A Novel Adaptive Weighted Loss Design in Adversarial Learning for Retinal Nerve Fiber Layer Defect Segmentation

Abstract

Glaucoma is a chronic eye disease that can cause permanent visual loss and is difficult to detect early. Retinal nerve fiber layer defect (RNFLD) is clinical evidence for the diagnosis of glaucoma. Classical deep learning based methods can be used to segment RNFLD from fundus images. However, the segmentation results of these methods do not have the specific geometry of RNFLD, and the segmentation errors of fundus images with special styles are large. In this paper, we present a novel conditional adversarial shuffle U-shaped network (CASU-Net) to segment RNFLD, which consists of a generator and a discriminator. For the generator, a mixed loss is designed, which consists of an adaptive weighted segmentation loss and an adversarial loss. This adaptive weighted segmentation loss can balance the segmentation accuracy of the target and background region, and assign more attention to the hard samples, thus ensuring the consistent improvement of the segmentation accuracy of all fundus images. The adversarial loss not only helps to improve the pixel-wise segmentation accuracy but also makes the geometry of the RNFLD segmentation closer to the ground truth. In addition, in the generator, a shuffle module was designed to fully mine the information of all channels to improve the feature extraction capability of the model. The proposed CASU-Net is verified on a RNFLD dataset from Beijing Tongren Hospital. The experiments show that the CASU-Net achieves state-of-the-art results on this dataset. INDEX TERMS Glaucoma, retinal nerve fiber layer defect segmentation, deep learning.

Accuracy of Using Generative Adversarial Networks for Glaucoma Detection: Systematic Review and Bibliometric Analysis

“…ImageGAN papers were [ 42 , 51 , 58 , 80 , 86 , 88 , 90 , 93 , 94 ], while PixelGAN papers were [ 49 , 73 , 74 , 76 - 78 , 82 , 91 , 92 ]. In addition, PatchGAN papers were [ 43 , 46 , 50 , 57 , 75 , 79 , 81 , 83 - 85 , 87 , 89 ].…”

“…In some cases, doctors are dissatisfied with deep learning segmentation performance, as it is not as real as their expectations. Taking RNFL segmentation as an example, the segmentation results do not have specific geometrical shape of RNFLD as the gold standards and large segmentation errors of fundus images [ 83 ]. Furthermore, the variability of shape and extremely inhomogeneous OD structure appearance result in inaccurate CDR measurement compared with ideal ones [ 115 - 117 ].…”

“…GANs are the framework that allows to create and use practical outputs as a gold standard [ 44 ]. Therefore, these frameworks were adopted by Lu et al [ 83 ] due to their ability to generate the required specific geometry of RNFLD, which is close to ground truth with high precisions, accuracy, and fewer segmentation errors, despite the existence of multiple pieces of RNFL or low-contrast images. Thus, its segmentation results are much more trusted by doctors than CNN’s.…”

“…Adopting GANs in the medical field remains in its infancy, with no breakthrough application yet clinically implemented for GAN-based approaches [ 95 ]. For better feature extraction, researchers must exploit full feature information on RBG channels, spatial structure, and geometry of landmarks [ 83 ]. Semantic segmentation may reduce manual labeling effort [ 50 , 74 ] and enhance model performance when incorporated with WGAN domain adaptation [ 79 ].…”

Accuracy of Using Generative Adversarial Networks for Glaucoma Detection: Systematic Review and Bibliometric Analysis (Preprint)

“…The subsequent layer in our taxonomy was to classify methods according to the generator's backbone (eg, U-Net based or CNN based) [69]. Papers [42,46,49,51,57,73,75,76,[80][81][82][83][84][85][86][87] represented about 50% of the studies (n=16) and were U-Net-based architectures. However, the other 50% of the papers [43,46,50,51,58,74,[77][78][79][88][89][90][91][92][93][94] were CNN-based generators (n=16).…”