Boundary and Entropy-Driven Adversarial Learning for Fundus Image Segmentation

Abstract

Accurate segmentation of the optic disc (OD) and cup (OC) in fundus images from different datasets is critical for glaucoma disease screening. The cross-domain discrepancy (domain shift) hinders the generalization of deep neural networks to work on different domain datasets. In this work, we present an unsupervised domain adaptation framework, called Boundary and Entropy-driven Adversarial Learning (BEAL), to improve the OD and OC segmentation performance, especially on the ambiguous boundary regions. In particular, our proposed BEAL framework utilizes the adversarial learning to encourage the boundary prediction and mask probability entropy map (uncertainty map) of the target domain to be similar to the source ones, generating more accurate boundaries and suppressing the high uncertainty predictions of OD and OC segmentation. We evaluate the proposed BEAL framework on two public retinal fundus image datasets (Drishti-GS and RIM-ONE-r3), and the experiment results demonstrate that our method outperforms the state-of-the-art unsupervised domain adaptation methods. Our code is available at https://github.com/EmmaW8/BEAL.

Virtual organelle self-coding for fluorescence imaging via adversarial learning

“…Thus transfer learning should be used if the ground truth of these corresponding channels exists. If not, unsupervised domain adaptation 40 would be a potential solution. In this study, we only performed autofocusing as a preprocessing step for images with blurred Golgi apparatus + F-actin signals in the U2OS-CPS dataset before training the Fluo-Fluo 2, 3, 4 models.…”

Unsupervised Domain Adaptation Based Image Synthesis and Synergistic Adversarial Learning for Optic Disc and Cup Segmentation

“…We consider all the samples in the Drishti-GS dataset and RIM-ONE-r3 dataset as target domain images. For the REFUGE challenge dataset, we only adopt the training set as source domain images following [9].…”

“…Domain adaptation methods [5,6] are explored to deal with the performance degradation caused by domain shift, which aim at effectively transferring the knowledge learned from the source domain to the target domain. Based on recent advances, the studies of unsupervised domain adaptation are mainly divided into three categories: input-level adaptation, feature-level adaptation and output-level adaptation [7][8][9].…”

“…For example, Wang et al [4] used adversarial learning to enforce the prediction results of the target domain that are similar to the source domain. Wang et al [9] devised the networks to obtain low-entropy prediction segmentation maps and boundary prediction maps using adversarial learning. In this paper, adversarial learning is utilized to align the edge prediction results and segmentation prediction results of query images and target domain images with the source domain images, which can improve prediction results of the target domain.…”

Semantic Constraint Based Unsupervised Domain Adaptation for Cardiac Segmentation

“…In addition, some works implement global alignment in the output space, taking every output map as a sample. The alignment of output maps provides a low computation way for feature alignment, which has been widely used in unsupervised domain adaptation segmentation [14] [15] [16].…”