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

Non-invasive optical detection of conjunctival vessels enables quantitative, real-time monitoring of ocular vascular conditions, which is crucial for early eye disease screening. However, the accuracy of this detection relies heavily on the precise segmentation of conjunctival vessels. Traditional segmentation methods often struggle to capture the intricate features of blood vessels, resulting in limited accuracy. We propose an effective conjunctival vessel segmentation model, CVNet, which utilizes a double-layer nested U structure integrated with a convolutional block attention module. This model emphasizes critical information, including the structural and textural features of the vessels and their contrast with surrounding tissues. In tests with ocular optical images, CVNet achieved a performance of 86.52%, outperforming other networks. Furthermore, we designed and built an optical system, collecting conjunctival vessel images for model training. Finally, a non-invasive optical detection experiment of the ocular surface was implemented based on the segmentation results of the model. The model enables dynamic recognition of conjunctival vessels, providing precise coordinate positioning for excitation beams in Raman spectroscopy systems, and has achieved excellent results in real-time eye detection experiments.