Circle- and Grid-Wise Analyses of Peripapillary Nerve Fiber Layers by Spectral Domain Optical Coherence Tomography in Early-Stage Glaucoma

Abstract

Purpose.: To study diagnostic performances of circle- and grid-wise analyses of peripapillary retinal nerve fiber layer thickness (RNFLT) using spectral domain optical coherence tomography (SD-OCT) in early-stage glaucoma.

Methods.: Eighty-nine open-angle glaucoma (OAG) eyes (mean deviation: 2.5 ± 1.8 dB) and 89 age-matched normal eyes were studied. Peripapillary RNFLT was analyzed using an SD-OCT raster scan in a 6.0 × 6.0-mm area. Averaged RNFLT was calculated over 0.1 × 0.1-, 0.21 × 0.21-, or 0.42 × 0.42-mm grids in the peripapillary area (grid method), or arcuate sector areas between 2.8- and 4.0-mm diameter circles (annulus method), or along a 3.4-mm diameter circle (circle method). Normative data-based cutoff values for averaged RNFLT and number of abnormal grid locations (grid method) or sectors (annulus or circle method) were varied.

Results.: The grid method showed the best power of sensitivity/specificity of 0.94/0.96 with any five contiguous 0.21 × 0.21-mm grid locations with a 2.5 percentile cutoff, followed by the annulus method of 0.81/0.98, and the circle method of 0.76/0.97, with 30° sectors. The sensitivity of the grid method was significantly higher than that of the other methods (P < 0.001), whereas the specificity was not. Coefficients of variation and interclass correlation coefficients of intervisit measurements of averaged RNFLT over each 0.21 × 0.21-mm grid were 3.1% to 11.3% and 0.937 to 0.760, respectively, in a separate OAG patient group.

Conclusions.: Grid-wise analyses of peripapillary RNFLT for differentiating early-stage glaucoma showed >0.90 sensitivity and ≥0.95 specificity.

Introduction

Diagnosis of early-stage glaucomatous optic neuropathy is important for minimizing the risk of visual field loss, especially in open-angle glaucoma (OAG) patients, who are unaware of their visual field defects, and who constitute the majority of glaucoma patients. ¹ Detailed examination of the optic nerve head (ONH) or the retina is essential because significant retinal ganglion cell loss can occur before evident visual field defects appear. ^2,3 Imaging techniques, including confocal scanning laser ophthalmoscopy (Heidelberg Retinal Tomograph; Heidelberg Engineering, Heidelberg, Germany), scanning laser polarimetry (nerve fiber analyzer and GDx, a scanning laser that measures the thickness of the retinal nerve fiber layer [RNFL]; Laser Diagnostic Technologies, San Diego, CA), and time domain optical coherence tomography (TD-OCT) allow quantitative evaluations of glaucomatous changes in ONH or peripapillary retina, with demonstrated clinical efficacy. ^4–10  

Among these techniques, approaches based on automatic RNFL segmentation and evaluation of its thickness (RNFLT), using OCT, are especially promising in diagnosis of glaucoma with early-stage visual field damage. ^4–6,9,11 Spectral domain OCT (SD-OCT), with faster retinal A-scan speeds and higher resolution and reproducibility of RNFLT measurements compared with TD-OCT, ^12–17 showed comparable or superior diagnostic performance for glaucoma when compared with TD-OCT. ^13,18–23  

A single examination, using raster-scan protocols for SD-OCT, provides RNFLT data in an area involving ONH, and enables the disc center to be automatically determined after data acquisition, of which correct determination is critical in glaucoma follow-up by circumpapillary RNFL (cpRNFL). ²⁴ Previous studies in differentiating glaucoma from normal eyes have evaluated RNFLT along the circumpapillary circle scan using averages in 360° or in 2, 4, or 12 sectors, each accounting for 180°, 90°, or 30°, respectively. ^{9,13,17–19,21–23,25–34} Glaucoma is a disease that should be diagnosed in its early stages, and SD-OCT could be especially useful in clinical practice if it could efficiently differentiate early-stage glaucoma eyes from normal eyes. Thus, the sensitivity/specificity when the patients are limited only to early-stage glaucoma ^{20,23,27,31–34} would be clinically important. 

Averaged RNFLT data over a subdivided area, which is more readily available with the SD-OCT raster scan over a peripapillary area, may provide more information needed for differentiating glaucoma in its early stages. Further, inherent increases in variation, by adopting narrower sectorization or smaller areas for analysis to effectively detect focal abnormalities, ^35–41 may be counteracted by adopting averaged data over an area to improve reproducibility of RNFLT measurements by SD-OCT. ^{12–17,42,43}  

We studied the potential for use of averaged RNFLT data over peripapillary grid locations or circumpapillary arcuate areas obtained with the SD-OCT in differentiating early-stage glaucoma eyes with average mean deviation (MD) of −2.5 dB from normal eyes. 

Methods

Participants

Normal subjects and OAG patients were separately recruited using identical inclusion criteria at The University of Tokyo, Gunma University, Kanazawa University, Kyoto University, Osaka University, and Tajimi Municipal Hospital after written informed consent was obtained. The study protocol was approved by the appropriate institutional review boards and adhered to the tenets of the Declaration of Helsinki. 

Self-reported ≥20-year-old healthy volunteers were invited as candidates for normal participants. Ocular examinations, including automatic refractometer (ARK-900; NIDEK, Aichi, Japan)-based measurements of refraction and corneal radius of curvature, best-corrected visual acuity (BCVA) measurements with the 5-meter Landolt chart (IOL Master; Carl Zeiss Meditec, Inc., Dublin, CA)-based axial length measurements, slit-lamp examination, Goldmann applanation tonometer–based intraocular pressure (IOP) measurements, dilated fundoscopy, and Humphrey Field Analyzer (HFA) 24‐2 Swedish Interactive Threshold Algorithm Standard (SITA-S) program (Carl Zeiss Meditec, Inc.)-based visual field testing, were performed at first visit. Exclusion criteria were contraindications to pupillary dilation, ≥22 mm Hg IOP, <20/25 BCVA, refractive error <−6.0 diopters (D)/>+3.0 D, unreliable HFA results (fixation loss or false negative >20%, false positive >15%), visual field defects suggestive of criteria-based glaucoma presented by Anderson and Patella, ⁴⁴ of which details will be described below, abnormal visual field loss consistent with ocular diseases, history of intraocular or refractive surgery, and history of ocular or systemic diseases that could affect results of HFA or OCT examinations, such as clinically significant cataract, diabetic retinopathy/maculopathy, age-related macular degeneration, or optic nerve/retinal abnormality. Volunteers outside the age range of the current OAG patients were excluded. 

OAG patients having glaucomatous changes in the ONH with/without apparent stereo fundus photography–based RNFL defects (confirmed by a panel of glaucoma specialists [MA, HA, and AI]), mean deviation (MD) of the 24‐2 SITA-S program >−6 dB, refractive error ≥−6.0 D/≤+3.0 D, and negative history of any other ocular pathological changes that could affect HFA or OCT examination results, including intraocular or refractive surgery, were included. Glaucoma was defined by the presence of reproducible HFA-confirmed glaucomatous visual field defects within 3 months of OCT examination. Glaucomatous visual field defects were defined by (1) a cluster of ≥3 points in the pattern deviation plot in a single hemifield (superior/inferior) with P < 5% and one of which had P < 1%; (2) glaucoma hemifield test result out of normal limits; or (3) abnormal pattern standard deviation (PSD) with P < 5%. ⁴⁴ If both eyes of a subject fulfilled the inclusion criteria, the SD-OCT data with better quality factor (given by the SD-OCT apparatus based on signal intensity) were included. 

A separate group of OAG subjects participated in the measurement reproducibility study. Inclusion criteria were the same as those mentioned earlier, except that OAG subjects were those with MD of >−10 dB. 

OCT Measurements

RNFLT was measured with commercial software (3D-OCT1000, version 3.20; Topcon, Inc., Tokyo, Japan) in which the SD-OCT technology is combined with a nonmydriatic fundus camera function equivalent to the commercially available nonmydriatic fundus camera (TRC-NW200; Topcon, Inc.). This system uses a superluminescent diode (central wavelength, 840 nm; bandwidth, 50 nm) as the light source and acquires 27,000 axial scans/s. All participants underwent SD-OCT measurements after pupillary dilation with 1% tropicamide. SD-OCT data sets were obtained with the raster scan protocol in which RNFLT data were obtained in a 6.0 × 6.0-mm area (512 × 128 pixels) centered on the optic disc in approximately 2.5 seconds, immediately after which a fundus photograph (45° visual angle) was automatically obtained by the nonmydriatic fundus camera function. To obtain accurate sizes on fundus images, the magnification effect was corrected according to the manufacturer's provided formula (modified Littmann's equation, ^45,46 which is based on refractive error, corneal radius, and axial length as previously measured). In this study, seven points were manually determined at the optic disc edge on the color fundus photograph, obtained as above. Registration of fundus photographs and OCT images was automatically confirmed using an OCT projection image (generated from the OCT data by summing signals at different retinal depth levels) and localization of major retinal vessels. The optic disc center was determined in fundus photographs as the barycenter coordinates of the closed spline curve fitted to the seven points determined earlier. The point was then extrapolated in all OCT images. 

Measurements were repeated three times with several second intervals, and data with the best quality factor (given by the apparatus according to signal strength of the image) were used. Images influenced by involuntary blinking or saccade indicated by breaks or shifting of the vessel or disc images or a straight line across the fundus OCT image, respectively, or those with quality factor <60% were excluded. RNFLT was determined in all B-scan images automatically, regardless of retinal vessels, and all images were confirmed after segmentation by an experienced examiner (AT). 

Grid Method

A 5.0 × 5.0-mm area centered on the disc barycenter coordinates was divided into 0.1 (5/50) mm × 0.1 mm, 0.21 (5/24) mm × 0.21 mm, or 0.42 (5/12) mm × 0.42 mm grids. Grids outside and not overlapping the 2-mm-diameter circle centered on the disc barycenter were used for averaged RNFLT measurements (Fig. 1A). Distribution of averaged RNFLT over grid locations in the peripapillary area in normal eyes was determined by additional data sets previously obtained using the same SD-OCT instrument in 272 eyes of 272 normal Japanese subjects. ^47,48 The five normative data-based cutoff values (<0.5th, <1st, <2.5th, <5th, or <10th percentiles) of distribution of averaged RNFLT values were adopted and an eye for which the averaged RNFLT value was abnormal in at least one or any plural contiguous grid locations was considered abnormal. Sensitivity/specificity was calculated per diagnostic criteria with varied grid sizes, normal averaged RNFLT cutoff values, and number of contiguous grid locations flagged as abnormal. 

Figure 1

 

(A) Diagram of grid method. (B) Diagram of annulus and circle method.

Annulus Method

Averaged RNFLT was measured in the annular area between two concentric circles centered on the disc barycenter, with diameters of 2.8 and 4.0 mm, which were adopted based on the relationship between the RNFLT measurement reproducibility and diameter of circumpapillary circle with this SD-OCT instrument ⁴⁷ ; and medium-sized measurement area of GDx scanning laser polarimetry ⁴⁹ (Fig. 1B); and analyzed in 180°, 90°, or 30° arcuate sector area. The distribution of averaged RNFLT in this arcuate sector area in normal eyes was also obtained from the above normative database. ^47,48 Eyes for which the averaged RNFLT value was abnormal in at least one or any separate or contiguous plural sectors from the 360° circle were considered glaucomatous. The sensitivity/specificity was calculated per diagnostic criteria with varied sector widths, normative data-based RNFLT cutoff values (<0.5th, <1st, <2.5th, <5th, or <10th percentiles), and number of sectors flagged as abnormal. 

Circle Method

To compare the two preceding methods, circumpapillary RNFLT (cpRNFLT) in sectors of 180°, 90°, or 30° was measured along a circle of 3.4-mm diameter centered on the disc barycenter using the raster scan data. The normal cpRNFLT distribution along this circle was also from the above normative database. ^47,48 An eye for which the cpRNFLT value was abnormal in at least one or any separate or contiguous plural sectors from the 360° circle was considered glaucomatous and sensitivity/specificity was calculated per diagnostic criteria with varied sector widths, normative data-based cpRNFLT cutoff values (<0.5th, <1st, <2.5th, <5th-, or <10th percentiles), and number of sectors flagged as abnormal. 

Measurement Reproducibility of the Optimum Method Giving the Highest Diagnostic Power

Intervisit measurement reproducibility of the averaged RNFLT, given by the method that yielded the highest diagnostic power (sum of sensitivity and specificity) with specificity ≥0.95 among methods tested, was calculated in a separate group of OAG patients. In those subjects, the measurements were repeated once again by the same protocol as above, at the same hour, on a separate day, within 3 months after the initial measurement. Reproducibility was evaluated as coefficients of variation (CV) and interclass correlation coefficient (ICC). 

Statistical Analysis

Statistical analyses were performed using commercial software (SPSS Statistics software, version19; IBM Japan, Ltd., Tokyo, Japan). Demographic data were compared between normal and OAG eyes using ANOVA and Tukey's post hoc test. Sensitivities and specificities were compared using McNemar's test. P < 0.05 was considered statistically significant. 

Results

Patients

After excluding eyes with deficient fixation, low data quality, inaccurate RNFL automated segmentation results, or eyes where any of the most peripheral grid locations were not included in the 5.0 × 5.0-mm analysis area centered on the disc barycenter, 89 normal and 89 OAG eyes from 89 normal and 89 OAG subjects, respectively, were included (Table 1). 

Table 1. 

 

Subjects' Characteristics of Sensitivity/Specificity Study

	Normal Group	Open-Angle Glaucoma Group
Number of participants, eyes	89	89
Age, y	58.1 ± 11.1	58.9 ± 11.5
Mean deviation, dB	−0.4 ± 1.2	−2.5 ± 1.8*
Refraction, diopters	−0.2 ± 1.3*	−1.5 ± 2.4*
Intraocular pressure, mm Hg	14.2 ± 2.3	14.5 ± 2.5
Axial length, mm	23.6 ± 0.9	24.0 ± 1.4
Corneal curvature, mm	7.69 ± 0.26	7.66 ± 0.27

 Values are mean ± SD. *P < 0.001.

The MD value averaged −2.5 ± 1.8 dB in the OAG eyes, and age, IOP at the OCT measurement, axial length, and corneal curvature showed no intergroup differences, whereas OAG eyes had worse MD and were more myopic (P < 0.001). The results yielding specificity ≥0.95 are discussed below. 

Grid Method

Given the same grid size, specificity was higher and sensitivity was lower with a greater number of contiguous grid locations and stricter cutoff. The diagnostic performance was compared among those criteria based on diagnostic power (sum of sensitivity and specificity). 

With a grid size of 0.42 × 0.42 mm, the highest diagnostic power of 1.76 (sensitivity/specificity: 0.80/0.96) was obtained using at least five contiguous grid locations outside the 1st percentile normative data-based cutoff, followed by 1.75 (0.76/0.99, at least four outside the 0.5th percentile cutoff). No other combination of the number of contiguous grid locations and cutoff values gave a diagnostic power ≥1.75 with a specificity ≥0.95. 

With a grid size of 0.21 × 0.21 mm, the highest diagnostic power of 1.90 (sensitivity/specificity: 0.94/0.96) was obtained using at least five contiguous grid locations outside the 2.5th percentile cutoff, followed by 1.89 (0.91/0.98, at least two outside the 0.5th percentile cutoff or 0.93/0.96, at least three outside the 1st percentile cutoff). No other combination of number of contiguous grid locations and cutoff values gave a diagnostic power ≥1.89 (Fig. 2). 

Figure 2

 

Diagnostic performance of the grid, annulus, and circle method. Sensitivity/specificity was obtained using 0.21 × 0.21-mm grids and at least five contiguous grid locations in the upper or lower hemifield outside the 0.5th, 1st, 2.5th, 5th, or 10th percentile cutoff of distribution of averaged retinal nerve fiber layer thickness obtained from the normative database by the grid method (solid triangle), and that obtained by the annulus (multiplier symbol) or circle method with 30° sector width (solid circle) and at least one sector outside the 0.5th, 1st, 2.5th, 5th, or 10th percentile cutoff of distribution of averaged retinal nerve fiber layer thickness obtained from the normative database.

With a grid size of 0.1 × 0.1 mm, no combination of number of contiguous grid locations and cutoff values giving the specificity ≥0.95, and a highest diagnostic power of 1.84 (sensitivity/specificity: 0.97/0.87 or 0.92/0.92) was obtained using at least three contiguous grid locations outside the 1st percentile cutoff or at least five contiguous grid locations outside the 0.5th percentile cutoff. 

Annulus Method

Given the same sector width, specificity was higher and sensitivity lower with a greater number of contiguous or separate sectors and stricter cutoff values both in the annulus and circle methods. With the sector width of 180° (upper or lower hemicircle), a specificity ≥0.95 was not obtained. 

With the sector width of 90° (superior, temporal, inferior, or nasal quadrant), the highest diagnostic power of 1.73 (sensitivity/specificity: 0.75/0.98) was obtained using at least one sector outside the 1st percentile normative data-based cutoff, followed by 1.65 (0.67/0.98, at least one sector outside the 0.5th percentile cutoff). No other combination of the number of contiguous or separate sectors and cutoff values gave a diagnostic power of ≥1.65 with specificity of ≥0.95. 

With the sector width of 30°, the highest diagnostic power of 1.79 (sensitivity/specificity: 0.81/0.98) was obtained using at least one sector outside the 0.5th percentile cutoff, followed by 1.60 (0.64/0.96, at least two separate sectors outside the 2.5th percentile cutoff). No other combination of the number of contiguous or separate sectors and cutoff values gave a diagnostic power of ≥1.60 with a specificity of ≥0.95 (Fig. 2). 

Circle Method

With the sector width of 180°, a specificity ≥0.95 was not obtained. With the sector width of 90°, the highest diagnostic power of 1.70 (sensitivity/specificity: 0.72/0.98) was obtained using at least one sector outside the 1st percentile normative data-based cutoff, followed by 1.61 (0.63/0.98, at least one sector outside the 0.5th percentile cutoff). No other combination of number of contiguous or separate sectors and cutoff values gave a diagnostic performance of ≥1.61 with specificity of ≥0.95. 

With the sector width of 30°, the highest diagnostic power of 1.73 (sensitivity/specificity: 0.76/0.97) was obtained using at least one sector outside the 0.5th percentile cutoff, followed by 1.57 (0.60/0.97, at least two separate sectors outside the 2.5th percentile cutoff). No other combination of number of contiguous or separate sectors and cutoff values gave a diagnostic power of ≥1.57 with specificity of ≥0.95 (Fig. 2). 

Measurement Reproducibility of the Grid Method With 0.21 × 0.21-mm Grid Size

A total of 46 eyes from 46 OAG patients were included (Table 2). CV and ICC values ranged between 3.1% and 11.3% and 0.760 and 0.937, respectively, at each 0.21 × 0.21-mm grid location. Reproducibility was worse, not only in the nasal, but also in the inferior temporal area, which is more likely to suffer glaucomatous damage (Fig. 3). 

Figure 3

 

Intervisit reproducibility (CV) of retinal nerve fiber layer volume measurement at each 0.21 × 0.21-mm peripapillary grid in early-stage glaucoma eyes.

Table 2. 

 

Patients' Characteristics of Reproducibility Study

Participants	Open-Angle Glaucoma Patients
Number of patients, eyes	46
Age, y	56.4 ± 12.7
Refraction, diopters	−1.4 ± 2.0
Intraocular pressure, mm Hg	14.1 ± 2.6
Mean deviation, dB	−2.5 ± 3.0

 Values are mean ± SD.

Discussion

Peripapillary RNFLT-based OCT diagnostic performance in early-stage glaucoma was studied based on various grid sizes with the grid method or sector widths by the annulus or circular method, and various normative data-based RNFLT cutoff values. The results obtained could be more readily applicable in daily clinical practices using a commercially available SD-OCT instrument. Trends were as expected. The specificity increased with stricter cutoff values, and greater grid size or sector width, with the reverse correlation for sensitivity. 

The optimum grid method, currently obtained at least five contiguous 0.21 × 0.21-mm grids outside the 2.5th percentile cutoff, yielded sensitivity/specificity of 0.94/0.96 in differentiating early-stage OAG eyes with mean MD of −2.5 dB from normal eye measurements. The optimum annulus or circle method analyzing circumpapillary RNFLT yielded sensitivity/specificity of 0.81/0.98 or 0.76/0.97, respectively, with sector width of 30° and at least one sector outside the 0.5th percentile cutoff. Among these three values, the specificity showed no significant difference ≥0.95, whereas the sensitivity yielded by the optimum grid method was significantly higher than that of the other two methods (P < 0.001). However, no significant difference was seen between the sensitivity measured by the optimum annulus method and the circle method (P = 0.22). 

Sensitivities to detect glaucomatous eyes were dependent on the severity of glaucomatous optic neuropathy of the subjects, ^31,50 and glaucomatous eyes of a relatively more advanced stage than those of the current study have been used in many of the previous studies. Using TD-OCT, sensitivities of 0.80 to 0.89 and specificities of 0.88 to 0.92 were reported in glaucomatous eyes with MD ranging from −4.6 to −9.0 dB, and with at least one or two abnormal clock hour (sector width of 30°) cpRNFLT outside of the 1st or 5th percentile normative data-based cutoff values. ^{13,18,20,51–53} SD-OCT generally yielded better diagnostic power than TD-OCT in distinguishing glaucomatous eyes from normal eyes. ^{13,18,20–23} In glaucomatous eyes with mean MD ranging from −5.0 to −9.2 dB, sensitivities of 0.63 to 1.0, and specificities of 0.72 to 1.0 have been reported based on various criteria using sectorized cpRNFLT and various cutoff values, or receiver-operating characteristic curves. However, sensitivity and specificity values of >0.90 have rarely been reported. ^{13,18,19,22,26,28,30} Leung et al. ²⁰ reported a sensitivity/specificity of 0.95/0.95 in glaucomatous eyes with a mean MD of −8.99 dB based on their original scoring system using raster scanned data and deviation maps. 

In early-stage glaucomatous eyes with a mean MD between −1.7 and −3.7 dB, ^27,32–34 and those eyes with MD > −6.0 dB, ²³ with preperimetric glaucoma, ²⁹ or with mild visual field damage, ³¹ sensitivities of 0.65 to 0.85 and specificities of 0.85 to 0.98 were reported. These values are similar to the results currently obtained using the annulus or circle method, which gave sensitivities of 0.76 to 0.81 and specificities of 0.97 to 0.98. We are not aware of previous reports with sensitivities ≥0.90 and specificities ≥0.95 in early-stage glaucoma eyes, based on cpRNFLT measurements. 

The current results obtained using the 0.21 × 0.21-mm grid-based averaged RNFLT analyses of the peripapillary RNFL yielded a sensitivity and specificity higher than 0.90 and 0.95, respectively, under several conditions in early-stage glaucoma eyes with a mean MD of −2.5 dB. A specificity ≥0.95 suggested that the current optimum grid method criterion may also be useful in clinical practices. The current small grid size (0.1 × 0.1 mm) roughly corresponds to that used to construct a deviation map in OCT (Cirrus OCT; Carl Zeiss Meditec, Inc.). Leung et al. ²⁰ used raster scanned data and deviation maps of RNFLT of the peripapillary area, which is an approach similar to that of the current one, and reported a sensitivity/specificity of 0.92/0.95 in glaucoma eyes with mean MD better than −6.0 dB (mean MD value not specified). The best result currently obtained, using the 0.1 × 0.1-mm grid size and three contiguous locations outside the 0.5th percentile cutoff, was a sensitivity/specificity of 0.93/0.92, similar to the result reported by Leung et al. ²⁰ However, no combination of number of contiguous grid locations and cutoff values gave a specificity of ≥0.95 in the current subjects. Taken together, it is possible that analyses of averaged RNFLT over small grid locations in the peripapillary area have more potential in differentiating early-stage glaucoma from normal eyes than analyses of cpRNFLT. 

The intervisit reproducibility of the averaged peripapillary RNFLT measurement, with a grid size of 0.21 × 0.21 mm at each grid locations, was 3.1% to 11.3% of CV and 0.760 to 0.937 of ICC, in OAG eyes with mean MD of −2.5 dB. This was similar to values reported for the circumpapillary measurement of clock hour (30° sector) RNFLT with SD-OCT. ^14,17 Reproducibility tended to be worse in the nasal and inferior temporal areas. A reason for this finding could be the relatively thin RNFLT in the nasal peripapillary area and early glaucomatous damage, which is more likely in the inferior temporal peripapillary area. Reasonable reproducibility of averaged RNFLT measurements at each 0.21 × 0.21-mm grid location and a sensitivity ≥0.90 and specificity ≥0.95 under several combinations of cutoff values and the number of contiguous grid locations flagged as abnormal in early-stage OAG eyes suggest clinical usefulness of the currently described optimum grid method. 

A limitation of this study was that the cohort contained participants only of Japanese ancestry. Thus, the results of this study should be cautiously extrapolated to other ethnic groups because RNFLT measured using TD-OCT or SD-OCT varies significantly between different racial and ethnic groups, ^54–56 suggesting that OCT-based glaucoma diagnostic methods should be based on the normal database established within the same ethnic group. Furthermore, whether the results of the current methods are reproducible with other SD-OCT devices with different normative databases needs to be determined. Another limitation of the current method is that it requires a longer measuring time, making the measurement vulnerable to unexpected ocular movement. This disadvantage may be overcome with the ongoing development of OCT devices that have faster scanning speeds and eye-tracking systems. Furthermore, it must be noted that many of the glaucoma patients at outpatient clinics are more likely to be complicated with senile cataract and/or poor mydriasis, which would result in insufficient image quality of SD-OCT. Thus, the sensitivity/specificity of the current optimum grid method may not directly apply to the entire glaucoma patient population at outpatient clinics. 

In the current study, the diagnostic performance has been compared among grid, annulus, and circle methods based on sensitivity and (1-specificity) in a two-dimensional chart mimicking receiver-operating characteristic (ROC) curve (Fig. 2), although it is not identical to a ROC curve. For example, sensitivity and specificity are dependent on multiple factors such as grid size, normative data-based cutoff values, or the number of grid locations flagged as abnormal in the current method, not only a cutoff value of a single parameter. In this case, sensitivity or specificity does not come up to 1.0, even with the smallest or largest grid size and thus a ROC curve could not be drawn. So, performances have been compared among the methods based on the diagnostic power (sum of sensitivity and specificity) with each criterion for grid size or number of grid locations flagged as abnormal, because area under the curve of each curve on Figure 2 could not be calculated. 

In summary, the best criteria for diagnosing early-stage glaucoma were determined based on the analysis of averaged RNFLT on grid locations with various grid sizes in the peripapillary area or over the arcuate sector areas with various sector widths between 2.8- and 4.0-mm circles centered on the disc barycenter using raster scans of the SD-OCT. Several combinations of normative data-based RNFLT cutoff values and number of contiguous abnormal grid locations of 0.21 × 0.21 mm in the peripapillary area gave a sensitivity of ≥0.90 and a specificity of ≥0.95, with the highest diagnostic performance of 1.90 (sensitivity/specificity: 0.94/0.96) in early-stage glaucoma with a mean MD value of −2.5 dB. 

Acknowledgments

Supported by a Grant-in-Aid for Scientific Research from the Ministry of Health, Labor, and Welfare of Japan (H18-Sensory-General-001) and Topcon, Inc. (Tokyo, Japan). All authors have been paid by Topcon, Inc. (Tokyo, Japan) to attend meetings regarding acquisition of the data (<$1000 in the last 12 months), and Drs. Araie and Yoshimura are paid members of the Advisory Board for Topcon, Inc. The authors alone are responsible for the content and writing of the paper. 

Disclosure: C. Mayama, Topcon, Inc. (R); H. Saito, Topcon, Inc. (R); H. Hirasawa, Topcon, Inc. (R); S. Konno, Topcon, Inc. (R); A. Tomidokoro, Topcon, Inc. (R); M. Araie, Topcon, Inc. (R, S); A. Iwase, Topcon, Inc. (R); S. Ohkubo, Topcon, Inc. (R); K. Sugiyama, Topcon, Inc. (R); T. Otani, Topcon, Inc. (R); S. Kishi, Topcon, Inc. (R); K. Matsushita, Topcon, Inc. (R); N. Maeda, Topcon, Inc. (R); M. Hangai, Topcon, Inc. (R); N. Yoshimura, Topcon, Inc. (R, S)