Retinal blood vessel segmentation using line operators and support vector classification.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Accurate retinal vessel segmentation is crucial for diagnosing various eye diseases.
• Existing methods face challenges in precise vessel delineation.

Purpose of the Study:

• To propose and evaluate novel line operator-based methods for retinal vessel segmentation.
• To compare unsupervised and supervised approaches for improved diagnostic accuracy.

Main Methods:

• A line detector, adapted from mammography, was applied to the green channel of retinal images.
• Two segmentation methods were developed: unsupervised thresholding and supervised Support Vector Machine (SVM) classification.
• Feature vectors for SVM were constructed using orthogonal line detector responses and pixel grey levels.

Main Results:

• Both unsupervised and supervised methods demonstrated effectiveness in retinal vessel segmentation.
• Receiver Operating Characteristic (ROC) analysis confirmed the performance on public fundus image databases.
• The supervised SVM approach, utilizing enhanced features, showed promising results.

Conclusions:

• Line operator-based techniques offer a viable approach for retinal vessel segmentation.
• The proposed methods contribute to advancing computer-aided diagnosis of eye conditions.
• Further research can explore optimizations for clinical application.