Computational Imaging of the Retina

A perennial problem in the design of instruments for imaging the retina, dating back over one hundred years, is to image the faint light reflected from the retina in the presence of strong reflections from the cornea and ocular lens. This becomes even more of a problem when it is desired to image over a wide field of view. One traditional solution is to scan a laser beam across the retina using optical scanners. The Optos ultrawide field imager uses this technique combined with some clever, but high-cost, optics. Various ophthalmic cameras, including fundus cameras and indirect ophthalmoscopes sidestep this problem by imaging over only a narrow field of view and employ highly optimised optics to remove glare. We propose a new computational-imaging approach for an ultrawide-field retinal camera, based on multiscale, multi-aperture imaging.
This technique involves segmenting the overall field of view into parallel narrow-field optical channels, to enable construction of retinal images with ultra-wide fields of view and using only inexpensive cameras - such as have been developed for mobile phones. Because each individual camera objective is required only to correct for the local eye aberrations over a small field of view, each channel can be optimised independently to achieve high-quality imaging. The local correction of aberrations that is proposed enables an otherwise impossible field of view.
The challenge to solve is to design and implement a new concept with the necessary and sufficient imaging performance. For example, there are field points that will be imaged partially through shared apertures, and the post-acquisition image reconstruction needs to account for these effects. Assuming the optimised individual channels are able to maintain a diffraction-limited correction over the entire retina using an effective aperture of 3.3mm, the approach promises to enable 50-megapixel imaging in a snapshot.
Additionally, the advantages of the proposed innovation also apply to the challenge of illuminating the retina. Because the field of view is segmented locally, so is the illumination if an illumination system is implemented for each channel, and the fundamental issues associated with reflections occurring at the cornea and anterior segment of the eye can be more easily solved.
The proposed innovation represents a step-change in the concept of retinal imaging: a computational imaging solution that cannot be achieved using conventional imaging. The quasi-simultaneous acquisition of the individual images will make it possible to mosaic an ultra-wide view of the retina without glare.