Comparison of quantitative parameters in Diabetics without clinical signs of diabetic retinopathy and healthy individuals using Swept-source Optical Coherence Tomography Angiography

Automated image analysis is the future for retinal imaging in order to early diagnose retinal diseases like diabetic retinopathy. Swept source coherence tomography angiography offers new insights in retinal vascularization. Data sets for each stage of diabetic retinopathy are crucial to compare findings to the healthy population and to develop further algorithms. We recruited 39 eyes of Type-2 diabetics without diabetic retinopathy comparing them to 43 eyes of age and gender matched individuals using swept source coherence tomography angiography. Vessel and perfusion density in the superficial and deep retinal plexus, as well foveal-avascular zone expansion, using Macular Density V 0.7 algorithm were evaluated. We found no significant change in vessel and perfusion density in both plexus, as well no difference in the foveal-avascular zone between healthy and diabetic eyes without clinical diabetic retinopathy. studies.


Introduction
Diabetes mellitus (DM) is a widespread disease affecting 366 million people worldwide 1 . Diabetic retinopathy (DR), a complication of DM, is the leading cause of vision loss amongst working age adults in developed countries 2 .
Optical coherence tomography angiography (OCTA) is a novel imaging technique using decorrelation between resampled images to detect blood-flow and thus, highlight 2-and 3-dimensional reconstructions of the retinal vasculature 3 .
Ultra-high speed swept-source optical coherence tomography (SS-OCT) with its higher wavelength and thus its deeper penetration into tissue coupled with the new modality of OCTA, identifies superficial and deep vascular retinal plexus noninvasively 4,5 .
Automated image analysis is the future of retinal imaging. There are multiple studies using OCT or fundus images and creating algorithms based on classic machine learning or deep neural networks 6 7 . Especially in DR there are multiple algorithms available to detect DR changes 8 .
The same image analysis methods can be used for the analysis of OCTA data. As this technique is rather new, there is less data published compared to conventional OCT. However, it was demonstrated that OCTA can be used for DR differentiation 9 10 . In the future screening methods using artificial intelligence will be of high importance 11 . In order to develop robust algorithms clean data sets are crucial.
Especially in the early stage of DM it is of great interest to investigate retinal microvasculature on its presence of early subclinical changes using automated algorithms. Therefor clean data sets of retinal microvasculature in diabetic patients without clinical signs of DR are needed for different SS-OCTA devices as results 4 regarding quantitative changes in retinal microvasculature are almost impossible to compare between different SS-OCAT devices 12 .
The aim of this study was to examine quantitative parameters of retinal microvasculature in diabetic patients without clinical signs of DR using SS-OCTA and comparing them to healthy volunteers. We present a complete data set of diabetic patients without DR and age-matched healthy controls.

Results
Thirty-nine eyes of 26 patients (48.1 %) with Type-2 diabetes and 43 eyes of 28 (51.9 %) healthy individuals were included. All participants were Caucasian. Table 1 shows demographic and clinical data as well as medication of diabetic patients. Table 2   In this study we compared quantitative parameters of the retinal microvasculature of patients suffering from diabetes without clinical signs of DR to non-diabetic participants using SS-OCTA.
The rational was, if the use the deeply penetrating SS-OCTA was able to reveal first preclinical changes in diabetic patients without DR before clinically mild NPDRP arise. Our study showed no significant difference in PD and VD, neither in the SRL, nor in the DRL between nondiabetic individuals and diabetic patients without clinical DR.
Automated image analysis is the future of retinal imaging. Clean data sets are crucial for different OCTA devices especially in the early stage of the disease to diagnose first microvascular changes before complications arise.
We present a complete and detailed data set of patients with diabetes, without fundus changes, and healthy controls. This data set is a valuable addition to current research, as it can be used as a reference data set for further studies using the SS-OCTA Plex Elite 9000. As results from different SS-OCTA can differ significantly it is important to have this data available to be able to compare results 12 .
Study groups using the same SS-OCTA device as our study but different algorithms present different results among Type-2 and Type-1 diabetic patients without DR.
One group found significant quantitative changes in PD and VD in Type-1 diabetics without DR using 6 x 6 mm OCTA scans and ImageJ software for analysis 13 .
Although the group used the same device it is difficult to compare results because of different image analysis method and scan size. Regarding scan size, a study group found the best predictability for detecting DR using SS-OCTA is using 3 x 3 mm scans 14 . Hence, we assume, scan size could play an important factor. In addition, the higher risk for developing DR earlier in Type-1 diabetics is well known 15  second with an A-scan depth of 3 mm in tissue and an axial resolution of 6.3 μm.
Differences between B-scans are used to generate relative reflectivity changes with 13 time secondary to motion. For each eye, fovea-centered 3 × 3 mm OCTA scans were acquired because of the highest scan quality 14 . Low-quality OCTA images with signal strength ≤ 6 and images with motion artefacts were excluded from the study.
Before starting quantitative analysis, all pictures were anonymized. En-face images of the superficial retinal layer (SRL), deep retinal layer (DRL) and full thickness retina images, using the built-in software segmentation algorithm, were exported. If the algorithm failed, segmentation was done manually. The SRL, extends from the inner limiting membrane to the inner plexiform layer, and the DRL, extends from the inner plexiform layer to the outer plexiform layer. Furthermore, the full-thickness slab is extending from the inner limiting membrane to 70 µm above the retinal pigment epithelium. Figure 1 showes the images of a diabetic patient without DR. Beside these quantitative parameters also the FAZ was investigated with the algorithm mentioned above. We used the FAZ raw size of the SRL to compare it between diabetics and healthy individuals.
The primary outcome measures were the quantitative parameters VD and PD compared between the diabetic and non-diabetic group. Furthermore, FAZ size between both groups was compared.
Statistical analysis was performed using SPSS (version 21.0; SPSS Inc., Chicago, Illinois, USA). Data were confirmed to be normally distributed using Shapiro-Wilk test. To compare the data an independent, two-sampled Student´s T-Test was applied. All 95%-Confidence intervals (95% CI) and p -values are two-sided. Data were assumed significant with a two-sided p < 0.05.