10.4.2   Resolution Limits of OCT

In contrast to traditional imaging systems, such as conventional flood illumination
systems and SLOs, transverse and axial resolutions for OCT are independently
controlled with the latter being dictated by the mean wavelength
and bandwidth of the illumination source rather than the numerical aperture
(NA). Specifically, the axial resolution of a traditional diffraction-limited
system is proportional to the square of the pupil diameter, while for OCT, the
full width at half height (FWHH) of the axial PSF is given by:

where λ is the center wavelength of the source and Δλ is the FWHH of its
spectrum. Equation (10.6) assumes a Gaussian spectral source and reveals
that increased axial resolution is realized with broader spectral sources. The
λ² dependence indicates that longer wavelength sources require disproportionately
broader spectral widths to achieve the same resolution. As examples,
the commercial Stratus OCT3 (Zeiss Meditec, Inc.) has an axial resolution
in retinal tissue of ~10 μm and ultra-high-resolution OCT recently demonstrated
≤3 μm [38, 40, 43, 44]. (See Figure 10.5 for specific resolution
examples.)

While the axial resolution in OCT is governed by the source spectral properties,
its transverse resolution is governed by the same focusing properties
(i.e., limited by diffraction and aberrations) that govern conventional flood
illumination systems and SLOs. It is for this reason that AO and OCT are
viewed as complementary technologies. The high transverse resolution
obtained with AO is fundamentally independent of the high axial resolution
from OCT.

The need for very high 3D resolution for imaging the retina microscopically
is illustrated in Figure 10.7. The figure shows point spread functions for
the major retinal ophthalmoscope architectures in combination with AO as
well as those for two commercial instruments (without AO). A scaled histological
cross section of the human retina is shown on the left for comparison.
In the figure, the three experimental AO-OCT point spread functions are
signifi cantly smaller than those of the other ophthalmoscope architectures
[37–39]. Currently, the smallest point spread volume is that of the Indiana
AO spectral-domain OCT ophthalmoscope at 51 μm³, which is 272 and 78
times smaller than that of the commercial SLO and OCT systems, respectively.
Note that this point spread function is at least as small as many of the
cell nuclei shown in the retinal cross section, suggesting that these cells could
be resolved and could be observed if there is sufficient backscatter signal.

The AO-OCT point spread functions depicted in Figure 10.7 assume the
axial (i.e., coherence gate) and lateral (i.e., focal plane) resolutions are super-
imposed at the same retinal depth, as is intrinsically and conveniently the case
for the conventional flood illumination and SLO systems. For OCT, this need
not be the case. The axial and lateral resolutions of OCT systems are independent
and therefore can be at physically separate locations in the retina.
This provides an additional degree of freedom. For example, the focal plane
could be strategically positioned axially in front of the coherence gate such
that cells lying at the focal plane are viewed at high lateral resolution and in
pseudo-transmission by light reflected from cells at the more deeply located
gate. An example of this scenario is shown in Figure 10.10 (right). This additional
freedom, however, necessitates additional ophthalmoscope complexity
in order to control both the focal plane and gate positions. This control must

account for instabilities in the eye, such as microfluctuations in accommodation
and head movements that unfortunately independently shift the focus
and gate positions. For example, axial head and eye motion shifts the gate,
but not the focus, while accommodation shifts the focus, but not the gate.

It may appear at first glance that to achieve the AO-OCT point spread
functions in Figure 10.7 for time-domain OCT, exquisite control is required
to precisely position the gate at the focal plane. Fortunately, the necessary
overlap is noticeably more relaxed due to the finite depth of focus that straddles
the geometric focal plane. More so, because depth of focus is proportional
to the square of the transverse resolution (which is inversely proportional
to pupil size for an unaberrated eye), placement of the gate is increasingly
less difficult for smaller pupil sizes, albeit at the expense of reduced lateral
resolution. Figure 10.8 shows the theoretical trade-off between lateral resolution
and depth of focus for an unaberrated eye at 0.83 μm. As an example for
a 6-mm pupil, the transverse resolution ranges from 2.8 μm (ω_o) to 4.0μm
( 2ω_o) across a narrow 60-μm depth of focus. For the same system, but with

a smaller 3-mm pupil, the transverse resolution is reduced by a factor of two
(5.6 to 8.0 μm), but the depth of focus increases fourfold (241 μm) and extends
across half to the full thickness of the retina, depending on retinal eccentricity.
The fundamental trade-off between transverse resolution and depth of
focus suggests that the two can be collectively optimized for specific AO
retinal imaging applications, for example, targeting specific cell layers such as
the ganglion cell layer, or the inner or outer nuclear layer, each of which is
<75 μm thick. Note that light collection efficiency (~pupil²) and speckle size
(~1/pupil) also depend on the pupil and should be taken into account when
optimizing the system.

For spectral-domain OCT, the overlap requirement of the gate and focal
plane is substantially more relaxed than that of time-domain OCT, as individual
A-scans are acquired in a single snapshot. This is a major performance
advantage of the spectral-domain approach.