An improved retinal vessel segmentation method based on high level features for pathological images.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Current retinal blood vessel segmentation methods often misidentify non-vessel structures in pathological images due to reliance on low-level features.
• Accurate segmentation of retinal vasculature is crucial for diagnosing and monitoring various eye diseases.

Purpose of the Study:

• To develop and evaluate a novel retinal blood vessel segmentation method utilizing high-level features.
• To improve the accuracy and reduce false positives in segmenting retinal blood vessels, particularly in pathological images.

Main Methods:

• A two-step segmentation approach was employed.
• The first step utilized low-level features for initial segmentation.
• The second step incorporated high-level features to refine segmentation by removing non-vessel components.

Main Results:

• The proposed method achieved an accuracy of 0.9536 and a false positive average of 0.0191 across the entire STARE database.
• On pathological images specifically, the method demonstrated an accuracy of 0.9542 and a false positive average of 0.0236.
• These results indicate superior performance compared to existing state-of-the-art methods, especially in minimizing false positives in challenging pathological cases.

Conclusions:

• The proposed high-level feature-based segmentation method offers a significant advancement in retinal blood vessel segmentation.
• The approach effectively processes vessel and non-vessel structures independently, leading to improved accuracy and reduced false positives in pathological retinal images.