A Novel Technique of Adjusting Segmentation Boundary Layers to Achieve Comparability of Retinal Thickness and Volumes between Spectral Domain and Time Domain Optical Coherence Tomography

Abstract

Purpose.: The quantitative assessment of retinal thickness and volume varies according to the optical coherence tomography (OCT) machine used due to differences in segmentation lines. We describe a novel method of adjusting the segmentation lines of spectral-domain OCT (SD-OCT) to enable comparison with time-domain OCT (TD-OCT), and assess factors affecting its accuracy.

Methods.: In a prospective study, SD-OCT (Spectralis OCT) and TD-OCT (Stratus OCT) were sequentially performed on 200 eyes of 100 healthy individuals. Central retinal thickness (CRT), central point thickness (CPT), and 1-mm volume of the Early Treatment Diabetic Retinopathy Study grid were compared between the two machines. The segmentation lines on SD-OCT were manually adjusted by a trained operator and the parameters compared again with TD-OCT.

Results.: The mean CRTs of Spectralis and Stratus were significantly different (268.2 μm vs. 193.9 μm, P < 0.001). After adjustment of segmentation lines, the mean adjusted Spectralis CRT was 197.3 μm, with the difference between SD-OCT and TD-OCT measurements decreasing from 74.3 μm to 3.4 μm (P < 0.001). The difference between the adjusted Spectralis and Stratus CRTs was smallest for high myopes (≤ −6.0 diopters [D]) compared with those with moderate and low myopia (1.5 μm vs. 3.5 μm and 4.6 μm, respectively; P < 0.001). Similar trends were obtained for central 1-mm volumes and CPT. Interoperator and intraoperator repeatability for adjustment of the segmentation lines were good, with an intraclass correlation of 0.99 for both.

Conclusions.: Manual adjustment of SD-OCT segmentation lines reliably achieves retinal thickness and volume measurements that are comparable to that of TD-OCT. This is valuable to allow comparisons in multicenter clinical trials where different OCT machines may be used.

Introduction

Optical coherence tomography (OCT) is a safe, noninvasive imaging modality that is used for both quantitative and qualitative assessment of the retina and choroid. OCT has assumed an increasingly important role in the diagnosis and monitoring of patients in both clinical practice and, especially, multicenter clinical trials. In some clinical trials, one of the study inclusion criteria is the central retinal thickness (CRT) measured using OCT, and an increase in CRT was an indication for retreatment of patients with AMD. ¹ Given the essential role played by OCT measurements in clinical studies, it is imperative that OCT measurements be accurate, reproducible, and comparable when using different OCT machines. 

Time-domain OCT (TD-OCT) became available for clinical use in 2002. For several years, TD-OCT was the main OCT modality for clinicians and was considered the “gold standard.” Currently, the Stratus OCT (Carl Zeiss Meditec, Dublin, CA) is still commonly used in many clinics around the world. Spectral domain OCT (SD-OCT), with increased scanning speed and higher image resolution, allows visualization of retinal morphology in greater detail. ² In studies where participants were scanned sequentially with different OCT machines, the retinal thickness measurements varied depending on the type of OCT machine used. ^2–13 The largest difference in retinal parameters was observed when TD-OCT and SD-OCT machines were compared, with differences in CRT ranging from 43 to 83 μm. ^{2,4–6,9–12} As a result, some authors have concluded that the retinal thickness outputs of different OCT machines are not comparable and cannot be used interchangeably. ^{3,5,7,13–15}  

The variability in retinal thickness measurements using different OCT machines presents difficulties in at least two common scenarios—the first of which occurs when various study sites in a multicenter clinical study use different models of OCT machines. This is of great importance to image reading centers of multicenter clinical trials, since the OCT parameters from different machines are currently not comparable. A second situation occurs in a clinical setting when a specific patient has an OCT scan performed using one machine, and is then subsequently imaged using a different OCT machine, either because of a change in OCT equipment, or when the patient is referred to another center for continued management. 

Our objectives are to describe a novel method of segmentation line adjustment to derive CRT and retinal volume measurements on SD-OCT scans that are comparable with those of TD-OCT machines, and to assess the reliability and repeatability of this technique. 

Methods

In a prospective study performed at the Department of Ophthalmology, Tan Tock Seng Hospital, Singapore, 100 healthy volunteers with no history of ocular disease underwent sequential OCT scans using SD-OCT and TD-OCT machines. This study was approved by the Institutional Review Board of the National Healthcare Group, and conformed to the tenets of the Declaration of Helsinki. Written, informed consent was obtained from all participants. Participants were examined by a trained ophthalmologist (CT) to exclude ocular pathology. 

Ocular Imaging Protocol

A standardized imaging protocol was used to obtain the OCT scans. SD-OCT was performed on both eyes using the Spectralis OCT (Heidelberg Engineering, Heidelberg, Germany), followed immediately by TD-OCT using the Stratus OCT. All OCT scans were performed by the same experienced operator under standardized mesopic lighting conditions. The OCT scan was first performed on the right eye, followed by the left eye. For Spectralis OCT, a 25-line raster scan (20° × 20°, 7.6 mm × 7.6 mm) centered on the fovea was performed, with 15 frames averaged to improve the image quality. Stratus OCT was performed using the fast macular scan protocol, where six radial 6-mm scans spaced 30° apart were centered on the fovea. All OCT scans were reviewed by the operator during scanning to ensure that the scan was centered on the fovea. If the scans were displaced or if the image quality was inadequate, they were immediately repeated and reviewed until satisfactory. 

The CRT, central point thickness (CPT), and retinal volumes were calculated by the software of the respective OCT machines, after errors in segmentation were manually corrected (described later). 

Refractive error and keratometry were measuring using a full autorefractor-keratometer (Canon RK-F1; Canon, Inc., Tokyo, Japan). 

Adjustment of Segmentation Lines

The segmentation lines of all Spectralis OCT scans were adjusted using medical imaging management software (Heidelberg Eye Explorer 1.7.0.0; Heidelberg Engineering, Heidelberg, Germany) by a trained operator (KL), who was masked to the retinal thicknesses and volumes of the corresponding Stratus OCT scans. First, each of the 5 horizontal B scans passing through the central 1-mm circle of the Early Treatment of Diabetic Retinopathy Study (ETDRS) grid was individually examined to adjust for errors in the upper segmentation line, which corresponds to the inner limiting membrane. The ETDRS grid was also manually centered on the fovea if there was any displacement. The lower segmentation line, which was automatically drawn by the software at the level of the RPE, was then manually adjusted to the upper border of the first hyper-reflective band, which corresponds to the inner segment–outer segment (IS-OS) junction (Fig. 1A). The adjustments were performed under high magnification (400%–800%) to ensure the best possible visualization of the retinal layers. The adjusted CRT, CPT, and 1-mm volumes were then automatically calculated by the software (Heidelberg Engineering). 

Figure 1. 

 

Segmentation boundary layers of SD-OCT and TD-OCT scans. (A) Segmentation of a Spectralis OCT scan, showing boundary layers drawn by the software at the inner limiting membrane and lower border of the retinal pigment epithelium (red lines). The position of the lower boundary layer was manually adjusted to the junction of the inner and outer segments (green line). (B) Automated segmentation of the Stratus OCT scan, showing boundary layers at the inner limiting membrane (ILM) and junction of the inner and outer photoreceptor segments (ISOS) (white lines).

To assess interoperator variability, the adjustments were repeated independently by several trained operators of differing levels of seniority and clinical experience (including attending ophthalmologists, residents, and medical students). In addition, to assess intraoperator variability, the main operator (KL) readjusted the segmentation of a series of participants 3 months after the original adjustments. When performing reliability assessments, the original (unadjusted) OCT scans were loaded into the software (Heidelberg Engineering) and all operators were masked to the actual position of the previous adjusted segmentation lines or the adjusted retinal thicknesses and volumes. 

For the Stratus OCT scans, all six radial B scans were reviewed using the Stratus OCT Review Software (version 6.02 [0562]) and any obvious errors in the position of both upper and lower segmentation lines were corrected. No other adjustments were made to the segmentation of Stratus OCT scans (Fig. 1B). 

Statistical Analysis

Statistical analysis was performed using statistical and data analysis software (SPSS 16.0; SPSS, Inc., Chicago, IL). The CRT, CPT, and 1-mm volumes of the Spectralis and Stratus OCT were compared using paired t-tests. Intraclass correlation (ICC) was used to assess the agreement between the CRT of the Stratus OCT and the adjusted equivalent CRT from the Spectralis OCT. 

Results

The mean age of the 100 participants was 25.9 years (range 21–60, SD ± 5.8), with 55 males (55%) and 45 females (45%). 

Before adjustment, the mean CRT was 268.2 μm (range 232–319, SD ± 16.2) using Spectralis OCT and 193.9 μm (range 152–240, SD ± 16.5) using Stratus OCT, with a mean difference of 74.3 μm (P < 0.001). Although the CRT measured using Spectralis and Stratus OCT had good correlation (Pearson correlation 0.933, P < 0.001; Fig. 2), the ICC for absolute agreement between Spectralis and Stratus CRT was 0.082. 

Figure 2. 

 

Scatterplot of the central retinal thicknesses of the unadjusted and adjusted Spectralis OCT plotted against corresponding Stratus OCT scans, demonstrating good correlation of the retinal thickness measurements. Blue – unadjusted Spectralis; red – adjusted Spectralis.

After Adjustment of Segmentation Lines

After adjustment of the segmentation lines on the Spectralis OCT scans, the mean adjusted Spectralis CRT was 197.3 μm, with the difference between the Spectralis OCT and Stratus OCT reduced to 3.4 μm (Fig. 2). The reduction in the difference between Spectralis OCT and Stratus OCT CRT after adjustment of segmentation lines was statistically significant (74.3 vs. 3.4 μm, P < 0.001). The ICC between the adjusted Spectralis OCT and Stratus OCT CRTs improved to 0.924. The Bland-Altman plots of the Spectralis OCT and Stratus OCT CRT before and after adjustment of the segmentation lines are shown in Figures 3A and 3B, respectively. Of note, the spread of the points was fairly close both before and after adjustment, demonstrating good correlation of Spectralis and Stratus CRT values. Before adjustment, however, the points are clustered around the mean of 74.3 μm, whereas after adjustment, the points are clustered around 3.4 μm, and include the line passing through zero. 

Figure 3. 

 

Bland-Altman plots of unadjusted and adjusted OCT thicknesses. (A) Bland-Altman plot of the central retinal thickness using unadjusted Spectralis OCT and Stratus OCT. (B) Bland Altman plot of the central retinal thickness using adjusted Spectralis OCT and Stratus OCT.

The difference between the adjusted Spectralis and Stratus CRT values varied with refractive error (P < 0.001), with the smallest difference (mean 1.5 μm) occurring in high myopes (spherical equivalent ≤ −6.0 diopters [D]). This difference increased progressively to 3.5 μm for moderate myopes (> −6.0 D to −3.0 D), 4.6 μm for low myopes (> −3.0 D to −0.5 D), and 6.8 μm for emmetropes or hyperopes (> −0.5 D). Performing linear logistic regression, spherical equivalent was a significant factor affecting the difference between the adjusted Spectralis and Stratus CRT values (coefficient 0.50, P < 0.001) 

Similar results were observed when assessing CPT and the retinal volume in the central 1-mm circle (Table), with a significant difference between the values derived from Spectralis and Stratus OCT. After adjustment of the segmentation lines, the adjusted Spectralis OCT CPT and retinal volumes were comparable with the corresponding Stratus OCT values. 

Table. 

 

Comparison of the Retinal Parameters of the Spectralis and Stratus OCTs before and after Adjustment of Spectralis OCT Segmentation Lines

	Spectralis OCT		Stratus OCT	Difference in Retinal Parameters
	Unadjusted	Adjusted		Unadjusted Spectralis & Stratus	Adjusted Spectralis & Stratus
CRT (μm)	268.2	197.3	193.9	74.3	3.4
CPT (μm)	223.6	143.9	154.9	68.4	−10.9
1-mm retinal volume (mm3)	0.21	0.16	0.15	0.06	0.01

The segmentation adjustment technique was reliable and repeatable. Between operators, the ICC was 0.99 and did not vary with the level of clinical experience of the operator. When the adjustment was repeated by the main operator 3 months later, a high level of intraoperator repeatability was observed, with a mean difference of 1.1 μm and an ICC of 0.99. 

Discussion

In this study, we have described a novel and reliable technique of adjusting the boundary layers of Spectralis OCT scans to achieve retinal thickness and volume measurements that are comparable with those obtained from corresponding Stratus OCT scans. The mean difference between the CRT of the adjusted Spectralis OCT and the Stratus OCT was 3.4 μm, which is clinically insignificant, and also well within the limit of resolution of SD-OCT machines (5–7 μm). 

Earlier studies that compared the retinal parameters measured obtained from various OCT machines reported significant differences in CRT measurements, ^2–13 even when these were compared between SD-OCT machines. ^3,6,12,13 However, the greatest difference occurred when the CRTs were compared between an SD-OCT and the Stratus OCT. The difference of 74.3 μm between the unadjusted Spectralis and Stratus CRTs in our study is consistent with previous reports comparing the two OCT machines. ^{4,7,9,10,13,16} The segmentation software of the Stratus OCT draws the lower retinal boundary at the level of the first hyperreflective line of the OCT scan, ¹⁷ corresponding to the junction of the inner and outer segments of the photoreceptors (Fig. 1B). In contrast, the software of the Spectralis OCT draws the lower boundary line at the junction of the RPE and Bruch's membrane. ¹⁰ It is believed that the difference in position of the boundary lines accounts for the differences in reported retinal thicknesses and volumes. ^7,10  

Some authors have stated that the CRTs between OCT machines are not comparable and cannot be used interchangeably. ^{3,5,7,13–15} This presents difficulties for reading centers of multicenter clinical trials, where various study sites may use different models of both SD-OCT and TD-OCT machines. Without adjustment, the retinal parameters reported by the reading center would vary depending on the model of OCT machine used, leading to inconsistencies in the treatment decisions and interpretation of study outcomes. ⁷ For example, the CRT of a patient is likely to be thinner on Stratus OCT, and hence this patient may not meet the inclusion or retreatment criteria for that study. However, if the same patient were assessed using an SD-OCT machine, the thicker CRT reported by that machine may fall within the inclusion or retreatment criteria, and the patient would have been managed differently. 

We are aware of one earlier paper ¹⁸ that compared the retinal thicknesses of SD-OCT machines by manually drawing boundary layers. The authors reported that the retinal thickness reported by different SD-OCT machines are comparable when a uniform position is used to locate the outer retinal boundary. ¹⁸ There are, however, several differences between that study and the current one. First, the comparisons in that study were made between SD-OCT machines, whereas we have reported a technique to achieve comparability of retinal thickness between a SD and TD-OCT machines. In addition, the earlier study adjusted the boundary layers using custom grading software. In contrast, in our study, we have achieved adjustment of the segmentation lines using the medical imaging management software (Heidelberg Engineering), which is available with the Spectralis OCT. Finally, the earlier study did not assess whether the retinal volumes were comparable. The results of the current study demonstrate that this technique can be used to simultaneously generate measurements of retinal volumes. 

In this study, we chose to adjust the segmentation lines of the Spectralis OCT to a retinal layer comparable with the Stratus OCT. At present, the Stratus OCT is still commonly used by many centers around the world. Therefore, it is reasonable to adjust the outputs of SD-OCT to compare with the Stratus OCT at this time. In the future, as SD-OCT machines become more common, it would be logical to standardize the boundary layers that define retinal thickness to include the photoreceptor outer segments. Manufacturers of OCT machines may also wish to design software that can automatically draw the segmentation lines at different layers of the retina to allow comparability between machines. 

The strengths of our study include the large number of participants (200 eyes of 100 participants). Previous studies were performed on groups of between 14 to 60 participants. With a smaller number of participants, it is possible that abnormalities inherent to some participants may have a greater effect on the results. We examined both interobserver and intraobserver variability and demonstrated that the adjustment technique has a high degree of repeatability, with ICCs of 0.99. In addition, this study included a group of high myopes (spherical equivalent ≤ − 6.0 D), and demonstrated the same degree of accuracy of the adjustment technique. Indeed, we found that the difference between the adjusted Spectralis and Stratus CRTs was smallest among the high myopes (1.5 μm). Although some studies have reported that the retinal thickness does not undergo diurnal variation, ^19–21 the OCT scans using SD-OCT and TD-OCT were performed consecutively on each participant to ensure that any variation in retinal thickness, if it did occur, would not affect the analysis. 

This study is not without limitations. We recruited only normal participants without eye pathology for this study. The study team felt that it was important to demonstrate that this technique is reliable and repeatable in normal eyes before applying it to eyes with ocular pathology, where the distortion of the retinal architecture by the disease process could make identification of retinal layers more challenging. In future studies, the accuracy of this technique when applied to individuals with different types of ocular pathology should be examined. 

Although radial scans, which are used by the fast macular scan protocol of the Stratus OCT, are also available on the Spectralis OCT, we chose to perform raster scans using the Spectralis OCT instead. Firstly, the raster scan gives greater detail due to the higher scan density and even spacing of the points scanned and is a common scan protocol for macular OCTs using SD-OCT machines. In addition, the medical imaging management software (Heidelberg Engineering) does not allow automated computation of the CRT or retinal volumes when radial scans are performed. 

In conclusion, we have described a novel technique to adjust the boundary layers on Spectralis OCT that generates retinal thicknesses and volumes that are comparable with scans performed on Stratus OCT. This technique has a high degree of accuracy and repeatability. It allows scans taken on Spectralis and Stratus OCT to be compared, either longitudinally or among different cohorts, thus enabling comparison of patient data in reading centers for multicenter clinical trials or in individual clinical practices. 

Acknowledgments

The authors would like to thank the participants who volunteered for the OCT scans. 

Footnotes

 Disclosure: C.S.H. Tan, None; K.Z. Li, None; T.H. Lim, Heidelberg Engineering (R)