Astigmatism Influences Quantitative and Qualitative Analysis in Optical Coherence Tomography Angiography Imaging

Abstract

Purpose: The purpose of this paper was to study the influence of astigmatism in optical coherence tomography angiography (OCTA) images in a quantitative and qualitative analysis.

Methods: This was a prospective, cross-sectional study. We included 110 eyes of 110 patients: 20 eyes without astigmatism and 90 eyes with astigmatism ≥0.5 diopters (D). We performed a macula centered OCTA as a reference image. In patients without astigmatism, registered follow-up scans were performed after induction of −1 and −2 D astigmatism. In patients with astigmatism, we performed the follow-up scan after astigmatism correction. We used a set of cylindrical lenses attached to the camera head of the SPECTRALIS (Heidelberg Engineering, Heidelberg, Germany). A quantitative and qualitative analysis of the superficial vascular complex (SVC) and deep vascular complex (DVC) was performed. The main outcome measures were vessel density (VD), image quality, and the presence of artifacts.

Results: Mean VD of the SVC was significantly higher in the reference images compared with the images after induction of −2 D. Differences with −1 D were nonsignificant. Higher degrees of astigmatisms had higher VD dropout (0.012–0.02 per diopter in SVC). Astigmatism axis showed no relevance in our cohort. Image quality assessed by two independent observers was graded as higher in images without astigmatism. Defocus and attenuation were more prevalent in images with astigmatism.

Conclusions: Astigmatism of −2 D affects quantification of VD in OCTA images, mainly affecting the SVC, as well as the subjective quality assessment. Correction of this refractive error might be necessary for an accurate quantitative assessment of OCTA images.

Translational Relevance: Correcting astigmatism of 2 D or greater appears to be necessary when analyzing OCTA images.

Introduction

Optical coherence tomography angiography (OCTA) offers a noninvasive visualization of the vasculature of the eye with valuable diagnostic and prognostic information.^1–3 As with other imaging modalities, the presence of artifacts might limit our ability to analyze and quantify OCTA images,⁴ thus, being aware of these artifacts is important for a correct interpretation of OCTA images. Image defocus is a common cause of OCTA image quality dropout,^5,6 affecting quantitative indicators, such as perfusion density (PD), vessel density (VD), and fractal dimension (FD). 

The impact of spherical refractive errors in OCT imaging quality was shown by Cheung et al. The authors described the influence of poor signal strength in tomographic retinal nerve fiber layer (RNFL) thickness measurements.⁷ Similarly, Yu et al. reported a direct relationship between OCTA image quality and VD measurements dropout, induced by beam attenuation and spherical defocus.⁸ These spherical errors are adjusted prior to optical coherence tomography (OCT) and OCTA imaging, however, astigmatism, a cylindrical refractive error,^9,10 cannot be corrected in commercially available devices. Hwang et al. demonstrated that induced with-the-rule and against-the-rule astigmatism can also affect OCT measurements of RNFL.¹¹ In another study, Jung et al. described the decrease on PD and VD affected by induced with-the-rule astigmatism.¹² To our knowledge, the degree of astigmatism affecting OCTA image quality and its correction, including the presence of artifacts and a subjective image evaluation, has not been described to date. 

The present study was designed to determine the influence of astigmatism on OCTA images at a qualitative and quantitative level, as well as to establish the degree of astigmatism in diopters (D) that significantly affects OCTA image analysis. 

Methods

Study Design

In this single-center, prospective, cross-sectional study, we recruited adult subjects presenting to our outpatient clinic between October 2020 and July 2021 for comprehensive ophthalmic evaluation. Good foveal fixation and clear media were inclusion criteria. We included patients regardless of ocular pathology or lens status. When both eyes met study criteria, only one eye for each patient was included, which were chosen randomly. 

Institutional review board (IRB)/ethics committee approval was obtained. All patients signed a written informed consent prior to their enrollment, and the study was conducted in accordance with the tenets of the Declaration of Helsinki. 

The patients were scheduled for OCTA images using SPECTRALIS HRA-OCT 2 (Heidelberg Engineering, Heidelberg, Germany). OCTA images including the macular central area were obtained using the 20 degrees × 20 degrees pattern (composed of 512 A-scans × B-scan and 512 B-scans, with an axial resolution of 3.9 µm/pixel). We excluded images with a Q value lower than 30, or in the presence of shadows or motion artifacts. The Q value indicates the signal-to-noise ratio, measured on a scale of 0 to 40, with 40 representing the best quality. 

Study Groups

Patients showing no astigmatic error, or −0.25 D of astigmatism measured by Topcon KR-800 keratometer (Topcon Corp., Tokyo, Japan), were included in the “astigmatism induction” group. Patients showing an astigmatism of −0.5 D or greater were included in the “astigmatism correction” group. Figure 1 shows an overview of the study protocol. 

Figure 1.

 

Study protocol. In patients without astigmatism (above), 3 OCTA images were acquired: baseline image without astigmatism, after induction of −1 diopter (D) (OCTA 1) and −2 D (OCTA 2). In each image, we collected vessel density (VD) values in the superficial vascular complex (SVC) and deep vascular plexus (DVC), and compared with the baseline using the Wilcoxon test. In patients with astigmatism (below), we performed an OCTA scan with astigmatism (baseline) and after correction (OCTA1), and compared VD values between them. We also stratified data according to astigmatism intervals in diopters and astigmatism axis (with-the-rule [WTR], against-the-rule [ATR], and oblique [OBL]). We used, in this case, the ANOVA test for comparison and simple regression analysis in order to quantify the relationship between diopters and VD.

In the group of astigmatism induction, a baseline OCTA image with no refractive correction was obtained and set as reference. Two additional scans were acquired enabling the follow-up function and inducing an astigmatic error of −1 and −2 D by using a set of cylindrical lenses provided by Heidelberg Engineering. The cylindrical lenses were attached on top to the lens of the camera head of the SPECTRALIS device, in a randomly chosen axis (90 degrees, 180 degrees, or 45 degrees). Images with none or induced astigmatism were performed in random order to avoid the fatigue bias. 

In the group of astigmatism correction, we obtained the OCTA images before and after astigmatism correction, using the same set of lenses, placing them in the corresponding axis in each case. Images with corrected and uncorrected astigmatism were performed in random order to avoid the fatigue bias. 

Image Analysis

Quantitative Analysis

A quantitative analysis was made using an OCTA analytics software provided by Heidelberg Engineering (version SP-X1902; Heidelberg, Germany). Comparison of the VD value (defined as vascular pixels/image area) obtained in 4 regions of the retina around the fovea of 2 mm² each: nasal, temporal, superior, and inferior to the fovea was performed. The foveal avascular zone was excluded as its interindividual and within-subject variability could mask the effects of the cylindrical error.¹³ 

• • Astigmatism induction in patients without astigmatism

In patients without astigmatism, we analyzed 3 images of each patient: baseline image without astigmatism, and after −1 and −2 D astigmatism induction. We analyzed data from all four areas studied and in both superficial vascular complex and deep vascular complex (SVC and DVC, respectively). 

• • Astigmatism correction in patients with astigmatism

In patients with natural astigmatism, we analyzed two images per patient (with astigmatism and after correction). We divided data according to the degree of astigmatism, having a total of 4 subgroups: −0.5 to −1.25 D, −1.5 to −2.25 D, −2.5 to −3.25 D, and −3.5 to 5.25 D. We also studied the influence of astigmatism axis, classifying data in three groups: with-the-rule (60–120 degrees), against-the-rule (0–30 degrees and 150–180 degrees), and oblique (120–150 degrees and 30–60 degrees). 

Qualitative Analysis

A blind qualitative analysis was performed comparing the en face reconstructions of the SVC and DCP from the reference OCTA image and the follow-up images. A number was assigned to each image according to their quality by two independent observers (authors R.D.M. and L.V.O.): maximum (1), intermediate (2), and poor (3). In addition, both observers assessed the presence of defocus, attenuation, and striping artifacts in a masked manner. A defocus artifact was detected when vessels or capillaries are seen with less definition or image sharpness; attenuation artifact is seen as a loss of OCT signal in a selected area and striping was detected as white lines in the direction of the scan. Grading was made separately for the SVC and the DVC images. Averaged data of the two graders was used for analysis. 

Statistical Analysis

Quantitative Analysis

In the astigmatism induction group, the quantitative analysis was made using Wilcoxon matched-pairs signed-rank test and corrected for multiple comparisons using the Benjamini-Hochberg method. The qualitative data were analyzed according to the degree of astigmatism in the three groups using the chi-square method for binomial variables (defocus and attenuation) and Friedman test for the quality assessment. 

In the astigmatism correction group, the quantitative analysis was made using the ANOVA test for paired samples. We also performed a simple linear regression analysis to estimate the changes in VD according to astigmatism diopters. The qualitative data was analyzed according to the degree of astigmatism in the two groups using the chi-square method for binomial variables (defocus and attenuation) and the Student t-test for the quality assessment. 

A P value of < 0.05 considered statistically significant. 

Wilcoxon matched-pairs signed-rank, chi-square, t-test, Friedman, and linear regression analysis were performed using Prism9 (GraphPad Software, LLC; version 9.3.1, 2021). Multiple comparison correction was performed using R (version 4.1.1) in RStudio (2021-09-0). 

Results

One hundred ten eyes of 110 subjects were included in the present study. Twenty eyes from 20 patients without astigmatism were included in the astigmatism induction group (mean age of 49.85 years [range = 19–68 years]; 12 were women [60%]). Six patients were then induced an astigmatic error with axis at 180 degrees (30%), 7 patients at 90 degrees (35%), and 7 at 45 degrees (35%). Ninety eyes from 90 patients with astigmatism equal or higher than −0.50 D were included in the astigmatism correction group (mean age = 64.6 years [range = 19–94 years]; 50 were female patients [55%]; mean astigmatic error = −1.43 D [range = −0.5 to −4.5 D; 95% confidence interval [CI] = 1.26–1.59]). Distribution of astigmatism axis and diopters intervals are shown in Figure 2. 

Figure 2.

 

Distribution of astigmatism axis (left) and diopters (right) in patients with astigmatism. WTR, with-the-rule; ATR, against-the-rule; OBL, oblique.

Quantitative Metrics

• • Induced astigmatism in patients without astigmatism

Mean VD values in the reference images (without astigmatism) showed nonsignificant differences with the images with −1 D astigmatic induction in all 4 analyzed areas for both the SVC and the DVC. However, VD values significantly decreased after −2 D astigmatism induction in all areas of the SVC and in superior and inferior areas in the DVC. Table 1 shows mean VD values, P values and percentage of VD reduction compared with baseline. We did not study the effect of the axis in this group due to the limited sample size. 

• • Astigmatism correction in patients with astigmatism

Table 1.

 

Vessel Density Values of the Superficial and Deep Vascular Complex in the Four Macular Areas Without Astigmatism (Baseline) and After −1 D and −2 D Astigmatism Induction (Shown as Mean and CI 95%)

 

After stratifying by degree of astigmatism, differences in VD were significant in the temporal, nasal, and superior areas in the SVC and the DVC (Fig. 3). However, we did not find any significant differences after stratifying by axis of astigmatism (see Supplementary Fig. S1). 

Figure 3.

 

Interleaved box and whiskers (mean and range) showing differences in vessel density (VD) in superficial (SVC) and deep vascular complex (DVC) according to astigmatism intervals in diopters in corrected and uncorrected images. *P < 0.05 using the ANOVA test.

Linear regression analysis showed a positive relationship between astigmatism diopters and VD differences. In the SVC, each 1 D increase in the astigmatism caused an increase in differences before and after correction of the VD ranging from 0.01267 to 0.0219 among sectors. However, the regression model showed only statistically significance in the nasal and the superior sectors. The DVC analysis showed similar results, with smaller differences, ranging from 0.008187 to 0.01566 among areas, being only statistically significant in the superior and the inferior sectors (Fig. 4). 

Figure 4.

 

Linear regression analysis between differences in vessel density (VD) in images with and without astigmatism correction. Higher degrees of astigmatism show greater differences in VD. The P values and the regression equation are shown in each case. *P < 0.05.

Qualitative Assessment

• • Induced astigmatism in patients without astigmatism

Image quality grading was statistically significant prior and after astigmatism induction of −1 and −2 D in both SVC and DVC. Figure 5 shows frequencies and P values using Friedman test. Figure 6 shows an example of OCTA images of the same patient with different astigmatism induction. 

Figure 5.

 

Image quality of the group without astigmatism. Graph with interleaved bars showing the image quality assessment and the presence of attenuation and defocus artifacts in superficial vascular complex (SVC) and deep vascular complex (DVC). Columns show the frequencies of each grade of image quality and the frequencies of each artifact in the different groups (without astigmatism [baseline]), with −1 diopter (D) astigmatism induction and with −2 D astigmatism induction. The P values are shown in each case. Q1 = maximum quality; Q2 = intermediate quality; and Q3 = poor quality. *P < 0.05.

Figure 6.

 

Upper row: En face images of the superficial vascular plexus (SVP) without astigmatism (left, baseline) and after induction of −1 D (center) and −2 D (left) of astigmatism. Lower row: En face images of the SVP before (right) and after (left) astigmatism correction. Less definition of capillaries and increased diameter of large retinal vessels can be seen in the images with astigmatism. D, diopters.

Differences were statistically significant in the three-group analysis (baseline and after astigmatism induction of −1 D and −2 D) for the presence of attenuation (P = 0.0026) and defocus (P = 0.0004) artifacts in the SVC. Similarly, in the DVC, group differences were significant for attenuation (P = 0.0007) and defocus (P = 0.0138). No statistically significant differences were found in the percentage of images showing striping on the SVC P = 0.392 and the DVC P = 0.122. 

• • Astigmatism correction in patients with astigmatism

In the astigmatism group, after astigmatism correction images had higher quality compared with uncorrected ones (Table 2, see Supplementary Fig. S2). Additionally, the percentage of images showing attenuation and defocus were also higher in the uncorrected images, compared with the corrected ones (see Table 2, Fig. 6). No statistically significant differences were found in the percentage of images showing striping on the SVC P = 0.289 and the DVC P > 0.999. 

Table 2.

 

Eyes With Astigmatism

 

Discussion

The present study evaluated the effect of astigmatism on OCTA qualitative and quantitative parameters in adult patients using the SPECTRALIS device. To the best of our knowledge, this is the first attempt to evaluate the effect of both induced and corrected astigmatism using a qualitative and quantitative approach, and taking into account the degree of astigmatism. 

Astigmatism causes an optical blur, being one of the causes for image defocus artifact. Although other artifacts that alter image quality might be attenuated using computational algorithms implemented in common OCTA software, defocus is harder to correct.^4,14–21 For example, Holmen et al. and Lujan et al. highlighted image defocus as one of the most prevalent artifacts in OCTA images.^20,22 In addition, Tomlinson et al. showed in a single-case report the impact of spherical errors as they may cause global vascular dropout and poor reflectivity.²² Furthermore, Yu et al. demonstrated that signal strength affects vessel density measurements. They found that 1 D of spherical defocus significantly influenced VD. Beam aberration, which includes defocus, astigmatism, and higher order aberrations, diminishes signal strength, and increases the focal spot size of the OCT beam. Using the AngioVue system, they hypothesized that the effect of astigmatism lower than 2 D were negligible. Although differences in the Rayleigh range between devices might influence the results, these theoretical data are consistent with our results.⁸ We found that −1 D of induced astigmatism had minimal effect on VD, and systematic correction may be less relevant with these degrees of astigmatic defect. However, −2 D of astigmatism significantly decreased VD measurements, as well as caused qualitative changes in OCTA en face images, mostly in the SVC. The results were constant in all the studied areas, suggesting that no other artifact has influenced our findings. In addition, we suggest that the degree of astigmatism might increase the VD dropout, after analyzing a larger cohort of 90 patients. These changes might not be clinically meaningful, as they change in order of 0.012 to 0.02 in a per diopter basis. However, it could be relevant in studies addressing quantification changes, and should be considered as a limiting factor. 

In the same direction, Jung et al. analyzed the influence of the induction of with-the-rule astigmatism in 15 cases in the SVC using contact lenses.¹² Although they used a different OCTA device (spectral domain [SD]-OCTA Cirrus 5000 Angioplex; Carl Zeiss Meditec, Dublin, CA, USA) with different segmentation boundaries and other settings, they also reported a reduction in the total area of perfused vasculature per unit area, in the range of 0.0064 to 0.0089 on a per diopter basis. Another difference from our study is that they used toric contact lenses for astigmatism induction, whereas we used a cylindrical lens attached to the OCTA device, which may be easier to use in the clinic for astigmatism correction. In addition, we stratified our data according to astigmatism axis, showing that it might have no influence in the quantitative assessment, although the small number of patients per group might have influenced our findings. We also found similar effects in the quantitative metrics, despite the axis of the steepest meridian, after inducing astigmatism. Finally, we also included the DVC in the analysis, although the influence of astigmatism was weaker in this vascular plexus but showed similar trends compared to the effect on the SVC. We hypothesized that other artifacts may play a role in this deep slab, and that differences in VD calculations on capillary vessels might be more difficult to assess. In addition, we found that the differences found in the SVC were significant in more areas compared with the DVC. The superficial plexus is the least affected by projection artifacts, which are usually present in deeper layers, such as the intermediate and deep plexus.^4,23 

Our data also showed that astigmatism significantly affects the quality of OCTA images. The qualitative assessment showed worse image quality in the images acquired in the presence of astigmatism (astigmatic induction in patients without astigmatism, mostly of −2 D, or corrected images in those whose astigmatic error was higher or equal to −0.50 D), both in the SVC and the DVC. In addition, the percentages of images showing attenuation and defocus artifacts were higher in those images acquired with astigmatism, which can also explain the vascular dropout demonstrated in the quantitative analysis. On the contrary, the incidence of striping artifact was similar in all groups. The striping artifact is related to eye movements and is therefore not influenced by astigmatism, but by patient characteristics. This suggests that patient cooperation was comparable in all acquisitions and may not have acted as a confounding factor. 

Several trials have demonstrated the predictive power of OCTA parameters in retinal vascular pathologies.^24–28 However, analysis of data according to astigmatism error and its correction is not protocolized. Up to date, it is possible to correct spherical errors using the manual or auto-focus tools in current OCTA devices,²⁰ but astigmatism remains uncorrected. Hence, astigmatism might stand as a limiting factor for high-quality OCTA imaging and accurate OCTA quantification in certain conditions. Based on our data, it seems likely that astigmatic correction, especially above −2 D, may improve image quality and therefore it may also improve vascular metrics. The prevalence of high astigmatism (above 2 D) defects in the general population is relatively low, 5% of the population.²⁹ As such, our cohort of naturally occurring astigmatism had a mean astigmatism of 1.43 D. Therefore, a systematic astigmatism correction in the clinical setting might not be necessary according to our results, as nearly 70% of adults have a corneal astigmatism equal or lower than 1 D according to other studies.^29,30 However, for certain conditions with high astigmatism, such as keratoconus, where accurate OCTA quantification may be limited, the option of astigmatism correction should be available and considered. Nonetheless, we do not know the effect of astigmatism between −1 and −2 diopters as the set of lenses used included only full diopters. In addition, our results suggest that astigmatisms of 2 D or more, if uncorrected, might serve as a limiting factor for the use of OCTA quantitative metrics in clinical practice or clinical trials. 

The main strength of our study is that we included two groups to test our hypothesis. Whereas the induction of astigmatism offers a clean experimental design with matched samples to increase statistical power, the larger group with naturally occurring astigmatism might be more applicable to the general population. In addition, we used a simple method for astigmatism correction, a single lens placed in front of the camera head, which is noninvasive and easy to use. 

Important limitations in our study include the limited number of patients, especially in the induced astigmatism group, which was too small to assess the influence of the axis and if there are any relationships between both variables (axis and diopters) affecting quantitative parameters, such as VD, PD, or FD. Although we did not find any sign that astigmatism axis might have an influence in these parameters in the natural astigmatism group, it could not be confirmed after astigmatism induction. In addition, we did not correct our results for axial length, which could also have influenced our results.³⁰ Another relevant limitation is that we only induced astigmatisms of −1 and −2 D. We did not experimentally study the influence of an induced astigmatism of −1.5 D, which is also frequent in the general population, especially in older patients.³¹ In addition, we could not perfectly correct natural astigmatism, as we only used lenses from −1 to −5 D. To solve this, we estimated the changes of VD according to the increase in astigmatism diopters using a regression analysis in the natural astigmatism group. 

In summary, we showed that the presence of −2 D of astigmatism resulted in significantly reduced VD measurements, as well as diminish image quality including large vessel and capillary blurring. This refractive error must be considered in studies addressing quantitative OCTA parameters. Further studies with a higher number of patients and a wider range of astigmatic defects are needed to create a more accurate model to estimate VD changes in relation with astigmatism diopters. 

Acknowledgments

Disclosure: L. Vidal-Oliver, None; R. Gallego-Pinazo, Novartis (C, F), Roche (C, F), Zeiss (C), Allergan (C), ORA (C), Heidelberg Engineering (F), IvericBio (F); R. Dolz-Marco, Heidelberg Engineering (C), Roche (F), Novartis (F), and IvericBio (F)