S3013
Invited Speaker
ESTRO 2026
5348 EPID in vivo dosimetry: Present and future IGOR OLACIREGUI RUIZ Radotherapy, Netherlands Cancer Institute, Amsterdam, Netherlands This presentation will examine the current state of EPID-based in vivo dosimetry (EIVD), its integration into online adaptive radiotherapy (oART) workflows, and the emerging role of AI in next-generation systems. EIVD is the most widely implemented form of in-vivo plan-specific quality assurance (PSQA). Two main algorithmic approaches exist, forward and back projection systems, yet broad clinical adoption remains limited. Key barriers include the vendor–user gap, EPID technology constraints, the lack of open access to EPID images and metadata, limitations in MV acquisition software, insufficient automation, and the resulting workload burden. In oART, in-room imaging is used to generate online patient models for contouring, dose calculation, and plan re-optimization. Several of these steps remain semi-automated and require human validation, making some degree of error unavoidable. Robust PSQA tools capable of detecting clinically relevant discrepancies—either before treatment (PT) or after treatment (offline IV)—are essential. PT PSQA for oART should be fast, automated, and performed in the narrow window between plan adaptation and delivery. Proposed approaches have combined independent online patient model generation with secondary dose calculation algorithms. Despite strong clinical interest, these methods remain challenging to implement due to their complexity and the need for essentially zero tolerance to false negatives or false positives. By contrast, extending current EIVD systems to support offline IV PSQA in oART workflows is far more straightforward, as it is primarily a software-engineering task. The system must simply receive the online patient model and the adapted plan to verify dose delivery, providing a final end-to-end check of the adaptive workflow similar to conventional implementations. Recent work has explored AI-based enhancements to EIVD. In forward-projection systems, deep learning (DL) methods are emerging as alternatives to Monte Carlo for predicting portal images. In back-projection systems, DL has been applied for direct 2D/3D dose reconstruction, to overcome algorithmic limitations, and to support error detection or classification. The key question is how close these AI-driven approaches are to clinical translation, given that many studies rely on synthetic data, show limited generalizability, and may face regulatory hurdles. By the end of the presentation, we hope to have a clearer perspective on how to prioritize future
research and guide the development of next-generation EIVD systems.
5349 In-vivo dosimetry in brachytherapy: From now to wow Kari Tanderup Danish Center for Particle Therapy, Aarhus University Hospital, Aarhus, Denmark Brachytherapy is characterised by steep dose gradients that enable the delivery of high doses to the target while sparing surrounding healthy tissues. However, these same gradients imply that even small deviations from the planned source position can lead to clinically significant dosimetric errors. In addition, brachytherapy workflows rely on multiple manual steps, making the modality inherently vulnerable to operational mistakes. At present, treatment verification is predominantly based on manual checklists and procedural controls, and modes of failures which are not included in these checks may go unnoticed – some of them latent errors that can persist over extended periods. Historically, in-vivo dosimetry in brachytherapy was primarily used for organ dose measurements, such as rectal dosimetry. With the widespread adoption of 3D image-guided brachytherapy, this application has become less relevant, as imaging now provides more comprehensive dose information. Nevertheless, this evolution has opened the door to a new paradigm: using in-vivo dosimetry as a “fingerprint” of the source stepping pattern during treatment delivery. By capturing temporal and spatial characteristics of the radiation signal, in-vivo systems can provide independent verification of source position and dwell sequence. Recent studies have demonstrated that time-resolved dosimetry methods significantly improve the ability to detect deviations from the treatment plan. A key advantage of these approaches is their capacity for real-time or near-real-time monitoring, enabling the identification of discrepancies during treatment delivery. This creates the opportunity to interrupt treatments in the presence of gross errors, thereby enhancing patient safety. Furthermore, approaches based on source tracking rather than direct dose measurements can reduce experimental uncertainties and provide more robust verification of source dynamics. Despite promising technological advances, clinical implementation of in-vivo dosimetry in brachytherapy remains limited. Moving from research prototypes to routine clinical use will require well-designed clinical trials, standardisation of methodologies, and systematic reporting of errors and near-misses.
Made with FlippingBook - Share PDF online