A novel method for blood vessel detection from retinal images

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Retinal blood vessel morphology is key for diagnosing conditions like diabetes, hypertension, and glaucoma.
• Accurate segmentation of retinal vasculature is vital for clinical diagnosis.

Purpose of the Study:

• To develop a robust method for segmenting retinal blood vessels, addressing challenges with varying contrast in large and thin vessels.
• To improve the accuracy and efficiency of retinal blood vessel segmentation for diagnostic purposes.

Main Methods:

• Adaptive local thresholding is employed to segment large vessels based on their contrast.
• Support Vector Machine (SVM) classification is used to identify thin vessel segments with poor contrast.
• A tracking growth algorithm is applied to reconstruct the complete vascular network.

Main Results:

• The algorithm achieved an average sensitivity exceeding 77% on the DRIVE database.
• The proposed method demonstrated an average accuracy of 93.2% in segmenting retinal blood vessels.
• The approach successfully segmented both large, high-contrast vessels and thin, low-contrast vessels.

Conclusions:

• The method effectively distinguishes large vessels using adaptive thresholding and identifies thin vessels via SVM classification.
• Thin vessel segments are enhanced and connected using a tracking algorithm, forming a complete vascular network.
• This novel approach avoids computationally intensive processes and manual intervention, offering a more efficient diagnostic tool.