S1676
Physics - Detectors, dose measurement and phantoms
ESTRO 2026
3 Department of Physics, University of Pisa, Pisa, Italy. 4 Department of Experimental and Clinical Biomedical Sciences Mario Serio, University of Florence, Florence, Italy. 5 Department of Computer Science, University of Pisa, Pisa, Italy. 6 Department of Radiation Oncology, Centro di Riferimento Oncologico di Aviano (CRO) IRCCS, Aviano, Italy. 7 Department of Radiation Oncology, Azienda Ospedaliero-Universitaria Careggi, Florence, Italy Purpose/Objective: In-vivo dose monitoring with Electronic Portal Imaging Devices (EPIDs) in radiotherapy can be performed by comparing an acquired EPID image with a reference expected image, either directly, or by first converting it into water equivalent dose [1,2]. We developed a Deep Learning (DL) model, that transforms EPID images into 2D water-equivalent dose images (see Figure 1). In this study, we test whether our DL-based framework is sensitive to the detection of treatment errors, and what metrics can best be used for error detection. Material/Methods: We irradiated different commercial radiotherapy phantom with different fields, and acquired EPID images without errors (reference) and with controlled treatment errors. Portal dose (PD) images were obtained from EPID images using our DL framework. To compare images with and without treatment errors, we applied various alert metrics, including commonly applied metrics in 2D in-vivo EPID dosimetry (like gamma-index) and less commonly applied metrics related to dose. We perform a systematic comparison, evaluating their efficacy for the detection of errors in several phantoms. Results: We found that the gamma-index alone, as frequently applied in comparisons, is not always enough to detect errors, particularly when using narrow fields some errors may be difficult to detect. For example, when irradiating a CIRS phantom with a 1 x 15 cm2 field, the Gamma Passing Rate was 95%, while there was a 5% excess in monitor units (Fig 2(a)). Applying the gamma analysis in combination with various dose difference- relatedmetrics, plus the dose profiles, can help in error detection (Fig 2(b)). The most important treatment errors can be detected by using a combination of metrics.
time was (16.0 ± 3.3) min, while total door-to-door times ranged from 74 to 114 min, depending on the anatomical extension and number of subplans.
Conclusion: The implementation of TMI/TMLI on the Halcyon G platform suggests stable and robust performance throughout the treatment workflow.The results indicate that combining pretreatment verification, transit dosimetry performed according to TG-307 guidance, and robustness analysis support accurate and consistent delivery, enabling a safe, efficient, and reproducible workflow for highly complex treatments. References: 1. Mancosu P, Cozzi L, Muren LP. Total marrow irradiation for hematopoietic malignancies using volumetric modulated arc therapy: A review of treatment planning studies.2. Shahid T, Mandal S, Biswal SS, et al. Preclinical validation and treatment of volumetric modulated arc therapy-based total bone marrow irradiation in Halcyon™ ring-gantry linear accelerator. Keywords: TMI/TMLI, multi-isocenter, robustness, consistency Digital Poster 3968 Usage of DL-based portal dose images for treatment error detection with transit dosimetry: a systematic comparison of analysis metrics Aafke Christine Kraan 1 , Emmanuel Uwitonze 1,2 , Rossana Lanzillotta 1,3 , Carlotta Mozzi 4 , Lorenzo Marini 1,5 , Michele Avanzo 6 , Francesca Lizzi 1 , Livia Marrazzo 4,7 , Icro Meattini 4,7 , Stefania Pallotta 4,7 , Giovanni Pirrone 6 , Alessandra Retico 1 , Cinzia Talamonti 4,7 1 Department of Physics, Istituto Nazionale di Fisica Nucleare, Pisa, Italy. 2 Department of Radiation Oncology, Rwanda Cancer Centre, Kigali, Rwanda.
Made with FlippingBook - Share PDF online