Accurate and efficient optic disc and cup segmentation from fundus images is significant for glaucoma screening. However, current neural network-based optic disc (OD) and optic cup (OC) segmentation tend to prioritize the image's local edge features, thus limiting their capacity to model long-term relationships, with errors in delineating the boundaries. To address this issue, we proposed a semi-supervised dual self-integrated transformer network (DST-Net) for joint segmentation of the OD and OC. First, we introduce a dual-view co-training mechanism to construct the encoder and decoder of the self-integrated network from the mutually enhanced feature learning modules of Vision Transformer (ViT) and convolutional neural networks (CNN), which are co-trained with dual views to learn the global and local features of the image adaptively. Moreover, we employ a dual self-integrated teacher-student framework, effectively utilizing large amounts of unlabeled fundus images through semi-supervised learning, thereby refining OD and OC segmentation results. Finally, we use a boundary difference over union loss (BDoU-loss) to optimize boundary prediction further. We implemented the comparative experiments on the publicly available dataset RIGA+. The OD and OC Dice values of the proposed DST-Net reached 95.12 ± 0.14 and 85.69 ± 0.27, respectively, outperforming other state-of-the-art (SOTA) methods. In addition, DST-Net shows strong generalization on the DRISHTI-GS1 and RIM-ONE-v3 datasets, proving its promising prospect in OD and OC segmentation.

The static convolutional network is designed with a restricted sense field, but it limits the global feature extraction. While dynamic convolution addresses the issue of limited static receptive fields, it struggles to perform well in discontinuous liver regions, liver tumor border regions, and microtumor segmentation. To alleviate the above issues, we proposed a network, GS_NeXt, based on graph theory and the Segment Anything Model (SAM) for liver and tumor segmentation from CT scans. First, we employed the feature extraction module of ConvNeXt-v2 to learn features across channels, enabling the network to focus on critical liver and tumor regions. Second, we utilized the SAM with frozen weights to extract more comprehensive global information, thereby enhancing feature representation. Third, we applied graph reasoning to globally model unstructured local features, improving the network's understanding of CT images in discontinuous liver regions, liver-tumor boundaries, and microtumor areas. Finally, we incorporated a deep supervision mechanism to facilitate the learning of multi-scale features throughout the network. We evaluated the proposed segmentation method for two publicly available abdominal liver tumor CT datasets. On the LiTS17 dataset, GS_NeXt achieved 97.74% and 87.25% on the Dice scores, 1.01 and 2.23 mm on the average symmetric surface distance (ASD), and 3.68% and 22.60% on the volume overlap errors (VOE) for liver and tumor segmentation, respectively. On the 3DIRCADb dataset, it achieved 97.31% and 87.36% on the Dice scores, ASD values of 1.01 and 2.12 mm, and VOE scores of 3.56% and 21.56% for liver and tumor segmentation, respectively.