ESTRO 2025 - Abstract Book

S3637

Physics - Quality assurance and auditing

ESTRO 2025

Keywords: dosimetric audit, electronic portal imaging

References: 1. Greer PB, Lehmann J, Moore A, Experience with remote electronic portal imaging device-based dosimetric auditing for static and rotational intensity modulated radiotherapy, in-press, Phys. Imag. Radiat. Oncol. 2024

2323

Digital Poster Assessment of Spatial Localization Accuracy of AI-based Synthetic CT Egor Borzov, Eleonora Kuptzov, Saray Yehuda, Raquel Bar-Deroma Oncology, RAMBAM, Haifa, Israel

Purpose/Objective: The use of MR simulation (MRsim) in radiotherapy is an emerging field, with significant advancements aimed at eliminating the need for traditional CT simulation (CTsim) and reduce tumour margins in stereotactic radiosurgery (SRS). AI-based algorithms enable continuous HU-based synthetic CT (CTsyn) generation from specific MR sequences, making it suitable for dose calculation. However, it is crucial to evaluate the dosimetric and geometric accuracy of the generated CTsyn images. This study aims to assess the spatial localization accuracy (SLA) of CTsyn images. Material/Methods: Patients who underwent intracranial stereotactic radiosurgery (SRS) with 1, 3, or 5 fractions were selected for the study. The patients underwent two consecutive simulations using a Somatom go.Sim CT scanner (Siemens AG, Germany) and an Magnetom Sola MR scanner (Siemens AG, Germany) using the same immobilization equipment for both simulations. AI-based CTsyn images were generated using the Singo.via software (v. 08.07, Siemens AG, Germany). The geometric accuracy of the CTsyn images was evaluated for each fraction by comparing between CTsyn to CBCT registration and CTsim to CBCT registration in the XVI software (v. 5.0.4, Elekta AB, Sweden). The automatic fusion settings were standardized with a Mask of "Brain + 1cm" and “Grey Value T+R” to unify conditions and to minimize human error. Results: Data from 32 patients and 81 fractions were analyzed. For each fraction, differences between (CTsyn - CBCT) fusion and (CTsim - CBCT) fusion were calculated along the main axes (x,y,z), as well as for rotational axes (rotX, rotY, rotZ). The differences were found to be nearly independent across fractions, with a maximum variation of ±0.3 mm and ±0.2 degrees. These values were averaged across fractions for each patient. The results of the SLA assessment, averaged across patients, were as follows:

Conclusion: The objective of this study was to evaluate the SLA for CTsyn images. A challenge arose from poor visual quality of CTsyn when imported to the XVI software due to compatibility issues between XVI and Singo.via. Thus, visual verification of the fusion results became rather difficult. The results indicate that additional margins might be required for CTV when using MRsim that consequently eliminate the main advantage of the MR-sim workflow. While the averaged results suggest acceptable accuracy, individual patients showed discrepancies large enough to raise