Assessing Age-Related Changes in Corneal Densitometry Parameters With Anterior Segment OCT Speckle

Abstract

Purpose: The purpose of this study was to assess in vivo regional variability in the densitometry parameters of corneal stroma and the modulating effect of age on those parameters using statistical characterization of optical coherence tomography (OCT) speckle.

Methods: OCT imaging of central and peripheral cornea was performed in a group of 20 younger (24 to 30 years old) and 19 older (50 to 87 years old) subjects. The sample size was estimated using normal assumptions and previously reported data on speckle parameter variability. Statistical parameters of corneal OCT speckle were calculated in the regions of interest (ROI) encompassing central and peripheral stroma as well as taking into account their anterior and posterior subregions. Both parametric (Burr-2 parameters: α and k) and a nonparametric approach (contrast ratio [CR]) were considered. Two-way analysis of variance was used to test for differences in densitometry parameters with respect to ROI position and age.

Results: Both approaches showed statistically significant differences within the ROI positions (all P < 0.001 for α, k, and CR) and age (P < 0.001, P = 0.002, and P = 0.003, for α, k, and CR, respectively) indicating substantial stromal asymmetry. Additionally, CR showed statistically significant differences between anterior and posterior subregions (P < 0.001).

Conclusions: Corneal OCT-based densitometry is inherently asymmetrical and are influenced by age. The results indicate that regional variability of stromal structure is not limited to the central and peripheral regions but that differences exist also between the nasal and temporal parts of the cornea.

Translational Relevance: The in vivo acquired parameters of corneal OCT speckle can be used to indirectly assess corneal structure.

Introduction

The human cornea is not rotationally symmetric and this feature applies not only to its topography and thickness but also to optical and biomechanical properties, in which structural elements play an important role. The topographic asymmetry of the corneal surface, omitting that with significant central astigmatism, increases toward the periphery.¹ This asymmetry is particular evident in the nasal and temporal regions of the cornea.² Additionally, both topography and thickness of cornea are modulated by age.^3–6 At the microstructural level, the optical anisotropy – also a form of asymmetry – of the human cornea has been revealed using polarizing microscopy,⁷ Mueller type polarimetry,⁸ and polarization sensitive optical coherence tomography (PS-OCT).⁹ Further, corneal anisotropy derives from the same molecular mechanism that determines the transparency, refractive function, and biomechanics of the cornea.^10,11 

Knowledge about corneal asymmetry is of clinical relevance when examining the biomechanical response of the peripheral corneal regions after photorefractive keratectomy,¹² assessing the risk of postsurgical ectasia after laser in situ keratomileusis,¹³ and evaluating the efficacy of surgical procedures involving small incision lenticule extraction,^14,15 to mention just a few. Clinically available instrumentation for assessing corneal asymmetry in vivo is usually limited to measuring corneal topography, thickness, and some biomechanical parameters,¹⁶ whereas the assessment of corneal asymmetry at the microstructural level is more difficult to attain. 

Recently, there has been increasing interest in using the parameters obtained from standard optical coherence tomography (OCT) B-scans of the cornea, whereby the statistics of stromal speckle are examined.^17,18 Speckle is a phenomenon occurring in imaging samples with coherent or partially coherent light. It manifests itself in the form of a granular noise affecting the image. It was shown that the parameters of OCT speckle are influenced by intraocular pressure and that they have the potential to be indirectly used for assessing changes in corneal stroma in both ex vivo and in vivo studies.^19,20 Changes in corneal OCT speckle induced by crosslinking have been directly linked to changes at the molecular level, assessed with spectroscopy.²¹ Moreover, OCT speckle analysis has been recently used to assess corneal densitometry parameters.²² Conventionally, densitometry is associated with estimating the mean pixel intensity in a specific region of an image (Scheimpflug or OCT), whereas the speckle approach requires image preprocessing that extracts the amplitude OCT signal from the image. Nevertheless, the statistical parameters of OCT speckle are often directly proportional to the first moment (mean) of the distribution. Consequently, it is expected that speckle parameters are strongly correlated with mean pixel-intensity densitometry. 

The aim of this study was to evaluate regional variations in the densitometry parameters of the cornea, using both parametric and nonparametric methods of OCT speckle characterization. Additionally, the modulating effect of age on those speckle characteristics was considered. 

Methods

The study was conducted with the participation of Caucasian volunteers in two age groups: a group of younger subjects, denoted in short as group Y, aged between 18 and 30 years, and a group of older subjects, denoted as group O, aged over 50 years. Using normal approximation, the sample size has been estimated at n = 19 based on the results of OCT speckle analysis reported earlier,¹⁷ assuming a 5% level of significance and 90% of power to discriminate less than 10% change in speckle parameters. Exclusion criteria encompassed any pathology in corneal scarring, such as dystrophies and trauma, any ectatic condition, such as keratoconus and medical history of cornea or intraocular surgery. Subjects who used contact lenses had to abstain from their use at least 24 hours prior to the measurements because speckle statistics can be influenced by corneal hypoxia or swelling.¹⁷ Subjects were excluded if they complained or were diagnosed with the dry eye disease. The study was conducted in accordance with the tenets of the Declaration of Helsinki and was approved by the institutional ethics board. Each subject was briefed on the purpose and conduct of the study and provided written consent to participate in the study. 

To study the potential regional variations in corneal densitometry parameters, both biometry and OCT were utilized. The statistics of corneal OCT speckle depend on intraocular pressure (IOP).¹⁹ Hence, to ensure that there were no substantial differences in IOP between the two considered groups, air-puff tonometry measurement with Corvis ST (Oculus Optikgeräte GmbH, Wetzlar, Germany) were performed. Three consecutive measurements with breaks of about 2 minutes were acquired for each of the subjects and the median IOP value was taken for further analysis. Biometry was primarily used to accurately measure the central corneal thickness (CCT) and it was performed with IOLMaster 700 (Carl Zeiss Meditec AG, Jena, Germany) whereas tomography was performed with SOCT REVO 80 (Optopol Technology, Zawiercie, Poland) with axial resolution of about 5 µm and lateral resolution ranging between 12 and 18 µm. The device has the ability to image the anterior segment of the eye by applying an appropriate measurement protocol. Both instruments were switched on in advance before conducting the measurements to stabilize their working temperatures. All measurements were taken at one time of the day between 9:00 AM and 12:00 PM, to minimize the effect of diurnal changes in biometry,²³ IOP, and OCT speckle parameters. 

The test eye was selected randomly. The measurement of CCT was performed according to the standard on-axis IOLMaster 700 procedure. OCT examination was performed for on- and off-axis eye positions to allow imaging central and corneo-scleral regions. The OCT imaging acquisition protocol was as follows. The anterior scan type, B-scan program, 5 mm scan width, and 12,032 A-scans comprising each B-scan were selected. Two scan angles (0 degrees and 90 degrees) were considered for on-axis imaging of the central part of the cornea and one scan angle (0 degrees) for off-axis imaging of the nasal and temporal corneo-scleral regions. At each scan angle and each eye position (i.e. central, nasal and temporal), three image acquisitions were performed. To facilitate imaging consistency and ensuring the correct imaging angle for the off-axis nasal and temporal acquisitions, subjects looked to both sides through appropriately positioned mirrors attached to the OCT device head, focusing their gaze at an external target behind their backs. For the central on-axis eye position, subjects were asked to fixate on the instrument's default target. All acquisitions of corneal cross-section were performed within the instrument's depth of focus using specially provided guiding lines. 

Estimation of Corneal OCT Speckle Parameters

Technical details of statistical characterization of corneal OCT speckle have been reported earlier.¹⁸ Below, a short summary that includes additional stages of the procedure is outlined. An OCT B-scan (see Fig. 1) saved as a bitmap image file is input into custom algorithm written in Matlab (MathWorks, Inc. Natick, MA, USA). During the preprocessing stage, the OCT images underwent a normalization process in which their pixel values were divided by the maximum value of 255. Subsequently, the inverse-log transformation was applied. The transformed pixel values were further subjected to normalization to lead speckle values between 0 and 1. The first step of the image processing procedure included delineation of the air/epithelium interface. In the case of an off-axis image (see Fig. 1B), it was rotated to its horizontal position (see Fig. 1C) according to a linear trend determined by the air/epithelium outline. After rotation, the scale of the image was appropriately adjusted. In the second step, a predefined region of interest (ROI) was selected. The ROI width is 1 mm whereas its depth is 350 µm. The ROI is located 60 µm from the estimated air/epithelium outline. The ROI size and its position have been selected experimentally taking into account its centration within the B-scan, the range of measured corneal thicknesses as well as the changing corneal microstructure toward the limbus. In the next step, the log-transformation, applied automatically to every OCT B-scan in the device software, was inversed. Additionally, two subregions of ROI were considered: the anterior ROI and the posterior ROI, whereby the ROI (further in the paper referred to as the full ROI) is divided horizontally into two equal parts.²⁰ 

Figure 1.

 

Illustrative B-scans for the central on-axis (A) and off-axis nasal (B, C) eye positions with the region of interest (ROI) outlined in red. The dashed yellow line indicates the air/epithelium interface. The nasal scan B undergoes a rotation C for further analysis (see full text).

The transformed image pixels collected within the ROI are used for both parametric and nonparametric characterization of stromal speckle. For the parametric model, the two-parameter Burr distribution (Burr-2) was considered, as it is both theoretically and practically justified.^18,24,25 The Burr distribution is characterized by scale and shape parameters (α and k). The mean of the Burr distribution is proportional to α and, hence, the latter can be interpreted as the densitometry parameter.²² For the nonparametric (distribution-free) approach, the speckle contrast ratio (CR), defined as the ratio of the standard deviation of the amplitude to its mean value, is considered. 

Statistical Analyses

Median values of the parameters α, k, and CR were obtained from each series of three measurements and used for further analyses that included testing for the hypothesis of normality using the Shapiro-Wilk test, testing for the equality of variances using Levene's test, and 2-way ANOVA with post hoc tests (including correction for multiple comparisons). The factors considered included: ROI position (temporal, central, and nasal), age (young and old), ROI type (anterior and posterior), and, for the central on-axis measurements, the B-scan position (horizontal and vertical). Correlation between OCT-based densitometry parameters and that based on sample mean was evaluated using Pearson's correlation coefficient. For the tests, the level of significance was set to 0.05 whereas when multiple comparisons were made, it was set to 0.017. 

Results

The scans that were analyzed were collected from a group of 20 younger subjects (12 women and 8 men), aged 24 to 30 years old (mean ± standard deviation = 25 ± 1.9 years) and a group of 19 older subjects (12 women and 7 men), aged 50 to 87 years old (61 ± 8.6 years). Average CCT for was 549 ± 24 µm and 548 ± 36 µm for group Y and group O, respectively, and there was no statistically significant difference in this parameter between the groups (t-test: P = 0.432). Median IOP ranged between 13 and 22 mm Hg (mean ± standard deviation = 16.95 ± 2.18 mm Hg for group Y and 16.43 ± 2.06 mm Hg for group O). No significant differences were found in IOP between the age groups (t-test, P = 0.456). 

Parametric Approach

For full ROI, when examining the B-scan position factor (for on-axis measurements) no statistically significant differences between the parameters of the Burr distribution acquired from horizontal and vertical scans were noted (t-test: P = 0.134 and P = 0.135, for α and k, respectively). When considering those parameters and two factors of ROI position (temporal, central, and nasal) and age, the 2-way ANOVA did not indicate statistically significant changes in age (P = 0.086 and P = 0.421 for α and k, respectively) but showed to differentiate the ROI position. Figure 2 shows the results of the mean ± standard error of the mean (SEM) of the Burr parameters for full ROI. For the shape parameter α (see Fig. 2A) statistically significant differences were noted between any two ROI positions, whereas for the shape parameter k (see Fig. 2B) statistically significant differences were noted between central and temporal and central and nasal ROI positions but not between temporal and nasal ROI position. In addition, the variability of the two parameters (in terms of SEM), in the considered population, was greater at the temporal and nasal ROI position than at the central ROI position. 

Figure 2.

 

Parametric approach, full ROI. Variations of the scale (A) and shape (B) parameters of the Burr distribution for both the group O and group Y combined according to the ROI position (temporal, central, and nasal).

When data from the anterior and posterior ROIs were treated separately, no statistically significant differences between the parameters of the Burr distribution acquired from anterior and posterior ROI were noted (t-test: P = 0.137 and P = 0.543 for α and k, respectively). Hence, in the following, the data from anterior ROI was concatenated with that of the posterior ROI. Further, when considering distributional parameters and two factors of ROI position (temporal, central, and nasal) and age, the 2-way ANOVA indicated statistically significant changes in age (P < 0.001 and P = 0.002 for α and k, respectively) as well as in the ROI position (both P < 0.001). Figure 3 shows the results of the mean ± SEM of the parameters for the two distinct ROIs considered separately. Additionally, the two considered age groups are shown separately. It is worth noting that both α and k differentiate the two groups in the temporal ROI position but not in the central one. In the nasal ROI position, the difference between the groups is evident but smaller than those in the temporal ROI position. In addition, α differentiates the age groups at the nasal ROI position whereas k does not. 

Figure 3.

 

Parametric approach, data from ROIs concatenated. Variations of the scale (A) and shape (B) parameters of the Burr distribution according to the ROI position (temporal, central, and nasal) for group O (solid red) and group Y (dashed green).

Nonparametric Approach

For full ROI, when examining the B-scan position factor (for on-axis measurements) no statistically significant difference between CR acquired from horizontal and vertical scans was noted (t-test: P = 0.315). When the two factors of ROI position (temporal, central, and nasal) and age were taken into account, the 2-way ANOVA indicated statistically significant changes in CR with age (P = 0.003) and ROI position (P < 0.001). Figure 4 shows the variations of CR according to the ROI separately for each age group together with the results of post hoc analysis. 

Figure 4.

 

Nonparametric approach, full ROI. Variations of CR according to the ROI position (temporal, central, and nasal) for group O (solid red) and group Y (dashed green).

Unlike in the parametric approach, when data from the anterior and posterior ROIs were treated separately, statistically significant difference in CR was apparent (t-test: P < 0.001). Hence, data from the anterior ROI were treated separately to that of the posterior ROI. Figures 5A and 5B indicate the variations of CR according to the ROI position for its anterior and posterior part, respectively, separately for each age group together with the results of post hoc analysis. 

Figure 5.

 

Nonparametric approach, data from ROIs treated separately. Variations of CR according to the ROI position (temporal, central, and nasal) for group O (solid red) and group Y (dashed green). (A) Anterior ROI, and (B) posterior ROI.

OCT-Based Versus Conventional Densitometry

Conventional densitometry, expressed as mean pixel intensity in a specific ROI, can be linked to the speckle parameters that represent OCT-based densitometry. In particular, those parameters that are proportional to (α) or inversely proportional (CR) to the sample mean. Correlation between the parameter α and densitometry value was high and statistically significant (R = 0.907 and R = 0.754, both P < 0.001, for group Y and group O, respectively). Similarly, high and statistically significant correlation was found between CR and densitometry value (R = −0.879 and R = −0.748, both P < 0.001, for group Y and group O, respectively). The parameter k, which is not directly proportional to the sample mean, had lower correlation with mean-based densitometry (R = 0.565, P < 0.001 and R = 0.355, P = 0.036, for group Y and group O, respectively). 

Discussion

Differences in corneal stroma structure between the temporal and nasal sides have been examined by Daxer and Fratzl,²⁶ who assessed the degree of human cornea collagen fibril orientation ex vivo, 5 hours postmortem (5 subjects with the median age of 71 years). From their results, it can be observed that the mean degree of orientation is higher in the nasal side than that in the temporal one. In addition, the nasal side has higher variability of this orientation. It is unknown whether the examined in this study CR parameter, describing here the assessed corneal OCT speckle, represents in any way that orientation. However, the correspondence between the CR results and those of collagen fibril orientation is evident, although not examined formally. In the future, it would be of interest to develop specially designed phantoms to confirm any such relationship. Differences between central and peripheral cornea regions is not only evident in its thickness² but also in the stroma microstructure.¹⁰ It is of interest to note that when corneal OCT speckle is analyzed parametrically, both scale and shape parameters of the Burr distribution have much less variation in the central part of the cornea than those in its peripheral parts, for both young and old subjects. This can be attributed to a more regular microstructure of central stroma than that of the peripheral one. The obvious benefit of the densitometry-type parameters derived from examining the corneal OCT speckle is they are in vivo nature. Although some of the OCT-based densitometry parameters exhibit high correlation with correspondingly evaluated mean-based densitometry, they offer additional statistical information on the distribution of the backscattered light and, most importantly, physically model the OCT speckle.²⁴ 

Corneal asymmetry may also have an iatrogenic character. For example, Serrao et al.¹² found that corneal asymmetry, expressed by its anterior curvature, changes after a photorefractive keratectomy. Similarly, Vinciguerra et al.²⁷ showed corneal asymmetry changes in terms of the thinnest cornea point after corneal cross-linking. Further, they suggest modifications to existing algorithms for cornea ablation to alleviate such differences. Having the opportunity to examine not only the corneal curvature but also its microstructure in vivo (here, in an indirect manner of examining the corneal OCT speckle) may help reduce the additionally induced corneal asymmetry due to the surgical treatment. 

Age-related changes to the corneal structures have been studied from different perspectives using a number of imaging techniques. It is also well established that corneal densitometry assessed with Scheimpflug imaging changes with age.²⁸ In general, increase in the corneal collagen fiber size as well as decrease in susceptibility to proteolytic degradation and the ability to swell are observed.²⁹ The latest has also been observed by analyzing age related changes to the corneal OCT speckle.¹⁷ In addition, Faragher et al.³⁰ noted that the aging process results in disorganization of collagen fibers. This study partially supports those findings because the CR parameter is shown to decrease with age whereas the decrease in CR corresponds to an increase in number of scatterers, as it was shown earlier for the phantoms with varying levels of particle concentration.¹⁸ It should also be noted that such disorganization of collagen fibers, here, indirectly evaluated with CR, appears mainly in the noncentral parts of the cornea with the nasal part mostly affected, where statistically significant changes between the two considered age groups were observed. 

Another aspect of cornea aging relates to the continuing exposure of the anterior eye to the ultraviolet (UV) radiation. Golu et al.,³¹ in an experimental study with rodent corneas (rats), noted that most of the UV energy is absorbed by the anterior layers of the cornea and that changes between the anterior and posterior corneal layers are evident after exposition to UV radiation. Similar findings have been reported by Kolozsvári et al.,³² who spectroscopically examined slices of human cornea (treating epithelium separately) and concluded that the UV-B absorption (range considered from 240 to 400 nm) is up to 80% higher in the anterior 100 µm of the cornea than that in the posterior layers. In the current study, differences in corneal OCT speckle parameters between the anterior and posterior ROI were also observed and those differences were statistically significant when a nonparametric approach to the speckle analysis was used. What is more intriguing is that age-related changes in corneal OCT speckle parameters follow the spectroscopy-based results of Doutch et al.,³³ who showed that the stromal UV transmission increases from the center of the cornea toward its periphery, and that this increase was most evident at a wavelength of 400 nm (UV-B). The observed differences in corneal OCT speckle parameters in different stroma regions may also be the results of changes in stroma hydration that is higher in the posterior part.³⁴ What is more, the differences in the parameters between the nasal and temporal sites of the stroma may also be an effect of UV radiation, which extreme rays mostly concentrate at the nasal region.³⁵ 

Some studies find asymmetry between left and right corneas. Boote et al.³⁶ mapped the orientation and distribution of fibrillar collagen across the human corneas using wide-angle x-ray scattering, discovering that the corneas of left and right eyes are structurally distinct, exhibiting a degree of midline symmetry with characteristic rhombus-shaped contour pattern. In this study, the tested eye of each subject was selected randomly, so in the context of the work of Boote et al. this could be considered as a limitation. However, for the statistical analysis, the images of the left eyes were flipped to correspond to those of the right eyes. 

A limitation of the study is the lack of examining the superior and inferior parts of the cornea. OCT imaging of those corneal parts was difficult to perform in a way to standardize the angle of gaze, because of the construction of the OCT casing. 

Summarizing, OCT speckle parameters can be utilized for assessing corneal densitometry parameters. The results of this study closely correspond to those of ex vivo studies performed with laboratory instrumentation, usually unavailable in clinical practice. 

Acknowledgments

Supported by Wroclaw University of Science and Technology statutory funds. 

Disclosure: A. Fojcik, None; A. Kościółek, None; D.R. Iskander, None