Intersession Repeatability of Macular Thickness Measurements with the Humphrey 2000 OCT

Abstract

purpose. This study was designed to determine intersession repeatability of measurements of macular thickness made with a commercially available optical coherence tomography (OCT) system. The images that can be routinely acquired with the commercial instrument differ significantly in quality from the images in the literature, which have mostly been acquired on prototype systems.

methods. Multiple OCT images of the nasal macula were obtained from the right eye during three independent measuring sessions, using the Humphrey 2000 OCT system (Humphrey, San Leandro, CA). Twenty-six volunteers with no history of ocular disease participated in this investigation. Eyes in all subjects were undilated during scan acquisition. Scans were horizontal, 3 mm long, and through the fovea. Five scans were used from each session, for a total of 15 scans per subject. Retinal boundaries were automatically detected using custom software. Statistical software was used to calculate intersession and intrasession repeatability. Manual correction was performed on the automatically detected boundaries, and a second analysis was performed using these boundaries.

results. When no manual correction of boundaries was performed, there were no significant effects between different sessions (P = 0.529) or between different scans within the same session (P = 0.509). Average retinal thickness was found to be 274 ± 17 μm for a 1-mm long region 0.75 mm from the fovea. Individual scan averages differed from overall patient averages by 0 ± 4.3 μm (99% confidence interval, 11.2 μm).

conclusions. OCT measurements of macular thickness made with the Humphrey 2000 OCT system are repeatable over different sessions with an expected variation of less than 11 μm (99% confidence interval).

Optical coherence tomography (OCT) is a new ophthalmic technology that has been used to image a multitude of retinal diseases. Macular lesions associated with optic nerve head pits, ¹ epiretinal membranes, ² central serous chorioretinopathy, ³ age-related macular degeneration, ⁴ choroidal neovascularization, ⁴ and diabetic macular edema ⁵ are just some of the diseases that have been studied using OCT. For many of these studies, OCT’s fine, 10-μm resolution in vivo ⁶ has given insight into disease mechanisms. 

OCT’s ability to provide quantitative measurements makes it complementary to traditional means of examination such as opthalmoscopy and slit lamp biomicroscopy. In particular, OCT has been used to measure macular thickness, and the variability of measurements taken during a single session has been reported. ^{5 6 7} To date, no studies have demonstrated the intersession repeatability of retinal thickness measurements obtained with OCT. For OCT measurements to be used to monitor patient progress over time, however, the variability of OCT measurements made during different sessions must be known. Determining the appropriate treatment for diabetic macular edema, for example, requires tracking retinal thickness changes in a patient through subsequent visits. ⁸ Therefore, one purpose of the present study was to determine the intersession repeatability of OCT measurements of macular thickness. 

The prototype OCT systems, which, to date, have dominated the literature, demonstrate significant differences with the commercially available system used in this study. Differences exist in the optics, internal amplification of the interference signal, output interface, and superluminescent diode (SLD) power range, as confirmed by Humphrey Instruments (San Leandro, CA). Although it is not clear exactly how these differences might affect the OCT images acquired, there is a substantial difference in quality between the OCT images reported ⁹ and those that can be obtained easily with the commercially available Humphrey 2000 OCT. There is therefore a need to demonstrate that the variability reported in the literature using the prototype ^{5 7} is applicable to commercially available systems. The ultimate purpose of this study was to assess both intrasession and intersession repeatability in patients using the commercially available Humphrey 2000 OCT system. 

Materials and Methods

Equipment

OCT is a new imaging technology with ophthalmologic applications based on the principles of laser interferometry. ^{10 11} The output from a low-power infrared SLD is split so that one half travels down a reference path and the other half travels into the eye and reflects back. ¹⁰ The reference and reflected beams are recombined, and heterodyning allows the optical interference of the two beams to be detected. ¹² The use of a low-coherence-length laser allows only the reflections from a narrow region of the retina to interfere with the reference beam, ¹² giving the theoretical resolution of less than 10 μm that has been reported. ⁶ This region is scanned through the retina at a single point to generate an A scan, and this scanning point is moved over the retina to generate a B scan. ¹¹ The B-scan images generated by OCT represent infrared reflectivity within the tissue and are thus analogous to ultrasound B scans, which depict ultrasound reflectivity. ¹¹  

The Humphrey 2000 OCT system used in this study is a commercial implementation of the prototype systems that have been reported. ^{1 2 3 4 5 6 7 11 12} It uses an 850-nm SLD source with a patient exposure range of 200 to 750 μW. Other studies have reported machines using SLD wavelengths of 830 to 850 nm, with incident power ranges of 200 to 1000 μW. ^{1 2 3 4 5 6 7 10 11 12}  

Subjects

Twenty-six volunteers (13 men and 13 women), ranging in age from 20 to 52 years (mean ± SD, 31 ± 9 years) participated in this study. The volunteers were selected from research associates and clinic staff. The study was conducted according to the tenets of the Declaration of Helsinki and participants provided informed consent after the intent of the study had been explained. The exclusion criterion was history of known retinal disease. Informed consent was obtained, and the protocol was approved by The Ohio State University Biomedical Sciences Institutional Review Board. 

Pupil Measurements

All eyes in this study were imaged through undilated pupils. Pupil size can vary considerably, however, between normal persons. Thus, to assess the applicability of this study’s results to other subject populations, each pupil was measured in both dark- and light-adapted states. A pupil gauge was held on the forehead while the subject fixated on a distant object. Pupil measurements were made in a darkened room using a transilluminator held obliquely from below to visualize each pupil before it contracted. The transilluminator was maintained shining into the pupil for the constricted pupil measurements. This method was selected because it is commonly used in routine practice. Estimates of pupil diameter were made to the nearest 0.5 mm. In all subjects but one, pupil measurements were performed by the same investigator (SEK). 

Refractive Error

Refractive error was assessed by using either an automatic refractometer (model 585; Allergan Humphrey) or by assessing the prescription from the subject’s glasses by lensometer (model 12603; American Optical, Buffalo, NY), if they had been obtained within the past 12 months. 

Scanning

All scanning was performed by the same investigator (DK). For each scanning session, the undilated right eye was aligned with the OCT machine. The subject was then asked to gaze at an internal fixation light within the machine. The machine was programmed to assume a normal, emmetropic eye and to scan along a horizontal line beginning at the temporal edge of the foveal pit and extending nasally 3 mm. Figure 1 illustrates the position of the scan on the retina of a study subject. In the figure, the white haze above and below the machine’s scan line is an artifact resulting from glare. The system was set to image along a line overlapping the fixation point, and scanning was begun. The scanning line position was then adjusted so that the image of the foveal pit appeared as deep as possible on the temporal side of the scan. It was assumed that the deepest part of the foveal pit was its center. A sample scan is shown in Figure 2 , in which the typical placement of the scan is seen relative to the foveal pit. 

The procedure used to acquire scans was consistent with the standard clinical use of OCT. The OCT system displays scans continuously at the rate of approximately one per second and updates the screen image accordingly. Acceptable scans were selected as quickly as they appeared. Acceptance criteria were based solely on signal strength and absence of artifacts due to motion and pupillary shadowing. These artifacts are obvious and are caused by subject motion; even small head or eye motions can cause jumps in the scans, as well as great reductions in signal strength. Nonetheless, subjects could typically be seated, positioned, and their eyes scanned within 5 minutes for each session. Each subject participated in three separate scan sessions, and all sessions were conducted within 1 day. To make each session independent, a minimum separation of 5 minutes was mandated, and the machine was realigned between sessions. 

Analysis

The acquired scans were exported to an SGI computer workstation (Mountain View, CA) for subsequent analysis using custom automatic boundary detection software written for the MATLAB software platform (The Mathworks, Natick, MA). The software automatically detects the vitreoretinal junction as the inner retinal boundary and the retinal–choroidal junction as the outer retinal boundary. The retinal thickness was then calculated as the distance between the two boundaries along each A scan. For the purpose of comparison, the automatic boundary location was manually verified for all scans and corrected when necessary. All inspection and correction was done by the same unmasked investigator (DK). Separate analyses were performed with both the corrected and the uncorrected boundaries. Although the Humphrey 2000 comes with its own automatic boundary detection software, its results cannot be exported for analysis. 

A sample measured scan is presented in Figure 2 with the marked retinal boundaries highlighted. Figure 3 shows the calculated retinal thickness contours for all five scans within a single session (subject 25, session 2). That is, Figure 3 shows how the thickness varies across the acquired scan. The fovea is represented by the point of minimum thickness toward the left side of the plot. More variability was found near the fovea, as expected, because the actual retinal thickness changes in this region. Thus, small changes in scan placement could cause very different scan profiles in the foveal region. 

The retinal thickness value for a scan was calculated for a 1 mm long section located 0.75 mm from the fovea. The mean and SD for each subject were calculated over each session (5 scans total per session) and over all sessions (15 scans total). Finally, the mean thickness value was calculated over all subjects using the mean of 15 scans in each subject. 

To examine the effects of the within-subject factors scan number (intrasession repeatability) and session number (intersession repeatability), a repeated-measures analysis was performed using a statistical software package (SAS, Cary, NC). For each subject there were 15 observations (three sessions with five scans per session). Separate analyses were performed on the thickness measurements derived from the corrected and uncorrected boundaries. The effect of boundary correction was measured by finding the difference between thickness measurements from corrected and uncorrected boundaries for each scan. The mean and SD of these differences was found. 

To quantify the typical variability occurring between sessions, average thickness values were found for each subject over each individual session and over all three sessions (the latter is assumed to be an estimate of the true value). Then, for each subject, the individual session averages were subtracted from that subject’s average over all sessions. These differences indicated how much session averages varied around the true value (the assumption was made that each subject’s true retinal thickness did not change between sessions). The differences in means were then averaged over all patients and all sessions as an indication of normal intersession variability in mean retinal thickness measurements. A similar analysis was performed comparing the retinal thickness measurements derived from individual scans to each subject’s average over all scans. This result indicated the normal variation between individual scans. 

Two additional statistics were computed for comparison with the literature. For each subject, the SD of thickness values was calculated for each of the three sessions. The intrasession SD was then averaged over all subjects for comparison with the value of Hee et al. ⁵ Each SD was also divided by its corresponding session mean to compute the intrasession coefficient of variation (CV). The CVs were then averaged over all subjects and all sessions for comparison with the results of Baumann et al. ⁷  

Two separate linear regressions were performed to study the effects of pupil size and refractive error on intrasubject variability. The variability for a subject was measured as the SD for that subject’s 15 scans. This SD was then regressed against mean pupil size and mean spherical correction. 

Results

For the uncorrected boundaries, the result of the repeated measures analysis testing showed no significant effects for either session number (intersession repeatability) or scan number (intrasession repeatability); P = 0.529 and 0.509, respectively. Correcting the retinal boundaries similarly yielded P = 0.567 for session number effects and P = 0.573 for scan number effects. The thickness measurements from corrected and uncorrected boundaries differed by a mean of 1.1 ± 3.3 μm (±SD), which was not significantly different from zero (P = 0.739). 

The mean and SD for retinal thickness measurements for all patients and all scans was 273 ± 17 μm for uncorrected boundaries and 274 ± 17 μm for corrected boundaries. The means (±SD) of retinal thickness measurements for all patients, grouped by session, are presented in Table 1 . Each of the 26 subjects had five scans per session, and the means are therefore determined from 130 values. The means of retinal thickness measurements for all patients, grouped by scan number, are presented in Table 2 . The 26 subjects participated in three sessions each, providing means from 78 values. 

The mean differences between the individual session averages and the average over all sessions, calculated for each subject, are presented in Table 3 . The mean of differences between individual scan averages and each subject’s average are presented in Table 3 as well. The mean intrasession coefficient of variation was 1.2% ± 0.7% for corrected boundaries and 1.1% ± 0.8% for uncorrected boundaries. The mean subject intrasession SD is 3.2 ± 2.1 μm for corrected boundaries and 3.0 ± 2.2 μm for uncorrected boundaries. 

Pupils ranged from 4.0 to 7.0 mm in the dilated state (mean, 5.9 ± 0.8 mm). Constricted pupils ranged from 1.5 to 3.5 mm (mean, 2.6 ± 0.5 mm). The range of refractive errors for the subjects (mean spherical correction) was +1.00 to −10.75 D (mean, −2.50 ± 2.75 D). No significant relationship was found by regressing either mean pupil size or mean spherical correction against the SD of each subject’s 15 scans. For the pupil size, the slope of the regression with corrected boundaries was 0.16 (P = 0.76), and for mean spherical correction, the slope of the regression was −0.03 (P = 0.83). 

Discussion

In this study, macular thickness measurements obtained in undilated eyes by the Humphrey 2000 OCT system were shown to be repeatable within a session and over different independent sessions. The measurements made in this study were performed on scans obtained through the fovea; repeatability for measurements made elsewhere would depend on how easily the scan could be consistently placed at the same location. Intersession repeatability suggests that OCT can be used to observe patients over time to measure the progression of disease. Furthermore, the commercially available Humphrey 2000 OCT system, while providing images qualitatively different from those in the literature, gives a satisfactory quantitative performance with the software used for this study. 

For this study, retinal thickness was calculated using both corrected and uncorrected retinal boundaries output by the software. The necessity of correcting the results of the automatic boundary detection algorithm may be questioned, because this increases the amount of time necessary to obtain retinal thickness information. The corrected averages differed from the original output of the algorithm by 1.1 ± 3.3 μm. This is not significantly different from zero (P = 0.739) and is less than 9.7 μm (99% CI). An advisable strategy may therefore be to view the algorithm’s output to verify that there are no gross errors but otherwise to accept its results. Software allowing correction of the boundaries could be a useful addition to the OCT system. Because the difference was not clinically significant, only the corrected results are discussed in this article. 

The average retinal thickness was 274 ± 17 μm over all 26 subjects, when calculated with corrected boundaries. Our subject population was generally young and with the exception of four subjects, the refractive errors ranged from +1.00 D to −5.00 D. Our average thicknesses are also consistent with the values presented by Hee et al. ⁵ They measured the retinal thickness within a 3-mm disc radius surrounding the fovea by interpolating the results of six foveal scans placed 60° apart. ⁵ All scans were acquired in one session. The average foveal thickness for all subjects was reported to be 174 ± 18 μm. Along a horizontal line segment extending nasally from the fovea, they reported the average in 73 subjects to be 260 ± 16 μm for the segment extending 0.5 to 1.5 mm from the fovea, and 255 ± 16 μm for the segment extending 1.5 to 3.0 mm from the fovea. ⁵  

The technique of Hee et al. provided six foveal thickness measurements for each subject, all within the same session. ⁵ The intrasession SD within the six measurements of each subject was reported to be distributed with a mean of 11 ± 6. In this study, each session yielded five thickness measurements; a similar calculation by Hee et al., ⁵ for the purposes of direct comparison, yields an average subject intrasession SD of 3.2 ± 2.1 μm with the corrected boundaries. 

For the present study, the data in Table 3 show that the differences between session averages and total patient averages are distributed with an SD of 2.7 μm. Similarly, the differences between individual scans and subject averages are distributed with an SD of 4.2 μm. These SDs best quantify the variability within this study and are of the same order of magnitude as those in Hee et al. ⁵ The larger size of Hee’s variability assessment may result from their measurement of thickness at a single point rather than within a region. 

The data from this study may be better interpreted by using confidence intervals. There is a 99% confidence that session averages will be within 7.0 μm of the true subject value using corrected boundaries. Similarly, there is a 99% confidence interval that individual scan averages will be within 11.2 μm, using corrected boundaries. Only a small decrease in the variability is achieved by averaging five scans per session. Therefore, it may be reasonable for a clinician to accept retinal measurements from only one or two scans. 

The intrasession reproducibility results found here can also be compared with those of Baumann et al. ⁷ They obtained six vertical scans of length 2.88 mm through the fovea of 18 eyes. ⁷ The scans were divided into seven sections, and the sections were treated individually. The CV (SD divided by the mean) for each scan segment was averaged over all subjects, and these averages were presented for both manual and automated retinal boundary determinations. The segments of their scans most relevant to this study were the most superior and inferior segments (1.12–1.44 mm from the fovea). For these they found mean CVs of 4.1% and 3.8%, respectively. The average CV of 1.2% achieved in the present study is comparable. 

The ability to perform OCT measurements in undilated eyes extends the potential uses of this instrument. Although Baumann et al ⁷ obtained their single-session OCT measurements in patients with undilated eyes, the pupil sizes were not reported. It is not uncommon for circumstances to make dilation either difficult or undesirable. Adequate dilation may be difficult in patients having extended topical pilocarpine therapy, exfoliation syndrome, ¹³ pseudoexfoliation syndrome, chronic diabetes, and, occasionally, old age. Subject variability does not seem to be related to pupil size, although the smallest pupil diameter for which macular thickness measurements can be made routinely with OCT was not determined in this study. The use of OCT in undilated eyes presents further limitations. In particular, the fundus is frequently difficult to visualize during scanning. Thus the patient’s cooperation in fixation or the presence of clearly identifiable OCT landmarks, such as the foveal pit, is necessary for scans to be located precisely on the fundus. Scans around the optic nerve head would most likely be compromised in undilated eyes for these reasons. 

The OCT measurements made in this study were made without knowledge of the subject’s axial length or refraction. Because of the design of the OCT system’s optics, these parameters are necessary for distances in the transverse direction (i.e., along the direction of the scan line) to be measured accurately. ⁵ A standard, emmetropic eye (plano, 23 mm axial length) was assumed for all subjects and probably caused some intersubject variations from the assumed scan length. However, these errors would remain constant for each subject from session to session, and therefore would not be expected to affect repeatability. Moreover, measurements in the axial direction (i.e., into the retina) do not depend on the refraction or axial length of the eye. Thus, thickness values are not affected. To date, no study has been published in which axial length and refraction were used to ensure accurate scan length measurements. 

In conclusion, OCT offers a way to make measurements of retinal thickness that are repeatable over different sessions. In particular, the repeatability obtained with the Humphrey 2000 OCT system is comparable to that of the prototype systems in the literature. The automatic boundary-marking system used here can be expected to provide repeatable measurements close to those marked manually. Moreover, this intersession repeatability has been demonstrated in undilated eyes, extending the noninvasiveness or OCT. 

 Supported by The Ohio Lions Eye Research Foundation, Columbus, Ohio; Sunbury Lions Club, Sunbury, Ohio; Humphrey Systems, Dublin, California; and The Whitaker Foundation, Rosslyn, Virginia.

 Submitted for publication May 7, 1999; revised October 26, 1999; accepted November 8, 1999.

 Commercial relationships policy: Cc5(CR).

 

Figure 1.

 

Fundus view showing OCT scan position as shown by the white line. The white haze above and below the line is artifact.

Figure 2.

 

Sample OCT scan showing measurement points and marked boundaries.

Figure 3.

 

Thickness measurements obtained in one session (five scans).

Table 1.

 

Retinal Thickness Measurements Grouped by Session

Session	Number of Subjects	With Uncorrected Boundaries	With Corrected Boundaries
1	26	272.8 ± 17.0	273.7 ± 16.9
2	26	272.8 ± 17.0	274.2 ± 17.6
3	26	272.1 ± 16.7	273.3 ± 17.1
Repeated measures test for intersession effects			P = 0.529	P = 0.567

 Data are mean micrometer ± SD obtained from five scans in each session.

Table 2.

 

Measurements Grouped by Scan

Scan	Number of Subjects	With Uncorrected Boundaries	With Corrected Boundaries
1	26	272.9 ± 16.9	274.3 ± 17.0
2	26	272.4 ± 16.8	273.7 ± 17.3
3	26	272.6 ± 17.2	273.3 ± 17.2
4	26	272.7 ± 17.1	273.5 ± 17.3
5	26	272.2 ± 16.6	273.8 ± 17.4
Repeated measures test for intrasession effects			P = 0.509	P = 0.573

 Data are mean micrometers ± SD obtained from three sessions per scan.

Table 3.

 

Intersubject Differences in Thickness Averages and Thickness Measurements from Individual Scans and Sessions

	Uncorrected Boundaries	Corrected Boundaries
Session differences averaged over all subjects and all sessions	0 ± 2.5	0 ± 2.7
Individual scan differences averaged over all subjects and all sessions	0 ± 4.2	0 ± 4.3

 Data are mean micrometers ± SD.