Skip to main content
NCBI home page
Radiation Oncology (London, England) logoLink to Radiation Oncology (London, England)
. 2026 May 28;21:100. doi: 10.1186/s13014-026-02863-4

Improving workflow efficiency during prostate stereotactic body radiotherapy using real-time adaptive planning associated with reduced intra-fractional target motion

Ting Chen 1, David Barbee 1, Hesheng Wang 1, Siming Lu 1, Sangkyu Lee 1, Ruth Afanador 1, Stavroula Kolitsopoulos 1, Matthew Long 1, Allison McCarthy 1, Paulina Galavis 1, Peter Schiff 1, Michael J Zelefsky 1,✉
PMCID: PMC13403397  PMID: 42210283

Abstract

Background

Magnetic resonance (MR) imaging-linear accelerator-based real-time adaptive planning for delivering ultra-hypofractionated stereotactic body radiotherapy (SBRT) has advanced clinical accuracy yet added complexity to the radiotherapy workflow. We retrospectively evaluated whether the addition of a Parallel Automated Contouring module with Enhancement of AI (PACE-AI) reduced contouring time and overall SBRT duration, and its impact on intra-fractional target motion.

Methods

This study included 250 fractions from 90 prostate cancer patients from which fraction time were tracked. All patients received definitive SBRT to the prostate on the 1.5-Tesla Unity (Elekta©) system, through the Adapt-To-Shape (ATS) workflow, which required real-time re-contouring of the normal tissues by the dosimetrists and target contouring by the physician for each of the fractions. We compared the fraction durations for 125 consecutive fractions treated with the incorporation of PACE-AI with 125 fractions previously treated without PACE-AI to determine whether PACE-AI improved efficiency and reduced treatment session duration.

Results

For the cohort treated without PACE-AI, the overall median duration was 67.6 min, including 23.9 min contouring time. With the incorporation of PACE-AI, the overall median duration was 51.2 min, representing a 24.3% reduction. The average contouring time was reduced by 55% to 10.8 min. In addition, the extent of positional shifts prior to beam delivery was significantly reduced from an average of 1.8 mm to 1.3 mm (p < 0.001) in both superior/inferior (range reduced from 0 to 6.3 mm to 0–4.2 mm) and anterior/posterior directions (range reduced from 0 to 6.6 mm to 0–5.0 mm).

Conclusions

The incorporation of PACE-AI improved workflow efficiency during SBRT. The reduction in treatment duration also helped reduce organ motion during real-time adaptive planning.

Keywords: Prostate cancer, Radiotherapy, Adaptive-planning, Stereotactic

Background

The use of a magnetic resonance imaging (MRI) linear accelerator (MRL) to deliver stereotactic body radiotherapy (SBRT) for the treatment of localized prostate cancer has added a new dimension of innovation and complexity to the delivery of ultra-hypo-fractionated external beam radiotherapy. With the availability of intra-fraction MRI to assist with planning and assessment of the target shape and position along with the normal anatomy, real-time adaptive planning has become possible, where plan optimization according to inter-fractional shift and deformation of organ at risks (OARs) and the clinical target can be accomplished before the delivery of the treatment fraction [1, 2]. Ongoing bladder filling and occasional rectal distention from gaseous build-up can lead to intra-fractional motion that could shift the target position, leading in some cases to higher doses deposited in regions not correlated with the location of the dominant tumor [3]. These intra-fractional changes may become more pronounced the longer the patient is on the treatment tables [4–11], and compared to standard SBRT, real-time adaptive planning on an MRL requires more input and longer planning times with patient on the treatment table [12, 13]. Therefore, efforts to increase the efficiency of the complex workflow of real-time adaptive planning would be clinically valuable.

For patients treated with SBRT on 1.5T MRL at our institution, we recently incorporated a parallel contouring script that allows physicians and dosimetrists to simultaneously delineate/edit different structures or regions of interest on the same set of adaptation MR scan in a streamlined workflow in MIM (MIM software). The workflow was further enhanced as we integrated it with an AI-powered auto-segmentation module: a UNet-based deep learning (DL) model, trained over 100 sets of structures contoured during previous treatment sessions, was used to generate initial contours of the normal tissue structures as well as the prostate target at the beginning of the parallel workflow, ultimately leading to more reliable and high-quality contours with a reduced contouring time. This enhancement is called Parallel Automated Contouring module with Enhancement of AI (PACE-AI). The purpose of this study was to evaluate the impact of PACE-AI on treatment duration during real-time adaptive planning.

Methods

The aim of this study was to report on the MR guided adaptive radiotherapy (MRgART) work-flow used for patients with prostate cancer treated with SBRT on an MR-Linac and to evaluate how efficiency of the workflow could be further improved with AI auto-segmentation and parallelization in contouring. This single-center study was performed after approval from the NYU Langone Medical Center institutional review board, which waived the requirement for informed consent due to the retrospective study design. The study was compliant with the Health Insurance Portability and Accountability Act and was conducted in accordance with the Declaration of Helsinki.

Overall fraction and individual task durations were consistently recorded for MR-guided prostate SBRT fractions using a web-based dashboard. The cohort reported here represents a subset of 90 patients and 250 fractions with total tracked fraction time and tasks. This subset included an initial group of 48 patients with 125 fractions using online contouring in MIM without parallel nor auto-segmentation for contouring (treated during February, 2025 to May 2025). We compared that group of patients with a more recent cohort (n = 42 patients and 125 fractions) incorporating PACE-AI (from August, 2025 to Dec. 2025). All patients were treated on the 1.5-Tesla (1.5-T) Elekta Unity MRL system (Elekta AB, Stockholm, Sweden) and had biopsy-proven prostate cancer. The subjects of this analysis all received definitive SBRT to prostate (40–45 Gy in 5 fractions). Treatment planning and delivery were carried out as previously described [14, 15] through the adapt to shape (ATS) workflow, in which the prostate target and OARs are contoured based on the daily MR before the adaptive planning and optimization is initiated. Contouring for all fractions included in this study were performed by the radiation oncologist (physician) and medical dosimetrists (planner) who have completed a comprehensive training program and have accumulated extended experience on MRgART. Descriptive statistics are reported as median and interquartile range (IQR) for contouring and total treatment time, and as mean (± SD) for patient motion data. For comparison between cohorts, Student’s t test was performed, and p-values are reported. Statistical significance was defined as p < 0.05.

For the cohort without PACE-AI, reference contours, which can either be contours in the offline reference plan or contours from previous adaptive sessions, were mapped to the adaptation T2w MR acquired on treatment day via deformable image registration (DIR) in MIM. For this earlier cohort without PACE-AI the planner completes the delineation of critical OARs including bladder and rectum first, saves the session in MIM, and then informs the physician to proceed sequentially with target delineation using the saved MIM session. Once the target contours are completed, planners will regain the session to work on other structures before sending the structure set to Monaco for plan optimization.

Parallel contouring and AI based auto-segmentation

A script-based MIM workflow was developed at our institution using the MIM workflow builder. The workflow supports parallel contouring that enables physicians and planners to contour simultaneously during the adapt to shape (ATS) session and to streamline the contouring process by providing a scripted automation solution. The workflow is standardized by contouring structures in a pre-programmed fixed order, selecting the appropriate contouring tool (e.g. 2D/3D paintbrushes) for the user, and automatically focus on the target and zoom the image view for enhanced visibility to facilitate contouring in MIM. At each contouring step, a text message of guidance is displayed in the MIM notification box to guide the user with the current contouring task. For the physician’s contouring session, a 4-view display consisting of the axial, coronal, and sagittal view of the T2w adaptation MR, and a view of lesion in functional MRI (fMRI) as seen on the diffusion weighted imaging (DWI) series is designed as the default display used in the workflow. Adjustments made to the target structure are updated in real time in both the T2W and fMRI views to facilitate multi-modality contouring.

The AI-driven auto-segmentation module was developed, tested, and then integrated with the parallel workflow for the clinical deployment of the current version of PACE-AI in August 2025. AI segmentation was performed using an in-house developed nnUNet v2 model [16]. The OAR model was trained on 127 T2w MR imaging studies (162 total including 35 held-out validation cases) with manual delineations of 13 structures: body, bilateral femurs, bones, rectum, bladder, bowel, penile bulb, urethra, hydrogel spacer, air cavities, and CTV prostate and node targets. Training employed 3D full-resolution architecture with 80 × 128 × 224 patch size, batch size 2, instance normalization, z-score intensity normalization, and no mirroring applied. Inference is performed in AWS SageMaker via tunnel from MIM, with output structures recombined and exported as DICOM RTSTRUCT to MIM for implementation into PACE-AI.

The design of the current PACE-AI workflow is summarized in Fig. 1. For cohort of patients treated using PACE-AI, two independent sources of initial contours were generated at the beginning of every ATS session when the adaptation T2w axial MRs were acquired: (a) structures derived from previous session via semi-automatic DIR between the current and previous adaptation MRs, and (b) structures automatically generated by the AI auto-segmentation model. These two structure sets were automatically merged into one initial structure set in MIM by replacing some DIR-generated structures with expert-verified AI-segmentation.

Fig. 1.

Fig. 1

Open in a new tab

Top: Diagram of the structure of PACE AI workflow. Middle from left to right: Initial structure set generated based on deformable registration from reference plan; initial structure set generated by AI; OAR structure set sent to physician review. Bottom left: structure set reviewed by physician with target volumes appended. Bottom right: comprehensive structure set after final review ready for adaptive planning. Color index for each structure is attached

At the beginning of the parallel workflow, the initial structure set is split into two subsets: one for the physician, which includes all the target volumes, urethra, penile bulb, and bowel; and one structure set for the planner, which includes bladder, rectum, bone structures, body, and air cavities. At this point, the physician and planner concurrently start to delineate/edit/review their corresponding subsets of structures in parallel on two different computers. The planner contours the critical OARs (particularly bladder and rectum) first and saved as a structure set in MIM and continue with other structures in the subset of structures. When the physician completes the contouring of target volumes, saved OAR structures are imported in the physician’s contouring session for review, edit, and approval together with target volumes. In the final step, the planner imported the physician-approved structure sets to merge with locally contoured structures and completes the workflow by generating motion monitoring structures necessary for real-time MR-guided beam delivery and performing a final review before exporting the completed structure set to Monaco for plan optimization.

Intra-fractional motion measurement

For both cohorts, upon physician approval, the primary adaptation T2w axial images and RTstruct DICOMs are exported to offline Monaco for insertion of immobilization devices, followed by IMRT plan optimization in online Monaco based on the anatomy-of-the-day. The optimization workflow is illustrated in Fig. 2.

Fig. 2.

Fig. 2

Open in a new tab

Diagram of the adaptive to shape workflow with intra-fractional target motion measurement

A T2w MR verification scan is acquired at the beginning of the plan optimization, followed by Comprehensive Motion Management (CMM) template acquisition and matching, as shown in Fig. 2. Intra-fractional target drifts during contouring were measured in the left/right, superior/inferior, and anterior/posterior direction through the CMM template matching. The shifts are verified by the physician in the verification scan to determine if accurate. If there is an intra-fractional prostate shift > 1.5 mm in any direction an additional Adapt-To-Position (ATP) plan will be initiated to correct for target misalignment by making the necessary positional shifts of the target. At the beginning of ATP, target contour (usually CTV prostate) is overlaid in the verification scan with CMM suggested shifts manually applied. Once the target shifts are confirmed, planner will begin quick optimization, during which MLC segments in the adaptive IMRT plan will be re-optimized to account for the target position shift, to re-achieve dose criteria. Verified target shifts are recorded for every fraction of treatment in the study for quantitative analysis.

Results

For the non-PACE-AI cohort (no parallel nor auto-segmentation), the median patient on-table time (from start of patient set-up to end of delivery) was 67.6 min (IQR: 51.2–92.3 min, n = 125). Table 1 demonstrates the median duration for each component of the treatment fractions for this cohort, which grouped the following elements: (1) patient entry and set-up; (2) MRI for the scout, primary axial T2w, and additional sequences; (3) image import to MIM, image fusion, and contouring of normal tissue structures; followed by physician confirmation of normal structures and contouring of the prostate target; (4) contour export to Monaco for planning and optimization by planner; (5) physician plan review; (6) physics pre-treatment review; and (7) treatment delivery. The tasks with the longest median durations were contouring (23.9 min, IQR: 20.1–27.4 min), followed by treatment planning and optimization (14.2 min) and treatment delivery (13.8 min). For this patient cohort, 76 of 125 (60.8%) fractions needed positional adjustment from initial target (CTV) and re-optimization (ATP). The median elapsed time from entry of the patient into the MRL room to completion of the treatment fraction was 65 min (IQR: 51.2–87.0 min, n = 49) for those patients who did not require a positional correction before the beam delivery compared to 69 min (IQR: 51.5–92.3 min, n = 76) for those patients who required a positional correction of ATP before the beam delivery. The comprehensive motion management (CMM) system recorded average CTV shifts of 1.8 mm (standard variation ± 1.5 mm, range 0–6.3 mm) in the superior/inferior direction, and 1.8 mm in the anterior/posterior direction (standard variation ± 1.3 mm, range 0–6.6 mm). Among the 125 fractions in this cohort, 60% (75 out of 125) received 45 Gy concurrent boost to the dominant intraprostatic lesion (DIL). With the additional contouring of DIL target and higher dose, there was a significant difference between prostate only, and prostate with DIL treatments in contouring time and total treatment duration, both for the no PACE-AI (p = 0.005) and PACE-AI (p = 0.004) cohorts.

Table 1.

Summary of median timein minutes of each step during the adaptive workflow

Setup Imaging Contouring Planning Physician Review Physics Review Beam delivery Total Duration
Without PACE AI 4.0 7.5 23.9 14.2 2.8 0.8 13.8 67.6
With PACE AI 3.7 6.5 10.8 12.0 2.2 0.6 13.8 51.2
Change in min 0.3 1.0 13.1 2.2 0.6 0.2 0 16.4
Change in % -15.0%* -8.0%* -54.8% -15.5% -21.4% -25.0% 0% -24.3%
Open in a new tab

* The reduction of setup and imaging time is independent of the use of PACE-AI. Setup time reduced because of therapists’ improved skills and accumulation of experience with Unity and the adaptive workflow. Imaging time reduction was mostly due to the implementation of C-Sense acceleration during adaptation T2w acquisition

When compared to the more recent PACE-AI cohort, as shown in Fig. 3a a significant reduction in overall treatment time was observed for the PACE-AI cohort. The median duration from the start of patient set-up to the end of treatment delivery was 51.2 min (IQR: 46.9–57.2 min, n = 125). With the incorporation of PACE-AI, a 24.3% reduction in duration was noted. The median contouring time was reduced by 54.8% from 23.9 min to 10.8 (IQR: 9.2–13.5 min) minutes. There was a difference in treatment duration for those patients who did not require a positional correction before the beam delivery (median: 48.2 min, IQR: 44.1–52.2 min, n = 52) and those who required (median: 51.7 min, IQR: 48.8–57.3 min, n = 73). It has been noted that for patients who received prostate-only treatment without DIL boost (72 out of 125), the median contouring time is 10.0 min, and the median treatment time is 49.7 min (IQR: 46.0–52.9 min). For those who received DIL boost (53 out of 125), the median contouring time is 11.2 min, and the median treatment duration was 52.1 min (IQR: 49.2–54.0 min), as shown in Fig. 4.

Fig. 3.

Fig. 3

Open in a new tab

Histogram comparison between treatment time of 125 fractions without PACE AI and 125 fractions with PACE AI treatment

Fig. 4.

Fig. 4

Open in a new tab

Change of contouring and total treatment time after the start of the clinical application of PACE-AI (separated into subgroup of prostate only and prostate + DIL)

The clinical implementation of PACE-AI was divided into 3 stages: (1) parallel workflow only, (2) initial AI integration, and (3) updated (and final) AI integration, over 7 weeks. The contouring and overall treatment time during the implementation were recorded and illustrated in Fig. 5. With the use of the parallel workflow alone, a roughly 9 min reduction of median contouring time has been observed (22min50sec to 13min48sec, or 39.6% decrease). The improvement of efficiency came from the standardization of the contouring process in the scripted workflow, text guidance at each contouring step, automated subroutines that significantly reduced manual operations and user interactions, and above all, the parallel structure of the workflow that allow the physician and planner contour concurrently to minimize the idling time during the process. As shown in Fig. 5, the integration of the AI contouring helped to reduce another 4 min in median contouring time (13min48sec to 9min28sec, 31.4% decrease from parallel only) by providing high quality initial contours. Considering this is after the implementation of parallel workflow, the actual impact of AI contouring can be more significant in sequential workflow as the AI contouring will expedite the contouring process for both the physician and the planner by providing high quality initial structure sets.

Fig. 5.

Fig. 5

Open in a new tab

The weekly contouring (top) and total treatment (bottom) time reduction during the implementation of PACE-AI. The data point at non-para are the average of 2 weeks’ time data before the implementation of the parallel workflow. The data point at para-only are the average of 2 weeks’ time data after the implementation of the parallel workflow before the integration of AI. The AI integration took place before the time point of AI_1, then the model upgraded to version 2.0 before the data point AI_4

To quantify the impact of multiple measurements from same patient, we recalculated the median contouring time and treatment time with and without PACE-AI by replacing multiple measurements from same patient with one measurement of the patient’s average value. No significant differences were observed for median contouring times when analyzing all fractions (including multiple measurements from the same patient) versus an analysis of all patients with utilizing one measurement of the average value across all the fractions for a patient’s treatment.

Finally, we observed that likely related to the efficiency of the PACE-AI workflow, the on-table treatment duration for the patient was significantly shorter compared to the non- PACE-AI cohort. This in turn was associated with less target positional variability that was observed. As shown in Fig. 6, for the PACE-AI cohort (n = 125 fractions), the average superior/inferior and anterior/posterior positional adjustments were both 1.3+/-1 mm, significantly less than the pre-PACE-AI cohort (p < 0.001 and p < 0.002, respectively).

Fig. 6.

Fig. 6

Open in a new tab

Change of intra-fractional target motion after the start of the clinical application of PACE-AI

Discussion

Real-time MR guided adaptive radiotherapy of prostate cancer entails an intricate workflow comprising multiple elements: image acquisition, contouring of target and normal tissues, treatment planning, and re-imaging with positional adjustments before beam delivery. Streamlining the workflow to reduce the duration of the treatment fraction helps to reduce potential target drift during real-time adaptive planning and beam delivery. We demonstrate that incorporating a parallel workflow and AI-based auto-contouring streamlined the process, reducing the amount of patient-on-the-treatment-table time. Auto-segmentation of normal tissue structures and targets with minor editing performed by the dosimetrist and physician reduced the duration of the contouring segment of the workflow by about 55%, which effectively reduced the fraction duration on the treatment table by > 24%. Using PACE-AI also significantly reduced the extent of positional shifts in both superior-inferior and anterior-posterior directions, presumably owing to the reduced duration of the contouring session. This is a logical conclusion based on previous findings [4–11].

These data suggest that the reduced contouring time afforded by the PACE-AI workflow may have minimized bladder filling, which can cause prostate drifts, and gaseous distention of the rectum, which may further impact upon the prostate target position [14] and lead to significant reduction of intra-fractional target motion prior to beam delivery. The reduced target motion may help to explain the reduction of plan optimization time shown in Table 1. Less target motion will help to reduce the probability of the ATP adjustment needed before beam delivery, and also reduce the complexity of the re-optimization, which is reflected in the reduction of planning time.

Since the data of the two cohorts were acquired sequentially before and after the implementation of PACE-AI, there are sources of potential bias in the analysis. Some of the major contributors of bias are: (1) the learning curve effect, (2) team familiarity and overall workflow maturation, and (3) multiple fractions of time from same patient may not be eligible as independent measurements. All these factors may contribute to the observed improvement of the post PACE-AI contouring and treatment time.

As part of our workflow, we have routinely incorporated full ATS workflows, where the prostate target and the surrounding normal tissues including bladder, urethra, rectum, and bowel are contoured, and then the daily adaptive plan is generated, which adheres to the established target and OAR constraints. Rather than using the MRL to employ only an ATP workflow, the ATS approach considers any deformation of the target as well as the normal tissue. While potentially allowing for shorter on-table time, an ATP-only workflow does not take complete advantage of the accuracy that the MRL can achieve regarding imaging, targeting, and delivery during MR-based online adaptive therapy that improves the therapeutic ratio.

Recently multiple strategies targeting the reduction of target/OARs delineation time during MR guided ART have been proposed. In [17, 18], a parallel contouring script was developed in MIM to improve the efficiency of MR guided ATS workflow. However, the workflow proposed was mostly based on manual delineation without auto-segmentation to accelerate the tedious contouring process. In [19–21], AI-assisted segmentation approaches have been proposed, validated, and implemented clinically for contouring in T2w MR images in Monaco during MRgART on prostate cancer. For patient with DIL, Monaco-based auto-segmentation has limited compatibility with fMRI, such as DWI, which plays a critical role in the delineation of DIL targets. PACE-AI is one of the first clinical implementations that integrated AI auto-segmentation with scripted contouring workflow in MIM to support multimodality segmentation, and with enhanced accuracy and efficiency.

In line with published benchmarks for full ATS SBRT treatments on MRL, our current experience indicates that the on-table time for the most frequently treated SBRT indication is approximately 50 min with the use of PACE-AI. Integration of MIM into daily online MRgART has afforded our program with versatility in remote treatment supervision, improved visualization and localization of prostate gland and DIL within using multiparametric imaging, more intuitive contouring tools, concurrent parallel contouring, and potential future scripting solutions. However, integrating this platform results in additional steps, including export to MIM, registration and adaptation of prior contours, fusion of additional image orientations/sequences, export to and import of images/structures in offline Monaco, and re-opening the case in online Monaco, which adds several minutes to our contouring and optimization times. The implementation of PACE-AI has effectively compensated for the duration added by these steps while maintaining, and even improving, the overall quality and efficiency of real time contouring during the highly intensive MRgART workflow. Although the contouring and treatment time reported here is slightly longer (but still well within the comparable range) than some of the works reported recently [19–21] due to the use of a MIM based workflow, there are potential benefits of using multiple MRI and fMRI sets for target contouring in MIM, especially for DIL cases, during MRgART for prostate.

It should be noted that the use of AI in clinical environments is a relatively new and controversial approach and still under investigation for its reliability and productivity. Accepting AI products without cautious validation and a routine secondary check in place may lead to discouraging consequences in cases when AI provides mediocre or unsatisfactory solutions. At our institute, all AI applications, including PACE-AI, are thoroughly validated by a testing group, which consists of experienced physicians, physicists, and dosimetrists, before clinical use. All results involving the use of AI must be reviewed by experienced senior physicians to confirm their integrity and accuracy in the clinical environment. For the study reported in this manuscript, all PACE-AI generated contours were reviewed by qualified radiation oncologists and went through independent check performed by certified medical physicists before plan optimization.

There are significant opportunities for further improving the efficiency of MRgART [22–28]. The main elements where streamlined workflows and software innovations can expedite the processes include acceleration of MRI image acquisition and more rapid treatment planning algorithms, possibly utilizing a library of AI-assisted planning solutions based upon the prior fractions for the individual patent. PACE-AI can be easily integrated with other technical improvements as a software solution that requires minimum hardware upgrades.

Conclusion

In conclusion, our current analysis has provided valuable insights into the current online adaptive workflow, establishing realistic task-based benchmarks for new MRL programs seeking contouring solutions outside Monaco, and has successfully highlighted areas for potential improvement and innovation. PACE-AI, as our latest clinical application enhancement, has demonstrated the feasibility of establishing an accurate and efficient workflow for daily ATS powered by latest technologies with less dependency on Monaco for contouring. Future efforts will focus on further optimization of the existing MIM-to-Monaco workflow, integrating image acceleration, automation tools, and planning optimization techniques to further streamline the process and reduce treatment times to < 40 min in the near future. These advancements are expected to enhance the clinical efficiency of our program, ultimately improving patient experience and increasing access to MRL technology.

Abbreviations

PACE-AI

Parallel Automated Contouring module with Enhancement of AI

MRI

Magnetic resonance imaging

SBRT

Stereotactic body radiotherapy

ATS

Adapt-To-Shape

OARs

Organ at risks

DL

Deep learning

PTVs

Planning Target Volumes

CMM

Comprehensive motion management

fMRI

Functional MRI

DWI

Diffusion weighted imaging

MgART

MR guided ART

ART

Real-time adaptive radiotherapy

Author contributions

TC developed the methodology, collected data, and performed data analysis, and drafted the manuscript. DB developed the methodology and collected data. HW collected data and helped with drafting. SL (Lu) collected data.SL (Lee) collected data. RA collected data. SK collected data. ML performed data analysis. AM helped coordinate the research. PG coordinated the research of the physics team. PS supervised data collection. MZ supervised data collection, data analysis, methodology development, drafted, revised, reviewed, and finalized the manuscript.

Funding

No Funding was received as part of research associated with this manuscript.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

MJZ has received speaking honorarium from Elekta,. The other authors have no relevant financial disclosures.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  • 1.Hall WA, Paulson ES, van der Heide UA, et al. The transformation of radiation oncology using real-time magnetic resonance guidance: A review. Eur J Cancer Oxf Engl 1990. 2019;122:42–52. 10.1016/j.ejca.2019.07.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Hall WA, Paulson E, Li XA, et al. Magnetic resonance linear accelerator technology and adaptive radiation therapy: An overview for clinicians. CA Cancer J Clin. 2022;72(1):34–56. 10.3322/caac.21707. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Brennan VS, Burleson S, Kostrzewa C, et al. SBRT focal dose intensification using an MR-Linac adaptive planning for intermediate-risk prostate cancer: An analysis of the dosimetric impact of intra-fractional organ changes. Radiother Oncol J Eur Soc Ther Radiol Oncol. 2023;179:109441. 10.1016/j.radonc.2022.109441. [DOI] [PubMed] [Google Scholar]
  • 4.Kusters J, Monshouwer R, Koopmans P, Wendling M, Brunenberg E, Kerkmeijer L, van der Bijl E. Prostate motion in magnetic resonance imaging-guided radiotherapy and its impact on margins. Strahlenther Onkol. 2025;201(8):799–806. https://doi.org/10.1007/s00066-024-02346-z. Epub 2025 Jan 14. Erratum in: Strahlenther Onkol. 2025;201(8):807. 10.1007/s00066-025-02390-3. PMID: 39808200; PMCID: PMC12283804. [DOI] [PMC free article] [PubMed]
  • 5.Wittmann D, Paulson ES, Banerjee A, Banla LI, Schultz C, Awan M, Chen X, Omari EA, Straza M, Li XA, Erickson B, Hall WA. Quantification and dosimetric impact of normal organ motion during adaptive radiation therapy planning using a 1.5 tesla magnetic resonance–equipped linear accelerator (MR-Linac), Advances in Radiation Oncology. 2025;10(5). [DOI] [PMC free article] [PubMed]
  • 6.Tocco Boris R, Kishan Amar U. Ma Ting Martin, Kerkmeijer Linda G. W., Tree Alison C. MR-guided radiotherapy for prostate cancer, frontiers in oncology; 2020;10. [DOI] [PMC free article] [PubMed]
  • 7.Peng S, Courtot T, Lye J, et al. Magnetic resonance guided radiation therapy (MRgRT) prostate motion and margins. J Med Radiat Sci. 2025:1–8. 10.1002/jmrs.70033. [DOI] [PMC free article] [PubMed]
  • 8.Benitez CM, Steinberg ML, Cao M, Qi XS, Lamb JM, Kishan AU, Valle LF. MRI-Guided Radiation Therapy for Prostate Cancer: The Next Frontier in Ultrahypofractionation. Cancers (Basel). 2023;15(18):4657. 10.3390/cancers15184657. PMID: 37760626; PMCID: PMC10526919. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.de Muinck Keizer DM, Kerkmeijer LGW, Willigenburg T, van Lier ALHMW, den Hartogh MD, van der Voort van Zyp JRN, et al. Prostate intrafraction motion during the preparation and delivery of MR-guided radiotherapy sessions on a 1.5T MR-Linac. Radiother Oncol. 2020;151:88–94. ISSN 0167–8140. 10.1016/j.radonc.2020.06.044. [DOI] [PubMed]
  • 10.Schaule J, Chamberlain M, Wilke L, et al. Intrafractional stability of MR-guided online adaptive SBRT for prostate cancer. Radiat Oncol. 2021;16:189. 10.1186/s13014-021-01916-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Cuccia F, Corradini S, Mazzola R, Spiazzi L, Rigo M, Bonù ML, Ruggieri R; Buglione di Monale e Bastia, Magrini M, Alongi SM. F. MR-guided hypofractionated radiotherapy: current emerging data and promising perspectives for localized prostate cancer. Cancers. 2021;13: 1791. 10.3390/cancers13081791. [DOI] [PMC free article] [PubMed]
  • 12.de Leon J, Crawford D, Moutrie Z, et al. Early experience with MR-guided adaptive radiotherapy using a 1.5 T MR-Linac: First 6 months of operation using adapt to shape workflow. J Med Imaging Radiat Oncol. 2022;66(1):138–45. 10.1111/1754-9485.13336. [DOI] [PubMed] [Google Scholar]
  • 13.Winkel D, Bol GH, Kroon PS, et al. Adaptive radiotherapy: The Elekta Unity MR-linac concept. Clin Transl Radiat Oncol. 2019;18:54–9. 10.1016/j.ctro.2019.04.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Byun DJ, Oh C, Kim J, Barbee D, Long M, Fuligni G, Chen T, Wang H, Lu S, Zelefsky MJ. Bladder filling dynamics during online adaptive prostate stereotactic body radiotherapy: Rationale for using an empty bladder workflow for treatment. Radiother Oncol. 2025. 10.1016/j.radonc.2025.110978. Epub 2025 Jun 2. PMID: 40466739. [DOI] [PubMed] [Google Scholar]
  • 15.Taneja S, Wang H, Barbee D, Galavis P, Serrano-Sosa M, Chen T. Commissioning and Implementation of a Pencil Beam Algorithm with a Lorentz Correction As a Secondary Dose Calculation Algorithm for an MR-Linear Accelerator. J Appl Clin Med Phys. 2024;26:e14590. 10.1002/acm2.14590. PMID: 39625056, PMCID: PMC11905237. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Isensee F, Jaeger PF, Kohl SAA, et al. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods. 2021;18:203–11. 10.1038/s41592-020-01008-z. [DOI] [PubMed] [Google Scholar]
  • 17.Paulson ES, et al. Int J Radiat Oncol Biol Phys. 105(1): E788.
  • 18.Eric S, Paulson E, Ahunbay X, Chen NJ, Mickevicius G-P, Chen C, Schultz B, Erickson M, Straza WA, Hall X, Allen Li. 4D-MRI driven MR-guided online adaptive radiotherapy for abdominal stereotactic body radiation therapy on a high field MR-Linac: Implementation and initial clinical experience. Clin Transl Radiat Oncol. 2020;23:72–9. ISSN 2405–6308. 10.1016/j.ctro.2020.05.002. [DOI] [PMC free article] [PubMed]
  • 19.Lorenzen EL, Celik B, Sarup N, Dysager L, Christiansen RL, Bertelsen AS, Bernchou U, Agergaard SN, Konrad ML, Brink C, Mahmood F, Schytte T, Nyborg CJ. An open-source nnU-net algorithm for automatic segmentation of MRI scans in the male pelvis for adaptive radiotherapy. Front Oncol. 2023;13:1285725. 10.3389/fonc.2023.1285725. PMID: 38023233; PMCID: PMC10654998. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Nachbar M, Lo Russo M, Gani C, Boeke S, Wegener D, Paulsen F, Zips D, Roque T, Paragios N, Thorwarth D. Automatic AI-based contouring of prostate MRI for online adaptive radiotherapy. Z Med Phys. 2024;34(2):197–207. Epub 2023 May 30. PMID: 37263911; PMCID: PMC11156783. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Konrad ML, Brink C, Bertelsen AS, Lorenzen EL, Celik B, Nyborg CJ, Dysager L, Schytte T, Bernchou U. Feasibility and time gain of implementing artificial intelligence-based delineation tools in daily magnetic resonance image-guided adaptive prostate cancer radiotherapy. Phys Imaging Radiation Oncol. 2025;33::100694. 10.1016/j.phro.2024.100694. [DOI] [PMC free article] [PubMed]
  • 22.Lustig M, Donoho D, Pauly JM, Sparse MRI. The application of compressed sensing for rapid MR imaging. Magn Reson Med. 2007;58(6):1182–95. 10.1002/mrm.21391. [DOI] [PubMed] [Google Scholar]
  • 23.de Leon J, Twentyman T, Carr M, Jameson M, Batumalai V. Optimising the MR-Linac as a standard treatment modality. J Med Radiat Sci. 2023;70(4):491–7. 10.1002/jmrs.712. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Elguindi S, Zelefsky MJ, Jiang J, et al. Deep learning-based auto-segmentation of targets and organs-at-risk for magnetic resonance imaging only planning of prostate radiotherapy. Phys Imaging Radiat Oncol. 2019;12:80–6. 10.1016/j.phro.2019.11.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Nachbar M, Lo Russo M, Gani C, et al. Automatic AI-based contouring of prostate MRI for online adaptive radiotherapy. Z Für Med Phys. 2024;34(2):197–207. 10.1016/j.zemedi.2023.05.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.McDonald BA, Cardenas CE, O’Connell N, et al. Investigation of autosegmentation techniques on T2-weighted MRI for off‐line dose reconstruction in MR‐linac workflow for head and neck cancers. Med Phys. 2024;51(1):278–91. 10.1002/mp.16582. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier-Hein KH. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods. 2021;18(2):203–11. 10.1038/s41592-020-01008-z. [DOI] [PubMed] [Google Scholar]
  • 28.Naccarato S, Rigo M, Pellegrini R, et al. Automated Planning for Prostate Stereotactic Body Radiation Therapy on the 1.5 T MR-Linac. Adv Radiat Oncol. 2022;7(3):100865. 10.1016/j.adro.2021.100865. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Articles from Radiation Oncology (London, England) are provided here courtesy of BMC

RESOURCES