FRE-GAN : Full-resolution efficient convolutional generative adversarial network for retinal vessel segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate retinal blood vessel segmentation is crucial for diagnosing and researching eye diseases.
• Challenges include varying vessel thickness, low contrast, and complex topology in medical images.
• Existing deep learning methods struggle to achieve excellent performance due to these complexities.

Purpose of the Study:

• To propose FRE-GAN, a full-resolution efficient convolutional generative adversarial network for improved retinal blood vessel segmentation.
• To enhance feature extraction and segmentation efficiency in retinal images.
• To improve the generation quality and topological continuity of blood vessel structures.

Main Methods:

• Designed a full-resolution parallel convolutional interactive generator.
• Developed a lightweight dual-domain discriminator for multi-scale feature learning and information fusion.
• Reconstructed a topological continuity loss function to address blood vessel topology.

Main Results:

• FRE-GAN demonstrated higher accuracy, continuity, and comprehensive performance on DRIVE, CHASE_DB1, and STARE datasets.
• Achieved superior results compared to classical and state-of-the-art methods with a small parameter count.
• Qualitative and quantitative evaluations confirmed the method's effectiveness.

Conclusions:

• FRE-GAN offers an efficient and accurate solution for retinal blood vessel segmentation.
• The proposed network architecture and loss function effectively handle challenges in medical image analysis.
• This method shows significant potential for clinical diagnosis and disease research.