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

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal vessel segmentation is crucial for diagnosing eye diseases.
• Challenges include scale variation, low contrast, and limited training data.
• Existing models struggle with intricate vascular patterns and edge precision.

Purpose of the Study:

• To develop a novel, lightweight network for precise retinal vessel segmentation.
• To enhance model robustness, real-time efficacy, and performance on complex retinal images.
• To address challenges like scale variation, low contrast, and data limitations.

Main Methods:

• Introduced LiViT-Net, a U-Net-like, lightweight Transformer network integrating MobileViT+ and a novel local encoder representation.
• Designed a joint loss function combining weighted cross-entropy and Dice loss for improved segmentation.
• Conducted experiments on three prominent retinal image databases.

Main Results:

• LiViT-Net demonstrated superior robustness and generalizability across datasets.
• The model achieved high precision in segmenting fine vessels and intricate vascular edges.
• LiViT-Net exhibited efficient, fast performance suitable for devices with limited computational power.

Conclusions:

• LiViT-Net effectively addresses key challenges in retinal vessel segmentation.
• The proposed network offers a robust, efficient, and accurate solution for clinical applications.
• A publicly accessible website demonstrates the model's real-time capabilities.