Retinal Vessel Segmentation Made Easy with DAU-Net

Segmenting blood vessels in fundus images is crucial for diagnosing eye diseases like diabetic retinopathy. Enter DAU-Net, a new network that's easier to use and more accurate than ever. It's divided into an encoder and decoder, with some clever tricks to make it lightweight and powerful. Traditional convolutional layers are swapped for ConvNeXt Block and SnakeConv Block, boosting its ability to recognize different types of blood vessels.

DAU-Net also features two attention modules: Local-Global Attention (LGA) and Cross-Fusion Attention (CFA). LGA highlights important vessel features and ignores background noise, while CFA ensures no vital details are lost during feature extraction. Tests on public datasets like DRIVE, CHASE_DB1, and STARE show that DAU-Net is a star performer, with impressive scores across the board. This makes it a promising tool for real-world clinical use.

Actions