Article Text
Abstract
There is an evolution in the demands of modern ophthalmology from descriptive findings to assessment of cellular-level changes by using in vivo confocal microscopy. Confocal microscopy, by producing greyscale images, enables a microstructural insight into the in vivo cornea in both health and disease, including epithelial changes, stromal degenerative or dystrophic diseases, endothelial pathologies and corneal deposits and infections. Ophthalmologists use acquired confocal corneal images to identify health and disease states and then to diagnose which type of disease is affecting the cornea. This paper presents the main features of the healthy confocal corneal layers and reviews the most common corneal diseases. It identifies the visual signatures of each disease in the affected layer and extracts the main features of this disease in terms of intensity, certain regular shapes with both their size and diffusion, and some specific region of interest. These features will lead towards the development of a complete automatic corneal diagnostic system that predicts abnormalities in the confocal corneal data sets.
- Cornea
- Imaging
Statistics from Altmetric.com
Introduction
Corneal confocal microscopy is a non-invasive technique for studying the cellular structure of corneal physiology and disease. It offers non-invasive visualisation of the living tissues and provides greyscale images with greatly increased resolutions over light biomicroscopy and biocytology, which both have been very rewarding approaches to improve definitions of disease phenotypes.1 This ability to provide high-resolution images of all layers in the living cornea has resulted in new discoveries in corneal pathology at the cellular level;2 it has been used in the detection and management of pathological and infectious conditions, corneal dystrophies, monitoring contact lens-induced corneal changes and for pre-surgical and post-surgical evaluation following laser vision correction (photorefractive keratectomy, laser-assisted in-situ keratomileusis, laser-assisted sub-epithelial keratectomy).3
The normal cornea is avascular, and in histological cross section, it consists of five distinct layers. The corneal epithelium lies at the anterior side and is about 50 µm thick at the central corneal region. Bowman’s layer is under the epithelium; it is 10–16 µm thick and is acellular except for the nerves that permeate it. The basal lamina of the epithelium is located on Bowman’s layer where the latter in turn separates the epithelium from the stroma. The stromal region is about 450 µm in thickness and contains large nerves, stromal keratocytes and orthogonal layers of collagen fibres. Descemet’s membrane lies posterior to the stroma. It is an acellular layer and about 15–20 µm in thickness. The limiting layer on the posterior side of the cornea is a single layer of corneal endothelial cells.4
Each corneal disease demonstrates significant qualitative and/or quantitative alterations in the corneal layers in the eyes of patients. In other words, any abnormality occurring in the cornea has its own signs relative to the involved layer. Qualitative changes relate to alterations in the morphology of the corneal layer, while quantitative changes relate to measurements such as cell diameter, cell density, nerve length, total number of nerves within a frame, keratocytes density, number of cells within a frame, corneal reflectivity and corneal thickness.5 In vivo confocal microscopy is rapidly becoming a useful method to evaluate the morphological characteristics of corneal abnormalities at the histological level and may be helpful in diagnosis, determination of progression and understanding the pathophysiology of disease.2
In ophthalmology, reliable and early diagnosis usually depends on the recognition of minute changes in normal structures. To the patient, the functional consequences of such pathological alterations may be the only aspect of concern; however, ophthalmologists look for morphological criteria that are essential in defining health and disease.1 We are working in collaboration with a group of clinical specialists and ophthalmologists to achieve more accurate diagnostics of corneal diseases that will help in developing our vision of an automatic diagnosis system for improving the standards of patient care.
In this paper, we list the most common corneal pathologies, classified according to the International Committee for Classification of Corneal Dystrophies (IC3D classification), and summarise the diagnosis of corneal diseases as they appear in corneal layers with an emphasis on the visual signs of each disease (qualitative changes). We analyse the images as well to extract their features in terms of intensity, region of interest and contained shapes with both their sizes and diffusion.
Our aim is to work towards the development of an automatic corneal diagnostic system that can distinguish between normal and abnormal cornea using the features and the visual signatures that are extracted from data sets of corneal images from confocal microscopes.
This paper is organised as follows. In section ‘Development of confocal microscopy’, the development of confocal microscopy is summarised. In section ‘Corneal layers in health’, a summary of corneal layers in health is presented. A summary of corneal layers in disease is presented in section ‘Corneal layers in disease’. This section displays the most common corneal diseases classified according to the level chiefly affected using the IC3D classification. The final section contains our conclusions and suggestions for future work.
Development of confocal microscopy
This review covers many research papers, review papers and case studies from the past two decades that worked with confocal microscope images of the cornea. We include a subsection to cover the history of, and the key people involved in, the development of confocal microscopy. The images in this review were produced by several types of confocal microscopy: tandem scanning confocal microscopy (TSCM), slit scanning confocal microscopy (SSCM) and laser scanning confocal microscopy (LSCM). For that reason, a subsection is included to indicate the differences between these three techniques and what distinguishes each type from the others.
Historical overview of confocal microscopy
The confocal microscope was originally developed by Marvin Minsky in the mid-1950s. The constructed instrument, which was called the ‘double focusing stage scanning microscope’, was patented in 1957 and focused the light on a small area of tissue.6 ,7
The illumination and observation pathways have a common focal point, and because of that this principle is termed ‘confocal’.6 After Minsky’s work, M. David Egger and Mojmir Petran created a multiple-beam confocal microscope in the late 1960s for examining unstained brain sections and ganglion cells. Egger continued in this arena and developed the first mechanically scanned confocal laser microscope. In 1973, he published the first recognisable images of cells produced by this technique.7 Because of the advances in computer and laser technology coupled to new algorithms for digital manipulation of images, during the late 1970s and the 1980s, a growing interest in confocal microscopy developed.7
Several investigators translated the practical laser scanning confocal microscope designs into working instrument. Dutch physicist G. Fred Brakenhoff developed a scanning confocal microscope in 1979, while at almost the same time Colin Sheppard contributed to the technique with a theory of image formation. Tony Wilson, Brad Amos and John White nurtured the concept and later (in the late 1980s) demonstrated the utility of confocal imaging in the examination of fluorescent biological specimens.7
Types of confocal microscopy
Tandem scanning confocal microscopy
The first real-time TSCM was developed by Petran and Hadravsky using point illumination and detection. The basic part of the system was modified by Nipkow, who invented a spinning disc composed of sets of conjugate pinholes arranged in spirals. After a specimen is scanned in parallel, a two-dimensional image is generated.1 ,8 To eliminate most of the scattered light, the pinholes need to be as small as possible. However, this results in very low light throughput (only 0.25–1% of light reaches the cornea), which necessitates a high-intensity light source and a low-light-level camera to enable image acquisition.8
This original microscope in its clinical version was produced by the Tandem Scanning Corporation (Reston, Virginia, USA) and later by the Advanced Scanning Corporation (New Orleans, Louisiana, USA), and is no longer in production.1
Slit scanning confocal microscopy
An alternative to point scanning, the SSCM was first developed by Thaer for observing the in vivo cornea. This involves illumination from a slit of scanned over the back focal plane of the microscope objective,8 which allows all points on the axis of the slit to be scanned at the same time and consequently greatly reduces scanning times over the TSCM.1 ,8 The slit illumination also allows a greater light throughput than the TSCM and thus a lower intensity light source may be used, allowing longer examination times.8 Contrast with the SSCM is greater in comparison to the TSCM, which provides clearer images of the stroma and enables imaging of the low reflecting layer of wing cells in the epithelium.8 The slit height may be adjustable, which would allow the user to control the thickness of the optical section and the amount of light throughput. However, the microscope is only truly confocal in the axis perpendicular to the slit and provides lower axial and transverse resolution than the TSCM.1 ,8
There are commercial sources that provide slit scanning in vivo confocal microscopes such as Tomey Corporation (Cambridge, Massachusetts, USA), Nidek Technologies (Gamagori, Japan) and Helmut Hund (Wetzlar, Germany).1
Laser scanning confocal microscopy
The LSCM was developed by Webb; it employs a coherent laser as a high-intensity light source. The laser beam is scanned using a set of galvanometer scanning mirrors, which provides fast scanning over the xy plane. The reflected light refocused by the microscope objective is rescanned by the galvanometer scanning mirrors and imaged onto a pinhole aperture located in front of a photomultiplier.1
One of the well-established in vivo confocal imaging systems used in ophthalmology is the Heidelberg Retina Tomograph (HRT) (Heidelberg Engineering, Heidelberg, Germany). HRT is a device that was designed to acquire and evaluate topographic measurements of the optic nerve head to detect glaucomatous damage using a 670 µm diode laser. The University of Rostock (Germany) modified the HRT to use a detachable objective system called the ‘Rostock Cornea Module’ (RCM). This converts it into a high-resolution LSCM for the visualisation of the anterior segment of the eye.1 ,8
Optical sections of only a few micrometres can be imaged with a high contrast and precisely measured because of the high-depth resolution.1 LSCM provides greater contrast than the TSCM or SSCM.8
Performance comparison between different confocal technologies
Niederer and McGhee8 mentioned that limited data are available for comparisons of performance between the different types of in vivo confocal microscopy.
A comparison between keratocyte density measured with TSCM and SSCM (Confoscan 3) demonstrated that, providing the effective depth of the sample volume was taken into consideration, comparable measurements of keratocyte cell density were achievable between the two devices.9
A further qualitative study compared the SSCM (Confoscan 2) and LSCM (HRT II RCM) appearance of the central cornea in normal subjects and in subjects with corneal dystrophy. The observed morphology was similar with both devices, but image contrast appeared greater with the LSCM.10
Endothelial cell density measurement has been compared between SSCM (Confoscan 3) and LSCM (HRT II RCM). All the results of cell density (for patients and normal subjects) that were performed with the RCM were higher than those with the Confoscan 3.11
The authors of refs.8 ,12 compared measurements of cell density and image contrast with contemporary SSCM (Confoscan 4, NIDEK, Gamagori, Japan) and LSCM (HRT II RCM). At the levels of the basal epithelium and the endothelium, they observed good agreement between the two devices for cell density measurement. Mid-stromal keratocyte density (measured in cells/mm2) was approximately double with SSCM compared with LSCM, presumably due to differences in optical section thickness. They observed as well that the contrast at the level of the mid-stroma was greater with the LSCM compared with the SSCM and the contrast levels were consistent throughout the image, whereas there was loss of image quality towards the edges of the SSCM image. Poor agreement was observed in measurement of sub-basal nerve plexus (SNP) density, possibly due to differences in image contrast, with values measured with LSCM higher than those measured with SSCM.
A study13 observed that SNP density measured with SSCM was two to three times higher than density measured with TSCM.
Corneal layers in health
Epithelium
Superficial epithelial cells
Superficial epithelial cells present a polygonal cell pattern, reflecting nuclei with bright illuminated cytoplasm, and perinuclear dark halos. In general, the size of the cell is up to 50 µm in diameter and about 5 µm in thickness (figure 1A).1 ,8 ,14 Broadly speaking, superficial epithelial cells have light cell boundaries and bright visible nuclei (figure 1B).15–17
Wing cells
The epithelial intermediate layer (wing cells) forms a regular mosaic with sharp and reflecting cellular borders. The sizes of wing cells (which are regular in form) are about 20 µm. They can also be subdivided into upper (figure 2A) and lower (figure 2B) wing cells; the latter are smaller.1 ,18 These cells, which can probably only be imaged with a scanning slit in vivo confocal microscopy,16 are identified by their bright cell borders and a bright cell nucleus, and unlike superficial epithelial cells, have no dark oval ring around the nucleus.8 They are characterised by bright cell borders and a bright cell nucleus with few organelles and the nuclei are usually not visible.3
Basal epithelial cells
Basal epithelial cells appear as a regular mosaic of dark cell bodies with light, narrow intercellular reflecting borders and they have a smaller diameter, which is 8–10 µm (figure 3),1 ,8 ,19 and without nuclei reflectivity such that cells show only cell borders.1 ,15 ,16
Sub-basal nerve plexus
The SNP appears as bright, well-defined long linear nerve fibre bundles, frequently demonstrating branches or anastomosis (figure 4A).8 ,16 ,20 It is characterised by the appearance of hyper-reflective fibres of lengths that range between 4 and 8 µm, and organised in a vortex pattern in the lower quadrant of the paracentral cornea (figure 4B).1 ,19 SNP may be identified in Bowman's layer.15
Bowman's membrane
Bowman's membrane (anterior limiting lamina) is an acellular hyper-reflective structure, where SNP may be identified.15 ,16 It appears as an ‘amorphous’ layer when viewed with confocal microscopy.8 It consists of randomly arranged collagen fibrils located in between the basal cells and the stroma. This cell-free area is 10–16 µm in thickness (figure 5).1 ,16
Stroma
Corneal stroma forms around 80–90% of the whole corneal volume. Stromal keratocytes appear as hyper-reflective cell nuclei, typically forming clusters, with poorly visualised cell processes as appeared in figure 6A. The corneal nerves enter the cornea in the mid-stroma and radiate towards the central cornea, dividing into branches (see figure 6B), which are thicker than at the subepithelial level.5 ,8 ,15 ,16 ,20 The corneal stroma consists of three main histological components: cellular, acellular and neurosensorial. The connective lamellae appear black (ie, transparent), whereas the keratocyte nuclei are visible as egg-shaped light reflecting corpuscles. Using cross-sectional analysis, keratocyte density, which measured in cells/mm2, is greatest immediately under Bowman's membrane (in the anterior stroma as seen in figure 6C), and it is reported to be around 500–1300 cells/mm2. The density declines sharply towards the central stroma (minimum 65 cells/mm2), and it increases again slightly in the posterior stroma as seen in figure 6D; it is reported to be around 250–900 cells/mm2.1 ,8 ,15 ,21 ,22 Keratocyte density decreases with age and appears to be independent upon sex.1 ,15 ,16 In ref.,15 it was mentioned that keratocyte density seemed to decrease 0.45% per year.
Descemet's membrane
Descemet's membrane is rather difficult to see under normal circumstances and appears as an acellular layer between the posterior stroma and the endothelium. It is usually indiscernible with confocal microscopy, particularly in younger subjects.8 ,15 Also, nerve plexus is not found.15 It is thin with thickness about 6–10 µm.1
Endothelium
The endothelial cells are located immediately posterior to Descemet's membrane and are characterised by a regular hexagonal hyper-reflective pattern or regular honeycomb mosaic, surrounded by hypo-reflective borders without obvious nuclei reflections (figure 7).1 ,8 ,15 ,16 It is easy to recognise this layer because of its clearly identifiable structure.8 Sometimes, the nuclei of the cells may be visualised.1
Corneal layers in disease
We use the IC3D Classification of the Corneal Dystrophies, which was introduced in 2008.23 Its goal was to develop a new classification system for corneal dystrophies, integrating the most recent information on clinical description, pathological examination and genetic analysis. A template was devised by the IC3D to review the current knowledge of each corneal dystrophy in an organised manner. Each template provides information that includes the name of the dystrophy, alternative names, genetic information (the gene involved and the gene locus of the responsible mutation), onset, signs, symptoms, course, light microscopy, transmission electron microscopy, confocal microscopy and clinical photographs. Each corneal dystrophy is also categorised by the committee based on the level of evidence available to support its existence. The categories are as follows:23
Category 1: A well-defined corneal dystrophy in which a gene has been mapped and identified and specific mutations are known.
Category 2: A well-defined corneal dystrophy that has been mapped to one or more specific chromosomal loci, but the gene (or genes) remains to be identified.
Category 3: A well-defined corneal dystrophy that has not yet been mapped to a chromosomal locus.
Category 4: This is reserved for suspected new (or previously documented) corneal dystrophies, although the evidence for it, being a distinct entity, is not yet convincing.
The IC3D classification system is organised anatomically according to which corneal layer is affected. In general, it classified dystrophies into four groups: epithelial and subepithelial dystrophies, Bowman layer dystrophies, stromal dystrophies, and Descemet's membrane and endothelial dystrophies. For the purposes of this review article, we will focus only on the organisation that depends on the affected corneal level.
Epithelial and subepithelial
Amiodarone-induced keratopathy
Amiodarone-induced keratopathy appears in the corneal epithelial cells. It is characterised by bright intracellular, highly reflective inclusions. In more advanced cases, other corneal layers may be involved: bright microdots arise within the anterior and posterior stroma and on the endothelial cell layer. Also, keratocyte density reduces in the anterior stroma.15 The basal cell layer from a patient during the period of amiodarone ingestion, clearly showing hyper-reflective cell inclusions, is shown in figure 8A. They are in almost the whole image, except the sides, and appear as white and adjacent small rounded-shape structures (each one forms almost 0.04% of the image) on high-intensity background. During the period of amiodarone ingestion, the anterior stroma as well shows marked increase in the number of microdots distributed almost evenly between nuclei of anterior stroma as seen in figure 8B. 24 LSCM (HRT II-RCM) was used to capture figure 8A, B with fields of view 400×400 pixel.
Advancing wave-like epitheliopathy
This disease is characterised by the presence of atypical elongated cells with centripetally oriented long axes (needle shape) that are surrounded by a grey hazy halo. There are hyper-reflective nuclei on darker grey background at the level of the abnormal epithelium, but confluent hyper-reflective regions are demonstrated at the subepithelial level where they appear as hazy reflections.25
Epithelial basement membrane dystrophy
Confocal microscopic images for epithelial basement membrane dystrophy reveal four main features. First, highly reflective irregular material, indicated by arrows in figure 9A, intermixed with the keratocytes of the anterior stroma and posterior epithelial interface, and this area is shown as a reflection of medium values in greyscale. Second, the cells of the basal epithelium have abnormal distended cytoplasm with very reflective nuclei that have grey double-walled aureole (the white circle forms 0.4% of the image), indicated by an arrow in figure 9B. Third, the cysts are elliptic with size of 50–400 µm, indicated by arrows in figure 9C. The borders of cystic lesions are poorly defined, and there is irregular reflective material within them, as indicated by an asterisk in figure 9C. A basement membrane bearing some long (at least 300 µm), highly reflective, linear structures that have diffuse posterior borders and well-delineated anterior borders are indicated by arrows in figure 9D. A non-reflective space appears within this linear lesion and the normal basement membrane.26 SSCM (Confoscan 2) was used to capture figures of epithelial basement membrane dystrophy features with horizontal field width of 610 µm.
Salzmann's nodular degeneration
The confocal images of this disease demonstrate adjacent polygonal and irregularly shaped basal epithelial cells with reflective borders and prominent nuclei appearing in each polygonal cell as a small grey dot as shown in figure 10A. Sub-basal nerve fibres exhibit an abnormal pattern with increased thickness and absence of branching as indicated by arrows in figure 10B. The nerve fibre appears as an undulating sloping line with high intensity. Moreover, there is a presence of reflective cellular elements near these fibres.27 ,28 SSCM (Confoscan 2) was used to capture the images of Salzmann's nodular degeneration.
Gelatinous drop-like dystrophy
The epithelial cells of gelatinous drop-like dystrophy are hyper-reflective adjacent polygonal shapes with irregular structure and often elongated as shown in figure 11A. At the level of the Bowman's membrane in figure 11B, a very small number of sub-basal nerves with increased background intensity can be seen. We can notice as well some highly reflective small dots spread over the image (each forms 0.08% of the image). Figure 11C shows large accumulations of brightly reflective amyloid materials beneath the epithelium and within the anterior stroma. Also, the nuclei of keratocytes were poorly identifiable.29 LSCM (HRT II-RCM) with field of view 400×400 pixel was used to capture the images of gelatinous drop-like dystrophy.
Thygeson epithelial keratitis
Thygeson keratitis shows highly reflective deposits (50–300 μm in diameter) with cotton-like appearance in the basal epithelial cell layer and Bowman's layer (see figure 12A). These cotton-like deposits accumulate in the middle of the image. Highly reflective material appears in figure 12B that directly damages epithelial cell connection while epithelial cells are swelling and losing polygonal structure, and cell gaps are widening as well. In other words, epithelial cells look like cracked ground. These cells are disconnected and appear as high-intensity curvy lines on low-intensity background.30–33 LSCM (HRT II-RCM) with field of view 400×400 µm was used to capture figure 12A, B.
Meesmann's dystrophy
The cystic lesions in this disease, which are 10–50 µm in diameter, appear in round, well-delineated shapes surrounded by reflective points in the cytoplasm, which probably correspond to cell nuclei, as indicated by arrows in figure 13A. Above an apparently normal basal membrane, some normal cell nuclei appear as reflective round regions, as indicated by arrowheads in figure 13A.26 Multiple cyst-like changes are seen at the level of the basal epithelium in figure 13B. They are very similar in size (each forms 3% of the image) and are distributed non-uniformly with white colour while the normal cells have very dark grey colour with unclear borders.8 Figure 13A was captured by SSCM (Confoscan 2) with horizontal field width of 610 µm, while figure 13B was captured by LSCM (HRT II-RCM) with field of view 400×400 pixel.
Recurrent erosion syndrome
In figure 14, superficial polygonal low reflective areas are seen surrounded by bright epithelial cells, and a mucus band appears as a bright line over the epithelial mosaic. The length of each line is approximately equal to the horizontal field width of the image, which is 610 µm. SSCM (Confoscan 2) was used to obtain this figure. No epithelial cystic lesions are found by confocal microscopy.26
Acanthamoeba infective keratitis
Acanthamoeba cysts and trophozoites have a typical structure that allows rapid diagnosis leading to early institution of treatment. The cysts appear as highly reflective, round-shaped particles (10–20 µm in diameter) or ovoid structures, ranging in diameter from 10 to 26 µm within the corneal epithelium and stroma. It is possible to find cysts as single-walled structures, as appeared in figure 15A, or conglomerates with double-walled structures, as indicated by arrows in figure 15B.1 ,8 ,16 ,34–36 Figure 15A was captured by LSCM (HRT II-RCM) with field of view 400×400 µm, and TSCM was used to capture figure 15B with field of view 475×350 µm.
Bacterial keratitis
Confocal microscope images of bacterial keratitis show typical hyper-reflective defects without recognisable structure, as indicated in figure 16A. The adjacent epithelium is oedematous.1 Bacterial keratitis is demonstrated as either distinct needle-like deposits or amorphous deposits at different epithelial depths as shown in figure 16B.8 ,37 An increased density of dendritic-like inflammatory cells within the epithelium, shown in figure 16C, has also been reported in a patient with bacterial keratitis.8 ,36 The study of ref. 38 used confocal microscopy to identify Nocardia, which are filamentous bacteria. The confocal scan of Nocardia in this study revealed multiple, thin and short filamentous structures that demonstrated right-angled branching within the epithelium as indicated by white arrow in figure 16D. These filamentous structures were surrounded by round to oval bright structures that represent inflammatory cell infiltration as indicated by black arrow in figure 16D. Figure 16A–C were taken by LSCM (HRT II-RCM) with field of view 400×400 µm, while figure 16D was taken by SSCM (Confoscan 3).
Viral keratitis: herpes simplex virus
Herpes simplex virus (HSV) infection is usually followed by an inflammatory response, which can seriously damage all corneal layers.39 A classification scheme has been proposed for HSV keratitis:39 ,40 infectious epithelial keratitis, neurotrophic keratopathy, herpetic stromal keratitis and endotheliitis. Superficial epithelium layer shows, as indicated in figure 17A, increase in cell size and hyper-reflectivity while there is a decrease in cell density.39 In HSV endothelitis,40 pseudoguttata look alike true corneal guttata but are less regularly shaped. As indicated by single arrows in figure 17B, instead of a round white dot in the centre, they possess a line of high reflection on the border of the elevated dark area. Intercellular gaps are characterised by small black dots at the vertices of endothelial cells and by pronounced intercellular borders as shown in figure 17B by double arrows. Some patients with HSV infection have alterations of the SNP. Other patients have no visible sub-basal nerve or reduced nerve density.8 ,41 An SSCM (Confoscan 4) with field of view 460×345 µm was used to take figure 17A, while figure 17B was captured by SSCM (Confoscan 3) with field of view 425×320 µm.
For easier reference to the appearance of epithelial and subepithelial diseases, we summarise the visual signature of each disease in online supplementary appendix table A1.
Bowman's layer
Reis–Bückler’s dystrophy
Images of patients with Reis–Bückler’s dystrophy show focal deposition of homogeneous, reflective material, as indicated by arrows in figure 18A, interspersed with the basal epithelial cells. This image shows medium-intensity branching and undulating lines in the middle on a hazy background. In the left of the image, there is an elongated reflective area that has many values of intensity in greyscale; the outside boundary is dark grey, and the inner side near the boundary is grey and the middle is light grey that approximates to values of white. The region of interest (which forms 20% of the image) is located in the right side (almost starts from the centre right and extends to the right top); it is a high-intensity area (arrows in the figure) with grey boundary that is interspersed with the basal epithelial cells. Highly reflective material at the level of Bowman’s layer is indicated by an asterisk in figure 18B. This reflective area covers almost the whole image except the sides. Boundaries of the reflective area have lower intensity. We can notice as well a low-intensity patch in the centre left, which forms 2% of the image and another two patches in the right centre and in the top centre (each forms 0.3% of the image). In figure 18C, showing the anterior stroma,42 asterisk indicates fine diffuse deposits interspersed between keratocyte nuclei, which are indicated by arrows.
Another study demonstrates the presence of focal highly reflective and large number of adjacent irregular and granular materials (on average each material forms 0.2% of the image) without dark shadows in the basal epithelial layer, shown in figure 18D. At the level of Bowman's layer, there is highly reflective small granular material actually replacing Bowman's layer shown in figure 18E.43 SSCM (Confoscan 4) with field of view 767×575 pixel was used to take figure 18A–C, while figure 18D, E were captured by LSCM with field of view 400×400 pixel.
Thiel–Behnke dystrophy
This disease shows the presence of deposits with homogenous reflectivity, rounded edges and dark shadows in the basal epithelium, as indicated in figure 19A. We see these deposits appear as haze of different intensity values in the right third of the image and also in the left third. In the central third, the basal epithelial cells appear clearly with hyporeflective borders. The edges that separate these three parts have a very low-intensity undulating structure but at the level of Bowman's layer. This homogeneous reflective material appears as haze with different values of intensity and there are some diffuse bright irregular spots (each forms on average 0.3% of the image). Bowman's layer is completely replaced as indicated in figure 19B.8 ,43 LSCM with field of view 400×400 pixel was used to take Thiel–Behnke dystrophy images.
For easier reference to the appearance of Bowman layer diseases, we summarise the visual signature of each disease in online supplementary appendix table A2.
Stromal
Lattice dystrophy
In figure 20A, confocal microscopy reveals the presence in the mid-stroma of unspecified, undulating and thinner string-like structures as indicated by arrows, apparently interacting with the keratocytes.44–46 Also, highly reflective deposits, which are indicated by arrows, are observed in figure 20B under the basal epithelial cells. In figure 20C, immediately under the basal cells of the epithelium, Bowman's layer is seen to be irregular in thickness, demonstrated as differences in greyscale, which are indicated by arrows. This may be due to accumulation of amyloid.45 ,46 Poorly demarcated, hyper-reflective, linear and branching structures of varying intensity are the main characteristics of amyloid deposition in lattice dystrophy shown in figure 20D8 LSCM (HRT II-RCM) with field of view 400×400 pixel was used to produce figure 20A–D.
A study on lattice dystrophy42 revealed reflective punctiform structures in the basal epithelial cell layer, indicated as arrows in figure 20E. In figure 20F in the anterior stroma, fine, diffuse, reflective deposits (indicated by asterisk) are interspersed between keratocyte nuclei (indicated by arrows). In figure 20G, showing the anterior and middle stroma, tubular structure filaments with well-defined edges are indicated by arrowheads interspersed with normal keratocytes indicated by arrows. SSCM (Confoscan 4) with field of view 767×575 pixel was used to take figure 20E–G.
Fleck dystrophy
In fleck dystrophy, hyper-reflective and largely intracellular dots with various shapes are distributed throughout the corneal stroma.8 ,15 ,47 They consist of spherical matter with diameters of 3–5 µm and are sometimes enclosed in cyst-like structures as indicated in figure 21, which was taken by SSCM.15 ,47
Granular dystrophy
In granular dystrophy confocal images, hyper-reflective breadcrumb-like deposits measuring approximately 50 µm in diameter are observed in the basal epithelial cell layer, as indicated by arrows in figure 22A. In the anterior stroma near the Bowman layer region, reflective diffuse deposits are noticeable in figure 22B (indicated by arrows). A nerve fibre without myelin appears as a bright line (indicated by arrowhead) in figure 22B. In the deep stroma, shown in figure 22C, keratocyte nuclei are indicated by arrows interspersed with punctiform reflective deposits, which are indicated by arrowheads.42 SSCM (Confoscan 4) with field of view 767×575 pixel was used to obtain figure 22A–C.
Avellino corneal dystrophy
In the basal epithelial of an Avellino corneal dystrophy patient, focal deposits of highly reflective granular materials without dark shadows are observed (see figure 23A). At the level of the superficial and middle stroma, figure 23B shows clusters of highly reflective granular materials with irregular edges.48 ,49 The images of Avellino corneal dystrophy were captured by SSCM (Confoscan 4).
Schnyder crystalline corneal dystrophy
A study analysing the corneal morphology in Schnyder dystrophy demonstrated that highly reflective elliptic material, presumably cholesterol or lipid, accumulates inside and around anterior keratocytes, which have hyporeflective irregular structure, indicated by arrows in figure 24A. Also, brightly reflective punctiform deposits indicated by arrows in figure 24B are associated with the subepithelial nerves. Figure 24A, B were captured by a TSCM (model 165A) with field of view 450×360 µm, and each image was cropped to 250×170 µm. Moreover, crystalline needle-shaped adjacent deposits appear in the anterior stroma,50 and the crystals are needle shaped as shown in figure 24C, or rectangular.51
Another study shows that, at the level of Bowman’s layer, a small number of subepithelial nerves are detectable with high background intensity in figure 24D, but in other cases these nerves are undetectable.51 LSCM with field of view 400×400 pixel was used to take figure 24C, D.
The study of Schnyder dystrophy52 confirms the presence of large or multiple deposits of brightly reflective crystalline material extending from the anterior to the mid-stroma. Some abnormal nerve branches with an irregularly curved appearance are also found in the anterior stroma as shown in figure 24E, which was taken by SSCM (Confoscan 2). The subepithelial nerve plexus are undetectable.
Macular corneal dystrophy
In the superficial stroma of macular corneal dystrophy as indicated by an asterisk in figure 25A, we can see highly reflective deposits without distinct borders. Figure 25B shows the middle stroma containing homogeneous reflective materials with dark striae-like features as indicated by an arrow.48 ,53 LSCM (HRT II-RCM) was used to capture the images of macular corneal dystrophy.
Central cloudy dystrophy of François
This disease reveals small highly reflective granules and deposits in the superficial stromal layer. In the deep stroma adjacent to the corneal endothelial layer, multiple dark acellular striae among extracellular matrices with increased intensities are observed.54
Pre-Descemet's membrane corneal dystrophy
Anterior stroma of patients with pre-Descemet's membrane corneal dystrophy shows normal keratocyte nuclei and a few irregular highly reflective particles as seen in figure 26A. Posterior stroma in figure 26B reveals hyper-reflective and non-homogenous structures with numerous tiny inclusions (indicated by arrows).55 SSCM (Confoscan 2) was used to capture figure 26A, B.
Posterior amorphous corneal dystrophy
The confocal microscope images of this disease demonstrate microfolds and diffuse hyper-reflective sheet-like opacities with spikes extending from the posterior stroma to the stroma immediately adjacent to the endothelial layer (see figure 27, which was taken by SSCM (Confoscan 3)).56
Corneal amyloidosis
Corneal amyloidosis appears as subepithelial and anterior stromal grey white deposits. The confocal microscopic images show these deposits as intercellular, hyper-reflective, cotton candy-like material, fibrillar amyloid material scattered throughout the anterior stroma as shown in figure 28, which was taken by TSCM.15 ,57
Fungal keratitis
LSCM is able to reveal numerous hyper-reflective elements resembling Fusarium, Aspergillus hyphae or Candida pseudofilaments in the anterior stroma even in the early phase of fungal keratitis.1 ,58
Fusarium solani reveals numerous highly reflective linear elements (hyphae) of lengths that range between 200 and 300 µm and of widths that range between 3 and 5 µm, with branches at 90° angles in the anterior stroma as indicated by white arrows in figure 29A. Round inflammatory cells are present also in the stroma as indicated by black arrows in figure 29A, which was taken by LSCM (HRT II-RCM) with field of view 300×300 µm. Aspergillus hyphae have the same characteristics as Fusarium solani hyphae with branches at 45° angles.1 ,58
Candida pseudophilaments are characterised by numerous high reflective elongated particles that measure 10–40 µm in length and 5–10 µm in width located in the anterior stroma as indicated by white arrows in figure 29B, which was captured by LSCM (HRT II-RCM) with field of view 400×400 µm.1 ,58
For easier reference to the appearance of stromal diseases, we summarise the visual signatures of each disease in online supplementary appendix table A3.
Descemet's membrane and endothelial
Fuchs’ endothelial dystrophy (Cornea guttata)
Guttae are characteristic of Fuchs’ endothelial dystrophy. At the level of the endothelium, corneal guttae appear in figure 30A as roundish hyporeflective areas of varying size with central bright spots and surrounded by hyper-reflective endothelial cells.8 ,15 ,59 ,60 TSCM (Topcon SP 2000 P) was used to capture figure 30A. Descemet's membrane displays significant fibrosis in figure 30B, which was captured by SSCM (Confoscan 2).59
Iridocorneal endothelial syndrome
The endothelium in iridocorneal endothelial (ICE) syndrome displays epithelium-like transformation, with prominent bright nuclei with indistinct cell borders.61 The studies in refs.8 ,62 confirm that endothelial cells appear as epithelioid-like cells with irregular borders and non-homogenous, diversely shaped nuclei as shown in figure 31. It is worth mentioning here that eyes with ICE syndrome also have some degree of iris atrophy.61 SSCM was used to capture figure 31.
Posterior polymorphous corneal dystrophy
Vesicles are observed at the level of endothelium in posterior polymorphous corneal dystrophy and appear as relatively big, and well-delineated roundish-shape lesions (having diagonal pattern) containing endothelial cells with surrounding haze, as shown in figure 32A (taken by LSCM (HRT II-RCM) with field of view 400×400 pixel). They appear as a prominent hyper-reflective band lesion at the level of Descemet's membrane, indicated by arrow in figure 32B.8,63
A case report in ref.63 observes the appearance of endothelial vesicular lesions in curvilinear pattern with associated endothelial pleomorphism and polymegathism (see figure 32C). In patients with more advanced posterior polymorphous dystrophy, very prominent corneal nerves at the level of Bowman's membrane are delineated (see figure 32D). Figure 32B–D were taken by SSCM (Confoscan 2).
Brown–McLean syndrome
In a Brown–McLean syndrome patient, the stromal structure was masked by the oedema, with no clear outline of the stromal cells (see figure 33A). At the level of the endothelium, pigmentation was observed in figure 33B as round, bright, hyper-reflective bodies measuring 5–120 µm.64 The images of Brown–McLean syndrome were taken by SSCM (Confoscan 2).
For easier reference to the appearance of Descemet's membrane and endothelial diseases, we summarise the visual signatures of each disease in online supplementary appendix table A4.
Conclusions and future work
The cornea has three main layers separated by two thin membranes. The epithelium is the outermost layer separated from the central stromal layer by Bowman’s membrane. Descemet’s membrane separated stromal layer from the innermost endothelial layer. Each layer and membrane has its distinguished histological structure.
Confocal microscopy is a powerful diagnostic technique that allows non-invasive in vivo cellular imaging of all layers of the cornea, enabling the clinical investigation of numerous corneal diseases that are extensive and fall under the areas of corneal dystrophies, degenerations and inflammations. In vivo confocal microscopy is helpful in evaluating the morphological characteristics of corneal abnormalities at the histological level and may be helpful in diagnosis, determination of progression and understanding the pathophysiology of disease. The ability of providing high-resolution images of all layers in the living cornea has enabled new discoveries of corneal pathology at the cellular level.
In this review, we summarise the diagnosis of most common corneal diseases that are typically classified by the layer of the cornea that is affected using IC3D classification, and we have focused on presenting the qualitative changes (visual signature) in the structure and morphology of the involved layer and some parameters of quantitative analysis, especially the reflectivity.
We detected the regions of interest in corneal diseases and extracted their shape-based and grey-level features manually in a step called knowledge base creation, and we categorised these extracted shape-based features of the reviewed diseases into five groups in online supplementary appendixtable B1. This step is going to help us in investigating the edge-and-region-based segmentation techniques that are able to detect these regions and then to select the most suitable and effective ones. The next stage is to extract features of the segmented regions; these features give a representative of the information that the image has to offer. We are planning to apply statistical texture-based feature extraction techniques that give feature vectors that will be used as input to our classifier aiming to identify the disease that is related to the abnormal cornea. We aim to investigate machine learning approaches to accurately identify the abnormal features of the proposed application.
Put briefly, our aim and future work from this review, as visual computing specialists, is to process the confocal microscopic corneal images and activate appropriate segmentation and feature extraction algorithms to develop an automatic system that can help the ophthalmologists in detecting abnormal cornea and identifying the disease that is related to this abnormality.
Acknowledgments
The authors express their gratitude to the following publishers: Elsevier, BMJ, Springer, Wiley, Molecular Vision, Nature and the Korean Ophthalmological Society for giving permission to reproduce images published in their journals, which were used in this study.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Data supplement 1 - Online supplement
Footnotes
Contributors All the authors have contributed equally to the paper, which is written by RA.
Competing interests None declared.
Provenance and peer review Not commissioned; internally peer reviewed.