To compare retinal nerve fiber layer (RNFL) thickness measurements between 3 different spectral-domain optical coherence tomography (SD-OCT) instruments (Spectralis; Heidelberg Engineering; Cirrus; Carl Zeiss Meditec; RTVue; Optovue, Inc) and one time-domain OCT (Stratus; Carl Zeiss Meditec).
Prospective, cross-sectional study.
RNFL thickness was measured on both eyes of 40 normal subjects using Stratus, Spectralis, RTVue, and Cirrus OCT on the same day. Scans were repeated 2 to 8 weeks later in the same fashion. Agreement with Stratus was evaluated for each SD-OCT and intervisit reproducibility was assessed for all machines.
Mean RNFL thickness determined by each of the 3 SD-OCT instruments was highly correlated with Stratus (r = 0.87–0.91). The mean RNFL thickness (mean ± standard deviation) was 110.1 ± 12.8 μm for Stratus, 106.6 ± 12.8 μm for Spectralis, 98.7 ± 10.9 μm for Cirrus, and 112.8 ± 13.2 μm for RTVue. The average differences between each SD-OCT and Stratus for mean RNFL thickness were all statistically significant ( P ≤ .001), as were most quadrant measurements. All 4 instruments demonstrated excellent intraclass correlation coefficient values for mean RNFL thickness (0.90–0.97). However, intervisit variability was lowest for RTVue as evidenced by reproducibility values, followed by Stratus, Cirrus, and Spectralis (6.59, 8.83, 8.89, and 11.72 μm, respectively).
RNFL measurements taken with Spectralis, RTVue, and Cirrus all have excellent correlation to Stratus, with good reproducibility in normal eyes. Despite high correlations, RNFL values are significantly different between instruments and should not be used interchangeably.
For clinicians managing glaucoma, retinal nerve fiber layer (RNFL) thickness is often measured by optical coherence tomography (OCT) and followed as a means of objectively monitoring glaucomatous damage. Given that functional visual field loss often cannot be detected until up to 40% of the RNFL is lost, OCT also offers the potential capability of earlier detection of glaucomatous damage. Third-generation time-domain OCT (TD-OCT, Stratus; Carl Zeiss Meditec, Dublin, California, USA) has emerged as a reliable platform for aiding in the diagnosis and management of glaucoma. Recently, the advent of Fourier- or spectral-domain OCT (SD-OCT) has brought forth significant improvements in resolution and software capabilities when compared to TD-OCT. SD-OCT affords a robust increase in speed of image acquisition, allowing for multiple parallel B-scans to be acquired and summated into 3-dimensional (3D) volume data sets. Depending on the machine used, this allows for scanning speeds of 29 000 to 55 000 A-scans/second, compared to an average of 400 A-scans/second in TD-OCT. For example, Mumcuoglu and associates demonstrated that a TD-OCT image consisting of 512 A-scans can be acquired in 1.3 seconds, while SD-OCT can acquire 2048 A-scans in 0.085 seconds. In addition to improved speed compared to TD-OCT, SD-OCT also decreases motion artifacts and increases image resolution. Commercially available SD-OCT devices produce images with an in-tissue axial resolution of 5 to 7 μm compared with 10 to 15 μm in TD-OCT. In prototype SD-OCT models using an upgraded titanium:sapphire laser source, an even higher resolution of up to 2 μm is possible.
Despite perceived advantages, questions regarding the reliability of SD-OCT as well as degree of agreement of RNFL measurements between TD-OCT and SD-OCT remain to be answered conclusively. The progressive nature of glaucoma demands that treatment decisions be based on a consistent and reproducible evaluation of structure and/or function. Therefore, if SD-OCT is to replace and improve upon TD-OCT, the reliability of measurements and the relationship to TD-OCT measurements must first be elucidated. In this study, we compare the RNFL measurements between 3 different SD-OCT instruments ([Spectralis; Heidelberg Engineering, Heidelberg, Germany]; [Cirrus; Carl Zeiss Meditec, Dublin, California, USA]; [RTVue; Optovue, Inc, Fremont, California, USA]) and Stratus TD-OCT in patients without ocular pathology. We also determine the reproducibility of RNFL measurements for each SD-OCT instrument as well as Stratus on the same group of study participants.
Forty subjects were recruited between June 27, 2009 and August 25, 2009 at the Rocky Mountain Lions Eye Institute, University of Colorado Denver. All subjects underwent screening examinations to ensure eligibility requirements were met. The examination included assessment of Snellen visual acuity, refractive error, intraocular pressure, and optic nerve, as well as Swedish Interactive Threshold Algorithm 24-2 full-threshold Humphrey Visual Field (Carl Zeiss Meditec) and gonioscopy examinations. Screening examinations were performed by one of 2 experienced ophthalmologists (L.K.S. and M.Y.K.). Visual field examinations were administered by a single experienced ophthalmic technician and interpreted by a glaucoma specialist (M.Y.K.). Both eyes were examined for each patient.
Inclusion criteria consisted of a best-corrected visual acuity of ≥20/40 or better, spherical equivalent between +3.00 and −6.00 diopters, absence of ocular hypertension, normal anterior chamber with open angle, normal and reliable visual field examination, optic nerves without abnormalities of the neuroretinal rim, absence of cup-to-disc asymmetry greater than 0.2, and no optic nerve head hemorrhages. Exclusion criteria included a history of glaucoma, suspicion of glaucoma, or any other ocular disease except for mild age-related cataracts.
Optical Coherence Tomography Technique
Both eyes of each subject were scanned on each of 4 different OCT instruments in random order. For each subject, all 4 scans were completed on the same day, within a 2-hour period, by a single experienced ophthalmic technician. All 4 scans were repeated in a similar fashion on each eye 2 to 8 weeks later by the same technician in random order. Timing of subject reexamination was dependent on convenience for each subject.
For Stratus OCT (software version 4.0.2), the Fast RNFL mode was used for analysis. This protocol consists of 3 consecutive circumpapillary scans of 3.4 mm in diameter. Each circle scan is composed of 256 A-scans acquired consecutively. The computer algorithm automatically delineates the anterior and posterior margins of the RNFL and thickness is calculated by assessing 768 data points along the borders of RNFL. Signal intensity is listed on a scale from 1 to 10. Only scans with signal strength of ≥6 out of a possible 10 were included for analysis.
The Cirrus OCT (software version 184.108.40.206) scans were obtained using the Optic Disc Cube 200 × 200 protocol. Under this protocol, a 3D cube of data is generated over a 6-mm-square grid of 200 horizontal scan lines, each composed of 200 A-scans. A cirrus software algorithm automatically detects the center of the optic disc and positions a 3.46-mm-diameter calculation circle over this point. From the 256 A-scans along this circle, the borders of the RNFL are delineated and thickness is calculated at each point along the circle. All scans were reviewed to ensure signal strength of ≥6 for inclusion in analysis.
For RTVue OCT (software version 220.127.116.11), the NHM4 and 3D Disc protocols were used for analysis. Using the 3D Disc protocol, a 4 × 4-mm raster scan is acquired of 101 B-scans, each composed of 512 A-scans. The resulting en-face image is then used to manually draw a contour line along the disc margin. The NHM4 protocol is composed of 12 radial scans 3.4 mm in length and composed of 452 A-scans each. Using the previously defined contour line, 6 concentric ring scans are taken ranging from 2.5 to 4.0 mm in diameter, centered on the optic disc. This generates 9510 A-scans in approximately 0.39 seconds. The ring scan with diameter of 3.4 mm was used for comparison analysis. All scans were reviewed to ensure a signal strength indicator (SSI) of ≥40 out of 100.
The Spectralis OCT (software version 4.0) uses dual-beam simultaneous imaging with SD-OCT and confocal scanning laser ophthalmoscope to generate an en-face image on screen. The operator then manually centers a 3.4-mm-diameter circle on the optic disc. The eye-tracking system was activated for all scans including repeat scans, and the AutoRescan feature was implemented, allowing accurate placement of the scan circle in the same position each time. Fifteen images were then acquired at the scan circle under high-resolution settings (1536 A-scans) and averaged automatically by the software. RNFL boundaries were automatically delineated according to software algorithms underneath the circumpapillary circle. All scans were reviewed to ensure a signal strength >15 dB.
For each instrument, the subject was seated and properly aligned in typical fashion as is done in everyday practice. If the subject blinked or moved, resulting in an unreliable scan, the test was repeated. Any scans of poor quality were discarded. All scans were reviewed at study completion to evaluate for accurate computer detection of inner and outer boundaries of RNFL. If a scan possessed an error in segmentation, the information was not included in the statistical analyses.
All statistical analyses were performed with STATA software (StataCorp, College Station, Texas, USA). The average and 4-quadrant RNFL thickness measurements from each SD-OCT device were compared with those of Stratus by use of Student paired t tests. For the comparison of RNFL thicknesses between machines, only the RNFL thicknesses from the first examination were used for each eye. Since in certain patients both eyes were used in the analysis, calculated standard errors were clustered at the patient level to account for the possibility that observations from the right and left eye of the same patient may not be independent. Scatter plots were generated for each SD-OCT RNFL thickness against Stratus RNFL thickness. Pearson correlation coefficients were also calculated from these plots. Bland-Altman plots were generated to assess agreement of measurements between each SD-OCT and Stratus. In these graphs, the differences in thickness between instruments were plotted against the mean RNFL measurement.
Mean intervisit differences were calculated for each subject on all machines for average and quadrant RNFL thickness measurements. The differences were compared with the Student paired t test. Intervisit reproducibility values were calculated using the following formula: 2.77 × the intervisit within-eye standard deviation (Sw). The coefficient of variation was also calculated for all measurements (100 × Sw/Overall mean). The intraclass correlation coefficient (ICC) was calculated as a ratio of between-eye variance to total variance (between-eye variance + within-eye variance). The within-eye variance of average and quadrant RNFL thickness was compared between each SD-OCT and Stratus by using the Student paired t test on the log-transformed variance (log Sw 2 ). The difference in log-transformed within-eye variance between each machine will then be positive if the SD-OCT variance is greater than Stratus and negative if the variance is less. Bland-Altman plots were used to assess the relationship between the within-eye variance and RNFL thickness. P < .05 was considered to be statistically significant.
Agreement with Stratus
Seventy-nine of 80 eyes (98.8%) of 40 subjects qualified for inclusion in the study. Of these 40 subjects, 16 (40%) were male and 24 (60%) female. The mean (± standard deviation) age of subjects was 37.1 ± 11.0 years (range 21 to 61 years). Mean best-corrected Snellen visual acuity was 20/20, with spherical equivalent of −0.8 ± 1.7 diopters. Mean IOP was 15.6 ± 2.3 mm Hg, and mean cup-to-disc ratio 0.3 ± 0.1. From visual field analysis, mean pattern standard deviation was 1.62 ± 0.47 decibels (dB), while mean deviation was 0.18 ± 1.25 dB. Among the 80 eyes undergoing baseline examination, 1 eye was excluded because of history of unilateral trauma with question of angle recession. Of the 79 qualifying eyes, 8 (10.1%) were withdrawn from the analysis of agreement between machines. Seven of the 8 were withdrawn because of poor scan quality and/or failure of the software to accurately delineate the RNFL/ILM boundaries, and 1 was attributable to technician failure to scan on all 4 machines. For the reproducibility analysis, an additional 4 eyes (total of 12/79; 15.2%) were withdrawn because of poor scan quality of the repeat scans. The majority of poor-quality scans withdrawn were produced by Spectralis (10/11; 91%), while only one was from Cirrus.
The mean average RNFL thickness was 110.1 ± 12.8 μm by Stratus, 106.6 ± 12.8 μm by Spectralis, 98.7 ± 10.9 μm by Cirrus, and 112.8 ± 13.2 μm by RTVue ( Table 1 ). Both Spectralis and Cirrus measured average RNFL thicknesses significantly lower than Stratus; however, the difference was more pronounced for Cirrus. RTVue average RNFL measurements were significantly higher than Stratus. The mean difference between each SD-OCT instrument and Stratus was −3.36 μm for Spectralis, −11.33 μm for Cirrus, and +2.81 μm for RTVue. The association of average RNFL thickness measurements from each SD-OCT to Stratus is demonstrated by the scatter plot in Figure 1 . Bland-Altman plots in Figure 2 demonstrate the relationship of average RNFL thickness measurements taken by each SD-OCT and Stratus. The difference in average RNFL thickness between each SD-OCT and Stratus is plotted against the mean RNFL measurement of both instruments. The difference between Cirrus and Stratus was noted to have an increasing trend with higher average RNFL measurements. Comparisons of Stratus to RTVue and to Spectralis showed a fairly consistent discrepancy in measurements at all thicknesses. Pearson correlation coefficients listed in Table 1 were found to be quite high between all 3 SD-OCT machines and Stratus.
|Mean ± SD||SE|
|Average||110.10 ± 12.81||2.08|
|Temporal||75.79 ± 13.03||2.07|
|Superior||133.46 ± 16.71||2.7|
|Nasal||87.57 ± 16.85||2.7|
|Inferior||143.59 ± 19.89||3.08|
|Mean ± SD||SE||Mean Difference||P Value||r|
|Average||98.68 ± 10.89||1.78||−11.33||<.001||0.911|
|Temporal||64.88 ± 10.37||1.68||−10.87||<.001||0.746|
|Superior||123.52 ± 16.18||2.5||−10.01||<.001||0.701|
|Nasal||74.88 ± 10.31||1.72||−12.48||<.001||0.755|
|Inferior||132.01 ± 18.91||3.09||−11.35||<.001||0.861|
|Mean ± SD||SE|
|Average||106.59 ± 12.82||2.08||−3.36||.001||0.869|
|Temporal||78.54 ± 14.22||2.33||3.01||.034||0.748|
|Superior||131.4 ± 18.45||2.92||−2.10||.222||0.780|
|Nasal||78.12 ± 13.13||2.15||−9.35||<.001||0.659|
|Inferior||137.37 ± 18.95||3.05||−5.92||<.001||0.854|