Automating screening and diagnosis in the medical field saves time and reduces the chances of misdiagnosis while saving on labor and cost for physicians. With the feasibility and development of deep learning methods, machines are now able to interpret complex features in medical data, which leads to rapid advancements in automation. Such efforts have been made in ophthalmology to analyze retinal images and build frameworks based on analysis for the identification of retinopathy and the assessment of its severity. This paper reviews recent state-of-the-art works utilizing the color fundus image taken from one of the imaging modalities used in ophthalmology. Specifically, the deep learning methods of automated screening and diagnosis for diabetic retinopathy (DR), age-related macular degeneration (AMD), and glaucoma are investigated. In addition, the machine learning techniques applied to the retinal vasculature extraction from the fundus image are covered. The challenges in developing these systems are also discussed.

The exact blood vessel trees segmented from fundus images provide important information required for screening and following‐up of diabetic retinopathy and age‐related macular degeneration. The trained deep neural network presents an automated prediction of the blood vessels in retinal fundus camera images in the publicly DRIVE database with accuracy up to 0.9533 and area under the receiver operating characteristic curve up to 0.9752, which is better than manual recognition by expert human eyes. A resizing technique is introduced and applied to the multi‐level network combining dropout and spatial‐dropout layers to obtain more generalised training. The proposed model has the potential for the classification of other types of images.

A superimposition method for alleviating the characteristic artifact structure of an imaging fiber bundle is demonstrated. This method can reduce the inherent honeycomb patterns of fiber bundle images and restore hidden information by using pixelated images captured from neighboring locations. The pixelation effect is suppressed when multiple adjacent images, with centers separated by one core radius and arranged in a square pattern, are superimposed for summation. The restored image has preserved edge sharpness in cellular and vascular structure boundaries, compared to the reference image without pixelation and identical imaging modality, as found by optical coherence tomography.

We present an iterative method of eliminating pixelation artifacts from an endoscopic image acquired from a coherent fiber bundle imager. Our proposed approach for decoupling the honeycomb effect from the obtained sample image was formulated by using the prior probability for an approximate Bayesian framework in which the ideal complete image can be estimated by maximizing the posterior probability from the observed image. The maximization of the posterior probability from the original mask image (the mirrored fiber bundle imager structure) and the observed image (the sample image of the United States Air Force chart) has been performed by learning the image priors in the space of Markov random fields. By iteratively estimating the probability distribution, we reduced the noise effects from the mask image and recovered the ideal shape of the image. This method was efficient for automatically learning the sliding patch from the combination of projected kernels. The mask and observed images were obtained from en face images of the Fourier domain optical coherence tomography based on a common path interferometry scheme.

Artificial intelligence (AI) has expanded by finding applications in medical diagnosis for clinical support systems