TiM-Net

TiM-Net segments retinal vessels in fundus images to improve segmentation accuracy and robustness for clinical assessment of cardiovascular and cerebrovascular diseases.

Key Features:

• M-Net backbone: Builds upon the M-Net architecture as the backbone for feature extraction in retinal vessel segmentation.
• Dual-Attention Mechanism: Incorporates channel and spatial attention to mitigate noise interference in fundus images.
• Transformer Integration: Embeds Transformer self-attention within skip connections to re-encode features and capture long-range relationships.
• Weighted Side Output Layers: Introduces weighted side output layers to optimally combine multi-stage features for the final segmentation.
• Validation: Demonstrated superiority to state-of-the-art baselines through quantitative and qualitative experiments on three public datasets.

Scientific Applications:

• Retinal vessel segmentation: Segmentation of retinal vasculature in fundus images to support diagnosis and monitoring of cardiovascular and cerebrovascular diseases.
• Robust vessel delineation: Enhanced robustness against noise in fundus imaging to enable more reliable downstream analyses of vascular structures.

Methodology:

Extends the M-Net architecture by integrating channel and spatial dual-attention, embedding Transformer self-attention within skip connections, and adding weighted side output layers; evaluated on three public datasets.