Comparison of the Abilities of SD-OCT and SS-OCT in Evaluating the Thickness of the Macular Inner Retinal Layer for Glaucoma Diagnosis

Purpose To compare the abilities of spectral-domain optical coherence tomography (OCT) (SD-OCT; Spectralis, Heidelberg Engineering) and swept-source OCT (SS-OCT; DRI-OCT1 Atlantis system, Topcon) for analyzing the macular inner retinal layers in diagnosing glaucoma. Methods The study included 60 patients with primary open-angle glaucoma (POAG) and 60 healthy control subjects. Macular cube area was scanned using SD-OCT and SS-OCT on the same day to assess the thicknesses of the macular retinal nerve fiber layer (mRNFL), ganglion cell layer plus inner plexiform layer (GCIPL), and total retinal layer in nine subfields defined by the Early Treatment Diabetic Retinopathy Study (ETDRS). The abilities of the parameters to discriminate between the POAG and control groups were assessed using areas under the receiver operating characteristic curves (AUCs). Results Glaucoma-associated mRNFL and GCIPL thinning was more common in the outer zones than inner zones for both SD-OCT and SS-OCT. The mRNFL and GCIPL measurements showed distinct pattern differences between SD-OCT and SS-OCT in each ETDRS subfield. Although the glaucoma-diagnosis ability was comparable between SD-OCT and SS-OCT for most of the parameters, AUC was significantly larger for SD-OCT measurements of the GCIPL thickness in the outer temporal zones (p = 0.003) and of the mRNFL thickness in the outer nasal zones (p = 0.001), with the former having the largest AUC for discriminating POAG from healthy eyes (AUC = 0.894). Conclusion Spectralis SD-OCT and DRI SS-OCT have similar glaucoma-diagnosis abilities based on macular inner layer thickness analysis. However, Spectralis SD-OCT was potentially superior to DRI SS-OCT in detecting GCIPL thinning in the outer temporal zone, where the glaucomatous damage predominantly occurs.

Introduction obtained from all subjects, and the study protocol was approved by the Seoul National University Bundang Hospital Institutional Review Board and followed the tenets of the Declaration of Helsinki.
Subjects for the POAG and healthy control groups were enrolled from patients who visited Seoul National University Bundang Hospital Glaucoma Clinic between September 2014 and April 2015. All of the participants underwent comprehensive ophthalmic examinations that included assessment of visual acuity, Goldmann applanation tonometry, refraction tests, slitlamp biomicroscopy, gonioscopy, and dilated stereoscopic examination of the optic disc. The following investigations were also performed: central corneal thickness measurement (Orbscan II, Bausch & Lomb Surgical, Rochester, NY, USA), axial length measurement (IOLMaster version 5, Carl Zeiss Meditec, Dublin, CA, USA), color fundus photography and red-free fundus photography (EOS D60 digital camera, Canon, Utsunomiyashi, Japan), SD-OCT (Spectralis OCT) scanning of the circumpapillary and the macular cube area, SS-OCT (DRI-OCT1 Atlantis system) scanning of the macular cube area, and standard automated perimetry (Humphrey Field Analyzer II 750, 24-2, Swedish interactive threshold algorithm, Carl Zeiss Meditec).
All participants had to meet the following inclusion criteria: a best-corrected visual acuity of greater than 20/40, a spherical equivalent ranging from -8.0 to +6.0 diopters, cylinder correction within -3.0 to +3.0 diopters, and the presence of an open angle as confirmed by gonioscopy. The exclusion criteria were retinal or neurologic diseases that could cause visual field defects, history of intraocular surgery other than uneventful cataract surgery, unreliable visual field tests (fixation loss rate >20%, or false-positive or false-negative error rates >25%), and poor-quality SD-OCT and SS-OCT images that did not allow the precise segmentation of each inner retinal layer. When both eyes were eligible, one eye was randomly chosen for data analysis.
POAG was defined as the presence of glaucomatous optic nerve damage (neuroretinal rim notching or thinning, or presence of an RNFL defect), an open angle and associated visual field defects without ocular disease or conditions that may elevate the intraocular pressure (IOP). A glaucomatous visual field defect was defined as (1) outside the normal limits on a glaucoma hemifield test; or (2) three contiguous nonedge points (allowing for the two nasal-step edge points), with a probability of <5% of being normal and one with a probability of <1% based on pattern deviation; or (3) a pattern standard deviation of <5%. Those visual field defects were confirmed on two consecutive reliable tests (fixation loss rate of 20%, and false-positive and false-negative error rates of 25%).
Healthy subjects were defined as those having an IOP of <21mmHg, a normal-appearing optic disc, and normal visual field results. The visual field was considered normal when there was no glaucomatous visual field defect or neurological defect.

cpRNFL and macular inner retinal layer analysis using SD-OCT and SS-OCT
The SD-OCT and SS-OCT images of the macular area were obtained on the same day. The cpRNFL was scanned using the circular scan mode of the Spectralis OCT system, which consisted of 768 A-scans. The scan circle subtended 12°, and the diameter in millimeters depended on the axial lengths.
The volumetric scanning of the macular area was performed using both SD-OCT and SS-OCT for each subject. For SD-OCT, 61 B-scan sections parallel to the foveo-disc axis were obtained in the perifoveal area covering 8.9 mm × 7.4 mm. Each section had nine OCT frames averaged, and adjacent sections were separated by 126 μm. Images are obtainable using the Spectralis OCT system only when the quality score is >15; at lower scores the image acquisition process automatically stops and the image of the corresponding section is excluded. This study only included eyes with a quality score of >15 in all sections. The new software provided with the Spectralis OCT system automatically segmented the individual retinal layers, and the thicknesses of the mRNFL, GCIPL (calculated as the summation of the GCL and IPL thicknesses), and total retinal layer (TRL) were recorded for comparison with SS-OCT results. For SS-OCT, a 3D imaging data set was acquired for each subject with a raster scan protocol of 512 × 256 A-scans per data set. Each 3D scan covered an area of 7 mm × 7 mm centered on the fovea, and adjacent sections were separated by 27 μm. The segmentation algorithm of DRI SS-OCT provides the mRNFL, GCIPL, and TRL thicknesses.
The accuracy of the segmentation of each retinal layer (mRNFL, GCIPL, and TRL) and adequate centering on the fovea were reviewed independently by two masked observers (K.M.L. and E.J.L.). Only images that were considered adequate by both observers were included in the analysis.
In both types of OCT, we used retinal thickness map analysis to display numeric averages of the measurements for each of nine ETDRS subfields (Fig 1). The inner, intermediate, and outer rings with diameters of 1, 3, and 6 mm, respectively, were considered for the analysis. The average thicknesses of the following nine zones were used in the analysis: central fovea (CF), inner superior (IS), inner nasal (IN), inner inferior (II), inner temporal (IT), outer superior (OS), outer nasal (ON), outer inferior (OI), and outer temporal (OT) (Fig 1).

Statistical analysis
Measurements recorded by a device reflect a combination of the true value (which is unknown) plus device-specific systematic error (relative bias) and a random error (imprecision). To account for the systematic and random errors present in each of the two devices, structural equation model (SEM) was used, and calibration equations, bias and imprecision of the measurements were calculated [28][29][30]. Fig 2 shows the path diagram illustrating the SEM that describes the relationship between the OCT measurements and the unknown true values. In the SEM, DRI SS-OCT measurement was chosen arbitrarily as the reference (path coefficient fixed to 1), and the method of full information maximum likelihood (FIML) was used to estimate the SEM parameters.
The measured mRNFL, GCIPL, and TRL thicknesses were compared between the two OCT systems using the paired t-test. The POAG and control groups were compared using the independent t-test. The threshold for statistical significance was set at p<0.05. Bonferroni correction was applied to raw data in the comparison of the subfield thicknesses and the areas under the receiver operating characteristics (ROC) curves (AUCs) on the basis of the number of comparisons within each analysis. The diagnostic ability of each measurement to differentiate between groups was determined by calculating the AUC. The ROC curve shows the trade-off between sensitivity and specificity. An AUC of 1.0 represents perfect discrimination, whereas an AUC of 0.5 represents chance discrimination. The following established five-category rating scale was used for interpreting AUC values: >0.90, excellent; 0.80-0.90, good; 0.70-0.80, fair; 0.60-0.70, poor; and 0.50-0.60, fail [31]. The statistical significance of the differences in the AUCs between the paired ROC curves of each system was calculated [32]. ROC regression analysis was performed using a probit model by maximum likelihood estimation to adjust the covariate effects of age, refractive error, gender, and glaucoma severity [33,34].
Statistical tests were performed with commercially available software (Stata version 13.0, StataCorp, College Station, TX, USA) and the OpenMx SEM package for the R language for statistical analysis to compute FIML estimates and likelihood-based confidence intervals. Except where indicated otherwise, the data are presented as mean±standard deviation values.

Results
Sixty-five POAG eyes and 65 control eyes were initially enrolled. Of these, five eyes (three POAG eyes and two control eyes) were excluded due to poor image quality, and five eyes (two POAG eyes and three control eyes) were excluded during the age matching, leaving a final study population of 60 POAG eyes and 60 control eyes. The demographic and clinical characteristics of the POAG and control groups are summarized in Table 1.
The TRL was thicker when measured using SD-OCT than using SS-OCT in all ETDRS subfields, with the thickness ratio between the SD-OCT and SS-OCT measurements being almost constant among all nine subfields (Table 2). This was because Spectralis SD-OCT set the outer boundary as the retinal pigment epithelium while DRI SS-OCT determined it as the junction between the photoreceptor outer segment and the retinal pigment epithelium (Fig 1).
The GCIPL was thicker in the inner zones and thinner in the CF and outer zones when measured using SD-OCT than when measured using SS-OCT (Table 3). The differences between the systems were significant in all of the subfields except for the IT zone in the POAG group (Table 3). The mRNFL thickness differed significantly between SD-OCT and SS-OCT only in the central and ON zones in the POAG group, which were larger when measured with SD-OCT (Table 4). In the control group, the difference in the mRNFL thickness between the systems was significant in six subfields (CF, IT, and all four outer zones): except for the IT and OT zones, the SD-OCT measurements were larger than the SS-OCT measurements (Table 4).
Calibration equations between the two devices obtained based on the SEM parameter estimates and the calibration curves are shown in  3B).
The glaucoma-diagnosis ability of each macular parameter was assessed and compared between the two OCT systems using AUCs (Table 5). For both SD-OCT and SS-OCT, good AUCs were found for the GCIPL thickness in the OI zone (AUC = 0.818 and 0.847, respectively) and OT zone (AUC = 0.894 and 0.825, respectively) and for the mRNFL thickness in the OI zone (AUC = 0.859 and 0.852, respectively). The AUC of Spectralis SD-OCT was largest for the GCIPL thickness in the OT zone [AUC = 0.894, 95% confidence interval (CI) = 0.839-0.950], which was significantly larger than the AUC of DRI SS-OCT in the corresponding subfield (p = 0.003). For DRI SS-OCT, the AUC was largest for the mRNFL thickness in the OI zone (AUC = 0.852, 95% CI = 0.783-0.921), but it was only comparable to the AUC of Spectralis SD-OCT in the corresponding sector (p = 0.770). The AUCs were comparable between the two systems in most of the subfields, but Spectralis SD-OCT performed better than DRI SS-OCT for the mRNFL thickness in the ON zone (p = 0.001) and for the GCIPL thickness in the OT zone (p = 0.003; Table 5). An ROC regression model was applied to determine the factors affecting the glaucoma-diagnosis ability for the GCIPL thickness in the OT zone; the factors analyzed were age, gender, spherical error, visual field mean deviation, and the type of OCT. The regression model revealed that a larger AUC was associated with a smaller visual field mean deviation and the use of Spectralis SD-OCT (Table 6). Spectralis SD-OCT was better than DRI SS-OCT for measuring the GCIPL thickness in the OT zone irrespective of the degree of visual field damage (Fig 4).

Discussion
The present study compared the diagnostic abilities of Spectralis SD-OCT-and DRI SS-OCTbased macular inner layer analysis to differentiate glaucomatous eyes from control eyes. The results showed that although the diagnostic abilities of the two OCT systems were comparable in most of the macular subfields, Spectralis SD-OCT was better in some areas. To our knowledge, this is the first study to compare the ability of new segmentation software for Spectralis SD-OCT in the macular area with that of DRI SS-OCT. While the agreement between the two systems was generally good (Fig 3), there was an inherent difference in the measurement of the macular retinal layer thickness between the two OCT systems. Firstly, TRL was universally thicker when measured using SD-OCT than SS-OCT. Secondly, GCIPL was thicker in the inner zones while the other zones were thinner for SD-OCT compared to SS-OCT. Thirdly, mRNFL thickness differences between the systems varied among sectors, but SD-OCT generally produced thicker measurements. The intersystem differences in the TRL thicknesses were due to the difference in the inherent segmentation algorithm between SD-OCT and SS-OCT. However, we currently do not have clear answers for the source of discrepancy in the GCIPL and mRNFL thicknesses between the systems, and which system would provide more accurate measurements. Several postulations could be made based on the findings of this study, as described below.
In some GCIPL (IT; Table 3) and mRNFL (IT, OS, OI and OT; Table 4) subfields, the significant differences between the two systems that were evident in the control group were not present in the POAG group. The pattern was also discernible in the calibration curves (see Fig 3), where the thickness differences between the two systems were generally larger in the control group than in the POAG group. This finding may indicate that the DRI SS-OCT underestimated the loss of RNFL or GCL, thereby producing measurements in which the RNFL or GCL in the damaged area was thicker than for the Spectralis SD-OCT, which may have reduced the inherent discrepancy between the OCT systems. The low detectability of the damaged inner retinal layers using DRI SS-OCT may be attributable to several factors. Firstly, DRI SS-OCT uses longer wavelength, which penetrates deeper into the tissue [35][36][37][38]. Although this deeper tissue penetration of SS-OCT may facilitate the examination of deep tissues [35][36][37][38], the longer wavelength also means that the resolution is not necessarily sufficiently higher in the inner retinal tissues. We speculate that the lower resolution of the inner retina in DRI SS-OCT images may have resulted in a lower accuracy in the inner retinal layer segmentation, and so reduced the possibility of detection of the thinned area. Secondly, Spectralis SD-OCT involves averaging nine OCT frames for a single B-scan image, while DRI SS-OCT does not involve averaging multiple images but uses denser sampling than that in Spectralis SD-OCT [39]. Although this denser sampling potentially yields images of higher quality, the averaging of multiple images in the Spectralis system may enhance the image contrast and thereby facilitate segmentation of Comparison of SD-OCT and SS-OCT Macular Analysis Table 5. AUC, sensitivity and specificity of the macular inner retinal layer parameters obtained by the SD-OCT and SS-OCT. Comparison of SD-OCT and SS-OCT Macular Analysis the inner retinal layer. In terms of indicating the presence of glaucoma, less-dense sampling with a higher signal-to-noise ratio (in the Spectralis system) may provide better stability of the segmentation compared to denser sampling with the associated likelihood of a higher variability of the measurements (in the DRI-OCT1 Atlantis system). In addition, the eye-tracking functionality of the Spectralis system may reduce motion artifacts in the course of acquiring multiple images [40]. Thirdly, it is also possible that the less-precise segmentation of SS-OCT was attributable to the intrinsic segmentation software of the DRI system. Despite the potential superiority of Spectralis SD-OCT in segmentation of the inner retinal layer, Spectralis SD-OCT and DRI SS-OCT exhibited very similar glaucoma-diagnosis abilities. However, it is noteworthy that the AUC was largest for the GCIPL thickness in the OT zone as measured by SD-OCT, and significantly larger than the AUC of SS-OCT in the corresponding zone. ROC regression analysis also showed that measurements of the GCIPL thickness in the  OT zone were better for SD-OCT than for SS-OCT, irrespective of the degree of visual field damage. This is suggestive of the superiority of Spectralis SD-OCT over DRI SS-OCT in the evaluation of the GCIPL, because the OT zone of the macula is one with the earliest sites of GCIPL damage in glaucoma [41].
In the present study, the AUCs were largest in the OT and OI zones for both SD-OCT and SS-OCT. The better diagnosis ability when using the outer zones than the inner zones is consistent with the results of previous studies [13,42]. The inferior arcuate fibers followed by the superior arcuate fibers are known to be the most susceptible to glaucomatous damage, while papillomacular fibers are generally the last to be involved in glaucoma [43]. Therefore, it is relevant that glaucomatous damage is more likely to be detected in the outer macular zones. On the other hand, the better diagnosis ability when using the inferior and temporal zones than the superior or nasal zones may be explained by the different distribution of the superior and inferior arcuate fibers. Most RGCs in the inferior macular region project their axons to the inferior optic nerve head (ONH) quadrant [44]. Therefore, glaucomatous damage at the inferior pole of the ONH may result in damage to the inferior and inferotemporal macular zones [41]. However, RGCs in the superior macular region project axons to the temporal ONH quadrant [44], which is generally not involved until end stage of disease [43]. Thus, significant thinning would not be detected in the superior macular zone. Meanwhile, axons entering the superior ONH quadrant are originated from the area which is not included in the mGCL analysis. Thus, axonal damage occurring at the superior pole of the ONH may not be detected in the macular analysis [41].
On the other hand, the location with the best AUC differed between the mRNFL and GCIPL parameters: only the OI zone had a good AUC for the mRNFL thickness parameters, while both the OI and OT zones showed good AUCs for GCIPL thicknesses; this may be attributable to the spatial discrepancy between the axons and their original RGCs [41,45]: Cell bodies located at the temporal retina project their axons mainly through the inferior retina, and so mRNFL damage in the OI zone may be associated with damage to the GCL in the OT zone. Meanwhile, it is also possible that the different topographies of the RNFL and GCL in the macular area influence the ability to detect thinning in each macular zone [46]: the mRNFL is originally thinner in the OT zone than in other macular zones, and thus thinning of the mRNFL in this area might not have been detected by the segmentation algorithm.
The present study was limited by several factors. Firstly, there were inherent differences in the segmentation algorithm between Spectralis SD-OCT and DRI SS-OCT. Other than the difference in the definition of the TRL, the method to obtain the GCIPL differed between systems: Spectralis SD-OCT measured the GCL and IPL separately, while in DRI SS-OCT the GCIPL measured GCL and IPL together as a single parameter. Secondly, neither of the two OCT devices provides a normative database of macular inner layer thicknesses, and so the diagnostic performance was evaluated based on the AUC. Thirdly, the AUC values presented in this study indicate that OCT macular inner retinal layer analysis alone might not be an adequate technique for glaucoma diagnosis; that is, while it could be a useful adjunct for evaluating the glaucomatous damage of the macula, it might not replace conventional methods in the diagnosis of glaucoma. Forthly, the thickness measurement can be influenced by various factors, such as the direction of the light source or the head posture of the patient. An oblique light source could result in overestimation [47], and cyclotorsion of the fundus may result in a different allocation of the mean thickness in each ETDRS subfield. Lastly, the anatomical orientation specific to each eye, such as the fovea-Bruch's membrane opening plane [48,49] or the direction of the temporal raphe [50], were not considered in the ETDRS subfield analysis. Although the Spectralis OCT system evaluates the macular thickness based on the fovea-Bruch's membrane opening plane, this concept is not incorporated in the DRI-OCT1 Atlantis system, and so ETDRS subfield based analysis, which does not consider the fovea-Bruch's membrane opening plane was chosen in both systems for the comparison.
In conclusion, the ability to diagnose glaucoma based on analyzing the macular inner layer thickness was largely comparable between Spectralis SD-OCT and DRI SS-OCT. The AUC for diagnosing glaucoma was largest for the GCIPL thickness in the OT zone as measured by Spectralis SD-OCT. This is the zone where glaucomatous RGC damage occurs most frequently, and hence the present results suggest the potential superiority of Spectralis SD-OCT over DRI SS-OCT in diagnosing glaucoma in the macular region.