Prediction Of Short-term OCT Image Changes After Anti-VEGF Therapy In Neovascular AMD Patients Using Generative Adversarial Networks

Aims:Age-related macular degeneration(AMD)is one of the main causes of irreversible vision loss in people over 50 years old.Intravitreal injection of anti-vascular endothelial growth factor(VEGF)is the first line strategy for neovascular AMD(nAMD),and has brought revolutionary changes to the current situation of nAMD treatment.However,there exist significant individual differences in terms of treatment efficacy.Artificial intelligence(AI)method has powerful information processing and data mining ability.The introduction of AI into researches of fundus diseases has brought new ideas and methods for traditional clinical research.The aim of this study was to generate and evaluate individualized post-therapeutic optical coherence tomography(OCT)images that could predict the short-term response of anti-VEGF therapy for typical nAMD patients based on pre-therapeutic images using generative adversarial network(GAN).Methods:We retrospectively reviewed the records of patients with nAMD who underwent an intravitreal injection of anti-VEGF drugs at the Depatment of Ophthalmology of Peking Union Medical College Hospital from November 1st 2018 to June 30th 2019.Pre-and post-therapeutic OCT images were collected.Patient selection were based on the inclusion and exclusion criteria.Then,pairs of pre-and post-therapeutic images were screened and matched according to the position of macula and intraretinal arteries,and were assigned into training set and test set at the ratio of about 10:1.Finally,a total of 524 pairs of pre-and post-therapeutic OCT images of nAMD patients were included in training set,while 58 pre-therapeutic OCT images were included in the tests set,and their corresponding post-therapeutic OCT images were used to evaluate the synthetic images.The process of model training was divided into unlabeled and labeled stages,and the labeled stage was further divided into training with original OCT images and training without original OCT images.During the labeled stage,the input information of the model included both original OCT images and the segmented retinal structures.The pix2pixHD method was adopted for image synthesis.Three experiments were performed to evaluate the quality,authenticity and predictive power of the synthetic images by retinal specialists.Results:Results of three different training stages improved successively.In terms of the training process with both lesion segmentation and original OCT image,100%of the synthetic OCT images had sufficient quality for further clinical interpretation.The rate to correctly identify synthetic images was only 12-16%,which indicated that the authenticity of synthetic images was good.The accuracy to predict macular status of wet or dry was 0.81(95%CI 0.71-0.91)with specificity of 0.97(95%CI 0.80-1.00).Nearly 80%of the treatment-related macular status transformation could be predicted.Wet macula was defined as the existence of intraretinal cystoid fluid or subretinal fluid.Conclusion:Our results revealed that GAN can generate post-therapeutic OCT images for nAMD patients based on pre-therapeutic OCT images with both good quality and high predictive accuracy.Thus,this model has great potential in predicting the treatment effect of anti-VEGF therapy and could assist with clinical decision making.