Combined Wide-Field Imaging in Grading Diabetic Retinopathy

Objectives: To detect retinal neovascularization elsewhere (NVE), of the optic disc (NVD) and intraretinal microvascular abnormalities (IRMA) in treatment naïve diabetic retinopathy (DR) and compare these �ndings by using 90° Wide-Field Color Fundus Photography (WF CFP), Wide-Field Spectral-Domain Optical Coherence Tomography Angiography (OCTA) and the combination of WF CFP and OCTA through overlay software. Methods: Patients with treatment naïve severe non-proliferative DR or proliferative DR were prospectively enrolled. All patients underwent WF-CFP and OCTA in the same day. Two readers independently analysed WF-CFP, SD-OCTA and the overlay of the two techniques. The degree of agreement between the two raters and between different techniques (WF CFP, OCTA, WF CFP combined to OCTA) were measured with Cohen’s Kappa coe�cient. Results: Thirty-one eyes from 21 patients (10 males, mean age 63 ± 15 years) were included. Inter-rater agreement by using WF-CFP in detection of NVE, NVD and IRMA was respectively 0.62, 0.22 and 0.55. OCTA scored values of inter-rater agreement of 0.86, 0.87 and 0.92 in detection of NVE, NVD and IRMA, respectively. By combining WF-CFP and SD-OCTA, inter-rater agreement in detection of NVE, NVD and IRMA was 0.93, 0.94 and 0.89, respectively. Conclusion: Inter-rater agreement in detection of NVE, NVD and IRMA was substantial, fair and moderate, respectively. OCTA provided almost perfect values of inter-rater agreement in NVE, NVD and IRMA detection. Combining WF-CFP and OCTA further empowered concordance values in detection of NVE and NVD. Combining OCTA and WF-CFP is the best performance to detect NVE and NVD.


Introduction
Early detection and treatment of different grades of diabetic retinopathy (DR) are widely recognized as fundamental goals to reduce visual impairment in patients with diabetes. 1,2 Consequently, a standardized and systematic classi cation for severity grading of DR is essential for therapeutic decisional process.
According to the Early Treatment Diabetic Retinopathy Study (ETDRS) ndings and the International Clinical Disease Severity Scale for DR, severity grading is univocally established based on color fundus photography (CFP) and direct fundus examination. 3 Although this international DR scale is widely recognized and simple to use in common clinical practice, there is an increasing need to implement fundus ndings with multimodal imaging examinations for therapeutic decision-making and DR severity grading. [4][5][6] Among the landscape of lesions and ndings of diabetic retinopathy, intraretinal microvascular abnormalities (IRMA) showed a fair inter-rater agreement at color fundus photography. 7 The complete agreement in IRMA detection between two independent ophthalmologists has been assessed at 51.5% (weighted kappa statistics = 0.49). 7 Concurrently, the agreement in the detection at CFP of retinal neovascularization elsewhere (NVE) was considered as substantial, while the agreement in the detection of retinal neovascularization of the optic disc (NVD) yielded less satisfactory results and has been ranked as moderate (weighted kappa statistics = 0.48). 7 As the severity of IRMA and retinal neovascularization is considered by ETDRS as an important factor for photocoagulation treatment strategy choice, it is fundamental to supplement these ndings with multimodal imaging techniques. 8 Particularly, spectral domain optical coherence tomography angiography (SD-OCTA) showed a higher detection rate of IRMA, compared to color fundus photography grading. In addition, structural spectral-domain optical coherence tomography (SD-OCT) allowed to a better characterization and differentiation of IRMA and NVE. 9 An overlay between CFP and angiographic data (SD-OCTA) allows a more comprehensive analysis of the landscape of pathological ndings in DR. Matching anatomical (CFP) and functional (SD-OCTA) data could improve both the detection rate and the inter-rater agreement of IRMA and retinal neovascularization indeed.
Purpose of the study is to detect intraretinal microvascular abnormalities, retinal neovascularization elsewhere and retinal neovascularization of the optic disc in treatment naïve diabetic retinopathy and to compare these ndings by using We included eyes with severe non-proliferative diabetic retinopathy (SNPDR) or proliferative diabetic retinopathy (PDR), according to the International Clinical Diabetic Retinopathy and Diabetic Macular Edema Disease Severity Scales. 3 Both patients with type 1 and type 2 diabetes mellitus were enrolled in the study. Exclusion criteria were prior laser or intravitreal treatments in the included eye, any retinal disease than diabetic retinopathy, DR related complications impeding retinal imaging (vitreous haemorrhage, retinal detachment, corneal oedema) and optic media opacities limiting image quality.
We collected age, gender, medical and ocular history. Each included eye underwent WF-CFP centred on the fovea and SD-OCTA, with a scan eld of 12x12 mm 2 centred on the fovea. WF-CFP was performed with Clarus® (ZEISS, Oberkochen, Germany). At fundus color photography, we de ned NVE as neovascularization that are on the surface of the retina or further forward in the vitreous cavity, except for those on the disc or within 1 optic disc diameter of its margin. 7 ( Figure 1A) Conversely, a neovascularization that ful l the abovementioned criteria, located on the optic disc or within 1 optic disc diameter, was indicated as NVD. (Figure 2A) IRMA were de ned as tortuous intraretinal vascular segments of variable diameter (within ¼ the width of a major vein at the disc margin). 7 (Figure 3A) SD-OCTA was performed with Cirrus 6000® (ZEISS, Oberkochen, Germany) and examined with Angioplex ® OCTA software (ver. 11.5.2, ZEISS, Oberkochen, Germany). SD-OCTA slabs were automatically segmented by SD-OCTA software and independently manually adjusted by two ophthalmologists (MM and RS). Vitreoretinal interface (VRI) slab, de ned as the region 10 to 300 µm above the internal limiting membrane (ILM), was selected to detect NVE and NVD ( Figure 1B; Figure 2B). Each suspected neovascularization was con rmed at B-scan as extraretinal proliferation. Super cial capillary plexus (SCP) slab was de ned as the region between the inner nuclear layer (INL) and the ILM and selected to detect IRMA, identi ed as dilated terminal vessels adjacent to areas of capillary loss. 10
Inter-rater agreement by using WF-CFP in detection of NVE, (Figure 1 Table 2A and 2B, respectively.

Discussion
Grading DR is an essential step for therapeutic decisional process. 1 Determining the presence of neovascularization and IRMA is still a challenge, since inter-rater agreement values in detection of these lesions do not provide adequate results. 7 OCTA could contribute in better characterization and detection of IRMA and neovascularization. We investigated whether combining anatomical data from WF-CFP and functional ow data from OCTA could improve the rate and the inter-rater agreement in detection of IRMA, NVE and NVD.
We found substantial (0. Distinguishing neovascularization from other ndings related to DR represents a challenge for non-retina specialists. 12 Vascular abnormalities or haemorrhages are often confused with neovascularization, and concurrently neovascularization is thin, slightly coloured and is frequently missed when examined from a non-retina specialist ophthalmologist. Particularly, NVD detection could represent a challenge due to the underlying vascular network in the optic disc and explain the low value of inter-rater agreement at WF-CFP. 13 A-scan and B-scan OCTA helps in detection of NVE and NVD; ophthalmologists are able to localize whether the lesion is inside the vitreoretinal interface or not, distinguishing neovascularization from IRMA, and to characterize the presence of ow inside the abovementioned lesions. In our experience, this tool has improved inter-rater agreement in detection of both NVE and NVD. On the other hand, OCTA alone does not provide anatomical data and wrong segmented layer or artefacts could lead to misinterpreted data. Combining OCTA and WF-CFP with the overlay software allows better characterization of the lesions, since complementary data are merged, ow signal at B-scan can be analysed in the counterpart WF-CFP, and artefacts or wrong segmented layers can be better highlighted. As a matter of fact, WF-CFP combined with OCTA provided better values of inter-rater agreement in NVE and NVD detection. IRMA detection at WF-CFP disclosed poor inter-rater agreement (0.22). Since IRMA appear as abnormal branching or dilation of existing blood vessels in low-blood supply retinal area, they can be often confused with neovascularization or misinterpreted. OCTA highlights their tortuosity and their location in non-perfusion areas, and most importantly they can be easily distinguished from neovascularization, since IRMA localize in the super cial capillary plexus. 14 According to our results, the combination of OCTA and WF-CFP did not improve inter-rater agreement.
The main limitation of the study is the relatively small sample size. In addition, inter-rater agreement of the considered imaging techniques could have been explored between ophthalmologists of different level of experience. Indeed, young ophthalmologists could eventually bene t the most from combining OCTA and WF-CFP.
In conclusion, combining OCTA and WF-CFP displayed good concordance values in detection of NVD and NVE, compared to OCTA. In addition, the overlay between the two techniques has shown better concordance values, particularly in NVE and NVD detection, as previously discussed. Thus, this tool could help ophthalmologists improving detection of clinical ndings of diabetic retinopathy, particularly NVD and NVE. Further studies are required to con rm whether combining OCTA and WF-CFP could help detecting other ndings (i.e. microaneurysms), and whether even retina specialists could improve their skills.