10.4.3   Light Detection

Light detection, or equivalently sensitivity, for OCT commonly refers to the
weakest retinal reflection that is detectable relative to the optical power that
enters the eye. It is commonly defined as the square of the signal amplitude
divided by the variance of the noise floor [49]. While the sensitivity of OCT
varies widely depending on the design, typical values are between −90 dB
(10^-9) and −100 dB (10^-10). Figure 10.5 shows sensitivity levels for specific OCT
instruments. To obtain a feel for what this means for retinal imaging, the
brightest retinal reflections start between −50 to −60 dB (10^-5 to 10^-6 less than
the amount of light that enters the eye) and typically come from the nerve
fiber, photoreceptor, retinal pigment epithelial, and choroidal layers. The
remaining 30 to 40 dB sensitivity of a typical OCT instrument (with −90 dB
sensitivity) is sufficient for detecting structures in essentially all of the stratified
layers of the retina. Assuming that reflections in the transparent retina
are due to refractive index mismatches, Δn, −90 to −100 dB corresponds to a
Δn of only 10^-4 to 10^-5. Such high sensitivity in an OCT instrument is achieved
via optical heterodyning in which weak tissue reflections are amplified into
the photon noise regime by means of a strong reference beam. This approach
is in stark contrast to that employed by conventional flood illumination and
SLO systems that rely on direct rather than interferometric imaging.

The sensitivity of time-domain OCT and spectral-domain OCT depends
on slightly different temporal parameters. The signal-to-noise ratio (SNR) of
time-domain OCT is given by:

where η is the detector quantum efficiency, P_retina is the optical power of a
retinal reflection that reaches the detector, hv is the energy of a photon, and
Δf_xis the acquisition bandwidth. Equation (10.7) reveals a fundamental trade-off
between sensitivity and acquisition rate (or bandwidth). Higher acquisition
rates (i.e., wider bandwidths) fundamentally decrease sensitivity, yet are
often necessary to mitigate eye motion artifacts or to achieve denser sampling
when, for example, a broader light source is used for higher axial resolution.
While power entering the eye can be increased to offset faster acquisition
speeds, a hard upper limit is imposed by light damage to the ocular tissue.

The signal-to-noise ratio of spectral-domain OCT is given by:

where T is the exposure time of a single A-scan [50]. In contrast to time-domain
OCT, spectral-domain OCT is independent of the acquisition rate
due to the fact that an entire A-scan is collected simultaneously rather than
sequentially. Sensitivity, therefore, does not degrade with increased resolution.
This is a major advantage, as it permits ultrahigh axial resolution (~3 μm)
images of the retina to be acquired at the same sensitivity as standard resolution
(~10 μm) OCT images. This is particularly relevant for AO retinal imaging
applications where speed and sensitivity are critical. Note that the signal-to-noise
benefit of spectral-domain OCT relative to time-domain OCT goes
as:

where N_A is the number of resolution elements that fit within a single A-scan.
This reveals that the sensitivity advantage of spectral-domain OCT over that
of time-domain OCT can be substantial (two to three orders of magnitude
greater) for A-scans composed of many resolution elements. Conversely, for
very thin samples that traverse few A-scan resolution elements, the sensitivity
advantage of spectral-domain OCT is, in principle, largely lost and both types
of OCT perform similarly in terms of sensitivity, speed, and resolution.

As specified in Eqs. (10.7) and (10.8), sensitivity is proportional to the
retinal reflection that reaches the detector, P_retina, which in turn depends on
the collection efficiency of the eye’s optics. The combination of a large pupil
with AO to correct the wave aberration of the eye will, in principle, permit
substantial increases in OCT sensitivity above that in current research and
commercial instruments. For example, the Vienna AO tomographic scanning
OCT instrument utilized a deformable mirror (OKO Technologies) for aberration
correction across a 3.7-mm pupil [38]. By adding this basic adaptive
optics system to correct mainly the defocus of the system (almost 1 D), they
found an increase in the SNR of 9 dB when using adaptive optics at the 3.7-
mm pupil size. The mirror used in this system was not optimal for higher
order aberration correction; improvements in sensitivity beyond what was
shown could be obtained by using a better wavefront corrector, such as the
Xinξtics deformable mirror. As an example, the Indiana AO line illumination
spectral-domain OCT ophthalmoscope employed such a mirror for the correction
of aberrations across a 6-mm pupil. For this ophthalmoscope, the
signal-to-noise ratio of the detected reflection from the photoreceptor layer
was highly sensitive to the level of ocular aberrations and defocus with changes
of 11.4 and 13.1 dB observed when the ocular aberrations (astigmatism, third
order, and higher) were corrected and when the focus was shifted by 200 μm
(0.54 D) in the retina, respectively. Over and above the sensitivity increase
obtained from adding adaptive optics, the sensitivity will increase with the
pupil size. For a diffraction-limited system, simply increasing the exit pupil
from 1 to 6 mm theoretically increases the sensitivity of the system by
15.6 dB.