Comparison of model-based and delineation-based target volume definition for ocular proton therapy

Background In recent years, eye-specific, high-resolution MRI protocols have become available for ocular tumours, enabling 3D tumour delineation. Despite these advances, ocular proton therapy typically still relies on model-based target volume definition. Therefore, the aim of this research was to compare model-based and delineation-based target volume definition for ocular tumours. Methods Tumours were delineated for 70 uveal melanoma patients. From the delineated tumour base, polynomial models of different degrees were generated. Additionally, model variations with expanded base and added thickness were generated. Models were considered acceptable at a coverage level of 98% or a maximum underestimation of 0.4 mm. The best model for each patient was defined as the acceptable model with the smallest volume. Results For 69/70 patients, an acceptable tumour model could be generated, out of whom 12/69 needed an expansion of the tumour base or added tumour thickness to produce at least one acceptable model. At a coverage level of 98%, tumours were overestimated with 20.5% (5th-95th percentile: 5.7%-59.7%), corresponding to a median absolute surface distance of 0.42 mm. Irregularly shaped tumours showed larger overestimations than dome-shaped tumours (16.4% vs 36.6%, p < 0.001). Conclusion Tumour models as used in ocular proton therapy treatment planning systems can achieve acceptable tumour coverage. However, to achieve this coverage, the tumour models overestimated the delineated tumour volume.