LiViT-Net: A U-Net-like, lightweight Transformer network for retinal vessel segmentation

• DOI: 10.1016/j.csbj.2024.03.003
• Publication Date: 2024-03-19T16:55:54Z
• Authors: Le Tong; Tianjiu Li; Qian Zhang; Qin Zhang; Renchaoli Zhu; Wei Du; Pengwei Hu
• Keywords: Robustness

The intricate task of precisely segmenting retinal vessels from images, which is critical for diagnosing various eye diseases, presents significant challenges models due to factors such as scale variation, complex anatomical patterns, low contrast, and limitations in training data. Building on these challenges, we offer novel contributions spanning model architecture, loss function design, robustness, real-time efficacy. To comprehensively address a new U-Net-like, lightweight Transformer network vessel segmentation presented. By integrating MobileViT+ local representation the encoder, our design emphasizes processing while capturing image structures, enhancing edge precision. A joint designed, leveraging characteristics weighted cross-entropy Dice effectively guide through task's foreground-background imbalance vascular structures. Exhaustive experiments were performed three prominent databases. results underscore robustness generalizability proposed LiViT-Net, outperforms other methods scenarios, especially environments with fine or edges. Importantly, optimized efficiency, LiViT-Net excels devices constrained computational power, evidenced by its fast performance. demonstrate this study, freely accessible interactive website was established (https://hz-t3.matpool.com:28765?token=aQjYR4hqMI), revealing performance no login requirements.