The HRF is a sensitive instrument producing a unitless index that
represents volumetric blood flow within capillary beds in the retina.
It is also sensitive to changes in the photodetector gain. Although the
beam entering the eye is held at a constant level by the HRF controls,
the user may change the sensitivity of the HRF photodetector manually.
Indeed, determining the proper photodetector setting requires a skilled
technician. Considering the results of the present study, previous
studies that did not control for variations in DC levels should be
interpreted with appropriate prudence. To accurately assess retinal
blood flow with the HRF, the sensitivity setting between longitudinal
images must remain unaltered. A target DC level of each image should be
known before acquiring image data and recorded in the subject’s
records. After imaging, the DC level should be checked. This may be
accomplished quickly and easily with a 50 × 50 pixel box
placed in the area of interest; with the sole purpose of measuring the
area’s DC level. The large box is inappropriate for flow measurements
because large vessels must be excluded for accuracy.
6 In longitudinal studies of disease, as retinal disease progresses the
optical properties of the tissue may change, and alterations in the
sensitivity setting may be necessary to maintain the target (baseline)
DC level.
A previous study of the acute test–retest reproducibility of the HRF
found a coefficient of variation of repeated measures of 6.6% for
blood flow.
5 In that study the subject sat back from the
HRF, and the headrest and all camera settings were randomized between
the two images. In the present study, only the sensitivity was changed.
The subject remained in the headrest with eyes closed between each of
the five images, and focus settings and camera alignment remained
constant. Further, in the original test–retest study, the default
100-pixel sample box was used to measure flow. In the present study,
pointwise analysis was used to analyze all well-focused pixels,
resulting in the analysis of approximately 700 pixels common to each of
the five images included in the flow measurement. The high control of
the imaging technique as well as an increased analysis area suggests
that the coefficient of variation in the present study is at the most
6.6%. It is likely that the differences in flow measurements observed
between the various flow measurements resulted from the difference in
illumination level and not physiological alterations or other HRF-based
sources of noise.
Flow is calculated from the moment of the corrected power spectrum
weighted by intensity, which is synonymous with brightness in this
context
2 :
\[Flow_{x,y}{=}\ \frac{{{\int}}\ f{\cdot}Pc_{x,y}df}{I_{x,y}}\]
where
\[f{=}\mathrm{frequency,}\]
\[Pc_{x,y}{=}\ corrected\ (for\ noise)\ power\ spectrum\]
\[I{=}\ intensity\ or\ pixel\ DC\ level.\]
An oversensitive photodetector may have two effects on the power
calculation. The first is to drive the pulsations of the interference
pattern into the saturation range of the HRF’s photodetector. The
effect of saturation on the flow calculation would be to drive the
power spectrum term Pc x,y toward 0. The
second effect is to drive the pixel intensity term, I,
toward a high value. These two effects work together in the power
equation to reduce the calculated flow value. Similarly, small
pulsations of the interference pattern will be exaggerated in
conditions of very low sensitivity, because the flow calculation
divides by the DC term, I. As I approaches 0, the
flow value will be increased.
Another potential source of sensitivity-derived errors in flow
measurements is the HRF’s use of a noise correction algorithm based on
image intensity.
2 Raw HRF measurements of Doppler shift
are altered on the basis of the assumed level of noise within the
measurement pixel. Pixels with high DC values are thought to contain a
high level of noise, and pixels with low DC values are thought to
contain less noise. Flow measurements in high DC images are therefore
reduced by a large correction factor, whereas low DC images are
corrected by only a small amount. This noise correction routine would
tend to alter flow values in the same direction as the intensity term
in the flow equation. It is possible that the combination of noise
correction and inclusion of DC level in the flow equation contributed
together to the different flow measurements observed in this study.
This DC-linked error in HRF measurements has been observed previously
in in vitro models but has not previously been demonstrated in the
human fundus. Further, in the in vitro model study, the changes in
intensity occurred in the background material and not in the area being
measured.
2
Further validation of the HRF or other instruments measuring retinal
capillary blood flow requires good models with known flow conditions.
Despite predicting photodetector sensitivity-based errors in HRF flow
measurements, there are several differences between an in vitro glass
capillary tube model and the human fundus. The HRF measures an
interference pattern created by Doppler-shifted and
non–Doppler-shifted light. It is important that the volume being
analyzed contain both moving and stationary scattering sources. In an
in vitro model, blood flowing through the glass tube features a
parabolic velocity distribution across the tube. Blood in contact with
the tube wall had a velocity of zero. Each pixel from the in vitro
model included a stationary scattering source, red blood cells at the
surface of the tube. Each pixel also contained a continuous spectrum of
velocities. The penetrating beam passed into the parabolic
distributions of velocities present in the tube. This is not the case
in the fundus. The inner diameter of a capillary is approximately equal
to the diameter of a red blood cell. Should a pixel be filled with a
capillary, there will be no stationary scatter source within that
pixel. Light will either strike a moving blood cell or plasma.
When the beam encounters plasma in the vessel, it will be scattered by
structures posterior to the capillary. Of course, and at 795 nm, the
only prominent light-scattering structure is hemoglobin, the
surrounding tissues being transparent.
8 Animal studies
have found that laser Doppler techniques are limited to surface
measurements within the fundus.
9 10 Therefore, an ideal in
vitro model should have (1) blood flowing though vessels with diameters
approaching those of capillaries so that there is no parabolic
distribution of velocities across the vessel but that all blood cells
travel with a velocity equal to the mean velocity; (2) vessels
surrounded by a material approximately transparent at 795
nm
8, similar to the optical-scattering
characteristics of human tissue, and (3) this vessel/material
combination should be mounted on top of a high flow layer that mimics
the choroid. Given the difficulty of creating such a complex and small
model, further HRF validation studies will depend on either highly
controlled, yet nonphysiological capillary tube models, or by human and
animal studies in which the actual flow conditions cannot be entirely
known.