Wide-Range Calibration of Corneal Backscatter Analysis by In Vivo Confocal Microscopy

Abstract

Purpose.: To report intra- and interinstrument calibration methods for corneal backscatter analysis by in vivo confocal microscopy.

Methods.: Applicability of two reference standards was evaluated for corneal backscatter calibration. Repeated measurements of four concentrations of AMCO Clear (GFS Chemicals, Inc., Powell, OH) suspension and three transparencies (26%, 49%, and 65%) of polymethylmethacrylate (PMMA) slabs were performed to assess image intensity acquisition in a wide backscatter range. Intra- and intersession repeatability and lot-to-lot variation were determined for both standards. The effect of light intensity (LI) variation on image intensity acquisition was evaluated by examination of PMMA slabs with nonreference (60% and 80%) and reference (72%) LIs. Both reference standards were implemented in the protocol. Intrainstrument calibration was verified by measuring three normal corneas with 60%, 72%, and 80% LIs. Interinstrument calibration was tested by measuring PMMA slabs on a second, similar confocal microscope.

Results.: AMCO Clear was used to express image intensity in absolute scatter units (SU), whereas the 49% transparent PMMA slab showed best repeatability, without image saturation, to adjust for LI variation. Intrainstrument calibration for LI variation reduced mean differences from −38.3% to 1.7% (60% LI) and from 33.9% to −0.6% (80% LI). The mean difference between similar microscopes decreased from 18.4% to 1.2%, after calibration of the second microscope.

Conclusions.: Large interinstrument differences necessitate calibration of corneal backscatter measurements. With AMCO Clear suspension and PMMA slabs, standardization was achieved in a wide backscatter range corresponding to normal and opaque corneas. These methods can easily be applied in ophthalmic practice.

In vivo confocal microscopy (IVCM) is an optical technique in which minute structures on the order of micrometers can be visualized. In ophthalmic practice, IVCM can be used for measurement of endothelial cell density, analysis of corneal and intracorneal thicknesses, and assessment of cellular morphology and histopathologic changes in corneal sublayers. ¹ Another important feature of IVCM is objective quantification of corneal backscatter. Currently, this feature is not widely used. Corneal backscatter determined by IVCM is based on image intensity measurements, composed of backscatter and reflectance, ^2,3 at different corneal depths. Corneal backscatter analysis by IVCM has been used primarily to define objective haze grading in refractive and lamellar corneal surgery. ^{1,4 –6} These reports however, lack a universal calibration method. Therefore, comparison between machines or over time is unreliable. Only recently, a calibration method has been described that expresses corneal backscatter measured by IVCM in absolute scatter units (SU). ⁷ This method has proven sufficient for measuring backscatter of normal corneas. Opaque corneas require a larger backscatter range and may need a different approach. For this study, two general methods for intra- and interinstrument calibration of corneal backscatter analysis by IVCM were developed. The performances of both methods were verified for a wide backscatter range corresponding to both normal and opaque corneas. 

Methods

Device Settings

Backscatter was measured with a confocal microscope (Confoscan 4; Nidek Technologies, Padova, Italy). At full-thickness setting, this device captures 350 images of 425 × 320 μm within 12 seconds at a focal plane along the z-axis. Scan depth can be adjusted manually in a range of 100 to 1500 μm and scan step in a range of 1.5 to 100 μm. The Confoscan 4 has a 100-W halogen light source (Xenophot HLX 64625; Osram Sylvania, Munich, Germany). A 360° rotary knob allowed analog adjustment of the light intensity (LI) between 0% and 100%. As in a previous study, ⁸ a reference LI of 72% was used, because at this setting the corneal endothelium was visualized optimally, without image saturation. The backscattered and reflected light at the focal plane was captured by a charge-coupled device (CCD) camera. As an immersion substance, one drop of viscous gel (Vidisic Carbomer, 2.0 mg/g; Dr. Mann Pharma, Berlin, Germany) was applied to the 40× objective lens before each examination. 

Setup and Reference Standards

We devised a setup (Fig. 1) to standardize corneal backscatter analysis. A customized holder was used to fix a reference standard at any appropriate angle. To position the reference standard perpendicular to the objective, we centered the specular reflection at the immersion gel to the reference standard interface in the confocal image. Two potential reference standards were compared: AMCO Clear (GFS Chemicals Inc., Powell, OH) and polymethylmethacrylate (PMMA, Perspex GS; Lucite International Ltd., Southampton, UK). 

Figure 1.

 

Calibration setup. (A) A customized holder was fixed to the chin rest of the confocal microscope (Confoscan 4; Nidek Technologies, Padova, Italy), so that PMMA slabs could be measured perpendicularly. (B) A rectangular sample cell was attached to a PMMA slab so that a standard turbidity suspension could be measured.

AMCO Clear is a stable suspension of styrene-di-vinylbenzene copolymer beads that is commercially available as a calibration standard for measuring turbidity. This suspension was examined through a transparent rectangular 10-mm sample cell (LPZ045; Hach Lange, Tiel, The Netherlands). First, the intensity profile of the sample cell with distilled water was measured, to correct for specular reflection at the glass–liquid interface. Next, this setup was measured with AC-4000 (AMCO Clear in a maximum concentration that accounts for 4000 nephelometric turbidity units [NTU]) and, after gradual dilution, with AC-2000, AC-1000, and AC-500. 

The PMMA examinations were performed with three 3-mm-thick slabs with different light transmittances: 26%, 49%, and 65% (opal/white 028, 040, and 030). According to the manufacturer, the light transmittances of the slabs vary by ±3%. 

To determine intra- and intersession repeatability and lot-to-lot variation, we examined three samples from different lots of AC-4000 and PMMA at all three transparencies in triplicate on three different dates. 

Fixed device settings were used for all examinations on reference standards: semiautomatic mode; 72% LI; scan depth, 1500 μm; scan step, 10 μm; and autoalignment function, off. The examination was started after the focal plane had been positioned at a depth of approximately 1400 μm in the reference standard. 

Influence of LI Variation

The PMMA slabs with 26%, 49%, and 65% transparency covered the whole range (grayscale) of image intensities, but AMCO Clear did not. Therefore, the slabs were used to assess the effect of LI variation on the image intensity acquisition. In addition to the 72% LI measurements, the three slabs were measured with 60% and 80% LIs. On the basis of the results, two functions were calculated for adjusting the image intensities measured with 60% and 80% LIs to reference values. 

Subsequently, three eyes randomly selected from three healthy volunteers were examined to verify the performance of these functions. The clear corneas were measured with 60%, 72%, and 80% LIs. To adjust image intensities acquired with 60% and 80% LIs, the corresponding functions (determined by the PMMA slabs) were applied to the image intensity data (measured in vivo on the three normal corneas). After determining mean corneal backscatter for each scan (Appendix A), backscatter values with and without adjustment for LI variation were compared with backscatter values acquired with reference LI. 

All examinations on subjects were performed according to a previously described method. ⁸ The present study was approved by the local Institutional Review Board and adhered to the tenets of the Declaration of Helsinki. Informed consent was obtained from each subject. 

Interinstrument Calibration

To assess the interinstrument difference in image intensity acquisition, we re-examined the PMMA slabs with 72% LI on a second, similar confocal microscope (Confoscan 4; Nidek) and the resulting image intensities were compared with the reference values measured with the first microscope. The mean absolute and relative differences between the two microscopes were determined by using the function that adjusts for LI variation. 

By measuring AMCO Clear on microscope 2 with gradually increasing LIs, we found the LI for microscope 2 that corresponded to the reference LI (72%) of microscope 1. Using this corresponding LI, we re-examined all three PMMA slabs on microscope 2. The results were compared with the 72% LI measurements on microscope 1. 

Data Processing

The z-scan curve is an image intensity–depth profile generated by plotting average pixel intensities of consecutive high-resolution images as a function of depth in the z-axis. ⁹ Exporting raw data from the z-scan curve yielded four main parameters per image: Z MotPos, W MotPos, intensity, and pressure. Only W MotPos and intensity were used to calibrate corneal backscatter measurements. W MotPos is measured by Hall sensors and expresses the distance in micrometers from the measurement starting point to the focal plane where the image was taken. Intensity, in this article expressed as image intensity, accounts for the mean gray level from 0 to 255 of an image pixel sample, sampling 1190 pixels from the image center. These pixels were obtained by sampling, respectively, 34 with 16-pixel spacing, and 35 with 15-pixel spacing in the X and Y directions. For analysis of the AMCO Clear measurements, different z-scan curves were aligned in depth with respect to the specular reflection at the sample cell–AMCO Clear interface. The center of this reflectance peak was set at a depth of 0 μm. Appendix B illustrates the alignment process. The z-scan curves of the PMMA slabs were aligned on the basis of the immersion gel–PMMA interface and the normal corneas were aligned on the basis of the specular reflection of the endothelium. 

With a scan depth of 1500 μm and a scan step of 10 μm, one z-scan curve always contained two complete passes through the reference specimen. Every reference specimen was examined in triplicate, so each one had six complete passes. After alignment, the raw image intensity data of these six passes were plotted against depth, whereupon a smoothing interpolation was applied. A smoothing spline was fitted by visually estimating the smoothing parameter (0.0001), by using a curve-fitting toolbox (MatLab R2009b; The MathWorks. Inc., Natick, MA). 

As the 49% transparent PMMA slab covered a large part of the image intensity range without image saturation, the smoothing spline corresponding to this slab was used to predict image intensities for every micrometer in a 1000-μm depth range (200–1200 μm). This method yielded 1000 image intensity pairs when two LIs or two microscopes were compared. These image intensity pairs were then plotted, and linear regression analysis was performed. The goodness of fit of the linear model was expressed as the coefficient of determination (R ²). This linear model was regarded as the function for adjusting image intensities measured with a different LI or another microscope to values measured with our reference LI. 

Statistical Analysis

Intra- and intersession repeatability and lot-to-lot variation were expressed as the mean coefficient of variation (COV) and as the mean coefficient of repeatability (COR). The COV was defined as the standard deviation divided by the mean image intensity. Because the set of depth values was not consistent between different passes in a z-scan curve, image intensity predictions per scan (obtained by smoothing interpolation) were used to calculate a COV for every micrometer in a 200- to 1200-μm depth range. The COR was derived by multiplying the SD by a t-value of 1.96. 

The three normal corneas were examined to verify intrainstrument calibration for LI variation. Before and after correction for LI variation, differences in mean corneal backscatter were compared with the SD of the mean corneal backscatter for repeated measurements in a normal population. In a pilot study in which the eyes of 19 healthy volunteers were measured eight times a day, the SD of the mean corneal backscatter for repeated measurements was 40 SU (data not shown). 

Results

Specular reflection at the sample cell–distilled water interface affected image intensity up to a depth of 173 μm. Hence, all image intensity data acquired up to a 200-μm depth in the AMCO Clear and PMMA measurements were excluded. AMCO Clear showed less inhomogeneities at the focal plane than did PMMA (Fig. 2). Image intensity in AMCO Clear appeared inversely proportional with depth (Fig. 3A). On closer consideration, however, image intensity profiles of all concentrations behaved as negative exponential functions. Higher concentrations of AMCO Clear showed more image intensity decrease per unit depth than did lower concentrations. Accordingly, the relation between image intensity and turbidity depended on the imaging depth (Fig. 3B). Only at a 200-μm depth did image intensity show a linear relationship with turbidity (R ² = 0.99995). This relationship was nonlinear at deeper levels in the suspension. 

Figure 2.

 

Confocal microscopy images of two different reference standards. (A) AMCO Clear in a concentration of 4000 NTU (GFS Chemicals, Inc., Howell, OH). (B) PMMA slab with 49% transparency. AMCO Clear showed less inhomogeneity than did PMMA.

Figure 3.

 

Characteristics of two different reference standards. (A) AMCO Clear (GFS Chemicals, Inc., Howell, OH) in four concentrations. The AMCO Clear concentrations covered only the lower half of the image intensity spectrum. (B) Effect of imaging depth on the relationship between image intensity and turbidity. Only at 200-μm depth was a linear relationship found. When the focal plane was chosen deeper in the suspension, the relation between image intensity and turbidity was nonlinear. (C) Image intensity profiles of PMMA slabs in three different transparencies. The PMMA slabs covered the whole range of image intensities. (D) Logarithm of the raw image intensity data of the three PMMA slabs showed artificial bending at the higher end of the gray level spectrum, due to image saturation (arrowhead). The lower end of this image intensity spectrum displayed enlarged distribution of the data due to quantization error (arrow). (■) Individual measurements; () smooth interpolation (smoothing spline).

Whereas AMCO Clear in different concentrations covered only part of the range of image intensities, the whole 8-bit range was covered by the three PMMA slabs (Fig. 3C). Similar to AMCO Clear, the image intensity profiles of the PMMA slabs consisted of negative exponential functions, except for the slab with 26% transparency. At the higher end of the image intensity range, the image intensity profile of this slab showed artificial bending toward maximum gray level (255). This phenomenon was attributed to image saturation and was clearly noticeable when image intensity was displayed in logarithmic scale (Fig. 3D). 

When repeatability was compared between the two reference standards, only minor differences were observed (Table 1). PMMA with 26% transparency showed highest intrasession and intersession repeatability. Lot-to-lot variation was comparable between reference standards, with the 65% transparent PMMA slab showing the least variation. Repeated measurements on the PMMA slabs with different transparencies showed a negative correlation between repeatability and the degree of transparency. 

Table 1.

 

Intrasession, Intersession, and Lot-to-Lot Variation in a Depth Range of 200 to 700 μm

	Intrasession		Intersession		Lot-to-lot
	COV	COR	COV	COR	COV	COR
AMCO Clear (4000 NTU)	0.014	2.9	0.016	3.4	0.062	13.8
PMMA (26% transparency)	0.012	4.5	0.004	1.5	0.057	20.7
PMMA (49% transparency)	0.020	3.3	0.006	1.1	0.065	10.3
PMMA (65% transparency)	0.030	1.1	0.029	1.1	0.033	1.2

Varying the LI of the confocal microscope (Fig. 4) resulted in a corresponding linear change in measured image intensity. However, image intensity was affected by saturation from gray levels of 200 and higher. Only below this saturation point did the function to adjust image intensities to reference values show a linear relation. Most of the image intensity range below the saturation point was covered by the 49% transparent PMMA slab. Therefore, this slab was used to calculate the two functions that adjust image intensities measured with 60% and 80% LIs to reference values. Linear regression showed excellent coefficients of determination for the 60% LI (R ² = 0.99995) and the 80% LI (R ² = 0.99989) functions (Fig. 4E). 

Figure 4.

 

Intrainstrument LI variation. (A–C) PMMA slabs with different transparencies measured with different LIs. Markers represent individual data points measured with 80% and 60% LIs. The reference image intensity profile measured with 72% LI is based on Figure 3C. (A) PMMA slab with 26% transparency. (B) PMMA slab with 49% transparency. (C) PMMA slab with 65% transparency. (D–F) Scatterplots of 60% and 80% versus 72% LI data. (D) PMMA slab with 26% transparency. Robust linear fits (gray lines) showed image saturation (arrows) occurred from an image intensity of approximately 200 gray levels. (E) PMMA slab with 49% transparency. Gray lines: standard linear fittings that represent the functions used to adjust image intensities measured with 60% (I _60%) or 80% (I _80%) to reference LI (I _R) values. (a) I _R = 1.537 × I _60% + 5.070; R ² = 0.99995. (b) I _R = 0.7738× I _80% − 2.890; R ² = 0.99989. (F) PMMA slab with 65% transparency. Below an image intensity of approximately 10 gray levels, readings became increasingly unreliable as was shown by robust linear fits (gray lines); I _R, image intensity acquired with reference LI; I _N, image intensity acquired with nonreference LI.

In accordance with the slab measurements, backscatter in the normal corneas increased with a higher LI setting (Table 2, Fig. 5). After correction for LI variation, mean differences reduced from −38.3% to 1.7% (60% LI) and from 33.9% to −0.6% (80% LI). 

Table 2.

 

Mean Corneal Backscatter of Three Normal Eyes Measured with Different Light Intensities

Light Intensity	72%	Nonadjusted			Adjusted
		60%	Difference	(%)	60%	Difference	(%)
Eye 1	1226	799	−428	(−34.9)	1296	70	(5.7)
Eye 2	1155	695	−460	(−39.9)	1177	21	(1.8)
Eye 3	1197	715	−482	(−40.3)	1167	−30	(−2.5)

Light Intensity	72%	Nonadjusted			Adjusted
		80%	Difference	(%)	80%	Difference	(%)
Eye 1	1226	1621	394	(32.1)	1203	−23	(−1.9)
Eye 2	1155	1569	414	(35.8)	1164	8	(0.7)
Eye 3	1197	1601	403	(33.7)	1188	−9	(−0.8)

 (Non-) adjusted: Without or with correction for measurement with nonreference light intensity. Data are expressed in scatter units.

Figure 5.

 

Adjustment of image intensity acquired with nonreference LI to reference values. (A–C) Image intensity profiles before adjustment. (A) Eye 1, (B) eye 2, and (C) eye 3. (D–F) Image intensity profiles after adjustment. (D) Eye 1. A small part of the z-scan curve corresponding with the superficial epithelial layers showed disagreement (arrow) between the reference 72% LI values and the adjusted values. (E) Eye 2 and (F) eye 3. (G) Four characteristic corneal layers of eye 1 measured with different LIs. The endothelium and stroma showing increasing image intensity with increasing LI. Images of the epithelium depicted the disagreement in the z-scan curve of (D), showing higher image intensity with 60% than with 72% LI.

Measurement of the PMMA slabs with similar device settings on both machines showed a large interinstrument difference in image intensity acquisition (Fig. 6A). On the basis of the AMCO Clear measurements, we found 78% LI on microscope 2 to correspond with 72% LI on microscope 1. After LI adjustment of microscope 2, the mean absolute difference decreased from 21.5 to 1.2 gray levels. The mean relative difference decreased from 18.4% to 1.2%. 

Figure 6.

 

Interinstrument calibration. (A) Image intensity data of three PMMA slabs measured with 72% LI by two similar confocal microscopes. (B) Image intensity of three PMMA slabs after LI adjustment of microscope 2 to 78%.

Discussion

Calibration of corneal backscatter measurement is mandatory, as manufacturer settings show large differences between microscopes. AMCO Clear and PMMA, two possible reference standards for calibration of corneal backscatter measurement, were assessed. AMCO Clear had one main advantage over PMMA. This turbidity standard, developed for reference purposes, enabled expression of image intensity in scatter units. In contrast with AMCO Clear, the inhomogeneous structure of PMMA made standardization to absolute values impossible. Nevertheless, this solid reference standard performed better than AMCO Clear on the correction for LI variation. After considering the pros and cons of these potential reference standards, we implemented both in our calibration methods. 

Repeatability

A sample cell was used to determine the image intensity profile of AMCO Clear. Contrary to PMMA measurements, the AMCO Clear setup displayed two interfaces: one from immersion gel to glass and one from glass to AMCO Clear. Part of the incoming light was reflected at the first interface, depending on the shape and regularity of the front surface of the sample cell. Furthermore, part of the incoming light was scattered by the sample cell itself, depending on the cell's transparency. These sample cell characteristics have a large effect on the standardization measurements. ⁷ For standardization of the AMCO Clear calibration setup, a commercially available sample cell should be used, as this improved repeatability. 

Repeated measurements on the three PMMA slabs indicated that the degree of transparency influenced repeatability. A lower COV (better repeatability) was found when transparency of the reference standard decreased. This reduction can be explained by a quantization error due to rounding off to stationary image intensity increments (Fig. 3D). For example, a difference of two gray levels in image intensity will have a larger effect, in terms of percentage, on a mean image intensity of 20 than on a mean of 200 gray levels. The negative correlation between the degree of transparency and repeatability is enhanced by image saturation above image intensity of 200 gray levels, artificially creating a lower COV of the 26% transparent PMMA slab. When comparing COVs of two different reference standards, one should take the effect of transparency on the repeatability into account. In a depth range of 200 to 700 μm, mean image intensity of AC-4000 showed best agreement with PMMA of 49% transparency. Comparison of these two materials showed a slightly better intrasession repeatability of the reference standard with least inhomogeneities (AC-4000). Because the sample cell affected repeatability in the AC-4000 setup, intersession repeatability was clearly in favor of the 49% transparent PMMA slab. Lot-to-lot variation was equal for both materials, despite the fact that only AMCO Clear was developed for reference purposes. 

When reporting corneal backscatter measurements in long-term follow-up studies, care should be taken to correct for small changes in detector sensitivity and illumination intensity. ¹ Although a previous IVCM study has indicated long-term stability of the backscatter detection system, ¹⁰ potential changes have to be quantified. In the authors' opinion PMMA with 49% transparency is first choice to detect changes over time, as this solid reference standard showed the lowest intersession repeatability without image saturation and was most practical in use. Unfortunately, backscatter changes of the PMMA slab over time are largely unknown. These changes however, probably do not outweigh the relatively large lot-to-lot variation for AMCO Clear (Table 1), which is important, as stability of the suspension is guaranteed for only 1 year. Moreover, the PMMA slab can be recalibrated against AMCO Clear to monitor and correct for aging of the PMMA. 

Limited information is available on the repeatability of corneal backscatter techniques. ¹¹ A daily reference scan of 2.5% hydroxypropyl methylcellulose with a tandem scanning confocal microscope showed a COV of 6%. ¹⁰ With a slit scanning confocal microscope, the present study showed better intersession repeatability (COV, 0.4%–2.9%) for solid as well as for liquid reference standards. Considering corneal backscatter measurement in humans, best intersession repeatability has been reported for slit lamp-based haze measurements (COV, 3%–7%). ^12,13 With IVCM, involuntary movements due to pulse, respiration, and ocular microsaccades, negatively affect repeatability of backscatter measurements. ¹¹ These motion artifacts in the z-axis can be reduced by using a z-ring adapter. However, this contact method has a low infection hazard. Without a z-ring adapter, tandem scanning and slit scanning confocal microscopes showed moderate repeatability, with a COV of 35% ¹ and a COR of 15.5 gray levels (Jalbert I, et al. IOVS 2002;43:ARVO E-abstract 1713), respectively. 

Expressing Image Intensity in SUs

Without calibration, a direct relation between image intensity expressed in gray levels and backscatter is lacking. For turbidity measurements the degree of haze has been standardized, defined by the turbidity of a reference suspension in NTU. The relation between turbidity and image intensity can be used to express image intensity in SUs, provided that 1 SU is equal to the image intensity measured in a 1-NTU suspension. ^2,7 To determine this relation in a turbidity range corresponding to normal corneas (200–1500 NTU), McLaren et al. ⁷ used a mean image intensity in a 400-μm depth range. By extending the turbidity range to concentrations that account for more opaque corneas (2000–4000 NTU), we showed that this relation was dependent on the imaging depth (Fig. 3B). Therefore, image intensity should not be averaged over a depth range. Only at a depth of 200 μm can the following linear model ⁷ be used to express image intensity in scatter units (I _SU)  

 

where image intensity (I _R) acquired with our reference LI setting (72%) was characterized by a = 30.13 and b = 116.2. 

Influence of LI Variation

The image intensity profiles of the PMMA slabs (Fig. 3C) showed that decreased light propagation and image saturation affected backscatter measurements in the range corresponding to more opaque corneas. A complex analytical model would be needed to describe these image intensity profiles. Instead, smoothing interpolation was used to obtain smooth image intensity profiles and to assess the influence of LI variation. LI variation was found to affect image intensity (Figs. 4A–C). This subsequently led to a change in model 1 for determining corneal backscatter. Therefore, every time a nonreference LI is used, it is mandatory to determine a new model that describes the relation between turbidity and image intensity specific for this LI. One way to derive this model is by re-examination of different concentrations of AMCO Clear and subsequent determination of a ratio that compensates for LI variations. ⁷ In ophthalmic practice, however, repetition of a reference scan every time a nonreference LI is used is undesirable. Single determination of functions for every relevant nonreference LI is more feasible. After the image saturation point was determined with the 26% transparent PMMA slab, we found these functions by measuring the 49% transparent PMMA slab with different LIs. In accordance with McLaren et al., ⁷ a ratio was found that adjusts for changes in the LI setting. However, an offset had to be included to complete this linear model  

 

where I _R is image intensity adjusted to reference LI, and I _N is image intensity acquired with nonreference LI. Linear regression analysis of a scatterplot with I _N on the x-axis and I _R on the y-axis (Fig. 4E) rendered the parameters a and b. 

First Method

Considering all observations of the two potential reference standards, the following method for intra- and interinstrument calibration is proposed:

1.  
   Choose a reference LI.
2.  
   On the basis of different concentrations of AMCO Clear, find the standardization function (model 1) that characterizes the reference LI and enables expression of image intensity in SU.
3.  
   On the basis of the 49% transparent PMMA slab, determine functions (model 2) for every relevant nonreference LI, that adjust for LI variation.
4.  
   Reduce LI when image saturation occurs (>200 gray levels) and adjust image intensity to reference values afterward.
5.  
   Every week, re-examine the 49% transparent PMMA slab with reference LI to detect degradation of the backscatter detection system.
6.  
   Once a year re-examine a new lot of AMCO Clear (4000 NTU) to detect changes in the transparency of the PMMA slab.

Verification of this calibration method indicated agreement between mean backscatter values after calibration for all three corneas. Differences after standardization were less than ±2 SD (±80 SU) for repeated measurements found in a normal population (data not shown). On closer examination, the first eye showed, after standardization, disagreement in a small part of the z-scan curve (Fig. 5D). This can be explained by morphologic changes in the corneal epithelial layer. Images of the superficial epithelial cells of the first eye (Fig. 5G) showed brightening, when acquired with 80% and, to a greater extent, 60% LI. As image acquisition took place in a fixed order ([1], 72% LI; [2], 80% LI; [3], 60% LI), and other corneal layers appeared unaffected, these changes are probably caused by development of punctate lesions during the repeated measurements on this eye in combination with epithelial exposure to preservatives in the corneal anesthetic eye drop and the immersion gel. 

Second Method

As shown in Figure 5G, LI variation influenced assessment of cell morphology. To enable interinstrument comparison of cell morphology assessment, LI should be matched between two microscopes. For this purpose, the method is as follows:

1.  
   After examination of AC-4000 with reference LI on microscope 1, determine the image intensity measured at 200 μm depth.
2.  
   Find the corresponding LI on microscope 2 by measuring AC-4000 by gradually varying the LI until similar image intensity is measured at a 200-μm depth.
3.  
   If necessary, verification is obtained by measuring the 49% PMMA slab with reference LI on microscope 1 and with corresponding LI on microscope 2.

This second method can be used as an alternative one for interinstrument calibration of corneal backscatter analysis. However, matching LI between microscopes is coarser than the first calibration method. Eventually, only 6% LI change was needed to correct for 18.4% difference in mean image intensity. Because LI was adjusted by a small 360° rotary knob, fine tuning was fairly difficult. Despite this coarse LI setting, a minor difference of 1.2% in mean image intensity after calibration was found.

With the availability of IVCM the number of reports describing corneal backscatter has increased in the past few years. Nevertheless, this feature is still not widely used in a clinical setting. Reports have been primarily focused on the evaluation of surgical techniques including photorefractive keratectomy, ^{1,10,14 –16} laser in situ keratomileusis, ^{4,7,17 –23} deep lamellar keratoplasty, ⁵ and Descemet's stripping (automated) endothelial keratoplasty. ^6,24,25 However, other interesting potential applications for corneal backscatter analysis, such as assessment of corneal hydration ²⁶ and quantification of the effects on corneal backscatter of diabetes, ^27,28 contact lens wear, ^29,30 and fluorescein drops, ³¹ have been suggested. Although our results point out the necessity of calibration of corneal backscatter analysis, most IVCM reports lack this standardization. Nonstandardized backscatter measurements can be used to compare relative values within one study. However, these studies cannot be reproduced by other researchers. Application of our calibration methods enables comparison of standardized absolute backscatter values between studies on a similar subject and allows introduction of backscatter measurements into prospective multicenter trails. 

Besides motion artifacts due to involuntary eye movements, two other major limiting factors in corneal backscatter analysis by IVCM remain. First, with a magnification up to 500 times, only a small portion of the cornea is imaged, 0.14% of the corneal surface. ⁸ In combination with a moderate positional repeatability in the paracentral and peripheral corneal regions, backscatter analysis by IVCM is limited to a miniscule area of the central cornea and therefore seems less suited for follow-up of nonhomogenous corneal opacities. Second, backscatter, as measured by IVCM, is affected by interface reflection of two layers with different refractive indices, such as the endothelium-to-anterior chamber interface. Only with polarizing filters can backscatter be separated from reflectance. ^2,32 However, no commercially available device that measures corneal backscatter is equipped with these filters. 

Several other imaging techniques, including Scheimpflug photography, ^{33 –40} slit lamp photometry, ^{2,13,41 –49} CCD camera systems, ^{12,32,50 –55} opacity lensometers, ⁵⁶ high-frequency ultrasound, ^57,58 and anterior segment optical coherence tomography (AS-OCT), ^59,60 have been used to measure corneal backscatter. Only some of these studies have standardized their measurements, using liquid (formazin, ⁶¹ sodium fluorescein, ¹³ latex microspheres, ^28,37 styrene-di-vinylbenzene copolymer beads ^{2,7,45 –48} ) or solid (neutral-density filters, ^32,51,52,54 fluorescent glass, ³⁰ and thermoplastic resin [Spectralon; Labsphere, Vinkeveen, The Netherlands] ^43,44,49 ) reference standards. This diversity indicates the lack of a universal reference standard for calibration of backscatter measurement. A complicating factor in finding such a universal reference standard is the use of different measurement angles and wavelengths by the different imaging techniques. Because backscatter shows dependency on these factors, ^62,63 these techniques are noninterchangeable. Currently, the best option for a universal reference standard is a turbidity standard solution. Yet, a solid homogenous material that is stable over time and has a low lot-to-lot variation would be more practical. 

Of the noncontact imaging techniques, AS-OCT shows the greatest potential for objective backscatter measurement, whereas this technique allows high depth penetration and high-resolution imaging in opaque corneas, when compared to other techniques. Repeatability will be even further enhanced when AS-OCT is equipped with an eye-tracking system. Despite its great potential for measuring corneal backscatter, AS-OCT is still surpassed by IVCM in terms of lateral resolution. Especially the unique combination of objective backscatter quantification and assessment of cellular morphology is essential for monitoring corneal diseases and consequent therapeutic actions. Supplementary to slit lamp examination, IVCM serves this purpose best. 

In summary, we have demonstrated with large interinstrument differences the necessity of standardized corneal backscatter analysis. Combination of two reference standards enabled intra- and interinstrument calibration of backscatter measurement in a wide range of corneal opacities. With standardization, IVCM combines objective backscatter measurement with assessment of corneal morphology, thus allowing clinical evaluation of treatment modalities for prevention of haze formation. 

Footnotes

 Presented at the annual meeting of the Association for Research in Vision and Ophthalmology Annual Meeting, Fort Lauderdale, Florida, May 2010.

Footnotes

 Supported by the Research Foundation SWOO Flieringa, Rotterdam; The Dutch Cornea Foundation, Rotterdam; the OOG Foundation, ‘s Gravenzande, The Netherlands; and an ARVO International Travel Grant. None of the funding organizations had a role in the design or conduct of this research.

Footnotes

 Disclosure: T. Hillenaar, None; V.A.D.P. Sicam, None; K.A. Vermeer, None; B. Braaf, None; L. Remeijer, None; R.H.H. Cals, None; J.F. de Boer, None

Appendix A

Mean Corneal Backscatter

Figure A1.

 

(A) With a scan step of 6 μm, a full-thickness scan of an average cornea (550-μm-thick) always contained three complete passes. When a scan was performed with a nonreference LI, image intensities were adjusted to reference values. (B) After adjustment for LI variation, image intensities were standardized to SUs, on the basis of the standardization function derived from the AMCO Clear measurements (model 1) (GFS Chemicals, Inc., Howell, OH). All backscatter values lower than 400 SU and also the incomplete fourth pass were discarded (shaded areas). A mean backscatter value was determined for each pass by summing all the backscatter values in that pass and dividing them by the number of frames. Mean corneal backscatter was the average of the three passes. This example: pass 1, 99730/83 = 1202; pass 2, 106789/87 = 1227; and pass 3, 100018/86 = 1163. Mean corneal backscatter: (1202 + 1227 + 1163)/3 = 1197 SU.

Appendix B

Alignment Process

Figure A2.

 

(A) The z-scan curve of the PMMA slab with 49% transparency measured on microscope 1. At fixed settings (semiautomatic mode; 72% light intensity; scan depth, 1500 μm; scan step, 10 μm; and autoalignment function, off) the z-scan curve of all reference standards always contained two complete passes. Alignment was based on the PMMA–coupling gel interface (arrow). The center of this reflectance peak was set to a depth of 0 μm. The following model was used to calculate depth for every image in the z-scan curve: depth of the current image = W MotPos of the interface minus W MotPos of the current image. All images in the shaded areas that corresponded to a negative calculated depth or were part of the third incomplete pass were discarded. (B) After alignment, image intensity data were plotted against depth. Because the specular reflection of the PMMA–coupling gel interface influenced the image intensity profile up to a 173-μm depth, all images below the 200-μm depth were discarded. In addition, all images above the 1200-μm depth were discarded, to create a standard 1000-μm depth range. After examining the PMMA slab in triplicate and the subsequent data processing, a smoothing interpolation was applied to the image intensity–depth plot (Fig. 3C).