Page 57 - Computational Retinal Image Analysis

3 Ophthalmic instruments 47

The axial size of the images is restricted now by the instantaneous coherence
length of the source, however this is not limited to a couple of mm as in the case of
CB-OCT. Typically MEMS based SSs can provide axial ranges from 3 to 4 mm to
1 cm depending on the speed data is sampled. In terms of axial resolutions, SSs have
a more limited spectral range, of maximum 100 nm. As current technology only al-
lows for fast (>100 kHz), stable SSs for wavelengths superior to 1060 nm, SS-OCT
instruments can only deliver image with axial resolutions of 4.8 μm in air. However,
SS-OCT instruments provide better sensitivity than CB-OCT and a better depth pen-
etration in the choroid as the scattering is reduced at longer wavelength. The use
of CB-OCT at longer wavelengths is restricted due to the unavailability of the fast
cameras.
3.8.5 Methods of generating images in SD-OCT
To produce an A-scan in SD-OCT, the integral of the product between an experimen-
tally acquired spectrum, obtained by interfering light from the sample and reference
arm of the interferometer and a kernel function is calculated. To eliminate the chirp-
ing due to nonlinear wavelength mapping and due to unbalanced dispersion, two
methods can be used: a wide spread method is based on FFT that operates on the
experimental spectra, or a novel method (Master/Slave) that operates on the kernel
function.
When using the FFT based method, each experimental spectrum is re-sampled
then multiplied by a function that cancel the effect of the unbalanced dispersion. A
relationship between the phase of the modified spectrum and a new wavenumber
distribution is obtained, then an A-scan compensated for broadening is produced
by calculating the FFT of the product between the re-sampled spectrum and an
apodization function. The process of calibration (computing the new wavenumber
distribution) is performed at each spectral acquisition, before imaging is carried
out. To produce accurate information on the reflectivity from within the sample,
each acquired experimental spectrum is resampled according, typically via a cubic
B-spline interpolation, then correct for the unbalanced dispersion. All these opera-
tions limit the capability of the FFT based instruments to operate in real-time as
they are executed sequentially. In addition, the resampling operation is time con-
suming and must be applied to each acquired experimental spectrum. In contrast
to the FFT method, the process of obtaining an axial reflectivity profile utilizing
the MS method consists in modifying the kernel function [56]. Thus, a transversal
reflectivity profile (T-scan), for a given axial position z i can be reduced to a simple
matrix multiplication [57]. The advantage of this mathematical representation of a
T-scan is that it can be generated in real-time. Collections of T-scans can be pro-
duced in real-time as now all operations are parallelizable. Furthermore, an entire
en-face MS based SD-OCT image can be produced in real-time, multiple en-face
images from different axial positions along depth can also be generated without
producing entire large 3D volumes. This allows for defining axial regions of in-
terest where B-scans can be generated without producing redundant data, hence