The Effect of Glaucoma on the Optical Attenuation Coefficient of the Retinal Nerve Fiber Layer in Spectral Domain Optical Coherence Tomography Images

Abstract

Purpose.: To demonstrate the effect of glaucoma on the optical attenuation coefficient of the retinal nerve fiber layer (RNFL) in Spectral Domain Optical Coherence Tomography (SD-OCT) images.

Methods.: We analyzed images of the peripapillary areas in 10 healthy and 30 glaucomatous eyes (mild, moderate, and advanced glaucoma, 10 eyes each), scanned with the Spectralis OCT (Heidelberg Engineering GmbH, Dossenheim, Germany). To calculate the RNFL attenuation coefficient (μ_att), determined by the scattering properties of the RNFL, we used a model that normalized the reflectivity of the RNFL by the retinal pigment epithelium. The analysis was performed at four preset locations at 1.3 and 1.7 mm from the center of the optic nerve head (ONH) (i.e., temporally, superiorly, nasally, and inferiorly) and on averages per eye. To assess the structure-function relationship, we correlated the μ_att to the mean deviation (MD) in standard automated perimetry.

Results.: The μ_att of the RNFL decreased up to 40% with increasing disease severity, on average as well as in each location around the ONH (Jonckheere-Terpstra test, P < 0.019 in all tests). The μ_att of the RNFL depended significantly on the location around the ONH in all eyes (Kruskal-Wallis test, P < 0.014) and was lowest nasally from the ONH. The μ_att correlated significantly with the MD in SAP (R ² = 0.337).

Conclusions.: The measurements clearly demonstrated that the μ_att of the RNFL decreased with increasing disease severity. The RNFL attenuation coefficient may serve as a new method to quantify glaucoma in SD-OCT images.

Introduction

In glaucoma, retinal nerve fibers are damaged and lost, leading to thinning of the retinal nerve fiber layer (RNFL). ^1,2 Optical coherence tomography (OCT), a technique used for in vivo retinal imaging, for example, can measure the thickness of retinal layers, such as the RNFL. Since its development ^3,4 for ophthalmology, this technology has been predominantly used for imaging the macula and macular diseases and, to a lesser extent, glaucoma. Time domain OCT (TD-OCT) yielded a gradually improving reproducibility of measurements over the past decade. ^5–9 Nevertheless, it never exceeded the diagnostic capabilities of other structural measurement techniques for glaucoma, such as scanning laser polarimetry (SLP) and confocal scanning laser ophthalmoscopy (CSLO). ^9–12 Since the recent development of Spectral Domain Optical Coherence Tomography (SD-OCT), ^13–16 OCT is increasingly used in glaucoma assessment because of its high image resolution, high reproducibility of measurements, ^17–23 and diagnostic accuracy. ^24–26  

The local signal strength of the OCT is used to measure thickness of retinal layers. Optical scattering properties of each retinal layer are closely related to the tissue properties. Pons et al. ²⁷ reported a decreased reflectivity of the RNFL in glaucoma measured with TD-OCT. In a recent study, we confirmed the diminished reflectivity of the RNFL in glaucomatous eyes compared to healthy eyes in SD-OCT images (van der Schoot J, et al. IOVS 2010;51:ARVO E-Abstract 212). Even focal defects are detected by this method (Vermeer KA, et al. IOVS 2011;52:ARVO E-Abstract 3666). Figure 1 illustrates the difference in reflectivity of the RNFL between a healthy and a glaucomatous eye. These findings suggest that glaucomatous damage is not only reflected by the thickness of the RNFL, but also by changes in its optical scattering properties. The latter may potentially lead to a new method to diagnose or follow glaucoma. 

Figure 1.

 

SD-OCT image. (A) Healthy eye. (B) Advanced glaucomatous eye. The two colored areas represent the manual segmentation of the areas of the RNFL (blue) and RPE (purple), used for analysis. Note the reduced brightness of the RNFL in the glaucomatous eye compared to the healthy eye, representing the diminished RNFL reflectivity in the glaucomatous eye.

To determine the optical scattering properties of a retinal layer, it is not sufficient to measure its reflectivity. The measured reflectivity of a layer is sensitive to media opacities and light collection geometry, for example. Therefore, a relative measurement is necessary to extract quantitative scattering properties from OCT images. In this article, a method to quantify the attenuation coefficient (μ_att) of the RNFL in SD-OCT images was presented, by taking the scattering properties of the retinal pigment epithelium as a reference. To investigate the μ_att of the RNFL as a clinical diagnostic parameter, the μ_att of the RNFL in healthy and glaucomatous eyes was measured. 

Patients and Methods

Participants

Datasets of 10 healthy subjects and 30 subjects with glaucoma were randomly selected from a larger population participating in a longitudinal prospective follow-up study into glaucoma imaging. The selection of the 30 glaucomatous subjects was based on the severity of their glaucoma per eye, stratified by the mean deviation (MD) of the visual field test (determined with the standard 24-2 W/W SITA test program of the Humphrey Field Analyzer; Carl Zeiss Meditec, Dublin, CA). The severity of glaucoma was defined as an MD between 0 dB and −6 dB, between −6 dB and −12 dB, and worse than −12 dB, for mild, moderate, and advanced glaucoma, respectively. Of each group, 10 eyes of different subjects were randomly selected. 

For inclusion, a healthy subject needed to have a normal visual field and an intraocular pressure below 22 mmHg. A normal visual field had, by definition, a Glaucoma Hemifield Test within normal limits and an MD and PSD within the 95% confidence intervals. Visual fields were considered as glaucomatous if they had at least two of the following on at least two separate occasions: a PSD significant at the 5% probability level, a Glaucoma Hemifield Test outside normal limits, and a cluster of more than 2 points below the 5% probability level or 1 individual point below the 1% probability level. All patients were recruited from the population treated for glaucoma at the Rotterdam Eye Hospital. 

Subjects were excluded from participation in the presence of any coexisting ocular or systemic disease known to possibly affect the visual field (e.g., diabetes mellitus); a history of intraocular surgery (except for uncomplicated cataract surgery or glaucoma surgery in glaucoma patients); uncontrollable arterial hypertension, and secondary glaucoma in case of glaucoma. 

All participants had a best-corrected Snellen visual acuity of at least 20/40 in both eyes, a spherical equivalent refractive error between −10.0D and +5.0D, and had unremarkable findings upon slit-lamp examination, including open angles on gonioscopy. 

The methods and procedures used in this study adhered to the Declaration of Helsinki. Informed consent was obtained from all participants. This study was approved by the Medical Ethics Committee of the Erasmus Medical Centre, Rotterdam, The Netherlands. 

All participants were scanned once with the Spectralis OCT (Heidelberg Engineering, Dossenheim, Germany) between January and September 2009. The optic disc and peripapillary area were scanned by means of a volume scan of 20 by 20 degrees. This scan contained 193 B-scans, each consisting of 512 A-scans. Each B-scan was an average of 5 scans on the same location (made possible by the built-in eye-tracking system on the Spectralis OCT with ART value 5). The lateral distance between adjacent B-scans was 30 μm. To minimize the decrease in sensitivity as a function of depth, all retinal cross sections were placed in the upper one-third of the scan depth. A volume scan was excluded, if the proprietary overall quality score provided by the Spectralis was below 15 dB or if the volume scan was incomplete due to the built-in maximum acquisition time of 300 seconds. 

Image Processing and Model

The exported raw data of optic nerve head scans was used for all analyses. To determine the reflectivity, the raw data was exported, which represents the measured intensity on a linear scale between 0 and 1. The device displays these intensities in a grayscale image after applying the following formula: Y = ⁴ √X; where X represents the raw data values; Y = 0 represents no reflectivity and Y = 1 maximum reflectivity. ²⁸  

The model developed for this study, was based on two different scattering layers: the RNFL and the RPE. Assuming that the backscatter properties of the RPE are constant, it was used to normalize the RNFL OCT signal. The model can be summarized as follows: The amount of light that is backscattered from the RNFL is proportional to the incident intensity and the amount by which the incident beam on the eye is attenuated due to scattering. The incident light that is not scattered by the RNFL propagates to the RPE, and is partially backscattered by the RPE. The ratio of total light backscattered by the RNFL (RNFL reflectivity) and the RPE (RPE reflectivity) is now related to the μ_att of the RNFL according to the following formula:  

 

 

in which μ_att represents the attenuation coefficient in mm⁻¹; R is the ratio between the reflectivity of the RNFL and RPE; and d is the thickness of the RNFL in mm. The model is derived in the Appendix. 

The analysis of the μ_att was performed in all four quadrants around the ONH at fixed locations. The center of the ONH was manually selected, taking the border of the RPE as a reference. From this center of the ONH, the areas of interest were selected 1.3 mm superiorly, inferiorly, temporally, and nasally. To correct for any failed B-scans, the best out of 3 B-scans closest to the selected distance was selected for analysis. Per B-scan, the area selected for analysis was 20 pixels wide. If the selected area of 20 pixels contained any blood vessels, the 20 pixels wide window was shifted along the B-scan toward an area without blood vessels, to avoid any scatter caused by the blood vessel. The μ_att in all 4 peripapillary locations was determined separately and the mean μ_att per patient was calculated by averaging the μ_att of these four locations. The error of the mean μ_att was calculated over the patient population. 

To obtain the reflectivity of each retinal layer separately, the layers were manually segmented and color coded by a manual segmentation tool, called ITKSNAP. ²⁹ The color coding was manually performed by a trained physician. The reflectivity of the RNFL and the RPE were used for analysis. An example of this color coding is presented in Figure 1A. 

Of the three selected B-scans (one superiorly; one inferiorly; and one running from nasally to temporally to the ONH, containing the nasal and the temporal 20-pixel wide area of interest), the quality scores (as given by the device) were reported ranging from 0 dB to 40 dB. 

We correlated this measure of structure (μ_att) with a summary measure of function (i.e., the MD of the visual field test), which was performed by all participants during the same visit as the Spectralis measurement. 

For comparison of the study outcomes to currently common clinical protocols of SDOCT, the mean μ_att was also calculated at a 1.7-mm distance from the ONH center by the same methods explained earlier. These results have been presented as supplementary material. 

Statistics

An analysis of generalized estimating equations was used to test for differences in the demographics. The Jonckheere-Terpstra test for trend was used for comparison between the μ_att of healthy eyes and of the three groups of glaucomatous eyes and to test for a trend in the μ_att between the various locations around the ONH. The Kruskall-Wallis test was used to test for the presence of an effect of location on μ_att. The coefficient of determination (R ² ) of exponential regression analysis between MD and μ_att was calculated. A significance level of 0.05 was considered to be statistically significant. 

Results

Between the four groups of randomly selected subjects, there were no statistically significant differences in age and scan quality (Table 1). The MD showed a statistically significant difference between the four groups, decreasing with increasing disease severity (Table 1). No patients were excluded because of bad scan quality (<15 dB) or incomplete scans. Examples of an SD-OCT image of a healthy eye and of a glaucomatous eye are shown in Figure 1. 

Table 1.

 

Patient Demographics and Scan Quality: Analysis of Generalized Estimating Equations for Differences between Age, Mean Deviation (MD), and Scan Quality

	Healthy	Mild Glaucoma	Moderate Glaucoma	Advanced Glaucoma	P Value
Mean age (years)	63.7 [SD 9.9]	65.3 [SD 1.4]	66.1 [SD 7.1]	71.8 [SD 9.3]	0.18
MD (dB)	0.15 [SD 1.10]	−2.80 [SD 1.39]	−9.05 [SD 3.64]	−17.65 [SD 4.31]	<0.001
Scan quality (dB)	26.5 [SD 3.6]	24.0 [SD 3.9]	24.4 [SD 2.9]	24.5 [SD 3.8]	0.2

The mean attenuation coefficients (μ_att) of the healthy and the mild, moderate, and advanced glaucomatous eyes, as well as for each location, have been presented in Table 2. The average μ_att of the RNFL showed a statistically significant trend for decreasing μ_att with increasing disease severity (Jonckheere-Terpstra test, P < 0.001 for both analyses at 1.3 and 1.7 mm from the center of the ONH). Figure 2 (see also Supplementary Fig. 1) shows the trend of decreasing average μ_att of healthy and glaucomatous eyes, in order of increasing in severity. 

Table 2.

 

Mean (± Standard Error) Attenuation Coefficient μ_att at 1.3 mm from the Center of the ONH in Healthy and Glaucomatous Eyes [mm⁻¹]

	Attenuation Coefficient of the Retinal Nerve Fiber Layer
	Average	Temporal	Superior	Nasal	Inferior
Healthy	4.78 ± 0.46	5.25 ± 0.59	5.31 ± 0.53	3.60 ± 0.53	4.96 ± 0.42
Mild glaucoma	4.09 ± 0.34	4.72 ± 0.72	4.41 ± 0.47	2.73 ± 0.46	4.51 ± 0.38
Moderate glaucoma	3.14 ± 0.22	3.33 ± 0.49	3.99 ± 0.49	1.74 ± 0.20	3.51 ± 0.46
Advanced glaucoma	2.93 ± 0.33	3.54 ± 0.54	3.67 ± 0.46	2.02 ± 0.26	2.48 ± 0.42

Figure 2.

 

Box plot of the average attenuation coefficient μ_att (mm⁻¹) at 1.3 mm from the center of the optic nerve head in healthy and glaucomatous eyes.

The location significantly affected the μ_att of the RNFL in all four groups (Kruskal-Wallis test, P < 0.001 for analyses at 1.3 mm and P < 0.019 for analyses at 1.7 mm from the ONH center). In these eyes, the μ_att of the RNFL was lower nasally than in the other locations. All locations showed a statistically significant trend for decreasing μ_att with increasing disease severity (Jonckheere-Terpstra test, P < 0.001 for analyses at 1.3 mm and P < 0.014 for analyses at 1.7 mm from the ONH center for temporal, superior, nasal, and inferior). These findings are illustrated by Figure 3 (see also Supplementary Fig. 2). 

Figure 3.

 

Plot presenting the trend of decreasing attenuation coefficient μ_att (mm⁻¹) with increasing disease severity for all locations separately at 1.3 mm from the center of the optic nerve head. Note the difference in μ_att between the nasal and the other locations.

A statistically significant relationship was also found between the MD and μ_att (R ² = 0.337 and 0.250, P ≤ 0.001, for 1.3 and 1.7 mm from the center of the ONH, respectively). The plot of this correlation has been presented in Figure 4 (see also Supplementary Fig. 3). 

Figure 4.

 

Plot presenting the correlation between the attenuation coefficient μ_att (mm⁻¹) at 1.3 mm from the center of the optic nerve head and the mean deviation (dB), from the visual field test. The coefficient of determination (R ² ) of exponential regression analysis between MD and μ_att is 0.337 (P < 0.001).

Discussion

The attenuation coefficient (μ_att) of the RNFL decreased with increasing disease severity in SD-OCT images of healthy and glaucomatous eyes. We created a model to quantify the optical scattering properties from OCT data by calculating μ_att. The model was based on the reflectivity of the RNFL and the RPE, on interaction between light and the tissue and on the thickness of the RNFL. This novel analysis confirms earlier conclusions (i.e., that the optical scattering properties of the RNFL change in glaucoma) ²⁷ (van der Schoot J, et al. IOVS 2010;51:ARVO E-Abstract 212). In this current study, the attenuation coefficient was examined, because it is based on relative measurements of the reflectivity. Absolute measurements of reflectivity might be affected by a number of factors, such as the particular alignment of the OCT beam on the cornea, and media opacities. Relative measurements are less sensitive to these factors and thus provide a more reliable measure of change related to pathological processes in glaucoma. 

The model developed for this study incorporated the backscatter signal of the RNFL and the RPE and not of the layers in between. These layers have a relatively low backscatter signal, compared to the RNFL and RPE, ³⁰ and they attenuate only a relatively small part of the incident light. Moreover, the attenuation of these layers is uncorrelated with disease severity and was therefore not incorporated in our model. The backscatter properties of the RPE were assumed to be constant. This assumption does not hold in case of peripapillary atrophy. In glaucoma, the RPE near the ONH can be affected by PPA, an important cause of irregular RPE around the ONH. PPA can be divided into two zones, the alpha and beta zone. The alpha zone is characterized by pigmentary irregularities in the RPE, while the beta zone, which is nearest to the ONH, correlates with a complete loss of RPE cells. ³¹ The alpha zone is nearest to the location of the analyses conducted for this study, which was 1.3 mm and 1.7 mm from the center of the ONH. It is sometimes difficult to exactly delineate the peripheral border of alpha zone PPA. In retrospect, only 1.0% of the regions of analysis of our included eyes potentially contained some alpha zone PPA. Therefore, we feel confident that the PPA did not significantly affect our dataset. 

It was hypothesized that μ_att indirectly measures the nerve fiber density of the RNFL. Evidently, there is loss of nerve fibers due to glaucoma. ² In a histology study, Quigley et al. ¹ measured a decrease of the fiber density in glaucomatous eyes compared to healthy eyes. A structural change due to glaucoma (i.e., the decreased density of the nerve fibers) is also thought to be associated with a decrease in birefringence of the RNFL. ^32,33 A decreased nerve fiber density will also have its effect on the optical scattering properties of the RNFL. The denser the tissue is, the more the incident light is attenuated by the RNFL. Study findings indicated that the μ_att of the RNFL diminished with increasing disease severity. A reduction of μ_att could be an early sign of glaucomatous damage. This hypothesis is supported by the findings of Huang and others. Firstly, they found that 50% of the RNFL reflectance is caused by microtubules ³⁴ ; and secondly, they found a change in reflectance of the RNFL to precede thinning of the RNFL. ³⁵ Future studies are needed to elucidate the relation between μ_att, reflectance, and glaucomatous damage. 

The location significantly affected the μ_att of the RNFL. μ_att was lower nasally from the ONH, compared to the other locations (temporally, superiorly, and inferiorly) in both healthy and in glaucomatous eyes. These differences in μ_att appear to be unrelated to glaucoma, because they occurred in both glaucomatous and healthy eyes. The difference between the retina nasally from the ONH and the other locations is that the RNFL is usually thinner nasally. More importantly, the incident light being reflected by the RNFL nasally is probably less compared to the other locations because of a different angle of this part of the retina. The nasal side of the ONH is more toward the periphery, leading to a higher retinal curvature perpendicular on the light beam. The less perpendicular the RNFL is to the light beam and—consequently—the larger the scan angle is, the smaller the reflectance of the cylindrical nerve fibers will be. ³⁶ This might explain the lower μ_att nasally compared to the other locations. 

Moreover, a structure-function relationship was assessed by correlating μ_att to the MD. In previously published studies, in which the average RNFL thickness was used as a structural measure, and an index of visual fields as the functional measure, the structure-function relationship was generally quite weak and variable (R ² from 0.08 to 0.55). ^37–40 In this current study, the structure-function relationship between μ_att and MD showed a correlation well within the same range (R ² = 0.337 and 0.250 for 1.3 and 1.7 mm from the ONH center, respectively). Compared with earlier structure-function studies, ^37–40 this result is promising, since the model in this study is a first attempt of characterizing RNFL scattering properties in SD-OCT images. Further development in modeling the relationship between the μ_att and an index of visual fields is needed to explore the use of μ_att as a measure for structure in characterizing a structure-function relationship. 

To our knowledge, SD-OCT currently does not consistently outperform several other available techniques in their diagnostic accuracy, such as SLP and CSLO.^9-12 With every available imaging technique, it is notably difficult to detect early glaucoma, because of the large biological variation between healthy eyes and the inherent overlap with glaucomatous eyes. A model was created for SD-OCT to quantify the attenuation of light, which is related to the optical scattering properties of the RNFL. With our method, we found a clear effect of glaucoma on the μ_att of the RNFL. The correlation of μ_att with standard automated perimetry is comparable to that of SLP and CSLO with SAP. ^37–40 We speculate that the reduced μ_att of the RNFL in SD-OCT images in glaucomatous eyes might serve as a discriminating feature for structural changes. Future studies are needed to clarify the role of the μ_att of the RNFL in diagnosing and following glaucoma with SD-OCT, potentially leading to a new method to quantify glaucoma in SD-OCT images. 

Supplementary Materials

Doc S1 

JPG S2 

JPG S3 

JPG S4 

Footnotes

 Supported in part by several Dutch foundations for Research in Ophthalmology (in random order): The Rotterdam Eye Hospital Research Foundation, Rotterdam, The Netherlands; Stichting Oogfonds Nederland, Utrecht, The Netherlands; Glaucoomfonds, Utrecht, The Netherlands; Landelijke Stichting voor Blinden en Slechtzienden, Utrecht, The Netherlands; Stichting voor Ooglijders, Rotterdam, The Netherlands; Stichting Nederlands Oogheelkundig Onderzoek, Nijmegen, The Netherlands.

Footnotes

 Disclosure: J. van der Schoot, None; K.A. Vermeer, None; J.F. de Boer, P; H.G. Lemij, None

Appendix

Model

The proposed model is based on two different scattering layers: the retinal nerve fiber layer and the retinal pigment epithelium. The former is the layer of interest while the latter will serve as an internal calibration layer. These layers are also the two most scattering layers in the retina. Other, weakly scattering retinal layers are not included. Assuming constant RPE backscatter properties, this layer is used to normalize the OCT signal from the RNFL. First, a model of the interaction between light and tissue is described. This model is then applied to the RNFL and the RPE, producing a method to determine the RNFL attenuation coefficient. 

Attenuation and Scattering of a Homogeneous Layer.

The incoming beam, with a power of I ₀ when it arrives at the front surface of the tissue, is attenuated by the tissue according to dI/dx = −μ _att · I(x). The power of the beam at depth x is then given by I(x) = I ₀ e ^−μx , where μ _att is the attenuation coefficient of the tissue. Combining these equations produces the attenuation of the light at depth x: μ _att ·I(x) = μ _att ·I ₀ e ^−μx . A fraction of the attenuated light, given by α, is backscattered. Before reaching the detector, this light is again attenuated by the tissue. The power of the light at the detector that was backscattered at depth x is then given by α · μ _att · I ₀ e ^−2μx . Integrating over a depth range d results in the total power of the backscattered signal at the detector 

 

 

 

RNFL.

The RNFL is the first retinal layer that the incident light beam interacts with. The total backscattered signal for the RNFL is given by 

 

 

 

RPE.

Light that reaches the RPE is first attenuated by the RNFL. After interacting with the RPE, the backscattered light is again attenuated by the RNFL. The total backscattered signal for the RPE is therefore given by 

 

 

 

Deriving μ.

Calculating the ratio of the total OCT signal of the RNFL and the RPE yields 

 

 

 

 

 

 

 

Assuming that α _RNFL and α _RPE are constant, and that the attenuation of the RPE is constant (i.e., μ _RPE and d _RPE are constant), this equation reduces to  

 

 

 

Solving this equation for μ _att results in 

 

 

 

Calculating β.

In equation 6, μ can be calculated if R, d, and β are known. Both R and d can be determined from a segmented OCT scan. Given that β is a constant in our model, it may be estimated from the data. For estimating β, equation 5 was used. The dataset contained R and d, determined from OCT data of healthy eyes, and μ and β were fitted to the model by minimizing the error 

 

 

 

The result is shown in Figure A1 and the optimal fit was found for β = 2.3 (and μ _att = 4.6 mm⁻¹). 

Figure A1 .

 

Scatter plot of the ratio (R) between the reflectivity of the RNFL and RPE against the RNFL thickness (d) for healthy eyes.