Simultaneous Segmentation of Layers and Fluids in Retinal OCT Images

Accurate quantification of retinal Optical Coherence Tomography (OCT) images provides important clinical information of the pathological changes present in age-related macular degeneration (AMD). Currently, monitoring the progress of AMD is mostly performed manually by ophthalmologists, which is time-consuming, difficult and prone to errors. In this work, we have developed a model Deep_ResUNet++ to address this issue and to provide an automatic solution to the problem of simultaneous segmenting retinal layers and fluid regions from retinal OCT images. We have evaluated the method on the Annotated Retinal OCT Images (AROI) dataset. Experimental results demonstrate that our method outperformed the baseline U-Net model, the current state-of-the-art models (UNet_ASPP, ResUnet and ResUnet++) and even the human experts' annotation results, and achieved the best performance by a clear margin with Dice Score above 90% in every single class.