A Microvascular Segmentation Network Based on Pyramidal Attention Mechanism.

Area of Science:

• Medical Imaging
• Computer Vision
• Ophthalmology

Background:

• Accurate retinal vasculature segmentation is vital for early detection of eye diseases like diabetic retinopathy.
• Extracting fine vessels and edge pixels presents a significant challenge due to complex vessel structures.

Purpose of the Study:

• To improve the segmentation accuracy of thin retinal vessels.
• To enhance the generalization ability of segmentation models.

Main Methods:

• Incorporated a pyramid channel attention module into a U-shaped network (U-Net).
• Optimized the standard convolutional block with a pre-activated residual discard convolution block to prevent overfitting.

Main Results:

• The proposed model demonstrated significant improvements in sensitivity scores across three benchmark datasets (DRIVE, CHASE_DB1, STARE).
• Achieved sensitivity score increases of 7.12%, 9.65%, and 5.36% on the respective datasets compared to the baseline.

Conclusions:

• The enhanced U-Net model effectively captures multi-level information and focuses on vessel-related features.
• The model shows a strong capability for fine vessel extraction, crucial for improved eye disease screening.