Background and Purpose:The closed-off nature of most treatment planning systems (TPS) limits the potential for using artificial intelligence (AI) tools during online adaptive treatments. The aim of this study was to develop an AI-driven pipeline (AutoAdapt) for online planning of adaptive radiotherapy usable in a closed-off setting, providing optimal plan constraints derived from a population-based dose prediction model. Material and methods:The AutoAdapt pipeline consists of a physics-aware Swin UNet transformer network for dose prediction trained on 266 magnetic resonance images from 25 prostate cancer patients treated with 60 Gy on a 1.5 T magnetic resonance linear accelerator. The predicted dose was used to calculate plan constraints that were subsequently fed into a commercial TPS. AutoAdapt was tested using ten unseen cases and compared to manual plans based on clinical objectives, time, and complexity. Results:While all plans were approved by a radiation oncologist, AutoAdapt met all clinical objectives in seven patients compared to ten when manually planned. AutoAdapt yielded a significantly lower D0.035cm3 to the rectum (p = 0.01). Manual plans achieved a median rectum V20Gy of 44% compared to 51% in AutoAdapt plans (p = 0.02). The pipeline only required a median of 29 s (7.5%) longer than the manual planners. Conclusions:The developed pipeline resulted in high-quality plans, ready for clinical use without further adjustments. AutoAdapt prioritized maximum rectum dose over D20% compared to manual planning, while requiring less manual work. In the future, AutoAdapt may be used to assist human planners and improve adaptive radiotherapy workflows.

INTRODUCTION:Biomarker-driven strategies are central to personalized oncology, yet local treatments such as radiotherapy still lack validated stratification frameworks. Conventional frequentist trial designs often require prohibitively large patient cohorts without incorporating previously validated data. We therefore propose a Bayesian trial framework, which utilizes information of a patient's biomarker and historical information of the biomarker effect to optimize the assigned dose. The aim of this work is to develop and illustrate such a framework for radiotherapy in rectal cancer organ preservation. METHODS:We designed a prospective two-arm Bayesian response-adaptive trial concept incorporating the principle of optimal stopping. Biomarker measurements obtained during treatment guided dose adaptation, illustrated here using the imaging-derived early regression index (ERI). Feasibility and statistical performance were evaluated through a fully simulated trial in rectal cancer organ preservation. Sensitivity analysis was applied to investigate the effect of prior information on the posterior distribution. RESULTS:Simulations modeled the patients' response using biomarker-dependent tumor control and biomarker-independent toxicity curves. One subgroup of patients would benefit most from moderate dose escalation, achieving improved tumor control without excessive toxicity. In contrast, poor and excellent responders gained limited additional benefit at clinically acceptable doses. Under the optimistic prior scenario, the proposed design could claim the benefit of the adaptive dose with fewer than 100 simulated patients, while with weakly informative priors less than 150 patients are needed. CONCLUSION:This Bayesian response-adaptive design provides a quantitative framework to integrate biomarkers such as ERI into radiotherapy personalization. It enables a structured evaluation of dose-response relationships and may help facilitate the translation of biomarker findings into local cancer therapy, acknowledging the underlying model assumptions.

Background and Purpose Neural networks promise fast dose modelling with high accuracy for challenging situations like magnetic resonance imaging (MRI)-guided radiotherapy. As they are data-driven, failure can occur and early identification of erroneous dose calculations is required.In this study, we implemented and evaluated three uncertainty estimation techniques to assess whether they can indicate increased prediction error. Materials and methods Using a dataset of 6713 radiotherapy segments from 130 1.5 T MRI linear accelerator plans and a 3D UNet for dose modelling, three techniques were implemented to assess uncertainty: Monte Carlo dropout (MCD), mean variance estimation (MVE) and a deep ensemble (DE).All methods were evaluated regarding calibration using the expected normalized calibration error (ENCE) and correlation to mean absolute error (MAE).Cumulative failure rates were calculated across uncertainty thresholds, with a 3 mm/3% gamma passing rate < 95% defining failure. Results After calibration, all three methods showed ENCE values of 0.14/0.05/0.10 for MCD/MVE/DE. A strong relationship between mean uncertainty and MAE was observed, reflected by high Spearman correlation coefficients (MCD: ρ = 0.64, MVE: ρ = 0.76, DE: ρ = 0.67).Failure rates increased with mean uncertainty across all methods, enabling thresholds targeting a 10% failure rate. On independent evaluation, 44%–77% of segments were accepted, with observed failure rates among accepted segments of 7.2%–9.5%. Conclusions All methods provided meaningful uncertainty estimates clearly associated with prediction error. MVE showed the lowest ENCE and strongest correlation with MAE while requiring the least computation. DE most closely matched the target failure rate while flagging the fewest segments.

Background and purpose:Large language models (LLMs) have shown growing potential for clinical text processing, but their systematic application in radiation oncology-especially for non-English clinical documentation-remains underexplored. This study investigated whether pretrained LLMs can automatically extract, analyze, and structure radiotherapy-relevant information from routine unstructured medical notes, with the goal of supporting automated population of electronic case report forms (eCRFs). Materials and methods:This study examined prostate cancer patients treated with the MR-Linac, for whom ground truth data exist in the MOMENTUM database. A total of 100 patients were included, with 90 used for prompt development and 10 for independent testing. Medical notes were extracted, anonymized, and categorized by time points. The Llama-3.1-8b model was used, with prompts designed using chain-of-thought (CoT) logic with five in-context examples. The model output was post-processed, and extracted data was compared against ground truth. Results:Medical notes were successfully processed, with predicted values generated in an average time of 16 s per note. The LLM achieved matching accuracies of 83.6% and 83.8% on the development and testing datasets. Analysis revealed that the model disagreed with specific values in 8.1% of development dataset cases and 8.6% of testing dataset cases. An independent manual review before model evaluation showed approximately 7.5% of routinely collected test data did not match reviewed values, indicating inaccuracies in the routinely acquired ground truth. Conclusion:This study demonstrated the effectiveness of LLMs in structuring clinical data from medical non-English notes, with high accuracy in extracting and categorizing information. While multi-institutional validation is needed, the results indicate a significant healthcare impact through efficient data management, processing notes in 16 s, and accurately populating CRFs with minimal staff involvement.

This roadmap provides a comprehensive framework for integrating diffusion-weighted imaging (DWI) into radiation therapy (RT), with an emphasis on its application in magnetic resonance imaging-guided radiotherapy and its potential for driving biological image-guided adaptive radiotherapy (ART). Developed through collaboration among experts in medical physics, magnetic resonance imaging science, and radiation oncology, the paper aims to bridge disciplinary gaps and foster a shared understanding across scientific, technical, and clinical domains. It benchmarks the current state of DWI in RT, identifies critical challenges, and highlights recent advancements in acquisition, reconstruction, biophysical modeling, quality assurance, clinical validation and translation, as well as emerging concepts. By outlining ongoing efforts and forecasting future developments, this roadmap supports the adoption of DWI as a quantitative imaging biomarker for personalized and ART in precision oncology.

Abstract Background For radiotherapy of head and neck cancer (HNC) magnetic resonance imaging (MRI) plays a pivotal role due to its high soft tissue contrast. Moreover, it offers the potential to acquire functional information through diffusion weighted imaging (DWI) with the potential to personalize treatment. The aim of this study was to acquire repetitive DWI during the course of online adaptive radiotherapy on an 1.5 T MR-linear accelerator (MR-Linac) for HNC patients and to investigate temporal changes of apparent diffusion coefficient (ADC) values of the tumor and subvolume levels. Methods 27 patients treated with curative RT on the 1.5 T MR-Linac with at least weekly DWI in treatment position were included into this prospective analysis and divided in four risk groups (HPV-status and localisation). Tumor and lymph node volumes (GTV-P/GTV-N) were delineated on b = 500 s/mm2 images while ADC maps were calculated using b = 150/200 and 500 s/mm2 images. Absolute and relative temporal changes of mean ADC values, tumor volumes and a high-risk subvolume (HRS) defined by low ADC tumor voxels (600 < ADC < 900 × 10−6 mm2/s) were analyzed. Relative changes of mean ADC values, tumor volumes and HRS were statistically tested using Wilcoxon-signed-rank test. Results Median pretreatment ADC value for all patients resulted in 1167 × 10−6 mm2/s for GTV-P and 1002 × 10−6 mm2/s for GTV-N while absolute pretreatment tumor volume yielded 9.1 cm3 for GTV-P and 6.0 cm3 for GTV-N, respectively. Pretreatment HRS volumes were 1.5 cm3 for GTV-P and 1.3 cm3 for GTV-P and GTV-N. Median ADC values increase during 35 fractions of RT was 49% for GTV-P and 24% for GTV-N during RT. Median tumor volume decrease was 68% and 52% for GTV-P and GTV-N with a median HRS decrease of 93% and 87%. Significant differences from 0 for mean ADC were observed starting from week 1, for tumor volumes from week 2 for GTV-P and week 1 for GTV-N and for HRS in week 1 for GTV-P and week 2 for GTV-N. Conclusion Longitudinal DWI acquisition in HNC is feasible on a MR-Linac during the course of online adaptive MR-guided radiotherapy. Changes in ADC and volumes can be assessed, but future work needs to explore the potential for biologically guided treatment individualization. Trial registration: NCT04172753, actual study start: 09.05.2018.

In recent literature, the output correction factor in a 1.5T magnetic field, kB,Q,Clin, were determined only for the central axis (CAX), including our last work. At an MR-linac, the determination of the CAX position relies on e.g. MV-imaging and is at small field sizes located at the penumbra, whereas the maximum of the lateral profile (MAX) position is directly measurable. This study determined kB,Q,Ratio - the influence of the magnetic field on the output correction factor without a magnetic field, kQ,Clin - and kB,Q,Clin for both CAX and MAX fully experimentally for two MR-optimized ionization chambers, their conventional counterparts, and a solid-state detector. Additionally, the uncertainty due to intra-type variation for tabulated kB,Q,Clin was estimated. Measurements were conducted using an experimental setup consisting of a mobile electromagnet positioned in front of a standard clinical linac (6MV, 1.5T). Intra-type variation for the solid-state detector was assessed using four detectors. The change of absorbed dose to water was determined with alanine. Within the uncertainty, no difference in kB,Q,Ratio was observed between the MR-optimized chambers and their conventional counterparts. For the solid-state detector, no difference between CAX and MAX was observed. For all detectors, kB,Q,Clin remained constant down to a field size of 3x3cm2. At smaller field sizes, for MAX and for all detectors, kB,Q,Ratio decreased linearly. The maximum intra-type variation standard uncertainty was 0.009 for the smallest field size, with and without a magnetic field. In summary, kB,Q,Ratio and kB,Q,Clin, including its uncertainty, was successfully determined for five detectors at both CAX and MAX. For ionization chambers, kB,Q,Clin is lower at MAX, whereas for the solid-state detector, it remains the same. For the tabulated correction factors uncertainty, intra-type variation must be considered.

Objective. Commissioning an MR-linac treatment planning system requires output correction factors, kB ->,Qclin,Qmsrfclin,fmsr, for detectors to accurately measure the linac's output at various field sizes. In this study, k((B) over right arrow ,Qclin,Qmsr)(fclin,fmsr) was determined at the central axis using two methods: one that combines the corrections for the influence of the magnetic field and the small field in a single factor ( k((B) over right arrow ,Qclin,Qmsr)(fclin,fmsr)), and a second that isolates the magnetic field's influence, allowing the use of output correction factors without a magnetic field, kQclin,Qmsrfclin,fmsr, from literature for determining kB ->,Qclin,Qmsrfclin,fmsr. Approach. To determine k((B) over right arrow ,Qclin,Qmsr)(fclin,fmsr) and examine its behaviour across different photon energies and magnetic flux densities B in small fields, measurements with an ionization chamber (0.07 cm(3) sensitive volume) and a solid-state detector were carried out at an experimental facility for both approaches. Changes in absorbed dose to water with field size were determined via Monte Carlo simulations. To evaluate clinical applicability, additional measurements were conducted on a 1.5 T MR-linac. Main results. Both methods determined comparable k((B) over right arrow ,Qclin,Qmsr)(fclin,fmsr) results. For field sizes >3 x 3 cm(2), B ranging from -1.5 to 1.5 T and photon energies of 6 and 8 MV, no change of kQclin,Qmsrfclin,fmsr as a function of the magnetic field was observed. Comparison with measurement results from the 1.5 T MR-linac confirm this. For <= 3 x 3 cm(2), kB ->,Q(clin,Qmsr)(fclin,fmsr) depends on photon energy and B. For 1.5 T and 6 MV, B reduces kQclin,Qmsrfclin,fmsr up to 3% for the ionization chamber and up to 7% for the solid-state detector. Significance. k((B) over right arrow ,Qclin,Qmsr)(fclin,fmsr) were successfully determined for two detectors, enabling their use at a 1.5 T MR-linac. For field sizes of >3 x 3 cm(2), k((B) over right arrow ,Qclin,Qmsr)(fclin,fmsr) is one for most detectors suitable for small field dosimetry for all available perpendicular MR-linac systems, as confirmed in the literature. For these field sizes and detectors, the correction factor accounting for the dosimeter response change in the reference field due to the magnetic field, k((B) over right arrow ,Qmsr)(fmsr), can be used for cross-calibration. Therefore, future research may only focus on small field sizes.