Hyperspectral imaging of the retina presents a unique opportunity for
direct and quantitative mapping of retinal biochemistry - particularly of the
vasculature where blood oximetry is enabled by the strong change of absorption
spectra with oxygenation. This is particularly pertinent both to research and
to clinical investigation and diagnosis of retinal diseases such as diabetes,
glaucoma and age-related macular degeneration. Spectral processing methods
such as linear spectral unmixing1enable semi-quantitative depiction of retinal
oximetry as shown in figure 1, but there is a clinical requirement for more
accurate oximetry within retinal blood vessels.
A liquid crystal tuneable filter has been incorporated into a conventional
fundus camera to enable computer-controlled, random-access spectral filtering
of the source with 10nm spectral resolution (figure 2). The available output
image has been reimaged to enable the full 60° field of view to be
when used with a cooled low-noise CCD. A computer running under LabView is
used for instrument control and data acquisition.
The process for calculation of blood oxymetry is illustrated as shown in
figure 4. This involves a non-linear fit of a physical model for light
propagation to each narrow-band image2. An
interpolated estimate of the illumination at the retina obviates the need for
calibration (figure 5).
Due to non-uniform illumination and random movement of the eye ball and
reflections from lens surface recorded images need to be calibrated and
co-registeredto correct for rotational
and translational offsets introduced between images. A typical set of images
is shown in figure 3 for wavelengths between 580 nm and 600 nm.
The oxymetry map (figure 6) can be rapidly calculated for a comprehensive
fraction of the retinal vasculature. Random variations in oxygenation along a
blood vessel of assumed constant oxygenation are typically less than 4%.
This technique enables a
highly flexible approach to acquiring a wide-field spectral data cube for
research and clinical exploration. The use of vessel tracking and fitting of
an analytical mathematical model to the transverse profiles suppresses the misregistration
and calibration artefacts that are normally an issue for time-sequential
techniques. For clinical application a wide-field snapshot technique3is desirable: the
investigative data recorded here has informed the optimisation and design of
the novel snapshot system spectral retinal imager4.Future work using this instrument will assess clinical
interpretation and refinement of the mathematical model for light propagation.
2 Ophthalmology Unit, Cheltenham General
Hospital, Cheltenham, United Kingdom.
580 nm590 nm600 nm
Illumination estimation on both sides of each
(T)& optical density( OD) estimation
Non-linear fit of
physical model to optical densities at different wavelengths to estimate oxygen saturation OS
Spatial dimension along a
D.J.Mordant,I.AlAbboud,A.R.Harvey,A.I.McNaught, Hyperspectral Imaging Of
the human retina: oximetric studies,ARVO 2007.
 I. AlAbboud ,A. R. Harvey ,Hyperspectral
imaging of the eye for early detection of retinal diseases ,
 Andrew R. Harvey, David W. Fletcher-Holmes,
Alistair Gorman, Kirsten Altenbach, Jochen Arlt, Nick D. Read, Spectral
imaging in a snapshot, Spectral Imaging: Instrumentation, Applications, and
Analysis III, Proc. SPIE Vol. 5694, p. 110-119, (2005).
G.Muyo,A.Corman,I.Alabboud,D.J.Mordant,A.I.McNaught,A.R.Harvey, En Face Snapshot Spectral
Imaging Of The Retina ,ARVO 2007.
RELATIONSHIPa: QinetiQ, AstraZeneca, b :Alcon;
Allergan; MSD; Pfizer.