S83
Brachytherapy - Gynaecology
ESTRO 2026
would be needed. Two such failing cases were noted; the result is thereby based on 47 patients. Results: The RMSE for the active dwell points were for the tandem channel 0.77 (±0.46) mm. The right and left ovoids had a RMSE of 1.06 (±0.48) mm and 1.14 (±0.45) mm, respectively. The mean maximum point-wise distance error for the tandem, and the right and left ovoid parts was 0.93 (+/- 0.45) mm, 1.20 (+/- 0.53) mm and 1.27 (+/- 0.51) mm. These results is also presented in the box plots in Figure 2.
Digital Poster Highlight 4672
Fully automatic applicator image alignment in cervical brachytherapy treatment planning Jakob Pettersson, Tom Niessen, Ola Weistrand, Lovisa Byman Research and Development, RaySearch Laboratories, Stockholm, Sweden Purpose/Objective: To develop and evaluate a fully automatic algorithm for alignment of intracavitary brachytherapy applicators on MR images. Material/Methods: An automatic alignment algorithm was developed in a research version of the RayStation treatment planning system. This was evaluated using a patient cohort of 50 MRI-guided clinical cervical brachytherapy plans with manually aligned tandem and ovoid applicators. The algorithm consists of three parts: (1) segmenting the outline of the applicator model in the MRI data using an analytical intensity-based approach, (2) computing the registration that gives optimal image similarity between the outline of the 3D applicator model and the segmented volume, and (3) a post- processing step to fine-tune the applicator position guided by the tandem tip. Figure 1 exemplifies the algorithm by visualizing the segmented applicator and the automatically aligned applicator on the MRI data.
Conclusion: The developed algorithm is proven to be a feasible method for automatic alignment of applicators to MRI data with millimeter precision, indicating no or minor manual adjustments would be needed. This can reduce the time and manual labor of aligning applicators during planning. Keywords: Applicator reconstruction, MRI, Cervical Multi-image modality AI-based automated applicator reconstruction in cervical cancer brachytherapy for various applicators Beatriz M Fernandes 1 , Andras Zolnay 1 , Ioannis Androulakis 1 , Theo van Walsum 2 , Remi Nout 1 , Ben Heijmen 1 , Linda Rossi 1 1 Department of Radiotherapy, Erasmus MC, Rotterdam, Netherlands. 2 Department of Radiology & Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands Purpose/Objective: While highly effective, image-guided adaptive brachytherapy for locally advanced cervical cancer is Proffered Paper 4691 time-intensive and leads to significant pain and anxiety experienced by patients [1]. In our center, treatment fractions last on average 9 hours. Speeding up in-fraction treatment preparation could significantly reduce fraction duration, and thereby improve patient experience and reduce personnel workload. Currently, manual applicator and needles reconstruction takes a median of 70 minutes [1]. In this study, we investigated fast, high-quality AI-based
To evaluate the model accuracy, the root-mean-square error (RMSE) and maximum point-wise distance were computed between the active dwell positions of the manually aligned applicator in the clinical plans and the corresponding dwell positions of the automatically aligned applicator. One patient was excluded for having detached ovoid parts, not supported by the proposed algorithm. Alignment was considered to have failed if the max error of any part was more than 10 mm off, assuming an extensive manual adjustment
Made with FlippingBook - Share PDF online