Patient-specific mapping of fundus photographs to three-dimensional ocular imaging

Background

Ocular proton beam therapy (OPT) planning would benefit from an accurate incorporation of fundus photographs, as various intra-ocular structures, such as the fovea, are not visible on conventional modalities such as Magnetic Resonance Imaging (MRI). However, the use of fundus photographs in OPT is limited, as the eye's optics induce a nonuniform patient-specific deformation to the images.

Purpose

To develop a method to accurately map fundus photographs to three-dimensional images.

Methods

Personalized optical raytracing simulations were performed for 27 subjects, using subject-specific eye models based on corneal topography, biometry, and MRI. Light rays were traced through the eye for angles of 0°–85° with respect to the optical axis, in steps of 5°. These simulations provided a reference mapping between camera angles and retinal locations and were used to develop a mapping method without raytracing. The accuracy of this and earlier proposed methods was evaluated. Finally, the most accurate method was implemented in RayOcular, an image-based OPT planning system, and the fundus photography-based tumor contour was compared with MRI.

Results

When a patient-specific second nodal point is taken as a reference to describe the retinal location, the camera, and retinal angles show a strong linear relation with a small variation between subjects. At a camera angle of 60°, for example, a corresponding retinal angle of 59.9° ± 0.4° (mean ± SD) was found. When this linear relation is used to predict the corresponding retinal location (without raytracing) of a camera angle of 40°, the mean (Euclidian distance) error in the retinal location was 0.02 mm (SD = 0.06 mm), which was significantly (p < 0.001) lower than earlier proposed methods including EYEPLAN 4.16 mm (SD = 0.25 mm), the Lamberth projection −0.12 mm (SD = 0.46 mm) or polar projection 0.26 mm (SD = 0.57 mm). When implemented in the fundus view of RayOcular, the median distance between contours based on MRI and fundus photography was 0.2 mm (IQR = 0.1–0.3 mm).

Conclusions

The second nodal point provides a patient-specific reference for an accurate mapping of fundus photographs to three-dimensional images with sub-millimeter errors.