Enhancing ocular brachytherapy

Abstract

Brachytherapy with ruthenium-106 applicators is an effective treatment for ocular tumours. However, the surface dose profiles of these applicators are not homogeneous, and the effects of hot and cold spots on tumour and organ at risk doses have not yet been fully characterised. Furthermore, there is a lack of dedicated software to support treatment planning and to visualise three-dimensional dose distributions, which are necessary to accurately assess tumour control and the impact on organs at risk. To address these challenges, the efficiency of an existing simulation workflow is improved through the implementation of a particle tracking method combined with a novel external binning approach. This enables flexible, three-dimensional scoring of dose distributions within complex geometries. The impact of inhomogeneous surface dose profiles is evaluated using Monte Carlo simulations for three patient cases and different applicator models. The inhomogeneities are modelled using weighted phase space files. The results indicate that tumour control doses are generally maintained for applicator positions fully covering the target volume. However, hot spot scenarios can increase the dose to adjacent structures, and cold spots may reduce the local dose. To accelerate dose estimation, a 3D U-Net is trained to predict three-dimensional dose distributions from simulation data. The model achieves mean differences of (4.3 ± 15.8) % for the minimum tumour dose and (0.6 ± 0.6) % for the maximum scleral dose compared to the simulation results, suggesting that the model provides clinically sufficient accuracy to support treatment decisions. Building on the improved simulation workflow and the neural network-based dose prediction, a prototype web-based treatment planning interface is presented.