Reliability and Reproducibility of Assessment of Corneal Epithelial Thickness by Fourier Domain Optical Coherence Tomography

Abstract

Purpose.: To analyze the intra-user reliability and inter-user reproducibility of assessment of corneal epithelial thickness by Fourier domain optical coherence tomography.

Methods.: In this consecutive cross-sectional case series performed at a tertiary ocular care institution, 210 eyes of 210 subjects underwent anterior segment Fourier domain optical coherence tomography (FDOCT). A caliper tool software was used to measure the corneal thickness. For the reproducibility measures, the examination was done by 2 examiners (user 1, user 2) within 30 minutes of each other. For the reliability measure, the retest was done by user 1 on the next day, within 30 minutes of the previous test's time. The total corneal thickness, epithelial thickness, and corneal thickness excluding the epithelium were measured.

Results.: The mean corneal thickness of the population measured by user 1 was 519.5 ± 31.1 μm, 58.6 ± 4.2 μm, and 460.95 ± 31.4 μm for total cornea, epithelium, and non-epithelial cornea, respectively. The difference in results between user 2 and user 1 was 0.8 ± 7.2 μm, 0.23 ± 3.3 μm, and 0.7 ± 8.2 μm for total, epithelium, and non-epithelial cornea, respectively, and the difference in results between the repeated series by user 1 was 0.49 ± 5.7 μm, −0.13 ± 2.7 μm, 0.61 ± 5.4 μm total, epithelium, and non-epithelial cornea, respectively (paired t-test, P > 0.05). Intraclass correlations ranged from 0.87 to 0.99, coefficients of repeatability from 4.5 to 14.11, and coefficient of variation from 2.3% to 11.1%.

Conclusions.: Fourier domain anterior segment optical coherence tomography is reproducible and reliable for the measurement of epithelial thickness at vertex.

Introduction

The corneal epithelium plays an important role in the optics of the eye. ^1,2 The thickness of the corneal epithelium variably contributes to the optical power of the cornea. ² After excimer laser ablation surgeries, the corneal epithelium might influence tear film instability and may be associated with local irregularities of corneal topography. ³ Epithelial hyperplasia after excimer ablation can cause an undesired refractive shift. ^4–5 Epithelial thinning can occur on continued usage of hydrogel contact lenses, due to overnight wear as in orthokeratology, or in ectatic disorders like keratoconus. ^6–8 Due to these multiple factors, accurate assessment of epithelial thickness is very important. In the past, Reinstein et al. ⁹ measured the corneal epithelium profile in normal eyes using a very high-frequency digital ultrasound device in normal eyes. In another study done on rabbit eyes, researchers used a confocal microscopy through-focusing (CMTF) methodology. ¹⁰ Furthermore, anterior segment optical coherence tomography (ASOCT) has been used in recent years to evaluate the corneal epithelial thickness. In a repeatability study based on the older technology of time domain optical coherence tomography, the authors found clinically reasonable outcomes. ¹¹ Their small study of 18 subjects (32 eyes) showed intraclass correlation (ICC) was 0.98 for cornea and 0.73 for epithelium, and the coefficients of repeatability were ±10.64 μm for cornea and ±6.53 μm for epithelium. ¹¹ Fourier domain (spectral domain)-based optical coherence tomography was initially used for posterior segment evaluation. ^12–16 However, recently the same has been applied into anterior segment applications. ^17–23  

In the authors' previous study based on anterior segment optical coherence tomography, they found that Fourier domain-based ASOCT has a better reliability than time domain ASOCT. ²⁴ Another recent study demonstrated that Fourier domain OCT can be used to measure the thickness of the epithelium and Bowman's membrane. ²⁵ However, to the best of the authors' knowledge, there is no study in published literature that has evaluated repeatability (inter user and intra user) of Fourier domain OCT to measure epithelial thickness. The current study looks at the reliability and reproducibility of a commercially-available Fourier domain platform for assessment of corneal epithelial thickness, and compares it with non-epithelial and total corneal thickness. 

Materials and Methods

This comparative trial was conducted at in a tertiary care eye hospital. It was approved by the institutional review board. All the tenets of the Declaration of Helsinki were adhered to. A total of 210 healthy volunteers were enrolled in the study (mean age 24.2 ± 2.6 years; 122 males and 88 females). 

None of them had any ocular morbidity except for mild refractive error, which was being treated by spectacle correction. None of them used contact lenses. All the volunteers underwent anterior segment Fourier domain optical coherence tomography (FDOCT) with Cirrus HD-OCT (Carl Zeiss Meditec, Inc., Jena, Germany). The technical specifications are as follows ²⁶ : The machine takes 27,000 A-scans per second. The axial resolution in tissues is 5.0 μm, transverse resolution is 15 μm, and A-scan depth is 2 mm. The optical source is a super-luminescent diode (SLD) with a scan beam wavelength of 840 nm. The exposure power at the cornea is <720 μW. The patient was positioned on the headrest. The infrared (IR) image of the cornea was seen directly on the examination screen. The patient was asked to look in the fixation light in the device. All scans were performed with the patient's eye wide open with his/her own effort. No topical anesthesia or lubricating drops were used. The instrument's anterior segment 5-line raster mode was used. It has a scan angle of 0 degrees, spacing of 0.25 mm between the lines, and a line length (scan length) of 3 mm. We centered the image capture on the corneal vertex. It was determined by observation on the dynamic output screen. The concentricity of the limbus and the capture annulus helped in determining alignment. In the image acquire mode, 5 lines for 5 scans were displayed on the screen on the real-time image of the cornea. Use of X–Y and defocus adjustments was done to align the central line with the corneal vertex. On satisfactory alignment of the central line and the vertex, the scan was captured. A repeat scan was taken if the first scan was not of satisfactory nature. The scan can be unsatisfactory when it is decentered due to the patient's eye movement, poor corneal apex reflection, or head tilt, or when poor corneal apex reflection undermines the quality of the image, preventing a good resolution and demarcation of corneal layers. There are no subjective guidelines of the same currently in peer-reviewed literature. However, the authors were stringent in terms of the scan selection, and in case even a minor tilt from was noticed, the scan was discarded and a repeat scan was done. Some figures are included to emphasize of the types of scan that were discarded from the study (see Fig. 2). It would be interesting to see future studies/guidelines or incorporation of software that does analysis of scan quality, as in devices like the Zywave aberrometer (Bausch and Lomb, Rochester, NY). 

The output image showed prominent layers of the cornea (Fig. 1), and a built-in software caliper tool was used to subsequently to measure the epithelial thickness and corneal thickness, excluding the epithelium (subsequently called “non-epithelial thickness”) and total corneal thickness. The same image was used for all 3 calculations. The images were stored as screenshots with first image showing the caliper readings for epithelial and non-epithelial corneal thickness (Fig. 3) and the second image showing the caliper reading for total corneal thickness (Fig. 4). The non-epithelial thickness consisted of Bowman's membrane, stroma, Descemet's membrane, and endothelium together. 

Figure 1.

 

Screenshot of the normal cornea with a well centered scan showing the layers of the cornea. The white arrows and legends were added later and not a part of the image.

Figure 2.

 

Screenshots of variations in scan outputs: left upper, normal scan well focused and centered (included in the study); left lower: a well focused but decentered scan (discarded from study); right upper, defocused and decentered scan (discarded from study); right lower: distorted scan due to patient's eye movement (discarded from study).

Figure 3.

 

Screenshot of the output screen for measurements for epithelium (“56 μm in cornea”) and the non-epithelial region (“432 μm in cornea”).

Figure 4.

 

Screenshot of the output screen for total corneal thickness (“488 μm in cornea”).

For the reproducibility measures, the examination was done by 2 examiners (user 1 and user 2) within 30 minutes of each other. For the reliability measure, the retest was done by user 1 on the next day, within 30 minutes of the previous test's time. All patients underwent the test on both eyes, after which 1 eye was randomly selected for evaluation based on a computer-generated random number allocation chart, and was used for reproducibility and reliability. 

The data was entered on a Microsoft Excel spreadsheet (Microsoft Inc., Redmond, WA) and transferred for analysis to SPSS 16.0 (SPSS Inc., Chicago, IL). Paired t-tests were used to analyze the difference of mean. Correlation coefficients and best fit linear equations were computed to assess the correlation between measures from retesting. For the repeatability measures, intraclass correlations, 95% limits of agreement (LOA), coefficient of reliability, and coefficient of variation were computed. 

The coefficient of reliability (for intra-user repeatability or inter-user reproducibility) was computed using the Bland–Altman method. It was computed as follows. ²⁷  

 

 

where  

 

 

The coefficient of variation (CV) was calculated to factor for the overall effect of the average of mean pachymetry of the 2 tests, on the standard deviation noted. It was expressed as a percentage. It was computed as follows. ²⁸  

 

 

where  

 

 

 

 

95% LOA were used to evaluate the overall spread of the data. ²⁶ They were computed as: 

Upper limit: mean of difference plus 2 standard deviation of difference 

Lower limit: mean of difference minus 2 standard deviation of difference 

Width of 95% LOA: upper minus lower limit 

Results

Corneal Thickness Readings.

Mean values of measurements by the first user, second user (performed at the same day, same time, as described in the Methods section), and repeat measurement by the first user (next day, same time, as described in the Methods section) for total corneal thickness, epithelial thickness, and non-epithelial thickness were analyzed and are given in Table 1. 

Table 1.

 

Corneal Thickness Outcomes for the 3 Measurements

		Mean ± SD	95% Confidence Interval
Total corneal thickness	User 1 Measurement 1	519.5 ± 31.1 μm	515.3 to 523.8 μm
	User 2	518.6 ± 32.5 μm	514.2 to 523.1 μm
	User 1 Measurement 2	519.0 ± 31.5 μm	514.7 to 523.3 μm
Epithelial thickness	User 1 Measurement 1	58.6 ± 4.2 μm	58.0 to 59.2 μm
	User 2	58.4 ± 5.3 μm	57.6 to 59.1 μm
	User 1 Measurement 2	58.7 ± 4.8 μm	58.1 to 59.4 μm
Non-epithelial thickness	User 1 Measurement 1	460.95 ± 31.4 μm	456.7 to 465.2 μm
	User 2	460.3 ± 31.2 μm	456.0 to 464.6 μm
	User 1 Measurement 2	460.2 ± 33.0 μm	455.8 to 464.8 μm

Difference from Measurement by User 2 (Reproducibility) Measure.

The difference in the epithelial thickness was 0.23 ± 3.3 μm (paired t-test, P = 0.3). There was a correlation of r = 0.78 (P = 9.4 × 10⁻⁴⁵). The best fit line was b = 0.98a + 0.835 (where b = user 2 epithelial thickness, a = user 1 measurement 1 epithelial thickness), with an R-squared of 0.6. The difference in the non-epithelial thickness was 0.7 ± 8.2 μm (paired t-test, P = 0.2). There was a correlation of r = 0.96 (P = 8.8 × 10⁻¹²⁸). The equation for the best fit line was b = 1.02a − 10.02 (where b = user 2 non-epithelial thickness, a = user 1 measurement 1 non-epithelial thickness), with an R-squared of 0.94. The difference in the corneal thickness was 0.8 ± 7.2 μm (paired t-test, P = 0.08). There was a correlation of r = 0.97 (P = 6.4 × 10⁻¹³⁸). The equation for the best fit line was b = 1.02a − 12.8 (where b = user 2 corneal thickness, a = user 1 measurement 1 corneal thickness), with an R-squared of 0.6. 

Difference on Repeat Measurements for Intra-user Reliability.

The difference in the epithelial thickness was −0.13 ± 2.7 μm (paired t-test, P = 0.3). There was a correlation of r = 0.88 (P < 0.001). The equation for the best fit line was b = 1.002a + 0.014 (where b = user 1 measurement 2 epithelial thickness, a = user 1 measurement 1 epithelial thickness), with an R-squared of 0.78. The difference in the non-epithelial thickness was 0.61 ± 5.4 μm (paired t-test, P = 0.1). There was a correlation of r = 0.98 (P < 0.001). The equation for the best fit line was b = 0.99a + 2.9 (where b = user 1 measurement 2 non-epithelial thickness, b = user 1 measurement 1 eye non-epithelial thickness), with an R-squared of 0.97. The difference in the corneal thickness was 0.49 ± 5.7 μm (paired t-test, P = 0.3). There was a correlation of r = 0.98 (P < 0.001). The equation for the best fit line was b = 0.997a + 0.87 (where b = user 1 measurement 2 corneal thickness, a = user 1 measurement 1 eye corneal thickness), with an R-squared of 0.97. 

Intraclass Correlations.

The intraclass correlations were computed for the comparison for reliability and reproducibility for the epithelial, non-epithelial, and total measurements. 

The intraclass correlations were high for all the comparisons. The values ranged from 0.87 to 0.99 (Table 2). 

Table 2.

 

Intraclass Correlation Coefficients for the Retest Measurements for Epithelial, Non-Epithelial, and Total Corneal Thickness

Comparison	Intraclass Correlation Coefficients	95% Confidence Interval		P Value
		Lower	Upper
Epithelial thickness – User 1 and User 1 repeat test (reliability)	0.93	0.91	0.95	P < 0.001
Epithelial thickness – User 1 and User 2 (reproducibility)	0.87	0.83	0.90	P < 0.001
Non-epithelial thickness – User 1 and User 1 repeat test (reliability)	0.99	0.99	0.99	P < 0.001
Non-epithelial thickness – User 1 and User 2 repeat test (reproducibility)	0.98	0.99	0.99	P < 0.001
Total corneal thickness – User 1 and User 1 repeat test (reliability)	0.99	0.98	0.99	P < 0.001
Total corneal thickness – User 1 and User 1 repeat test (reproducibility)	0.98	0.98	0.99	P < 0.001

Coefficients of Repeatability.

Coefficients of repeatability were computed for the corneal, epithelial, and non-epithelial thickness. For intraobserver reliability (same user test 1 and repeat test) measures, the values were: 4.5, 10.6, and 11.2 for epithelial, non-epithelial, and total corneal thickness, respectively. For interobserver reproducibility (user 1 and user 2) measures were 6.5, 16.1, and 14.1 for epithelium, stroma, and total cornea, respectively. 

Coefficient of Variation.

Coefficients of repeatability percentage fraction values were computed for the corneal, epithelial, and non-epithelial thickness. For intraobserver reliability (same user test 1 and repeat test) measures the values were: 7.7%, 2.3%, and 2.15% for epithelial, non-epithelial, and total corneal thickness, respectively. For interobserver reproducibility (user 1 and user 2) measures were 11.1%, 3.5% and 2.7% for epithelial, non-epithelial, and total corneal thickness, respectively. 

95% LOA.

Mean difference and standard deviation values were used to compute 95% LOA between 2 measurements (Table 3). The width of 95% LOA was lesser for intra-user reliability compared to inter-user reproducibility. 

Table 3.

 

95% Limits of Agreement for the Retest Measurements for Epithelial, Non-Epithelial, and Total Corneal Thickness

		Mean of Difference (μ)	Standard Deviation of Difference (μ)	95% Limits of Agreement
				Upper Limit (μ)	Lower Limit (μ)	Width (μ)
Inter-user reproducibility	Epithelial	0.23	3.3	6.83	−6.37	13.2
	Non-epithelial	0.7	8.2	17.1	−15.7	32.8
	Total	0.8	7.2	15.2	−13.6	28.8
Intra-user reliability	Epithelial	0.13	2.7	5.53	−5.27	10.8
	Non-epithelial	0.6	5.4	11.4	−10.2	21.6
	Total	0.49	5.7	11.89	−10.91	22.8

Discussion

Optical coherence tomography works on the principle of low coherence interferometry. It gives clinically acceptable results on single-session and 2-session repeat examination for anterior segment evaluation. ^11,24 However, previous to this study, the reliability and reproducibility of Fourier domain-based anterior segment optical coherence tomography systems for epithelial thickness measurement had not been established. The mean epithelial thickness noted in the present cases was similar to that noted by Tao, et al. (52.5 ± 2.4 μm) on FDOCT. ²⁵ Sin and Simpson on time domain ASOCT (52 ± 3 μm. ¹¹ and that by Reinstein, et al. on a very high-frequency digital ultrasound (53.4 ± 4.6 μm). ⁹  

The corneal thickness at the vertex was 519.5 ± 31.1 μm in the present study. In a previous study by the authors on a different Fourier domain platform (Optovue, Fremont, CA) the mean central corneal thickness was comparable to this value (521.12 ± 34.5 μm). ²⁴ Another study by Ishibazaw. ¹⁷ showed the mean central Fourier domain ASOCT-based pachymetry to be 530 ± 33 μm. Therefore, racial- and instrument algorithm-based differences may occur in pachymetric analysis in general, and therefore, the same can apply to isolated epithelium measurement. Researchers should keep this in mind when comparing results across different platforms. Fair reliability and reproducibility values were seen for epithelial thickness, as well as for the mean non-epithelial thickness and overall thickness. 

CR is 1.96 times the standard deviation (SD). For repeatability between 2 observers, for the total corneal thickness (mean ∼ 519 μm), the value was 14.11. 97.5% of the data (based on 1.96 × SD) should be within ≤14.11 microns and 68% should be ≤7.19 microns on repeat testing. Clinically, this translates to a 97.5% chance of difference between the 2 observers of <2.7 μm per 100 μm of corneal thickness and a 68% chance of chance of difference <1.4 μm per 100 μm of corneal thickness, and thus is a clinically acceptable result. 

CV is computed as the percentage fraction of CR over the mean thickness. As the mean thickness is approximately 9 times lesser for the epithelium compared the total cornea and the CR is around 3 times more for total cornea, the overall result is a 3-fold increase in the CV for epithelium (11.1% compared to 2.7% for total cornea). The CV is thus useful for comparison between similar layers, for example, the epithelium, as measured by the same user or by a different user. For clinical interpretation between layers, the CR seems to have more practical overall utility. 

The high . ² values suggested a good data fit in a linear fashion, suggesting that both types of repeat measurements (by the same user and by a different user) produced a good concordance with the initial values. The values for reliability were slightly better compared to those achieved for reproducibility. However, on paired tests for mean differences, this did not show a significant difference in the mean of 2 different users. 

OCT measures the optical path length, by definition. The optical path length is the physical path length times the index of refraction. Therefore, when interpreting OCT measurements, there is an assumption that the index of refraction is constant. It is possible that some of the variation in the inferred thicknesses is caused by fluctuations in the index of refraction. Additionally, use of a manual caliper tool may also introduce difference in measurements. The location of the caliper, and the position of the starting and ending points of the caliper are all possible factors. Therefore, it would be interesting to have additional advances in the measurement algorithm like automated positioning, vertex determination, and a more detailed caliper with progression of 0.5 μm steps. Unlike the posterior segment Fourier domain analysis, where there are landmarks like the optic disc, macula, and the arcades to help in pattern recognition, there are no similar tools currently for the anterior segment Fourier domain system. Iris or limbal recognition-based corneal analysis could be a welcome tool in such a scenario, as it would increase the repeatability and reliability further. 

Furthermore, it remains to be seen whether the variability would increase in pathological corneas. 95% LOA also indicate at the amount of data spread. As with the other measures of reliability, the values of intra-user retesting were better than those with inter-user testing. 

To conclude, Fourier domain anterior segment optical coherence tomography is reproducible and reliable for measurement of epithelial thickness at the vertex. A slight, non-significant advantage for intra-user reliability was noted compared to inter-user reproducibility in this study. 

Footnotes

 Disclosure: G. Prakash, None; A. Agarwal, Bausch and Lomb (C), Staar Surgical (C), Abott Medical Optics (C); A.I. Mazhari, None; M. Chari, None; D.A. Kumar, None; G. Kumar, None; B. Singh, None