Adaptive contour prediction in postoperative OCT imaging using domain-adaptive generative adversarial networks

Optical Coherence Tomography (OCT) imaging plays a pivotal role in diagnosing ophthalmic diseases. This study introduces a domain-adaptive generative adversarial network (DaCGAN) to predict postoperative OCT contours from preoperative images, addressing the challenges of limited paired data and incomplete contour structures. DaCGAN integrates data from different diseases and employs preoperative contours to enhance prediction accuracy. Experimental results on Diabetic Macular Edema and Retinal Vein Occlusion datasets demonstrate the superiority of DaCGAN over existing methods, with improvements in SSIM, MSE, and PSNR metrics. Our approach offers a cost-effective solution for postoperative image prediction, potentially benefiting clinical diagnosis and treatment planning.