The CO-stained sections were homogeneously back illuminated, and
a digital image was captured at a 1200-pixel resolution with a camera
(DCM1; Polaroid, Cambridge, MA), which was linear with optical density,
with an R = 0.99. The density of the COR was scaled 0
to 255, where 0 = opacity and 255 = the incident light.
Measurements were taken before any filtering or contrast enhancement.
COR = I − T, where I was the incident light (nominally, a
value of 255), and T was the light transmitted through the tissue
containing the CO-reaction product. The value of COR from a blob
connected with the glaucomatous experimental eye
(CORG) was always compared in the same section
with the value of COR measured in the directly adjacent blob connected
with the normal companion eye (CORN). The ratio,
CORG/CORN, constituted the
primary data from which mean values and variances were calculated. The
mean COR and SD of a closed contour, drawn by eye to best outline the
blob, were recorded for each blob and compared with the values obtained
in the companion blob with primary input from the opposite eye. Thus,
the ratio of CORG/CORN was
formed. A minimum of 10 such ratios was then averaged for each tissue
location. Most often, the ratio
CORG/CORN was expressed as
a percentage reduction of COR, relative to that in the companion site
that had input from the normal eye.
For the relative size measurement, the numbers of pixels contained
within the closed contour of the tracing of the CO blob image was
compared. It is recognized that there is no generally accepted
objective criterion for drawing the CO blob boundaries, because there
is a characteristic decreasing COR gradient from the blob center toward
the edge.
14 Therefore, we present a statistical comparison
of the subjective sizes of pairs of CO blobs, collected by a technician
naive to the purposes of the study, and the results evaluated by the
paired
t-test.
The brain section image was processed using the University of Texas at
San Antonio Image Tool 2.0 image analysis software. With the
reticulated CO pattern characteristic of layer 4A as a reference, the
perimeters of the CO blobs in layer 3 were traced by eye and the
numbers of pixels within the enclosed perimeter taken as an index of
the relative area of blobs with input from the normal and the
glaucomatous eye. The average area and relative COR were measured for
pairs of CO blobs, one from the normal left (ODC), paired with the
adjacent CO blob from the ODC of the glaucomatous right eye. A minimum
of 10 pairs was measured from the V1 cortices of each of five of the
experimental monkeys. The primary data were imported into a spreadsheet
(QuattroPro; Corel, Ottawa, Canada) for computation and graphing,
whereas statistical comparisons were made using the paired t-test (SigmaStat; Jandel Scientific, San Rafael, CA).
To describe the distribution of relative COR within the blob, pairs of
normal and companion glaucomatous blobs were scanned along a line
orthogonal to the course of the eye dominance columns. Smoothing the
resulting noisy curves was achieved by subjecting the data to a 3-point
rolling average to generate a 48-point profile for 10 pairs of blobs.
To present a different and enhanced view of the density, size, and
distribution of COR within the blobs, the following operations were
performed on selected blob fields. First, the density range of the
image was inverted so that the low-intensity blob COR density signal,
represented by the dark COR blobs, became high-intensity values. Next,
the pixel values in the holes representing the blood vessels were
replaced by the average value of the immediately surrounding pixels. A
surface plot was then made so that the pixel intensity was coded both
by height of the surface and by color. An azimuth and elevation was
chosen to optimize inspection of differences between rows of blobs. In
practice, low-pass filtering was often performed to accentuate the
major features over high-frequency variations.
To fill and replace the blood vessel holes, two different protocols
were used with approximately equal success. In the first protocol, the
holes were selected in the image management software (Photoshop; Adobe,
San Jose, CA) by selecting that range of intensities. A mask was
created corresponding to the holes. The remainder of the image without
the holes was then blurred by several passes of a median filter. The
calculation facility of the software program was then used to replace
the masked area corresponding to the holes with corresponding portions
of the blurred image. This replaced the white holes with a local
average of the surrounding area. In the second protocol, a custom“
zippering” routine was implemented in a statistical analysis
software program (The MatLab; MathWorks, Natick, MA). By this means, an
upper threshold was imposed that distinguished the high-intensity holes
from the darker tissue. The routine then replaced pixels exceeding
this threshold with the value of their nearest neighbor not exceeding
the threshold. By this mechanism, the surrounding regions “flooded”
the holes with their values. This procedure was more robust in dealing
with a wide variety of tissues but sometimes replaced the holes with
values somewhat lighter than the near background, probably because of
light scattering near the holes.