|Home | About | Journals | Submit | Contact Us | Français|
To determine the agreement between peripapillary retinal nerve fiber layer (RNFL) thickness measurements from Stratus time domain optical coherence tomography (OCT) and Cirrus spectral domain OCT (Carl Zeiss Meditec, Dublin, CA) in normal subjects and glaucoma patients.
Evaluation of diagnostic test or technology.
One hundred thirty eyes from 130 normal subjects and glaucoma patients were analyzed. The subjects were divided into Normal (n=29), Glaucoma Suspect (n=12), Mild Glaucoma (n=41), Moderate Glaucoma (n=18), and Severe Glaucoma (n=30) by visual field criteria.
Peripapillary RNFL thickness was measured with Stratus Fast RNFL and Cirrus 200 x 200 Optic Disc Scan on the same day in one eye of each subject to determine agreement. Two operators used the same instruments for all scans.
Student paired t-testing, Pearson’s correlation coefficient, and Bland-Altman analysis of RNFL thickness measurements.
The average age of the glaucoma group was significantly older at 68.3±12.3 years versus 55.7±12.1 years. The average RNFL thickness (mean ± SD, in μm) for each severity group with Stratus OCT was 99.4 ± 13.2, 94.5 ± 15.0, 79.0 ± 14.5, 62.7 ± 10.2, and 51.0 ± 8.9, corresponding to normal, suspects, mild, moderate, and severe subjects, respectively. For Cirrus OCT, the corresponding measurements were 92.0 ± 10.8, 88.1 ± 13.5, 73.3 ± 11.8, 60.9 ± 8.3, and 55.3 ± 6.6. All Stratus-Cirrus differences were statistically significant by paired t-testing (p < 0.001) except for the moderate group (p = 0.11). For average RNFL, there was a highly significant linear relationship between Stratus minus Cirrus difference and RNFL thickness as well (p < 0.001). Bland-Altman plots showed that the systematic difference of Stratus measurements are smaller than Cirrus at thinner RNFL values but larger at thicker RNFL measurements.
RNFL thickness measurements between Stratus OCT and Cirrus OCT cannot be directly compared. Clinicians should be aware that measurements are generally higher with Stratus than Cirrus except when the RNFL is very thin as in severe glaucoma. This difference must be taken into account if comparing measurements made with a Stratus instrument to those of a Cirrus instrument.
Optical coherence tomography (OCT), first described by Huang et al,1 is a high-resolution imaging device that uses laser light to acquire in vivo images of the retina. OCT applies the principle of interferometry to interpret reflectance data from a series of multiple side-by-side A-scans combined to form a cross-sectional image. An automated segmentation algorithm, based on reflectivity changes between adjacent retinal layers, calculates the retinal nerve fiber layer (RNFL) thickness.2 OCT has been shown to be a highly reproducible imaging modality3–8 and to correlate positively with ex vivo histological measurements of the retina.9–10 The third-generation instrument, Stratus OCT (Carl Zeiss Meditec, Dublin, CA), relies upon time domain technology, a method of acquiring images by evaluating the interference pattern created by the echo time delays of backscattered light from the subject’s retina and those from a moving reference mirror. The most recent development in OCT imaging employs spectral domain technology. With spectral encoding of light via diffraction, spectral domain OCT does not use an adjustable reference mirror and does not need to measure reflectivity changes between retinal layers in each A-scan one at a time. Instead, the spectrometer detects the relative amplitudes of many optical frequencies simultaneously within the backscattered light. Thus many points are sampled at the same time, with all of the retinal layer depths of each A-scan calculated using Fourier transformation. This new technology vastly improves data acquisition speed, making it approximately 50 times faster.11 The commercially available spectral domain instrument from Carl Zeiss Meditec is called Cirrus OCT. It acquires a 6 mm by 6 mm cube of data consisting of 40,000 (200x200) points over the optic nerve in less time than it takes Stratus OCT to perform a Fast RNFL 3.46 mm circle around the optic nerve consisting of only 768 points.
There are over 8,000 Stratus OCT units and 1,000 Cirrus units in use worldwide (Carl Zeiss Meditec, direct communication, August 28, 2008) so many patients who have been scanned with Stratus OCT over the past six years are now potentially going to be scanned with Cirrus OCT. The purpose of the current study is to determine the relationship between peripapillary RNFL thickness measurements from Stratus OCT and Cirrus OCT in normal and glaucoma patients to identify any systematic differences.
The Human Subject Research Office (HSRO) of the University of Miami Institutional Review Board approved this study. The study protocol adhered to the Declaration of Helsinki and all subjects signed an informed consent as well as a separate Health Insurance Portability and Accountability Act (HIPAA) consent. Normal and glaucoma subjects were recruited between October 2007 and March 2008 at the Anne Bates Leach Eye Hospital, Department of Ophthalmology, Miller School of Medicine, University of Miami. Each subject underwent a screening eye exam to determine eligibility. The exam included assessment of visual acuity (VA), intraocular pressure (IOP), and the optic nerve as well as reviewing previously acquired Humphrey Visual Field (HVF, Carl Zeiss Meditec, Dublin, CA) exams.
Inclusion criteria for normal subjects were a best-corrected VA of 20/40 or better, normal IOP, and normal appearing optic nerves without asymmetry, cupping, hemorrhages, or notches. Exclusion criteria for normal subjects included a history of glaucoma or any other ocular disease or systemic disease such as diabetes. Glaucoma subjects were included if they had a definitive diagnosis of glaucoma, defined as glaucomatous visual field loss with an accompanying optic nerve abnormality typical of glaucoma. The glaucoma needed to be stable, and the patient required at least two reliable SITA Standard 24-2 HVF examinations, with the most recent exam within 12 months of the enrollment date. Glaucoma subjects were excluded if they had any history of previous retinal disease such as macular degeneration or optic nerve disease, including non-glaucomatous optic neuropathy. Both normal and glaucoma images were also excluded if they had unusable OCT scans due to poor signal strength (< 6) or algorithm failure, in which the boundaries for the RNFL layer were not delineated correctly.
Criteria for glaucomatous visual field defect were as follows:12,13 Glaucoma Hemifield test (GHT) outside normal limits, pattern standard deviation (PSD) with p values < 5%, or a cluster of 3 or more points in the pattern deviation plot in a single hemifield (superior or inferior) with P values < 5%, one of which must have a p value < 1%. Any one of the preceding criteria, if repeatable, was considered sufficient evidence of a glaucomatous visual field defect. An attempt was made to recruit glaucoma subjects across the entire spectrum of severity, including those with mild, moderate, and severe visual field defects, according to standard visual field severity grading.14 The glaucoma suspect group either had elevated intraocular pressure or suspicious looking optic discs without qualifying abnormalities on visual field testing.
One eye fitting the study criteria was chosen for the study. This was done randomly for normal subjects, but with the intent to create three subgroups for the glaucoma subjects based on prior visual fields. After pharmacologic dilation, the study eye was scanned with both the Cirrus OCT and Stratus OCT on the same day in a random order. The same Stratus and Cirrus instruments were used for all subjects by two trained operators (OK, RC).
Three separate Fast RNFL Scans were acquired with Stratus OCT (software version 4.0.7). The subject was seated and properly aligned. The OCT lens was adjusted for any refractive error. After bringing the optic nerve into view on the live fundus camera using the internal fixation light, the Z-offset and enhancement (polarization) were optimized. The aiming circle was adjusted to be centered over the optic nerve head so that an OCT TSNIT (temporal, superior, nasal, inferior, temporal) profile of an even 1.73 mm radius could be obtained. If the subject moved or blinked during the scan, the image was repeated. Scans were discarded and retaken if the signal strength was less than 6. Each Fast RNFL Scan captured 3 successive circular scans around the disc with A-scan measurements at 256 points per revolution over a total of 1.92 seconds. The resulting RNFL thickness output is an averaged value computed from a combined set of 768 A-scans. From the exported group of 3 scans, the most reliable scan was chosen for comparison (defined as the highest percentage of acceptable A-scans with the highest signal strength).
Three individual 200 x 200 cube Optic Disc Scans were obtained with Cirrus OCT (software version 18.104.22.168). After the subject was seated and properly aligned, the iris was brought into view. The scanning laser ophthalmoscopic (SLO) image was focused by adjusting for refractive error. Once the optic nerve head was centered on the live SLO image using the internal fixation cross, the centering (Z-offset) and enhancement (polarization) were optimized. A 6 mm x 6 mm square of data was captured. If the subject had an involuntary saccade within a 1.73 mm radius during the scan, it was discarded and retaken. The minimum acceptable signal strength was 6. Each Optic Disc Scan captured a 6 mm x 6 mm x 2 mm “cube” of data consisting of 200 A-scans from 200 linear B-scans (40,000 points) in ~1.5 seconds (27,000 A-scans/sec). As with Stratus OCT, the single scan with the highest signal strength and least movement was selected for comparison.
Since the data collected by Cirrus is a 6mm by 6 mm “cube,” rather than a single circle as in Stratus, a comparable circle (radius 1.73 mm) must be extracted from Cirrus data. The first step is for the software to locate the center of the optic disc. This is achieved by taking several slices and looking for the break in the RPE to map the central dark spot. Next, automated segmentation generates a complete cube RNFL thickness map. The data points along the circular path of a centered 1.73 mm radius are processed via bilinear interpolation and smoothing to create 512 A-Scans of 2D data. This TSNIT profile map is equivalent to the Stratus peripapillary RNFL scan. For both instruments, then, the reported RNFL values for the peripapillary scan circle are mean, quadrants (temporal, superior, nasal, inferior), and clock hours (12 starting temporally).
All analyses were performed using SPSS 15.0 (SPSS Inc., Chicago, IL). The RNFL comparison used values from one Stratus and one Cirrus scan, including the mean, temporal, superior, nasal, and interior quadrants. Again, of the three scans taken for each, the single best scan was defined as the one with the best signal score and least motion artifact. Student’s paired t-tests were used to compare the Cirrus and Stratus RNFL measurements. A Bland-Altman plot was graphed to assess the agreement between the two instruments. The same analysis was used for the mean, quadrant, and clock hour RNFL thickness values. An analysis of covariance test of slope*group interaction was used to determine whether the relationship between Stratus-Cirrus differences and (Stratus+Cirrus)/2 averages differed between diagnosis groups. Linear regression relationships were created for use in converting Cirrus to Stratus measurements for mean RNFL and each quadrant. Finally, the Cirrus Stratus variance from measurements made in this study was compared to the Stratus test-retest measurement variance, from a previously published study by this research group.8 Measurements in that study were collected by a different operator than the current study using the same Stratus OCT machine.
One hundred and fifty eyes of 150 subjects were enrolled. One hundred and thirty of the enrolled subjects had useable OCT scans. Table 1 summarizes the demographic and clinical characteristics of subjects. Subjects were placed into the following categories: Normal (n=29), Glaucoma Suspects (n=12), Mild Glaucoma (n=41), Moderate Glaucoma (n=18) and Severe Glaucoma (n=30). As one would expect, the average age of the glaucoma subjects was significantly older than the normal and suspects group and the average C:D ratio was larger in the glaucoma versus the normal and suspects group (P < 0.001 for both).
The average RNFL thickness for the normal subjects as measured by Cirrus was 92.0 ± 10.8 μm, while the average RNFL as measured by Stratus was 99.4 ± 13.2 μm. The average RNFL thickness for glaucoma suspects was 88.1 ± 13.5 μm using Cirrus and 94.5 ± 15.0 μm using Stratus. With the Cirrus, mean RNFL thicknesses in mild, moderate, and severe glaucoma were 73.3 ± 11.8, 60.9 ± 8.3, and 55.3 ± 6.6, respectively (mean ± SD, in μm). Using the Stratus, mean RNFL thickness in mild, moderate, and severe glaucoma were 79.0 ± 14.5, 62.7 ± 10.2, and 51.0 ± 8.9, respectively (mean ± SD, in μm). In comparing Stratus versus Cirrus, p-values by paired t-test were as follows: Normals Stratus > Cirrus, p<0.001; Suspects Stratus > Cirrus, p=0.001; Mild Stratus > Cirrus, p<0.001; Moderate Stratus > Cirrus, p=0.11; Severe Stratus < Cirrus, p<0.001.
The Pearson correlation coefficient for mean RNFL thickness between the two instruments was 0.958. The high degree of correlation is shown graphically in Figure 1. This level of correlation is expected given that the two instruments essentially measure the same thing. Using a Bland-Altman plot to detect agreement between Cirrus and Stratus, however, there is a small systematic difference (Figure 2). The Stratus minus Cirrus difference is proportional to RNFL thickness. For thinner RNFLs, Stratus measurements tend to be thinner than Cirrus, while for thicker RNFL thicknesses, Stratus measurements tend to be thicker than Cirrus. Figure 2 illustrates this relationship using the measured Cirrus minus Stratus difference plotted against the average of the two measurements. For mean RNFL, there is a highly significant linear relationship between Stratus minus Cirrus difference and RNFL thickness (p<0.001, linear regression test for non-zero slope). Further inspection of Figure 2 reveals that for mean RNFL, the variability around the regression line is consistent over the range of RNFL thicknesses. The slope relating Stratus minus Cirrus differences and mean RNFL thickness did not differ between the overlapping diagnosis groups (analysis of covariance test of group * slope interaction p=0.94). Interestingly, however, while there was no difference in the Cirrus minus Stratus difference (after accounting for the slope) between normal, suspect, mild and moderate groups (all p>0.47 post-hoc LSD test), the severe group had a highly significantly 6.3 micron (95% CI: 3.5, 9.1) greater Stratus minus Cirrus difference from the other diagnosis groups (p<0.001 after accounting for the slope).
The residual mean square Stratus minus Cirrus variance after the linear regression relationship is accounted for is 34.7, which is significantly greater than Stratus intra-session test-retest variance of (4.8/2)2 = 5.8 after multiplying the latter by a factor of two (due to differences in the way mean squares were calculated in these two studies) to get 11.6 (p<0.001).3 This finding was confirmed using data from our past reproducibility study providing same day test-retest observation pairs of Stratus RNFL average.8 These pairs were subtracted as if they were the Stratus-Cirrus differences in our current study. The variance of the differences was 14.3, which compares well with 11.6 from the data in the IOVS article.8 Thus, after correcting for the systematic effect (with linear regression), the Stratus/Cirrus difference variance of 34.7 is significantly larger (about 2x, so the SD is about 40% larger) than the Stratus test-retest difference variance. Again, this confirms that Cirrus and Stratus are different not just based on the test-retest variance of Stratus
Figure 3 shows Bland-Altman plots for each individual quadrant across the study population. There is a greater variation in the superior and inferior graphs, as evidenced by the diffuse spread, which have more measurements made in the thicker regions of the RNFL. Figure 4a displays a similar Bland-Altman plot combining all information from all clock hours of all patients and figure 4b provides the same information but indicating data from eyes with severe glaucoma. The accompanying non-parametric loess regression fits make clear that the systematic difference in Stratus and Cirrus measurements are mostly proportional to the RNFL thickness when RNFL thickness is less than 100 micrometers. This is the reason that the linear relationship is more evident in the nasal and temporal quadrants.
Table 2 supplies regression relationships which may be use to convert Stratus to Cirrus measurements. Figure 4 also suggests that the complex relationship between individual clock-hour Stratus minus Cirrus differences and RNFL thickness cannot be well characterized by a simple linear relationship, perhaps related to the high degree of test-retest variability in clock hour measurements, even with the same instrument.3,8
The results of the current study demonstrate that Cirrus RNFL measurements correlate well with those from Stratus OCT. However, there is a systematic difference in measurement values between the two instruments. This is shown in Figure 2, where the normal group is all above zero, but the severe group is almost all below zero. Essentially, Cirrus measures thicker at thinner “glaucomatous” RNFL values and measures thinner at thicker “normal” RNFL values. Therefore, the systemic difference is not in a single direction. This finding was also confirmed by Zangwill et al. in their ARVO 2008 poster (Zangwill LM, Vizzeri G, Bowd C, et al. RNFL Thickness Measurements with Cirrus HD-OCT and Stratus OCT. ARVO 4629/A255 poster 2008). However, because Zangwill’s glaucoma group did not include any moderate or severe glaucoma subjects, their difference of Cirrus being larger at thinner RNFL values was less pronounced.
To our knowledge, no previous studies have been published comparing RNFL thickness in normal and glaucoma patients scanned with Stratus OCT and Cirrus OCT. Two studies in particular have compared previous versions of OCT, namely OCT1 or OCT 2000 and OCT3 (Stratus) for RNFL measurements.15–16 Also, there have been a few published studies comparing macular retinal thickness across OCT instrument generations.17–20
Bourne and colleagues,15 found that RNFL thicknesses were thinner with OCT3 compared to OCT 2000. However, their study had very few glaucoma patients and those that they studied had early glaucoma as evidenced by low mean deviation indices, so the effect at varying RNFL thicknesses was not seen. They did conclude after application of a correction factor, though, that the variability exceeded the instrument’s limit of resolution (10 um). Therefore, they felt the agreement was unacceptably low from a clinical standpoint. Because the difference in our study changed at varying thicknesses in more than one direction, applying a correction factor would be more complex and difficult to use clinically.
In a study by Monteiro et al, 16 RNFL measurements were smaller with OCT3 compared to OCT1. The authors had proposed that a higher variability observed for OCT1 should lead to both greater and smaller RNFL values when compared to later versions, i.e. OCT3. This is the case when we compare Stratus with Cirrus across all thicknesses. Their study also concluded that the discrepancies between the two instruments worsened at thinner RNFLs, again, mirroring our Cirrus Stratus results but to a lesser degree. The Bland-Altman plot would have appeared more similar to ours had they plotted the band atrophy patients and normals on the same graph.
We could not find an exact explanation to explain the difference observed between Cirrus and Stratus OCT. Each patient was scanned by one examiner on the same two instruments in a single day. Perhaps the observed difference may be accounted for by modifications to hardware and/or software. The segmentation algorithm is different; Cirrus aims to identify the bottom of the nerve fiber layer whereas Stratus attempts to localize the top of the ganglion cell layer. Since we excluded subjects with scans that had signal strengths less than 6, one cannot generalize the results of this study to patients with poor signal strength.
Another possible explanation for Cirrus measuring thicker in severe disease is due to differences in measuring the exposed blood vessels as the RNFL thins out. Or, because Cirrus has less smoothing than Stratus, it may drop through to the next layer at times, resulting in thicker measurements. This relates back to the segmentation algorithm. Likely, the instruments appear to reach a “floor effect” in severe disease (though the floor of the two instruments may be different), so the variance of RNFL across all thicknesses may be the sum of variance of two instruments with a wider variation at normal RNFL values and less variation at the “floor.”
Because there is no external validation for measurements taken from either instrument, i.e., histology or manual segmentation, it is impossible to know which instrument is more accurate. However, given that a cube of data is now obtained over the optic nerve head, the reproducibility of Cirrus is likely to be better because the circle is extracted from the same location all the time, as opposed to manual placement, where patient movement may change the location each time. It is important for clinicians to be aware that there are differences between the Cirrus and Stratus OCT instruments. Because of this, patients should not be scanned back and forth with each instrument, which would introduce too much variability to detect any change over time. One of the difficulties that the clinician faces in following patients with newer imaging technologies is that the hardware and software changes every five to ten years requiring purchase of new equipment and establishing of a new baseline with the new technology. In the case of switching from Stratus to Cirrus OCT technology, the best strategy for switching from one to the other during the follow-up of a patient may be to get a follow-up Stratus scan to determine whether there has been change from prior Stratus scans and then getting a new baseline scan with Cirrus for future comparison.
A paper by Ishikawa et al.21 did note that refining the retinal segmentation algorithm to define the RNFL better had a major effect on the thickness results, and their algorithm found significantly lower RNFL values compared to the then standard OCT1. Should this be the case, each newer successive generation instrument may find their RNFL values to be smaller than the current version due to a more accurate algorithm. Because it is more difficult to measure RNFL at thinner measurements, we may be close to accurate values for thicker RNFL but not yet for thinner RNFL.
In conclusion, the current study demonstrates that RNFL thickness measurements in normal and glaucoma patients scanned with Stratus OCT correlate well with those from Cirrus OCT. However, correlation does not mean that the measurements are the same. Clinicians need to be aware that there are systematic differences between measurements made by the two instruments that vary with RNFL thickness.
unrestricted grant from the Research to Prevent Blindness, NIH P30 EY014801, Heed Ophthalmic Foundation (RC) Unrestricted grant from Carl Zeiss Meditec
The sponsor or funding organization had no role in the design or conduct of this research
Presented at the annual meeting of the Association for Research in Vision and Ophthalmology, April, 2008
Conflict of interest: The department for which all authors work has received research support from Carl Zeiss Meditec
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.