Retinal imaging with a Linnik interferometer
In an ophthalmic interferometer the eye is the sample-arm objective: its own optics focus the light, the retina sends some of it back, and a lens with a mirror reproduces the same arrangement in the reference arm.
Open in the canvas →Background
Optical coherence tomography (OCT) images the layers of the retina by interfering light returned from the eye with light from a reference mirror. In full-field OCT the whole field is illuminated at once and imaged onto a fast camera instead of scanning a single spot. A Linnik interferometer is a Michelson interferometer with an objective lens in each arm; for retinal imaging, the sample-arm objective is the eye itself, so the reference arm carries a lens of similar focal length and a mirror at its focus to match it[1].
The same lens-and-mirror pair is also how such instruments are tested without a volunteer: Auksorius, Borycki and Wojtkowski chose a 2.5 cm lens in front of a resolution target as a close optical approximation of a human eye, and an eye focal length of 2.5 cm in their system estimates[1]. Their system then destroyed the spatial coherence of the swept laser with a fast deformable membrane, so that multiply scattered light from the retina no longer interferes and corrupts the image with crosstalk.
What this setup demonstrates
The layout is a simplified version of the paper's Fig. 1, with the deformable membrane switched off (a flat mirror, left out here) and the relay lenses omitted. An 840 nm source splits at a 50/50 beamsplitter. The sample arm is a Human eye with a 25 mm focal length, a 6 mm dilated pupil and 5 % retinal reflectivity: the collimated beam focuses on the retina, and the reflected part leaves the pupil collimated again. The reference arm is the same thing built from parts — a 25 mm lens with a mirror at its focus — behind a neutral density filter that leaves about 6 % after the double pass, as in the paper. Both returns recombine at the beamsplitter and a tube lens focuses them onto the camera.
Change the eye's Retinal reflectivity to see the sample return and the camera reading change, or its power (in millimetres or dioptres) to give it a refractive error: the return then leaves the eye converging or diverging instead of collimated. Raise its Diffuse share to make the retina scatter like a real fundus: only the light aimed at the 6 mm pupil leaves the eye, about 1.4 % of what is scattered, so the sample return drops far below the reference — the reason the paper turns its reference arm down.
The retina is a single plane, mirror-like here and optionally diffuse, not a layered fundus, so the scene has no depth structure to image and no crosstalk to remove. The source is a single fixed wavelength rather than a sweep, and the carrier phase through the eye is not modeled, so the camera adds the two returns' powers instead of forming OCT fringes. The deformable membrane, its spatial-coherence gating, and the paper's three-lens defocus compensation are not modeled.