Analysing the translatability of macular hole size measurements between high-density horizontal and radial OCT scan patterns

Introduction

The size of full-thickness macular holes (MHs), defined by their minimum linear diameter (MLD) on optical coherence tomography (OCT) scans, is the critical factor for treatment selection. Based on their MLD, MHs can be classified into small (<250 µm), medium (250–400 µm) and large (>400 µm). 1 While less invasive treatment options such as enzymatic or pneumatic vitreolysis are available for small to medium-sized MHs,2 3 pars plana vitrectomy (PPV) is the main surgical approach to treat MHs of all sizes since significant ocular complications have been reported with ocriplasmin and pneumatic vitreolysis.1 2 4 Based on the size of the MH, some surgeons decide to perform the PPV with or without internal limiting membrane (ILM) peeling and with or without an inverted ILM-flap.5–8 Therefore, accurate measurement of the MLD is essential for determining the appropriate management strategy in patients with aMH.

However, MLD measurements are not automatically provided by current OCT devices but need to be determined manually by clinicians using a calliper tool integrated in the OCT device software. Even though these measurements are obtained according to set guidelines,1 they are subjective and prone to intra-observer and inter-observer variations. Furthermore, the morphology of a MH is not uniform, exhibiting variability in the location of the largest diameter along the horizontal, vertical or oblique axes.9 10 As a result, horizontal OCT scans potentially underestimate the largest MLD in MHs with vertical or diagonal orientation, potentially resulting in misclassification of the MH size. For instance, MLD measurements using horizontal volume scans with an approximately 240 µm distance between scans underestimated the MH size by approximately 13% and resulted in a different MH classification in 22% when compared with radial scans with a 7.5° interval.11 Moussa et al found that high-density horizontal scans with a Triton inter-scan spacing of 50 µm yield lower variability and a reduced discrepancy in MLD measurements for smaller MHs compared with low-density scans with a Spectralis inter-scan spacing of 124 µm.12 Schneider et al demonstrated that the largest MLD detected with 24-line radial scanning was significantly greater than that obtained with 6-line radial patterns, with a clinically relevant rate of missed MHs associated with the lower radial density.13

High-density horizontal and radial scan patterns (narrow inter-scan spacing) provide a more accurate assessment of the true MLD of a MH and may potentially minimise the difference in MLD size measurements obtained in the two different scan patterns, which is important as the horizontal scan pattern is still the preferred standard for some clinicians and researchers, and some relevant publications on MHs are based on horizontal line scans,8 14 or the scan mode and the distance between the B-scan images have not been specified at all.15

The aim of this study was, therefore, to compare MLD measurements between high-density horizontal and high-density radial scan patterns and analyse their interoperability.

Materials and methods

This is a prospective case series of patients with a full-thickness MH who had two radial (radial scans 1 and 2), and two horizontal (horizontal scans 1 and 2) high-density high-resolution volume OCT scans of the macula (with the follow-up mode deactivated) obtained in the same examination session (Heidelberg Spectralis, Heidelberg Engineering, Heidelberg, Germany) at the Ophthalmology Department, TUM University Hospital, Technical University of Munich (TUM), Germany. The second scan in each mode was obtained to measure the inter-scan variability and device error. The horizontal volume scan consisted of 145 B-scans with approximately 30 µm inter-scan spacing and the radial volume scan of 48 cross-sectional B-scans with an angular 3.75° inter-scan spacing. Both scan patterns, including the number of sections and inter-scan spacing, can be routinely selected from the Spectralis interface, and the acquisition of a high-resolution high-density OCT scan takes less than 1 min in either mode. To obtain a steady fixation, patients were asked to fixate a target light presented to their contralateral eye while the OCT scans were taken in the affected eye. In both scan modes, the OCT raster was manually centred on the centre of the MH.

Inclusion criteria required that two sets of horizontal and radial volume scans could be performed concurrently. Exclusion criteria were obvious decentration of the MH on OCT, low-quality OCT images (quality index ≤30) and patients with coexisting macular pathology other than MHs.

Five moderately to highly experienced raters were instructed on how to use The International Vitreomacular Traction Study Group Classification System method to measure the size of the MH.1 In brief, using the built-in software calliper tool in the Heidelberg Eye Explorer viewing module (1:1 µm mode), the rater selected the scan displaying the largest aperture of the MH and measured the narrowest distance between the two edges of the MH (parallel to the RPE) as MLD size, under avoidance of the area of the operculum if present.

The raters measured with at least a 2-weekly interval 6 sets each of 60 OCT scans (horizontal volume scan 1, radial volume scan 1, horizontal volume scan 2, radial volume scan 2, repeat horizontal volume scan 1 and repeat radial volume scan 1). The raters were blinded to their previous measurements and to those of the other raters. For each radial OCT scan, the raters noted the scan number they used to measure the MLD as an indication of the localisation of the largest MLD. A MH was defined as having a horizontal orientation if the widest MLD was measured within 15° of the horizontal axis, to have a vertical orientation if the largest MLD was measured within 15° of the vertical axis and as oblique for the remaining MHs.

Results

60 eyes of 60 consecutive patients with a MH were included in this study. Mean MLD of all measurements (n=1800) was 432.94 (±185.99) µm (range 125–924 µm). Homogeneity of variance of the differences in MLD measurements across the included range of MH sizes was confirmed for rater 1, 2, 3, 4 and 5 in the horizontal (p=0.987, p=0.997, p=0.998, p=0.992 and p=0.971) and radial (p=0.972, p=0.945, p=0.949, p=0.998 and p=0.987) scan modes and between the five raters in the horizontal (p=0.994) and radial (p=0.0.988) modes. This suggests that the differences in the measurements are unrelated to their mean, that is, to the magnitude of the MH size, which is further supported by the absence of any obvious systemic bias in the ten BAPs for the MLD measurements of each rater in both scan modes (online supplemental figure 1).

The repeatability of measurements for rater 1, 2, 3, 4 and 5 was assessed using the coefficient of repeatability (CR=1.96×SD), which assumes that two repeat measurements of the same MH will be within 1.96×SD of the differences for 95% of MHs.16 The intra-rater CRs were for the horizontal mode 25 µm, 9 µm, 35 µm, 32 µm and 41 µm, and for the radial mode 31 µm, 33 µm, 58 µm, 69 µm and 49 µm, respectively. The inter-rater CR (calculated for the first set of measurements of each rater and taking rater 1 as nominal reference standard) was for the horizontal mode 16 µm, 42 µm, 36 µm and 51 µm, and for the radial mode 41 µm, 65 µm, 75 µm and 72 µm, respectively. The intra-rater ICCs were all ≥0.997 in the horizontal and ≥0.991 in the radial mode, and the inter-rater ICC was 0.998 in the horizontal and 0.994 in the radial mode.

There were no statistically significant differences between the primary and repeat measurements of all raters (n=300 each) in scan 1 in the horizontal (429.39 (±187.67) µm and 426.75 (±188.25) µm, respectively, p=0.10) and in the radial mode (438.21 (±186.26) µm and 437.72 (±185.68) µm, p=0.74), and between the measurements in scan 1 and scan 2 in the horizontal (429.39 (±187.67) µm and 427.59 (±185.71) µm, p=0.21) and in the radial mode (438.21 (±186.26) µm and 437.97 (±183.45) µm, p=0.90).

Furthermore, there was no correlation between the differences in the measurements and their mean and no obvious systemic bias in the corresponding BAPs for the repeat measurements of all raters (n=300 each) in the horizontal (r=−0.04, p=0.53; figure 1A) and in the radial scan mode (r=0.02, p=0.69; figure 1B), and for the primary measurements in scan 1 and the measurements in scan 2, either in the horizontal (r=0.08, p=0.17; figure 1C) or in the radial scan mode (r=0.08, p=0.15; figure 1D), suggesting that the differences in the measurements are unrelated to the magnitude of the MH size.

However, the variability between the differences in the repeat MLD measurements in the horizontal and radial scan 1 (figure 1A,B) and between the differences in the MLD measurements between scan 1 and scan 2 (figure 1C,D) was smaller in the horizontal than in the radial mode as reflected by their corresponding 95% LOAs (−28.29 to 33.59 vs −49.97 to 50.97 µm and −47.18 to 50.78 vs −66.40 to 66.90 µm, respectively).

Horizontal versus radial OCT scan mode

There was a statistically significant difference of −10.05 (±34.32) µm between the mean MLD of all measurements in the horizontal (n=900) and in the radial (n=900) mode (427.91 (±187.01) vs 437.97 (±184.93) µm; p<0.001). The variability of these differences around the mean MLD in both modes, measured as the corresponding 95% LOA, was −77.31 to 57.21 µm, which is large (figure 2), even though the ICC between the mean MLD values measured in the horizontal and radial mode was 0.991, which suggests excellent reliability.

Figure 2

Request permission

The x-axis illustrates the mean of all MLD measurements obtained from five raters for each of the 60 eyes with a MH (scan 1, repeat scan 1 and scan 2 in both the horizontal and radial mode) while the y-axis shows the difference between the mean of all horizontal and the mean of all radial measurements for each MH. The green, red and blue colours of the data points refer to the different orientations of the radial scan section containing the maximum MLD in which the raters’ measurements in the radial mode were conducted. Green data points represent the MHs in which the largest MLD was measured within 15° along the horizontal meridian, red data points represent the MHs in which the largest MLD was measured within 15° along the vertical meridian, and blue data points those with an oblique orientation. The maximum MLD was identified in 17/60 (28.3%) in a horizontal, in 7/60 (11.7%) in a vertical and in 36/60 (60.0%) in an oblique orientation. The difference between the mean of all MLD measurements in the horizontal and the mean of all MLD measurements in the radial mode was 0.77 (±13.88) µm for MHs with a horizontal, -10.39 (±34.62) µm for MHs with an oblique and −34.43 (±55.22) µm for MHs with a vertical orientation. MH, macular hole; MLD, minimum linear diameter; OCT, optical coherence tomography.

The difference between the mean of all MLD measurements in the horizontal and the mean of all MLD measurements in the radial scans was plotted against the mean of all MLD measurements in both the horizontal and radial mode for each MH (figure 2). The BAP does not show any obvious systemic pattern, but larger MHs (≥600 µm) are only represented by a few data points. Pearson’s correlation did not demonstrate an association between the difference in MLD measurements obtained in the horizontal and radial mode across the range of MLD sizes included in this study (r=0.05, p=0.16), formally supporting the graphical impression that there is no systemic pattern or bias in the BAP.

The mean difference between all horizontal and all radial MLD measurements in a MH was 0.77 µm (±13.88) µm for all MHs that had their widest MLD within 15° of the horizontal meridian in the radial scan, −10.39 (±34.62) µm for those with an oblique orientation, and −34.43 (±55.22) µm for all MHs that had their widest MLD within 15° of the vertical meridian in the radial scan.

A post hoc two-sided power analysis was performed, and with a power of 0.9 and an effect size (Cohen’s d) of 0.428 (80/187) a sample size of 60 would allow detecting a difference in MLD measurements of ≥80 µm between the horizontal and radial high-density scans.

Discussion

OCT has markedly improved the evaluation of MHs.11 17 18 However, routine MH size measurements are still performed manually and do not always provide consistent results,9 10 as MHs often exhibit irregular shapes with varying diameters across different meridians.10 11 Previous studies reported that the mean MLD obtained from horizontal scans was significantly smaller than that derived from radial scans,11 13 suggesting a potential tendency to underestimate the true MLD size when horizontal scanning techniques are used. However, the densities of the horizontal raster applied were only moderate (61 horizontal raster lines with an inter-scan spacing of 125 µm, along with 24 radial cross-sections)13 to low (19 horizontal raster lines with inter-scan spacing of 240 µm and 24 radial cross-sections).11

We demonstrate that MLD measurements obtained from high-density horizontal scans (ie, 145 B-scans with 30 µm inter-scan spacing), and from high-density radial scans (ie, 48 B-scans with an angular 3.75° inter-scan spacing) did not significantly differ within raters, between raters and between scans in the same mode. The intra-individual and inter-individual repeatability measured as ICC was high, and the intra- and inter-rater CRs were comparable to those in previous publications.11 18

We found a statistically significant difference of −10 µm between the mean MLD of the measurements obtained in the horizontal and in the radial mode. However, the variability of the differences around the mean MLD of both modes with the corresponding 95% LOA is much wider (range 135 µm), suggesting that this difference is not practically relevant. The large variability of these differences and the wide LOA are a combination of intra-rater and inter-rater as well as inter-scan or device error. The inter-scan error is particularly obvious for the radial scan mode as there is no automated control of pattern centration on the centre of the MH during scan acquisition (figure 3).

The mean difference between all horizontal and all radial MLD measurements in a MH was close to zero for MHs that had their widest MLD within 15° of the horizontal meridian, but negative for MHs that had their widest MLD within 15° of the vertical or oblique meridian, suggesting that horizontal and radial scans are only interchangeable for MHs with their widest MLD in the horizontal plane, but horizontal scans systematically underestimate the maximum MLD if located vertically or diagonally (figure 4).

However, the variability (intra-rater and inter-rater as well as inter-scan) between horizontal scans is less than between radial scans, due to the lack of perfect centration of the radial raster on the centre of the MH. Therefore, the two scan modes are not interchangeable but rather complement each other and should both be performed to get the best appreciation of the widest MLD. Ultimately, automated software-controlled centration of the high-density radial scan pattern on the centre of the MH would be required to make the additional acquisition of a high-density horizontal scan redundant.

In our study, using high-density scanning patterns, one MH was misclassified in the radial (≤250 µm) compared with the horizontal mode (>250 µm), and three MHs were misclassified in the horizontal (≤400 µm) compared with the radial mode (>400 µm). We could decrease the rate of misclassification to 6.7% (4/60) as compared with the rate of 22% previously reported when low-density scanning patterns were used,11 which may be important if the choice of treatment is based on the classification of small, medium and large MHs.1–3 5 8 14 15 This emphasises the need for standardisation of OCT scan patterns and inter-scan spacing to enhance the precision and comparability of MLD measurements in MHs.

Our study has several limitations: Our findings regarding MLD measurements between the two high-density scan modes may be restricted to the investigated range (125–924 µm); however, this range covers most MH sizes we see in clinical practice. Larger MHs (≥600 µm) were represented by fewer data points. Employing moderately to highly experienced raters for the MLD measurements, we have obtained a realistic reflection of real-world measurements. However, the results may differ for measurements obtained by less experienced raters.11 We have acquired two separate volume scans in both the horizontal and the radial mode to analyse the effect of device and pattern centration errors; however, these may vary between different OCT devices12 and technicians.

In summary, our findings indicate that both high-density horizontal and radial scanning modes contribute to the detection of the largest MLD in a MH and, therefore, to the accuracy of MLD size measurements. While a horizontal OCT scan offers advantages in terms of less intra-rater and inter-rater as well as inter-scan variability, relying solely on this scanning mode will lead to an underestimation of the true MLD size of a MH when its largest diameter aligns with the vertical or oblique axis. Ultimately, there is a need for the development of software applications that are capable of automated detection and tracking of pattern centration during scan acquisition, and that provide automated MLD measurements.