View
216
Download
0
Category
Preview:
Citation preview
ww.sciencedirect.com
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x
Available online at w
journal homepage: www.elsevier .com/locate/survophthal
Diagnostic and surgical techniques
Quantitative analysis of in vivo confocal microscopy images:A review
Dipika V. Patel, PhD, MRCOphth*, Charles N. McGhee, PhD, FRCOphth
Department of Ophthalmology, New Zealand National Eye Centre, Faculty of Medical and Health Sciences, University of Auckland, Auckland,
New Zealand
a r t i c l e i n f o
Article history:
Received 2 August 2012
Received in revised form
9 December 2012
Accepted 11 December 2012
Neelakshi Bhagat and David Chu,
Editors
Keywords:
In vivo confocal microscopy
cornea
quantitative analysis
repeatability
corneal endothelium
corneal epithelium
keratocytes
sub-basal nerves
corneal thickness
* Corresponding author: Dipika V Patel, PhMedical and Health Sciences, University of A
E-mail address: dipika.patel@auckland.ac0039-6257/$ e see front matter ª 2013 Elsevhttp://dx.doi.org/10.1016/j.survophthal.2012.
a b s t r a c t
In vivo confocal microscopy (IVCM) is a non-invasive method of examining the living
human cornea. The recent trend towards quantitative studies using IVCM has led to the
development of a variety of methods for quantifying image parameters. When selecting
IVCM images for quantitative analysis, it is important to be consistent regarding the
location, depth, and quality of images. All images should be de-identified, randomized, and
calibrated prior to analysis. Numerous image analysis software are available, each with
their own advantages and disadvantages.
Criteria for analyzing corneal epithelium, sub-basal nerves, keratocytes, endothelium,
and immune/inflammatory cells have been developed, although there is inconsistency
among research groups regarding parameter definition. The quantification of stromal nerve
parameters, however, remains a challenge. Most studies report lower inter-observer
repeatability compared with intra-observer repeatability, and observer experience is
known to be an important factor. Standardization of IVCM image analysis through the use of
a reading centerwould be crucial for any future large,multi-centre clinical trials using IVCM.
ª 2013 Elsevier Inc. All rights reserved.
In vivo confocal microscopy (IVCM) is a noninvasive method
quantification is crucial for objectively assessing the effects ofof examining the living human cornea under high magnifi-
cation in healthy and pathological states. These attributes
make it a powerful clinical and research tool.45 Themajority of
early studies using this technique have been qualitative in
nature. For example, the diagnosis of infectious keratitis
typically requires qualitative analysis of images by an expe-
rienced observer, and no quantitative studies are currently
available. The recent trend towards quantitative studies using
IVCM has led to the development of a variety of methods for
quantifying image parameters. As well as establishing
the normal range of cell densities in healthy corneas,
D, MRCOphth, Departmenuckland, Private Bag 920.nz (D. V. Patel).ier Inc. All rights reserve12.003
pathology or therapeutic interventions on these parameters.
We highlight IVCM image parameters that may be quan-
tified, including discussion of analysis techniques, limita-
tions, and repeatability.
1. Image selection and analysis
When selecting IVCM images for quantitative analysis, it is
important to be consistent regarding the location (central vs
peripheral cornea) and depth of images. Consistency in
t of Ophthalmology, New Zealand National Eye Centre, Faculty of19, Auckland,New Zealand. 1142.
d.
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x2
corneal location may be maximized by the use of fixation
targets–internal for slit-scanning IVCM17 or external for laser
scanning IVCM.47 Accuracy in determining section depth can
be maximized by using fixed landmarks (e.g. measuring ker-
atocyte density immediately posterior to Bowman’s layer) or
by using devices such as the “z-ring encoder” (see the Corneal
Thickness section).
Currently, there is no consensus regarding the minimum
number of images required for representative quantitative
analysis, although a single image is generally considered
insufficient. Themajority of published studies have used up to
five images per layer per eye.
The quality of the selected images is key. Obviously,
blurred or non-tangential images should be excluded
(Fig. 1),34 and once image selection has been completed, all
images should be de-identified and randomized by an
independent investigator prior to analysis to avoid
observer bias.
All IVCM images must be appropriately calibrated, but
the parameters will vary depending on the type of micro-
scope used. The standardization and choice of frame size
are important factors given the differing contrast distribu-
tion across images from different types of IVCM. In partic-
ular, slit scanning IVCM images exhibit decreased contrast
towards the lateral edges of the image, and thus fewer of the
measured structures may be visualized in these regions.43,48
In such cases, therefore, restricting the analysis frame to the
central, higher contrast region of the image may be appro-
priate. When analyzing cell densities, to standardize meth-
odology the majority of published studies exclude all cells
that overlap two predefined borders of the selected
frame.19,26,32
2. Image analysis software
2.1. Proprietary software
2.1.1. Nidek Advanced Vision Information System softwareNidek Advanced Vision Information System (NAVIS) Endo-
thelial Analysis Software (Fig. 2), available for use with the
Fig. 1 e Laser scanning in vivo confocal microscopy images sho
involuntary movements of the patient’s eye at the time of imagi
stroma, sub-basal nerve plexus, and basal epithelium.
ConfoScan IVCM software (Nidek Technologies, Fremont, CA)
enables quantitative analysis of in vivo confocal images. The
region of interest is easily defined in terms of area, the
dimensions of which may be adjusted as required prior to
analysis. For corneal endothelial images, analysis may be
performed manually, automatically, or using a combination
of both techniques. Manual analysis only provides data
regarding endothelial cell density, whereas automatic anal-
ysis of endothelial images has the advantage of providing data
regarding endothelial density (and the normal range for the
subject’s age), the mean cell area, the coefficient of variation
in area, the mean number of sides, the coefficient of variation
in the number of sides, and the percentage of hexagonal cells.
In some cases, however, cell bordersmay be incorrectly traced
by the automated system. This can be rectified by manual
adjustment of cell border tracings. Lengths and areas of
objects larger than 1 mm may also be measured using this
software.
2.1.2. Rostock Corneal Module proprietary softwareThe Heidelberg Retina Tomograph II Rostock Corneal
Module (RCM; Heidelberg Engineering, GmBH, Germany) has
proprietary software for manual analysis of cell densities
(Fig. 3). Although the region of interest is easily defined, the
area of this region is not displayed until after selection,
making it difficult to select a region with a fixed area.
Additionally, there is no facility for automated image anal-
ysis or for making linear measurements.
2.2. Other commercially available software for IVCMimage analysis
A wide range of software is available for quantitative anal-
ysis of biological images. Commonly used software in IVCM
studies include: Image J (National Institutes of Health,
Bethesda, MD),25,55 a free public domain open source soft-
ware; Adobe Photoshop (Adobe Systems Inc, San Jose,
CA)28,36; AnalySIS (Soft Imaging System GmBH, Munster,
Germany)37,46; and AMIRA (Visage Imaging GmbH, Berlin,
Germany).62
wing (A) blurring and distortion of keratocyte nuclei due to
ng, and (B) an oblique optical section of the anterior corneal
Fig. 2 e User interface for Nidek Advanced Vision Information System software (NAVIS).
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x 3
3. Corneal epithelium
The corneal epithelium typically consists of five to seven
layers of cells, including superficial epithelial cells, wing cells,
and basal epithelial cells. Superficial cells have most
commonly been quantified in terms of cell diameter (mm) and
cell density (cells/mm2); images obtained by IVCMmay not be
sufficient to demonstrate full detail, however.39 Mocan et al39
have objectively shown that topical fluorescein application
prior to IVCM enhances the visualization of the superficial
epithelium, thus enablingmore accurate quantification of this
layer. It is therefore important to ensure consistency
regarding the use of fluorescein prior to IVCM imaging.
Wing cells may be quantified in terms of cell diameter (mm)
and cell density (cells/mm2). Whereas the superficial cell layer
is one to two cells thick and the basal cells form a monolayer,
thewing cell layer is three to six cells thick and has a lower cell
Fig. 3 e User interface for Rostock Corneal Module proprieta
density than that of the basal epithelium.7 It is therefore
important to distinguish wing cell layer images from the
similar basal cell layer.
Basal epithelial cells are the most easily and reproducibly
imaged and quantified of the epithelial layers as the result of
their location immediately anterior to Bowman layer and their
monolayer configuration. The majority of published studies
performed computer assisted manual analysis of cell diam-
eter (mm) and cell density (cells/mm2).1,3,7,44 Harrison et al12
suggested using the NAVIS automated endothelial analysis
software to analyze ConfoScan images of the basal epithe-
lium. This technique involves first inverting the image, then
performing automated analysis. Adjustment of the image
brightness and contrast is often required to aid the detection
of cell borders. The software is usually only able to accurately
identify 50% of basal epithelial cells, so extensive manual
adjustment is necessary. Unlike corneal endothelial cells,
basal epithelial cells have a more rounded appearance;
ry software for manual analysis of endothelial density.
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x4
therefore, although the software will automatically count the
number of sides for each cell, this feature is not accurate or
reliable for basal epithelial cells.12 The intra-examiner
repeatability of computer-assisted manual analysis of basal
cell density of RCM images is good, with an intra-class corre-
lation coefficient (ICC) of 0.89.3
4. Corneal sub-basal nerves
Corneal sub-basal nerves are easily and reproducibly imaged
because of their location and orientation on Bowman’s layer.
Nonetheless, the manner in which nerve density is defined
has been somewhat inconsistent in the literature. The
majority of studies have defined sub-basal nerve density as
the total length of nerves visible within a defined area (mm/
mm2 or mm/mm2),44,48 but some investigators have only
included nerve branches longer than 50 mm in their
measurements.9 Others have analyzed the total number of
nerves within a frame; the definitions for this vary, however,
from the number of long nerve fibre bundles57 to the sum of
nerve branches present.42 These differences in analysis make
reliable comparisons between studies difficult.
Efron et al8 reported high intra- and interexaminer
repeatability in nerve fibre length (NFL) measurement from
ConfoScan images in eyes with diabetes mellitus; the stan-
dard deviation was large (up to 2.33 mm/mm2), however.
Interestingly, the overall spread of data was smaller for the
observer who was more experienced at analyzing nerve fiber
images and for the second measurement. The reason why
both observers tended to assign a lower NFL value on the
second measurement than the first occasion is unclear, but
may indicate a shift in criteria toward a more conservative
approach when identifying nerve fibers.8
Other studies using slit-scanning IVCM images have also
shown good intra- and interoperator reproducibility in NFL
measures (intraclass correlation [ICC] ¼ 0.96, and ICC ¼ 0.94,
respectively).35 Grupcheva et al11 reported 93% intra-observer
repeatability and an inter-observer variation within 12% for
nerve density measurement. The latter is of similar magni-
tude to the variation (10%) noted by Benitez del Castillo et al.1
Investigations using the RCM have demonstrated excellent
intra-observer reliability (ICC ¼ 0.86) and very good inter-
observer reliability (0.74) for corneal nerve fibre density for
single images. For a set of 40 images, however, although intra-
observer reliability was excellent (ICC ¼ 0.83), interobserver
reliability was only moderate to very good (ICC ¼ 0.60). For the
more clinically relevant analysis of repeated patient exami-
nations (“study-level reproducibility”) intra-observer reli-
ability (ICC ¼ 0.57) and inter-observer reliability (ICC ¼ 0.61)
were only moderate.14 Intra-observer ICC for the number of
sub-basal nerves may be high (0.9) for RCM images.3,5
Semi-automated analysis of sub-basal nerves is both labor
intensive and subjective, requiring the investigator to manu-
ally trace all visible nerves. Scarpa et al59 made the first step in
addressing this issue with the development of a fully auto-
mated algorithm for analyzing sub-basal nerve length. The
average percentage of correctly recognized nerves length,
with respect to completely manually traced lengths of visible
nerves, was 80.4% in normal corneas and 83.8% on abnormal
corneas. The average rate of false nerve length recognition
using the automated software was 6.5% in normal corneas
and 9.1% in abnormal corneas. Other researchers have since
developed automated methods for detecting nerves in IVCM
images.
Prior to quantitative analysis, preprocessing of the raw
IVCM images is usually required to distinguish nerves from
background data, and this may be achieved using segmenta-
tion and skeletonization algorithms.18,69 Dabbah et al4 re-
ported a coefficient of variation for manual analysis of RCM
images of 0.34 compared with 0.29 for automated analysis
using a dual-model detection algorithm. Ferreira et al10 used
phase symmetry techniques in their algorithm to aid the
identification the linear structure of corneal nerves while
rejecting the regular oval shape of stromal keratocytes nuclei
appearing in slit-scanning IVCM images of the sub-basal nerve
plexus. Their method segmented corneal nerves with a
sensitivity near 90% and an average of 5.3% false recognitions.
Currently, there is no commercially available software for
automated analysis of IVCM images.
Formeasurements of the diameter of thin, highly reflective
structures (such as sub-basal nerves) or beading frequency to
be comparable, all images need to be acquired using a fixed
illumination intensity because illumination intensity affects
the apparent thickness of corneal nervesdparticularly as they
approach the limit of resolution. It follows that comparisons
of the dimensions of reflective objects in images obtained by
in vivo confocalmicroscopy are only validwhen the same type
of in vivo confocal microscope is used and illumination
intensity is constant.48 Interobserver variation in measuring
beading frequency has been reported to be 14%.1
Kallinikosetal22were thefirst todescribeanobjective, semi-
automated technique for quantifying sub-basal nerve tortu-
osity (tortuosity coefficient); the degree of agreement with
subjective grading had not been assessed, however. Although
intra- and interobserver repeatability using this technique is
very good for single images or image sets,14 intra- and inter-
observer reliability for analysis of repeated patient examina-
tions (“study-level reproducibility”) is poor (ICC¼ 0.23 and0.29,
respectively).
Scarpa et al60 modified their algorithm for the automatic
recognition of corneal nerve structures59 to enable automated
tracing of nerves and also devised a new method for auto-
mated analysis of nerve tortuosity. They evaluated the reli-
ability of their new system by comparing it to subjective
tortuosity grading by an experienced observer and showed
that their method achieved the lowest number of incorrect
classifications and the best concordance coefficient and class
separation.
5. Corneal stroma
5.1. Stromal nerves
Quantitative analysis of stromal nerves imaged by IVCM
remains controversial. A wide range of values for stromal
nerve diameter have been reported, and this variation is due
to a number of factors. First, stromal nerves commonly
traverse obliquely relative to the en face section of IVCM
Fig. 5 e Laser scanning in vivo confocal microscopy image
of the corneal mid-stroma. Although nuclei with the
highest contrast and sharpest edges can be consistently
identified, those with lower contrast and blurred edges
(arrowhead ) will be inconsistently identified by different
observers.
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x 5
images. These nerves will therefore appear shorter than one
whose path is parallel to the plane of the image (Fig. 4). The
visible nerve length per frame area will also depend on the
axial resolution of themicroscope used (i.e., a greater length of
nerve will be visible with a thicker optical section). Therefore,
measuring the length of nerve per frame does not necessarily
relate to the true stromal nerve density.48 The quantification
of stromal nerve parameters remains a challenge.
5.2. Keratocytes
Each two-dimensional IVCM image frame actually repre-
sents a volume with z-depth equal to the optical section
thickness.51 Keratocyte density has therefore been expressed
as density per unit area or per unit volume. In order to
calculate the volumetric density of keratocytes, stereological
principles must be applied that take into account the optical
section thickness and the mean size of cells normal to the
counting plane.53 The depth of field must therefore be
known when estimating volumetric cell density and,
because of variations in design, the depth of field must be
measured on the particular microscope used to record the
images of the cells.32
Classification of the stromal depth is also an important
aspect of keratocyte density analysis. The anterior stroma is
often defined as the first image posterior to Bowman’s layer,
and the posterior stroma as the first layer anterior to Desce-
met’s membrane.26 Further subdivision of the stroma varies
between studies and may be described in terms of corneal
depth or as a percentage of total corneal thickness.30,34
Manual analysis of keratocyte density involves marking
each clearly defined cell or nucleus in a predefined rectangular
frame.53 Although nuclei with the highest contrast and
sharpest edges can be consistently identified, those with
lower contrast and blurred edges will be inconsistently
Fig. 4 e Laser scanning in vivo confocal microscopy image
of a corneal stromal nerve leaving the optical section
(arrows) due to its oblique orientation.
identified (Fig. 5).34 This method is therefore both time
consuming and subjective and is hindered by high intra- and
interobserver variation.30 Estimates of human keratocyte
density in healthy corneas vary over a wide range, from as low
as 17,384 cells/mm3 in the posterior stroma38 to as high as
39,442 cells/mm3 in the anterior stroma.52
One study using tandem scanning IVCM highlighted that
mean cell densities assessed manually three times in the
same frames differed from each other by as much as 3,160
cells/mm3.34 Similarly, manual analysis of ConfoScan IVCM
images compared to histological images of donor human
corneas showed a difference of 2,377 cells/mm3, with 95%
confidence of �10,742 to 5,989 cells/mm3.53
Multiple studies have reported the repeatability of manual
analysis of keratocyte density from RCM images. One showed
mean a difference between repeated measurements of kera-
tocyte cell density of 4,300 cells/mm3 (95% limits of
agreement � 20,900 cells/mm3),64 whereas another demon-
strated intra- and interexaminer coefficients of variation of
5.2% and 2.9%, respectively.26 The ICC for anterior stromal
keratocyte density has been reported to be excellent (0.9).3
Automated analysis of keratocyte density was first intro-
duced by Prydal et al.54 Patel et al51 subsequently used
a custom program integrated into an image analysis program
to quantify keratocyte density automatically from tandem
scanning IVCM images of rabbit corneas. The coefficients of
variation of repeated estimates of keratocyte density using
this method ranged from 0.032 to 0.075. The mean difference
between cell density estimated by IVCM and histology was
1,184 cells/mm3, and the 95% limits of agreement were �9,329
cells/mm3 to 11,699 cells/mm3. Interestingly, agreement was
better for the anterior than for the posterior stroma.
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x6
McLaren et al34 further developed and refined the auto-
mated algorithm described by Patel et al51 and tested it on
tandem scanning IVCM images. Cell densities determined by
the automated method were compared with densities deter-
mined by manually counting keratocyte nuclei in the same
images, in a population of patients who underwent laser in
situ keratomileusis. The mean difference was �363 � 5214
cells/mm3. The same group also developed a program that
used image-processing routines to identify stromal cell nuclei
in ConfoScan 4 IVCM images.30
The presence of stromal nerves complicates computer
analysis, making measurements unreliable.54 Additionally,
image-processing programs developed for a particular
microscope cannot be directly applied to images from other
microscopes because the optical properties of each micro-
scope uniquely affect the cell selection criteria of the
program.30 Despite the development of automated methods
for keratocyte density analysis, none have yet been used
consistently in clinical studies.
6. Corneal endothelium
Manual analysis of endothelial cell density simply involves
counting the number of cells within a predefined frame. This
traditional form of analysis only provides information
regarding the cell density and number of cells counted and
does not supply any morphometric data. In contrast,
contemporary software applications such as the NAVIS soft-
ware of the ConfoScan IVCM (see the Image Analysis Software
section) provide data regarding the number of cells counted,
cell density, age-matched normal range of endothelial
density, mean cell area, coefficient of variation in cell area,
mean number of sides per cell, coefficient of variation in
number of sides per cell, and percentage of hexagonal cells.
A fundamental issue with automated endothelial analysis
is a failure to identify correctly endothelial cell borders,23,24
leading to overestimation of endothelial density compared
with manual analysis.20 The accuracy of cell border delinea-
tionmay be checked by the operator, and any inaccuracies can
be corrected by manual adjustment of cell borders (semi-
automated image analysis). Possible adjustments include
fusing cells (changing cell vertices), dividing cells, and erasing
cells. Kitzman et al23 have demonstrated that after manual
correction of cell borders detected by the NAVIS automated
software, the endothelial density, coefficient of variation of
cell area, and percentage of hexagonal cells were not different
from those determined by the Corners method.23,50 The
authors therefore concluded that when the ConfoScan and its
proprietary program are used to determine cell density,
investigators must manually correct cell boundaries on the
images. Doughty et al6 reported that at least 75 cells should be
counted per image for an acceptable level of inter-subject
variance.
Although semi-automated analysis improves the accuracy
of quantification of endothelial parameters,23,61 it is time-
consuming and therefore usually impractical in the clinical
setting.
Whenmanual analysis using the RCMproprietary software
was compared with automated analysis of non-contact
specular microscopy images, endothelial density differed
significantly. The RCM method underestimated endothelial
density in eyes with low cell density and overestimated
endothelial density in eyes with high cell density, although
intra and inter-observer reproducibility was good.58 Conflict-
ing data from Rieth et al56 showed no significant differences
between the two groups. Comparisons between the RCM
method and the semi-automated NAVIS method in healthy
eyes showed RCM endothelial density was significantly higher
and the degree of overestimation increased with higher
densities.63
In eyes that have undergone corneal transplantation (full
thickness or deep anterior lamellar), NAVIS software yields
a modest to substantial overestimation of endothelial
density.21 The main reason is over-segmenting of endothelial
cells by the software, commonly due to the presence of bright
endothelial nuclei (Fig. 6).49 The NAVIS manual counting
method provides lower endothelial density estimates than
planimetry (a technique that involves printing of the endo-
thelial image file following careful visual inspection and
manual outlining of the cellecell borders within the region of
interest with a pen) resulting in an underestimation of cell
density.21 Therefore, in corneal transplantation the semi-
automated NAVIS method is probably the best for
quantification.
An alternative, although perhaps less convenient, method
of quantitative endothelial assessment involves importing
IVCM images into the KSS-400 image analysis program
(Konan, Inc, Torrance, CA). Image rescaling and appropriate
calibration are required prior to analysis. Four endothelial
analysis techniques are described: the center method, the
flex-center method, the Corners method, and the variable
frame method. Patel et al50 showed good agreement in endo-
thelial density between the first three methods, although
agreement in morphometric data was poor.
7. Corneal reflectivity
Measurement of corneal reflectivity is an objective method of
assessing corneal haze.40 Most investigators have expressed
corneal haze in terms of the specific units of image intensity
from the instrument used for measurement; however, the
image “brightness” of light back-scattered from corneal haze
can only be compared with brightnessmeasured at a different
time in longitudinal studies or across laboratories if the IVCM
instruments are standardized so that units that express haze
intensity are equivalent. Unfortunately, brightness expressed
as a digitized video signal is only useful for relative
measurements from the same camera and varies depending
on its settings. Measurements can be standardized in two
ways. First, image intensity should be expressed in terms that
are meaningful and can be compared across laboratories.
Second, measurements of haze must be adjusted for differ-
ences or variations in the brightness of the light source and
the sensitivity of the light detectordvariables that may
change over time.31
The basal cell layer (BCL) index was developed by Mor-
ishige et al41 for the quantitative evaluation of corneal
epithelial edema. ConfoScan images (obtained with a fixed
Fig. 6 e In the presence of bright endothelial nuclei, automated endothelial analysis using NAVIS software usually
overestimates endothelial density due to oversegmentation of cells. Subsequent manual adjustment of cell borders gives
greater accuracy.
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x 7
light intensity of 60% of maximum) were exported into their
own program, which works by automatically defining a 100-
pixel diameter circle in the center of the image, measuring
the 8-bit pixel intensity inside the circle, and summing all the
pixel intensities. The BCL index is defined as the total pixel
intensity divided by 1,000. The intra-observer repeatability for
this method was reported to be good with a coefficient of
variation (CoV) of 3.4%. Mocan et al39 evaluated epithelial
reflectivity using the Z-scan mode of the ConfoScan. In this
method, the intensity levels of the brightest epithelial layers
were recorded in terms of intensity units with a range of
0 to 255.
When measurements are standardized with a standard
turbidity suspension, corneal back-scatter measurements
(expressed in scatter units) of ConfoScan IVCM images had
intra-observer CoV of 4.8e6.8% and there was no difference
between intra- and inter-session repeatability.15 Good inter-
session repeatability (CoV ¼ 0.4e2.9%) is present for solid as
well as for liquid reference standards.16 The brightness of the
illuminating light source in the RCM is usually varied auto-
matically by the instrument in order to maximize image
quality. Measurement of reflectivity using these images is
therefore inappropriate.
8. Corneal thickness
The slit-scanning IVCM has poor repeatability for corneal
thickness measurements, exhibiting the widest 95% limits
of agreement both within and between sessions when
compared with ultrasound, optical coherence tomography,
and Orbscan.67
This is because the position of the cornea relative to the
objective lens varies throughout the scan acquisition, and the
error is compounded by the several second length of scanning
time during which involuntary axial motion of the eye is
inevitable.33
The relatively recently developed “z-ring” for the Con-
foScan may significantly improve the accuracy of thickness
measurements.32 The z-ring remains in contact with the
surface of the cornea during scanning to stabilize the eye and
allows the objective to move with the cornea in the ante-
rioreposterior direction. Using the graph of intensity of the
confocal images, corneal thickness is determined using the
distance between the brightest image of the surface epithe-
lium and the endothelium. Using polymethyl methacrylate
contact lenses of known thickness, measurements using the
ConfoScan 4 with the z-ring were accurate to within 5 mm.2,33
When imaging human corneas, the intra-instrument repro-
ducibility of corneal thickness measurements using this
technique was excellent (ICC ¼ 0.989). There was also no
significant difference in corneal thickness measurements
between scans taken by two different operators
(ICC ¼ 0.896).2 This method, however, tends to underestimate
corneal thickness when compared with ultrasound pachy-
metry by a mean of 24.82 mm to 38 mm.33 Limits of agreement
were also large, extending from 7.2 to 69 mm.33 Also of
concern is that the standard deviation of differences between
consecutive measurements by IVCM was almost four times
greater than that of consecutive measurements by ultra-
sound pachymetry.
The plane of focus in the RCM is altered by movement of
the objective lens relative to the applanating cap. The focal
plane position (in mm) is automatically displayed on the
computer screen and recorded for each image saved. Corneal
thickness may therefore be obtained from a full thickness
scan of the cornea. Salvetat et al58 has shown that RCM tends
to overestimate corneal thickness comparedwith non-contact
specular microscopy by a mean of 6.5 � 17 mm. The 95% limits
of agreement between instruments ranged between �25.6 mm
and 38.7 mm.58
Examination with the RCM involves corneal contact and
compression, and this inevitably induces artifacts associated
with corneal flattening that may give rise to excessive thick-
ness measurements. Moreover, the accuracy of using this
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x8
technique for corneal pachymetry is limited by ante-
rioreposterior eye movements during the scan. Intra- and
inter-examiner reproducibility has been reported to be
good, although lower than that for non-contact specular
microscopy.58
Currently a number of techniques such as ultrasound, slit-
scanning tomography, and scheimpflug tomography have
been widely accepted as accurate methods to assess corneal
thickness. Thus the current limitations of thickness
measurement inherent to IVCM mean that these other tech-
niques are much more widely used for pachymetry in clinical
practice.
9. Immune/inflammatory cells
Zhivov et al68 were first to report in vivo evaluation of Lang-
erhans cells within the human corneal epithelium. Quantifi-
cation of Langerhans cell density is achieved by counting the
number of these cells per image frame.29,66,68 Keratic precip-
itates have also been quantitatively evaluated in Fuchs het-
erochromic cyclitis. Characteristics such as density, diameter,
area, the ratio between the total size and the body size, and
number of pseudopodia are analyzed.13,27
10. Reading centers
The inter-observer variability of many of the quantitative
parameters discussed in this review raises the question
whether there should be reading centers for IVCM image
analysis. A reading center is a central facility specializing in
the standard evaluation of images. Reading centers are
commonly used in multicenter clinical trials. Usually, each
study center requires certification by the reading center
and must use the center’s imaging protocol.65 In the case of
IVCM, a minimum level of expertise in imaging at each site
and a minimum standard of image quality would therefore
be important factors. Digital images would need to be
transmitted electronically. Therefore, all patient identifying
information must be removed to protect patient confidenti-
ality. Additionally, images must not be compressed as this
would degrade image quality. Transmission via the Internet
requires robust security and fast upload speeds. To our
knowledge, the University Hospitals Eye Institute Reading
Center (Cleveland, OH) is currently the only reading center
that provides standardized analysis of corneal endothelial
cell density and morphology and more recently qualitative
and quantitative assessment of the corneal epithelium and
stroma from IVCM images.
11. Conclusion
Quantitative analysis is an increasingly common feature of
studies using IVCM. It is clear that appropriate image selection
and randomization are crucial prior to analysis. There are
several software applications available for quantifying IVCM
images, each with specific advantages and disadvantages.
There are also multiple ways of defining morphological
parameters and there is currently no consensus regarding
“gold standard” definitions of parameters such as sub-basal
nerve density, making comparisons between different
studies difficult. Standardization of IVCM image analysis
through the development of reading centers will be crucial for
any future large multicenter clinical trials.
12. Method of literature search
Searches were performed using PubMed and Medline, and all
years were searched. The following search terms were used:
In vivo confocal microscop, AND cornea, corneal endothelium,
corneal epithelium, keratocyte, sub-basal nerves, corneal haze,
corneal reflectivity, corneal thickness, repeatability, and
reproducibility.
All articles judged to be of relevance to quantitative anal-
ysis of IVCM images were included, and case reports were
excluded. English articles and non-English language articles
with English abstracts were included and relevant non-
English language papers were translated. Pertinent articles
referenced by retrieved articles were also included.
13. Disclosure
The authors reported no proprietary or commercial interest in
any product mentioned or concept discussed in this article.
r e f e r e n c e s
1. Benitez del Castillo JM, Wasfy MAS, Fernandez C, et al. Anin vivo confocal masked study on corneal epithelium andsubbasal nerves in patients with dry eye. Invest OphthalmolVis Sci. 2004;45:3030e5
2. Brugin E, Ghirlando A, Gambato C, et al. Central cornealthickness: Z-ring corneal confocal microscopy versusultrasound pachymetry. Cornea. 2007;26:303e7
3. Ceresara G, Fogagnolo P, De Cilla S, et al. Corneal involvementin crohn’s disease: an in vivo confocal microscopy study.Cornea. 2011;30:136e42
4. Dabbah MA, Graham J, Petropoulos I, et al. Dual-modelautomatic detection of nerve-fibres in corneal confocalmicroscopy images. Med Image Comput Comput Assist IntervInt. 2010;13(Pt 1):300e7
5. De Cilla S, Ranno S, Carini E, et al. Corneal subbasal nerveschanges in patients with diabetic retinopathy: an in vivoconfocal study. Invest Ophthalmol Vis Sci. 2009;50:5155e8
6. Doughty MJ, Muller A, Zaman ML. Assessment of thereliability of human corneal endothelial cell-densityestimates using a noncontact specular microscope. Cornea.2000;19:148e58
7. Eckard A, Stave J, Guthoff RF. In vivo investigations of thecorneal epithelium with the confocal rostock laser scanningmicroscope (rlsm). Cornea. 2006;25:127e31
8. Efron N, Edwards K, Roper N, et al. Repeatability of measuringcorneal subbasal nerve fiber length in individuals with type 2diabetes. Eye Contact Lens. 2010;36:245e8
9. Erie JC, McLaren JW, Hodge DO, et al. The effect of age on thecorneal subbasal nerve plexus. Cornea. 2005;24:705e9
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x 9
10. Ferreira A, Morgado AM, Silva JS. A method for corneal nervesautomatic segmentation and morphometric analysis.Comput Methods Programs Biomed. 2012;107(1):53e60
11. Grupcheva CN, Wong T, Riley AF, et al. Assessing the sub-basal nerve plexus of the living healthy human cornea byin vivo confocal microscopy. Clin Experiment Ophthalmol.2002;30:187e90
12. Harrison DA, Joos C, Ambrosio R Jr. Morphology of cornealbasal epithelial cells by in vivo slit-scanning confocalmicroscopy. Cornea. 2003;22:246e8
13. Hasler S, Thiel MA, Becht CN. In vivo confocal microscopy ofkeratic precipitates in fuchs heterochromic uveitis syndrome.Klin Monbl Augenheilkd. 2009;226:237e40
14. Hertz P, Bril V, Orszag A, et al. Reproducibility of in vivocorneal confocal microscopy as a novel screening test forearly diabetic sensorimotor polyneuropathy. Diabet Med.2011;28:1253e60
15. Hillenaar T, Cals RHH, Eilers PHC, et al. Normativedatabase for corneal backscatter analysis by in vivoconfocal microscopy. Invest Ophthalmol Vis Sci.2011;52:7274e81
16. Hillenaar T, Sicam VADP, Vermeer KA, et al. Wide-rangecalibration of corneal backscatter analysis by in vivoconfocal microscopy. Invest Ophthalmol Vis Sci.2011;52:2136e46
17. Hillenaar T, Weenen C, Wubbels RJ, et al. Endothelialinvolvement in herpes simplex virus keratitis: an in vivoconfocal microscopy study. Ophthalmology.2009;116:2077e86
18. Holmes TJ, Pellegrini M, Miller C, et al. Automated softwareanalysis of corneal micrographs for peripheral neuropathy.Invest Ophthalmol Vis Sci. 2010;51:4480e91
19. Hurmeric V, Sahin A, Ozge G, et al. The relationship betweencorneal biomechanical properties and confocal microscopyfindings in normal and keratoconic eyes. Cornea.2010;29:641e9
20. Imre L, Nagymihaly A. Reliability and reproducibility ofcorneal endothelial image analysis by in vivo confocalmicroscopy. Graefes Arch Clin Exp Ophthalmol.2001;239:356e60
21. Jonuscheit S, Doughty MJ, Ramaesh K. In vivo confocalmicroscopy of the corneal endothelium: comparison of threemorphometry methods after corneal transplantation. Eye.2011;25:1130e7
22. Kallinikos P, Berhanu M, O’Donnell C, et al. Corneal nervetortuosity in diabetic patients with neuropathy. InvestOphthalmol Vis Sci. 2004;45:418e22
23. Kitzmann AS, Winter EJ, Nau CB, et al. Comparison of cornealendothelial cell images from a noncontact specularmicroscope and a scanning confocal microscope. Cornea.2005;24:980e4
24. Klais CMC, Buhren J, Kohnen T. Comparison of endothelialcell count using confocal and contact specular microscopy.Ophthalmologica. 2003;217:99e103
25. Kojima T, Matsumoto Y, Dogru M, et al. The application ofin vivo laser scanning confocal microscopy as a tool ofconjunctival in vivo cytology in the diagnosis of dry eyeocular surface disease. Mol Vis. 2010;16:2457e64
26. Ku JYF, Niederer RL, Patel DV, et al. Laser scanning in vivoconfocal analysis of keratocyte density in keratoconus.Ophthalmology. 2008;115:845e50
27. Labbe A, Dupas B, Offret H, et al. Evaluation of keraticprecipitates and corneal endothelium in fuchs’heterochromic cyclitis by in vivo confocal microscopy. Br JOphthalmol. 2009;93:673e7
28. Ledbetter EC, Scarlett JM. In vivo confocal microscopy of thenormal equine cornea and limbus. Vet Ophthalmol.2009;12(Suppl 1):57e64
29. Lin H, Li W, Dong N, et al. Changes in corneal epithelial layerinflammatory cells in aqueous tear-deficient dry eye. InvestOphthalmol Vis Sci. 2010;51:122e8
30. McLaren JW, Bourne WM, Patel SV. Automated assessment ofkeratocyte density in stromal images from the confoscan 4confocal microscope. Invest Ophthalmol Vis Sci.2010;51:1918e26
31. McLaren JW, BourneWM, Patel SV. Standardization of cornealhaze measurement in confocal microscopy. InvestOphthalmol Vis Sci. 2010;51:5610e6
32. McLaren JW, Nau CB, Kitzmann AS, et al. Keratocyte density:comparison of two confocal microscopes. Eye Contact Lens.2005;31:28e33
33. McLaren JW, Nau CB, Patel SV, et al. Measuring cornealthickness with the confoscan 4 and z-ring adapter. EyeContact Lens. 2007;33:185e90
34. McLaren JW, Patel SV, Nau CB, et al. Automated assessmentof keratocyte density in clinical confocal microscopy of thecorneal stroma. J Microsc. 2008;229(Pt 1):21e31
35. Midena E, Cortese M, Miotto S, et al. Confocal microscopy ofcorneal sub-basal nerve plexus: a quantitative and qualitativeanalysis in healthy and pathologic eyes. J Refract Surg.2009;25(1 Suppl):S125e30
36. Mimura T, Amano S, Fukuoka S, et al. In vivo confocalmicroscopy of hereditary sensory and autonomicneuropathy. Curr Eye Res. 2008;33:940e5
37. Misra S, Craig JP, McGhee CN, et al. Interocular comparison byin vivo confocal microscopy of the 2-dimensionalarchitecture of the normal human corneal subbasal nerveplexus. Cornea. 2012;31(12):1376e80
38. Mitooka K, Ramirez M, Maguire LJ, et al. Keratocyte density ofcentral human cornea after laser in situ keratomileusis. Am JOphthalmol. 2002;133(3):307e14
39. Mocan MC, Irkec M. Fluorescein enhanced confocalmicroscopy in vivo for the evaluation of corneal epithelium.Clin Experiment Ophthalmol. 2007;35:38e43
40. Moller-Pedersen T, Cavanagh HD, Petroll WM, et al. Stromalwound healing explains refractive instability and hazedevelopment after photorefractive keratectomy: a 1-yearconfocal microscopic study. Ophthalmology.2000;107:1235e45
41. Morishige N, Takahashi N, Chikamoto N, et al.Quantitative evaluation of corneal epithelial oedema byconfocal microscopy. Clin Experiment Ophthalmol.2009;37:249e53
42. Oliveira-Soto L, Efron N. Morphology of corneal nerves usingconfocal microscopy. Cornea. 2001;20:374e84
43. Patel DV. In Vivo Confocal Microscopy Of The Cornea InHealth And Disease. Saarbruecken, Germany, LAP LAMBERTAcademic Publishing GmbH & Co, 2011
44. Patel DV, Ku JY, Johnson R, et al. Laser scanning in vivoconfocal microscopy and quantitative aesthesiometry revealdecreased corneal innervation and sensation in keratoconus.Eye. 2009;23:586e92
45. Patel DV, McGhee CN. Contemporary in vivo confocalmicroscopy of the living human cornea using white light andlaser scanning techniques: a major review. Clin ExperOphthalmol. 2007;35:71e88
46. Patel DV, McGhee CN. In vivo laser scanning confocalmicroscopy confirms that the human corneal sub-basal nerveplexus is a highly dynamic structure. Invest Ophthalmol VisSci. 2008;49:3409e12
47. Patel DV, McGhee CNJ. Mapping of the normal humancorneal sub-basal nerve plexus by in vivo laser scanningconfocal microscopy. Invest Ophthalmol Vis Sci.2005;46:4485e8
48. Patel DV, McGhee CNJ. In vivo confocal microscopy of humancorneal nerves in health, in ocular and systemic disease, and
s u r v e y o f o p h t h a lmo l o g y x x x ( 2 0 1 3 ) xex x10
following corneal surgery: a review. Br J Ophthalmol.2009;93:853e60
49. Patel DV, Phua YS, McGhee CN. Clinical and microstructuralanalysis of patients with hyper-reflective corneal endothelialnuclei imaged by in vivo confocal microscopy. Exp Eye Res.2006;82:682e7
50. Patel SV, McLaren JW, Bachman LA, et al. Comparison of flex-center, center, and corner methods of corneal endothelial cellanalysis. Cornea. 2010;29:1042e7
51. Patel SV, McLaren JW, Camp JJ, et al. Automatedquantification of keratocyte density by using confocalmicroscopy in vivo. Invest Ophthalmol Vis Sci. 1999;40:320e6
52. Patel SV, McLaren JW, Hodge DO, et al. Confocal microscopyin vivo in corneas of long-term contact lens wearers. InvestOphthalmol Vis Sci. 2002;43:995e1003
53. Popper M, Morgado AM, Quadrado MJ, et al. Corneal celldensity measurement in vivo by scanning slit confocalmicroscopy: method and validation. Ophthalmic Res.2004;36:270e6
54. Prydal JI, Franc F, Dilly PN, et al. Keratocyte density and size inconscious humans by digital image analysis of confocalimages. Eye. 1998;12(Pt 3a):337e42
55. Rao K, Leveque C, Pflugfelder SC. Corneal nerve regenerationin neurotrophic keratopathy following autologous plasmatherapy. Br J Ophthalmol. 2010;94:584e91
56. Rieth S, Engel F, Buhner E, et al. Comparison of data from therostock cornea module of the heidelberg retina tomograph,the oculus pentacam, and the endothelial cell microscope.Cornea. 2010;29:314e20
57. Rosenberg ME, Tervo TM, Immonen IJ, et al. Corneal structureand sensitivity in type 1 diabetes mellitus. Invest OphthalmolVis Sci. 2000;41:2915e21
58. Salvetat ML, Zeppieri M, Miani F, et al. Comparison betweenlaser scanning in vivo confocal microscopy and noncontactspecular microscopy in assessing corneal endothelial celldensity and central corneal thickness. Cornea. 2011;30:754e9
59. Scarpa F, Grisan E, Ruggeri A. Automatic recognition ofcorneal nerve structures in images from confocal microscopy.Invest Ophthalmol Vis Sci. 2008;49:4801e7
60. Scarpa F, Zheng X, Ohashi Y, et al. Automatic evaluation ofcorneal nerve tortuosity in images from in-vivo confocalmicroscopy. Invest Ophthalmol Vis Sci. 2011;52(9):6404e8
61. Sheng H, Parker EJ, Bullimore MA. An evaluation of theconfoscan3 for corneal endothelial morphology analysis.Optom Vis Sci. 2007;84:888e95
62. Stachs O, Zhivov A, Kraak R, et al. In vivo three-dimensionalconfocal laser scanning microscopy of the epithelial nervestructure in the human cornea. Graefes Arch Clin ExpOphthalmol. 2007;245:569e75
63. Szaflik JP. Comparison of in vivo confocal microscopy ofhuman cornea by white light scanning slit and laser scanningsystems. Cornea. 2007;26:438e45
64. Twa MD, Giese MJ. Assessment of corneal thickness andkeratocyte density in a rabbit model of laser in situkeratomileusis using scanning laser confocal microscopy. AmJ Ophthalmol. 2011;152:941e53
65. Vargo P, Curtin R. Digital reading center requirements forretinal images from clinics in multi-center trials.J Ophthalmic Photogr. 2010;32:92e5
66. Wang Y, Zhao F, Zhu W, et al. In vivo confocal microscopicevaluation of morphologic changes and dendritic celldistribution in pterygium. Am J Ophthalmol.2010;150:650e5
67. Williams R, Fink BA, King-Smith PE, et al. Central cornealthickness measurements: using an ultrasonic instrument and4 optical instruments. Cornea. 2011;30:1238e43
68. Zhivov A, Stave J, Vollmar B, et al. In vivo confocalmicroscopic evaluation of langerhans cell density anddistribution in the normal human corneal epithelium.Graefes Arch Clin Exp Ophthalmol. 2005;243:1056e61
69. Zhivov A, Winter K, Peschel S, et al. [Quantitative analysis ofcorneal subbasal nerve plexus with in vivo confocal laserscanning microscopy]. Klin Monbl Augenheilkd.2011;228:1067e72
Recommended