Retinal vascular segmentation using superpixel-based line operator and its application to vascular topology...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Automatic analysis of retinal vascular networks is crucial for diagnosing and treating eye diseases.
• Existing methods face challenges in precise segmentation, topology estimation, and artery-vein classification.

Purpose of the Study:

• To develop a novel framework for accurate retinal vascular segmentation and topology construction.
• To enhance the classification of arteries and veins within the retinal vascular network.

Main Methods:

• A nonlocal total variation inspired Retinex model for image enhancement.
• A superpixel-based line operator for improved segmentation generalizability.
• Dominant sets clustering for retinal vessel topology estimation and artery-vein classification.

Main Results:

• Achieved competitive segmentation accuracy (0.954-0.964) on public datasets (STARE, DRIVE, IOSTAR).
• Demonstrated high accuracy in topology estimation (0.830-0.928) across five public databases.
• Obtained superior artery-vein classification accuracy (approx. 0.909) using vascular topology.

Conclusions:

• The proposed framework effectively addresses segmentation and topology reconstruction challenges, including the crossover problem.
• Vascular topology information significantly enhances the accuracy of artery-vein classification.