Spectral-domain optical coherence tomography (SD OCT) has had a profound impact on clinical and experimental ophthalmology, but also presented practicing eye specialists with the important and expensive decision of whether to invest in SD OCT technology. Unlike time-domain (TD) OCT, for which only one company had commercial propriety, there are several SD OCT systems currently available. Also, unlike other previous potential competitors of TD OCT, all SD OCT devices provide fast, high-resolution, retinal cross-sectional scans. Therefore, a method to differentiate these systems is desirable, especially in nonexperimental settings.
In their study published in this issue, Pierro and associates addressed this quandary by prospectively evaluating the macular thickness measurements and interobserver and intraobserver reproducibility of 7 commercially available OCT devices, including 6 SD OCT platforms, in healthy eyes. They determined, as have previous studies, that the devices measure macular thickness differently and have variable coefficients of correlation (intraclass correlation) and variation. These differences may help to provide the basis for individualizing and qualifying each OCT device.
Differences in Macular Thickness Measurement
Although a conversion factor has been postulated by some studies examining the eyes of healthy volunteers and finding strong correlation between different OCT systems, most research suggests that no such direct quantitative transfer is possible, because the correlation and variation is wide ranging in eyes with clinically relevant macular pathologic features. Indeed, in a recent statistical analysis comparing Stratus OCT (Carl Zeiss Meditec, Dublin, California, USA) with several SD OCT systems, the authors point out that although a strong correlation is present between Cirrus (Carl Zeiss Meditec) and Stratus OCT measurements in healthy eyes ( r = 0.89), spanning 95% confidence intervals for these regression plots indicate that a straightforward conversion equation to convert SD OCT and TD OCT measurements interchangeably is not feasible because of a predictive error of approximately 26 μm per central subfield thickness measurement observed between systems. The bottom line is that various instruments measure different portions of the central macula and the remaining unmeasured parts have their own intrinsic thickness variation. The implicit assumption with a theoretical correction factor is that the unmeasured parts have no variation in thickness, which is not true for both healthy and diseased retinas. These data and the current study support the belief that different OCT systems cannot be used interchangeably for the measurement of macular thickness.
However, definite, reproducible differences in macular thickness measurement are present between devices because of segmentation boundary variation between each system’s software algorithms. Although the inner limiting membrane acts as an inner reference boundary, the outer reference boundary is variable between systems. In a previous study of 6 SD OCT systems, one group found that the Stratus OCT measured the least thick on average (approximately 204 μm) with an outer boundary approximately at the photoreceptor inner segment/outer segment junction, followed by Spectral OCT/SLO (approximately 244 μm; OPKO/OTI, Miami, Florida, USA), RTVue-100 (approximately 247 μm; Optovue, Fremont, California, USA), and SOCT Copernicus (approximately 248 μm; Optopol, Zawiercie, Poland), which all have an outer boundary near the inner border of the retinal pigmented epithelium (RPE), followed by Cirrus (approximately 277 μm) and Spectralis (approximately 289 μm; Heidelberg Engineering, Vista, California, USA), which seem to have an outer retinal boundary near the mid-RPE (Cirrus), near the outer RPE, or near the junction of Bruch membrane and the choriocapillaris (Spectralis). The current study confirms this interdevice measurement variation, with the exception of SOCT Copernicus, which measured an average of approximately 172 μm, nearly 75 μm less than in the previous study. Although our group has noted a trend toward differences in macular thickness among populations of different ethnicities, in this case, it is more likely because the authors chose a different outer reference boundary location. In the newer SOCT Copernicus software (currently version 4.1), retinal thickness can be calculated either from the inner limiting membrane to the photoreceptor inner segment/outer segment junction, similar to TD software, or from the inner limiting membrane to the inner boundary of the RPE. The inclusion of a greater retinal area for quantification theoretically may allow for more potential change in pathologic conditions to be appreciated, such as the presence of subretinal fluid. In practice, clinicians should analyze the entire OCT image along with each individual scan section, and should not base decisions on macular thickness value alone.
There is a paucity of data examining whether SD analysis may yield improved correlation because of the greater number of points tested with SD OCT in healthy and diseased eyes. We believe that these differences may be important because, especially if comparing data between devices, factious differences may be induced by algorithm variation that change the point of outer retinal reference and, therefore, the macular thickness relative to the observer. This is especially relevant in practices transitioning from TD to SD OCT; unfamiliarity with the new system may lead to overrepresented macular thickness in newer SD OCT systems with more outer retinal reference boundaries. A system-specific nomogram may benefit those practices that are transitioning between TD and SD systems. At least 1 study has analyzed a large number of healthy and diseased eyes with 1 system and devised a relative nomogram for such processes. This sort of data may be of benefit for other individual SD OCT devices because it may provide a clinically relevant reference point. Furthermore, SD OCT en face visualization of the retina, particularly of the inner segment/outer segment photoreceptor junction, may provide improved correlation with retinal structural findings and visual function (Kiernan DF, et al. IOVS 2010;51:ARVO E-Abstract 3864).
The Importance of Reproducibility
Interoperator reproducibility reflects the learning curve necessary to become proficient in using a SD OCT device and is a measure of the variability in macular thickness measurement acquired by a trained user. Higher reproducibility may be especially desirable in the setting of clinical trials with strict protocols regarding macular thickness quantification, as was the case with the Variable-Dosing Regimen with Intravitreal Ranibizumab for Neovascular Age-Related Macular Degeneration study, commonly referred to as PrONTO, which used TD OCT technology. The fact that Spectralis and Cirrus SD OCT seem to have the highest interoperator reproducibility in the present and a previous study may influence future clinical trials. In fact, the ongoing Study of Efficacy and Safety of 0.5 mg and 2.0 mg Ranibizumab Administered Monthly or on an As-Needed Basis in Patients with Subfoveal Neovascular Age-Related Macular Degeneration is only using SD OCT Cirrus for gathering tomographic patient data. On the other end of the spectrum, as the authors of the current study point out, if the fact that the 3D OCT 1000 (TOPCON, Paramus, New Jersey, USA) had the lowest intraoperator reproducibility was the result of fundus alignment and focusing taking longer than with other devices, this may be a serious consideration for future company hardware modifications. A Diabetic Retinopathy Clinical Research Network ( DRCR.net ) trial, which recently completed recruitment, is prospectively comparing TD Stratus with either SD OCT Cirrus, Spectralis, or 3D OCT 1000; this likely will help to demonstrate the importance of device reproducibility in the setting of diabetic macular edema in a variety of devices.
Intraclass correlation is the ratio of how closely associated sequential macular thickness measurements are in the same eye. Thus, an ideal intraclass correlation should be close to 1.0 when scanned with the same machine. This value may be lower depending on variables such as patient movement, blinking, poor fixation, or off-center targeting by the operator. These variables can be minimized with software that stops the OCT scan when loss of target fixation is detected and resumes when the fovea again is centered. With a separate eye-tracking laser, it perhaps is not surprising that Spectralis has the highest intraclass correlation in the current and a previous study. Also, despite arbitrary values based on statistics, it is not clear what a significant difference in intraclass correlation might be for this data. For example, the intraclass correlation for the RTVue-100 has a mean of 0.82 but a 95% confidence interval of 0.59 to 0.96, which is quite large and spans most measurements shown for other devices. What is of particular interest is that TD Stratus had a higher intraclass correlation than 4 of the SD OCT systems tested with both operators. This may reflect a learning curve bias, because the Stratus is the most widely used OCT system and likely was used much more by the operators than the other systems; although the 2 operators had similar practical OCT experience, it is not stated how much experience on each system they had. Regardless, these data indicate that tried-and-true TD Stratus demonstrates intraclass correlation that is at least as good as many SD OCT systems.
With regard to coefficient of variation, based on statistics and probability theory, values less than 1 are considered low variance, whereas those more than 1 are considered high variance. In the present study, only 2 of the devices (Spectralis and Cirrus SD OCT) may be considered as having low variation, and the TD Stratus showed the highest variation, possibly because of the longer acquisition time associated TD technology. The authors point out that the eye-tracking laser of the Spectralis and the fast acquisition speed of the Cirrus may contribute to their lower coefficient of variation. However, the SOCT Copernicus, which also had an eye-tracking system used in this study, had a much greater degree of variation. Despite this, the authors do not offer statistical data indicating whether the different coefficients of variation observed between systems was of statistical significance. Because the coefficient of variation compares dispersion of data, the values should have been put into perspective by providing a reference value based on the current data.