RF‐GANs: A Method to Synthesize Retinal Fundus Images Based on Generative Adversarial Network

Abstract

Diabetic retinopathy (DR) is a diabetic complication affecting the eyes, which is the main cause of blindness in young and middle-aged people. In order to speed up the diagnosis of DR, a mass of deep learning methods have been used for the detection of this disease, but they failed to attain excellent results due to unbalanced training data, i.e., the lack of DR fundus images. To address the problem of data imbalance, this paper proposes a method dubbed retinal fundus images generative adversarial networks (RF-GANs), which is based on generative adversarial network, to synthesize retinal fundus images. RF-GANs is composed of two generation models, RF-GAN1 and RF-GAN2. Firstly, RF-GAN1 is employed to translate retinal fundus images from source domain (the domain of semantic segmentation datasets) to target domain (the domain of EyePACS dataset connected to Kaggle (EyePACS)). Then, we train the semantic segmentation models with the translated images, and employ the trained models to extract the structural and lesion masks (hereafter, we refer to it as Masks) of EyePACS. Finally, we employ RF-GAN2 to synthesize retinal fundus images using the Masks and DR grading labels. This paper verifies the effectiveness of the method: RF-GAN1 can narrow down the domain gap between different datasets to improve the performance of the segmentation models. RF-GAN2 can synthesize realistic retinal fundus images. Adopting the synthesized images for data augmentation, the accuracy and quadratic weighted kappa of the state-of-the-art DR grading model on the testing set of EyePACS increase by 1.53% and 1.70%, respectively.

Diabetic Retinopathy Classification With Deep Learning via Fundus Images: A Short Survey

“…Chen et. al [78] proposed dubbed retinal fundus image generative adversarial networks (RF-GANs), which consist of two generation models, that is, RF-GAN1 and RF-GAN2.…”

Unveiling visionary frontiers: a survey of cutting-edge techniques in deep learning for retinal disease diagnosis

“…The DRGraduate model for DR grading was trained with this data augmentation technique, and experiments were performed to assess its impact [64]. Chen et al [65] introduced RF-GANs, comprising two generative models, RF-GAN1 and RF-GAN2. RF-GAN1 addresses the domain gap between semantic segmentation datasets and EyePACS.…”

k-SALSA: k-anonymous synthetic averaging of retinal images via local style alignment

“…Niu et al [40] proposed a method to generate an image consistent with the given pathological descriptors. Both Zhou et al [62] and Chen et al [6] developed GAN models to synthesize retinal images conditioned on a semantic segmentation to improve disease classification performance. Yu et al…”

“…We also note that the existing work on GANs for retinal images (e.g. [6,40,61,62]) leverages additional labeled information such as vessel segmentation, and thus are not directly applicable to our setting where we use the raw fundus images.…”

“…Despite the promise of these approaches, several key challenges remain in applying these methods to retinal images. First, while several works have explored the use of GANs to generate synthetic retinal images [4,5,6,7,40], none to our knowledge have addressed the problem of effectively summarizing these images in the latent space, making the feasibility of this approach for retinal images an unknown. Next, the difficulty of capturing fine-grain visual patterns of retinal images (e.g.…”