IterNet++: An improved model for retinal image segmentation by curvelet enhancing, guided filtering, offline hard‐sample mining, and test‐time augmenting

Abstract

In clinical medicine, the segmentation of blood vessels in retinal images is essential for subsequent analysis in clinical diagnosis. However, retinal images are often noisy and their vascular structure is relatively tiny, which poses significant challenges for vessel segmentation. To improve the performance of vessel segmentation, an improved model IterNet++ based on the architecture of IterNet is proposed. First, curvelet signal analysis is applied to enhance retinal images. Second, residual convolution (ResConv) blocks and guided filters are introduced to utilise the encoder features of previous iterations in the model to reduce overfitting. Third, offline hard‐sample mining is used to improve segmentation performance by utilising training samples with low segmentation accuracy as many possible on a few‐sample training set. In addition, a test‐time augmentation method is applied to testing samples in test dataset during inference. Extensive experiments show that this model achieves Dice scores of 0.8313, 0.8277, and 0.8372 on DRIVE, CHASE‐DB1, and STARE datasets, respectively, demonstrating the best performance compared with IterNet and other baseline models.

TD Swin-UNet: Texture-Driven Swin-UNet with Enhanced Boundary-Wise Perception for Retinal Vessel Segmentation

“…To further validate the superiority of TD Swin-UNet, we compared it with 13 retinal vessel segmentation methods proposed over the past ten years on the DRIVE and CHASEDB1 datasets. These methods include SegNet [8], UNet [9], Att-Unet [33], UNet++ [11], CE-Net [34], AA-UNet [20], Efficient BFCN [35], PSP-UNet [36], AMF-NET [37], IterNet++ [38], TiM-Net [39], CAS-UNet [40], and LMSA-Net [41]. We conducted comparative experiments on the first five methods, employing identical training strategies and environments across all experiments.…”

TRACL: Temporal reconstruction and adaptive consistency loss for semi‐supervised video semantic segmentation

“…Some transformerbased segmentation models [4][5][6][7][8] are proposed and acquired more impressive performance than current convolutional neural network (CNN)-based networks. In the field of clinical medicine, IterNet++ [54] has presented some strategies, such as using curvelet signal analysis to enhance the retinal images, utilizing the encoder features of previous iterations, and offline hard-sample mining, to improve the performance of vessel segmentation, and achieved the best performance. However, these supervised approaches still heavily rely on large-scale pixel-wise annotated data, which can be too costly to acquire in practice.…”