Predicting OCT Images Of Short-term Response To Anti-VEGF Treatment For Retinal Vein Occlusion Using Generative Adversarial Network

Objective: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(RVO)to anti-vascular endothelial growth factor(anti-VEGF)therapy.Methods:Real-world imaging data were retrospectively collected from 1 May 2017,to 31 December 2022.A total of 515 pairs of pre-and post-therapeutic OCT images of patients with RVO were included in the training set,while 68 pre-and post-therapeutic OCT images were included in the validation set.A pix2pixHD method was adopted to predict posttherapeutic OCT images in RVO patients after anti-VEGF therapy.The quality and similarity of synthetic OCT images were evaluated by screening and evaluation experiments.We quantitatively and qualitatively assessed the prognostic accuracy of the synthetic posttherapeutic OCT images.In the optimization model,we retrospectively analyzed paired OCT images from patients with RVO-ME at baseline and 1 month after treatment,the contours of crucial retinal structures on post-therapeutic OCT images were annotated and extracted by LabelMe and OpenCV,respectively.We trained the GAN model using 622 pairs of images to synthesize the postoperative OCT contours based on the preoperative images.The average retinal thickness at the fovea,0.5mm,1.5mm,3mm away from nasal and 0.5mm,1.5mm,3mm away from temporal of synthetic and real post-therapeutic contours in the testing set were measured and compared by paired t test to quantitatively evaluate the model’s predictive performance.Result:The post-therapeutic OCT images generated by the pix2pixHD algorithm were comparable to the actual images in edema resorption response.Retinal specialists found most synthetic images(63/68)difficult to differentiate from the real ones.The mean absolute error(MAE)of the central macular thickness(CMT)between the synthetic and real OCT images was 26.33 ±15.81 μm.There was no statistical difference in CMT between the synthetic and the real images.In the contours GAN algorithm,there were no significant differences in retinal thickness at the central fovea,0.5mm,1.5mm,3mm nasally and 0.5 mm,1.5mm,and 3mm temporally to the fovea between the real and synthetic post-therapeutic contours(p>0.05)The MAE values were 73.97μm,60.48μm,42.36μm,48.76μm,60.13μm,45.00μm and 37.98 μm,respectively.The accuracy of synthetic contours in predicting the degree of regression of individualized macular edema after therapy was 92.5%.Conclusion:in this study,pix2pixHD model was applied to accurately predict the shortterm efficacy of anti-VEGF treatment in RVO patients based on OCT images,which is conducive to objective assessment of patients’ conditions and prognosis,and to identify individuals who may not be sensitive to anti-VEGF treatment in advance,so as to assist clinicians to adjust the therapies for patients with RVO.