Glaucoma Diagnostic Ability of the New Circumpapillary Retinal Nerve Fiber Layer Thickness Analysis Based on Bruch's Membrane Opening

Abstract

Purpose: To compare the diagnostic ability of the new spectral-domain optical coherence tomography (SD-OCT) algorithm for measuring circumpapillary retinal nerve fiber layer (RNFL) thickness centered on Bruch's membrane opening (BMO), with the conventional circumpapillary RNFL thickness measurement centered on the optic disc.

Methods: In 75 eyes with primary open-angle glaucoma (POAG) and 71 healthy control eyes, circumpapillary RNFL thickness was measured with SD-OCT, first by centering the scan circle on the optic disc (RNFL_Di), and then on the BMO (RNFL_BMO). Difference between the topographic profiles of RNFL_Di and RNFL_BMO was compared and factors influencing any discrepancies between methods were investigated. Glaucoma diagnostic abilities of each method were assessed using the areas under receiver operating characteristic curve (AUCs).

Results: Axial length did not differ between POAG and healthy eyes. A longer axial length and larger width of externally oblique border tissue (BT) associated with tilted optic disc were the two major factors influencing discrepancies between RNFL_BMO and RNFL_Di (both P < 0.001). Compared with RNFL_BMO, RNFL_Di tended to result in a thinner nasal RNFL in eyes with externally oblique BT, while RNFL_BMO and RNFL_Di were comparable in eyes without externally oblique BT. The glaucoma diagnostic capabilities were generally comparable, but RNFL_BMO was superior to RNFL_Di in eyes having a larger width (>250 μm) of externally oblique BT (AUC = 0.933 vs. 0.843, respectively, P = 0.027).

Conclusions: The new circumpapillary RNFL scanning algorithm centered on BMO may provide a more reliable RNFL profile in eyes with tilted optic discs, with a largely comparable glaucoma diagnostic ability to the conventional algorithm.

A precise evaluation of retinal nerve fiber layer (RNFL) is crucial for diagnosing and monitoring glaucoma. The evaluation of circumpapillary RNFL thickness using optical coherence tomography (OCT) has been shown to provide objective and reliable information on glaucomatous optic nerve damage, and has been widely used over several decades.^1–8 The recent advent of spectral-domain OCT (SD-OCT) has enabled a more precise measurement of RNFL thickness and improved its performance compared with earlier time-domain OCT.^9–13 

A precise evaluation of circumpapillary RNFL thickness requires correct centering of the OCT scan circle.^14,15 The conventional SD-OCT technique requires manual positioning of the scan circle by the device operator, normally on the optic nerve head (ONH). However, correct delineation of the optic disc margin is sometimes challenging and varies among observers,¹⁶ which may produce measurement variability, or false-positive and false-negative circumpapillary RNFL thickness results. 

It has recently been suggested that Bruch's membrane (BM) opening (BMO) is a more reliable reference than the clinical disc margin.¹⁷ While the clinical disc margin does not have a single anatomic structure and thus its determination can vary among observers, BMO is a clear anatomic structure and thus may provide a more consistent reference.^17–20 Based on this concept, new software was developed for an OCT device (Spectralis; Heidelberg Engineering, Heidelberg, Germany) that automatically detects the BMO and measures the circumpapillary RNFL thickness centered on BMO. The topographic orientation of RNFL for thickness evaluation is based on the foveal–BMO axis, taking into account the cyclotorsion of individual eyes. 

The present study compared measurements of circumpapillary RNFL thickness centered on BMO and the clinical optic disc, and determined whether the former measurement could help to diagnose glaucoma. 

Methods

Participants

This study was based on a review of data from patients with primary open-angle glaucoma (POAG) included in the investigating glaucoma progression study (IGPS), which is an ongoing prospective study that has been underway since August 2011 at the Glaucoma Clinic of Seoul National University Bundang Hospital. Healthy subjects were also included as a control. This study was approved by the institutional review board of Seoul National University Bundang Hospital and conformed to the tenets of the Declaration of Helsinki. All of the subjects provided written informed consent to participate. 

The database of patients included in the IGPS between September 2011 and March 2015 was reviewed. Subject medical records were reviewed for medical history and results of a complete ophthalmic examination, which included visual acuity assessment; refraction; slit-lamp biomicroscopy; gonioscopy; Goldmann applanation tonometry; dilated stereoscopic examination of the optic disc; central corneal thickness measurement (Orbscan II; Bausch & Lomb Surgical, Rochester, NY, USA); axial length measurement (IOL Master version 5; Carl-Zeiss Meditec, Dublin, CA, USA); corneal curvature measurement (KR-1800; Topcon, Tokyo, Japan); stereo disc photography; red-free fundus photography; SD-OCT (Heidelberg Engineering); and standard automated perimetry (Humphrey Field Analyzer II 750, 24-2 Swedish interactive threshold algorithm; Carl-Zeiss Meditec). 

Primary open-angle glaucoma was defined as the presence of glaucomatous optic nerve damage (neuroretinal rim notching or thinning, or RNFL defect) and associated visual-field defects without ocular disease or conditions that may elevate intraocular pressure (IOP). A glaucomatous visual-field defect was defined as being present when one or more of the following criteria were fulfilled: (1) values outside the normal limits on the glaucoma hemifield test; (2) ≥3 abnormal points, with a probability of being normal of P < 0.05, and 1 point with a pattern standard deviation (PSD) of P < 0.01; or (3) a pattern standard deviation of P < 0.05. Those visual-field defects were confirmed on two consecutive reliable tests (fixation loss rate, ≤20%; false-positive and false-negative error rates, ≤25%). 

The inclusion criteria for the control group were eyes with a baseline IOP of < 22 mm Hg without antiglaucoma medication, a normal-appearing optic disc, and a normal visual field. A normal-appearing optic disc was defined as the absence of glaucomatous optic neuropathy and pallor or swelling of the optic disc. A normal visual field was defined as the absence of glaucomatous visual-field defects and neurologic field defects. Healthy control patients were matched with the glaucoma group in terms of age and axial length, using a frequency-matching method. 

For both POAG patients and healthy controls, eyes with a visual acuity of <20/40, a spherical refraction of < −10.0 or > +3.0 diopters (D), and a cylinder correction of > ±3.0 D were excluded. Eyes with any abnormalities that affected the circumpapillary scan ring where the SD-OCT RNFL thickness measurements were obtained; a history of ocular surgery other than cataract extraction; intraocular disease (e.g., diabetic retinopathy or retinal vein occlusion); or neurologic disease (e.g., pituitary tumor) that could cause visual-field loss were also excluded. If both eyes were eligible, one eye was randomly selected as the study eye. 

The baseline IOP was defined as the mean of at least two measurements made within 2 weeks before initiating IOP-lowering treatment for the POAG group and the mean of at least two measurements made within 4 weeks of each other for the control group. Disc ovality was defined as the ratio of the longest diameter to the shortest diameter of the optic disc as measured on a color disc photograph.²¹ 

Measurement of Circumpapillary RNFL Thickness

The peripapillary area was imaged by SD-OCT (Heidelberg Engineering), first by centering the circular scan on the optic disc (RNFL_Di) and then on BMO (RNFL_BMO; Fig. 1). The corneal curvature of each eye was entered into the OCT system (Heidelberg Engineering) before performing SD-OCT scanning to compensate for potential magnification error. We measured the RNFL_Di using the conventional circular scan mode, in which the operator centered the scan circle approximately on the ONH. The scan circle spanned 12° of arc, with the diameter in millimeters depending on the axial length. The software of the OCT device (Heidelberg Engineering) provides a global average RNFL thickness and a mean RNFL thickness for each of the six sectors relative to the foveal–disc axis, as follows: nasal–superior (NS, 90–135°); nasal (N, 135–225°); nasal–inferior (NI, 225–270°); temporal–inferior (TI, 270–315°); temporal (T, 315–45°); and temporal–superior (TS, 45–90°). 

Figure 1

 

Measurements of the circumpapillary RNFL thickness using the new method (A–D) and the conventional method (E). (A) Determination of the center of the BMO. (A-1) Infrared image indicating the 24 locations where the radial B-scan images were obtained. (A-2) Radial B-scan image at the location indicated by the light-green arrow on the infrared image. The boundary of the BMO (red dots) is determined by defining the two BM termination points (red squares) in each of the 24 radial B-scan images centered on the optic disc. Blue asterisk indicates the center of the BMO demarcated by the red dots. Circumpapillary RNFL thicknesses are then measured using the scan circles centered on the BMO (blue asterisks) with diameters of 3.5, 4.1, and 4.7 mm, which measure (B) RNFL_BMO1, (C) RNFL_BMO2, (D) RNFL_BMO3, respectively. (E) Conventional RNFL_Di measurement uses the scan circle manually located by the examiner (light-green asterisk), which spans 12° of arc (the diameter in millimeters depends on the axial length).

We measured the RNFL_BMO using the new software, which obtains circular scan images centered on BMO. In this mode, the OCT device (Heidelberg Engineering) automatically detects BMO in 24 high-resolution, 15° radial scans of the ONH, each averaged from 20 to 30 individual B-scans, with 1536 A-scans per B-scan acquired with scanning speed of 40,000 A-scans/second.^22,23 Then, three circular scans along peripapillary circles with diameters of 3.5, 4.1, and 4.7 mm measure three sets of circumpapillary RNFL thicknesses centered on BMO (RNFL_BMO1, RNFL_BMO2, and RNFL_BMO3, respectively). Each scan circle produces a global average, and the mean thickness for each of the six sectors relative to the foveal–BMO axis, is as follows: NS (85–125°); N (125–235°); NI (235–275°); TI (275–315°); T (315–45°); and TS (45–85°). 

The foveal–disc and foveal–BMO axes are obtained automatically at the time of RNFL_Di and RNFL_BMO scanning, respectively. The foveal–disc (or foveal–BMO) axis was defined as the angle between the fovea and the optic disc (or BMO) center relative to the horizontal axis of the image-acquisition frame. 

Eyes were excluded when the OCT did not detect the proper location of BMO during RNFL_BMO measurement, and when the diameter of the scan circle in the conventional RNFL_Di scan differed by more than 1 mm relative to the diameter of the RNFL_BMO1 scan (3.5 mm). 

Distance between the centers of the scan circles for the RNFL_Di and RNFL_BMO1 was assessed by overlapping the two infrared fundus images provided by the circumpapillary RNFL thickness measurement protocol of the OCT device (Heidelberg Engineering). The distance was measured using ImageJ software (http://imagej.nih.gov/ij/; provided in the public domain by the National Institutes of Health, Bethesda, MD, USA). 

Determination of the Width of the β-Zone Parapapillary Atrophy and the Externally Oblique Border Tissue

The widths of β-zone parapapillary atrophy (PPA) and the externally oblique border tissue (BT) of Elschnig were measured on the SD-OCT optic disc scans, as described elsewhere.²⁴ Based on previous reports that the clinical characteristics differ according to the extent of BM within the PPA area,^24,25 the PPA area was divided into β-zone PPA area (the area with BM) and the area associated with externally oblique BT (PPA without BM). The distance from the disc margin to the boundary of BM termination, and that from the BM termination to the PPA margin were defined as the widths of the externally oblique BT and the β-zone PPA, respectively. Measurements were obtained on the radial scan images that had been obtained at the time of RNFL_BMO scanning. The three radial scans containing the largest dimension of the PPA were selected from 24 images. The widths of the externally oblique BT and the β-zone PPA were measured in each of the three radial scans, and the mean values of the three measurements were defined as the maximum widths of the externally oblique BT and the β-zone PPA, respectively. 

Measurements were obtained by an observer (EJL) who was masked to the subjects' clinical information. Excellent interobserver reproducibility for the measurement of PPA width has been reported previously.²⁴ 

Statistical Analysis

The intermethod agreement between the global RNFL_BMO and the global RNFL_Di was assessed by calculating the intraclass correlation coefficient and its 95% confidence limits, and also by Bland-Altman plot. Comparison between groups was performed using the independent-samples t-test. 

Factors influencing the difference between the global RNFL_BMO and the global RNFL_Di were determined using univariate and multivariate linear regression analyses. The multivariate analysis was performed for variables with statistical significance of P < 0.10 in the univariate analysis. To avoid problems of multicollinearity, the multivariate analysis was performed in several ways when there were factors correlated with each other. Multicollinearity between factors was assessed by calculating variance inflation factors (VIF). 

The pattern of sectorial RNFL thicknesses was compared between RNFL_BMO and RNFL_Di, first using the paired t-test and then using generalized estimating equation analysis to account for the regional differences between the RNFL thicknesses in each sector, independent of the systemic difference between RNFL_BMO and RNFL_Di.^26–28 In the generalized estimating equation analysis, the T-sector RNFL thickness was set as a reference, because the RNFL thickness in the T sectors has been reported to be the least variable.^29,30 

The ability of RNFL_BMO and RNFL_Di to discriminate between POAG and control eyes was assessed using the area under the receiver operating characteristic (ROC) curves (AUC) and compared between methods. The receiver operating characteristic regression model was used to account for the multiple covariates that could influence on the diagnostic abilities of each method.^31–34 

Statistical analyses were performed using statistical software (SPSS, version 20.0; SPSS, Chicago, IL, USA, and Stata, version 13.0; StataCorp, College Station, TX, USA). Except where stated otherwise, the data are presented as mean ± standard deviation values. The cutoff for statistical significance was set at P < 0.05. The raw data for t-tests were subjected to Bonferroni correction on the basis of the number of comparisons within each analysis. 

Results

The study initially included 141 POAG and 120 healthy subjects. Of these, 102 were excluded because of errors in BMO detection (n = 77), and a difference in the scan circle diameter of more than 1 mm between methods (n = 25). Of the remaining 159 eyes, a further 13 were excluded because of error in the RNFL segmentation using either method, leaving 75 POAG and 71 control eyes in the final analysis. 

Table 1 gives the baseline demographics and clinical characteristics of the participants. Subjects in the POAG group had a higher baseline IOP (P < 0.001); worse visual-field mean deviation and pattern standard deviation (both P < 0.001); a larger maximum β-zone PPA width (P = 0.001); and a more-oval disc (P = 0.039). There was no difference in the axial length between the POAG and healthy eyes. 

Table 1

 

Baseline Characteristics of the Participants

 

The intermethod intraclass correlation coefficient between the RNFL_Di and the RNFL_BMO measurements was 0.965 (95% confidence interval [CI], 0.952–0.975) for RNFL_BMO1; 0.947 (95% CI, 0.927–0.961) for RNFL_BMO2; and 0.899 (95% CI, 0.863–0.926) for RNFL_BMO3. Figure 2 shows the Bland-Altman plot demonstrating the agreement between the global RNFL thicknesses measured using the conventional scan circle centered on the optic disc (RNFL_Di) and that measured using the three scan circles centered on BMO (RNFL_BMO1, RNFL_BMO2, and RNFL_BMO3). The best agreement was found for RNFL_BMO1 and RNFL_Di (Fig. 2A). 

Figure 2

 

Bland-Altman plots showing the agreement between the conventional and new methods in measuring the global RNFL thickness. Graphs comparing the conventional (RNFL_Di) and new (RNFL_BMO) methods for different scan circle diameters ([A] RNFL_BMO1; [B] RNFL_BMO2, and [C] RNFL_BMO3). The solid line indicates the ordinary least-squares line of best fit, with the two dotted lines showing the upper and lower 95% limits of agreements.

Table 2 gives the factors associated with discrepant global RNFL_BMO1 and RNFL_Di measurements. In the univariate analysis, younger age, a longer axial length, a larger maximum width of externally oblique BT, a larger disc ovality, and a larger distance between the scan circles for RNFL_BMO1 and RNFL_Di were more associated with larger global RNFL_BMO1 than global RNFL_Di measurements (all P < 0.001). Because the maximum width of the externally oblique BT and the distance between the center of the scan circles had relatively large VIFs (5.288 and 3.784, respectively) the multivariate analysis was performed in two ways to account for the multicollinearity between the two variables: each of the two factors was separately included in multivariate analyses 1 and 2, respectively. The multivariate analysis revealed a significant influence of a longer axial length, a larger maximum width of the externally oblique BT, and a larger distance between the center of the scan circles (all P < 0.001). 

Table 2

 

Factors Associated With the Difference Between the Conventional and the New Methods to Measure the Global RNFL Thickness

 

The pattern of sectorial RNFL thicknesses was also compared between RNFL_Di and RNFL_BMO1. To this end, the eyes were assigned to four subgroups according to the presence of the externally oblique BT, since this was one of the two independent factors found to be independently associated with the discrepancy between the global RNFL (i.e., maximum width of externally oblique BT and axial length; Table 2), and it was considered a clearer cutoff. In both the healthy and POAG groups, eyes with externally oblique BT were younger (43.3 ± 15.0 vs. 58.5 ± 13.2 years, P < 0.001, and 49.8 ± 13.0 vs. 58.3 ± 13.4 years, P = 0.010, in the healthy and POAG groups, respectively) and had a longer axial length (25.9 ± 0.9 vs. 23.9 ± 1.2 mm, P < 0.001, and 25.6 ± 1.1 vs. 24.2 ± 1.4 mm, P < 0.001, in the healthy and POAG groups, respectively) and a more-oval optic disc (1.19 ± 0.13 vs. 1.08 ± 0.06, P < 0.001, and 1.28 ± 0.18 vs. 1.09 ± 0.06, P < 0.001, in the healthy and POAG groups, respectively) compared with those without externally oblique BT. 

Table 3 compares the global and sectorial RNFL thicknesses between eyes with externally oblique BT and those without in each healthy and POAG group. We found the RNFL_BMO1 did not differ between the groups either globally or in any of the sectors in both the healthy and POAG groups. However, the RNFL_Di was thinner in healthy eyes with externally oblique BT than in those without both globally (P = 0.006) and in the NI (P = 0.006) and N (P = 0.001) sectors. In the POAG group, RNFL_Di did not differ between eyes with and without externally oblique BT in all areas. 

Table 3

 

Comparison Between the Conventional and New Methods to Measure the Circumpapillary RNFL Thicknesses Depending on the Presence of the Externally Oblique BT in Healthy and POAG Groups

 

Figure 3 compares between RNFL_Di and the RNFL_BMO1 in healthy eyes with externally oblique BT and those without externally oblique BT. Paired comparison of the absolute circumpapillary RNFL thicknesses obtained by the two methods revealed a relatively thicker RNFL_BMO1 than RNFL_Di in both eyes with externally oblique BT and those without externally oblique BT. Generalized estimating equation analysis revealed that the pattern difference between the two methods existed only in eyes with externally oblique BT. When the T-sector RNFL thickness was set as a reference, RNFL_Di was significantly thinner in the NI, N, and NS sectors than RNFL_BMO1 in the corresponding sectors, only in eyes with externally oblique BT (P = 0.042, P < 0.001, and P = 0.002, respectively, by generalized estimating equation analysis; Fig. 3A). The generalized estimating equation analysis did not reveal such differences in any of the sectors in eyes without externally oblique BT (Fig. 3B). 

Figure 3

 

Comparison between the conventional and new methods in measuring the circumpapillary RNFL thicknesses in healthy eyes, depending on the presence of externally oblique BT. Asterisks indicate significant differences between the circumpapillary RNFL thicknesses (paired t-test: P ≤ 0.0071, 0.05/7). Double daggers indicate significant differences between the circumpapillary RNFL thicknesses relative to the difference in the temporal sector (dagger).

Figure 4 illustrates a representative case of a healthy myopic eye with externally oblique BT in which the differential RNFL profiles between the methods resulted in a false-positive finding in the TS sectors using the conventional algorithm. 

Figure 4

 

Comparison between the two methods in measuring the circumpapillary RNFL thicknesses in a nonglaucomatous healthy eye. (A) Color disc photograph showing a myopic optic disc with tilt and torsion. (B) Standard automated perimetry revealing a normal visual field. (C) Circumpapillary RNFL thickness measurement using the scan circle centered on the BMO (RNFL_BMO). (C-1) Infrared image showing the three scan circles centered on BMO. The innermost light-green circle has a diameter of 3.5 mm and measures the RNFL_BMO1. (C-2) Circular diagram showing borderline abnormality in the NS sector. (C-3) However, the RNFL thinning is unremarkable in the temporal–superior–nasal–inferior–temporal (TSNIT) curve. (D) Circumpapillary RNFL thickness measurement using the scan circle centered on the optic disc (RNFL_Di). (D-1) Infrared image showing the scan circle centered on the disc margin. The diameter of the scan circle is set at 12° and was 3.6 mm in this case. Both the circular diagram (D-2) and TSNIT curve (D-3, arrow) show significant RNFL thinning at the TS sector.

Paired comparison between RNFL_Di and RNFL_BMO1 in the POAG group in eyes with externally oblique BT and those without externally oblique BT yielded similar results to those obtained in the healthy group (Fig. 5). The generalized estimating equation analysis revealed that when using RNFL thickness in the T sector as a reference, RNFL_Di in the NI, N, and NS sectors was significantly thinner than RNFL_BMO1 in the corresponding sectors in POAG eyes with externally oblique BT (Fig. 5A), while the significant difference was found only in the N sector in POAG eyes without externally oblique BT (Fig. 5B). 

Figure 5

 

Comparison between the conventional and new methods in measuring the circumpapillary RNFL thicknesses in glaucomatous eyes, depending on the presence of externally oblique BT. Asterisks indicate significant differences between the circumpapillary RNFL thicknesses (paired t-test: P ≤ 0.0071, 0.05/7). Double daggers indicate significant differences between the circumpapillary RNFL thicknesses relative to the difference in the temporal sector (dagger).

Figure 6 illustrates a representative case of a glaucomatous eye with externally oblique BT in which the differential RNFL profiles between the methods resulted in a false-negative finding in the TI sector and a false-positive finding in the N and NI sectors using the conventional algorithm. 

Figure 6

 

Comparison between the two methods in measuring the circumpapillary retinal nerve fiber layer thicknesses in a glaucomatous eye. (A) Color disc photograph showing a myopic, tilted optic disc. (B) Standard automated perimetry revealing a superonasal step. (C) Circumpapillary RNFL thickness measurement using the scan circle centered on the BMO (RNFL_BMO). (C-1) Infrared image showing three scan circles centered on BMO. The innermost light-green circle has a diameter of 3.5 mm and measures RNFL_BMO1. (C-2) Circular diagram showing borderline RNFL thinning at in the TI sector. (C-3) The localized RNFL thinning in the TI sector is more definite in the TSNIT curve (red arrow). (D) Circumpapillary RNFL thickness measurement using the scan circle centered on the optic disc (RNFL_Di). (D-1) Infrared image showing the scan circle centered on the disc margin. The diameter of the scan circle is set at 12°, and was 3.6 mm in this case. (D-2) Circular diagram showing borderline RNFL thinning in the N and NI sectors, neither of which corresponds with the location of the visual-field defect (B). (D-3) Curve of TSNIT showing RNFL thinning in the TI sector (red arrow), which corresponds with the RNFL_BMO1 measurement (C-3, red arrow). However, borderline RNFL thinning was also observed at nonrelevant locations (black arrows).

The ability to discriminate between healthy and POAG eyes was comparable between the two methods for global RNFL thicknesses (AUC = 0.892 for both global RNFL_BMO1 and global RNFL_Di; P = 0.958). The area under the curve did not differ between the two methods for measuring the RNFL thickness in any of the six sectors (Table 4). However, when the analysis was performed in each subgroup divided by the externally oblique BT width using 250 μm (the median value of the BT width in subjects having the BT) as the cutoff, the global RNFL_BMO1 performed better than the global RNFL_Di in eyes with an externally oblique BT width >250 μm (Table 5). There was no significant difference in the AUC for RNFL_BMO1, RNFL_BMO2, and RNFL_BMO3, (all P > 0.05, data not presented), while RNFL_BMO1 had the largest AUC in almost all sectors. 

Table 4

 

Comparison of the AUCs Between the Conventional and the New Methods to Measure Circumpapillary RNFL Thicknesses

 

Table 5

 

Comparison of AUCs Between the Conventional and the New Methods to Measure Circumpapillary RNFL Thicknesses Depending on the Width of the Externally Oblique BT

 

The receiver operating characteristic regression model was applied to explain the factors affecting the glaucoma diagnostic ability of each parameter, such as age, sex, refractive error, visual-field mean deviation, presence of externally oblique BT, and the circumpapillary RNFL measurement method (i.e., RNFL_Di versus RNFL_BMO1). The regression model revealed that the use of RNFL_BMO1 had a marginally significant influence on a larger AUC in global (P = 0.066) and N-sector (P = 0.059) RNFL thickness measurements, and a significant effect on the larger AUC in NS-sector RNFL thickness measurement (P = 0.035). Figure 7 shows the ROC curves for the global RNFL_Di and RNFL_BMO measurements, and those in the N and NS sectors, as calculated from the regression model. Figure 8 shows the ROC curves for global RNFL_Di and RNFL_BMO1 for arbitrary values of externally oblique BT width (Figs. 8A–C), and visual field mean deviation (Figs. 8D–F), according to the regression model. The area under the curve was larger when using the global RNFL_BMO1 than the global RNFL_Di for a larger externally oblique BT width (Figs. 8A–C). There was no difference in the AUCs between global RNFL_Di and RNFL_BMO1 according to the differed disease severity (Figs. 8D–F). Figure 9 shows the sensitivities at fixed specificities of 80% for the RNFL_Di and RNFL_BMO, throughout the range of the width of the externally oblique BT calculated based on the regression model. The sensitivity of the RNFL_BMO increased as the externally oblique BT width increased, while that of the RNFL_Di decreased. 

Figure 7

 

Receiver operating characteristic curves for RNFL thickness measurements using the conventional and new methods. Graphs comparing the two methods for the (A) global, (B) nasal, and (C) nasal-superior RNFL thicknesses, as calculated from the regression model. Values of P indicate the statistical difference between the areas under the curve. Note that RNFL_BMO1 tends to have larger areas under the curve than RNFL_Di, and the difference is statistically significant for the NS sector.

Figure 8

 

Receiver operating characteristic curves for the global RNFL thickness measurements using the conventional and new methods for arbitrary values of the externally oblique BT width (A–C) and the visual field mean deviation (D, E), according to the regression model.

Figure 9

 

Sensitivities at a fixed specificity of 90% for RNFL_Di and RNFL_BMO1, according to the width of the externally oblique BT.

Discussion

This study investigated the performance of the new circumpapillary RNFL scanning algorithm centered on BMO and compared it with that of the conventional algorithm, which centers on the clinical disc margin. The results of the study can be summarized as follows: 

•  
  We found the RNFL_BMO1 tended to be larger than RNFL_Di, and the discrepancy was greater in elongated eyes with externally oblique BT.
•  
  The discrepancy between the two methods was significant in both healthy and glaucomatous eyes with externally oblique BT, but not in those without externally oblique BT.
•  
  The discrepancy was due mainly to the difference in N-sector RNFL thicknesses: RNFL_Di tended to be thinner than in RNFL_BMO1 in the NS, N, and NI sectors in eyes with externally oblique BT, while the two methods recorded comparable RNFL thicknesses in eyes without externally oblique BT.
•  
  The ability to discriminate between glaucoma and healthy eyes was comparable between the two methods. However, the global RNFL_BMO1 performed better than the global RNFL_Di in eyes with an externally oblique BT width >250 μm.
•  
  According to the ROC regression model, RNFL_BMO1 performed better than RNFL_Di in the NS sector.
•  
  According to the ROC regression model, AUC was larger for the global RNFL_BMO1 than for the global RNFL_Di, as the externally oblique BT width increased.

Studies have consistently suggested that the circumpapillary RNFL profile is less consistent in myopic eyes than nonmyopic counterparts.^35–39 This has been explained by the difference in the retinal vessel topography associated with axial elongation in myopia,^35,36 and the influence of ocular magnification.³⁷ However, the effect of the OCT scan circle location on the circumpapillary RNFL profile may also have to be considered. Gabriele et al.¹⁴ demonstrated that horizontal movement of the OCT scan circle significantly affected the RNFL thickness profile. Later, Chung and Yoo⁴⁰ subsequently showed that the RNFL thickness measured along the scan circle centered on the clinical optic disc is unreliable in myopic eyes with tilted optic discs, and suggested that the neural canal opening is a more appropriate landmark for the OCT scan circle in eyes with myopic, tilted optics. This issue is particularly relevant in tilted optic discs with externally oblique BT, in which a large disparity is frequently observed between the neural canal opening or BMO, and the clinical optic disc margin.^18,20 In these eyes, the margin of BMO might have been the original disc margin when the disc tilt had not yet occurred; thus, externally oblique BT had not yet developed.⁴¹ Therefore, in eyes with a tilted optic disc, the RNFL profile obtained centered on BMO is potentially closer to the normative profiles of eyes without tilted optic discs. 

The new circumpapillary RNFL scanning algorithm sets the scan circle at the center of BMO. The concept of this algorithm is based on the findings of Chauhan and Burgoyne¹⁷ and other researchers,^18–20 suggesting that BMO is a more consistent anatomic reference than the clinical disc margin, because the latter lacks a solid anatomic structure and there is a large interindividual variability in its determination. The Bruch's membrane opening–based RNFL analysis is particularly advantageous in myopic eyes with a tilted optic disc for several reasons: 

•  
  The discrepancy between the disc margin and BMO is relatively large due to ONH distortion.
•  
  Tilted optic discs are commonly accompanied by a large degree of PPA, which may interfere with the OCT scan path when the circle is centered on the clinical disc margin. However, by centering the scan circle on BMO, the PPA can be bypassed from the scan path.
•  
  The interoperator variability can be reduced because the location of the scan circle is determined automatically by the supplied OCT software (Heidelberg Engineering).

The relevance of circumpapillary RNFL measurement made by centering the scan circle on BMO in myopic eyes is supported by the present study finding that the RNFL profile in healthy eyes was similar between those with externally oblique BT and those without, when the circle was centered on BMO, while it differed significantly when the circle was centered on the disc margin. In addition, the two methods gave comparable measurements in eyes without externally oblique BT, but differed significantly in eyes with externally oblique BT, in both the healthy and POAG groups. Together these findings suggest that the RNFL_BMO measurements will be more reliable than RNFL_Di measurements in eyes with externally oblique BT. 

The discrepancy between RNFL_BMO1 and RNFL_Di was found mainly in the N sectors. BMO is usually deviated temporally relative to the disc margin, and thus the scan circle measuring RNFL_BMO is set more temporally than that measuring RNFL_Di. Therefore, the N-sector RNFL_Di measurements may be underestimated compared with the N-sector RNFL_BMO1 measurements, while the opposite is true for the T-sector RNFL measurements. The difference found only in the N sectors in the present study suggests that the discrepancy in RNFL measurements with horizontal movement of the scan circle may be smaller in the T sectors than in the N sectors. 

Despite the potential advantage of RNFL_BMO measurements in myopic eyes, the ability to discriminate glaucoma from healthy eyes was only comparable between RNFL_BMO1 and RNFL_Di. The regression model of ROC suggested a superiority of the RNFL_BMO1 over RNFL_Di, at least in the N sectors; but this difference was significant only in the NS sector. The relatively low superiority of the RNFL_BMO may be attributable to exclusion of a large number of eyes with externally oblique BT (n = 77, 29.5%), because Spectralis-based OCT detection of BMO frequently failed in such eyes. The larger the width of externally oblique BT, the greater the discrepancy in the location of the scan circle between methods, resulting in larger intermethod differences in RNFL profile, which may have been able to better differentiate the diagnostic performance of each method. However, a more detailed analysis revealed that RNFL_BMO was superior to RNFL_Di in global RNFL measurements, at least in the subset of eyes with an externally oblique BT width >250 μm. In addition, RNFL_BMO was potentially superior to RNFL_Di when the externally oblique BT was larger. 

This study was subject to several limitations. First, the size of the scan circle measuring the RNFL_Di varied according to the individual axial length (which is usually 3.3∼3.7 mm), while that measuring the RNFL_BMO1 was fixed at 3.5 mm. The bias caused by this difference in scan-circle size was minimized by excluding eyes when the scan-circle diameters of the two methods differed by more than 1 mm. In addition, the systemic difference between the methods was dealt with by introducing the generalized estimating equation analysis. 

Second, the topographic orientation of the sectorial RNFL thickness analysis differed between the two methods: the foveal–BMO axis was used for RNFL_BMO, while the foveal–disc was used for RNFL_Di. This may have influenced the results of the sectorial comparisons between the two methods. However, we believe that the influence would have been minimal, because the foveal–BMO and foveal–disc angles did not differ. 

Third, 77 eyes were excluded due to failure of the OCT algorithm to detect BMO. The peripapillary OCT scanning may produce poor quality images in myopic tilted eyes because of the large area of PPA and externally oblique BTE, which may cause false detection of BMO, less accurate segmentation, and subsequent less reliable measurement. Given that a large proportion of eyes were excluded, the result of the present study may not be generalizable to all myopic eyes. An improved algorithm to detect correct BMO location would be able to expand the usefulness of BMO-based circumpapillary RNFL examination. 

In conclusion, this study has revealed that the new circumpapillary RNFL scanning algorithm centered on BMO exhibits great potential, particularly in myopic eyes with externally oblique BT. By bypassing the inherent anatomic characteristics of myopic, tilted optic discs, the new algorithm produces RNFL profiles that are comparable to the normative database. Although the glaucoma diagnostic ability of the new algorithm was only comparable with that of the conventional algorithm, with improved BMO detection ability it may potentially serve as a more useful tool to assess glaucomatous RNFL thinning, specifically in myopic eyes with externally oblique BT. 

Acknowledgments

Supported by Basic Science Research program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science, and Technology (No. 2016R1D1A1B02011696) (HK). 

The authors alone are responsible for the content and writing of the paper. 

Disclosure: E.J. Lee, None; K.M. Lee, None; H. Kim, None; T.-W. Kim, None