CLINICAL SCIENCE
Evaluation of Possible Error Sources in Corneal Endothelial
Morphometry With a Semiautomated Noncontact
Specular Microscope
Michael J. Doughty, PhD
Purpose: To assess the corneal endothelium, particularly the
polymegethism feature, using the Topcon SP-3000P specular
microscope with newer center-dot software.
Methods: Forty-eight healthy, normal weight, noncontact lens
wearers of Asian ethnicity were assessed. Single endothelial images
from each subject were processed with center-dot software, reeval-
uated after correction of obvious errors, and then by manual border
marking and planimetry. Endothelial cell density based on average cell
area and the coefficient of variation (COV) of cell area (polymegeth-
ism) were calculated.
Results: Error sources are associated with erroneous location of cell
borders (usually creating larger or smaller “cells”) or failure to assign
cell borders to a marked cell. On the initial application of the center-
dot software, the endothelial cell density values ranged from 1822 to
3244 cells per square millimeter (mean, 2644 cells/mm2); this range
was reduced (eg, 1955–3054 cells/mm2; mean 2690 cells/mm2 on
editing or in manual planimetry). The COV values ranged from
17% to 39% (mean, 27.5% 6 5.5%), with one third of the endothelia
yielding COV values of greater than or equal to 30%. On editing or in
manual planimetry, the COV values were reduced to between 17%
and 29% (mean, 24.5% 6 3.2%; P , 0.001).
Conclusions: In the use of a center-dot endothelial analysis
program with cell border identification, it is likely that at least 1
set of editing steps is required to produce reasonable results.
Key Words: endothelial cell density, specular microscopy, polyme-
gethism, image analysis software
(Cornea 2013;32:1196–1203)
T he normal cornea is lined on its posterior aspect by a contin-
uous layer of endothelial cells,1,2 easily viewed at high mag-
nification by either a contact or noncontact specular microscope
and confocal microscope. The most widely used measure of the
corneal endothelium is to assess the endothelial cell density
Received for publication February 7, 2013; revision received April 4, 2013;
accepted April 5, 2013.
From the Department of Vision Sciences, Glasgow Caledonian University,
Glasgow, United Kingdom.
The author has no funding or conflicts of interest to disclose.
Reprints: Michael J. Doughty, Department of Vision Sciences, Glasgow
Caledonian University, Cowcaddens Road, Glasgow G4 0BA, United
Kingdom (e-mail: m.doughty@gcal.ac.uk).
Copyright © 2013 by Lippincott Williams & Wilkins
1196 | www.corneajrnl.com
(ECD)2–4 as the number of cells per square millimeter. This
measurement is usually undertaken by using image analysis
software incorporated into the instruments.
Other morphometric indices have been used to assess the
endothelial mosaic,5 especially assessment of polymegethism.
This is based on a calculation of the variability in the area of
the individual cells, usually presented as the coefficient of var-
iation (COV) of the cell area.5 Even in the early studies of the
corneal endothelium by specular microscopy, it was generally
considered that the most normal and healthy endothelium would
be composed of uniformly sized cells with a highly symmetrical
shape (the “hexagonal” cell concept).6 Any increase in the var-
iability in cell size in relation to age, for example, could be
considered to indicate some alterations in the functional state
of the cell layer. A low value for the COV is likely accepted
as the “norm” unless there are confounding factors, such as
corneal disease (degenerations), or various types of inflamma-
tion associated with surgery or contact lens wear, for example.1,2
It has been shown that the actual average COV value
obtained from an image can be highly dependent on the number
of cells used for the calculations; although this, in itself, should
not lead to a systematic error that results in predictably higher or
lower values.7 The repeatability of the COV estimates, however,
can be expected to get worse if lower numbers of cells are
analyzed.4,7,8 Even with evaluation of a reasonable number of
cells per endothelial image and the application of a flexible image
analysis software (which will allow for substantial editing,
including cell–cell border relocation), COV assessments may still
only be within around 65%,8,9 or perhaps even worse than this.10
This editing is important because at least some commer-
cially available image software systems are prone to error in
that much larger COV values (from what might be considered
normal) can be readily obtained. This can sometimes be the
result of just a single error in a set of cells.11,12 These analyses
were undertaken using a within-instrument software that does
not allow for cell–cell border editing, a system that is likely
a common option nowadays. With this type of software avail-
able on the Topcon SP-2000P noncontact specular microscope,
it was not possible to easily identify why the COV errors had
occurred11 because the older software only provided the user
with an output that identified which cells had been marked
(dotted) for cell area measurements. Reliance was placed on
the user examining the reported COV and cell minimum (MIN)
and maximum (MAX) area values.
From this perspective, an opportunity was taken
to reevaluate the center-dot software on the updated Topcon
Cornea  Volume 32, Number 9, September 2013
Cornea  Volume 32, Number 9, September 2013
SP-3000P, which now provides the user with an overlay of the
cell–cell borders identified after the cells have been selected with
the center-dot procedure.
METHODS
Subjects
After approval by the local ethics committee, subjects
were recruited from the university-wide student population by
word of mouth. Any Asian student aged between 18 and 30
years was considered suitable providing that they were in
general good health and had a negative history for contact
lens wear or any surgical or parasurgical procedures on the
eye. All individuals were ambulatory and presented with no
major health problems (eg, cardiovascular disease, diabetes,
thyroid disorders, obesity). The ethnic grouping of the
subjects was as communicated by them, along with a request
for information on their age and whether they were born in
the United Kingdom or in their country of ethnic identity.
Specular Microscopy
The protocol was to obtain a single image of the central
region of the corneal endothelium using a Topcon SP-3000P
noncontact specular microscope, which also provides a mea-
sure of central corneal thickness as a general indicator of
corneal health.13 The images were sent to a thermal printer
(Sony Video Graphic Printer, UP-897).
Image Analysis and Processing
As the first step, a rectangular area of the image deemed to
include well-resolved cells was selected with the “ANALYZE”
command of the menu. In normal corneas and on the use of the
default option on the instrument, this rectangular box size can be
expected to contain a little more than 100 cells. This was from
FIGURE 1. An example of a uniform
endothelium of an Indian man (A),
as found after the application of the
center-dot morphometry algorithm
(B), and with the same region of cells
identified by manual planimetry (C).

2013 Lippincott Williams & Wilkins
Semiautomated Corneal Endothelial Morphometry
the central region of the image or just slightly higher or lower
than this (see Results). The instrument software is a “center-dot”
algorithm in which a “dot” is placed in the vicinity of the middle
of a cell by a mouse “click” operation. This procedure was
repeated for a contiguous set of cells within the region of interest
(ROI) until 100 cells had been marked. The “CALC” menu step
then returns the image of the cells to the screen with an overlay
to show where the cell borders have been considered to be
present (Fig. 1B). In some cases, a few cells that had been dotted
did not result in a cell domain (border) being drawn, but the
location of the “dot” was still marked on the cell (see Results).
This outcome meant that the “N” for the cell count was less than
100. In some other images, even though all cells in a group had
been “dotted” (as reported by the “N” value before the CALC
command was applied), some borders were not drawn in. This
results in small locations within the contiguous set of fully tes-
sellated cells where a single “cell” is now made of 2 or more
actual cells and again the “N” was lower than 100. A variant on
this outcome was that a border of a cell on the periphery of the
cell set was drawn over onto an adjacent cell that had not been
dotted. In this outcome, the “N” could remain the same, but
a cell border identification was obviously in error (again result-
ing in larger and even very large “cells”).
As the primary outcome measure, the results from the first
presentation of the center-dot algorithm were recorded. This
generated data for ECD (designated as CD on the printout) and
the COV (designated as CV on the printout). These 2 primary
outcomes were therefore termed ECD1 and COV1.
For most of the images (45 of 48), at least a single error in
cell border identification was obviously found (see Results). As
far as possible, the ANALYZE command was reapplied to the
marked images (stored temporarily in the instrument software),
efforts were made to correct these errors, and the CALC
command was again used to generate a second result where
the “N” was as close to 100 as could be obtained. This generated
the second set of data designated as ECD2 and COV2. This was
managed for all of the 48 images analyzed for this report. The
www.corneajrnl.com | 1197
Doughty
machine software allows for the entire outcome to be undone by
“undotting” some or even all the cells and starting again. Even
with this being tried, 2 other images (see Results) were rejected
as being unanalyzable.
An ID was attached to the image prints, and a JPEG
image file was obtained by scanning at 400 dpi. An A3-sized
print (39 · 26 cm) was made, on which the endothelial cell
borders that were included within and deliberately slightly
extended beyond the ROI used in the center-dot analysis were
marked with a pen ( Fig. 1C). The cell border marking on the
prints was deliberately done without looking at center-dot out-
come. The cells with marked borders were numbered, and then,
their areas were measured with a digitizer pad in stream mode
(GridMaster DigiPro; Elstree Computing, London). An average
of 156 contiguous cells (range, 137–182) were measured, and
the areas were converted into absolute values using the scale bar
on the image instrument mask. This generated the third set of
data, designated as ECD3 and COV3. As with the center-dot
processed images, the ECD was calculated from the average cell
area value (ie, ECD3 = 1,000,000/AV), whereas the COV was
calculated from the SD of the cell area values (ie, COV = SD/
AV) and multiplied by 100 to give COV3. This approach
ignores whether the area measurements from each endothelium
were normally distributed or not, and no assessment of such
normality was considered simply because a data set of the areas
is not produced by the center-dot software that was used.
Statistical Analyses
The data were entered into spreadsheets in Systat v. 11
(Systat, Evanston, IL) to allow generation of global statistics
and graphical output. The data sets were checked for
normality using the Shapiro–Wilk default option in Systat,
and then, comparisons were made where appropriate, either
using paired t tests (assuming unequal variance) or a signed
rank-order Friedman analysis of variance.
RESULTS
Subject Details and Overall Outcomes From
Specular Microscopy
The 48 Asian subjects were aged between 18 and 29
years (mean 6 SD, 23 6 3 years). Most of the subjects were
born in the country identified as their ethnic grouping and had
been temporarily resident in the United Kingdom for 1 to 3
years. Overall, the subjects were from Mainland China (12),
Malaysia (12), India (12), Srilanka (3), Bangladesh (2), Thai-
land (3), South Korea (2), and Philippines (2). All corneas
appeared clear with a quality tear film and had central corneal
thickness values basically within a normal range between
0.478 and 0.585 mm (mean, 0.527 6 0.024 mm). Gross
external eye evaluations indicated no substantial abnormali-
ties of the conjunctiva or eyelid margins. Although several of
the Indian subjects did have low-grade hyperemia of the bul-
bar conjunctiva, they indicated (on specific questioning) no
major problems in terms of ocular discomfort and so on. All
of the endothelia showed an uninterrupted mosaic pattern
with no signs of inflammation (blebs, pseudoguttae, etc).
1198 | www.corneajrnl.com
Cornea  Volume 32, Number 9, September 2013
For the most part, the images were considered to be the best
(grade 1) in terms of clarity.
Outcome Measures from Morphometry
Figure 1A shows a uniform endothelial mosaic of an 18-
year-old man of Indian origin. The outcome of the center-dot
algorithm applied to the image is shown in Figure 1B and
yields an ECD value of 3042 cells per square millimeter,
a COV1 of 17.0%, and a percentage of 6-sided cells (HEX)
of 74%. Essentially, the same outcome was obtained by man-
ual planimetry of the same region of cells (Fig. 1C). It should
be noted, however, that the cell–cell borders marked by the
center-dot algorithm are not identical to those manually marked
for the planimetry, but the agreement appears reasonable.
Small Errors in Border Location and Cells Not
Identified With Borders
Figure 2A shows high density of cells (3052 cells/mm2)
of a 27-year-old Chinese man, but the reported COV1 was
higher at 26%. Evaluation of the cell borders reveals a very
unusual-shaped “cell” in the lower left and several unusual
slightly larger and elongated cells in the upper right. In the
same vicinity are 2 “dots” showing cells that were marked,
but no borders were drawn. The manual cell border marking
(data not shown) revealed the slightly elongated cells to be
errors, and COV3 was just 23%. Figure 2B shows the cells
from a Chinese woman aged 26 years. The ECD1 was just
2401 cells per square millimeter, and quite a few cells have
rather irregular outlines. The COV1 of the center-dot algo-
rithm was again slightly higher at 28%, and the cell overlay
again indicates a number of slightly irregular cells and dotted
cells without border identification. The manual outcome (data
not shown) revealed more regular shaped cells and a COV3 of
just 22.5%. Figure 2C shows the endothelium of similar den-
sity (reported as 2401 cells/mm2) of a 26-year-old Indian man
where the center-dot algorithm has clearly produced not only
a slightly larger irregular shaped cell in the middle of the ROI
but also an adjacent smaller cell and a gap. The software
reported a COV1 of 29%, but on manual planimetry (not
shown), the COV3 was just 24.5%.
The algorithm error of not identifying cell–cell borders
is sometimes easily correctable and sometimes not. Figure 3A
shows a high value for ECD1 (of 3244 cells/mm2) of a 19-
year-old Malay man. The main error is in a small group of
cells (upper right of ROI) with no borders marked and the
correction was to “undot” these cells and replace them with
cells from lower down in the image. With all cells obviously
being of similar size, the COV2 was the same as COV1 (at
21%) with no notable change in the estimated ECD. Figure
3B shows a higher ECD of 2919 cells per square millimeter of
a 21-year-old woman from Bangladesh, but there are 3 obvi-
ous errors of cell border identification with 3 larger cells
(lower left and just to the right of this and also in the upper
left of the ROI) being produced by 2 “dots” resulting in just 1
cell. The COV1 was 19% and reduced by 2.0% on the editing
(ie, COV2 was 17%) to be the same as that from manual
planimetry. Figure 3C shows slightly more cells that have

2013 Lippincott Williams & Wilkins
Cornea  Volume 32, Number 9, September 2013
FIGURE 2. Examples of the outcome
of application of the center-dot soft-
ware showing small possible errors
that are usually easily correctable. See
text for subject details.
not had their borders marked and with 1 cell (just below the
middle of the ROI) with an unusual shape in a 23-year-old
Malay man. The ECD1 was 3055 cells per square millimeter,
and the COV1 was 25%. With 2 sequential correction steps,
all cells were obtained with borders and the COV2 went down
slightly to 22% with a small increase in ECD2.
More Substantial Errors in Border Location
and Cells Not Identified With Borders
Figure 4A shows that the center-dot algorithm clearly
produces several unusual very irregularly shaped cells (see
lower left and upper right of ROI) of an Indian student aged
18 years, and in 1 place (see upper region of ROI), what appears
to be a very small cell. This was recognized in the cell MIN
value, and the overall COV1 was 31% despite a reported ECD1
of 3021 cells per square millimeter. After what were considered
to be acceptable corrections (eg, to remove the very irregular
cells), the COV2 was still 29%, whereas, planimetry yielded
a COV3 of just 24.5%. Figure 4B shows that the center-dot
algorithm missed several cells of a 20-year-old Chinese man.
FIGURE 3. Examples of the outcome
of application of the center-dot soft-
ware with differences in extent of
missed cells that take time to correct.
See text for subject details.

2013 Lippincott Williams & Wilkins
Semiautomated Corneal Endothelial Morphometry
The reported ECD1 was 2916 cells per square millimeter, and
the COV1 was 33%. In this example, the correction of the large
cells (see upper right and lower left of ROI) and replacing the
cells without borders (see right-hand side of ROI) changed the
ECD2 to 2994 cells per square millimeter and substantially
reduced COV2 to 24%. Planimetry yielded a COV3 of
22.5%. In Figure 4C, from a 23-year-old Indian man, the
ECD1 was lower at 2566 cells per square millimeter but again
shows some larger cell domains (now up to 930 mm2) where
borders were not drawn. The COV1 was 32%, which was
reduced to just 25% on editing. That the center-dot software
can produce some very peculiar cell domains with the border
marking can be seen in the upper left of the ROI.
Multiple Substantial Errors Difficult to Correct
In a few cases, the editing needed to be repeated several
times before a satisfactory outcome was obtained, and in 2
cases, the outcome of the center-dot routine was considered
uncorrectable. Figure 5A shows the endothelium of a 21-year-
old Malay man, with a reported ECD1 of 2611 cells per
www.corneajrnl.com | 1199
Doughty
FIGURE 4. Examples of the outcome
of application of the center-dot
software with errors in cell border
assignment and missed cells that are
not easily correctable. See text for
subject details.
square millimeter. The COV1 was “normal,” yet inspection of
the overlay shows 2 regions (upper left and lower left) where
there are large spaces including several cells. In this case, the
“dot” has not been recognized, and so these large domains
have not been computed either (with the cell MAX only being
670 mm2). After 3 steps of correction (undotting and redot-
ting), the resultant COV2 was only slightly altered (at 22%),
but the reported ECD2 increased to 2786 cells per square
millimeter. Figure 5B shows the outcome from applying the
center-dot algorithm to the endothelial image of a 24-year-old
woman from Srilanka. The reported ECD1 was 2486 cells per
square millimeter, and the COV1 was higher at 34%. There
are clearly several larger cell domains (with the cell MAX
being 767 mm2), and again, several dotted cells without bor-
ders. Four steps of correction increased the ECD2 to 2587
cells per square millimeter and reduced the COV to 27%.
Planimetry yielded a slightly lower COV3 of 24.5%.
Figure 5C shows the endothelium of a 20-year-old Indian
man, with reported ECD1 of 2786 cells per square millimeter
and the COV1 of 38%. Subjective evaluation of the original
image revealed no signs of polymegethism. However, an inspec-
tion of the overlay reveals multiple irregularly shaped cell
domains and some very irregular cell–cell borders. Even after
several attempts to correct the errors and then a complete
FIGURE 5. Examples of the outcome
of application of the center-dot
software showing substantial errors
that are hard to correct. See text for
subject details.
1200 | www.corneajrnl.com
Cornea  Volume 32, Number 9, September 2013
reprocessing of the image, the resultant outcome was considered
unusable despite the numerical outcomes being more reasonable.
Systematic Comparison of Outcome
Measures From Uncorrected and Corrected
Center-Dot Algorithm and Comparison to
Manual Planimetry
All the images analyzed for the present study were of
good quality and with no obvious abnormalities or overt signs
of polymegethism. Although most cells appeared to have
similar sizes, the overall outcome from the above-detailed
examples is that it is not uncommon for differences in COV to
arise from a first run of the center-dot algorithm on the SP-
3000P. For the most part, these differences are readily
correctable, providing of course time is taken to identify them.
These differences are far easier to identify by visual inspection
of a printout. The overall results are provided in Table 1.
The overall outcome for ECD assessments was that the
first run of the center-dot algorithm yielded slightly lower
values (mean, 2644 cells/mm2) compared with the corrected
version obtained by 1 or more steps of manual editing (mean,
2690 cells/mm2). A very similar value was obtained by

2013 Lippincott Williams & Wilkins
Cornea  Volume 32, Number 9, September 2013
TABLE 1. Comparison of Morphometry
Morphological First Operation Corrected
Parameter of Center-Dot Center-Dot Planimetry
ECD (mean 6 SD) 2644 6 266 2690 6 236 2662 6 229
ECD range 1822–3244 1955–3054 2022–3027
COV (mean 6 SD) 27.5 6 5.5 24.5 6 3.2 24.5 6 3.0
COV range 17–39 17–29 18–29
COV (in percent) is designated as CV on Topcon SP-3000P. ECD (in cells per
square millimeter) is designated as CD on Topcon SP-3000P.
manual morphometry (Table 1). The net difference of just 46
cells per square millimeter between ECD1 and ECD2 was
only 1.7% in relative terms but was statistically significant
(P , 0.001; Friedman). A step-wise comparison of ECD2
with ECD1 is shown in Figure 6. This Bland–Altman analysis
shows what appears to be reasonable agreement for most of
the images, but there are clearly 3 very substantial outliers. As
noted above, although the net difference was only 46 cells per
square millimeter, the SD on the difference was 78 cells per
square millimeter. Based on the differences, a 95% confi-
dence interval for ECD of 2107 to +199 cells per square
millimeter was obtained. There was a modest trend (regres-
sion analysis on the difference in relation to the mean, not
shown in the figure; P = 0.005, r = 0.395) for the larger
differences to be seen at both lower and higher cell densities.
For COV, the difference was more substantial at 3.0%
(ie, a relative difference of 12.2%), with a mean COV1 of
27.5% (range from 17% to 39%) compared with just 24.5%
(range, 17%–29% for COV2) when corrections were made.
FIGURE 6. Assessment of the net differences in ECD estimates
when comparing the edited output from the center-dot soft-
ware (ECD2) with that of the first output (ECD1). The middle
line shows the mean difference, whereas the top and bottom
horizontal lines show the 95% limits of agreement (61.96 SD).

2013 Lippincott Williams & Wilkins
Semiautomated Corneal Endothelial Morphometry
This difference was statistically significant (P , 0.001; paired
t test). The variability on the COV values (as 6SD) was also
substantially higher from the first assessment (ie, 65.5%
vs. 63.4%; P , 0.001; Friedman). That the COV values of
the uncorrected center-dot routine are often higher is supported
by the results from the planimetry where the mean COV was
just 24.5%. A Bland–Altman analysis for the differences
between COV2 and COV1 (Fig. 7) highlights not only this
substantial variability but also a very notable systematic error
in that the higher the initial COV1, the more likely it was to be
in error (regression analysis, not shown in the figure, P ,
0.001; r = 0.710). The SD on the differences was large at
3.3%, yielding a 95% confidence interval for COV
of 29.5% to +3.5%.
DISCUSSION
The general aim of the present study was to systematically
assess the reliability of the newer center-dot software on the
Topcon SP-3000P specular microscope. This revised on-machine
software provides the user with a very clear visual output to
indicate the fidelity of the cell border/cell domain identification.
This offers the user a substantial advantage over earlier versions
of this type of software and is different from older output formats
from the IMAGEnet software that can be used along with this
specular microscope (as a separate option).8,9,14
This systematic analysis of possible sources of error in
endothelial morphometry broadly confirms previous studies,
carried out on individuals of white Northern European origin.11,12
The same overall result has been found for other computer-based
programs linked to both specular microscopes8,9,15,16 and
FIGURE 7. Assessment of the net differences in coefficient of
variation, COV, estimates when comparing the edited output
from the center-dot software (COV2) with that of the first
output (COV1). Other details are as in Figure 6.
www.corneajrnl.com | 1201
Doughty
confocal microscopes.17,18 The present studies extend previous
ones in that, due to the new additions to the instrument-based
software, some progress can be made toward identifying when
these errors occur and then to be able to easily demonstrate the
result of correcting these errors.
The most substantial impact of the errors is on the
COV. As evaluated in detail elsewhere,19 the overall COV
cannot identify whether the increased variability in cell area
values is due to the presence of smaller or larger cells. How-
ever, even just 1 extremely small or large “cell” can result in
a skewed distribution (either negative or positive) for the cell
area values, which then results in an increased COV being
reported for the image analysis outcome. The present assess-
ment, of a software option on a specular microscope for cor-
neal endothelial analyses, further indicates that errors in COV
are not uncommon. A specific average value for “normal”
COV has yet to be defined. There have, however, been
numerous publications on younger adults (ie, between 20
and 30 years of age) designated as healthy “normal” individ-
uals and where specular or confocal microscopy has indicated
values of between 20% and 30% for the COV.14,18–29 These
studies cover a wide range of ethnic groupings. Although
some of these reports on those of Asian origin indicate
a slightly higher ECD, they do not indicate an unusual
COV as based on both subjective assessments of published
images and values reported for the average COV.14,22,25,28
From this perspective, the reporting of average COV
values rather higher than 30% for normal individuals of
a younger age therefore represents special exceptions or
contradiction to the accepted “norm.”30–36 The group mean
COV from these studies was 42.5% (range, 33.6%–58.3%),
and all of these studies were reported from Asian locations
(India, Malaysia, Philippines, Taiwan, Thailand, and Turkey).
It needs to be specially noted that none of these reports30–36
provide images to illustrate the substantial polymegethism
reported from the morphometry. As an overall outcome, it
is recommended that corneal endothelial morphometry stud-
ies include at least a representative example of the image
output along with the results of the application of morphom-
etry software for that particular image.
The present studies were undertaken not only to assess
the newer Topcon SP-3000P software but also to use this
assessment as a platform on which to further assess the
endothelia of a group of individuals that some have reported
to show quite remarkable endothelial polymegethism, that is,
a COV of over 40% and which should be overtly obvious on
visual inspection of the cell layer images. Of particular
interest were those from the Indian subcontinent or Eastern
Asia. Although the present study is of too small a scale to
draw definitive conclusions, it does further demonstrate that
young adult individuals of Asian origin can have what appear
to be normal appearing corneal endothelia, that is, a mosaic of
cells with moderate-to-high cell density and having a gener-
ally uniform cell size with no overt evidence of polymegeth-
ism. Despite this overall appearance, a semiautomated
software analysis reported rather high COV values in one
third of the endothelial images. This outcome is presented in
detail to show how it is not difficult to obtain outcome
measures from semiautomated specular microscopy software
1202 | www.corneajrnl.com
Cornea  Volume 32, Number 9, September 2013
that could be taken to indicate the presence of some degree of
endothelial polymegethism, that is, a COV of $30% as found
in 16 of the 48 subjects in this study in the initial image
processing with the center-dot software. On editing and with
also a comparison to the results from manual morphometry,
none of the endothelia analyzed were then considered to have
a COV above 30%.
The present study was one of convenience to take
advantage of foreign national students attending a metropol-
itan university, any of whom could be suitable subjects for
various cornea or contact lens wear–related studies. The sub-
jects were considered eligible for the study provided that they
met the criteria especially that of not being contact lens wear-
ers. Because most of the subjects evaluated in the present
study were born in the country of their ethnic origin, they
would have been exposed to the environment and lifestyle of
their particular Asian region.
In overall conclusion, the newer software option avail-
able for the Topcon SP-3000P noncontact specular micro-
scope provides the user with a very useful on-screen
indication of the fidelity of the automated cell–cell border
identification after application of a center-dot endothelial
analysis program. As illustrated in the present analyses, such
an option should greatly assist the user in identifying obvious
errors. It is likely, however, that at least 1 set of editing steps
is required to produce reasonable results.
REFERENCES
1. Doughty MJ. Comparative anatomy and physiology of the cornea and
conjunctiva. In: Martin Herranz R, Corrales RM, eds. Ocular Surface.
Boca Raton, FL: CRC Press; 2012: Chapter 3:32–78.
2. Edelhauser HF. The resiliency of the corneal endothelium to refractive
and intraocular surgery. Cornea. 2000;19:263–273.
3. Doughty MJ, Müller A, Zaman ML. Assessment of the reliability of
human corneal endothelial cell-density estimates using a noncontact
specular microscope. Cornea. 2000;19:148–158.
4. McCarey BE, Edelhauser HF, Lynn MJ. Review of corneal endothelial
specular microscopy for FDA clinical trials of refractive procedures,
surgical devices, and new intraocular drugs and solutions. Cornea.
2008;27:1–16.
5. Doughty MJ. Toward a quantitative analysis of corneal endothelial cell
morphology: a review of techniques and their application. Optom Vis Sci.
1989;66:626–642.
6. Doughty MJ. Concerning the symmetry of the ‘hexagonal’ cells of the
corneal endothelium. Exp Eye Res. 1992;55:145–154.
7. Doughty MJ, Fonn D, Trang Nguyen K. Assessment of the reliability of
calculations of the coefficient of variation for normal and polymegethous
human corneal endothelium. Optom Vis Sci. 1993;70:759–770.
8. Ding X, Huang Q, Zheng Y, et al. Measurement area and repeatability of
semiautomated assessment of corneal endothelium in the Topcon specu-
lar microscope SP-2000P and IMAGEnet system. Cornea. 2012;31:
1111–1118.
9. Doughty MJ, Aakre BM. Further analysis of assessments of the coeffi-
cient of variation of corneal endothelial cell areas from specular micro-
scopic images. Clin Exp Optom. 2008;91:438–446.
10. Abib FC, Holzchuh R, Schaefer A, et al. The endothelial sample size
analysis in corneal specular microscopy clinical examinations. Cornea.
2012;31:546–550.
11. Doughty MJ. Could the coefficient of variation (COV) of the corneal
endothelium be overestimated when a centre-dot method is used? Clin
Exp Optom. 2008;91:103–110.
12. Doughty MJ, Oblak E. A comparison of two methods for estimating
polymegethism in cell areas of the human corneal endothelium. Ophthal-
mic Physiol Opt. 2008;28:47–56.

2013 Lippincott Williams & Wilkins
Cornea  Volume 32, Number 9, September 2013
13. Doughty MJ, Zaman ML. Human corneal thickness and its impact on
intraocular pressure measures: a review and meta-analysis approach. Surv
Ophthalmol. 2000;44:367–408.
14. Cheung SW, Cho P. Endothelial cell analysis using the Topcon SP-1000
non-contact specular microscope and IMAGEnet system. Clin Exp Optom.
1998;81:1–7.
15. Siertsema JV, Landesz M, van den Brom H, et al. Automated video
image morphometry of the corneal endothelium. Doc Ophthalmol.
1993;85:35–44.
16. Szalai E, Németh G, Berta A, et al. Evaluation of the corneal endothe-
lium using noncontact and contact specular microscopy. Cornea. 2011;
30:567–570.
17. Jonuscheit S, Doughty MJ, Ramaesh K. In vivo confocal microscopy of
the corneal endothelium: comparison of three morphometry methods
after corneal transplantation. Eye (Lond). 2011;25:1130–1137.
18. Price MO, Fairchild KM, Price FW Jr. Comparison of manual and auto-
mated endothelial cell density analysis in normal eyes and DSEK eyes.
Cornea. 2013;32:567–573.
19. Doughty MJ. The ambiguous coefficient of variation: polymegethism of
the corneal endothelium and central corneal thickness. Int Contact Lens
Clin. 1990;17:240–247.
20. Deleted in proof.
21. Carlson KH, Bourne WM, McLaren JW, et al. Variations in human
corneal endothelial cell morphology and permeability to fluorescein with
age. Exp Eye Res. 1988;47:27–41.
22. Hashemian MN, Moghimi S, Fard MA, et al. Corneal endothelial cell density
and morphology in normal Iranian eyes. BMC Ophthalmol. 2006;6:9.
23. Kim KS, Park SY, Oh JS. Morphometric analysis of the corneal endo-
thelial cells in normal Koreans [in Korean]. J Korean Ophthalmol Soc.
1992;33:24–28.
24. Lee JS, Park WS, Lee SH, et al. A comparative study of corneal endo-
thelial changes induced by different durations of soft contact lens wear.
Graefes Arch Clin Exp Ophthalmol. 2001;239:1–4.

2013 Lippincott Williams & Wilkins
Semiautomated Corneal Endothelial Morphometry
25. Matsuda M, Yee RW, Edelhauser HF. Comparison of the corneal endo-
thelium in an American and a Japanese population. Arch Ophthalmol.
1985;103:68–70.
26. Tseng SH, Chen FK. Morphological and fluorophotometric analysis of
the corneal endothelium after radial keratotomy. Cornea. 1998;17:
471–475.
27. Yee RW, Matsuda M, Schultz RO, et al. Changes in the normal corneal
endothelial cellular pattern as a function of age. Curr Eye Res. 1985;4:
671–678.
28. Yunliang S, Yuqiang H, Ying-Peng L, et al. Corneal endothelial cell density
and morphology in healthy Chinese eyes. Cornea. 2007;26:130–132.
29. Nieto-Bona A, González-Mesa A, Nieto-Bona MP, et al. Short-term
effects of overnight orthokeratology on corneal cell morphology and
corneal thickness. Cornea. 2011;30:646–654.
30. Chiang CC, Lin JM, Bair PJ, et al. Effects of laser in situ keratomi-
leusis on the corneal endothelium. Kaohsiung J Med Sci. 2005;21:
272–276.
31. Mohammad-Salih PA. Corneal endothelial cell density and morphology
in normal Malay eyes. Med J Malaysia. 2011;66:300–303.
32. Padilla MD, Sibayan SA, Gonzales CS. Corneal endothelial cell density
and morphology in normal Filipino eyes. Cornea. 2004;23:129–135.
33. Rao SK, Ranjan Sen P, Fogla R, et al. Corneal endothelial cell density
and morphology in normal Indian eyes. Cornea. 2000;19:820–823.
34. Simaroj P, Kosalprapai K, Chuckpaiwong V. Effect of laser in situ
keratomileusis on the corneal endothelium. J Refract Surg. 2003;19
(suppl 2):S237–S240.
35. Sopapornamorn N, Lekskul M, Panichkul S. Corneal endothelial cell
density and morphology in Phramongkutklao Hospital. Clin Ophthalmol.
2008;2:147–151.
36. Yeniad B, Yilmaz S, Bilgin LK. Evaluation of the microstructure of
cornea by in vivo confocal microscopy in contact lens wearing and
non-contact lens wearing keratoconus patients. Cont Lens Anterior
Eye. 2010;33:167–170.
www.corneajrnl.com | 1203