Fluorescein CorneoGraphy (FCG): Use of a Repurposed Fluorescein Imaging Technique to Objectively Standardize Corneal Staining

Abstract

Punctate epithelial erosion (PEE) is a corneal sign of dry eye disease (DED), which is observable via staining with fluorescein on slit lamp as a standard of clinical practice and clinical research. There are currently many limitations for detecting PEE, such as lack of optimal excitation with cobalt blue light, difficulty to focus on the whole cornea, observer dependency and no available automatized quantification. We propose to reduce these limitations through repurposing the use of fluorescein angiography mode of optical coherence tomography (OCT, Heidelberg Spectralis II, Germany), as “fluorescein corneography” (FCG) for imaging PEE. A standard methodology was created using 50 patients with corneal staining and 10 healthy volunteers who were imaged on both slit lamp and FCG on two centers. Subsequently, two masked cornea specialists graded slit lamp and FCG images of 15 patients using the National Eye Institute (NEI) scale.

FCG showed both a higher interobserver agreement (IOA), and a higher intraclass correlation coefficient (ICC) than slit lamp (0.96 vs 0.86, p<0.001). Light- colored iris patients showed a statistically significant lower mean of epitheliopathy on slit lamp compared to FCG (6.11 vs 8.94; p=0.026), which was not the case with dark-colored iris patients (8.16 vs 8.25; p=0.961)

In conclusion we present an OCT-FA system for robust detection of PEE which has major implications in both clinical practice and research endpoints since it is highly sensitive, rigorous, reproducible in different facilities and offers potential for a numerical quantification and automatization of dry-eye corneal staining

Dry eye disease (DED) is an ocular surface and tear film disorder that commonly causes punctate epithelial erosions (PEE) as one of the end results of the syndrome. The corneal damage is commonly documented via fluorescent staining of its damaged epithelium, which becomes visible at the slit lamp after instilling a fluorescein drop and exciting the dye with the use of a cobalt-blue filter permitting visualization of PEE. This method has been the mainstay for evaluation of epithelial lesions on the slit lamp since 1882, when discontinuity in the corneal epithelium was first identified by using sodium fluorescein.¹ Not only is this epitheliopathy a sign for assessment of the vitality of the corneal epithelium on clinical practice, but importantly this sign is used as a major endpoint in clinical trials that test drugs for DED as well.² Despite the relevance of corneal staining, there are significant limitations in present day in both image acquisition with fluorescein and corneal staining grading³.

Capturing corneal staining with the use of a cobalt-blue filter poses several limitations, related to the excitation-emission properties of fluorescein molecule, which has a maximal absorption at approximately 490nm and maximum emission at approximately 525nm⁴. Meanwhile, the cobalt-blue filter has peak transmission of 390-410nm wavelength, providing suboptimal excitation of fluorescein. Additionally, the lack of an emission filter to remove reflected light of other wavelengths, particularly in the blue spectrum, results in poor visualization of fluorescein staining. These issues are particularly problematic when attempting to perform quantitative analysis of PEE, such as during a clinical trial.

Significant challenges in grading of PEE are that these are usually manually scored with discrete ordinal variables, which can skew the results towards lesser or greater staining scores, as opposed to using a more precise continuous scale. To overcome this, digital photographs of corneal fluorescein staining ought to be obtained in a standardized fashion that allows softwares to quantitate the area of interest. Currently, the methods of image acquisition for PEE are usually via direct visualization of the epitheliopathy under the slit lamp or with a colored digital photography, both of which are usually manually graded by an observer. Limitations associated with these are: 1) A difficulty to delineate the corneal staining areas, as the color of the iris (i.e.: blue iris), background, and use of different filters may hinder and confound the grader’s stain interpretation; 2) The corneal curvature represents a topographic obstacle since slit lamp pictures images fail to focus on both the central and peripheral epithelium because of the limited depth of focus when imaging a curved surface under magnification; 3) Common commercially available blue exciting filters measure approximately 8 mm in diameter, which is less than the entire corneal diameter and therefore cannot capture the totality of the epithelial anatomy. The ideal corneal staining image extraction should be sensitive, illuminate the entire cornea, virtually documented, easy to reproduce and have increased contrast to facilitate computer automated analysis.

In order to overcome these limitations, we repurposed the optical coherence tomograph (OCT, Heidelberg Spectralis II OCT, Heidelberg, Germany) associated fluorescein angiography (FA) function, commonly intended for retinal imaging, in order to capture the corneal epithelium staining patterns. The Spectralis imaging system has several advantages over typical imaging with a cobalt-blue filter. First, use of a 490nm excitation laser, which is imaged via scanning laser ophthalmoscopy (SLO) that enables superior excitation of fluorescein and more detailed imaging of the corneal surface. Second, the device uses an appropriate barrier filter to enable specific imaging of fluorescein emission around 525nm. Here, we present a standard operative procedure utilizing SLO optimized for fluorescein imaging to visualize corneal staining, which we term Fluorescein CorneoGraphy (FCG). In this proof-of-concept study, we describe the methodology for FCG to test the ability to image corneal epithelial erosions secondary to DED in comparison to slit lamp images and measure interobserver agreement between cornea specialists at two different clinical sites.

Methods

This study was approved by the Institutional Review Boards (IRB) of Duke University School of Medicine and the University of Miami Miller School of Medicine (Protocol Numbers: 00095121 and 20200432, respectively). The protocols conformed to the requirements of the United States Health Insurance Portability and Accountability Act, and the tenets of the Declaration of Helsinki. All study patients were seen at one of the academic centers: Duke Eye Center (Durham, NC) and Bascom Palmer Eye Institute (Miami, FL). Patients with ocular surface disorders with different degrees of corneal staining and healthy volunteers (total n=60) were recruited for corneal imaging by FCG and imaged using both slit lamp and FCG. These images included the evaluation of endogenous corneal autofluorescence, the observation of timings after fluorescein instillation for optimal corneal staining detection and the protocol was defined as such. We then conducted a second study on which two masked cornea specialists graded slit lamp and FCG images and we calculated the interobserver agreement (IOA).

Imaging protocol for FCG

Imaging was effectuated through using the Spectralis II OCT (OCT, Heidelberg Spectralis II OCT, Heidelberg, Germany) in FA mode to detect corneal staining. First, the inferior lid of the imaged eye was gently pulled down, and 2 microliters of 0.25% fluorescein sodium (Altaire pharmaceuticals) were instilled on patient’s lower tarsal conjunctiva using a 2-20 microliter micropipette and its tip, that was discarded after the procedure. One minute after fluorescein instillation, an ocular wash using sterile PBS was performed using the micropipette. For that, the patient was asked to incline the head to the contralateral side of the eye of interest, and 300 microliters of sterile PBS were instilled through a 100-1000 microliter micropipette, aiming at the ocular surface from the lateral sides in order to reduce fluorescein pooling and debris/tear film on the surface. The same technique was then replicated to the other eye.

The images were taken using the FA mode. To obtain a complete corneal area visualization from “limbus to limbus”, the 55° lens was used. The image frame was focused on a midpoint between the lacrimal caruncle and the corneal limbus in order to obtain a thorough view of the corneal epithelium. After that, the distance knob was locked so that all images taken for a patient were standardized and consistent in terms of distance. The lens was focused on the corneal epithelium in such a way that both the central and peripheral epithelium were visualized in clear focus and then, consecutive images were taken of the whole cornea. The sensitivity knob was kept constant during imaging, with minute compensation for maximal clarity (Figure 1).

Figure 1.

OCT Spectralis II (Heidelberg, Germany) elements diagram

Imaging protocol for slit lamp photographs

After FCG imaging, digital corneal photography was performed using a slit-lamp (Haag-Streit BX 900) with a digital camera attachment (Canon 7D Mark II) and a flash-through-the-slit illumination system. The photographer captured the entire cornea using diffuse illumination, 10x magnification and a flash intensity to obtain a clear image. First, a regular image of the iris color with diffuse lighting was taken. After that, a second image was done using the cobalt-blue filter, and finally a yellow-filter is applied in front of the camera lens as it acts as a barrier filter at approximately 525 nm and permits easier visibility of the staining. Flash intensity was increased slightly to compensate for a darkening field of view, and the optimal image is captured accordingly.

Interobserver Grading and Agreement

In order to evaluate the FCG imaging sensitivity as compared to slit lamp photographs, two blinded cornea specialists (A.L.S and V.L.P) were instructed to grade slit-lamp and FCG images of 15 patients using the National Eye Institute (NEI) scale for corneal staining⁵, with corneal scores ranging from 0 to 15.

The grading was done on two occasions, separated by a total of seven days, where the FCG and slit-lamp images of each patient were alternated between the two occasions (i.e.: FCG of patient 1 on occasion 1, Slit-lamp of patient 1 on occasion 2). Interobserver agreement (IOA) was calculated for the total corneal staining (the sum of all the quadrants) between both observes for slit lamp and FG images using Intraclass Correlation Coefficient (ICC), measured using analysis of variance methods for unbalanced data. The level of statistical significance was set at P>0.05. All statistical analysis were performed using SPSS Statistics version 25.0 (IBM Corp, Armonk, NY).

Results

FCG technique optimization: Focal distance and Lens selection

A total of 60 subjects (50 individuals with dry eye and 10 normal controls) were imaged in order to optimize the methodology of the FCG image extraction. Mean age was 55.6 +/− 20.6 years old, and 41/60 (68.3%) of patients were female.

The field of view in conjunction with the selection of the right lens was the first studied parameter. For that, different positions of the camera were trialed, which included a close, middle, and far field of view. A distance that balances optimal resolution and maximal field of view was possible when positioning a midpoint between the lacrimal caruncle and the corneal limbus on the borders of the image frame (Figure 2), which was selected as the standardized distance.

Figure 2. Distance optimization for FCG.

(A) Close-up view: borders of image frame coincide with corneal limbus. (B) Middle and optimal view: borders of image frame correspond to midpoint between lacrimal caruncle and corneal limbus. (C) Farther view: image frame encompasses totality of eye and eye margins

Additionally, three lenses were tested as possible objectives. We tested the 35° lens, the 55° lens and the “anterior segment” lens as an objective for image visualization. The 55° lens was selected and was the basis of pictures taken in this paper, as it offered the best focus on the whole epithelium without compromising a complete corneal field of view. Images with the 35° lens presented less resolution in comparison to the 55° lens when zoomed in to the adequate distance. When the anterior segment lens was used, although images were encompassing of the full corneal limbus, they were distorted and had an applanated field of view (Figure 3)

Figure 3. Examples of the different lenses trialed for FCG.

A) Anterior segment lens; B) 35° lens.

FCG Technique Optimization: Timings and Absence of Corneal Autofluorescence

First, we assessed for the presence of corneal auto-fluorescence by subjecting the different subjects involved to images with the OCT-FA filters before incorporation of fluorescein dye into their eyes. This was derived from the methodology used for initial implementation of retinal angiographies, in order to showcase that our observations of staining on further images were solely derived from the incorporated fluorescence, and not from endogenous autofluorescence. As seen in figure 4, we observed that in all the 60 patients imaged, there was no exhibition of endogenous corneal autofluorescence prior to instilling the drop. This prompted removal of autofluorescence sequencing in our imaging protocol for further images.

Figure 4. Autofluorescence and timed FCG images.

A) Image taken prior to instilling fluorescein for detection of autofluorescence, which was negative. B) After instillation of fluorescein, images were taken at 1, 3 and 5 minutes respectively

Second, we aimed to obtain optimal timings for image acquisition. Timed FCG protocols were performed, whereby the first image was taken with the FA mode of the Spectralis II OCT without the use of previous fluorescein drops. After this, fluorescein was instilled as described in the methodology, and timed images were taken at the 1^st, 3^rd, 5^th, and 7^th minute. Washing of the surface was done at minute 1 as described in the methods section. The timing for optimal corneal staining observation was assessed with focus on the presence of corneal staining (“signal”) and its correlation with the background staining, tear debris and tear evaporation (“noise”). The data demonstrated that images showed similar pattern staining and background confounders between different timepoints, making it possible to image the patients directly after the wash without timing further images (Figure 4).

Comparison of slit lamp with FCG images

Two masked cornea specialists quantified the corneal staining of 15 DED subjects with the described FCG and slit lamp image methodologies. The population mean age was 62 +/− 6 years old and 10/15 (66%) were female.

Comparison of slit-lamp images with yellow filter and FCG images showed a pattern of discrepancy in the detection of corneal staining according to the iris color: in fact, while FCG images showed similar staining to the ground truth regardless of iris color, slit-lamp images of light-colored irises obscured the visualization of corneal staining, which was not the case with dark colored irises as demonstrated in Figure 5.

Figure 5. Image strips of A) patient with dark-colored iris and B) patient with light-colored iris.

The images are arranged in the following order: regular slit-lamp image, slit-lamp with blue-light visualization, slit-lamp with blue-light visualization and yellow filter, FCG imaging.

Inter observer agreement (IOA) of slit lamp images versus FCG

We conducted an IOA agreement on the slit images of 15 subjects with differing degrees of epitheliopathy (Figure 6). Of these, 8 had a light-colored iris and 7 had a dark-colored iris. Mean ± SD total NEI corneal staining for slit lamp for observers 1 and 2 were 7.7 ± 3.9 and 6.2 ± 3.2. FCG gradings were 8.9 ± 4.2 and 8.4 ± 4.2 respectively. FCG grading had higher corneal scores in 23/30 (77%) of the combined cases. On another hand, sub analysis of iris color showed that, in light colored iris patients, there was a statistically significant lower mean on slit lamp images as compared to FCG (6.11 vs 8.94; p=0.026), explaining the discrepancy. Results for dark-colored irises showed no significant differences in score grading between slit-lamp and FCG (8.16 vs 8.25; p=0.961). Total corneal staining ICC was 0.86 (p<0.001) for slit lamp images and 0.96 (p<0.001 for FCG.

Figure 6. Image strips of 6 different representative patients.

This are 6 representative images of the 15 that were analyzed for interobserver agreement comparing slit-lamp imaging with blue-light visualization and yellow filter to FCG imaging

Discussion

In this proof-of-concept study, we present the use of FCG, a practical and robust repurposed OCT-FA imaging technique for the detection of epitheliopathy in eyes with different corneal staining severities. The benefits of this method include the ability to focus on all the regions of the cornea in one image, high contrast visualization and higher sensitivity that slit lamp images. As per our preliminary results, we have demonstrated that 1) FCG images detected corneal staining more frequently that slit lamp images and present higher IOA for total corneal staining; 2) FCG detection of staining is irrespective of the color of the iris, 3) This technology is repeatable at a different center.

The basics of fluorescein angiography were studied by Harold and Novotny who developed the FA methodology for retinal vessel imaging⁶. The sodium fluorescein molecule contained within this dye emits green light at a wavelength of around 525 nm when stimulated with blue light that usually ranges between 465 and 490 nm. As such, two filters are needed for FA: a blue exciter filter to create the fluorescence and a green barrier filter to obtain only green fluorescence and to absorb the reflected blue light⁶. There have been minimal reports of corneal epithelium images with fluorescein angiography techniques^7,8 although these typically use retinographs with adapted contrasted fluorescein like filters which are uncommon in private practice centers and therefore unpractical to adopt and escalate. A novelty of our report is the use of OCT-FA spectralis II, a widely commercially available device for corneal staining detection which has the mentioned filters in its hardware. We have also reevaluated and standardized the fluorescein instillation parameters, the camera distance, the correct lens, the fluorescein timings, and its washing for optimal signal to noise ratio detection of corneal staining. Moreover, we have developed a standardized and practical approach that has been successfully repeated at a different center, which indicates a possibility for escalating the technology.

In this study, we conducted a comparison of slit lamp electronic images with FCG. Our data demonstrates that the observers on FCG noted higher corneal staining than slit lamp image grading in most cases. This suggests a higher sensitivity for detection of corneal erosions, presumably due to the FCG ability to focus and illuminate all the different corneal topographies, unlike a routine slit lamp image. Additionally, a major limitation was encountered with the slit lamp images for patients with light iris eyes, on whom the blue light emission confounded the light-colored iris with the visualized fluorescein-stained PEE and hence, decreased the detection of PEE, unlike the FCG, which was not dependent on the color of the iris.

The FCG technology demonstrated a higher IOA, with an excellent ICC of 0.96, as opposed to the slit lamp images ICC of 0.86, which demonstrates a lower yet solid reliability. On a study with 122 slit lamp images graded by 11 observers using the NEI scale, the ICC was 0.991⁹, which was slightly above our FCG ICC, and below our slit lamp ICC. The minor difference might be due to a larger number of patients (122 vs 15) seen with a higher number of experienced observers (11 vs 2), which decreases the overall variance in grading and increased the ICC reliability. For example, on that same study the observers analyzed as well the Ocular Staining Score in the Sjögren’s International Clinical Collaborative Alliance (SICCA) corneal grading, which had an ICC of 0.981⁹. Yet, on a different study of 49 patients graded by 5 ophthalmologists on the slit lamp, the SICCA score resulted in a corneal staining ICC of 0.90-0.91¹⁰, which is inferior to the previous SICCA results and to our FCG reliability, suggesting that the numbers of graders, patients and methodology might impact the final ICC, although in all the cases these were considered excellent, including our FCG reliability scores. Larger studies detecting the same patient on different visits, with increased number of observers, are warranted for a more precise reliability calculation.

As detailed in the methods, many challenges were encountered and later fine-tuned while developing the FCG standard operative procedure. A current limitation involves patient compliance and collaboration for a perfectly centered image. This can impact the possibility of acquiring a still image with a complete field of view of the corneal epithelium, altering total corneal grading. The OCT device has an external fixating light attached which aids with patients’ centration, while a second assistant is useful to obtain a lid aperture, such that both the inferior and superior corneal limbus are completely observed. Equally, although ocular washing appears as an arduous process to do on a day-to-day clinical basis, no patient in the study felt pain or complained about the process. Despite these limitations, we have demonstrated that this technology is reproducible, scalable and practical. Further evaluation and optimization of the FCG should be done in a standardized methodology to use it with real-life patient encounters.

The implications of FCG adoption are substantial for the ophthalmologic practice and for clinical research protocols, since OCT devices with angiography functions are widely available in both academic and private practice centers, rendering the device accessible for use and for recruitment of patients. Overall, FCG imaging can serve as a more reliable and robust tool to follow up corneal epitheliopathy changes. It also allows the possibility for an automatized software that bypasses human biases and assists in unriddling the commonly found miscorrelation between signs and symptoms in dry eye disease.^11,12 Significantly, this technology can be replaced and included in studies that evaluate corneal staining in other contexts, such as ocular dryness in the setting of refractive surgery, cataract surgery, related to oculoplastic disorders and others.

In summary, fluorescent staining of the cornea is the mainstay for assessment of the vitality of the epithelium in ophthalmology practice and for clinical research purposes. Despite this, there is lack of agreement for extracting, grading, and quantifying this biomarker. This is partly due to the image acquisition protocols which rely on slit lamp images that are prone to illumination and sensitivity biases. We have developed a repurposed standardized methodology using an OCT-FA system for robust detection of PEE, coined FCG. This technique has major implications in both clinical practice and research endpoints since it is highly sensitive, rigorous, reproducible in different facilities, commercially available, safe and offers potential for a numerical quantification and automatization of dry-eye corneal staining.