Margin Calculation for Fundus Image Registration In Ocular Proton Therapy Planning
Abstract
Purpose
The delineation of uveal melanoma for proton beam therapy can be guided by 2D fundus photography registered to a patient-specific 3D eye model. However, the mapping suffers from camera-specific distortions and is further impacted by manual registration and calibration uncertainties. This study presents a computational analysis to quantify the necessary spatial margins by reprojecting retinal images through Monte Carlo simulations.
Methods
An algorithm was developed to map fundus features onto a spherical eye model using user-selectable projection models, implemented in MATLAB (v2024). The model calibrates pixel-to-millimeter scaling based on the anatomical distance between the fovea and optic disc (OD). To assess spatial fidelity, a Monte Carlo engine simulates geometric variations by perturbing input parameters including landmark e.g. positions/ distances, eye radius, and 3D rotation of the eyeball. Spatial uncertainty for tumor contour was quantified using distance to the 90% confidence boundary ultimately leading to an extra margin. The approach was tested on typical uveal melanoma cases for proton therapy with estimated uncertainties on single input quantities.
Results
A contour drawn on a registered fundus photo needs to be expanded when registration uncertainties of the fundus image are considered. Simulations reveal that geometric uncertainty is non-uniform, increasing towards the retinal periphery. For a typical uveal melanoma case an extra margin of >1 mm was calculated to allow for adequate coverage. Sensitivity analysis identified the optic disc position and fovea-OD distance as dominant sources of required margin.
Conclusion
The analysis demonstrates that a fundus photo registration on a 3D model as used in current treatment planning systems requires consideration due to registration and model uncertainties. A corresponding margin can be effectively calculated via probabilistic sampling in a Monte Carlo calculation, taking the relevant uncertainties into account.