Application Of Molecular And Imaging Markers In Diabetic Retinopathy

Chapter 1 Study on molecular markers of diabetic retinopathy and its correlation with renal functionPurpose:To analyze the levels of B-cell-produced antibodies in the vitreous humor of patients with or without diabetic retinopathy(DR)both qualitatively and quantitatively,and to And explore their correlation with renal function.Methods:A total of 52 type 2 diabetes mellitus(T2DM)with DR patients,and 52 control subjects without diabetes mellitus or inflammatory diseases were included in this prospective study.The levels of immunoglobulin(Ig)A,IgM and IgG subtypes were measured using magnetic color-bead-based multiplex assay.Results:The concentrations of IgA,IgM and total antibodies in the DR group were significantly higher than those in the control group(all p<0.001),but there was no significant difference in the 4 IgG subtypes between the two groups after Bonferroni correction.Pearson’s correlation analysis revealed low negative correlations between levels of antibodies(IgA,IgM)and estimate glomerular filtration rate(eGFR,r=-0.443,r=-0.377,respectively,both p<0.05).Furthermore,multiple linear regression analysis yielded three equations to predict the concentrations of IgA,IgM and total antibodies in the vitreous humor according to eGFR and other clinical variables(r=0.542,r=0.461 and r=0.312,respectively,all p<0.05).Conclusions:Increased levels of IgA,IgM and total antibodies produced by B cells were observed in the vitreous humor of T2DM patients with DR.There were low negative correlations between levels of antibodies(IgA,IgM)and eGFR.Chapter 2 Automatic prediction of diabetic kidney disease via retinal photographs marker of diabetic retinopathyPurpose:Screening for diabetic kidney disease(DKD)is a challenge in diabetes populations,even in high-income countries.Severity of DKD has been shown to be associated with retinal status.Thus,this study aimed to determine DKD grading according to retinal photographs.Methods:In this four-center,cross-sectional study,a deep learning(DL)system was developed for automated grading of DKD.Ground truth labels of DKD grading were determined according to the urine albumin to creatinine ratio(UACR;stage 1,UACR<30mg/g;stage 2,UACR 30-300mg/g;stage 3,UACR≥300mg/g)and estimated glomerular filtration rate(eGFR;stage 1,eGFR≥45 ml/min/1.73m2;stage 2,eGFR 15-45 ml/min/1.73m2;stage 3,eGFR<15 ml/min/1.73m2).The DL model was trained and internally validated with 4898 retinal photographs from two ophthalmic centers in Guangzhou.A total of 1550 photographs from Zhujiang Hospital(ZJH)and 171 photographs from the First Affiliated Hospital of Kunming Medical University(FAHKMU)were used to as external validation set.An area under the receiver operating characteristic curve(AUC)was used to evaluate the performance of the DL system,while heatmaps were applied to visualize features critical for accurate grading of the DL system.Results:In UACR grading,the AUC in internal validation ranged from 0.926 to 0.948 for stage 1,0.879 to 0.900 for stage 2,and 0.932 to 0.969 for stage 3.The AUC in external validation was 0.857,0.810 and 0.892 in ZJH,and was 0.848,0.806 and 0.877 in FAHKMU,respectively.In eGFR grading,the AUC in internal validation ranged from 0.823 to 0.852 for stage 1,0.816 to 0.847 for stage 2,and 0.788 to 0.938 for stage 3.The AUC in external validation was 0.792,0.733 and 0.801 in ZJH,and was 0.751,0.710 and 0.752 in FAHKMU,respectively.Heatmaps showed that the DL system achieved correct grading based on diabetic retinal changes(e.g.,hard exudate,soft exudate,retinal hemorrhage and proliferative epiretinal membrane).Conclusions:Using multi-center datasets,our DL system demonstrated good accuracy and transparency in the automatic grading of DKD based on retinal photographs.The DL system may be added to the current DKD screening systems.Chapter 3 Intelligent prediction of the efficacy of diabetic macular edema treatment based on optical coherence tomography image markerPurpose:This study aimed to predict the treatment outcomes in patients with diabetic macular edema(DME)after 3 monthly anti-vascular endothelial growth factor(VEGF)injections using machine learning(ML)based on pretreatment optical coherence tomography(OCT)images and clinical variables.Methods:An ensemble ML system consisting of four deep learning(DL)models and five classical ML(CML)models was developed to predict the posttreatment central foveal thickness(CFT)and the best-corrected visual acuity(BCVA).A total of 363 OCT images and 7,587 clinical data records from 363 eyes were included in the training set(304 eyes)and external validation set(59 eyes).The DL models were trained using the OCT images,and the CML models were trained using the OCT images features and clinical variables.The predictive posttreatment CFT and BCVA values were compared with true outcomes obtained from the medical records.Results:For CFT prediction,the mean absolute error(MAE),root mean square error(RMSE),and R2 of the best-performing model in the training set was 66.59,93.73,and 0.71,respectively,with an area under receiver operating characteristic curve(AUC)of 0.90 for distinguishing the eyes with good anatomical response.The MAE,RMSE,and R2 was 68.08,97.63,and 0.74,respectively,with an AUC of 0.94 in the external validation set.For BCVA prediction,the MAE,RMSE,and R2 of the best-performing model in the training set was 0.19,0.29,and 0.60,respectively,with an AUC of 0.80 for distinguishing eyes with a good functional response.The external validation achieved a MAE,RMSE,and R2 of 0.13,0.20,and 0.68,respectively,with an AUC of 0.81.Conclusions:Our ensemble ML system accurately predicted posttreatment CFT and BCVA after anti-VEGF injections in DME patients,and can be used to prospectively assess the efficacy of anti-VEGF therapy in DME patients.