Detection of Choroidal Neovascularisation in Flat Irregular Pigment Epithelial Detachment in Central Serous Chorioretinopathy Using Optical Coherence Tomography Angiography: en-face Image Combined With Cross-Sectional Image

Background: Central serous chorioretinopathy (CSC) is a disorder characterised by serous detachment of the retina and is associated with retinal pigment epithelium (RPE) alteration. Flat irregular pigment epithelium detachment (FIPED) is one of the pigment epithelial detachment (PED) patterns of CSC and it is associated with chronic CSC. This study investigated choroidal neovascularisation (CNV) in FIPED with en-face optical coherence topography angiography (OCTA) and cross-sectional OCTA, to evaluate the incidence of CNV and compare the ecacy of each method. Methods: We retrospectively studied OCT and OCTA images of 328 eyes with CSC. OCTA B-scans and macular cube were primarily reviewed for the detection of FIPED and CNV. En-face OCTA and cross-sectional OCTA with Angio-B view, which is an image that combines an OCT B-scan with a ow signal were analysed to evaluate the presence of CNV in FIPED. Results: CNV was observed in 23 eyes on en-face OCTA and 21 eyes on cross-sectional OCTA, among 93 eyes of 88 patients with FIPED. There were eight discrepant cases, in which result of en-face OCTA was not equal to that of cross-sectional OCTA. Conclusions: The CNV and FIPED lesions were well detected on cross-sectional and en-face OCTA. Integration of these two methods can potentially improve the utility and diagnostic accuracy of OCTA.


Background
Central serous chorioretinopathy (CSC) is a disorder characterised by serous detachment of the neurosensory retina and is associated with retinal pigment epithelium (RPE) alteration. It is predominantly encountered in young or middle-aged individuals [1,2].
In uorescein angiography (FA), diffuse RPE defects can be visualised and a single point or multifocal leakage points are observed according to the degree of chronicity. Indocyanine green angiography (ICGA) shows dilation of large choroidal veins in the mid-phase. Choroidal vascular hyperpermeability and congestion are thought to be the main causes of CSC development [3]. Optical coherence tomography (OCT) can identify various aspects of RPE alterations, according to the CSC chronicity [4].
Flat irregular pigment epithelium detachment (FIPED) is one of the pigment epithelial detachment (PED) patterns of CSC. It is mainly observed in chronic CSC compared with acute CSC. It has an irregular RPE elevation compared to a dome-shaped PED. Most FIPEDs are known to be avascular, however, choroidal neovascularisation (CNV) in FIPED has been reported in long-standing CSC cases [5,6]. Because of multiple RPE changes and dilated pachy-vessels of the choroid, it is often di cult to identify CNV on FA or ICGA [7].
Optical coherence tomography angiography (OCTA) is a new imaging modality that can detect blood ow without invasive dye injection. It can effectively identify the presence of CNV in FIPED as an en-face image [8,9]. Swept-source OCTA can detect blood ow under the RPE and can be identi ed by layer, which is advantageous over FA and ICGA.
However, the analysis using an en-face image in the existing OCTA protocol may result in errors, such as motion artefacts, projection artefacts, layer segmentation errors, and ow signal masking that can occur during the calibration of projection artefacts. En-face imaging is challenging because the size of FIPED is small to be accurately divided and analysed into layers. Manually correcting the layer segmentation for each cross-sectional image may be feasible only for research purposes; however, it is impractical to rely on automated computer analysis.
To overcome the limitations of this clinical application, the hardware and software of OCTA was upgraded, and cross-sectional OCTA was introduced. Cross-sectional OCTA provides an image obtained by adding the ow signal to the existing OCT B-scan with false code. It is possible to observe the ow signal in the PED without introducing errors in the process of creating the en-face image.
This study investigated CNV in FIPED by two methods, en-face OCTA and cross-sectional OCTA, to evaluate the incidence of CNV and compare the e cacy of both.

Study design and ethical statement
We retrospectively reviewed the medical records of 290 patients (328 eyes) with CSC who visited the Department of Ophthalmology at Hanyang University Hospital between February 1, 2016, and February 1, 2018. This study was conducted according to the tenets of the Declaration of Helsinki and was approved by the Institutional Review Board at Hanyang University Seoul Hospital.
Eyes with other retinal abnormalities, such as polypoidal choroidal vasculopathy, age-related macular degeneration, idiopathic choroidal neovascularisation or other retinal vascular diseases, intraocular in ammation, and posterior segmental tumour were excluded. In addition, we excluded patients who had steroid-induced CSC and organ transplant-associated CSC. According to the manufacturer's recommendation [10], images of poor quality (image quality score lower than 45) provided by the onboard OCT software were excluded.

De nitions
The de nitions used in this study were adapted from previous reports [2]. CSC was de ned as a retinal disease characterised by serous detachment of the neurosensory retina secondary to one or more focal lesions of RPE [11]. Chronic CSC was de ned as presence of symptoms for more than six months, history of recurrence, background fundus change, such as atrophic retinal and RPE changes, visible drainage tract or other atypical dye leakages during FA initial examination, and chronic CSC [4,12]. Intravitreal anti-Vascular endothelial growth factor (VEGF) injection and photodynamic therapy (PDT) history were not excluded from the study criteria.
FIPED was de ned as a small, at, and irregular surface PED on the OCT B-scan. Unlike semi-circular PED, RPE elevation is not large but the RPE layer is separated from Bruch's membrane and a "double layer sign" is observed [13]. The sub-RPE space at the site of FIPED showed either hypo-re ection (not optically lled) or at least partial hyper-re ection, as previously described [7].

Study protocol
Patients' medical records were reviewed, including best-corrected visual acuity, spherical equivalent, dilated fundus biomicroscopy ndings, and multimodal imaging data including OCT, FA, ICGA, and OCTA.
All subjects underwent OCT and OCTA with Swept-source optical coherence tomography (SS-OCT) (Deep range imaging OCT, Triton, Topcon, Tokyo, Japan). We identi ed the presence of FIPED on OCT B-scans (Topcon SS-OCT parameters: 100 KHz, A-scan rate, wavelength 1050 nm). This study used OCTA ratio analyses employed by Topcon, which is an intensity ratio analysis, not based on amplitude-decorrelation. The OCT B-scans were then combined with the corresponding en-face OCTA images, and scrolling was enabled through the sections similar to that of OCT cube scan. In the cross-sectional OCTA, we used the Angio-B view provided by the Topcon, which is an image that combines an OCT B-scan with a ow signal coated with a false-colour code. The cross-sectional view helped identify the ow signal in the FIPED by providing scrolling function. It was possible to distinguish between projection artefacts and the real ow signals by checking the sequential sections.

Study population
This study retrospectively analysed the medical records of 290 patients (328 eyes), of which 256 patients (279 eyes) were actually included in the nal analysis. In the remaining 15 patients (26 eyes), OCTA was not performed or the image quality was found to be poor. Age-related macular degeneration (AMD) was suspected on FA and ICGA in seven patients (10 eyes were eight discrepant cases, in which CNV was observed on en-face OCTA, but not on cross-sectional OCTA. In the remaining cases, however, cross-sectional OCTA, but not en-face OCTA, could detect CNV. cases were identi ed as projection artefacts and 1 due to a segmentation error because of the choriocapillaris layer appearing under the FIPED. Figure 3 shows discrepancy of the CNV on en-face OCTA in 3 eyes (3.23%). A clear ow signal was observed in the cross-sectional OCTA; however, PED size was small, and it was masked in the process of removing projection artefacts in the en-face OCTA. Thus, among the 93 eyes, 21 cases of CNV in FIPED were identi ed and 8 cases showed discrepant results.

Discussion
In this study, of the 93 eyes with FIPED, CNV was observed in 21 eyes. The prevalence of CNV was 22.6%, which was closer to the lower border reported in previous studies. Previous reports have con rmed the presence of CNV in 18.9-58% of chronic CSC cases [7,8,13].
Using OCTA, several studies have quanti ed vessel density and ow index [14], as well as CNV [15,16]. Accurate segmentation is important to interpret and quantify OCTA ndings. However, in eyes with retinal pathology such as chronic CSC, the margins of the retinal layer are distorted and precise boundary segmentation can be di cult. Although many researchers have attempted to improve the quality of automated segmentation in diseased eyes [17,18], accurate automated segmentation is not achieved in all clinical cases. Therefore, manual correction of segmentation is often required [19,20]. In cases of chronic CSC, most FIPEDs are known to be avascular, although CNV in FIPED has been reported [5,6]. However, because of multiple RPE changes and dilated pachy-vessels of the choroid, it is often di cult to detect CNV on FAG or ICGA [7].
Recently, several studies on PED using en-face views in OCTA have been performed [8,9,13,21]. However, analysis using en-face images may result in errors, such as motion artefacts, projection artefacts, layer segmentation errors, and ow signal masking that can occur during the calibration of projection artefacts. To complement this, cross-sectional OCTA obtained by adding the ow signal to the existing OCT B-scan with false code, was introduced. It is possible to observe the ow signal in the PED without being affected by the errors that may occur in creating the en-face image. This results in continuous sectional views on scrolling. In cases of PCV and PED in AMD, some studies using cross-sectional OCTA as well as en-face view have been published [22][23][24][25].
It is important that the entire morphology of a CNV lesion is thoroughly analysed to qualitatively de ne a choroidal neovascular network, based on the shape, branching, anastomoses, type of vessels termini, and presence of hypointense perilesional halo [26]. However, there could be some errors in en-face OCTA due to the abovementioned di culties. Since CNV is not a coplanar structure, it may show a different morphology at each level of depth. En-face OCTA provides depth-resolved images; therefore, manual con rmation of the cross-sectional B-scan allows a layer-by-layer tomographic visualisation of the entire neovascular feature.
This study is signi cant because of the identi cation of CNV on FIPED of chronic CSC using both en-face and cross-sectional OCTA. The CNV in FIPED was observed in 23 eyes on en-face OCTA and 21 eyes on cross-sectional OCTA. The diagnostic e cacy of both the methods was similar; however, eight cases showed discrepant results on en-face OCTA. Among them, ve were false-positive cases, i.e., en-face OCTA detected the presence of CNV but there was no actual CNV in these eyes ( Figure 2). Four cases were identi ed as projection artefacts and one case as a segmentation error with the choriocapillaris layer under FIPED. The remaining 3 eyes had actual CNV con rmed on cross-sectional OCTA, but no CNV was seen on en-face OCTA. The reason for the false negativity was the small size of PED that could not be detected on en-face OCTA and masked PED in the process of removing projection artefacts. A total of eight eyes showed discrepant results on en-face OCTA that could not be ignored, and cross-sectional OCTA was required to compensate for these errors. With cross-sectional OCTA, a more complete layer-bylayer analysis of the entire structure could be performed to detect the precise location of the CNV lesion relative to the RPE layer.
When FIPED is observed in chronic CSC, it is not necessary to consider anti-VEGF treatment for the possibility of type 1 CNV. In this study, CNV in FIPED was found in 21 out of 93 eyes, and less than onethird of cases of active CNV. Previous studies have reported that OCTA ndings in PED were caused by the severe choriocapillaris alteration in those with AMD [16,27]. However, the vascular signal in FIPED with chronic CSC seems to be due to compensatory choriocapillaris vascular remodelling owing to a weak Bruch's membrane. The CNV lesion described as a "darker halo" or total absence of choriocapillaris with loss of both the inner and deeper choroidal vessels encircling the CNV lesions could not be found in this study. Therefore, careful observation in these cases compared to AMD is recommended.
Limitations of this study are the retrospective design and relatively small sample size. Additionally, this study investigated the strength of cross-sectional views, and unlike for the en-face view, clinicians need to observe each CNV lesion using manual scrolling. In addition, longitudinal studies are needed to con rm changes in CNV lesions in FIPED in chronic CSC patients. However, this study has value in that it is the rst study to analyse CNV in FIPED in chronic CSC patients using both en-face and cross-sectional OCTA.

Conclusion
In conclusion, this study demonstrates the potential feasibility of using cross-sectional OCTA to detect CNV. To identify CNV in FIPED without dye angiography, the reliability of detecting ow signal within any pathological lesion is critical. The clinical use of cross-sectional OCTA, as well as en-face OCTA to interpret FIPEDs may help clinicians in decision-making. Integration of these two methods can potentially improve the utility and diagnostic accuracy of OCTA.    Representative cases of discrepancy contrary to Figure 2. The en-face views (A and B) do not show any CNV, but in C and D, ow signals (arrowhead) can be seen in the cross-sectional view. A clear ow signal can be observed in the cross-sectiona l OCTA, but PED size appears very small, and it is masked in the process of removing projection artefacts in the en-face Optical coherence tomography angiography (OCTA).