Robust Retinal Layer Segmentation Using OCT B-Scans: A Novel Approach Based on Pix2Pix Generative Adversarial Network

Various eye diseases, such as age-related macular degeneration (AMD), are caused due to the structural dysfunction of the retinal sublayers. In the current clinical practice, clinicians screen ubiquitous optical coherence tomography (OCT) scans to visualize the retina's structural changes and diagnose the associated diseases. Quantitative assessment of such structural changes assumes significance in accurate disease management and treatment response monitoring. To this end, various attempts based on image processing, machine learning, and deep learning methods have been reported. However, they have not fully leveraged the unique structural characteristics of retinal sublayers in OCT B-Scans. In this work, we proposed a Pix2Pix-GAN-based algorithm capable of synthesizing look-alike target images from source images by mapping pixel-wise structural information. In particular, we envisioned the Pix2Pix network to synthesize OCT images with each boundary of interest labeled in the same color as the corresponding boundary in ground-truth images. On a dataset of 180 EDI OCT B-Scans with accurate ground-truth segmentation for six retinal sublayer boundaries, we trained and tested the Pix2Pix-GAN model. In particular, we achieved an overall mean Dice coefficient of 96.84 % and 96.85 %, respectively, for train and test dataset, indicating close agreement between ground-truth and proposed Pix2Pix-GAN-based segmentation.