OCT Signal Enhancement with Deep Learning

To establish whether deep learning methods are able to improve the signal-to-noise ratio of time-domain (TD) OCT images approach that spectral-domain (SD) images. Method agreement study and progression detection in a randomized, double-masked, placebo-controlled, multicenter trial for open-angle glaucoma (OAG), United Kingdom Glaucoma Treatment Study (UKGTS). The training validation cohort comprised 77 stable OAG participants with TD SD imaging at up 11 visits within 3 months. testing 284 newly diagnosed patients from 516 recruited 10 centers between 2007 2010. An ensemble generative adversarial networks (GANs) was trained on image pairs dataset applied dataset. Time-domain were converted synthesized segmented via Bayesian fusion output GANs. Bland-Altman analysis assessed average retinal nerve fiber layer thickness (RNFLT) measurements RNFLT. Analysis distribution rates RNFLT change two treatment arms UKGTS compared. A Cox model predictors time-to-incident visual field (VF) computed 95% limits 26.64 –22.95; 8.11 –6.73; 4.16 –4.04. mean difference rate placebo 0.24 (P = 0.11) 0.43 0.0017). hazard slope regression modeling time incident VF 1.09 (95% confidence interval [CI], 1.02–1.21; P 0.035) 1.24 CI, 1.08–1.39; 0.011) OCT. Image enhancement significantly improved measurements. difference, its significance, enhanced became stronger predictor progression.