TDCAU-Net: retinal vessel segmentation using transformer dilated convolutional attention-based U-Net method.

Area of Science:

• Medical imaging analysis
• Computer vision in healthcare
• Ophthalmology diagnostics

Background:

• Retinal vessel segmentation is crucial for diagnosing conditions like diabetic retinopathy and glaucoma.
• The U-Net model shows promise but struggles with segmenting fine and dense retinal vessels.
• Accurate segmentation aids in early disease detection and management.

Purpose of the Study:

• To introduce a novel TDCAU-Net model for enhanced retinal vessel segmentation.
• To improve the precision of segmenting fine branching and dense vessels.
• To evaluate the proposed model against state-of-the-art methods.

Main Methods:

• Developed a TDCAU-Net model based on the U-Net architecture with Transformer-based dilated convolution attention.
• Implemented a five-step preprocessing and image segmentation pipeline.
• Trained and tested the model on the DRIVE and CHASEDB1 eye fundus image databases.

Main Results:

• The TDCAU-Net model achieved high sensitivity, specificity, accuracy, and AUC on both datasets.
• Achieved 0.8187 sensitivity, 0.9756 specificity, 0.9556 accuracy, and 0.9795 AUC on the DRIVE database.
• Achieved 0.8243 sensitivity, 0.9836 specificity, 0.9738 accuracy, and 0.9878 AUC on the CHASEDB1 database.

Conclusions:

• The TDCAU-Net model significantly outperforms U-Net and other mainstream methods in retinal vessel segmentation.
• The proposed approach demonstrates superior performance in segmenting fine branching and dense vessels.
• TDCAU-Net offers a promising advancement for automated diagnosis of retinal diseases.