Early Detection And Artificial Intelligence Screening Research For Diabetic Retinopathy

Part ?.Ultra-widefield Optical Coherence Tomography Angiography in Preclinical Diabetic RetinopathyPurpose:To investigate alterations in retinal microvasculature in eyes with preclinical diabetic retinopathy(PCDR)using ultra-widefield swept-source optical coherence tomography an giography(UWF-SS-OCTA).Methods:Prospective cross-sectional study.Fifty-five eyes of thirty diabetic patients without clinical retinal signs were included.All subjects underwent OCTA examination with a 12 mm×12 mm field of view of 5 visual fixations(1 central fixation and 4 peripheral fixations)to compose a UWF-OCTA image.In the UWF images,the central area corresponded to the original central image obtained using central fixation,and the peripheral area was the remaining area.Lesions,including nonperfusion areas(NPAs),microvascular tortuosity and dilation,and neovascularization(NV),were recorded in different areas.Diabetes history and laboratory test results were also recorded.Results:UWF-OCTA revealed retinal microvascular lesions in thirty-two eyes(58.2%).NPAs were more frequently detected in the peripheral areas than in the central areas,but the difference was nonsignificant(P=0.085).Peripheral areas presented significantly more microvascular dilation and tortuosity than central areas(P=0.024).The number of lesion types was associated with HbAlc levels in the peripheral and overall areas(all P values<0.001).Conclusions:UWF-OCTA is a promising imaging method for detecting vascular alterations in diabetic eyes without clinical signs to reveal retinal microvascular alterations.These alterations were correlated with systemic conditions.Part ?.Quantitative Evaluation of Microvascular Changes Using Widefield Optical Coherence Tomography Angiography in Preclinical Diabetic RetinopathyPurpose:To quantitatively analyze the changes in microcirculation in the eyes of type 2 diabetes patients with preclinical diabetic retinopathy(PCDR)using widefield optical coherence tomography angiography(WF-OCTA).Methods:A prospective,cross-sectional,comparative study.PCDR patients and healthy control subjects underwent standard clinical examinations and WF-OCTA imaging using 12 mm × 12 mm scans to examine the vessel density(VD)of the superficial capillary plexus(SCP)and deep capillary plexus(DCP).Regular 3 mm × 3 mm scans were also performed to analyze the foveal avascular zone(FAZ)area,perimeter,acircularity index and foveal density(FD-300)of the inner retina.Results:Fifty eyes from twenty-five PCDR patients and fifty eyes from twenty-five controls were included.No significant difference was detected in the VD of the SCP or FAZ parameters,including the area,perimeter,acircularity index of the FAZ and the FD-300,between the two groups(all P>0.05).The deep temporal VDs of the parafoveal,perifoveal,annular1-6,annular1-9,annular1-12,annular3-9,annular3-12,annular6-9 and annular6-12 regions were all lower in the NDR patients than in the controls(all P<0.05).There were also statistically significant differences between the groups for the deep VD of subfield0-3 and for the average,inferior,nasal parafoveal VD of the DCP(all P<0.05).Conclusions:Retinal microvascular impairments in type 2 diabetes patients with PCDR might start from the temporal region of the DCP of the posterior pole and can be detected early by WF-OCTA.Part ?.Artificial Intelligence Screens Optical Coherence Tomography-Derived Diabetic Macular Edema from Color Fundus Photographs:A Pilot StudyPurpose:To develop artificial intelligence(AI)models for screening diabetic macular edema(DME)from monoscope color fundus Photographs(CFP)based on the gold standard of optical coherence tomography(OCT)retinal thickness maps.Methods:507 CFPs of diabetic patients with the corresponding OCT retinal thickness maps were collected retrospectively,and divided into training set(294 cases),validation set(116 cases)and test set(97 cases).The corresponding OCT retinal thickness maps were used as the gold standard for DME diagnosis and grading in CFPs.The one-step whole image recognition and improved one-step macular recognition,two-step whole image recognition methods were used to train the Efficientnet_B3_Pruned deep learning model to detect DME based on the monoscope CFPs.Then,the detection efficacy of AI models and three attending physicians were evaluated respectively according to the test set.Results:In the overall results of ternary classifications of no DME,no center-involved DME(Nci-DME)and center-involved DME(ci-DME),the accuracy and Cohen's kappa value of the one-step whole image model were higher than those of the one-step macular model and two-step whole image model(accuracy:0.66 vs 0.59,0.60;Cohen's kappa:0.47 vs 0.39,0.40).Its accuracy,average F1 and Cohen's kappa value were also superior to the average level of the three physicians(accuracy:0.66 vs 0.58,average F1:0.68 vs 0.61,Cohen's kappa value:0.47 vs 0.34),When performing binary classifications,the accuracy,sensitivity,specificity,positive predictive value,negative predictive value,F1 and Cohen's kappa value of the one-step whole image recognition model were not lower than the average level of the three physicians.Conclusions:The DME-AI screening model based on the OCT gold standard and CFPs could effectively identify DME from the monoscope CFPs,and its detection efficacy was not inferior to that of fundus attending physicians.