MFA-UNet: a vessel segmentation method based on multi-scale feature fusion and attention module.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal vessel segmentation is crucial for diagnosing retinal diseases.
• Complex vessel structures, especially microvessels, pose challenges for current segmentation methods.

Purpose of the Study:

• To develop an advanced vessel segmentation method to overcome limitations in segmenting complex retinal vasculature.
• To improve the accuracy and reliability of retinal vessel segmentation, particularly for microvessels.

Main Methods:

• A novel multi-scale feature attention network (MFA-UNet) was developed, incorporating preprocessing steps like gamma correction and contrast-limited adaptive histogram equalization.
• The MFA-UNet utilizes a Multi-scale Fusion Self-Attention Module (MSAM) for enhanced feature adjustment and global dependency establishment.
• A multi-branch decoding module based on deep supervision (MBDM) and a parallel attention mechanism were employed for targeted segmentation and improved feature exploitation.

Main Results:

• The MFA-UNet achieved competitive performance across multiple datasets (DRIVE, STARE, CHASEDB1, HRF, IOSTAR, FIVES).
• Specific performance metrics include Dice scores ranging from 78.60 to 84.17 and accuracies from 95.71 to 97.10.

Conclusions:

• The proposed MFA-UNet demonstrates high efficacy in segmenting retinal vessels, including challenging microvessels.
• The method is expected to offer reliable segmentation results, supporting clinical diagnosis of retinal diseases.