Efficacy And Ai Prediction Of Anti-VEGF For Different Types Of Macular Edema In Retinal Vein Occlusion

Objective:To analyze the effect of the first anti-Vascular endothelial growth factor(VEGF)treatment on OCT images of different types of macular edema in Retinal vein occlusion(RVO).To generate and evaluate post-therapeutic optical coherence tomography(OCT)images based on pre-therapeutic images with generative adversarial network(GAN)to predict the short-term response of patients with retinal vein occlusion to anti-vascular endothelial growth factor therapy.Methods:This study retrospectively included 139 patients who first received anti-VEGF injections for retinal vein obstruction at the Yijishan Hospital affiliated to Wannan Medical College from December 1,2017,to March 31,2022.The disease duration,age,hypertension,OCT images,central macular thickness(CMT),and best corrected visual acuity(BCVA)before and after 4-6 weeks of treatment were collected.Patients were divided into four groups according to OCT images: Cystoid macular edema(CME),Sponge-like diffuse retinal thickening(SDRT),Serous retinal detachment(SRD),and mixed type(FULL).The changes in CMT and visual acuity before and after treatment were compared among the groups to analyze the difference in the effect of different edema types on anti-VEGF treatment and the effect of baseline CMT and visual acuity on post-treatment visual acuity.SPSS 26.0 was used for statistical analysis.The measures were expressed as mean ± standard deviation.Statistical differences between clinical data before and after anti-VEGF treatment were assessed using the Wilcoxon test.Differences between the four groups were analyzed by one-way ANOVA if the variances were equal and by the Kruskal-Wallis test if the variances were not equal.Correlations between pre-treatment CMT and visual acuity and post-treatment visual acuity were analyzed using Pearson correlation coefficients.190 pairs of images were preprocessed,139 pairs of images were randomly selected for data augmentation to be included in the training set and test set,and 51 pairs of images were included in the validation set.The Pix2 pix HD method was used to predict OCT images of patients with RVO after anti-VEGF treatment,and the prediction effect was evaluated by qualitative and quantitative experiments,and the quality of OCT image synthesis for different edema types was analyzed.Accuracy and 95% confidence interval(CI,95% CI)were used for categorical variables,and measures were expressed as mean ± standard deviation,and mean absolute error(MAE)was used as an index for quantitative evaluation of the synthesized images.Results:139 pairs were classified for macular edema,including 54 pairs in the CME group,23 pairs in the SDRT group,22 pairs in the SDR group,and 40 pairs in the FULL group.There was no significant difference in the duration of disease or age between the groups(P=0.92,P=0.61).There was a significant difference in preoperative CMT between groups(P=0.01,one-way ANOVA),with the CMT in the FULL group being significantly greater than that in the SDRT group(P=0.03).There was no significant difference in pre-treatment visual acuity between the four groups(P=0.26),and after anti-VEGF treatment,the macular central recess thickness was reduced and visual acuity was improved in all four groups,among which the CMT in the CME and FULL groups was reduced significantly(P<0.05)compared with the other two groups.The postoperative visual acuity was negatively correlated with preoperative CMT(P=0.044)and positively correlated with preoperative visual acuity(P<0.01).139 pairs of OCT images were included in the prediction model and expanded to2538 pre-treatment and 2514 post-treatment images for training,and 51 pairs for model validation.The post-treatment OCT images generated using the Pix2 pix HD algorithm were similar to the real images in terms of edema regression performance.The probabilities of correctly determining the true images were 0.51 and 0.31 for the two fluorophores,with 95% confidence intervals of(0.368,0.652)and(0.182,0.446),respectively,51 images generated by the prediction model in the validation group had a MAE of 106.22 ± 95.95 μm,and subgroup analysis based on different macular edema classifications showed that images from macular cystoid edema,diffuse retinal edema,placoid retinal detachment,and mixed patients had MAE of 105.71 ± 96.33 μm,97.27 ±88.68 μm,141.77 ± 121.26 μm,and 82.23 ± 85.59 μm for the post-treatment OCT images,respectively.Conclusion:The efficacy of anti-VEGF in the treatment of RVO macular edema may be related to the type of edema under OCT images,in which the efficacy is best in patients with CME,but poor in patients with SDRT and FULL.The generative adversarial network technique was able to generate OCT images of RVO-ME patients after anti-VEGF treatment,and the predictive effect of different types of macular edema in OCT images varied.