose contribution, because this may provide guidelines regarding online navigation and adaptation strategies.
For any dose accumulation study, the accuracy of the algorithm to calculate the deformable image registrations is a matter of concern.
The Eclipse algorithm which we used was evaluated according to the recommendations of the AAPM Task Group No. 132. 39  sis. An average residual error of 1.7 mm between the landmarks identified on the planning CT and on the deformably registered CBCT was derived. This is an acceptable accuracy for the described usage.
According to the commonly used margin recipe proposed by van Herk, 15 a posterior PTV margin of 8.8 mm (9.7 mm) is required to ensure a minimum dose to the CTV of 95% for 90% (95%) of our patients. This margin is markedly larger than the 6-mm margin proven to be sufficient for our dataset. A combination of several factors may explain the discrepancy: Most importantly, our data showed that a criterion on the minimum dose is not mandatory to maintain the target EUD SF within tight limits when reasonable cell survival model parameters are chosen. Moreover, time series analysis of the dose at point P worst showed outliers not predicted by normal distributions. Hence, the prerequisites for the computation of the cumulative dose distribution underlying the margin recipe are not fulfilled. 15 Van Herk's formula infers the PTV margin for the individual patient from random deviations of the target position and the scatter of the systematic mean deviations over all patients. However, we demonstrated that residual deviations quantified by various spatial and dosimetric parameters differ significantly from patient to patient (Table S3). Applying uniform margins, derived under the inclusion of worst case patients, to the population will be overly conservative for the majority of patients. Our analysis has shown that a 2-mm margin is still sufficient to maintain EUD SF > 90% (95%) in 75% (58.3%) of the patients. When daily image guidance is used, the systematic setup error derived from a patient population is not relevant for the individual patient, instead the daily individual systematic deviation matters. It is reasonable to determine the PTV margin for the 80%-90% of patients with smaller deformations and use larger margins with or without adaptive replanning for the remainder, as identi-