Segmentation Errors and Off-Center Artifacts in SS-OCT: Insight from a Population-Based Imaging Study

Abstract Purpose To evaluate the frequency and associated factors of artifacts in swept-source optical coherence tomography (SS-OCT) imaging. Methods This was a population-based cross-sectional study. Individuals aged 35 years or older, residing in the Yuexiu district of Guangzhou, China, were recruited by random cluster sampling. Nearly half of the participants were randomly selected for SS-OCT imaging centered on the optic nerve head. Six types of artifacts in the peripapillary choroidal layer and retinal nerve fiber layer (RNFL) were graded and identified. Univariate and multivariate logistic regression analyses were used to investigate the association between the presence of artifacts and clinical characteristics. Results Out of the 616 eligible individuals who underwent SS-OCT imaging, 18.3% and 13.6% of subjects exhibited at least one artifact in peripapillary RNFL (pRNFL) and peripapillary choroidal thickness (pCT) measurements, respectively, with posterior segmentation error and off-center artifact ranked as the most common artifacts. The presence of artifacts was significantly associated with age (odds ratio [OR], 1.03; 95% confidence interval [CI], 1.01–1.06; p = .003), refractive error (OR, 0.80; 95% CI, 0.71–0.89; p < .001), and signal strength (OR, 0.95; 95% CI, 0.90–0.997; p = .039) in pRNFL thickness measurement. Similarly, the presence of artifacts in pCT measurement was significantly associated with age (OR, 1.05; 95% CI, 1.03–1.08; p < .001), and refractive error (OR, 0.76; 95% CI, 0.68–0.86; p < .001). Conclusion Nearly one-fifth of the eyes were noted with at least one artifact in the population-scale SS-OCT study. Age was a risk factor for the presence of artifacts and should be considered in clinical settings.


Introduction
The widespread use of optical coherence tomography (OCT) has revolutionized the clinical practice in ophthalmology.Before any attempts for retinal or choroidal layer quantifications, the scan artifacts impose substantial challenges.2][3] Moreover, these artifacts occur in subjects without obvious retinal pathologic features (6-44% for Cirrus, 38% for RTVue, and 53% for Topcon 3D). 4,5Thus, recognition of artifacts and associated factors is essential for correct interpretation and accurate quantification.
Swept-source optical coherence tomography (SS-OCT) improves tissue penetration through the retinal pigment epithelium (RPE), which allows for better visualization of the choroidal-scleral interface. 6Accumulated studies have demonstrated the great value of SS-OCT in detecting retinal and choroidal abnormalities. 6Moreover, the use of automated measurement via SS-OCT built-in software provides an efficient and objective method for quantitatively evaluating retinal or choroidal structures, which is also highly reproducible. 6Previous studies have summarized a variety of artifacts, mainly including misidentification of retinal layers, incomplete or complete segmentation error, degraded image, off-center, cut edge, out of register, motion artifact, and blink artifact in macular spectral domain OCT (SD-OCT). 3However, relevant information regarding scan artifacts in SS-OCT remains unclear.
The quantitative evaluation of the peripapillary choroid and retinal nerve fiber layer (RNFL) plays an important role in the diagnosis and monitoring of glaucoma and non-glaucomatous optic neuropathies.However, the prerequisite of quantification, artifact evaluation, has not been fully determined in SS-OCT imaging.To fill the knowledge gap, the aim of study was to assess the frequency and associated factors of artifacts in SS-OCT imaging among population-based individuals.

Study population
This was a population-based observational study.Details of the sampling and recruitment methodologies have been reported previously. 7In brief, a total of 1817 subjects aged 35 years or older, residing in the Yuexiu district of Guangzhou, China, were selected and recruited by random cluster sampling in 2008 at baseline.Demographic information including age and gender was documented.In 2014, 1427 subjects underwent follow-up ophthalmic examinations and were randomly assigned to two groups: 662 subjects with SS-OCT conduction and 765 subjects without SS-OCT conduction (Figure 1).The study was approved by the Zhongshan Ophthalmic Center Ethics Committee, and performed in accordance with the tenets of the Declaration of Helsinki.Written informed consent to participate in this study was obtained from all subjects.

SS-OCT imaging
The SS-OCT device (DRI-OCT, Topcon Inc., Tokyo, Japan) was adopted for optic nerve head (ONH) imaging, which uses a wavelength of 1050 nm with a scan speed of 100,000 A-scans per second, yielding axial and lateral resolutions of 8 mm and 20 mm in tissue, respectively.In this study, SS-OCT was used to scan the peripapillary area in subjects by a well-trained technician blinded to the research protocol.The peripapillary region was scanned using a three-dimensional optic disk scanning mode, with a 360 , 3.4-mm-diameter circle that was centered on the optic disc.Scans were obtained using internal fixation.During the examination, any SS-OCT images with signal strength below 45, which is defined as poor quality, would be discarded.Subsequently, the technician would repeat the SS-OCT imaging process on the subject until achieving the desired image quality.The internal limiting membrane and inner border of the retinal ganglion cell layer were delineated using the segmentation algorithm implemented in peripapillary RNFL (pRNFL) thickness measurement by SS-OCT.Bruch's membrane and the choroidal-scleral interface were delineated using the segmentation algorithm implemented in peripapillary choroidal thickness (pCT) measurement by SS-OCT.The peripapillary RNFL and choroid layer were automatically segmented into the temporal, superior, nasal, and inferior quadrants.

Definitions of artifacts in SS-OCT
Six types of artifacts were included based on our observations and previous studies.The definition of those artifacts was listed as follows: (1) "anterior segmentation error", defined as the misidentification of the anterior pRNFL boundary (the internal limiting membrane) and misidentification of anterior pCT boundary (the Bruch's membrane); (2) "posterior segmentation error", defined as the misidentification of the posterior pRNFL boundary (the inner border of ganglion cell layer) and misidentification of the posterior pCT boundary (the choroidal-scleral interface).According to the proportion of involvement in pRNFL thickness and pCT, these artifacts were further classified as less than 1/3, between 1/3 and 2/3, and more than 2/3 of segmentation errors. 8Segmentation errors at the temporal, superior, nasal, and inferior quadrants were also evaluated; (3) "out of register" was defined when the portion of the retina or choroid was vertically located out of the range of the scan 9 ; (4) "degraded image" was defined when part of the scan was missing, which apparently looked like the broken retina or choroid 3 ; (5) "cut edge" artifact referred to an inappropriately truncated edge of the scan 9 ; (6) "off-center" artifact was defined when the center of the ONH was more than approximately 10% off the center of the peripapillary circular scan. 2 All SS-OCT images were screened for quality inspection, then an experienced ophthalmologist (W.W.) identified scan artifacts without any knowledge of participants' data.

Other variables
The demographic data were obtained by standardized questionnaire.Noncycloplegic automated refraction was performed on all participants using the same device after proper calibration (KR-8800; Topcon Corp., Tokyo, Japan).Five consecutive measurements were performed for each eye and the mean was recorded as the final value.The spherical equivalent (SE), calculated as spherical power plus half of cylindrical power, was used to represent refraction.

Statistical analysis
Data from the right eye of each participant were analyzed.Independent samples t-test or Chi-square test was used to compare the clinical characteristics of participants with and without SS-OCT, and those with and without artifacts in the RNFL and choroidal layers.Univariate or multivariate logistic regression analysis was used to investigate the association between the presence of artifacts in pRNFL thickness and pCT measurements with clinical characteristics.All statistical analyses were performed using Stata software (Stata Corp, College Station, TX), and statistical significance was set at p < .05.

Results
Among the 1427 participants, 662 of them were randomly selected for SS-OCT imaging in this study.Table S1 shows the clinical characteristics of participants with and without conduction of SS-OCT.Data from 46 participants were excluded because they were unable to complete the SS-OCT examination.Therefore, 616 participants with eligible OCT images were included in the final analysis.
Table 1 presents the frequency of artifacts in pCT and pRNFL thickness measurements as determined by SS-OCT.Among 616 participants, 84 subjects (13.6%) had at least one artifact in pCT measurement.Posterior segmentation errors (11.2%) were the most common artifacts, followed by off-center artifacts (7.3%), anterior segmentation errors (6.7%), and degraded images (0.5%). Figure 2 shows the examples of pCT artifacts in SS-OCT.In addition, anterior or posterior segmentation errors at the nasal and inferior quadrants were more common than those at the temporal and superior quadrants in pCT measurement.Similarly, 113 subjects (18.3%) had at least one artifact in RNFL thickness measurement, with posterior segmentation errors (17.2%) being the most common artifacts, followed by off-center artifacts (10.2%), anterior segmentation errors (5.2%), and degraded images (1.8%).Figure 3 shows the examples of pRNFL artifacts in SS-OCT.
Table 2 shows the comparison of clinical characteristics of participants with and without artifacts.Significant differences in age, refractive error, and signal strength were noted between 84 participants with artifacts and 532 participants without artifacts (p < .001,p < .001,and p < .001,respectively) in the choroid layer.Similarly, age, refractive error, and signal strength differed between 113 participants with artifacts and 503 participants without artifacts in the RNFL layer (p ¼ .009,p < .001,and p < .001,respectively).

Discussion
To the best of our knowledge, this is the first populationbased study to evaluate the scan artifacts in SS-OCT imaging with a relatively large sample size.We found that approximately 18.3% and 13.6% of subjects appeared at least one artifact in pRNFL and pCT measurements, respectively, with posterior segmentation error and off-center artifact seen as the most common artifacts.Moreover, age, refractive error, and signal strength were significantly associated with the presence of artifacts, and should be considered in clinical settings.These findings may be helpful in preventing erroneous clinical interpretations related to essential diagnostic and prognostic information in medical decision-making.
Currently, few studies have evaluated the scan artifacts in SS-OCT.Among 64 patients with glaucoma or suspected  glaucoma, Lee et al. reported that 35.9% of them had artifacts in the disc area by SS-OCT, with boundary misidentification ranked as the most common artifact. 10Mansouri et al. reported that artifacts were present in up to 9% of scans in SS-OCT imaging from 27 healthy subjects. 11Most of previous studies investigated the pRNFL measurement artifacts in SD-OCT and reported that 19.9-61.7% of scans had at least one artifact in eyes with glaucoma or suspected glaucoma, 1,2,10,[12][13][14][15][16][17] which is more common than that in normal eyes (14.7-25.4%). 14,16Likewise, the overall frequency of pRNFL measurement artifacts in this study (18.3%) was lower than that reported by Lee et al. (35.9%), which may be due to the difference in study population.Most of our participants were in the condition of physiological aging alone, and had less obvious RNFL damage than those patients with glaucoma or suspected glaucoma.Our findings suggest that artifacts in pRNFL thickness measurement, as determined by SS-OCT, were relatively common among adults and should not be overlooked.On the other hand, operator-related artifacts, including out of register and cut-edge artifacts, were not observed in both pCT and pRNFL thickness measurements, which strengthened the necessity of a proper operating procedure when using SS-OCT.
To date, little information on scan artifacts in pCT measurement is available.Pablo et al. briefly reported that 3.7% of the eyes were excluded because of incorrect segmentation of choroidal boundaries using SS-OCT. 18This study firstly evaluated the frequency of six different artifacts in pCT measurement using SS-OCT.We found that 13.6% of subjects had at least one artifact in pCT measurement.The posterior segmentation errors were the most common artifacts (11.2%), followed by off-center artifacts (7.3%), and anterior segmentation errors (6.7%) and degraded images (0.5%).This may be explained by the fact that reflectivity of the posterior boundary of pCT is attenuated by densely pigmented RPE and dense choriocapillaris microvasculature, whereas additional studies are warranted.
In addition, we found that segmentation errors of the choroid were more common at the nasal and inferior quadrants.Jiang et al. reported that the thinner pCT was associated with increased age in a large Chinese population-based study. 19Meanwhile, pCT at inferior and nasal regions was relatively thin. 19,20In the present study, the mean age of participants who had artifacts in pCT measurement was 60.8 ± 12.8 years.Thus, we inferred that the thinning of pCT at the nasal and inferior quadrants may cause reduced reflectivity of pCT more easily and influence the ability of automatic segmentation.
In glaucomatous eyes, posterior segmentation errors (41.7-46.2%)were found to be the most common artifacts, followed by off-center artifacts (10.4-20%) and anterior segmentation errors (5.0-9.4%). 14,16However, Liu et al. found that off-center artifacts (27.8%) were the most common artifacts in pRNFL thickness measurement, which was higher than that of posterior segmentation errors (7.7%) and anterior segmentation errors (3.16%). 2 In normal eyes, some researchers found that off-center artifacts (8.4-19.3%)were more common than posterior segmentation errors (5.3-7.0%) and anterior segmentation errors (0.9-3.2%). 14,16verall, based on previous studies, both posterior segmentation error and off-center artifact are the common artifacts; in particular, the posterior segmentation error occurs more frequently than anterior segmentation error does, which was in accord with our findings.
Glaucomatous RNFL thinning and loss of reflectivity of the RNFL hinder the ability of automated segmentation. 21,22hus, some researchers inferred that the posterior segmentation of the RNFL was primarily affected, as it became less easily distinguished from the underlying retinal layers. 2,16It has been suggested that pRNFL thickness decreases with increasing age. 23,24In the present study, the mean age of participants who had artifacts in pRNFL thickness measurement was 59.1 ± 11.7 years, which seemed plausible to explain the high incidence of posterior segmentation errors.Segmentation errors of pRNFL measurement at nasal and inferior quadrants were more common in this study.Ye et al. reported that the inferotemporal and nasal sectors were the most common locations showing segmentation errors, 25 partially attributed to the relatively thin pRNFL thickness at the nasal quadrant in non-glaucomatous eyes. 20,26However, pRNFL thickness at the inferior quadrant is relatively thick in normal eyes, 20,26 we thus inferred that signal attenuation due to shadows caused by retinal vessels crossing the circumpapillary scan location may be another cause.Given that segmentation errors are software-related artifacts, 3 our findings suggest that software algorithms for automated measurement of pRNFL thickness in elderly people remain to be improved.As previously described, off-center artifacts occur due to patient-related fixation errors (eccentric fixation, low vision, and attention deficit), 3,9,27 even with the prevalence rate of 8.4-19.3% in normal eyes. 14,16Thus, reminding subjects of looking at internal/external fixation location before and during image acquisition should decrease the occurrence of off-center artifacts.Additionally, the OCT operator may need to manually correct the scan location when encountering patients with unstable fixation stability.
The presence of scan artifacts in SS-OCT was significantly associated with age, refractive error, and signal strength.Previous SD-OCT findings showed that the presence of artifacts in pRNFL thickness measurement was significantly associated with age, 12,15 while Liu et al. found that it was not related to age and refractive error with a large sample size. 2 It may be due to the fact that the eyes of our participants differed from glaucomatous eyes that typically had severe RNFL thinning in the study by Liu et al. 2 Our results indicate that clinical interpretation of pRNFL thickness should be done cautiously and may require manual correction when encountering the elderly.In addition, poor signal strength has been validated as the source of artifacts in other OCT devices, 28,29 which was consistent with our findings in SS-OCT.This finding may be due to that media opacities of any kind (such as cataract) can degrade signal strength, and thus exert significant impacts on RNFL measurement. 28,30his study had some limitations.First, this study comprised relatively fewer clinical characteristics, which may not be sufficient to contain other factors associated with the presence of artifacts.Second, we did not investigate the relationship between certain ocular pathology and different types of artifacts because only a very small number of participants had clinically significant retinal disorders.Third, the results cannot be applied to other SS-OCT devices.Notably, our findings suggested the evaluation of artifacts is highly recommended before clinical diagnosis or interpretations of measurements.Fourth, the clinical value of this study may be somewhat limited due to the fact that the pCT assessment is less commonly performed than that of macular choroidal thickness in clinical practice.

Conclusion
In summary, this population-based study demonstrated that approximately 18.3% and 13.6% of subjects had at least one artifact in pRNFL thickness and pCT measurements using SS-OCT, respectively.The major artifacts were posterior segmentation errors and off-center artifacts.Moreover, age, refractive error, and signal strength were significantly related to the presence of artifacts, which should be considered in the interpretation of OCT-derived measurements.

Figure 1 .
Figure 1.Flowchart of the participants.

Figure 2 .
Figure 2. Examples of artifacts in peripapillary choroidal thickness (pCT) measurement as determined by SS-OCT.(A) Off-center (left) and degraded image(right); (B) Degraded image shows that portions of the retina and the choroid are missing; (C) Anterior segmentation error involves in less than 1/3 of pCT; (D) Posterior segmentation error involves in less than 1/3 of pCT; (E) The segmentation line incorrectly delineates the entire posterior boundary of pCT; (F) Anterior segmentation error involves in between 1/3 and 2/3 of pCT, which identifies incorrectly as posterior boundary of pCT.Meanwhile, approximately half of the posterior boundary of pCT was segmented incorrectly.

Figure 3 .
Figure 3. Examples of artifacts in peripapillary retinal nerve fiber layer (pRNFL) thickness measurement as determined by SS-OCT.(A) Off-center, accompanied by posterior segmentation error of pRNFL; (B) Anterior segmentation error involves in less than 1/3 of pRNFL thickness; (C) Anterior segmentation error involves in more than 2/3 of pRNFL thickness; (D) Posterior segmentation error involves in less than 1/3 of pRNFL thickness.(E) Posterior segmentation error involves in between 1/3 and 2/3 of pRNFL thickness.(F) Posterior segmentation error involves in more than 2/3 of pRNFL thickness.

Table 1 .
Frequency of artifacts in SS-OCT imaging in the ONH region.

Table 2 .
Comparison of clinical characteristics of participants with and without artifacts in SS-OCT imaging.

Table 3 .
Logistic regression analyses on the presence of artifacts in the choroid layer by SS-OCT imaging.

Table 4 .
Logistic regression analyses on the presence of artifacts in the RNFL layer by SS-OCT imaging.