This retrospective cross-sectional study included 45 eyes of 45 consecutive Korean glaucoma patients with RNFL defects as seen in red-free photographs from May 2005 to July 2006. Informed consent was obtained from all patients. Because all measurements were obtained by methods routinely used in patients with glaucoma, and this study protocol did not include surgical or medical interventions, approval from the institutional review board was not needed according to our regulations. The experimental protocol was performed in accordance with the tenets of the Declaration of Helsinki.

All patients provided their ocular and systemic histories and received ophthalmologic examinations, including best corrected visual acuity measurements, slit-lamp examinations, Goldmann applanation tonometry, ophthalmoscopy, and standard automated perimetry (SAP) with the Humphrey automated field analyzer HFA II 750 (Humphrey Instruments, a division of Carl Zeiss, Inc., San Leandro, CA, USA). In addition, all patients underwent an evaluation for RNFL defects, including RNFL photography, GDx-VCC, and OCT. All examinations were conducted within a period of six months.

Glaucoma was diagnosed in this study when a patient had glaucomatous optic disc damage and RNFL defects with or without glaucomatous visual field loss. All eyes had open angles. Glaucomatous optic disc damage was defined by evidence of excavation, neuroretinal rim thinning or notching, or an asymmetry of the vertical cup-to-disc ratio more than 0.2.

Patients were included when their eyes had diffuse or focal RNFL defects as seen in a red-free photograph, a best corrected visual acuity of 20/40 or better, a clear ocular media, and well-controlled intraocular pressure at the time of evaluation. One eye per patient was randomly selected if both eyes met our inclusion criteria. Eyes with a history of glaucoma surgery, secondary glaucoma, retinal diseases such as diabetic retinopathy, peripapillary atrophy extending to 1.7 mm from the disc center, an atypical birefringence pattern on GDx-VCC imaging (typical scan score < 60),24 a quality score grade less than 8 in GDx-VCC imaging, or poor image quality in OCT (signal-to-noise ratio < 35; accepted A-scans < 95%) were excluded from this study.

RNFL photography was performed after the pupil was fully dilated with a solution containing 0.5% tropicamide and 2.5% phelyephrine.25 A CF-60UV fundus camera (Canon Inc., Utsunomiya, Japan) with TMAX 100 film (Kodak Inc., Seoul, Korea) or an FF 450 plus IR fundus camera with the VISIPAC™ system (Carl Zeiss Meditec, Inc., Jena, Germany) was who had no knowledge of the patients' clinical information. When we could detect both the upper and lower borders of an RNFL defect in a red-free photograph, we regarded it as a focal RNFL defect (Fig. 1A). On the other hand, when we found decreased RNFL density in the whole superior and/or inferior temporal quadrants and could not define any borders of the defect, we categorized it as a diffuse RNFL defect (Fig. 2A). Decreased density in diffuse defects in red-free photography was determined by comparing the superior and inferior temporal quadrants in the same or the contralateral eye. If findings from both investigators (YYK, JHO) did not agree or the quality of the red-free photograph was poor, we excluded those eyes from the study.

SLP imaging was performed using GDx-VCC (software version 5.4) without pupillary dilation. GDx-VCC used a circular scan (3.2 mm diameter) centered on the optic disc. GDx-VCC software-provided parameters were obtained. We defined 3S (three squares) GDx RNFL defects as a cluster of three or more color-coded squares along the traveling line of the retinal nerve fiber in the GDx-VCC deviation map. Clusters of four, five, six, or more color-coded squareswere defined as 4S, 5S, and 6S defects, respectively (Fig. 3). A cluster of two or fewer squares was not regarded as a GDx RNFL defect because two or fewer squares could not show whether the defect traveled along the RNFL distribution. We also excluded any squares starting outside of a single disc diameter from the disc margin (Fig. 3C). In addition, 5% of the 3S GDx RNFL defects included any cluster with three or more squares outside of the 95%, 98%, 99%, and 99.5% normal limits. Two percent of the 3S GDx RNFL defects had three or more squares of 2% (outside of the 98% normal limit), 1% (outside of the 99% normal limit), or 0.5% (outside of the 99.5% normal limit). The principles were applied for other combinations in 1% 3S, 0.5% 3S, 5% 4S, 2% 4S, 1% 4S, 0.4% 4S, 5% 5S, 2% 5S, 1% 5S, 0.5% 5S, 5% 6S, 2% 6S, 1% 6S, and 0.5% 6S GDx RNFL defects.

The Stratus OCT (software version 3.0) was used with or without pupillary dilation to a minimum diameter of 4.5 mm. The circular 360° OCT scans were obtained using the fast RNFL thickness scan mode, which consisted of three scans, each having 256 test points measured along a nominal 1.73 mm radius circle on the peripapillary of the RNFL. OCT softwareprovided parameters were obtained. Segments of the line graph located below the yellow band (outside of the 95% normal limit) and within the red band (outside of the 99% normal limit) of the OCT RNFL analysis map were defined as 5% and 1% OCT RNFL defects, respectively.

The red-free photographs and GDx-VCC deviation maps were divided into four quadrants (superior temporal, inferior temporal, superior nasal, and inferior nasal) along the reference lines, which were the horizontal and vertical lines from the disc center. The OCT RNFL analysis map was also divided into four quadrants at the corresponding point of the reference lines on the red-free photograph and the GDx-VCC deviation map. The superior nasal quadrant and the inferior nasal quadrant were excluded in this study. Further, the superior temporal and inferior temporal quadrants in an eye were analyzed independently using the red-free photographs, the GDx-VCC deviation maps, and the OCT RNFL analysis maps.

When the location of GDx RNFL defects or OCT RNFL defects overlapped with the RNFL defects seen in the red-free photographs for localized RNFL defects, we defined those as the corresponding GDx-VCC or OCT RNFL defects (Fig. 1 and 2). For the comparison of diffuse defects, in which their borders could not be determined in the red-free photographs, we assumed that the RNFL defects existed outside the papillomacular bundle (central 40° divided into superior and inferior sectors each 20° from the horizontal reference line),26 which is relatively resistant to glaucomatous RNFL damage.

Because the deviation map uses a grayscale fundus image of the eye as a background, determining the overlap between the red-free photographs and GDx-VCC RNFL defect was straightforward. However, in the OCT RNFL analysis map, the minimum and maximum angles of the line graph segments located below the colored band were measured to determine the overlap between photographic RNFL defects and OCT RNFL defects, respectively. Two-hundred fifty-six test points were converted to 360 degrees in the OCT analysis maps (Fig. 1C).20,27 In the red-free photographs, the minimum and maximum angles between the horizontal reference line and a line from the center of the disc to the disc margin where the RNFL defect met the disc were measured (Fig. 1A). The range and location of angles in defects seen in the red-free photographs were then compared with those seen in OCT RNFL defects whether the measured angles overlapped or not.

All statistical analyses were performed using SAS version 9.1.3 (SAS Institute Inc., Cary, NC, USA). For an evaluation of the agreement between the use of RNFL photography and GDx-VCC or OCT, Cohen's κ value was calculated with 95% confidence intervals (CIs).28 The κ values were interpreted as follows: 0=no agreement, less than 0.4=poor agreement, 0.40-0.59 fair agreement, 0.60-0.74=good agreement, 0.75-0.99=excellent agreement, and 1=perfect agreement.

The mean age of the 45 patients was 50.5±14.3 years (range, 22-76 years). Twenty-nine patients (64%) were men and 16 (36%) were women. Fifteen (33%) right eyes and 30 (67%) left eyes were included in the data set. Patient demographics are summarized in Table 1.

Of the 90 quadrants (45 eyes), 31 (34%) had no apparent RNFL defects, 29 (32%) had focal RNFL defects, and 30 (33%) had diffuse RNFL defects in the red-free photographs. Table 2 shows the agreements and kappa valuesfor 3S, 4S, 5S, and 6S GDx RNFL defects and OCT RNFL defects with photographic RNFL defects. The highest agreement between the GDx-VCC and red-free photography was 73% when we defined GDx-VCC RNFL defects as a cluster of three or more color-coded squares (p< 5%) along the traveling line of the retinal nerve fiber in the GDx-VCC deviation map (kappa value, 0.388; 95% CI, 0.195 to 0.582) (Fig. 4A). The highest agreement between the OCT and red-free photography was 85% (kappa value, 0.666; 95% CI, 0.506 to 0.825) when a value 5% outside the normal limit for the OCT analysis map was used as a cut-off value for OCT RNFL defects (Fig. 4A).

When we used the 5% 3S GDx RNFL defects and the 5% OCT RNFL defects as the RNFL defects, the rate of agreement with focal RNFL defects seen in the red-free photographs (29 quadrants) was 66% with GDx-VCC and 79% with OCT. Three of 29 quadrants were detected only by the use of GDx-VCC, seven quadrants only by the use of OCT, and three quadrants were not detected by the use of either instrument (Fig. 4B). For diffuse RNFL defects seen in the red-free photographs (30 quadrants), the agreement with GDx-VCC was 80%, and with OCT it was 90%. Two of 30 quadrants were detected only by GDx-VCC, five quadrants were detected only by OCT, and one quadrant was not detected by either instrument (Fig. 4C).

The purpose of our study was to assess the agreement between GDx-VCC and OCT, and photographic RNFL defects. According to the kappa values, we found that the agreement between GDx-VCC deviation maps and red-free photography was poor, whereas the agreement between OCT RNFL analysis maps and red-free photography was good. Our results showed that the agreement between GDx-VCC and RNFL photography had the highest kappa value of 0.388 in the 5% 3S GDx RNFL defects. Similarly, agreement between GDx-VCC and OCT had the highest kappa value of 0.666 in the 5% OCT RNFL defects.

The OCT RNFL analysis map can be used to evaluate agreements between red-free photography and OCT. Hwang et al.27 reported a strong topographic correlation between characteristics of localized RNFL defects as determined by OCT and RNFL photography. Jeoung et al.20 reported that the sensitivity and specificity of OCT using RNFL analysis maps for detecting localized RNFL defects were 85.9% and 97.4%, respectively. Our results also indicated a good agreement between RNFL photography and OCT analysis maps.

We used GDx-VCC deviation maps to compare the use of GDx-VCC with RNFL photography. Because GDx-VCC deviation maps use a fundus image of the eye as a background, they may well reflect the locality of an RNFL defect. In addition, GDx-VCC deviation maps include a normative database, as do OCT RNFL analysis maps. Kook et al.22 used a deviation map algorithm for the detection of RNFL defects. The number of abnormal superpixels was counted in each quadrant, and each superpixel was assigned a severity value as follows: 0, normal areas; 1, p<0.05; 2, p<0.02; 3, p<0.01; and 4, p<0.005. A final severity score for each eye was determined by summing the severity values of all of the abnormal superpixels. Although their system was useful for categorizing the severity of the disease, it may not sufficiently indicate the location of an RNFL defect. Photographic RNFL defects indicated whether or not there was a defect and, if it was present, showed the location of the defect. Therefore, to reveal the location of RNFL defects with GDx-VCC deviation maps, we developed a novel algorithm. However,used statistically-obtained pixels, nothave representedan actual RNFL defect.

Localized RNFL defects are most often found in the inferior temporal quadrant, followed by the superior temporal quadrant. In the nasal fundus quadrant, localized RNFL defects are only rarely seen, possibly because normal RNFL are less detectable in the nasal fundus than in the inferior temporal and superior temporal fundus quadrants.29,30 In fundus regions in which the RNFL is physiologically thin, localized defects are harder to find than in regions with a thick RNFL.31 Therefore, we excluded the superior nasal and inferior nasal quadrants among the four quadrants in this study.

Generally, the agreement between GDx-VCC and red-free photography, and between OCT and red-free photography was greater in diffuse RNFL defects than in focal defects, possibly because the coincidence criteria were less stringent for diffuse RNFL defects in our study. In addition, by using the 5% 3S GDx RNFL defects and the 5% OCT RNFL defects (showing the highest agreements and highest kappa values) as RNFL defects, focal RNFL defects in three of 29 quadrants were detected only by the use of GDx-VCC, and seven quadrants were detected only by the use of OCT (Fig. 4B). For diffuse RNFL defects, two of 30 quadrants were detected only by the use of GDx-VCC, and five quadrants were only detected by the use of OCT (Fig. 4C). We suggest that the use of GDx-VCC and OCT instrumentation can be complementary in the evaluation of RNFL defects.

Our study has some limitations. First, all participants in our study were Korean, but the normative database for GDx-VCC and OCT consisted of several ethnic groups. Therefore, further study with other ethnic groups is needed. In addition, the interpretation of our photographs might be subjective because of the inherited nature of red-free photography. Moreover, RNFL defects cannot be ophthalmoscopically detected if less than 50% of the thickness of the RNFL is lost.32 As a result, the detection of glaucomatous RNFL defects using RNFL photography may be less accurate for a less damaged RNFL.

In conclusion, according to the kappa values, the agreement between GDx-VCC deviation mapping and red-free photography was poor, whereas the agreement between OCT RNFL analysis mapping and red-free photography was good. In addition, our study suggests that the use of GDx-VCC and OCT may be complementary for evaluating RNFL defects in some cases.