Validation of a Novel Confocal Microscopy Imaging Protocol With Assessment of Reproducibility and Comparison of Nerve Metrics in Dry Eye Disease Compared With Controls

Abstract

Purpose:

The purposes of this study were to assess the reproducibility of a novel standardized technique for capturing corneal subbasal nerve plexus images with in vivo corneal confocal microscopy and to compare nerve metrics captured with this method in participants with dry eye and control participants.

Methods:

Cases and controls were recruited based on their International Statistical Classification of Diseases and Related Health Problems (ICD-10) diagnoses. Participants completed the following 3 ocular symptom questionnaires: the Ocular Surface Disease Index, Neuropathic Pain Symptom Inventory, and Dry Eye Questionnaire 5. A novel eye fixation-grid system was used to capture 30 standardized confocal microscopy images of the central cornea. Each participant was imaged twice by different operators. Seven quantitative nerve metrics were analyzed using automated software (ACCmetrics, Manchester, United Kingdom) for all 30 images and a 6-image subset.

Results:

Forty-seven participants were recruited (25 classified as dry eye and 22 controls). The most reproducible nerve metrics were corneal nerve fiber length [intraclass correlation (ICC) = 0.86], corneal nerve fiber area (ICC = 0.86), and fractal dimension (ICC = 0.90). Although differences were not statistically significant, all mean nerve metrics were lower in those with dry eye compared with controls. Questionnaire scores did not significantly correlate with nerve metrics. Reproducibility of nerve metrics was similar when comparing the entire 30-image montage to a central 6-image subset.

Conclusions:

A standardized confocal imaging technique coupled with quantitative assessment of corneal nerves produced reproducible corneal nerve metrics even with different operators. No statistically significant differences in in vivo corneal confocal microscopy nerve metrics were observed between participants with dry eye and control participants.

The corneal subbasal nerve plexus is a rich neuronal network derived from the trigeminal nerve ophthalmic division (V1) and arranged between the corneal epithelium and Bowman’s layer. These nerve fibers provide sensory neural function and also regulate corneal epithelial integrity, proliferation, and wound healing.¹ The nerve fibers within this region represent terminal small nerve fibers, similar to small nerve fibers present in the skin epidermis.² In vivo corneal confocal microscopy (IVCCM) of the corneal subbasal nerve plexus offers a noninvasive means of imaging these nerves with the possibility of multiple examinations on the same tissue area. Alterations in the subbasal corneal nerve plexus occur in the setting of normal aging, systemic disease, corneal disease, and after corneal surgery.^3–6 The morphologic metrics of these nerves have been investigated as a surrogate marker for earlier detection of diabetic sensorimotor neuropathy.⁷ The results have been promising, and quantification of corneal nerve metrics have correlated with severity of diabetic neuropathy.^8–10 Quantitative morphologic investigation of the corneal subbasal nerve plexus in other forms of peripheral neuropathy have since been investigated.^11–17

One of the greatest limitations preventing the broader use of IVCCM nerve metrics is the lack of a standardized, reproducible, efficient, and well-tolerated imaging and analysis method. The development of a standardized protocol would be beneficial in disease detection, monitoring progression, assessment of therapeutic intervention, and use in longitudinal studies or randomized controlled trials. The use of a fully automated imaging software analysis was one of the first steps in achieving such a standard and was subsequently found to have comparable efficacy with manual and semiautomated approaches of nerve metric analyses.^18–21 However, techniques to standardize image capture and provide interexaminer reproducibility are lacking.

The purpose of this study was to introduce a novel 30-dot eye fixation grid and investigate the reproducibility of 7 subbasal corneal nerve plexus imaging metrics. In addition, we sought to assess whether participants with dry eye exhibited different nerve metric values as compared to control participants. Finally, using dry eye symptomology questionnaires, we examined whether an increase in dry eye responses correlated with corneal nerve metric values.

MATERIALS AND METHODS

Study Subjects

A convenience sample of patients with dry eye was recruited from the Proctor Foundation at the University of California, San Francisco (UCSF), and a convenience sample of healthy controls was recruited from the UCSF optometry clinics. Patients with dry eye were eligible if they had a previous diagnosis of Sjögren syndrome (M35.0), ocular graft-versus-host disease (D89.81), meibomian gland dysfunction (H02.88), ocular rosacea (L71.8), keratoconjunctivitis sicca (H16.223), or ocular pain syndrome (H57.10).

Exclusion criteria for participants included diagnoses of diabetes mellitus, corneal ulcer in the study eye, previous diagnosis of herpetic or zoster keratitis, limbal stem cell deficiency, corneal epithelial membrane dystrophy, or any previous refractive or nonrefractive corneal surgery. Patients with use of previous autologous serum, platelet rich plasma, or topical nerve growth factors were excluded. This study was approved by the UCSF Research Ethics Committee and adhered to the tenets of the Declaration of Helsinki. Informed and written consent was obtained before study participation.

Questionnaires

Before IVCCM, participants completed the following 3 ocular symptom questionnaires: the Ocular Surface Disease Index (OSDI), Dry Eye Questionnaire-5 (DEQ-5), and Neuropathic Pain Symptom Inventory (NPSI) modified for ocular pain.²² Cumulative scores from both the OSDI (possible range of scores: 0–100) and DEQ-5 were calculated (possible range of scores: 0–22). For the NPSI, raw scores for the 3 questions—”Does your pain feel like burning,” “Does your pain feel like squeezing,” and “Does your pain feel like pressure”—were assessed independently (possible range of scores: 0–10).

Corneal Confocal Microscopy

Four different operators (3 fellowship-trained ophthalmologists and one fellowship-trained optometrist) performed the confocal imaging. Two operators were experienced with use of the confocal microscope, and 2 had received a single confocal training session. One eye of each study participant was imaged by 2 confocal operators on the same day, approximately 5 minutes apart.²³

All participant corneas were scanned with a laser IVCCM (Heidelberg Retinal Tomograph III Rostock Cornea Module; Heidelberg Engineering GmbH, Heidelberg Germany). Topical proparacaine hydrochloride 0.5% was instilled into the study eye for anesthesia immediately before examination. A hypromellose gel (GenTeal Gel; Novartis Ophthalmics, East Hanover, NJ) was applied directly to the IVCCM lens acting as a coupling agent bridging the space between the lens and a single-use sterile applanating cap. Participants placed their chin on the chin rest and maintained their forehead against the forehead rest with their examined eye gently pressed against the applanating cap. Participants were first asked to fixate their nonexamined eye on the target red dot of a 6 × 5 dot grid system (Fig. 1). If correction was required to view the grid system, a loose trial lens was held over the nonstudy eye. While the participant fixated on the target red dot, the confocal operator identified the central whorl of the subbasal nerve plexus of each participant’s study cornea.²⁴ When the central whorl was identified, the operator began acquiring images using the “section” setting on the confocal microscope. Participants were then instructed to look along the first line on the grid, starting with the first dot on the left and moving to each subsequent dot to the right on this top line. A single “section” scan was acquired as the participant fixated on each dot. At the end of each row, participants were instructed to return to the center red target dot to recapture the central whorl. The participant was then instructed to perform the same sequence, starting with the leftmost dot on the second row, and so forth, until completion of the entire grid. The image set allowed construction of a montage of the entire subbasal corneal nerve plexus (representative example shown in Fig. 2).

FIGURE 1.

6 × 5 eye fixation grid used for sequential imaging during IVCCM. The dot highlighted in red, denoted B3 represents the point of reference encompassing the central whorl. The dotted box highlights the 6 central images fully encompassing the central whorl that was used as part of the sensitivity analysis (note: letters not visible on the fixation grid used during the study). (The full color version of this figure is available at www.corneajrnl.com.)

FIGURE 2.

Representative 30 image montage from a healthy control participant (note: this collection of images has been rotated 180 degrees to orient the whorl in its anatomically correct inferior position).

Image Analysis

All images were assessed using a fully automated corneal nerve analysis software from ACCmetrics.^19,25,26 For each image analyzed, the software calculated 7 parameters: 1) nerve fiber density (CNFD)—number of fibers/mm², 2) nerve branch density—number of branch points on the main fibers/mm², 3) nerve fiber length (CNFL)—total length of nerves (mm)/mm², 4) nerve fiber total branch density (CTBD)—total number of branch points/mm², 5) nerve fiber area (CNFA) —total CNFA mm²/mm², 6) nerve fiber width (CNFW)—average CNFW (in millimeters)/mm², and 7) nerve fiber fractal dimension. Fractal dimension is measured using a box-counting method, in which different sizes of boxes are used to scan the image, and the number of boxes containing any part of the detected nerve fibers is counted.^27–29 The fractal dimension is measured as the slope of the line when the value of log (n) is plotted against log (r), where n = number of boxes that cover the nerve fiber, and r is the inverse of the box size.

When possible, all 30 images were used from each imaging session. If an image was obtained with no visible nerve plexus (as determined by J.S.T.), it was excluded from the analysis.

Statistical Analysis

Statistical analysis was conducted using Stata version 15.0 (College Station, TX). Intraclass correlation (ICC) was calculated to assess interrater reliability of corneal nerve metrics in 2 ways. First, an analysis was performed that compared metrics from each individual scan (ie, ICC calculated from a mixed effects linear regression (LR) of data from all readable scans, with scan position nested in participant identifier as random intercepts). Second, the mean of each nerve metric was computed across all 30 images for each grader, and an ICC calculated for this mean value (ie, ICC calculated from a mixed effects LR with participant identifier as a random intercept). For comparisons between the dry eye and control groups, a grand mean was calculated for each metric of each participant and then compared with a t test. Because the software computed 7 metrics, we Bonferroni adjusted the significance level of the study to 0.007. As a sensitivity analysis, the same statistical analysis was repeated on the 6 IVCCM images adjacent to the central whorl (Fig. 1). Finally, LR analysis was performed to assess the correlation between survey questionnaire scores and grand mean nerve metrics.

RESULTS

Forty-seven subjects in total were recruited for this study: 25 participants with dry eye and 22 healthy controls. Of the participants with dry eye conditions, 8 were diagnosed with ocular rosacea, 3 with meibomian gland dysfunction, 4 with Sjögren syndrome, 3 with keratoconjunctivitis sicca, 2 with graft-versus-host disease, and one with ocular pain syndrome. Four participants carried multiple dry eye diagnoses. Dry eye diagnosis was confirmed using the Tear Film and Ocular Surface Society Dry Eye Workshop (TFOS DEWS II) algorithm—14 participants were classified as evaporative dry eye disease, 10 as aqueous deficient dry eye disease, and 1 participant was classified as both evaporative and aqueous deficient dry eye disease.³⁰ Of the 25 participants with dry eye, 20 were using artificial tears ranging from as needed to every hour, whereas 4 used a night-time lubricating ointment. Fourteen participants were being treated with topical corticosteroids, 2 with cyclosporine 0.05%, 3 with scleral lenses, and 4 participants were taking oral fish oil therapy for dry eye symptoms. An additional 4 participants were using topical antihistamine eye drops, and one participant was using topical brimonidine. Of the 22 control participants, 16 were not using any form of ocular surface therapy. Four control participants were using artificial tears as needed, 3 were using antihistamine eye drops, and 1 participant was using multiple topical intraocular pressure–lowering agents.

Participant demographic and DEQ results are summarized in Table 1. Confocal operator 1 was responsible for scanning 43 participants, operator 2 scanned 31 participants, operator 3 scanned 13 participants, and operator 4 scanned 7 participants. When evaluating the 30-image montage set, in total 59 images from 27 participants were excluded from the analysis (ranging from 1 to 6 images per participant) because there was no contact between the cornea and confocal lens. When evaluating the 6-image montage set, 6 images from 3 participants were excluded from the analysis (2 images per participant).

TABLE 1.

Participant Characteristic Summary

	Dry Eye Participants	Control Participants	Total
No. participants (N)	25	22	47
Age (yr ± SD)	58.5 ± 11.7	55.5 ± 13.0	57.3 ± 12.31
Female (N, %)	20 (80)	16 (73)	36 (77)
Male (N, %)	5 (20)	6 (27)	11 (23)
OSDI score (score ± SD)	40.29 ± 28.80	10.23 ± 11.46	26.01 ± 26.78
DEQ-5 score (mean ± SD)	11.92 ± 4.95	4.62 ± 3.35	8.66 ± 5.63
NSPI–burning (mean ± SD)	3.38 ± 3.54	0.52 ± 1.75	2.10 ± 3.20
NPSI–squeezing (mean ± SD)	0.81 ± 1.86	0.00 ± 0.00	0.45 ± 1.42
NPSI–pressure (mean ± SD)	2.27 ± 2.81	0.28 ± 0.78	1.36 ± 2.35

ICC results for the corneal nerve metrics are presented in Table 2 for the 30-scan image sets. Inter-rater reliability was higher when the metrics were first summarized as a mean of all 30 images, and lower when the metrics of each scan position were compared. The mean-summarized nerve metrics with the highest reproducibility were CNFL [ICC = 0.86, 95% confidence interval (CI), 0.76–0.92], CNFA (ICC = 0.86, 95% CI, 0.77–0.91) and fractal dimension (ICC = 0.90, 95% CI, 0.83–0.94). By contrast, the metric with the highest inter-rater reliability when assessing individual scan position was CNFL (ICC = 0.57, 95% CI, 0.49–0.64). A subanalysis was performed to investigate whether reproducibility metrics improved as confocal operators gained more experience. To do so, we examined the scans completed by the 2 operators who completed the imaging sessions for this study (operator 1 and operator 2). The first 10 imaging sessions of each operator were classified as “early” scans, and all subsequent sessions were classified as “late.” Participants who were scanned by both operator 1 and operator 2 were included in the analysis and all others excluded. Based on these criteria, 6 scans qualified for inclusion in the “early” cohort and 19 in the “late” cohort. ICC measurements were similar for the early and late scans and did not provide evidence for a substantial learning curve (see Supplemental Tables 1 and 2, Supplemental Digital Content 1, and 2, http://links.lww.com/ICO/B111 and http://links.lww.com/ICO/B112).

TABLE 2.

ICC Between 2 Confocal Microscopy Operators for Their Entire Imaging Montage (2 Sets of 30 Images Per Participant)

Nerve Metric	Mean ± SD	ICC (Average of all Image Taken) N = 47	ICC (Each Individual Image of Montage), N = 47
CNFD (mm/mm2)	11.60 ± 0.19	0.69 (0.52–0.82)	0.33 (0.26–0.42)
CNBD (main fiber branch points/mm2)	15.17 ± 0.38	0.62 (0.44–0.78)	0.30 (0.22–0.38)
CNFL (mm/mm2)	9.79 ± 0.09	0.86 (0.76–0.92)	0.57 (0.49–0.64)
CNFA (area of nerve: mm2/image area: mm2)	0.005 ± 0.00005	0.86 (0.77–0.91)	0.49 (0.42–0.56)
CNFW (mm/mm2)	0.02 ± 0.00005	0.68 (0.51–0.81)	0.29 (0.22–0.38)
CTBD (total branch points/mm2)	31.32 ± 0.59	0.74 (0.59–0.84)	0.38 (0.31–0.46)
Fractal dimension	1.43 ± 0.001	0.90 (0.83–0.94)	0.51 (0.42–0.59)

Column 3 presents the overall ICC for each nerve metric considering all montage images. Column 4 represents ICC for individual images by each operator within the montage (ie, A1 of operator 1, to A1 of operator 2, etc).

CNBD, nerve branch density.

The results for the 6-scan central subset of images are shown in Table 3. Corneal nerve metric means for each of the 7 parameters were higher for the 6-image subset than for the 30-image set. ICCs displayed a similar pattern to the full analysis, although with universally lower inter-rater reliability.

TABLE 3.

ICC Between 2 Confocal Microscopy Operators for the Inferior Corneal Nerve Plexus Whorl Subset (2 Sets of 6 Images Per Participant)

Nerve Metric	Mean ± SD	ICC (Average of all Images Taken), N = 47	ICC (Each Individual Image of Montage), N = 47
CNFD (mm/mm2)	11.92 ± 0.42	0.48 (0.28–0.69)	0.35 (0.25–0.47)
CNBD (main fiber branch points/mm2)	21.33 ± 1.10	0.35 (0.15–0.62)	0.32 (0.22–0.43)
CNFL (mm/mm2)	10.79 ± 0.21	0.71 (0.56–0.83)	0.58 (0.47–0.69)
CNFA (area of nerve: mm2/image area: mm2)	0.006 ± 0.0001	0.69 (0.53–0.82)	0.47 (0.36–0.60)
CNFW (mm/mm2)	0.02 ± 0.0001	0.40 (0.20–0.65)	0.27 (0.16–0.42)
CTBD (total branch points/mm2)	45.39 ± 1.75	0.57 (0.38–0.75)	0.41 (0.29–0.55)
Fractal dimension	1.44 ± 0.003	0.78 (0.64–0.87)	0.58 (0.47–0.68)

Column 3 presents the overall ICC for each nerve metric considering all montage images. Column 4 represents ICC for individual images by each operator within the montage (ie, B2 of operator 1, to B2 of operator 2, etc).

CNBD, nerve branch density.

The t test results comparing quantitative nerve metrics in participants with dry eye and control participants are presented in Table 4 (30 montage images) and Table 5 (6 central whorl images).

TABLE 4.

A Student t test Comparing Mean Corneal Nerve Fiber Metrics for Entire Montage Image (30 Images Per Participant, Refer to Image of Montage)

Nerve Metric	Dry Eye Participants, N = 25	Control Participants, N = 22	P
CNFD (mm/mm2)	11.93 (9.42–14.45)	11.30 (9.61–12.98)	0.67
CNBD (main fiber branch points/mm2)	14.60 (10.06–19.13)	15.95 (12.23–19.67)	0.64
CNFL (mm/mm2)	9.54 (8.05–11.04)	10.15 (9.15–11.15)	0.50
CNFA (area of nerve: mm2/image area: mm2)	0.0047 (0.0040–0.0052)	0.0056 (0.0050–0.0062)	0.03
CNFW (mm/mm2)	0.023 (0.022–0.024)	0.023 (0.023–0.024)	0.30
CTBD (total branch points/mm2)	28.50 (21.81–35.19)	34.91 (28.49–41.33)	0.16
Fractal dimension	1.42 (1.40–1.44)	1.44 (1.43–1.45)	0.20

CNBD, nerve branch density.

TABLE 5.

A Student t test Comparing Mean Corneal Nerve Fiber Metrics for Smaller Montage Image Bounded by B2 to B4 to C2 to C4 (6 Images Per Participant, Refer to Image of Montage)

Nerve Metric	Dry Eye Participants, N = 25	Control Participants, N = 22	P
CNFD (mm/mm2)	12.32 (9.52–15.11)	11.49 (8.92–14.06)	0.66
CNBD (main fiber branch points/mm2)	20.14 (13.43–26.86)	22.74 (15.90–29.58)	0.58
CNFL (mm/mm2)	10.33 (8.61–12.05)	11.35 (9.91–12.78)	0.36
CNFA (area of nerve: mm2/image area: mm2)	0.0059 (0.0050–0.0067)	0.0070 (0.0061–0.0079)	0.073
CNFW (mm/mm2)	0.023 (0.023–0.24)	0.024 (0.023–0.024)	0.42
CTBD (total branch points/mm2)	39.51 (29.03–49.99)	52.29 (39.59–64.99)	0.11
Fractal dimension	1.44 (1.42–1.46)	1.45 (1.44–1.47)	0.20

CNBD, nerve branch density.

No statistically significant differences between the 2 study populations were observed where P < 0.007. Despite this, all mean nerve metrics with the exception of CNFD were higher in control participants compared with participants with dry eye.

A sensitivity analysis investigating whether a single image, B3, representing the subbasal corneal nerve plexus whorl was also performed. Reproducibility results between 2 operators are provided in Supplemental Digital Content 3 (see Supplemental Table 3, http://links.lww.com/ICO/B113), and t test results between control and dry eye participants are presented in Supplemental Digital Content 4 (see Supplemental Table, http://links.lww.com/ICO/B114). Corneal nerve metrics were greater when assessing this single image, whereas ICCs was comparable with the results for ICCs when comparing reproducibility between operators for each grid dot.

LR analysis of the relationship between dry eye symptomology questionnaires and nerve metric outputs are presented in Table 6. Values describe the change in corneal nerve fiber metric associated with a one-unit increase in each survey parameter. Although no statistically significant relationships were found, increasing dry eye scores were in general negatively correlated with nerve metrics (ie, negative regression coefficients, indicating that eyes with increased dry eye symptoms had lower nerve metrics). OSDI scores in particular exhibited an inverse correlation with nerve metrics that nearly approached significance, whereas the magnitude of the regression coefficients of the DEQ-5 and NPSI analyses were frequently near zero.

TABLE 6.

LR Results Using DEQ Results as Predictors of ACCmetric Corneal Nerve Metric Measurements

	CNFD	CNBD	CNFL	CNFA	CNFW	CTBD	Fractal Dimension
Nerve metrics from all 30 images
DEQ-5 score	−0.038 (−0.37 to 0.29)	−0.057 (−0.23 to 0.11)	−0.19 (−0.73 to 0.37)	−756.32 (−1909.29 to 396.64)	−172.86 (−1371.09 to 1025.38)	−0.05 (−0.16 to 0.05)	−18.67 (−56.72 to 19.39)
OSDI	−0.65 (−2.30 to 0.99)	−0.58 (−1.43 to 0.27)	−1.16 (−3.93 to 1.60)	−3070.55 (−8930.16 to 2789.06)	1787.07 (−4099.90 to 7674.05)	−0.37 (−0.90 to 0.16)	−66.81 (266.35 to 132.73)
NPSI (burning)	−0.03 (−0.22 to 0.16)	−0.02 (−0.12 to 0.08)	−0.02 (−0.34 to 0.29)	−86.96 (−754.30 to 580.39)	255.47 (−421.83 to 932.78)	−0.01 (−0.076 to 0.046)	−1.25 (−23.11 to 20.60)
NPSI (pressing)	0.01 (−0.13 to 0.15)	−0.02 (−0.10 to 0.05)	−0.02 (−0.25 to 0.21)	−233.09 (−719.87 to 253.70)	−134.93 (−635.23 to 365.36)	−0.02 (−0.064 to −0.026)	−0.94 (−17.03 to 15.16)
NPSI (squeezing)	0.02 (−0.06 to 0.10)	0.01 (−0.04 to 0.05)	0.05 (−0.09 to 0.19)	86.71 (−210.07 to 383.49)	−81.71 (−384.78 to 221.36)	0.01 (−0.02 to 0.03)	3.52 (−6.17 to 13.22)
Considering the 6 image inferior whorl
DEQ-5 score	−0.01 (−0.28 to 0.26)	−0.28 (−0.13 to 0.08)	−0.16 (−0.60 to 0.29)	−357.40 (−1128.00 to 413.19)	−36.26 (−1145.80 to 1073.27)	−0.03 (−0.09 to 0.028)	−12.87 (−51.85 to 26.11)
OSDI	−0.79 (−2.12 to 0.54)	−0.40 (−0.91 to 0.12)	−1.57 (−3.77 to 0.62)	−2353.99 (−6207.55 to 1499.58)	2164.98 (−3300.19 to 7630.14)	−0.26 (−0.55 to 0.03)	−109.97 (−311.32 to 91.39)
NPSI (burning)	−0.07 (−0.22 to 0.08)	−0.03 (−0.09 to 0.02)	−0.13 (−0.38 to 0.12)	−187.67 (−626.25 to 250.90)	363.35 (−257.76 to 984.46)	−0.02 (−0.06 to 0.01)	−8.80 (−30.90 to 13.31)
NPSI (pressing)	−0.04 (−0.15 to 0.07)	−0.02 (−0.06 to 0.03)	−0.10 (−0.28 to 0.09)	−199.42 (−519.48 to 120.63)	33.79 (−430.46 to 498.05)	−0.01 (−0.04 to 0.01)	−6.82 (−23.09 to 9.44)
NPSI (squeezing)	0.02 (−0.04 to 0.09)	0.01 (−0.02 to 0.03)	0.03 (−0.08 to 0.14)	37.90 (−159.02 to 234.82)	−136.37 (−414.68 to 141.94)	0.01 (−0.01 to 0.02)	1.50 (−8.42 to 11.42)

Values in this table represent the associated change in resultant nerve metrics with a 1-unit increase in each survey parameter. A negative number suggests that as survey scores increase there is a resultant decrease in nerve parameters.

CNBD, nerve branch density.

DISCUSSION

This study investigated the reproducibility of IVCCM subbasal nerve plexus imaging using a novel eye fixation grid system and multiple (4) confocal operators. The highest ICC between confocal operators was found when assessing CNFL, CNFA, and fractal dimension measurements. ICC values were considerably higher when images were first summarized as a mean compared with when the metrics from each image position were assessed independently. When comparing study populations, control participants as a whole had a greater CNFA, CTBD and CNFL compared with dry eye participants, although not to a statistically significant level.

Previous IVCCM studies have used the terms reproducibility and repeatability interchangeably. Many of these studies were conducted by a single confocal imaging operator and one or multiple “raters” or “examiners” who evaluate quantitative nerve metrics using semiautomated (ie, CCmetrics or NeuronJ) or automated analysis (ie, ACCmetrics). A small subset of these studies repeated IVCCM imaging and nerve metric analysis of their participants which are summarized in Table 7.

TABLE 7.

A Summary of the Methods and Reproducibility Results of Previous Confocal Microscopy Studies That Involved Repeat Confocal Imaging Sessions of Participants

	Hertz et al7	Petropoulos et al28	Kim et al42	Smith et al45
No. participants	46 healthy and T1DM subjects	19 health subjects	10 healthy subjects	11 healthy subjects
No. confocal operators	—	1	1	1
Imaging setting	Volume-scan	Section-mode	Volume-scan	Volume-scan
Interval between sessions	Same day	1 wk	1 mo	1–4 wk
Imaging protocol	Participants fixated on a targeted position behind the device	Participants focused on an outer fixation light while images were taken in 1 μm increments	Modified Amsler grid located with 5 designated fixation positions	As per Kim et al
Image analysis program	CCmetrics (semiautomated)	CCmetrics (semiautomated)	CCmetrics (semiautomated)	CCmetrics (semiautomated)
No. images for analysis	2	10 (5 per eye)	5	5
CNFL (ICC)	0.72	0.70	0.89–0.90	0.92
CNBD (ICC)	0.61	0.61	0.80–0.81	—
CNFD (ICC)	0.57	0.74	0.67–0.75	—

CNBD, nerve branch density.

The methodology of this current study differs from previous analyses in at least 4 ways: 1) using multiple confocal operators with varying levels of experience, 2) standardizing imaging protocol in a reproducible manner, 3) imaging a larger proportion of the subbasal corneal nerve plexus, and 4) removing human error from the nerve metric quantification analysis. The results from our analysis demonstrate comparably strong CNFL reproducibility (ICC = 0.71–0.86) compared with previous studies shown in Table 7. In addition, CNFA (ICC = 0.69–0.86) and fractal dimension (ICC = 0.78–0.90) also demonstrated high reproducibility, although there was no available data to make comparisons to for these metrics. Other metrics such as nerve branch density and CNFD demonstrated strong reproducibility when analyzing the mean of 30-image montage but poorer reproducibility with the 6-image montage. This could be attributed to software limitations in discerning primary nerve fibers from branch fibers at nerve junctions.¹⁸ It could also be explained by greater reproducibility of peripheral aspects of the nerve plexus where nerves tend to be more delineated and less dense.

Previous studies have suggested CNFL to be the most robust parameter in quantitatively discriminating cases of neuropathy from cases without neuropathy.^{7,8,23,31–34} Eyes with neuropathic pain symptoms tend to demonstrate a decrease in CNFL density.^13,35–37 CNFL in this study cohort was greater in control participant eyes compared with participants’ eyes with dry eye, although not to a statistically significant level. Of note, quantitative values for CNFL in this study were lower than normative values published using CCmetrics software (mean CNFL 10.15 mm/mm² in control participants vs. 19.97 mm/mm² in healthy patients >65).²⁹ However, our values more closely approximate a recent reproducibility study published by Kalteniece et al,³⁸ which analyzed IVCCM using ACCmetrics (mean CNFL: 14.44–14.97 mm/mm²). A downside of using only CNFL to identify neuropathic disease is that CNFL does not inherently discern nerve degeneration and regeneration, which might have importance especially in the context of monitoring treatment effect.²⁸ Based on the findings presented, nerve metrics to consider when stratifying healthy from dry eye disease include CNFA and CTBD. Both parameters were greater in healthy eyes compared with participants with dry eye and also nullify the inherent challenge in defining nerves as a primary fiber or as a branching fiber. A study recently published by Giannaccare et al¹³ suggested that corneal nerves in neuropathy may be thicker as reflected by a greater CNFW compared with healthy eyes. The authors postulated that neuropathy causes loss of smaller nerve fibers while preserving thicker bundles, and pathology may induce nerve fiber swelling. Although these results were not reproduced in our study, the concept merits further investigation.

A lack of standardized imaging protocols limits the utility of reimaging and reanalysis in longitudinal studies because of uncertainty as to whether the same corneal region is captured at subsequent imaging sessions. Our novel grid system allows for reproducible imaging with the addition of a corneal mapping system that is simply printed on a standard letter-sized 8.5″ × 11″ paper. We believe that having patients view a single dot on our grid system in a sequential manner allows confocal operators to have the opportunity to capture the same areas of interest to identify changes in the subbasal nerve plexus of the central cornea over time. Using a fully automated software to quantitatively analyze nerve metrics further enhances its utility in an unbiased and faster manner compared with semiautomated programs.¹⁸

A limitation of previous IVCCM studies is that they often rely on a single experienced operator to acquire images. This study included 4 different IVCCM operators with varying levels of experience. Images were acquired using the “section” setting because it provided greater control of image quality and limited corneal nerve overlap between images analyzed. Use of sequence mode and volume scan mode on the contrary offer greater ease of use at the expense of precision. Image quality is dependent on image contrast, pressure artifact, depth, and presence of cells (ie, nerve sprouts and dendritic cells).³⁹ Poor quality images have a bias toward pathologic findings when assessed quantitatively, further highlighting the importance of image quality.

This novel grid protocol enabled us to image most of the subbasal corneal nerve plexus including its most peripheral aspects. However, we found the 6-image subset ranging from grid point B2 to C4 (Fig. 1) demonstrated greater absolute nerve metric characteristics as demonstrated in Tables 4 and 5. Overall reproducibility was greater when assessing the 30 image montage compared with the 6-image subset (Tables 2 and 3, column 3); however, individual image ICC was comparable and even marginally improved for multiple nerve metrics in the 6-image subset (Tables 2 and 3, column 4). For this reason, the 6-image subset corresponding to the inferior corneal whorl, which given its distinct appearance, may be the most optimal site to provide standardized and reproducible image capture.^40,41 Averaging multiple images provides a more accurate representation of a patient’s generalized corneal nerve anatomy, although the optimal number of images is unknown.⁴²

Limitations of the study include the lack of clinical correlates of dry eye disease severity (ie, Schirmer testing and tear break-up time). It is possible that some dry eye cohort participants may have had milder forms of disease resulting in smaller discrepancies between controls and patients with dry eye. Furthermore, the dry eye cohort used in this study comprised participants with a variety of dry eye disease diagnoses. Although the results from a heterogenous group allows for broad generalizations, future studies focusing on a single particular dry eye condition would be useful in specifically characterizing disease-related nerve plexus changes. The control patient sample, on the other hand, contained 6 participants who were using some form of ocular surface therapy. In particular, 1 participant was on several forms of intraocular pressure–lowering medications that have previously shown to subclinically effect subbasal nerve plexus measurements, such as nerve length, tortuosity, and dendritic cell density.⁴³ The remaining participants were using topical artificial tears or antihistamine drops intermittently on an as needed basis, although less frequently than daily. Although the effect of these medications and their preservatives on ocular surface nerve metrics is not well established, limiting all forms of ocular surface therapy would have been most ideal. Several operator and participant factors can limit the utility of IVCCM in general. Participant study compliance for one is difficult to accommodate for and may be limited in the setting of a 30-image examination. Use of the section imaging mode increases the demand for both participant and operator cooperation for the duration of the examination to produce high quality images. The full contribution of operator variability was difficult to assess in this study because not all operators performed the test on all participants. Finally, this study was completed on a small study sample of 47 participants—most whom were women, limiting the generalizability of the results. A larger sample size could provide a more robust statistical analysis to draw conclusions. In addition, future studies should consider utilization of a variety of ocular pain assessment tools, including both the NPSI and the Ocular Pain Assessment Survey.⁴⁴

To conclude, the use of this novel and simple eye fixation grid to capture IVCCM images has the potential to address some of the issues in the standardization of IVCCM. The results demonstrated good reproducibility among 4 confocal operators when averaging the nerve metrics of multiple images taken. Imaging the most dense and identifiable regions of the cornea is promising for better qualitative description of corneal disease, especially with reexamination on repeated examinations. Quantitative analysis of nerve metrics between participants with dry eye and healthy controls warrants further investigation with more stringent inclusion criteria and clinical correlates.