Retinal blood vessel segmentation by using the MS-LSDNet network and geometric skeleton reconnection method

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal blood vessel segmentation is vital for diagnosing ophthalmic diseases.
• Deep learning models achieve high segmentation accuracy but struggle with vascular structural connectivity.
• Maintaining vessel connectivity is essential for reliable diagnostic interpretation.

Purpose of the Study:

• To propose a novel, multi-stage strategy for retinal blood vessel segmentation.
• To address the challenge of maintaining vascular structural connectivity in automated segmentation.
• To improve the detection and reconnection of fine and broken blood vessels.

Main Methods:

• A multiscale linear structure detection network (MS-LSDNet) was developed for enhanced fine vessel detection.
• An adaptive hysteresis thresholding method was employed for robust vascular extraction and binarization.
• A geometric skeleton-based algorithm was utilized for reconstructing and reconnecting broken vessel segments.

Main Results:

• The proposed strategy demonstrated superior performance in maintaining retinal vascular tree connectivity compared to state-of-the-art methods.
• Experimental results on three public datasets validated the effectiveness of the three-stage approach.
• The MS-LSDNet improved the detection of hierarchical features, aiding in fine vessel identification.

Conclusions:

• The novel three-stage retinal blood vessel segmentation strategy effectively preserves vascular structural connectivity.
• This method offers a significant advancement for accurate ophthalmic disease diagnosis.
• The integrated approach of detection, extraction, and reconnection provides a robust solution for vessel segmentation challenges.