基于对抗学习和引导机制的视盘和视杯联合分割

Joint Optic Disc and Optic Cup Segmentation Based on Adversarial Learning and Guidance Mechanism

准确的视盘(OD)和视杯(OC)分割能够有效地辅助青光眼的诊断和监测,从而进一步提高治疗效果。然而,现有方法没有考虑到眼底图像不同通道之间的差异,并且难以实现对OC边界的精确分割。针对这个问题,提出一种基于对抗学习和引导机制的网络框架ALG-Net,旨在提高OD和OC的分割性能。ALG-Net由分割网络和鉴别器两部分组成。在分割网络中,构建引导融合模块(GFM),该模块将单通道特征信息与RGB图像特征融合,使网络充分学习眼底图像不同通道之间的差异信息,引导分割网络聚焦于关键区域。ALG-Net网络框架还采用了鉴别器,通过对抗学习的方式促进分割网络生成更真实的分割结果。在REFUGE和Drishti-GS数据集上进行广泛的实验评估,实验结果表明,ALG-Net在RUFUGE数据集上OD和OC分割的平衡精度分别达到了98.6%和95.9%,在Drishti-GS数据集上也表现出优异的性能。此外,ALG-Net的分割结果应用于青光眼分类任务,在RUFUGE数据集上ROC曲线下面积(AUC)为0.983,相较于经典UNet算法提高了0.015,为青光眼的早期诊断和监测提供了有力的支持。

Accurate segmentation of the Optic Disc(OD)and Optic Cup(OC)effectively assists in diagnosing and monitoring glaucoma,thereby improving treatment outcomes.However,existing methods do not consider the differences between the various channels of fundus images,making it challenging to achieve accurate segmentation of the OC boundary.To address this problem,a network framework based on adversarial learning and a guidance mechanism,termed ALG-Net,is proposed to improve OD and OC segmentation performance.ALG-Net comprises two main components:a segmentation network and a discriminator.The segmentation network includes a Guidance Fusion Module(GFM)designed to merge single-channel feature information with RGB image features.This allows the network to learn the differences among the various channels of the fundus image,guiding the segmentation network to focus on key regions.The framework also incorporates a discriminator,which encourages the segmentation network to generate more realistic results through adversarial learning.Extensive experimental evaluations are conducted on the REFUGE and Drishti-GS datasets.The results show that ALG-Net achieved balanced accuracies of 98.6%and 95.9%for OD and OC segmentation,respectively,on the RUFUGE dataset and demonstrated better performance on the Drishti-GS dataset.In addition,applying ALG-Net's segmentation results to glaucoma classification tasks yielded an Area Under the ROC Curve(AUC)of 0.983 on the REFUGE dataset,surpassing the classic UNet algorithm by 0.015.This demonstrates ALG-Net's strong support for early diagnosis and monitoring of glaucoma.