Retinal Vessel Segmentation Using Minimum Spanning Superpixel Tree Detector.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Retinal vessel analysis is crucial for ophthalmic examinations.
• Automatic segmentation of retinal vessels, especially from low-quality images, presents significant challenges.
• Detecting low-contrast and narrow vessels remains a difficult task in automated retinal image analysis.

Purpose of the Study:

• To develop a robust and effective approach for automatic retinal vessel segmentation.
• To qualitatively improve the detection of low-contrast and narrow retinal vessels.
• To provide an automated method for extracting vessels from fundus images.

Main Methods:

• Utilizing superpixels as the elementary unit for vessel segmentation, moving beyond traditional pixel grids.
• Regularizing the segmentation scheme by integrating geometrical structure, texture, color, and spatial information within a superpixel graph.
• Refining segmentation results using an efficient minimum spanning superpixel tree for capturing both global and local image structures.

Main Results:

• The proposed method achieved superior connectivity-area-length (CAL) scores of 80.92% on the DRIVE dataset and 69.06% on the STARE dataset.
• Demonstrated improved detection of vessels, particularly around pathological areas, due to the structure-aware tree detector.
• Outperformed existing state-of-the-art segmentation methods in experimental evaluations on public and challenging retinal image datasets.

Conclusions:

• The superpixel-based approach offers a significant advancement in automated retinal vessel segmentation.
• The method effectively handles low-quality images and improves the detection of challenging vessel structures.
• This technique provides a reliable automated solution for vessel extraction from fundus images, aiding ophthalmic diagnostics.