SUD-GAN: Deep Convolution Generative Adversarial Network Combined with Short Connection and Dense Block for Retinal...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Retinal blood vessel morphology is crucial for diagnosing ophthalmological diseases.
• Accurate retinal vessel segmentation from fundus images is essential for disease screening and diagnosis.

Purpose of the Study:

• To propose a novel deep learning model, SUD-GAN, for enhanced retinal vessel segmentation.
• To improve the accuracy and precision of blood vessel detection in fundus images.

Main Methods:

• Developed SUD-GAN, a deep convolution adversarial network incorporating short connection and dense blocks.
• The generator uses a U-shape encode-decode structure with short connections to mitigate gradient dispersion.
• The discriminator employs convolution blocks with dense connections to improve feature spread and discrimination ability.

Main Results:

• SUD-GAN achieved state-of-the-art performance on the DRIVE and STARE datasets.
• Achieved high sensitivity (0.8340 on DRIVE, 0.8334 on STARE) and specificity (0.9820 on DRIVE, 0.9897 on STARE).
• Demonstrated superior ability in detecting small vessels and accurately locating vessel edges.

Conclusions:

• SUD-GAN offers a significant advancement in retinal vessel segmentation accuracy.
• The proposed method holds promise for improved early diagnosis and monitoring of retinal diseases.
• The integration of short and dense connections effectively enhances deep network performance in medical image analysis.