Clinical features and prognosis of polypoidal choroidal vasculopathy with different morphologies of branching vascular network on optical coherence tomography angiography

This study highlights the clinical features and treatment response of polypoidal choroidal vasculopathy (PCV) among three different branching vascular network (BVN) morphologies in optical coherence tomography angiography (OCTA), and further correlates the BVN features with those under fluorescent angiography (FA) and indocyanine green angiography (ICGA). In total, we reviewed 70 eyes with PCV followed up for > 12 months. OCTA, ICGA and FA images were obtained at baseline and post-treatments. BVN was assessed using OCTA and divided into three types by a previously described BVN classification: type 1 (trunk), type 2 (glomeruli), and type 3 (stick). At baseline, type 1 BVN had the poorest vision and thinnest subfoveal choroidal thickness (SFCT), whereas type 3 had the best vision and thickest SFCT. The aforementioned trend sustained after treatments. Each BVN morphology in OCTA showed typical features in FA + ICGA and encompassed significant correlation (p = 0.004). In conclusion, OCTA is an innovative imaging tool for the detection and classification of BVN in PCV. Furthermore, OCTA has advantages of being noninvasive and free of systemic toxicities. The BVN can be divided into three types based on morphological characteristics in OCTA, which play crucial roles in clinical presentations and treatment outcomes.

www.nature.com/scientificreports/ structural differences, Coscas et al. observed that the visual acuity was significantly better in the idiopathic PCV group. However, no further implications were stated regarding visual prognosis in the aforementioned studies. Tan et al. proposed an innovative classification system of dividing PCV lesions into three BVN subtypes based on ICGA and fluorescent angiography (FA) images (types A, B, and C), which successfully predicted visual prognosis at 5 years 17 . Nevertheless, both ICGA and FA involve intravenous injection of dye for image retrieval. Severe anaphylactic reactions, although rare but detrimental, have been sporadically reported with the dye injection [18][19][20] . Conversely, OCT for choroidoretinal pathology assessment improved patients' safety and satisfaction revolutionarily 21 . OCT advancement further aided the visualisation of chorioretinal microvascular structures and provides en face imaging of high quality 22,23 . With the detection of red blood cell movement through decorrelation motion contrast between repeated scans, optical coherence tomography angiography (OCTA) can retrieve vascular images noninvasively [24][25][26] .
Huang et al. reported a novel BVN morphologic classification system with clinical relevance for treatment outcomes for PCV lesions that used OCTA 27 . However, clinical implications for long-term visual prognosis of PCV revealed through OCTA were not provided yet. In this study, the authors aimed to analyse the correlation between different BVN morphologies and biomarkers in PCV lesions through OCTA and FA + ICGA. Furthermore, the discrepancies in the demographic factors and prognosis between different BVN morphological types were re-validated and investigated.

Methods
This retrospective cohort study was performed in National Taiwan University Hospital (NTUH, Taipei, Taiwan, R.O.C.). The study was approved by the International Review Board of the National Taiwan University Hospital (20170803RINC) and was performed in accordance with the tenets of the Declaration of Helsinki. Owing to the nature of a tertiary referral and a clinical study centre in NTUH, the informed consents were obtained from the patients upon every visit and treatment episode.
Study design and image analysis. The authors recruited patients with PCV and followed them up for > 12 months in NTUH from January 2015 to December 2017. According to the predetermined study protocols, all included patients underwent OCTA, ICGA, and FA examination and received the combination therapy of photodynamic therapy (PDT) with a single anti-vascular endothelial growth factor (VEGF) agent (bevacizumab, ranibizumab, or aflibercept). OCTA images were retrieved using AngioVue (RTVue XR Avanti, Optovue Inc., Fremont, CA, USA), whereas ICGA + FA images were captured using SPECTRALIS HRA + OCT (Heidelberg Engineering, Heidelberg, Germany). The treatment protocol for the combination therapy of PDT has been reported previously 28 . The patients received an intravitreal injection of a single anti-VEGF agent followed by a PDT session. Furthermore, anti-VEGF intravitreal injection was subsequently administered monthly for the next 2 months. For patients who deferred PDT and only received anti-VEGF therapy, we omitted them from further study and statistics according to the study protocols and submitted IRB.
The patients with polyps occupying more than 3 × 3 mm area from the central macula, macular scarring, or choroidal neovascularisation (such as proliferative diabetic retinopathy and myopic choroidal neovascularisation) caused by factors other than PCV were excluded. The authors recorded age, sex, and PDT re-treatment status of all the enrolled patients. Moreover, the best-corrected visual acuity (BCVA) and imaging features in OCTA and ICGA+FA were recorded at baseline and 3, 6, and 12 months post-treatment to assess the treatment response. The BVN area were calculated by AngioVue OCTA system and were further analysed. The assessed biomarkers on structural OCT included subfoveal choroidal thickness (SFCT), central retinal thickness (CRT), subretinal fluid (SRF), retinal pigment epithelial detachment (RPED), and intraretinal cyst (IRC). All OCTA and ICGA+FA images were interpreted by two independent retinal specialists (S.-T.M. and C.-H.H.) to classify the PCV lesions. All disparities in interpretation between the two reviewers were arbitrated by another retinal specialist (C.-H.Y.).
Furthermore, we performed re-treatment and rescue therapy under appropriate clinical scenarios depending on clinicians' personal preferences in accordance to the PCV treatment consensus in Taiwan 29 . PCV reactivation was considered if new onset of vision loss ≥ 1 line or equivalent, and at least one additional criterion as follows was met: (1) subretinal or intraretinal fluid, (2) RPED, (3) subretinal or sub-retinal pigment epithelial (sub-RPE) haemorrhage, or (4) obvious fluorescent leakage. If the reactivation was confirmed by two independent specialists, the patients received further re-treatment or rescue therapy 30 .

BVN morphological classifications.
In the present study, the authors complied with the BVN morphological classification system presented by Huang et al. 27 Three distinct BVN patterns were identified in PCV lesions through OCTA and categorised as follows: Type 1 (trunk form), characterised by a major vascular trunk that further radiates into smaller vessel calibre pointing to the periphery; Type 2 (glomeruli form), characterised by prominent anastomoses of vascular calibre; and Type 3 (stick form), characterised by localised, fine, and thin neovascular network without identifiable main feeders or anastomoses. In this manuscript, we named this as NTUH morphologic classification.
As previously reported, Tan et al. divided PCV lesions into three subtypes (types A, B and C) based on BVN characteristics in ICGA and FA images 17 . Type A PCV was characterised by BVN with interconnecting vessels in ICGA without an obvious feeding origin, whereas the other two types of PCV consisted of distinct BVN and feeding vessels on ICGA, without (type B) or with (type C) detectable leakage through FA. Considering that both classification systems focused on the features of BVN in PCV lesions, the authors thus tried to disclosure the associations in both classification systems, if any.

Results
The authors found 78 eyes of 78 patients with PCV between 2015 and 2017, and only 8 patients underwent monotherapy of anti-VEGF and were thus omitted from further analysis according to the predetermined study protocols. Distinct BVN morphological patterns were detected in all the eyes after manual adjustment of segmentation lines on OCTA.
Demographics and the correlation of imaging features on OCTA and ICGA + FA. Of Table 1. The average BCVA at baseline in logarithm of the minimum angle of resolution (LogMAR) was 0.75 ± 0.57, and 34 (48.6%) patients needed re-treatment or rescue therapy within 1 year of the first PDT regimen. Men were more predominant (95.5%) in the type 2 BVN group than in the other two groups (p = 0.007). The mean BCVA at baseline was significantly worse in the type 1 BVN group than in the other two groups (p = 0.036). Regarding OCT structural biomarkers, the SFCT was significantly thick in the type 3 BVN group and thin in the type 1 BVN group (p < 0.001). Moreover, the BVN encompassed area was the largest in type 1 BVN (trunk form) subgroup (p = 0.024) and foveal involvement was seen in all the 30 patients. No statistical discrepancies were observed between the aforementioned three BVN types for CRT, SRF, RPED, or IRC. The multivariate linear regression analysis showed that vision at baseline was significantly related to the BVN morphological type revealed through OCTA and the existence of IRC (p = 0.029 and 0.040, respectively).
In the present study, we found that the NTUH morphological classification of BVN through OCTA and the features of BVN in ICGA + FA proposed by Tan et al. had significant correlations ( Table 2, p = 0.004, chi-square test).
For the majority of patients, types 1, 2, and 3 BVN in OCTA correlated with types C, B, and A in ICGA and FA, respectively (Figs. 1, 2, 3). The BVN types did not convert from one type to another after treatment in our studied cases.

Comparisons of visual outcomes and OCT structural biomarkers.
In the present cohort, all the enrolled patients were followed up for > 12 months after the first treatment. Overall, the BCVA improved from 0.75 ± 0.57 to 0.66 ± 0.60 (logMAR, p < 0.001) at 12 months post-treatment. The comparisons of the BCVA between the three BVN subtypes at each time point are depicted in Fig. 4a. The BCVA significantly improved in type 2 (from 0.67 ± 0.49 to 0.48 ± 0.42 at 12 months post-treatment, logMAR, p < 0.001) and type 3 (from 0.53 ± 0.43 to 0.39 ± 0.30 at 12 months post-treatment, logMAR, p = 0.012). By contrast, the BCVA showed limited improvement in type 1 BVN (from 0.98 ± 0.66 to 0.88 ± 0.69 at 12 months post-treatment, logMAR, p = 0.01). www.nature.com/scientificreports/ The patients with type 3 BVN had the most advantageous BCVA among the three types not only at baseline but also at 12 months post-treatment (p = 0.005). After the first treatment, 34 (48.6%) patients needed re-treatment within 1 year follow-up period ( Table 1). The PDT re-treatment rate was non-significantly higher in patients with type 1 and type 2 BVN (50.0% and 59.1%, respectively, p = 0.263). Among the biomarkers on structural OCT, the SFCT showed a common trend among the different BVN types at baseline and after treatment. The SFCT ultimately remained the lowest, intermediate, and highest in types 1, 2, and 3, respectively, at each time point (p < 0.05 for all, Fig. 4b). Furthermore, the SFCT decreased significantly in each BVN type at 12 months after treatment compared with that at baseline (p < 0.05 for each type). Conversely, the CRT did not significantly differ among the three morphologic types at each time point (Fig. 4c). Nevertheless, the CRT after PDT continued to decline after treatment in all BVN types, and the findings were significant in patients with type 2 BVN (3 months post-treatment compared with baseline, p = 0.026) and type 3 BVN (all time points compared with baseline, p < 0.01 for all).
The pattern of chorioretinal fluid compartments, namely SRF, RPED, and IRC, in different PCV types was thoroughly investigated. Of the 70 eyes studied at baseline, we found 51 (72.9%), 68 (97.1%), and 29 (41.4%) eyes with SRF, RPED, and IRC, respectively. At baseline, SRF was the most prevalent in type 2 BVN (77.3%), whereas RPED and IRC were the most prevalent in type 1 BVN (RPED and IRC: 100% and 50%, respectively). After treatment, the presence of SRF and IRC decreased universally, but no significant difference was observed among the three types at each time point. Multivariate linear regression analysis for visual outcomes at 12 months post-treatment is shown in Table 3. Overall, BVN morphology and the presence of IRC significantly affected the initial visual performance and final visual prognosis.

Discussion
In the present study, the authors reappraised the BVN morphological classification system with an extended observation period and successfully demonstrated that PCV outcomes can be prognosticated using OCTA 27 . We not only extended the follow-up duration to 12 months post-treatment but also investigated the demographic factors, imaging characteristics on OCTA, initial visual acuity, and re-treatment rate of patients with each BVN type. In addition, the correlations of BVN imaging features between OCTA and ICGA + FA were validated.
Owing to more experience with manual segmentation using OCTA, the detection rate of BVN was 100% in our study. Many studies have demonstrated that BVN detection using OCTA might be more sensitive than detection through traditional ICGA 31,32 . Recent efforts were made to diagnose PCV lesions without ICGA. Chaikitmongkol et al. 33 observed that the existence of more than two of four imaging criteria (RPED on colour fundus photography, peaked RPED, notched RPED, and hyperreflective ring with RPED in OCT) could successfully diagnose PCV lesions with satisfactory sensitivity and specificity. In addition, the Asia-Pacific Ocular Imaging Society (APOIS) PCV workgroup reached a universal consensus to diagnose PCV without ICGA. The presence of sub-RPE ring-like structures, complex RPE elevation in enface OCT, and sharp-peaked PED in cross-sectional OCT could measure up to a positive predictive value of 0.93 34 . Additionally, OCT has more advantages, such as dye-free procedure, noninvasive, prompt, and safe, which broadens its use.
To date, the pathophysiology and terminology of PCV lesions are controversial. Although its correlations with type 1 choroidal neovasculopathy and pachychoroid spectrum disease were disclosed by the APOIS PCV workshop consensus 34 , the clinical presentation and prognosis still varied immensely. Many classification systems have been attempted to categorise PCV lesions into clinically relevant groups 10,14,15 . In a novel comparative study, Bo et al. observed 43 polypoidal lesions through ICGA, which all corresponded to the "clusters of tangled vasculature" on OCTA 32 . In addition, all polypoidal lesions decreased in size and complexity after anti-VEGF treatment. These thin-walled, tangled vasculature, which stemmed from the BVN, were assumed to potentially account for the clinical features of polypoidal lesions in PCV 32 . Therefore, the authors speculated that BVN patterns, which are the origins of polypoidal lesions, act as potent indicators of disease behaviour and natural course.
In our current study, we divided PCV lesions into three BVN types based on their morphological patterns with clinical relevance. Patients with type 1 BVN had the worst vision at baseline, and the trend persisted until 12 months post-treatment (Fig. 4a, p < 0.05 at baseline, 6-and 12-months post-treatment). Furthermore, type 3 BVN encompassed the smallest BVN area among the three types and was thus associated with the most favourable visual outcome after treatment 27 .
Considering the biomarkers (SFCT, CRT, SRF, RPED, and IRC) on OCT that aided in the prediction of the treatment outcome, most of our experience stemmed from the neovascular AMD studies 35 www.nature.com/scientificreports/  www.nature.com/scientificreports/ disclosed that pachychoroidal neovasculopathy was associated with decreased VEGF level in the aqueous and better anatomical outcomes compared with neovascular AMD. Hereby, we found that type 3 BVN had significantly thick SFCT (p < 0.001) and less, although not significant, pathologic fluid compartment retention (SRF, RPED, and IRC) at baseline. Multivariate linear regression analysis further confirmed the influences of BVN morphology and IRC on baseline VA in our current study. Tan's classification system for PCV focused on the BVN characteristics in ICGA + FA and was first proposed in 2014 17 . The application and prediction of visual prognosis had been constantly verified in several studies 38,39 . In our study, the authors found a significantly high correlation between the two BVN classification systems (p = 0.004, Table 2). The result was not so surprising because both systems targeted BVN characteristics, which were regarded as the major pathological features in PCV lesions.
The possible explanations for the worst vision and treatment outcomes in type 1 (trunk form) BVN included the following two aspects. First, the authors noted that trunk form BVN encompassed the largest area over macula under OCTA and foveal involvement was almost inevitable in these cases (Table 1, Fig. 1), and thus posed greater impacts on vision. Second, according to Tan's proposal, significant and obvious leakage from type C BVN also contributed to less favourable outcomes. Our study also showed high correlation between type 1   Fig. 1). On the other hand, our study also demonstrated that visual improvements are more significant in type 2 rather than in type 3 BVN at 12 months post-treatment (Fig. 4a). Patients with type 3 BVN had better BCVA at initial presentation, and thus "ceiling effects" might be subsequently encountered as previous literature speculated 39 . The final BCVA at 12 months post-treatment was significantly related to the BVN types and the presence of IRC. Notably, the SFCT at 12 months post-treatment showed borderline significant effect on the final BCVA (p = 0.064, Table 3). The major limitations of the present study are the nature of the retrospective study design and small case number. The patients in our cohort study were enrolled in 2015 to 2017, when intravitreal anti-VEGF monotherapy was still less frequently adopted as a potent treatment option for PCV. To obey the predetermined study protocols and IRB, the authors thus mainly focused on patients who underwent the combination therapy of PDT and anti-VEGF injections to ensure study population homogeneity. Furthermore, we only included patients with polyps within a 3 × 3 mm central macular area. The aforementioned factors led to certain selection bias. Additional studies with larger population cohorts including both combination therapy and anti-VEGF monotherapy along with prospective design are warranted.
In summary, OCTA provided an innovative method of evaluating PCV lesions and their underlying BVN morphology. Moreover, aging is usually accompanied by increased systemic comorbidities, which should be considered during imaging studies. The main advantages of OCTA are its non-invasiveness and lack of systemic toxicity, making it a safe procedure. Through this study, the authors demonstrated that a strong correlation exists in the BVN imaging morphologies between OCTA and ICGA + FA and further supported the notion that BVN plays a major role in the pathophysiology of PCV lesions. Furthermore, we demonstrated that BVN morphology has important clinical implications, which help clinicians successfully predict the final visual prognosis. www.nature.com/scientificreports/