A Review of Imaging Biomarkers of the Ocular Surface

Abstract

A biomarker is a “characteristic that is measured as an indicator of normal biological processes, pathogenic processes, or responses to an exposure or intervention, including therapeutic interventions.” Recently, calls for a biomarker for ocular surface diseases have increased, and advancements in imaging technologies have aided in allowing imaging biomarkers to serve as a potential solution for this need. This review focuses on the state of imaging biomarkers for ocular surface diseases, specifically non-invasive tear break-up time (NIBUT), tear meniscus measurement and corneal epithelial thickness with anterior segment optical coherence tomography (OCT), Meibomian gland morphology with infrared meibography and in vivo confocal microscopy (IVCM), ocular redness with grading scales, as well as cellular corneal immune cells and nerve assessment by IVCM. Extensive literature review was performed for analytical and clinical validation that currently exists for potential imaging biomarkers. Our summary suggests that the reported analytical and clinical validation state for potential imaging biomarkers is broad, with some having good to excellent intra- and inter-grader agreement to date. Examples of these include NIBUT for dry eye disease (DED), ocular redness grading scales, and detection of corneal immune cells by IVCM for grading of inflammation and monitoring. Further examples are nerve assessment by IVCM for monitoring severity of diabetes mellitus and neurotrophic keratitis, and corneal epithelial thickness with anterior segment OCT for diagnosis of early keratoconus. However, additional analytical validation for these biomarkers is required prior to clinical application as a biomarker.

A biomarker has been defined by the United States National Institutes of Health (NIH) as a “characteristic that is measured as an indicator of normal biological processes, pathogenic processes or responses to an exposure or intervention, including therapeutic interventions”, while a biomarker signature is a collection of measures that together serve as a biomarker.¹ Molecular-, epigenetic-, cellular-, and tear film protein-based biomarkers have previously been proposed.^2-4 Further, quantification of medical images has also been proposed for potential biomarker application.⁵ The Quantitative Imaging Biomarkers Alliance (QIBA) has defined a quantitative imaging biomarker as “an objective characteristic derived from an in vivo image measured on a ratio or interval scale as indictors of normal biological processes, pathogenic processes, or a response to a therapeutic intervention”.⁵

Biomarkers provide an objective, measurable method of evaluating a disease process. Some biomarkers aim to diagnose disease, which has an obvious translation to clinical care, but can also help researchers to more precisely identify the disease and thereby also identify the risk factors and early signs of the disease. Identification of the early signs of disease could lead to earlier treatments and better health outcomes for patients. Other biomarkers will allow drug developers to measure the effectiveness or safety of treatments. The most desirable biomarkers will be easily and rapidly quantifiable and cost-effective.

For assessment of ocular surface disease, many of our current metrics do not provide the specificity necessary to differentiate these conditions from each other, or allow proper differentiation of severity grades. For example, superficial punctate keratitis can be associated with blepharitis, ocular allergies, dry eye disease (DED), demodex, or contact lens complications, Therefore, biomarkers that have better specificity may allow patients to receive more targeted therapies. In addition, many patients have more than one ocular surface condition or several different or concurrent underlying etiologies. Therefore, clinical trials treating one condition and using only subjective symptom questionnaires may underestimate the effectiveness of treatments because the trial drug is only targeting one of the conditions or one underlying factor that results in patient symptoms. Thus, objective biomarker are needed for ocular surface disease that could allow for better diagnosics, and for monitoring effectiveness of therapeutic interventions.

Any proposed biomarker should thus be evaluated for its analytical and clinical validation (Tables 1 and 2).^1,6 Analytical validation can be thought of as an evaluation of the imaging device and/or classification scheme developed from image parameters and requires demonstration of accuracy, precision, and feasibility of the biomarker. Clinical validation can be thought of as the evaluation of the imaging parameter’s ability to work as a biomarker in the proposed population and/or its association with clinical endpoints. Types of biomarkers include susceptibility/risk biomarkers, diagnostic biomarkers, monitoring biomarkers, prognostic biomarkers, predictive biomarkers, pharmacodynamics/response biomarkers, and safety biomarkers.¹ The statistics associated with the validation of biomarkers is discussed in more detail within the methods section. In each specific section below, the analytical and clinical validation status of potential biomarkers have been defined separately and their respective current states has been summarized.

Table 1-.

Validation Characteristics ^1,6

Analytical validation	Demonstration of the accuracy, precision, and feasibility of biomarker measurement.
• Accuracy	Closeness of agreement between the image acquired and the actual anatomical appearance
• Reference normal value	Normal distribution of values in a healthy population
• Precision	Closeness of agreement between a series of images obtained from multiple sampling of the same homogeneous sample under the prescribed conditions:
∘ Repeatability/Intra-grader reliability	The agreement between successive measurements made under the same conditions.
∘ Reproducibility/Inter-grader reliability	The agreement between successive measurements made with varying conditions, such as location or operator.

Table 2-.

Biomarker Definitions^1,6

Clinical validation	A process to establish that the test, tool, or instrument acceptably identifies, measures, or predicts the concept, which is the aspect of an individual’s clinical, biological, physical, or functional state, or experience that the assessment is intended to capture or reflect, of interest (the aspects necessary for clinical validation are dependent on the type of biomarker being developed)
•  Diagnostic biomarker	A biomarker used to detect or confirm presence of a disease or condition of interest or to identify individuals with a subtype of the disease
∘ Sensitivity	Measures the ability of a diagnostic biomarker to correctly classify the presence of a particular condition/disease in patients that have the condition/disease.
∘ Specificity	Measures the ability of a diagnostic biomarker to correctly classify the absence of a particular condition/disease in patients that do not have the condition/disease.
•  Monitoring biomarker	A biomarker measured serially for assessing status of a disease or medical condition or for evidence of exposure to (or effect of) a medical product or an environmental agent
•  Prognostic biomarker	A biomarker used to identify likelihood of a clinical event, disease recurrence or progression in patients who have the disease or medical condition of interest
•  Predictive biomarker	A biomarker used to identify individuals who are more likely than similar individuals without the biomarker to experience a favorable or unfavorable effect from exposure to a medical product or an environmental agent
•  Pharmacodynamic/ response biomarker	A biomarker used to show that a biological response has occurred in an individual who has been exposed to a medical product or an environmental agent
•  Safety biomarker	A biomarker measured before or after an exposure to a medical product or an environmental agent to indicate the likelihood, presence, or extent of toxicity as an adverse effect
•  Susceptibility/risk biomarker	A biomarker that indicates the potential for developing a disease or medical condition in an individual who does not currently have clinically apparent disease or the medical condition

Methods

The potential imaging biomarkers selected for inclusion in this review were based on imaging parameters that could be obtained with devices currently available to clinicians and not experimental devices or prototypes. Electronic searches of PubMed were used to analyze the literature for each potential biomarker. Search terms included “tear break up time AND repeatability,” “tear break up time AND reliability,” “tear break up time AND sensitivity,” “tear break up time AND specificity,” “tear break up time AND treatment,” “tear break up time AND correlation,” “tear break up time AND dry eye disease,” “non-invasive tear break up time,” for NITBUT; “tear meniscus height,” “tear meniscus area,” for tear meniscus measurement; “meibography,” “meibomian gland drop out,” “anterior segment optic coherence tomography,” for meibomian gland morphology; “in vivo confocal microscopy,” and its different combinations with; “corneal nerve,” “corneal density,” “nerve morphology,” “nerve tortuosity,” “nerve beading,” “nerve reflectivity,” “nerve thickness,” “neuroma,” “neuropathic corneal pain,” “diabetes mellitus,” “systemic disease,” “sensitivity,” “specificity,” “keratitis,” “acanthamoeba,” “fungal keratitis,” “demodex,” for corneal nerve alterations and qualitative biomarkers; “epithelial corneal thickness,” “keratoconus AND anterior segment oct,” “limbal stem cell deficiency AND anterior segment oct,” “subclinical keratoconus AND anterior segment oct,” “dry eye disease AND anterior segment oct,” for corneal thickness and epithelial thickness; “ocular redness,” “ocular redness grading,” “ocular redness scale,” “conjunctival hyperemia,” “bulbar redness,” “bulbar hyperemia,” “dendritic cell,” “dendritiform cell, “palpebral inflammation,” “in vivo confocal microscopy” for ocular surface inflammation. The results of this search were supplemented by abstracts from the authors in order to provide the most updated data available for each potential biomarker. Manuscripts/abstracts were examined for the inclusion of one aspect of analytical or clinical validation discussed below. After completion of the manuscripts/abstracts, the results were used to evaluate each potential biomarker’s analytical and clinical validation.

Analytical validation should include verification of the accuracy of the biomarker. For imaging biomarkers, this verification should prove that the structure within an image is being named the correct anatomical structure. Quantitatively, analytical validation includes repeatability and reproducibility. Repeatability and reproducibility address the variability that is possible between visits in the same subject due to differences in examiner, environment, etc. These are often quantified using intraclass correlation coefficients (ICC), kappa values (only for categorical data), coefficients of variability/variation (CoV), and the repeatability coefficient. If ICC is used, poor, moderate, good, and excellent agreement is defined by values of less than 0.50, 0.50 to 0.75. 0.76 to 0.90, and greater than 0.90, respectively. If Kappa is used, no, minimal, weak, moderate, strong, and almost perfect agreement is defined by values of less than 0.20, 0.21 to 0.39, 0.40 to 0.59, 0.60 to 0.79, 0.80 to 0.90, and greater than 0.90, respectively. Weighted kappa can also be used, which is typically larger than unweighted kappa values, because it takes into consideration the amount of disagreement. For example, grades of 1 and 2 for the same image are not considered the same amount of disagreement as grades of 1 and 4. This extent of disagreement is taken into consideration in a weighted kappa value but not an unweighted kappa. CoV is calculated by dividing the standard deviation by the mean. This ratio, which is often represented as a percentage, can be any positive number including zero. If necessary, CoV values reported in manuscripts were converted to percentage by the authors. The repeatability coefficient is derived from the within subject standard deviation. Ranges defining poor, moderate, good, and excellent repeatability do not exist for CoV and repeatability coefficient values; however, they can be compared to each other within the same biomarker provided the measures are absolute instead of relative.

Statistics required for clinical validation is dependent on the biomarker type being tested. It should be noted that a biomarker validated for one purpose does not necessarily mean that it can be used for another purpose. For example, a biomarker may be ideal for use as a diagnostic biomarker, but not as a monitoring biomarker. Diagnostic biomarkers typically are validated by measures of sensitivity, specificity, and/or receiver operating curves. Manuscripts reporting significant differences between groups are valuable in identifying imaging parameters that could be pursued as potential diagnostic biomarkers. However, the diagnostic validation step should consider if more than one disease/condition is capable of causing a reduction/increase in a potential biomarker compared to a control group. If this is true, the potential biomarker may not be useful in diagnosis due to low specificity. Monitoring biomarkers could be validated by correlating changes in the biomarker with changes in the status of the disease. Prognostic, predictive, safety, and susceptibility/risk biomarkers can be validated with survival analyses and/or odds ratios. Alternatively, these biomarker types may be able to be validated by comparing the response of those who were administered a therapy to those who were not when the two groups are further divided by presence/absence of the biomarker. Pharmacodynamic biomarkers may be validated by comparing the change in the biomarker in a group of patients who were given a control/standard treatment to that of a test treatment/environmental agent.

This review aims to provide an overview of the state of validation of current and potential imaging biomarkers for ocular surface diseases. The text provides a descriptive and interpreted analysis for each potential biomarker. Supportive quantitative data is provided in Tables 3 and 4.

Table 3.

Summary of the Biomarkers’ Analytical and Clinical Validation.

		Analytical Validation		Clinical Validation
Biomarkers	Normal Range	Repeatability/ Intra-observer reliability	Reproducibility/ Inter-observer reliability	Diagnostic		Monitoring	Predictive
				Sensitivity	Specificity
NIBUT	≥10 seconds	CoV= 7.3-36.6%18,26,27 ICC= 0.750-0.93018,23,25-27	CoV= 15.4-21.85%18,27 ICC= 0.88027	80.0-87.0% 29-31	86.0-100% 29-31	✔	✔
TMH
•  Time Domain	230-332 μm42,45,51	CoV= 21.3-49.6%42,51,292,293	Sp= ±29μm36,51	67.0%42	81.0%42	✔
•  Spectral Domain	319-354 μm36,51	CoV= 9.0-28.4%39,40,51,53 ICC= 0.900-0.98139,40,51,53	Sp= ±15μm36,51 CoV= 7.0-17.1%39,40,51,53 ICC= 0.920-0.93039,40,51,53	80.5%	89.3%	✔
•  Swept-Source	183-339 μm39,40,51,53	CoV= 23.9-28.0%48,294 ICC= 0.909-0.98148,294	CoV= 14.5-28.0%48,294 ICC= 0.932-0.94848,294	67.0%	88.0%	✔
TMA
•  Time Domain	3,414-35,283 μm2 15,39,48-50					✔
•  Spectral Domain	25,383-62,100 μm2 15,39,48-50	CoV = 17.0-44.0%53,54,292	CoV = 19.5-35.4%53,54,292	86.1%	85.3%	✔
•  Swept-Source	12,459-27,900 μm2 15,39,48-50	CoV= 45.5-52.0%48,294 ICC= 0.873-0.99048,294	CoV= 27.7-29.4%48,294 ICC=0.813-0.98248,294	62.0%	92.0%	✔

NIBUT: non-invasive tear break-up time, TMH: tear meniscus height, TMA: tear meaniscus area, Sp= pooled standard deviation, CoV: coefficient of variability, ICC: intra- or inter-class correlation coefficient.

Table 4.

Summary of the Biomarkers’ Analytical and Clinical Validation.

		Analytical Validation		Clinical Validation
Biomarkers	Normal Range	Repeatability/ Intra-observer reliability	Reproducibility/ Inter-observer reliability	Diagnostic		Monitoring	Predictive
				Sensitivity	Specificity
IR Meibography				49-88%48,292	65-100%48,292
•  % parcial gland		kw= 0.15 to 0.9181	Kw= 0.36-0.7781,82
•  % gland drop out		kw= 0.40-0.8182	Kw= 0.21-0.9082,85			✔
•  software analysis		ICC= 0.570-0.74082	ICC= 0.520-0.87085,93			✔
OR
•  Scales	0 grade [0-3/4]	CoV= 17-20%100-102,104 CoR= 0.31-0.59100-102,104 ICC= 0.980100-102,104	CoR= 0.49-0.80100-102,104			✔	✔
•  VBR	-	CoV= 3.9-9.1%107,295-297	CoR= 5.0-5.9 ICC= 0.992-0.999107,295-297			✔	✔
•  ORI	-	-	ICC= 0.995-0.998108			✔	✔
Dendritiform Cells	Central: 9-64 cells/mm2 130,158,159,298 Peripheral: 0-208 cells/mm2 130,158,159,298	CoV= 4.4-11.9%75,131-139	ICC= 0.991-0.99775,131-139			✔	✔
Palpebral Inflammation						✔
•  Immune cells on IVCM	Epithelial: 123.3-123.7 cells/mm2 74,178 Stromal: 36.7-38.8 cells/mm2 74,178	ICC= 0.960-0.97074,75	ICC= 0.920-0.99774,75	94.0%75	92.0%75	✔
Nerve Length	17.1-31.7 mm/mm2 179,184,185,187-189	ICC= 0.700-0.980182,189 DM ICC=0.980189	ICC= 0.610-0.940182,189	DED: 81.9%185	DED: 85.0%185	✔	✔
Nerve Density	8.4-31.9 mm/mm2179,184,185,187-189	ICC= 0.670-0.920182,189 DM ICC=0.840189,189	ICC= 0.520-0.960182,189	DM: 82.0%239-242 DED: 80.2%185	DM: 52.0%239-242 DED: 85.0%185	✔	✔
Nerve Beading	90-198 beads/mm182,190,194	ICC= 0.930187,195,196	ICC= 0.870187,195,196			✔
Nerve Tortuosity	1.1 grade [0-4] 192	ICC= 0.200-0.690 179,193,194	ICC= 0.600-0.930 179,193,194	51.9 - 58.8%250	81.1 - 84.7%250	✔
Nerve Reflectivity	1.1-2.6 grade187,196,197					✔
Nerve Thickness	0.52-4.68 μm182,190,194					✔
Epithelial Corneal Thickness	50.4-53.4 μm182	Central CoV= 1.7-2.4%256-260 Peripheral CoV=1.8-3.2%256-260 ICC= 0.957-0.990	Central CoV= 0.4-3.6%256-260 Peripheral CoV=2.4-3.8%256-260 ICC= 0.891-0.970	AUC= 0.640-0.880 AUCPSD= 0.985-1.000		✔	✔

Tear Film Assessment

Non-Invasive Tear Break Up Time (NIBUT)

The tear film plays a key role in maintaining ocular surface health and quality vision. The tear break-up time (TBUT) reflects the quality and stability of the tear film and is reduced in patients with an evaporative component of DED, compared to normal subjects.^7-9 Most commonly, reduction in TBUT is associated with a compromised tear lipid layer; however, mucin and tear dynamics have also been cited as contributors to TBUT changes.^9,10 Traditionally, assessment of TBUT is performed after instillation of fluorescein dye, which enables visualization of the tear film; however, the amount and concentration of fluorescein, as well as timing of assessment and humidity can cause variability and lack of adequate reproducibility.^11-15

The NIBUT attempts to eliminate the variability of TBUT caused by fluorescein dyes through imaging. It utilizes disruptions in a regularly patterned image (grid or concentric circles typically), projected onto the tear film to identify areas of tear film disruption (Fig. 1). The measurement of NIBUT initially required a subjective measurement by an examiner and results show a longer break-up time for a subject when compared to their fluorescein TBUT.¹⁶ More recently, the development of new devices with automated softwares for NIBUT quantification through detection and mapping of tear film break-up location over time, has attempted to eliminate inter-grader variability observed with subjective examiner assessment.^17,18 Initial studies with these softwares reported shorter break-up times compared to TBUT;^17,19 however, longer break-up times were also reported afterwards.²⁰ Despite different comparative results with TBUT, NIBUT with a cut off value less than or equal to 10 seconds has now been identified by the Dry Eye Workshop II (DEWS II) as an indicator for DED diagnosis with 82-84% sensitivity and 76-94% specificity.²¹

Figure 1.

Non-invasive tear film break-up time measurement by Oculus Keratograph 5M (Oculus, Arlington, WA, USA). a) Normal tear film break-up time with regular pattern. b) First and average break-up times were lower than normal values based on color scale in a patient with evaporative dry eye disease (movie of the same patient can be seen as supplementary data).

Analytical Validation

Accuracy of NIBUT was determined by comparing NIBUT values to those obtained via the gold standard TBUT. Significant correlations between these values have been reported.^18,22,23 However, while some studies have reported that TBUT is longer than the NIBUT,^18,20,22 others have reported the opposite.^23,24 The reason for the varying reports may be due to different subject cohorts (DED vs. control subjects, age, etc.).

The cut-off value for abnormal NIBUT can be assumed to be less than 10 seconds. There was less variability when the determination of NIBUT was made by software instead of the clinician.²⁵ As such, the Reported ICC values suggest good to excellent repeatability for automated NIBUT; however, there was a large CoV range suggesting unknown amount of variability, which means NITBUT does not exhibit consistent repeatability, and a wide range of repeatability has been reported by different groups (Table 3).^18,23,25-27 It was also noted that the variability of the NIBUT was greater at greater values of TBUT.²⁴ Overall, ICC values suggest good reproducibility CoV ranged from 15.4% to 21.85% (Table 3).^18,26 However, low ICC values suggest there was little agreement between two devices that provide NIBUT,²⁸ suggesting that measures from different devices should be compared with caution.

Clinical Validation

The sensitivity and specificity of NIBUT provides an assessment of its ability to act as a diagnostic biomarker. When the DEWS II cut-off of 10 seconds is used to differentiate DED from controls (Fig. 1), NIBUT provides a sensitivity of at least 80.0% and a specificity of at least 86.0.^29-31 Other cut-off values have been proposed, but the evidence does not support a clinically significant improvement in sensitivity and specificity compared to the 10 second cut-off. ^23,19,32.¹⁸ In addition, NIBUT has been used to show improvements (increase) with treatment, suggesting that it could be used as a monitoring biomarker,^33,34 albeit inconsistently.³⁵ Thus, additional data is needed to support the utilization of NIBUT as a monitoring biomarker, a process that can be accelerated by encouraging more studies to employ NIBUT methods, rather than the currently prevalent fluorescein TBUT approach.

In summary, the validation of the NIBUT as a diagnostic biomarker for DED has been investigated, and its’ good sensitivity and specificity values suggest that it is appropriate for utilization. Further, NIBUT shows good repeatability and reproducibility as a diagnostic biomarker. However, the larger variability it suffers is derived from the larger variation at higher values. This is common in other measures as well and should be taken into account during biomarker validation and when applied in a clinical setting. Further, these variations could also be reflecting disease processes. Evidence also suggests that NIBUT has potential as a monitoring or predictive biomarker in DED; however, further work is necessary to support its utilization in this regard. Current literature to support the potential of NIBUT to be used as a prognostic, safety, susceptibility/risk, or pharmacodynamic biomarker does not currently exist.

Tear Meniscus Measurement by Anterior Segment Optical Coherence Tomography (OCT)

The lower eyelid tear meniscus can be measured for tear meniscus height (TMH) and/or tear meniscus area (TMA),³⁶ which have been found to be reduced in aqueous-deficient DED.³⁷ As such the DEWS II report suggested that TMH be used for diagnosis of aqueous-deficient DED.^36,38-40 The utilization of techniques with infrared light and anterior segment OCT, using low coherence light, allows TMH and TMA to be measured without reflex tearing, which has improved its accuracy.^41-44 The main anterior segment OCTs used to date are time-domain (TD-OCT) and spectral-domain OCTs (SD-OCT),⁴⁵ with the latter performing scans at a much faster rate. Spectral-domain OCTs have improved resolution and a significantly higher acquisition speed, resulting in improved image quality, and minimized motion artefacts compared to TD-OCTs.⁴⁶ Moreover, a different OCT modality called swept-source (SS-OCT) has more recently become available.⁴⁵ One of the advantages of SS-OCT imaging over SD-OCT is the faster scanning speed, which allows for denser scan patterns and larger scan areas compared with SD-OCT scans for a given acquisition time. Additional advantages of SS-OCT technology are its use of a longer wavelength and its reduced sensitivity roll-off, resulting in better detection of signals from the deeper layers.⁴⁷

Analytical Validation

Reported mean TMH values in samples of normal subjects are varied, but are less than 0.354 mm, regardless of type of anterior segment OCT (Table 3). The measures gained via anterior segment TD-OCT and SD-OCT are typically lower³⁹ and SS-OCT is higher,⁴⁸ compared to those obtained via slit-lamp. Therefore, normative database and cut-off values for each device are recommended and different measurements between OCTs and slit-lamp are explained by nature of imaging technique and image processing procedures.³⁹ The range of inter-grader ICC values is smaller for anterior segment OCT and are considered good to excellent.^15,39,48-50 When compared to time domain OCT, spectral domain OCT has shown better intra- and inter-grader agreement in most studies based on reported CoV values and narrower 95% confidence intervals. In addition, the ICC values suggest excellent repeatability.⁵¹ Further, swept-source OCT has shown excellent repeatability as well; however, the CoV values for intra- and inter-grader repeatability are comparable or better to spectral domain OCT (Table 3).³⁹

The mean normal TMA range has been reported with high variability based on the anterior segment OCT methods used (Table 3).^{37,40,42,48,52,53} TMA measurements tend to be larger with time domain OCT as compared to spectral domain OCTs. Overall, based on the current literature, spectral domain OCTs provide superior intra- and inter-grader repeatability compared to time domain and swept-source OCTs, even though the intra- and inter-grader ICCs with swept source OCTs were moderate-to-high (Table 3).

Clinical Validation

TMH as a diagnostic biomarker for DED using TD AS-OCT has shown a 67.0% sensitivity and 81.0% specificity (Table 3). In comparison, SD AS-OCT demonstrates 80.5% sensitivity and 89.3% specificity for TMH and 86.1% sensitivity and 85.3% specificity for TMA. The diagnostic ability is best for Sjögren’s aqueous-deficient DED and is moderately acceptable (i.e., just greater than 65%) for non-Sjögren’s aqueous-deficient DED. In comparison, the diagnostic ability for evaporative DED is less than 50%.⁵⁴

Like TMH, TMA has better sensitivity with spectral domain compared to swept-source anterior segment OCT, but worse specificity (Table 4). TMA has shown better diagnostic ability than TMH in evaporative DED, non-Sjögren’s aqueous-deficient DED, and Sjögren’s aqueous-deficient DED.⁵⁴ However, the accuracy is above 70% only in non-Sjögren’s aqueous-deficiency and Sjögren’s aqueous-deficiency DED. Interestingly, El-Fayoumi et al. have shown a decreased TMH and TMA in rheumatoid arthritis patients, compared to healthy controls, despite the absence of clinical diagnosis of DED.⁵⁵ Because TMH and TMA changes were present prior to a diagnosis of DED, they may be useful as suspectibility/risk biomarkers; however, extensive additional work would be needed to validate TMH and TMA for these purposes.

The TMH and TMA on all OCT devices have been used as a monitoring biomarker to assess changes in tear meniscus following DED treatment^56-58 or to assess the effect of contact lens wear.⁵⁹ Increased tear meniscus measurements after installation of artificial tears, diquafasol sodium and rebamipide were reported in DED.^56,60 Furthermore, decreased tear meniscus height with contact lens wear and improvement after installation 3% diquafosol solution has also been reported.⁵⁹

Standardization of TMH and TMA measurements and potentially automated methodologies would be helpful in ensuring additional analytical validation. Furthermore, when using TMH or TMA in a clinical trial setting, some limitations should be considered when interpreting the outcomes. These include the presence of conjunctivochalasis, time-from-blink, and the prevailing environmental conditions, all of which may cause additional variability of both TMH and TMA measures.²¹

In summary, TMH and TMA have acceptable diagnostic ability in DED, especially in aqueous deficiency DED. Additionally, these parameters can be helpful in early diagnosis of DED in systemic conditions. However, normative databases, cut-off values and standardized acquisition methods should be described for each device. Altogether, the use of TMH and TMA as diagnostic and monitoring biomarkers appears promising.

Assessment of Meibomian Gland Morphology

Meibomian glands (MG) are modified sebaceous glands situated in the tarsal plate of the upper and lower eyelids and produce meibum, which is essential to stabilize the tear film and decrease evaporation. MG dysfunction (MGD) was defined by The International Workshop on Meibomian Gland Dysfunction (MGD Workshop) as “chronic, diffuse abnormality of meibomian glands that is commonly characterized by terminal duct obstruction or qualitative or quantitative changes in glandular secretion”.⁶¹ MGD can result in more rapid tear film evaporation and may lead to DED. Therefore, evaluation of MG function and/or morphology is required for the diagnosis of MGD.⁹ To date, MG morphology has been assessed anatomically by infrared meibography and at a cellular level by in vivo confocal microscopy (IVCM).

IVCM parameters that could be used as biomarkers, include MG orifice width (Fig. 2a), MG acinus descriptors, such as short diameter, long diameter, and density.⁶² However, the images provided in publications identified by authors as presumed MG acini vary widely in morphology.^62-78 In fact, a recent study provided histological evidence that image appearance refined in some studies as presumed MG acini were actually rete ridges of the conjunctiva.⁷⁹ This highlights the importance of accuracy in analytical validation and the lack of its current availability regarding MG acini and IVCM. Given the lack of consensus regarding MG acini morphology by IVCM, parameters related to the MG acini will not be discussed in this review.

Figure 2.

a) In vivo confocal microscopy of the orifice of meibomian glands (red arrow) of the eyelid and b) presence of immune cells (yellow arrows) surrounding acinar unit.

Infrared Meibography

MG dropout is defined as areas showing a loss of acinar tissue as identified by meibography.⁶¹ An example of MG dropout can be seen in Fig. 3. Even though evaluation of MGs by transillumination has been described decades ago, MG dropout has been more widely used since the introduction of non-contact infrared meibography, which has a shorter acquisition time, higher image quality, and less patient discomfort.⁸⁰ Several devices have been developed that can acquire the images necessary for the assessment of MG dropout, and several grading scales for MG dropout have since been presented.^80-82 Two grading scales, the Gestalt grading scale and Meiboscale, were recommended by the MGD Workshop and will be evaluated in this review. In the Gestalt grading scale, a partial gland was defined as “one that is incomplete and present in clumps or clusters.” Each lid is graded based on the percent of partial glands, such that Grade 1 is no partial glands, Grade 2 is less than 25% partial glands, Grade 3 is 25-75% partial glands, and Grade 4 is greater than 75% partial glands. In the Meiboscale, each lid is graded based on percent dropout as follows: Grade 0: no loss of meibomian glands, Grade 1: area loss less or equal to 25%, Grade 2: area loss less than or equal to 50%, Grade 3: area loss less than or equal to 75%, and Grade 4: area loss less than or equal to 100%.⁸⁰ More recently, continuous grading scales based on a semi-automated software have been developed. This semi-automated software is used to outline the total MG area and the part of the area that has MG dropout. The dropout area is then divided by the total area to give a percent of dropout that ranges between 0% and 100%.⁸³ One of the obvious advantages of the continuous scale is that small changes can be reflected with the continuous grading scale that may not be shown in the categorical grading scales. For example, a subject with 10% drop out who progresses to 20% drop out would be in the same meiboscore grading, but this change would be reflected in the continuous grading scale.

Figure 3.

Meibography images (Infrared 840 nm) recorded by Oculus Keratograph 5M (Oculus, Arlington, WA, USA) a) Meibomian glands with moderate dropout and increased tortuosity of the glands are seen on the upper eyelid. b) Lower eyelid: severe dropout and dilatation of meibomian glands can be seen. Meibography images (nearinfrared 735 nm) of c) upper and d) lower eyelid recorded by Optovue RTVue-XR Avanti OCT System (Optovue, Fremont, CA, USA)

Additional parameters associated with meibography include MG bend, MG thickness,⁸⁴ MG distortion, tortuosity, hooking, shortening, thickening, thinning, overlapping, and ghosting. One study clearly defined these parameters and assessed the intra- and inter-grader agreement of these parameters and reported low values of agreement, suggesting that they may not be ideal biomarkers.⁸⁵ This lack of agreement suggests little promise for MG bend, MG thickness, MG distortion, tortuosity, hooking, shortening, thickening, thinning, overlapping, and ghosting to become biomarkers. Given the lack of additional publications assessing these parameters, we will not further assess these parameters here.

Analytical Validation

The accuracy of infrared meibography is supported by a histological study that included analysis of light transmission through the MG.⁸⁶ This report showed that the meibum within MG acini cells scattered infrared light and caused the MG to appear white in the infrared meibography images.⁸⁶ The authors also noted that the extent of light scattering appears to be greatest in areas that contain small lipid droplets, which were identified using neutral lipid fluorescent staining, as opposed to areas with coalesced lipid droplets or areas without lipid droplets.⁸⁶ Although it should be noted that other abnormalities, such as concretions, can also reflect light in a similar way to MGs and can therefore be confused with them.⁸⁷ The extent of MG dropout is age dependent even in non-DED patients.⁸⁸ Therefore, normative values are different for various ages. The MGD Workshop proposed the following cut-off values: patients 20 years of age or less should have no dropout and patients over 20 years of age may have less than or equal to 25% dropout.⁸³

Repeatability of MG dropout is dependent on the repeatability of 1) image acquisition and 2) the grading scale. Image acquisition can be affected by the positioning of the lid during eversion and the patient’s direction of gaze, both of which can change even within a single session.⁸⁹ It can also be affected by the ability of the eyelid to evert, which is inconsistent across subjects.⁸⁵ The intra-grader weighted kappa values for the Gestalt grading show a large range from no agreement to almost perfect agreement (Table 4). ⁸¹In addition, less agreement was present in patients with greater MG loss [weighted kappa = 0.79 (95% CI: 0.69-0.88)] compared to those with less/no MG loss [weighted kappa = 0.82 (95% CI: 0.73-0.91)].⁸¹ For the meiboscale, the reported intra-grader agreement ranged from weak to strong when weighted kappas were evaluated and minimal when unweighted kappas were considered.In one study that compared these subjective grading scales, the meiboscale showed better intra-grader agreement, compared to Gestalt grading.⁸² When a continuous grading scale was used, the intra-grader ICC values showed moderate agreement.⁸⁵ The ICC was higher (although not statistically significant) when areas of indistinct MG appearance were classified as drop out, compared to when they were not, suggesting that such directives, training, or computer automation that reduces grader subjectivity, may assist in improving intra-grader agreement.⁸⁵

Reproducibility studies have shown that meibography devices can provide varying images due to variability in amount of lid everted and image quality, causing differences in MG drop out gradings⁹⁰ and quantitative image analysis.⁹¹ This suggests that if MG dropout is used as a biomarker to monitor disease progression, the same device should be used consistently throughout all visits. In addition, the upper and lower eyelids should be graded separately based on evidence that there is a significant difference in MG area/total analysis area between upper and lower lids, especially in eyelids with lower level of dropout.^84,92 The weighted and unweighted inter-grader kappa statistics for the Gestalt grading and the meiboscale showed large agreement ranges for upper and lower eyelids.^81,82 In general, inter-grader agreement was better for the upper eyelids compared to the lower eyelids.⁹³ However, for the continuous grading scale, the inter-grader ICCs show moderate to good agreement^85,93 and there was approximately equal amount of agreement for the upper and lower eyelids.⁹³

While the accuracy for infrared meibography has been addressed, the repeatability and reproducibility is inconsistent. This inconsistency, especially with the categorical grading scales, suggests that a more standardized protocol should be used across multiple studies, in order to improve the consistency of image acquisition and grading, and thereby the analytical validation. In addition, the utilization of a continuous grading system appears to be better than a categorical system. Incorporation of the continuous grading scale into a clinical setting may necessitate the use of artificial intelligence in order to address problems of feasibility.^61,82

In summary, MG dropout may be affected by image acquisition methods and grading scales. To minimize image acquisition related problems, same devices should be used during monitorization. Methods to standardize eyelid eversion should be developed and upper/lower eyelids should be graded and compared separately. To decrease grading scores related problems, continuous grading systems seems to be promising and may be implemented into clinical practice. MG dropout may be a good candidate as a diagnostic biomarker; however, its use as a biomarker for monitoring treatment response needs to be established further.

Clinical Validation

MG dropout was a significant discriminator between DED and control subjects when they were diagnosed based on OSDI score.⁸⁴ This parameter has also proven to be useful in discriminating between aqueous-deficient and evaporative DED, suggesting that MG dropout may be a good diagnostic biomarker.⁹⁴ A proposed meiboscore cut-off value of three for diagnosis of MGD provided 49.3% sensitivity and 64.5% specificity.⁹⁵ However, in a recent study based on a different subjective grading scale,⁸¹ the sensitivity and specificity of a cut-off value of 0.75 were reported with 87.9% and 100%, respectively. Moreover, MG dropout did not predict effectiveness of liposomal spray, latent heat goggles, and warm compresses,⁹⁶ suggesting that it is not a good predictive biomarker. In summary, both upper and lower eyelid drop out has been correlated with lipid layer thickness, NIBUT, OSDI, and age, suggesting it may be useful as a monitoring and/or pharmacodynamic biomarker. Currently, literature to suggest that MG drop out could be used as a prognostic, safety, or susceptibility/risk biomarker is not available.

Ocular Surface Inflammation

Conjunctival Redness

Bulbar conjunctival hyperemia is caused by vasodilation, increased blood flow and capillary permeability associated with ocular surface inflammation, and is one of the most common clinical signs perceived by the patient that can correlate to inflammation of the ocular surface and is often the main cause for seeking medical care.^97-99 However, inflammation can be subclinical, such as in DED, and may not be detected by slit-lamp examination or perceived in early disease.⁷⁴ Previously, ocular redness assessment relied on ordinal scales to classify the severity ranging in a 4 or 5 category scale from normal or trace to severe (Fig. 4).^100,101 When these scales are applied to photographs, they become part of the potential biomarker, and thus, should be validated like other imaging biomarkers.

Figure 4.

Example of ocular redness grading scales and classification according to severity.

Analytical Validation

The analytical validation of the ocular redness as a biomarker is dependent on consistency of image capture (illumination, white balance, magnification or image color resolution) and validation of the grading scales. The most common ocular redness scales, such as Efron, CCLRU, Vistakon and Annunziato have shown no statistically significant bias between the test/retest grading within observers (i.e. intra-grader agreement)^100,102.The inter-grader coefficient of repeatability ranges from 0.49 to 0.80 grading units,^102,104,105 which suggest that two graders should grade the same image within approximately 1.5 steps of another grader. Given that the scales are only typically 4 or 5 steps, there is a significant inter-grader variability with these scales. However, such scales have a high subjectivity and are prone to: 1) variations of photographic devices, such as illumination, white balance, magnification or image color resolution, 2) inter-grader variability and bias, such as judgment influence bias based on previous grading, and 3) inherit limitations of the grading scales, which have low discriminatory degree of severity (4 or 5 stages) to detect small incremental changes that occur in OR and show inconsistency with regards to the proportional severity between the representative image within the same scale or between different scales.^102,105,106 As such, the same grade in one scale might be perceived differently in another scale, not allowing comparison between studies with different scales.

One attempt to address these issues was with the Validated Bulbar Redness (VBR) grading scales that are based on the psychophysical attributes of graders during the validation process, by choosing representative images from a continuous grading scale ranging from 0-100 with equal steps of severity.¹⁰⁷ This scale showed an excellent intra-grader repeatability and inter-grader reproducibility for clinical assessment of ocular redness and has been validated against previous scales (Table 4).¹⁰⁷ Further, the Ocular Redness Index (ORI) performs a white-balance that corrects for lighting and color imbalance of slit-lamp photographs to provide an automated redness quantification of the selected region of interest.¹⁰⁸ This method has not only shown a high interobserver agreement and a strong correlation with subjective ocular redness and VBR scales, but also no significant disagreement or bias between graders. Amparo et al. compared the automated ORI to validated bulbar redness grading scales. This study showed that although both systems were capable to identify significant changes in mean redness scores, there was a significant level of disagreement in validated bulbar redness scores among experienced clinicians.¹⁰³

In summary, ocular redness grading scores have some limitations caused by image capturing, intergrader differences, and intragrader bias, based on previously assessed images (use of previous image as a reference), and low discriminatory degrees of severity. Automated systems may be useful to overcome these limitations.

Clinical Validation

Ocular redness provides a metric to record and monitor the natural course of ocular surface diseases. The subjective ocular redness scales have been the most commonly used and in a wide range of diseases, specifically to compare the adverse effects of different contact lens materials and solutions on the ocular surface,^102,109,110 as well as topical and surgical interventions for glaucoma,^111,112suggesting that it could be used as a safety biomarker. These scales have also been used to assess the effects of topical anti-allergic drops^113-115 and anti-inflammatory drops in various ocular surface diseases,^116-119 suggesting that it could be used as a monitoring biomarker. Recently, the ORI has been used to report conjunctival hyperemia changes in patients undergoing pterygium surgery, showing a potential to discriminate statistically significant ocular redness changes (26%) that were imperceptible to clinicians,¹⁰³ thus supporting the utilization of such a system.

Due to the numerous conditions that present with ocular redness, its utilization as a diagnostic biomarker seems improbable. In addition, evidence does not support the notion that ocular redness could be used as a predictive, susceptibility/risk, or pharmacodynamic biomarker. However, evidence does exist that ocular redness could be used as monitoring and safety biomarkers. Although the use of ocular redness grading systems is well established in ophthalmology and currently used for clinical trials, the more recent ORI measures have not reached the same level of evidence for clinical trials and require randomized and multi-center studies for further clinical validation. However, the ability of the ORI to detect smaller changes in ocular redness is promising and further work could elicit the utilization of ocular redness as a biomarker. Nevertheless, the main limitation that remains is the inability of this parameter to detect subclinical inflammation.

Cellular Quantification by In Vivo Confocal Microscopy

Although the normal cornea was thought to be immunologically privileged due to its lack of immune cells during homeostasis, the presence of resident immature dendritic cells and other resident corneal leukocytes has been clearly demonstrated by many independent groups.^120-127 These resident leukocytes can be visualized at the basal epithelial layer by IVCM as dendritiform cells (DCs; Fig. 5a). During inflammation, DC density increases along with elongation of their dendrites, which is associated with DC maturation (Fig. 5b-c).^121,122,128 Therefore, not only DC density, but also DC size, and DC field are potential parameters to evaluate the inflammatory state of the cornea.

Figure 5.

In vivo confocal microscopy images in dry eye disease: a) Healthy individual. b) Increased number of dendritiform cells in patients with dry eye disease (red arrows show representative cells). c) Increased number of presumed mature dendritiform cells in patient with dry eye disease (yellow arrows show representative cells).

Analytical Validation

IVCM is a non-invasive tool capable of microscopically imaging DCs of the ocular surface with a direct correlation to immunohistochemical DC observations.^123,129 This supports the accuracy of IVCM in imaging DCs. Further, it has been shown that the structural analyses of three representative central corneal images are comparable to wide-field composite images of the subbasal layer in chronic inflammatory conditions.¹³⁰ In healthy individuals, DCs are located mostly in the subbasal epithelial layer with a reported average density 9-64 cells/mm² in the central cornea and 0-208 cells/mm² in the peripheral cornea.^75,131-139 The assessment of the average DC density has shown intra-grader repeatability with low CoV (4.4-11.9%) and excellent inter-grader reproducibility (Table 4).^{74,134,135,137-139} However, increased variability was reported with increased DC densities.¹⁴⁰ A more recent study reported promising results for detection of DCs through artificial intelligence. Initial sensitivity rates for DC detection and DC segmentation were reported as 85% and 60%, respectively.¹⁴¹

Clinical Validation

An increase in mean corneal DC density above normal values is considered an indicator for active immune response or inflammation in the cornea.^137,142 Increase in DC density has been reported in DED,^{64,136,144-147} allergic keratoconjunctivitis,¹⁴⁸ infectious keratitis,^149,150 herpetic keratitis,¹³⁸ contact lens wear,¹⁵¹ post-LASIK cases,¹⁵¹ glaucoma medication use,^152-155 thyroid ophthalmopathy,¹⁵⁶ and systemic diseases.^157-161 It has been demonstrated that aqueous-deficient DED exhibits higher DC density compared to evaporative DED and that aqueous-deficient DED related to systemic immune diseases showed higher DC density compared to non-immune conditions.¹³⁶ In addition to DC density, DC size, and DC field were significantly increased in Sjögren’s DED compared to non-Sjögren’s DED.¹³⁶

It has been shown that there was a strong significant correlation of DC density with symptoms (OSDI) (r=0.37), Schirmer’s test (r= −0.47), TBUT (r= −0.61 to −0.25), corneal fluorescein staining (r= 0.48 to 0.53) and levels of various inflammatory tear cytokines (r= 0.31 to 0.55), which demonstrates its strong correlation with inflammation in DED patients.^{140,144,150,162-165} In a prospective, cross-sectional study it was further reported that at peripheral corneal quadrants DC density had different correlations with TBUT (r= −0.49 to −0.28), corneal fluorescein staining (r= 0.33 to 0.40), and conjunctival staining (r= 0.24 to 0.35).¹⁶⁶ Therefore, In addition to DC density, the morphology and distribution of DCs with their aforementioned correlations may further help establishing this parameter as a diagnostic biomarker for the DED subgroups.

From the perspective of a predictive biomarker, it has been shown that treatment response, based on change in OSDI and DC density, to topical steroids was correlated with baseline DC density and patients with increased DC density at baseline yielded a pronounced decrease in DC density with topical steroid treatment. Villani et al reported in a prospective, open label, masked study, that baseline DC density was correlated with OSDI and DC density changes in DED patients treated with topical steroid.¹⁴⁴ Moreover, a recent, prospective, randomized, placebo controlled, clinical trial confirmed that a DED population enriched with increased DCs for inclusion, yielded a pronounced decrease in DC density and significant improvement in signs and symptoms with topical steroid treatment.^163,167

Furthermore, DC density has also been shown a reliable tool in quantitative detection and monitoring of many ocular inflammatory diseases.^{149,150,168-173} In conditions such as DED, bacterial keratitis and post-LASIK ectasia correlation between proinflammatory cytokines (e.g. for bacterial keratitis IL-1β, IL-6, and IL-8, r= 0.40, r= 0.55, and r=0.31, respectively; p<0.002) and DC density was defined.^150,174,175 In the context of recurrent herpetic keratitis, DC density by IVCM has been shown to be a powerful tool to detect early signs of intracorneal inflammatory activity, which allows early detection and treatment, leading to improved prognosis.^170,176 Therefore, in inflammatory conditions, DC density may be helpful as a monitoring and pharmacodynamics biomarker.

Assessment of DC density through IVCM has been reported with extensive analytical validation, showing excellent repeatability and reproducibility. Recent studies on DED report the presence of subclinical immune cells in the ocular surface as a possible explanation for the symptom-sign disparity in DED and symptom refractory despite improvement of current clinical tests.^74,75,144 These results suggest that IVCM can detect early subclinical inflammation, allow treatment stratification, and measure therapeutic response according to baseline DC density as a predictive and/or response biomarker. Lastly, the DC density measurement has shown growing evidence as a diagnostic biomarker for the presence of inflammation in the ocular surface and is valuable both as a monitoring and pharmacodynamics biomarker to be used in clinical trials for various ocular surface diseases. Thus, it may serve as a useful monitoring biomarker for inflammatory conditions, such as in DED,^143,177 ocular allergies,¹⁴⁸ and contact lens wear, and pharmacodynamics biomarker for treatment response of those conditions.¹⁵¹

Palpebral Inflammation

Laser IVCM is a non-invasive technique that has been utilized to assess the immune cell density in IVCM images at epithelial and stromal levels (Fig. 2b).¹⁷⁸ Immune cell density is typically completed using semi-automated software that allows the cells to be marked and counted. That count is then typically standardized to cells per millimeter square.

Analytical Validation

Normative values for immune cell density have been reported as 123.3-123.7 (standard error = 17.2-19.2) cells/mm² for the epithelium and 36.7-38.8 (standard error = 9.5-10.2) cells/mm² for the stroma (Table 4).^74,75 The intra-grader repeatability showed excellent agreement for the epithelial immune cell (EIC) density and for the stromal immune cell (SIC) density (Table 4).⁷⁵ The reproducibility of the image analysis showed no statistically significant bias suggesting that the determination of the immune cell density does not vary considerably between graders.^74,75 There was also no significant difference between eyes.⁷⁵ This inter-grader consistency is supported by the ICCs for the epithelial density and for the stromal density,⁷⁴ and concordance correlation coefficient (Table 4).⁷⁵

Clinical Validation

The EIC density has been reported to be increased in MGD, with a cut-off of 195.8 cells/mm² which shows 94% sensitivity and 92% specificity suggest that EIC density could be used as a diagnostic biomarker to differentiate MGD from normal patients (Table 4).⁷⁵ In treatment studies there was a significant reduction in inflammatory cell density within the eyelid margin in groups treated with anti-inflammatory therapy compared to those without this treatment.^72,169

In summary, EIC density and SIC density detected by IVCM exhibit excellent intra/interobserver agreement with a high sensitivity and specificity, which makes it a good candidate as a diagnostic biomarker. Decreases in the inflammatory cell density by treatment suggest that they can be useful as a pharmacodynamics biomarker.

Assessment of Corneal Nerve Alterations by IVCM

The cornea is the most densely innervated tissue in the human body, and this high density of nerves is crucial for the protection of the eye through reflex blinking, maintenance of corneal epithelial integrity and regulation of wound healing.^179,180 Over the last years, IVCM has shown the capability of assessing the corneal subbasal nerve plexus (Fig. 6a) ^179,181 and the following total, trunk and branch nerve density and length (Fig. 6 b,c), tortuosity (Fig. 6d), beading (Fig. 6e), nerve reflectivity and thickness (Fig. 6f), and microneuromas (Fig. 6f) have been used to evaluate corneal nerve alterations.^182-185

Figure 6.

Subbasal nerves detected by IVCM. a) Subbasal nerves in a normal indiviual; b) Decreased main nerve density (pink traces) and decreased nerve branching (blue tracing) in herpes simplex keratitis; c) Absence of subbasal nerve plexus in neurotrophic keratitis secondary to herpes simplex keratitis; d) Increased nerve tortuosity in dry eye disease (two semicircles on the same nerve branch shows most representative areas); e) Increased beading in a patient with neuropathic corneal pain (most representative beading areas are bordered by blue stars); e, f) Hyperreflective abnormal bulging of corneal nerve ending, microneuromas in patient with neuropathic corneal pain (green arrow heads).

Analytical Validation

Total, trunk and branch nerve density and length

The normal range for nerve fiber length has been reported between 17.1-31.7 mm/mm² and 28.8-38.3 numbers/mm² for nerve density (Table 4).^{143,162,179,182,183,185-189} The range of nerve branching density was reported between 37.2-120.0 numbers/mm² in healthy individuals.^182,189 Nerve fiber length has been most extensively reported in the literature with moderate to excellent intra- and inter-grader ICCs (Table 4).^{182,183,189-191} In diabetes mellitus, the intra-grader repeatability was good for nerve density and excellent for nerve fiber length.¹⁸⁹ Also, they have shown no significant bias or trends with adequate CoV (<20%).^182,183 Software with automated image analysis for nerve fiber length is currently being used and shows a good inter-grader reproducibility (ICC = 0.870-0.890), however performed consistently lower measurements (−6.7mm/mm²) when compared to manual assessment.^182,183 However, it was reported that automated analysis might exclude nerve segments from slightly oblique images, images with mild pressure artifacts and short nerves with reduced contrast and might include some DCs as nerves,¹⁹² but still should provide compatible results with manual tracing.¹⁹² While data on nerve density are promising, additional validation for nerve density and branching are required.^{182,183,185,192} For instance, some research groups quantify nerve branches by the branching points, while other groups quantify according to the branch length.¹⁸⁵

Tortuosity

Nerve tortuosity is a morphological parameter of the corneal subbasal nerves that have been assessed by IVCM (Fig. 6d,e). The average normal value for tortuosity is reported as 1.1 grade (grade range 0-4) and tortuosity coefficient, a unitless measure that reflects the degree of curvature of nerve fibres, was reported as 20.0-38.2 in healthy individuals.^179,193,194 The intra-grader and inter-grader ICC values for tortuosity showed a wide difference between studies (Table 4).^182,190,194

Beading

Subbasal nerve beading frequency varied from 90-198 beads/mm in healthy individuals.^187,195,196 For beading (Fig. 6e), the intragrader ICCs were excellent- and inter-grader ICCs were good (Table 4).^182,190,194

Nerve Reflectivity and Thickness

The thickness of the subepithelial corneal nerves ranges between 0.52 and 4.68 μm^187,196,197 and reflectivity grade was 1.1-2.6 (grade range 0-4) in health.^{179,195,198,199}

Microneuromas

Microneuromas are currently described as stumps of severed nerves and abrupt, hyperreflective endings of nerve fibers on IVCM and they are evaluated for their presence or absence in different diseases, especially in neuropathic corneal pain.^200-202

Clinical Validation

The most frequent ocular conditions investigated for nerve density and length via IVCM have been DED,^64,143,147 herpes simplex keratitis (Fig. 6b),^203-205 herpes zoster ophthalmicus,^138,206 neurotrophic keratitis (Fig. 6c)²⁰⁷ neuropathic corneal pain and post-refractive surgery;^184,200,201 however, other anterior segment diseases, such as keratoconus,²⁰⁸ corneal dystrophies,¹³⁹ and vernal²⁰⁹ and atopic keratoconjunctivitis²¹⁰ and limbal stem cell deficiency^211,212 have also been studied. Furthermore, systemic diseases including endocrine disorders, such as diabetes^213,214 and thyroid diseases,²¹⁵ multiple sclerosis,^160,216,217 Parkinson disease,^216,218 small fiber neuropathy²¹⁹ graft versus host disease¹⁷¹ and mucous membrane pemphigoid¹⁵⁷ have also been investigated.

Total, trunk and branch nerve density and length

Most of the previous studies reported decreased corneal nerve density and length per frame in most ocular and systemic conditions.¹⁸⁵ The cut off values of nerve density and nerve length to differentiate normal controls from DED were determined as 36.5 nerves/mm² with 80.2% sensitivity and 85.0% specificity and 12.5 mm/mm² with 81.9% sensitivity and 85.0% specificity, respectively.²²⁰ In DED, nerve density and corneal sensation showed conflicting results where some reported negative and others positive correlations.^147,221,222 However, nerve density was negatively correlated with symptom severity^143,223 and fluorescein corneal staining and positively correlated with corneal sensation in DED.^221,224 Furthermore, DED patients with high nerve density responded to treatment better than the patients with low nerve density suggesting it can be used as a predictive biomarker for treatment stratification and likelihood of treatment response.¹⁶² Moreover, increased nerve density was reported as a pharmacodynamic biomarker for topical treatment in DED, such as cord blood serum and cyclosporine.^202,225,226

Similar to DED, in herpes simplex keratitis nerve density revealed positive correlation with corneal sensitivity.^185,203,227 Furthermore, it was shown that unilateral herpes simplex and herpes zoster keratitis results in decreased nerve density compared to controls, even in the contralateral non-affected eye.^206,228 Moreover, decreased corneal nerve length has been reported in patients who developed neurotrophic keratitis after neurosurgical trigeminal damage, where IVCM assessment was proposed for all patients with trigeminal impairment.²²⁹ The reported correlations herein support the use as a monitoring biomarker in DED, infectious keratitis and neurotrophic keratitis.

IVCM also showed nerve regeneration in DED, neuropathic corneal pain, herpes simplex keratitis, and neurotrophic keratitis with treatment, showing its potential as a pharmacodynamic biomarker.^{162,202,225,230-232} In contrast, diabetes mellitus showed decreased corneal nerve density, and its negative correlation with the severity of peripheral neuropathy^233-238 and retinopathy were established.^239-242 For the diagnosis of neuropathy in diabetes the sensitivity and specificity of nerve density were 82% and 52%, respectively.²³⁷ In addition to correlation with diabetic peripheral complications, it was showed that reduced corneal nerve density might be present before development of those complications.^243,244 And, the negative correlation between mean nerve density and HbA1c and improvement of nerve density after good glycemic control were also reported.^236,245-247 As a result of these reports, corneal nerve density in diabetes has shown to be an important predictive and monitoring biomarker.

Tortuosity

In DED, increased tortuosity is reported consistently in numerous studies;^195,196,223 Previous studies reported increased nerve tortuosity in different subtypes of DED, which was significantly different when compared to controls.^195,196,223 Nerve tortuosity was reported positively correlated with corneal sensation and negatively correlated with fluorescein staining in dry eye.²²⁴ Negative correlation (r= −0.179) between Schirmer’s I test and nerve tortuosity is also reported. ^{196,198,220,248} However, some studies showed no significant correlation of nerve tortuosity with signs and symptoms.^221,223,249 Therefore, it seems that nerve tortuosity does not seem promising as a diagnostic biomarker. However, decreased nerve tortuosity with treatment was also reported in dry eye, which suggests that tortuosity may be used as a pharmacodynamic biomarker.^{181,202,225,226}

In diabetes mellitus, increased nerve tortuosity and its correlation with the severity of peripheral neuropathy and retinopathy was established.^{233-237,239,240} Different definitions of tortuosity have been used to develop some qualitative and quantitative grading scales; however, previously used grading scales still need improvement because of their subjectivity and unfeasibility for clinical use and insufficient assessment of its complex mathematical definition.^179,233 Annunziata et al developed a promising, fully automated, hybrid segmentation system to assess tortuosity objectively and definition free, which might increase our knowledge about nerve tortuosity and its correlation with clinical symptoms and signs.²⁵⁰ This fully automated nerve tortuosity quantification showed a sensitivity of 51.9 - 58.8% and specificity of 81.1 - 84.7%.²⁵⁰

Beading

In DED, increased beading is reported consistently in numerous studies;^195,196,223 Number of beading were reported as 332-387/mm in Sjogren related DED and 186-323/mm and in non-Sjogren related dry eye.^195,223,224 Decreased nerve beading with topical chord serum tears and cyclosporine treatment was also reported in DED.^{181,202,225,226} Thus literature about nerve beading and its change with treatment is limited. For being a biomarker it still needs to be investigated.

Reflectivity and Thickness

Nerve reflectivity and thickness did not show consistent results.^143,185 Correlation between nerve reflectivity and OSDI¹⁴³ and decrease in nerve reflectivity with topical cyclosporine treatment was also reported.²²⁵

Microneuromas

Microneuromas have been proposed as a tool to assist in the diagnosis of neuropathic corneal pain patients.^{200,201,251,252} Microneuromas are known to be the source of ectopic spontaneous excitations in sensory fibers in patients with postsurgical pain.²⁵³ In 2015, Aggarwal et al. first described microneuromas as stumps of severed nerves and abrupt endings of nerve fibers on IVCM and reported their presence in photoallodynia patients with, which decreased with treatment of autologous serum tears.²⁰⁰ In recent study, Aggarwal et al, showed that in neuropathic corneal pain patients with different etiologies, such as post-refractive surgery, DED, autonomic neuropathy, herpes zoster ophthalmicus and UV light exposure showed presence of neuromas in 100% of the patients and decreased to 6.25% with treatment.²⁰¹ Additionally, a more recent study showed that microneuroma presence had a 100% sensitivity and 100% specificity in differentiating DED and NCP and warrants additional validation studies in larger cohorts.²⁵² Due to the lack of validated biomarkers in this ill-defined disease, microneuromas detected by IVCM are currently being validated, since current studies suggest them as a diagnostic and/or monitoring biomarker in the near future.^{200,201,251,254}

In summary, the nerve fiber density (total, trunk and branch) and length have been the most validated parameters to date. Other nerve parameters such as nerve tortuosity and beading require further validation and standardization of method due to their variability and subjectivity, whereas nerve reflectivity, nerve thickness require full analytical validation. In addition, the clinical value of these morphological findings is yet to be established, the exception are microneuromas that have shown to be highly prevalent and a potential biomarker in neuropathic corneal pain patients. In this context, nerve density and length are unlikely to be a diagnostic biomarker when used in isolation, but they can become a useful non-invasive biomarker to monitor the level of nerve damage and regeneration.¹⁸⁶ To date, clinical use of these biomarkers, most importantly in diabetic neuropathy, is focused on demonstrating severity of disease, on endpoints for treatment response, on monitoring and on predicting disease progress.

Corneal Thickness and Epithelial Thickness

Anterior segment OCT provides high-quality cross-sectional images based on the reflectivity of ocular structures through low-coherence interferometry. By performing high resolution anterior segment OCT scans on 8 meridians of the cornea, an automated software can reconstruct the cornea 3-dimensionally, generating 10-mm radius pachymetry; the latter can be further segmented to generate epithelial thickness maps by identifying the hypereflective basement membrane and surface of the epithelium (Fig. 7).²⁵⁵

Figure 7.

Anterior segment Optical Coherence Tomography (AS-OCT) corneal thickness and corneal epithelial thickness maps in a healthy eye. a) Reference infrared image and direction of selected scan (arrow). b) AS-OCT line of the cornea. c) Pachymetry and epithelial thickness color-coded maps.

Analytical Validation

The reported mean normal epithelial thickness is 50.4-53.4 μm.^256-260 When comparing ultra-high resolution ultrasound and anterior segment OCT for measuring epithelial thickness, they was close inter-grader agreement with an average measurement difference of 4 microns and 95% limits of agreement between −4.6μm and +3.2μm.²⁶¹ Epithelial thickness as measured by AS-OCT showed high repeatability for central and peripheral cornea and excellent reproducibility (Table 4).^{256,259,260,262,263}

Clinical Validation

Anterior segment OCT epithelial thickness evaluation has been utilized as a diagnostic biomarker for ectasias, DED, and limbal stem cell deficiency. In cases of keratoconus, the epithelial thickness is thinner in areas overlying areas of corneal steepening^256,257 and showed a high sensitivity of 88.9% and specificity of 59.5%, with an optimal cutoff of 52 μm at the thinnest area of the cornea, in early disease (form fruste) due to the potential epithelial changes that occur prior to topographical changes.^258,264 More specifically, anterior segment OCT epithelial thickness showed high accuracy (AUC= 0.840 to 0.985) in detecting keratoconus even in form fruste cases.^256,265,266

In keratoconus patients, the corneal epithelial thickness ranged from 40.0 μm to 51.9 μm and its parameters such as central, inferior, minimal-maximum and standard deviation showed various AUC values, however the pattern standard deviation showed 100% sensitivity and specificity in detecting keratoconus (Table 4).^256,264-266 Furthermore, the pattern standard deviation of epithelial thickness maps increased the accuracy in diagnosing subclinical keratoconus (AUC= 0.985), higher than pachymetry map-based keratoconus risk score (AUC= 0.735);²⁶⁵ even in subclinical disease with normal pachymetry-based keratoconus percentage index (KISA%) values the diagnostic power was high (AUC= 0.961).

In DED^267,268 and limbal stem cell deficiency^269-271 patients have shown epithelial thickness thinning as measured by AS-OCT at the limbus and central cornea compared to controls. In limbal stem cell deficiency the corneal epithelial thickness correlated with visualization of palisades of Vogt (r= 0.82-0.89).^269,271

The AS-OCT epithelial thickness has shown sufficient analytical validation with good repeatability and reproducibility even in diseases patients. In addition, its non-invasive and non-contact nature makes this a promising biomarker particularly for the diagnosis of keratoconus using the pattern standard deviation and should be considered for future clinical trials for early diagnosis and pre-operative screening for keratoconus. Therefore, anterior segment OCT epithelial thickness has the potential to be a good diagnostic and monitoring biomarker.

Qualitative Biomarkers

IVCM allows direct visualization of infectious pathogens and cellular changes of cornea in patients with keratitis, and provides comparable results with corneal cultures and PCR.²⁷² Therefore, they may be thought as diagnostic and pharmacodynamics biomarkers in demodex folliculorum blepharitis and infectious keratitis. These are different than quantitative measure described above, therefore outlined in a separate section:

Demodex Folliculorum Blepharitis

Demodex folliculorum and D. brevis are microscopic parasites that can infest the eyelashes and are a common cause of blepharitis. A pathognomonic finding on slit-lamp examination is collaret formation or crusts on the eyelashes, and diagnosis requires eyelash sampling with visualization of the parasites under the microscope.²⁷³ However, performing IVCM on the eye lids/eyelash can also provide diagnosis for Demodex blepharitis since these parasites are directly seen on the eyelash and confirmed by analyzing depilated eyelashes under light microscope.^274,275 The mites are generally imaged with longitudinal section. D. brevis and D. follicularis are seen as structures with hyperreflective legs around 250 μm and 350 μm length, respectively in longitudinal sections (Fig. 8a-b).^274,275 It is possible to show these parasites inside the follicle, inside the follicle next to MG, between two eyelashes and at the bottom of a follicle with accompanying follicle distension, epithelium hyperreflectivity, MG obstruction and reactional epithelial proliferation.²⁷⁴ A prospective study has shown that sensitivity and specificity of IVCM to detect Demodex infestation based on the definition of light microscopy positive demodex were 100% and 98.8% when performed and analyzed by experienced observers.²⁷⁶ Therefore, IVCM can be used in the diagnosis of ocular Demodex infestation and related structural changes. Therefore, it can be thought as a diagnostic biomarker.

Figure 8.

In vivo confocal microscopy images in infectious keratitis. a) Ovoid hyperreflective structure at the base of eyelash follicle of Demodex follicularis (red arrows) and b) Demodex brevis (yellow arrows). c) Double walled cysts (blue arrow heads) with hyperreflective center and hyporeflective surrounding halo and hyperreflective ovoid trophozoites in Acanthomoeba keratitis; d) Hyperreflective septated thin filamentous structures at anterior stroma in filamentous fungal keratitis (orange arrow heads).

Acanthamoeba Keratitis

Both cysts and trophozoites of Acanthamoeba can be visualized by IVCM (Fig. 8c). Cysts are seen highly reflective double-walled round structures in 10-25 μm diameter. Trophozoites appear as hyperreflective ovoid or irregular structures with pseudopods measuring 25-40 μm.^272,277,278 The accuracy to detect on IVCM with experienced observers on culture positive or light microscopy positive ulcers was 94.8%.²⁷⁹ The sensitivity and specificity of IVCM to diagnose Acanthamoeba are reported 69.7-100% and 84.0-100%, respectively.^278,280 Thus, IVCM can provide fast and definite diagnosis for Acanthomoeba keratitis and may serve as a diagnostic biomarker for this disease.

Fungal Keratitis

Aspergillus sp and Fusarium sp are the most common pathogens of fungal keratitis. Aspergillus spp was described as 200-400 μm hyperreflective structures with 5-10 μm septate hyphae with 45° dichotomous branches (Fig. 8d); conversely, Fusarium spp was described with branching at 90° in most of the IVCM studies on infectious keratitis.^281,282 However, in a recent study, the mean branching angle was measured as 59.7° and 63.3° for Fusarium spp and Aspergillus spp, respectively and the same study reported that dichotomous branching can be rarely seen with IVCM.²⁸³ Moreover, previous studies reported that IVCM is not only a diagnostic tool for infectious keratitis (accuracy of 94.8%) but also useful and valuable to follow treatment response and early detection of relapse in severe cases.^279,284,285 The reported sensitivity and specificity of IVCM for fungus filaments detection were 71.4-94% and 78-81.4%, respectively.^{278,279,281,282}

In summary, IVCM can serve as a diagnostic biomarker for fungal keratitis. Moreover, when it is applied to assess treatment response, it may be helpful as a monitoring biomarker.

Summary

This review assessed the current state of NITBUT, TMH, TMA, MG dropout, ocular redness grading scores, inflammatory cell density, DC density, nerve density, and epithelial thickness to as potential optical biomarkers in the scope of analytical and clinical validation. To date, none of the parameters can be described as a certain biomarker based on biomarker development processes shown in Table 1 and 2. However, NITBUT, TMH and TMA may become diagnostic biomarkers for DED once analytical and clinical validation are completed. Furthermore, DC density and palpebral immune cell density reflect the state of ocular surface inflammation with a good sensitivity and specificity. Although they cannot differentiate a specific disease, ocular inflammation can be diagnosed by these parameters and treatment response to anti-inflammatory theraies can be evaluated, highlighting their potential to additionally become monitoring and pharmacodynamics biomarkers. Like inflammatory cell density, nerve density may not address a specific disease but can be a promising candidate for being a monitoring or pharmacodynamic biomarker in ocular (DED, neurotrophic keratitis, neuropathic corneal pain etc.) and systemic conditions, such as in diabetes. Lastly, epithelial thickness measured by AS-OCT has a high potential to be a diagnostic and monitoring biomarker for keratoconus patients.

Future Directions

The imaging parameters discussed herein have been summarized in Table 3 and are new or emerging tools that may provide potential biomarkers for ocular surface diseases. There is a strong need for analytical or further clinical validation to provide the necessary evidence (or fit of purpose) to qualify these imaging parameters for clinical use. The NIBUT has had extensive analytical validation supported by many groups. This has provided clinicians with an understanding of its limitations, which are lack of reproducibility across devices and larger variation at higher values. In addition, NIBUT has shown diagnostic potential via clinical validation. More extensive clinical validation is necessary if NIBUT is to be used as any other biomarker type. These additional clinical validations are likely to be pursued if NIBUT is used more widely within the clinic. To date, most studies appear to favor the traditional fluorescein TBUT.

The ocular redness grading scales have been extensively validated and implemented in clinical practice. However, most of the recent, automated, more reliable and continuous grading methods presented herein have not reached the same level of evidence for clinical trials as previous illustrative scales; in part due to the high costs of these high-resolution/specialized devices or computer software development.¹⁰⁸ New methods to quantify ocular redness or the use of automated analysis, such as ORI, are warranted with the latter showing promising results that require further clinical validation. Future clinical studies to address this issue ideally should present a practical approach with an objective continuous scale, low cost, high repeatability and reliability, high sensitivity to detect small incremental changes and should be comparable to previous studies.

Moreover, IVCM has emerged as an important non-invasive tool for ocular surface diseases. More specifically, the microscopic visualization of DCs and subbasal nerve plexus of the cornea has provided a novel approach to detect inflammation and nerve abnormalities of the ocular surface that could be secondary to systemic diseases such as in diabetes mellitus. Therefore, as highlighted in this review, the biomarkers with the highest level of evidence to date are DC density to diagnose and monitor inflammation of the ocular surface, particularly in DED, and total nerve density to monitor severity or progression for diabetes and neurotrophic keratitis

The use of artificial intelligence in ophthalmology is rapidly growing and might greatly impact imaging analysis in the near future.²⁸⁶ Most recently, its use on IVCM image analysis has been shown to be capable of detecting DCs and micro-neuromas in a fast and automated process with high sensitivity (66.0% and 91.3%, respectively) and specificity (99.0% and 90.0%, respectively), further facilitating its implementation as a complimentary tool for ophthalmologists in research and clinical settings.^141,287 There is growing evidence of the neuro-immune crosstalk that occurs in the cornea, highlighting the importance of these non-invasive biomarkers for monitoring in ocular surface disease.^140,137

Overall, the epithelial thickness maps as measured by anterior segment OCT is a novel reliable parameter to differentiate diseased corneas from healthy corneas.²⁶³ Since many ocular surface diseases affect the corneal epithelium, either directly or indirectly, assessing changes to this layer can be a promising monitoring or predictive biomarker for conditions such as keratoconus. Nevertheless, studies confirming how the degree of inflammation interferes with the epithelial thickness are warranted, as the former can be a confounding factor when establishing diagnostic thickness cut-off values. Furthermore, advances in OCT technology with novel applications for anterior segment are emerging. In particular, en face anterior segment OCT and anterior segment OCT angiography have been explored in ocular surface diseases and are capable of imaging corneal neovascularization and conjunctival-limbal vessels.^5,288,289 For example, in limbal stem cell deficiency anterior segment OCT angiography has shown an increase in corneal vessel extension and depth in disease correlating to disease severity.²⁹⁰ In conjunctival tumors, the anterior segment OCT lesion features as well as its vascularization on anterior segment OCT angiography were used to differentiate benign from malignant tumors.²⁹¹ The advantage of non-invasively providing quantitative parameters of the ocular surface blood flow makes anterior segment OCT angiography a promising tool for diagnosis and monitoring.