3-D retinal surface inference: stereo or monocular fundus camera?

Three-dimensional (3-D) imaging of the eye fundus, and in particular of the optic disc, is widely used to assess glaucoma progression over time. In the literature, 3-D images of the optic disc have been obtained from stereo and monocular fundus cameras. While stereo systems are the gold standard for optic disc examination, monocular systems are less expensive, and therefore of more practical use. This stimulated a thorough investigation of the limits and advantages of these two imaging modalities. Our conclusion is that monocular imaging is generally not suitable for 3-D estimation. This is attributed to the fact that monocular systems do not allow a change in the vantage point from which the retinal surface is observed, despite variations in the relative pose between the eye and the fundus camera. To validate this analysis we carry out several experiments on both stereo and monocular fundus cameras with standard 3-D reconstruction algorithms. Furthermore, we devise a calibration procedure to quantify experimentally the highest accuracy achievable with a stereo system.