SRV-GAN: A generative adversarial network for segmenting retinal vessels.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Retinal diseases often present complications requiring accurate analysis of retinal vasculature.
• Effective segmentation of retinal blood vessels is critical for diagnosing and monitoring various ophthalmological conditions.

Purpose of the Study:

• To propose an improved generative adversarial network (GAN) model for enhanced retinal blood vessel segmentation.
• To improve the accuracy and effectiveness of retinal blood vessel segmentation for better disease diagnosis.

Main Methods:

• An improved GAN architecture based on R2U-Net was developed for retinal blood vessel segmentation.
• The generator incorporates channel and spatial attention mechanisms to minimize information loss and extract robust features.
• The discriminator utilizes dense connection modules to mitigate gradient disappearance and facilitate feature reuse.
• Mean squared error was introduced into the traditional GAN loss function to generate more realistic blood vessel structures.

Main Results:

• The proposed model demonstrated high performance in retinal blood vessel segmentation across three public datasets (DRIVE, CHASE-DB1, STARE).
• Achieved Area Under the Curve (AUC) scores of 0.9869 (DRIVE), 0.9894 (CHASE-DB1), and 0.9885 (STARE).
• The experimental results showed significant improvements compared to previous state-of-the-art methods.

Conclusions:

• The developed improved GAN model effectively segments retinal blood vessels with high accuracy.
• The integration of attention mechanisms and dense connections enhances feature extraction and model stability.
• This approach offers a promising tool for the early detection and management of retinal diseases.