2.1 Patient characteristics

This study included 189 patients with prostate cancer who underwent MR-Linac treatment. Of these, 17 patients without post-irradiation MRI, 15 whose treatment was interrupted due to MR-Linac malfunction, 21 with baseline IPSS > 20, 8 who received only two fractions, 1 who received a prescribed dose of 35 Gy, and 13 with large bladder-volume variations per fraction resulting in significant errors in DIR for dose accumulation were excluded. Consequently, 114 patients were included in the final analysis (low- or intermediate-risk group, n = 67; high-risk group, n = 47). Detailed patient characteristics are provided in Supplementary Table S1. None of the high-risk patients received pelvic lymph node irradiation. Low- or intermediate-risk patients were treated with 36.25 Gy in five fractions. High-risk patients received 36.25 Gy to the planning target volume (PTV) and 40 Gy to the clinical target volume (CTV) in five fractions. Nearly all patients underwent SpaceOAR (Boston Scientific Corporation, Marlborough, MA, USA) hydrogel insertion to create a separation between the prostate and rectum. IPSS assessments were performed on the first treatment day, once during treatment, and once or twice at 1–3-month intervals following treatment initiation. Data were collected using patient self-administered questionnaires.

This study was approved by the institutional ethics committee. Given its retrospective design, the requirement for informed consent was waived. All procedures were conducted in accordance with institutional and national ethical standards and regulations.

2.2 Organs-at-risk (OARs) and target contouring on MRI simulation

Reference plans were created on the simulation MRI for all patients. In the low-risk group, the CTV encompassed the prostate. In the intermediate- and high-risk groups, the CTV included both the prostate and the seminal vesicles, extending 1.5 cm from the prostate base into the seminal vesicles. The PTV margin was generally set at 5 mm in all directions. However, for patients in whom meeting bladder dose constraints was deemed difficult by the radiation oncologist or medical physicist, the superior margin was reduced from 5 mm to 3 mm. Radiation oncologists delineated the rectum, femoral heads, urethra, small bowel, and bladder as organs-at-risk (OARs). The entire bladder (including the lumen) was contoured, rather than the bladder wall alone. The urethra was contoured with a nominal diameter of 2 mm, and a 2-mm planning OAR volume (PRV) margin was added, resulting in a total diameter of 6 mm.

2.3 Treatment planning for simulation MRI

Treatment planning was performed using the Monaco treatment planning system (Elekta AB, Stockholm, Sweden). Electron density values were assigned to the CTV, bladder, rectum, bilateral femoral heads, bone, and body contour. A step-and-shoot intensity-modulated radiotherapy plan employing nine beams was generated. Dose calculations were performed using the Graphics Processing Unit–based Monte Carlo algorithm, with a calculation grid size of 3 mm for the low- and intermediate-risk groups and 2.5 mm for the high-risk group, and with a statistical uncertainty of 1%. A calculation grid size of 2.5 mm was used for high-risk patients to facilitate compliance with the maximum urethral PRV dose constraint of < 40 Gy. However, excessively small grid sizes substantially increased optimization time; therefore, a 2.5 mm grid size was used to balance good dose distribution with computational efficiency. A 7-MV FFF X-ray beam was used for all plans. The maximum number of segments per plan was limited to 60 for low- and intermediate-risk patients and 100 for high-risk patients. The minimum monitor units per segment, minimum segment width, fluence smoothing, plan quality, and minimum segment area were set to 5, 0.5 cm, low, 5, and 4 cm2, respectively. Institutional dose constraints are summarized in Supplementary Tables S2 (low- or intermediate-risk group) and S3 (high-risk group).

2.4 Online adaptive radiotherapy (ART) workflow

The standard bladder-filling protocol consisted of drinking 200 mL of water approximately 30 min before treatment initiation. The online adaptive radiotherapy (ART) workflow used at our institution is illustrated in Fig. 1. In this study, most patients underwent an initial ATS workflow on the treatment planning MRI before position verification, while one patient underwent an initial ATP workflow. Figure 1 illustrates the ATS workflow, representing our institution’s primary ART workflow. At the start of each treatment session, an MRI scan was performed and defined as the pre-MRI. The structures delineated on the simulation MRI were transferred to the pre-MRI via DIR, after which radiation oncologists manually reviewed and corrected the target and OAR contours to generate clinically acceptable structures. A treatment plan was then created based on the pre-MRI. Because prostate motion can occur during treatment preparation, an additional MRI was acquired immediately before irradiation, referred to as the position verification (PV) MRI. If target displacement was observed on the PV-MRI, plan adaptation was performed using an additional ATS procedure to correct for the motion. In this study, this sequence of pre-irradiation position adjustments was collectively defined as motion correction, which was performed using ATS in most patients. Motion correction encompasses all the involved procedures, including re-fusion of targets with positional displacement on PV-MRI, recontouring of the target and OARs, and treatment replanning using ATS optimization. No predefined quantitative threshold was established for motion correction of prostate displacement; implementation was determined by the physician. Specifically, motion correction was performed when the target was near the PTV boundary on PV-MRI and was likely to move outside the PTV during irradiation. When motion correction was required, the first PV-MRI was redefined as “the second pre-MRI,” and a subsequent MRI acquired for repeated PV was designated as “the second PV-MRI.” The ATS adaptive workflow loop triggered by motion correction required recontouring and re-optimization of the treatment plan on the second pre-MRI (hereafter referred to as secondary ATS-based recontouring and re-optimization), thereby extending the overall time spent on the treatment couch. If no prostate motion was detected on the PV-MRI, irradiation proceeded using the existing plan. The MRI acquired immediately after irradiation was defined as the post-MRI.

Fig. 1The alternative text for this image may have been generated using AI.

Workflow of adaptive radiotherapy for patients with prostate cancer treated at our institution. Standard workflow: Pre-magnetic resonance imaging (pre-MRI) refers to the first MRI performed on treatment day, which is used to create the treatment plan. After contouring and planning, position verification (PV)-MRI is acquired immediately before irradiation to confirm whether the target is positioned correctly within the planning target volume. Post-MRI is obtained immediately after irradiation to assess for any movement of the target or organs at risk during irradiation. Motion correction workflow: If target displacement is detected on the first PV-MRI, the PV-MRI is treated as a second pre-MRI, and recontouring and replanning are performed on the second pre-MRI. Subsequently, the target position is reevaluated using a second PV-MRI, and irradiation is delivered if no displacement is observed. Post-MRI is acquired after irradiation. T1–T4 represent time intervals for each process: T1 indicates the time from pre-MRI acquisition to contouring, T2 from contouring to planning, T3 from planning to irradiation, and T4 the irradiation time

2.5 Contouring and dose recalculation for PV- and post-MRI

Structural data and dose distributions were not inherently available for the PV- and post-MRI scans. Therefore, the structures delineated on the pre-MRI were transferred to the PV- and post-MRI using DIR. Medical physicists subsequently reviewed and modified the OAR and target contours as necessary to complete the structure sets on the PV- and post-MRI. The treatment plan information was also transferred to both the PV- and post-MRI datasets. Dose distributions were then reproduced on these images by recalculating the treatment dose using the same radiation fields and isocenter positions as those in the original treatment plan, without any modification. This dose recalculation was performed using the “original segments” function in the Monaco treatment planning system.

2.6 Creation of cumulative dose distribution

Cumulative dose calculations were performed using the RayStation software, Version 6.2.0 (RaySearch Laboratories, Stockholm, Sweden). Hybrid DIR employing the ANACONDA algorithm was used for dose accumulation [16]. This algorithm integrates both intensity-based and structure-based DIR methods. The CTV, bladder, urethra PRV, and rectum were defined as control regions of interest (ROIs), and DIR registration was centered on these ROIs. The DIR calculation grid was set to 2.5 mm. Each MRI from fractions 2–5 was registered to the MRI of fraction 1. The dose distributions from fractions 2–5 were then deformed according to the corresponding vector fields. Finally, the cumulative dose across all five fractions was generated on the MRI from fraction 1. This process was repeated for each of the three MRI types: pre-, PV-, and post-MRI. To verify the accuracy of the deformation, the average Dice similarity coefficient (DSC) between the deformed bladder (based on vector fields from MRI fractions 2–5) and the true bladder in MRI fraction 1 was calculated. Additionally, the average 95% Hausdorff distance was calculated as a supplementary measure of registration accuracy.

2.7 Extraction of dose constraints and evaluation

From the cumulative bladder dose distributions, the percentage volume and absolute volume (cc) receiving doses ≥ 10 Gy (V10Gy), V15Gy, V18.1 Gy, V25Gy, V30Gy, V35Gy, V37Gy, and V40Gy, as well as the maximum dose (Dmax), were extracted. V18.1 Gy and V37Gy were specifically included in the analysis because they represent institutional dose constraints (Supplementary Tables S2 and S3). In accordance with previous studies, patients exhibiting an increase in IPSS of ≥ 10 points from baseline within 3 months after radiotherapy were defined as having clinically significant acute urinary toxicity [13, 14]. The ability of each dose constraint to predict acute urinary toxicity was assessed using receiver operating characteristic analysis, and predictive performance was quantified by the area under the curve (AUC).

2.8 Extraction of factors other than dose constraints and evaluation

In addition to bladder dose constraints, specific factors related to the online ART workflow were extracted and evaluated for their association with acute urinary toxicity. The factors analyzed included prostate movement, bladder volume, bladder filling, irradiation time, online ART workflow time, and the number of motion corrections. Online ART workflow time was defined as the interval from pre-MRI acquisition start to irradiation start (T1–T3 in Fig. 1). In the ATS workflow loop triggered by motion correction, the online ART workflow time spanned from the start of acquisition of the second pre-MRI (first PV-MRI) to the start of irradiation (T1–T3 of motion correction workflow in Fig. 1). The number of motion corrections was defined as the total number of times motion correction was applied across the five ART workflows (range, 0–5). Detailed descriptions and calculation methods for these parameters are provided in Supplementary Table S4.

Additionally, clinical factors previously reported to be associated with urinary toxicity were extracted, including age [17], baseline IPSS [18,19,20], T stage [19], prostate volume [17], smoking status [17], and diabetes [17]. Herein, we also included prostate-specific antigen level, neoadjuvant chemotherapy, and risk group classification and evaluated their associations with acute urinary toxicity.

Univariate analyses were conducted to assess associations between potential factors and acute urinary toxicity. The Mann–Whitney U test was used for continuous variables, and the chi-square test was applied for categorical variables. Spearman’s correlation coefficients were calculated to evaluate multicollinearity among the variables. Factors with p < 0.1 in the univariate analysis were further examined for correlations with bladder dose indices to identify potential confounding factors. Only those with p < 0.1 and not exhibiting significant collinearity were included in the multivariate analysis, which was performed using linear regression.

2.9 Additional analyses regarding motion correction

Additional analyses were conducted to evaluate the impact of the number of motion corrections on bladder dose. Initially, correlations between the number of motion corrections and bladder dose constraints for the PV- and post-MRI datasets were examined using Spearman’s correlation coefficients. Furthermore, to explore the underlying mechanisms of these associations, a series of hypothesis-driven analyses were performed.

Hypothetically, secondary ATS-based recontouring and re-optimization with motion correction (repeating the ART process) offers three potential advantages for reducing bladder dose.

First, longer treatment couch time relaxes patients, which may lead to reduced body movement during irradiation. Repeating the ART workflow involves recontouring the displaced target and re-optimizing the treatment plan. Although this extends the time the patient is on the couch, patient motion decreases as the session progresses, resulting in more stable positioning before irradiation. Reduced motion limits bladder displacement into high-dose regions and may consequently decrease bladder dose exposure. To test this hypothesis, inferior and posterior prostate displacements during irradiation were compared between treatment fractions with and without motion correction. Prostate motion during irradiation was defined as the displacement of the prostate between the PV-MRI and post-MRI. To further assess the relationship between prostate motion and bladder dose escalation during irradiation, Spearman’s correlation analyses were performed between posterior and inferior prostate movements and the increase in bladder dose across all treatment fractions. The increase in bladder dose during irradiation was defined as the difference between bladder doses measured on the PV-MRI and post-MRI. The sample size included all treatment fractions, resulting in a total of 570 fractions (114 patients × 5 fractions).

Second, secondary ATS-based recontouring on the second pre-MRI may require less time than the initial contouring, potentially shortening the interval between MRI acquisition and irradiation (T1–T3 in Fig. 1). In the ATS workflow, initial ATS-based contouring without motion correction uses structure data propagated via DIR from the simulation contours to the daily MRI. Because these MRIs are acquired on different days, anatomical changes often require substantial modifications. For secondary ATS-based recontouring with motion correction, contours from the immediately preceding pre-MRI can be propagated to the second pre-MRI using DIR. As the anatomy is largely unchanged, only minor adjustments are typically needed. This reduces recontouring time and shortens the interval from the second pre-MRI to irradiation compared to sessions without motion correction. A shorter MRI-to-irradiation interval may help mitigate anatomical changes in the bladder occurring over time, thereby reducing bladder dose exposure. To test this hypothesis, the MRI-to-irradiation interval was compared between treatment fractions with and without motion correction. To ensure a fair comparison of the MRI-to-irradiation interval, only treatment fractions utilizing the ATS workflow were analyzed. The interval was defined as the duration from the second pre-MRI to irradiation for fractions with motion correction and from pre-MRI to irradiation for those without motion correction (T1–T3 in Fig. 1). Additionally, correlations between the MRI-to-irradiation interval and the increase in bladder dose from pre- to PV-MRI were analyzed using Spearman’s correlation coefficients. This analysis included all treatment fractions from 114 patients (570 fractions: 114 patients × 5 fractions). Fractions with missing time measurement data or those using ATP for the ART workflow were excluded, resulting in a final analysis cohort of 499 fractions. Of these, motion correction was applied in 112 fractions, while 387 fractions were delivered without motion correction.

Third, bladder dose may be reduced through secondary ATS-based re-optimization, adjusting the dose distribution to the enlarged bladder shape. Three treatment plans were retrospectively developed to test this hypothesis. For the first plan, the initial plan generated on the pre-MRI using the ATS workflow was not modified; however, the dose was recalculated on the PV-MRI using the same radiation field segments and fixed monitor units. This approach simulated a scenario in which, despite bladder changes due to filling between the pre-MRI and PV-MRI, no secondary ATS-based re-optimization was conducted, and the initial plan created on the pre-MRI was delivered to the bladder on the PV-MRI unchanged. This plan was defined as the “plan without replanning.” The second plan was re-optimized on the second pre-MRI, adjusting the dose distribution to the enlarged bladder via a secondary ATS-based re-optimization applied after motion correction. This was deemed as “replanning with ATS.” The third plan replaced the second plan’s ATS-based optimization with ATP-based re-optimization after applying motion correction. This approach was described as “replanning with ATP.” Bladder dose constraints were compared between these three plans. In this analysis, 10 patients were retrospectively selected from a cohort of 114 patients. These 10 patients exhibited minimal posterior and inferior prostate movements (average posterior and inferior displacements of 0.8 mm and − 0.2 mm, respectively, between pre- and PV-MRI) and rapid bladder filling (average pre-PV bladder volume increase of 66 cc [96%]). Among them, five patients were in the low- or intermediate-risk group, whereas five were in the high-risk group. Statistical comparisons were performed using the Wilcoxon signed-rank test.

All statistical analyses were performed using MATLAB (MathWorks, Inc., Natick, MA, USA), and a p value < 0.05 was considered statistically significant.