| OCR Text |
Show Original Contribution Section Editors: Clare Fraser, MD Susan Mollan, MD Ganglion Cell Layer Thickness Variance Using SPECTRALIS Optical Coherence Tomography Paul Mallory, MD, Sushant Wagley, MD, John Chen, MD, PhD, Collin M. McClelland, MD, Peter A. Downie, BA, Bruce Lindgren, MS, Michael S. Lee, MD Background: To determine the normal variance of the mean macular ganglion cell layer (GCL) volume among subjects without significant ocular pathology using SPECTRALIS optical coherence tomography (OCT). Methods: Fifty subjects underwent a baseline scan using SPECTRALIS OCT followed by 2 more studies with (reg-ON) and without (reg-OFF) eye registration all taken at the same session. The mean GCL volume was measured using built-in SPECTRALIS software. Eyes with macular pathology were excluded. The reproducibility of the measurements of the GCL volume was evaluated with Bland–Altman plots and limits of agreement, intraclass correlation coefficient (ICC), and the coefficient of repeatability (CR). Results: A total of 98 eyes met criteria for the analysis. The mean GCL volume difference was 0.0002 ± 0.029 and 20.0005 ± 0.035 mm3 for scans 1 versus 2 (baseline vs reg-ON) and 3 (baseline vs reg-OFF), respectively. The ICCs were 0.985 and 0.977 for the baseline vs reg-ON and regOFF groups. The CR for baseline vs reg-ON was 0.056 while CR for baseline vs reg-OFF was 0.069. Ninety percent of eyes fell within 0.04 mm3 of test–retest reliability. Conclusions: Our model found a predictable threshold of 0.07 mm3 or less for SPECTRALIS OCT mean GCL volume variance, which did not significantly change with eye registration in eyes without macular pathology. Clinicians may also consider a threshold of 0.04 mm3 when determining stable vs progressive changes in mean GCL volume using this device. Journal of Neuro-Ophthalmology 2022;42:310–315 doi: 10.1097/WNO.0000000000001569 © 2022 by North American Neuro-Ophthalmology Society O ptical coherence tomography (OCT) is an accurate and noninvasive tool for assessing ocular anatomy with high resolution. It is especially important in the evaluation of the optic disc and retina. Computerized algorithms can separate individual image layers of interest, such as the retinal nerve fiber layer (RNFL) or ganglion cell layer (GCL) (1). The GCL contains ganglion cells and displaced amacrine cells that represent projection neurons which synapse to subcortical nuclei. The GCL volume correlates well with visual function and may herald early optic nerve dysfunction before RNFL loss in the setting of glaucoma. In addition, serial GCL monitoring may identify disease change (2). A normative database of the average thickness of the GCL and/or inner plexiform layer (IPL) and intratest reliability of the mean GCL/IPL thickness for some spectral domain OCT machines exist (3). However, OCT machines use different algorithms to identify the GCL/IPL complex vs the GCL volume, and their findings are not interchangeable. Previous studies have suggested good reproducibility for SPECTRALIS (Heidelberg Engineering Inc, Heidelberg, Germany) OCT for both intraobserver and interobserver studies. These studies have provided general interclass correlated coefficients and have not provided quantifiable data to assist clinicians to detect significant change outside normative variability (2,3). METHODS Department of Ophthalmology and Visual Neuroscience (PM, SW, CMM, PAD, MSL), University of Minnesota, Minneapolis, Minnesota; Departments of Ophthalmology and Neurology (JC), Mayo Clinic, Rochester, Minnesota; and Clinical and Translational Science Institute (BL), University of Minnesota, Minneapolis, Minnesota. The authors report no conflicts of interest. Address correspondence to Michael S. Lee, MD, Department of Ophthalmology and Visual Neuroscience, University of Minnesota, 420 Delaware St SE, MMC 493, Minneapolis, MN 55455; E-mail: mikelee@umn.edu 310 This study was approved by the Institutional Review Board at the University of Minnesota and followed all tenants of the Declaration of Helsinki. Consenting patients from the Department of Ophthalmology and Visual Neurosciences at the University of Minnesota were invited to participate in the study. Trained imaging technicians acquired 3 sets of macular images using a 20 · 20° capture mode and 25 raster cuts each with 16 automated retinal time frames on the SPECTRALIS OCT machine (version 6.13.1.0). First, a baseline scan was obtained for each eye. The baseline image was registered with the AutoRescan function—an option Mallory et al: J Neuro-Ophthalmol 2022; 42: 310-315 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution available with SPECTRALIS software, which aligns OCT scans at identical locations during repeat examinations. After a short break, a second scan with AutoRescan registration on (reg-ON) was obtained. Finally, a third retinal scan was obtained without the AutoRescan function (regOFF) after another short break. Using the GCL segmentation analysis feature in SPECTRALIS software, the mean GCL thickness volume was measured for all 3 capture modes. The mean thicknesses volume within a 6 mm area around the fovea was used. An image was only considered acceptable when signal-to-noise ratio (Q) was equal to or above 25 (per manufacturer recommendation). Studies with significant macular pathology (geographic atrophy and macular fluid), resulting in obvious segmentation errors were excluded. Every study was examined for the integrity of the segmentation process. Manual centration of segmentation boundaries was applied by a singular experienced operator. Statistical analyses were run to compare the GCL volume between the 3 capture modes. A predictable threshold for mean GCL volume variance was calculated by analyzing the frequency of difference in the mean GCL volume between the 3 capture modes. Mean comparisons were performed using the t test and analysis of variance (ANOVA) statistical tests. intraclass correlation coefficient (ICC) were calculated comparing baseline images with reg-ON and reg-OFF and between reg-ON and reg-OFF groups. Bland–Altman plots (4) compared the difference in volume with the average of 2 readings and included the average bias and the interval of agreement. As further evidence of agreement, the coefficient of repeatability (CR) was calculated. RESULTS There were 100 eyes of 50 subjects. Demographic information was not collected. Of the 100 eyes, 2 were excluded (poor Q value, geographic atrophy, and presence of macular fluid) leaving 98 eyes for the final analysis. In the 98 sets of scans analyzed, the mean GCL volume for the baseline, reg-ON, and reg-OFF groups were 1.04683 mm3 (SD 0.162), 1.04663 mm3 (SD 0.166), and 1.047347 mm3 (SD 0.165), respectively. The absolute mean difference in GCL volume between the baseline the and regON group was 0.0002 mm3 (SD 0.029) while the difference between the baseline and the reg-OFF group was 0.0005 mm3 (SD 0.035). Absolute mean difference in GCL volume between the reg-ON and reg-OFF group was 0.0007 mm3 (SD 0.036). There was no statistically significant difference between the mean GCL volume in 3 groups (ANOVA, P = 0.722). There was no significant difference between baseline vs reg-ON (P = 0.225), baseline vs reg-OFF (P = 0.243), and reg-ON vs reg-OFF (P = 0.480) groups. Summary of the mean differences are given in Table 1. ICC between baseline vs reg-ON group was 0.985, P = ,0.001, and 0.977 for baseline vs reg-OFF group, P = ,0.001. Correlation between reg-ON vs reg-OFF group was 0.977 P = ,0.001. The distribution of the frequency of variance is shown in Figure 1. Figure 1A shows the average total variance in all groups (baseline–reg-ON, baseline–reg-OFF, and reg-ON– reg-OFF). Figure 1B shows the frequency of total variance between baseline and reg-ON, and Figure 2B shows frequency of variance between baseline and reg-OFF. In the 98 eyes analyzed, there were zero differences between baseline, reg-ON, and reg-OFF scans in 22.4% of cases. The most common variance in this subset was a GCL volume difference of 0.01 mm3 which was seen in 36.7% of cases. A difference of 0.02 mm3 was seen in 19.0% of cases, 0.03 mm3 in 8.5% of cases, 0.04 mm3 in 3.1% of cases, and 0.05 mm3 in 2.0%. No study that met quality criteria for segmentation had a micron variance greater than 0.19 mm3. A total of 89.7% of cases fell within an absolute difference of less than or equal to 0.04 mm3 in all 3 groups. Figure 2 shows the Bland–Altman plot highlighting the distribution of mean differences compared with the average of the measurements. The CR was calculated for each group with CR for baseline vs reg-ON is 0.0557, CR for baseline vs regOFF is 0.0691, and CR for reg-ON vs reg-OFF is 0.0693. CONCLUSIONS To the best of our knowledge, this is the first numerical report of reproducibility of GCL volume measurements TABLE 1. Ganglion cell volume measurements with and without eye registration GCL Volume Mean (SD) Difference in GCL volume Mean (SD) P-value intraclass correlation coefficient (ICC) P-value CR Baseline Registration ON Registration OFF 1.047 (0.162) Baseline-reg ON 1.047 (0.166) Baseline-reg OFF 1.047 (0.165) Reg ON–reg OFF 0.0002 (0.029) P = 0.722 0.985 P , 0.001 0.0557 mm3 20.0005 (0.035) 20.0007 (0.036) 0.977 P , 0.001 0.0691 mm3 0.977 P , 0.001 0.0693 mm3 CR, coefficient of repeatability; GCL, ganglion cell layer. Mallory et al: J Neuro-Ophthalmol 2022; 42: 310-315 311 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution FIG. 1. Distribution of GCL thickness variance. A. Baseline vs reg-ON. B. Baseline vs reg-OFF. C. Reg-ON vs reg-OFF. GCL, ganglion cell layer 312 Mallory et al: J Neuro-Ophthalmol 2022; 42: 310-315 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution FIG. 2. Bland–Altman plots. A. Baseline vs reg-ON. Mean bias = 0.0002 (SD = 0.0286); correlation between difference and average = 20.13. B. Baseline vs reg-OFF. Mean bias = 20.0005 (SD = 0.0355); correlation between difference and average = 20.10. C. Reg-ON vs reg-OFF. Mean bias = 20.0007 (SD = 0.0355); correlation between difference and average = 0.01. Mallory et al: J Neuro-Ophthalmol 2022; 42: 310-315 313 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution using the SPECTRALIS OCT machine. We found that a variance of #0.04 mm3 in the mean GCL volume for both reg-ON and reg-OFF for normal eyes is within the realm of test–retest reliability for the cohort for 89.7% of the eyes. Accuracy and precision are foundational components to any imaging modality, and enhanced reproducibility allows clinicians to reliably discern changes when they occur. Imaging studies, such as OCT, often drive clinical decisionmaking, but parameters of reliability are required. OCT has a well-defined role in many areas of ophthalmic disease diagnosis and monitoring. Segmentation analysis of the GCL is of special interest in various glaucomatous and neuro-ophthalmic disease processes. A high dependency on imaging modalities exists because there is no way to reliably discern the integrity of the GCL without the aid of such tools. To further aid clinicians in determining significant changes in the GCL, our study assesses the reproducibility of SPECTRALIS OCT. Other studies have reported on the reproducibility of the OCT layer analysis in both pathologic and healthy states, suggesting that certain macular pathologies may lead to lower repeatability (5,6). Our study reinforces the diagnostic accuracy and reproducibility of SPECTRALIS OCT in normal eyes. In total, 98 eyes successfully underwent reproducibility testing with segmentation analysis of the GCL layer. Our findings show that the SPECTRALIS GCL scan is extremely repeatable in measuring the GCL as highlighted by the Bland–Altman plots. We then calculated the CR which showed that a measurement of 0.07 mm3 or more, on scans with or without registration, is statistically likely to be true anatomic change rather than variation in testing. This is also consistent with the distribution of variation graph for each of the groups. Additional diagnostic power does not occur by using registration because there was no significant difference between reg-ON and reg-OFF scans. Excluded outliers coincided with severe macular pathology, specifically significant macular edema or atrophy. The layers of the innermost retina include the RNFL, GCL, and IPL. Spectral domain OCT machines differentially report the macular ganglion cell analysis (GCA) as GCL alone (SPECTRALIS), GCL/IPL (Cirrus, Zeiss, Oberkochen, Germany), or ganglion cell complex (combination of RNFL, GCL, and IPL (RTVue, Optovue, Fremont, CA)) (7). This makes comparing GCA across machines nearly impossible. Studies of mean GCL/IPL thickness using the Cirrus (Zeiss, Oberkochen, Germany) have shown high levels of intratest reliability (8,9). We are not aware of any study that has provided a numericaldefined change in the mean GCL volume to assess test– retest reliability. Limitations of this study include the lack of correlation between volume variance and pathological state. Study eyes with significant macular pathology (geographic atrophy and macular fluid) were excluded because the main goal of the study was to discern the reproducibility of 314 follow-up scans which are amenable to reliable segmentation. In eyes with pathology, registration may have led to better reproducibility than imaging without registration. Different operators were used to obtain patient scans, which may have influenced the reg-OFF group because there was no computerized matching of scans. All patients followed the same scan protocol which may be subject to testing fatigue or errors associated with AutoRescan. However, we believe that this provides greater generalizability to real-world conditions where different technicians obtain results. Previous studies have outlined that the results of this study, as they pertain to SPECTRALIS OCT imaging, cannot directly be applied to other OCT machines, such as Cirrus, because there are significant differences between the OCT machines and algorithms (10). We did not collect demographic information. Although this may affect the mean GCL, it should not influence repeatability. Finally, the overall clinical status of the patient must still be considered as age-related loss of the GCL occurs and must be factored in determining clinically significant change over time. It has been reported that the GCL-IPL layer decreases by 0.12 mm microns per year on the Cirrus OCT (11). Other factors such as the presence of ocular pathology and quality of studies must also be taken into consideration while interpreting segmentation results. In summary, this study was performed to determine a threshold for SPECTRALIS OCT GCL volume variance, within the proper clinical context, to aid clinicians in determining significant change between 2 macular OCT scans. To date, no threshold for significant GCL change using the SPECTRALIS OCT exists. We estimate that repeat mean GCL volume measurement of 0.07 mm3 or more is statistically unlikely to represent test–retest variability. Given that 90% of eyes fell within 0.04 mm3 of test–retest reliability, clinicians may consider using this threshold to increase sensitivity of catching a true anatomical change but reduce specificity. STATEMENT OF AUTHORSHIP Category 1: a. Conception and design: S. Wagley, P. Mallory, and M.S. Lee; b. Acquisition of data: S. Wagley, P. Mallory, M.S. Lee, and P.A. Downie; c. Analysis and interpretation of data: S. Wagley, P. Mallory, M.S. Lee, P.A. Downie, B. Lindgren, J. Chen, and C.M. McClelland. Category 2: a. Drafting the manuscript: S. Wagley and P. Mallory. b. Revising it for intellectual content: M.S. Lee, P.A. Downie, B. Lindgren, J. Chen, and C.M. McClelland. Category 3: a. Final approval of the completed manuscript: S. Wagley, P. Mallory, M.S. Lee, P.A. Downie, B. Lindgren, J. Chen, and C.M. McClelland. REFERENCES 1. Huang D, Swanson EA, Lin CP, Schuman JS, Stinson WG, Chang W, Hee MR, Flotte T, Gregory K, Puliafito CA, Fujimoto JG. Optical coherence tomography. Science. 1991;254:1178– 1181. Mallory et al: J Neuro-Ophthalmol 2022; 42: 310-315 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution 2. Çetinkaya E, Duman R, Duman R, Sabaner MC. Repeatability and reproducibility of automatic segmentation of retinal layers in healthy subjects using Spectralis optical coherence tomography. Arq Bras Oftalmol. 2017;80:378–381. 3. Hashmani N, Hashmani S, Murad A, Mahmood Shah SM, Hashmani M. Assessing reproducibility and the effects of demographic variables on the normal macular layers using the Spectralis SD-OCT. Clin Ophthalmol. 2018;12:1433–1440. 4. Altman D, Bland M. Measurement in medicine: the analysis of method comparison studies. The Statistician. 1983;32:307–317. 5. Lim HB, Shin YI, Lee MW, Koo H, Lee WH, Kim JY. Ganglion cell - inner plexiform layer damage in diabetic patients: 3-year prospective, longitudinal, observational study. Sci Rep. 2020;10:1470. 6. Xu X, Guo X, Xiao H, Mi L, Chen X, Liu X. Reproducibility of macular ganglion cell-inner plexiform layer thickness in normal eyes determined by two different OCT scanning protocols. BMC Ophthalmol. 2017;17:37. 7. Mohammadzadeh V, Fatehi N, Yarmohammadi A, Lee JW, Sharifipour F, Daneshvar R, Caprioli J, Nouri-Mahdavi K. Mallory et al: J Neuro-Ophthalmol 2022; 42: 310-315 8. 9. 10. 11. Macular imaging with optical coherence tomography in glaucoma. Surv Ophthalmol. 2020;65:597–638. Lee H-J, Kim M-S, Jo Y-J, Kim J-Y. Ganglion cell–inner plexiform layer thickness in retinal diseases: repeatability study of spectral-domain optical coherence tomography. Am J Ophthalmol. 2015;160:283–289.e1. Mwanza JC, Oakley JD, Budenz DL, Chang RT, Knight OJ, Feuer WJ. Macular ganglion cell-inner plexiform layer: automated detection and thickness reproducibility with spectral domainoptical coherence tomography in glaucoma. Invest Ophthalmol Vis Sci. 2011;52:8323–8329. Hafner J, Prager S, Lammer J, Kriechbaum K, Scholda C, Pablik E, Schmidt-Erfurth U. Comparison of ganglion cell inner plexiform layer thickness by Cirrus and Spectralis optical coherence tomography in diabetic macular edema. Retina. 2018;38:820–827. Huo YJ, Guo Y, Li L, Wang HZ, Wang YX, Thomas R, Wang NL. Age-related changes in and determinants of macular ganglion cell-inner plexiform layer thickness in normal Chinese adults. Clin Exper Ophthalmol. 2017;46:400–406. 315 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. |
