Vessel extraction from non-fluorescein fundus images using orientation-aware detector

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate blood vessel segmentation in non-fluorescein fundus images is vital for diagnosing conditions like diabetic retinopathy.
• Retinal vascular structures present complex variations, making precise modeling and extraction challenging.

Purpose of the Study:

• To develop a novel, robust method for automatic blood vessel extraction from non-fluorescein retinal fundus images.
• To improve the accuracy and reliability of retinal vessel segmentation, particularly in challenging image areas.

Main Methods:

• Developed an orientation-aware detector (OAD) leveraging the locally oriented and linearly elongated properties of blood vessels.
• Employed a two-scale segmentation approach using line operators and a Gabor filter bank to capture both wide and thin vessels.
• Implemented a post-processing step using path opening operations to eliminate false positives from non-vascular structures and pathologies.

Main Results:

• The OAD approach achieved competitive performance with CAL measurements of 80.82% on the DRIVE database and 68.94% on the STARE database.
• Demonstrated superior performance compared to existing segmentation methods.
• Showcased high accuracy and robustness in complex regions, including central reflex, closely spaced vessels, and crossover points, even with illumination noise.

Conclusions:

• The proposed orientation-aware detector (OAD) offers an effective and robust solution for blood vessel segmentation in non-fluorescein fundus images.
• This method holds significant potential for enhancing automated screening tools for diabetic retinopathy and other vascular-related eye conditions.