Quantification of Corneal Neovascularization by Segmentation of Blood Vessels in Slit-Lamp Microscopy Images

The cornea can form blood vessels as a result of inflammatory disease. To assess the evolution of vascularization and assist clinical decision making, ophthalmologists can assess slit lamp microscopy images, but this is qualitative and subjective, and more detailed analysis requires a time-consuming manual process. In this work, we propose automated image segmentation algorithms using classical methods and deep learning for 20 images, manually annotated by specialists. Artificial intelligence models, known as convolutional neural networks, based on the U-Net architecture, were trained and tested in three experiments and results compared with those from B-COSFIRE, a traditional image processing approach. The results with U-Net were the most promising, achieving Sørensen-Dice Similarity values of 48%.