Purpose
Spectral images of albino rats were recorded using a modified conventional clinical non-mydriatic fundus camera. Integration of a liquid crystal tuneable filter, as shown in figure 1, spectrally filters the flash illumination to enable time-sequential and spectrally randomised acquisition of spectral images of rat retinas with a spectral resolution of 7nm. Automated computer algorithms correct for image-to-image rotational and translational misregistration and for spatial variations in illumination. Pixel-based processing of the spectral data cube using commercial spectral processing software (ENVI) enables a map of spectral signatures to be constructed that provides a semi-quantitative map of biochemical chromaphore concentrations, whilst a physical model of light propagation in the retina enables quantification of retinal blood oxygenation [1,2,3].
We have developed a hyperspectral imaging technique for imaging the rat ocular fundus based on time-sequential recording of narrowband images. Processing methods for calculation of chromaphor concentrations effectively attenuates the artefacts that can be introduced by time-sequential recording. Accurate quantification requires the development of a rigorous model for light propagation. Measurements obtained for blood oxygenation are as expected for arterial and venous blood, but improvements in accuracy require refinement of this model, particularly for albino rats where the absence of a highly absorbing retinal pigment epithelium, as is found in pigmented mammals, results in a very challenging calibration.
Spectral Imaging of the Rat Retina
Andy R Harvey1*, Eirini Theofanidou1, Ied Al-Abboud1, Mark Graham2 and Andy C Hargreaves2.
1School of Engineering and Physical Sciences, Heriot-Watt University, Edinburgh, United Kingdom
*www.ece.eps.hw.ac.uk/~arharvey; a.r.harvey@hw.ac.uk;
2 AstraZeneca R & D Charnwood, Loughborough LE11 5RH, UK
Methods
Results
Conclusion
            580 nm                             600 nm
Figure 2 Narrow-band images of the retina of an albino rat
Figure 4 Application of a Physical model for calculation of blood oxygenation
 Vessel  tracking
Profile  extraction
 Illumination estimation on both sides of each profile.
 Transmission (T)  & optical density( OD) estimation
 Non-linear fit of physical model to optical densities at different wavelengths to estimate oxygen saturation OS
1
2
3
4
5
Pixel value
T
OD
OD
OD
λ[nm]
λ[nm]
ODs  values
Fitted model
Vein ,OS=55%
Artery, OS=95%
Figure 5  Method for measuring optical density of blood vessels and for fitting blood oxygenation to measured optical density.
Figure 6 Calculated oxygenation along a blood vessel and artery using the non-linear fit methods depicted in figures 4 and 5.
OS%
Spatial dimension along a vessel (pixels)
References
COMMERCIAL RELATIONSHIP a: QinetiQ, AstraZeneca,
OS
The ability to detect retinal disease in rats at an early stage is of great significance to drug discovery programmes where toxicity-induced retinal disease may be evident at an earlier stage in spectral images of the retina, than from time-consuming and costly histological studies of the retina. We report here on the development of instrumentation and methodology for recording and processing spectral images of rat retinas.
Figure 1 Hyperspectral rat fundus camera
400-720 nm
LCTF
CCD
400-700
   nm
Absolute and accurate determination of chromaphor concentrations requires the incorporation  of a physical model for light propagation in the retina as depicted. We have incorporated a physical model into the algorithm depicted in figures 4 and 5: an a automated process extracts and tracks blood vessels, a non-linear fit of a physical model for light propagation to each narrow-band image is applied to transverse intensity profiles of blood vessels. A key  parameter derived from the non-linear fiot is the blood oxygenation [3].
Calculated blood oxygenation is shown in figure 6 as a profile along a vein and along an artery. As expected arterial blood is calculated to be approximately 100% oxygenated and venous blood is approximately 60% oxygenated. The considerable variation along the blood vessels is an artefact introduced by the strongly varying spectral environment of the choroid.
Figure 3 Spectrally unmixed image of a rat retina indicating blood oxygenation
A typical set of narrowband images is shown in figure 2. Supervised  or unsupervised linear spectral unmixing of the full retinal spectral data cube enables a semi-quantitative mapping of chromaphor concentrations. For example, figure 3 shows a false-colour image of blood oxygenation: oxygenated arterial blood appears as red whereas deoxygenated venous blood appears blue. Since the imaged rat is an albino, the choroidal blood circulation is also visible.
A
V
1. A.R. Harvey et al., “Spectral imaging of the retina”, SPIE Vol. 6047 (2006).
2.D. J. Mordant et al., “Hyperspectral imaging of the human retina - Oximetric studies”, ARVO 148-B257 (2007).
3.I. Alabboud et. al, “Quantitative spectral imaging of the retina”, ARVO 2581-B658 (2007).
A.