CVNet: segmentation-assisted laser diagnosis of conjunctiva vessels based on deep learning

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Medical Imaging

Background:

• Accurate segmentation of conjunctival vessels is essential for non-invasive optical detection and early eye disease screening.
• Traditional segmentation methods lack the precision needed for intricate vascular features, limiting diagnostic accuracy.

Purpose of the Study:

• To develop an effective conjunctival vessel segmentation model (CVNet) for enhanced non-invasive ocular monitoring.
• To improve the accuracy and real-time capabilities of conjunctival vessel detection in ocular imaging.

Main Methods:

• Proposed CVNet, a novel model featuring a double-layer nested U structure with a convolutional block attention module.
• Developed an optical system for collecting conjunctival vessel images and trained the CVNet model.
• Implemented non-invasive optical detection experiments using the CVNet segmentation results.

Main Results:

• CVNet achieved a segmentation performance of 86.52% on ocular optical images, surpassing existing networks.
• The model effectively captures structural, textural, and contrast features of conjunctival vessels.
• Demonstrated successful dynamic recognition and precise coordinate positioning for Raman spectroscopy systems.

Conclusions:

• CVNet significantly enhances the accuracy of conjunctival vessel segmentation for non-invasive ocular diagnostics.
• The developed optical system and CVNet model enable real-time eye detection and monitoring.
• This approach offers a promising tool for early eye disease screening and targeted optical interventions.