Comparison of fundus fluorescein angiography and optical coherence tomography angiography findings of retinal ischemia with the assistance of artificial intelligence in patients with diabetic retinopathy

Purpose: The aim of the study is to evaluate the presence of ischemia in optical coherence tomography angiography (OCTA) images corresponding to the aneurysm areas determined in fundus fluorescein angiography (FFA) in diabetic retinopathy (DR) patients, to compare whether there are differences between superficial and deep capillary plexuses, and to investigate the relationship between aneurysm development and ischemia, to provide significant advantages in diagnosis and treatment by making analyzes faster, more accurate and more efficient with the use of artificial intelligence (AI) technologies in the evaluation of advanced retinal imaging methods such as FFA and OCTA. Materials and Methods: In this study, cases who underwent FFA and OCTA with the diagnosis of diabetic retinopathy at Karadeniz Technical University Faculty of Medicine, Department of Ophthalmology, Retina Unit between March 2023 and September 2023 were included. Patients whose age range was compatible with the study group and in whom no pathology was detected after detailed examinations and tests were included in the control group of the study. The data of all patients included in the study were examined retrospectively and 6*6 mm² OCT-Angiography images were obtained on the Solix OCT/OCTA (Optovue Inc, Freemont, CA, USA) device and fundus and fluorescein angiography on the Optos (California, Optos plc, UK) device. In OCTA images, total vessel densities in the superficial and deep capillary plexuses in the foveal and parafoveal regions of 6*6 mm² and grid-based vessel densities in the areas of the same region divided into 9 equal parts were automatically measured by the software of the device. In the same device, the FAZ area and the vessel density in the foveal and parafoveal choriocapillary regions were also automatically measured by the software of the device. The 6*6 mm² area in the OCTA image was placed exactly over the same area in the FFA images using software. Ischemic areas in FFA were marked manually by a single researcher, and area dimensions were determined with the software of the Optos device. The number of aneurysms in the units divided into 9 areas in OCTA was counted in FFA images. The ischemic areas in FFA and the vascular density in both superficial and deep plexuses in OCTA were statistically compared. In addition, correlations between vessel densities in areas with aneurysms and the number of aneurysms, and vessel densities in areas with and without aneurysms were compared. Affine transform technique which is one of the image fusion techniques, was used to ensure that FFA and OCTA images were geometrically aligned and merged onto FFA images. Results: When we compared patients with diabetic retinopathy with the control group, it was observed that foveal and parafoveal vascular densities were lower in SCP and DCP both in total and in 9 subregions (p<0,05). In the DR group, vessel density in areas other than superonasal and inferonasal was greater in DCP than in SCP and there was a statistically significant difference (p<0,05). DCP vessel density was significantly smaller in superonasal and inferonasal areas. Although the foveal avascular zone (FAZ, p=0.464) was enlarged in the DR group compared to the control group, no statistical difference was observed. However, the flow area (p=0.000) was significantly lower in the DR group. When the ischemic area with and without aneurysm measured by FFA was compared, ischemia in the area with aneurysm was significantly greater. There was a strong positive correlation between the number of aneurysms and the extent of ischemia. There was a positive correlation between ischemic area and FAZ area, a negative correlation between vessel density in DCP and SCP, and a negative correlation between FAZ area and DCP and SCP. No significant correlation was found between the number of aneurysms and DCP and SCP vessel density. No significant correlation was found between the age, gender and duration of DM of the patients and the ischemic area, number of aneurysms, FAZ area, DCP and SCP vascular density. Artificial intelligence has been used to automatically adjust affine transform parameters. In this way, after AI was used in the SCP, where the clearest vascular structure can be seen from the OCTA layers, it automatically enabled the same fusion process to be performed in the DCP, outer retina, and choriocapillaris, even if the vascular structure was not visible. Discussion and Conclusion: With increasing ischemia in diabetic retinopathy, vascular densities in the superficial and deep plexuses, flow in the choriocapillaris and the FAZ area are also affected. Additionally, the superonasal and inferonasal regions in the deep capillary plexus decreased more than the superficial plexus. Ischemia in the area with aneurysm is higher than ischemia in the area without aneurysm. According to our results, OCTA and FFA can be used routinely in monitoring DR in diabetic eyes. Different vascular layers in FFA and OCTA can be transformed only on the FFA of SCP with artificial intelligence, and thus, fusion can be performed automatically in deeper layers such as DCP, outer retina and choriocapillary layer, even if the vascular structure is not visible. Key words: Affine Transformation, Aneurysm, Deep Capillary Plexus, Diabetic Retinopathy, FAZ Area, Ischemia, Leakage, OCTA, Superficial Capillary Plexus, Choriocapillary Flow Area, Artificial Intelligence