Page 37 - Computational Retinal Image Analysis

2 Optics of the eye 27

the center of the retina and slightly displaced from the optical axis of the eye
ball, contains the highest concentration of photoreceptor cones. There are widely
varying optical characteristics for each component of the retina. With regard to
the effect of the paths that light travels, the following simplification of layers
may be considered: the retina, the retinal pigment epithelium (RPE), the choroid,
and the outer sclera. The retina is in turn multi-layered, is highly transparent and
has retinal blood vessels embedded within it. It is about 200 μm to 300 μm thick
with a depression at the fovea. The retinal pigment epithelium is a single layer of
cells, highly absorbing (depending on retinal pigmentation) of about 10 μm. The
choroid is a vascular layer of about 250 μm thick containing highly oxygenated
blood and also melanin. The sclera (the white of the eye) is the outer and opaque
shell that protects the eye.
Light incident upon the retina is scattered and absorbed, and eventually diffused
through the volume. To model such diffusion it is convenient, if not necessary, to
employ a Monte-Carlo approach in which light rays are traced, using a stochas-
tic model, according to scattering and absorption models at the retinal structures
[19–21]. These models are defined by a mean-free path and a scattering phase
function, which define stochastically the distance a light ray will travel before it is
scattered (i.e., it changes direction of propagation) and the angular probability den-
sity function of scattering towards a new direction, respectively. Effectively, light
rays enter the tissue, propagate, scatter (change direction), propagate, and so on.
On propagation, the intensity associated with each light ray is absorbed according
to the Lambert-Beer law [11].
After tracing sufficient rays, the macroscopic light transport, both in the ballistic
and diffuse regimes (few and many scattering events respectively) are reproduced
as illustrated in Fig. 4. Defining mean-free paths, phase functions, absorption co-
efficients, and refractive indices for all involved tissues, and noting that many of
these parameters are wavelength-dependent, enables numeric but accurate model-
ing of light diffusion within the retina. The power of Monte-Carlo modeling is that,
given these parameters are known, it enables the computation of light diffusion
under conditions of arbitrary geometrical complexity. Importantly, it is possible to
track the polarization of each individual light ray. An alternative use of Monte-Carlo
simulation becomes useful if the optical properties of the tissue are not known but
experimental data is available: Monte-Carlo simulations may then be used to actu-
ally compute what optical parameters explain experiments, in what is called inverse
Monte-Carlo approach.
As discussed in the next section the optical properties of the retina vary
greatly with wavelength: in particular red light is less absorbed than blue light.
This results in a higher fundus reflectance but also a lower contrast in red light
and so it is common to employ red-free filters in fundus cameras to increase the
contrast of the retinal vasculature. More generally spectral imaging of the retina
can be used to detect vascular oxygenation to aid classification of arteries and
veins [22–27].