Deformable image registration (DIR) in Swiss radiotherapy: Usage patterns survey and multi-institutional deformable dose accumulation comparison.

Background and purpose: Deformable image registration (DIR) enables advanced applications for image-guided and adaptive radiotherapy. However, DIR has not yet been widely adopted in clinical settings. This study addresses two challenges: (1) evaluating DIR practices and identifying barriers through a survey of Swiss radiotherapy institutes, providing Swiss-specific insights, and (2) assessing multi-institutional deformable-based dose accumulation (DDA) uncertainties by analyzing dose discrepancies on the accumulated dose. These findings aim to inform consensus guidelines, quantify DDA uncertainties, and support standardized future DIR integration into routine RT.

Material and methods: All Swiss radiotherapy institutes with available contact details were invited to participate in the study's two parts: a survey and a multi-institutional DDA evaluation. The survey consisted of 25 questions and was conducted from August 23 to October 31, 2024. Survey results were compared to surveys from other countries. For the DDA evaluation, participants used shared datasets, including a lung cancer case with two dose distributions, one conventional VMAT dose and one artificially created cuboid dose, to perform DIR and accumulate doses. The cuboid dose was used to have a standardized dose with a "perfect" dose gradient. The level of inter-institutional dose discrepancies was assessed visually and by calculating dose differences on a voxel level.

Results: Eighteen out of 26 institutes (69%) completed the survey. All 18 institutes use rigid image registration (RIR) clinically, 11 (61.1%) use DIR. Five of those institutes use DIR in research projects. RIR is primarily applied for image fusion (100% of the institutes), patient positioning (88.9%), and as a preliminary step to DIR or adaptive RT (55.6%). DIR use cases include image fusion, dose accumulation, and adaptive workflows. Barriers to DIR adoption in clinical routine include software limitations and a lack of quality management methods. Standardized quality management of DIRs is largely missing in Swiss radiotherapy institutes so far. Six institutes participated in the dose accumulation part, submitting dose accumulation results for a VMAT and a cuboid dose distribution. Visual inspection revealed inter-institutional differences, particularly in steep dose gradient areas. Quantitative analysis showed average voxel-wise absolute differences of 0.01-0.03 Gy (VMAT), corresponding to 0.1%-0.3%, and 0.18-0.45 Gy (cuboid), corresponding to 1.8%-4.5%, for a total dose of 10 Gy. Differences were more pronounced in steep dose gradients and, consequently, in proximity to the PTV. The voxel-wise maximum-minimum analysis highlighted variability in gradient regions, even for the same algorithms used by different institutes, emphasizing the impact of software, software usage, and workflow choices on dose accumulation outcomes. The cuboid dose showed increased interpretability of the uncertainty compared to VMAT doses.

Conclusions: The usage of DIR in Swiss RT institutes is primarily hindered by workflow inefficiencies in commercially available software and insufficient quality management options. Specifically, a lower level of standardized validation, commissioning, and uncertainty handling for DIR was observed in the Swiss RT institutes compared to survey outcomes in other countries. The multi-institutional DDA comparison increased the knowledge on DDA uncertainty areas and magnitude. Cuboid doses might help to evaluate uncertainties on the dose level in a standardized way in the future.