Simultaneous segmentation and classification of the retinal arteries and veins from color fundus images

The study of the retinal vasculature is a fundamental stage in screening and diagnosis many diseases. A complete vascular analysis requires to segment classify blood vessels retina into arteries veins (A/V). Early automatic methods approached these segmentation classification tasks two sequential stages. However, currently, are as joint semantic task, results highly depend on effectiveness vessel segmentation. In that regard, we propose novel approach for simultaneous A/V from eye fundus images. particular, method that, unlike previous approaches, thanks loss, decomposes task three problems targeting arteries, whole tree. This configuration allows handle crossings intuitively directly provides accurate masks different target trees. provided ablation public Retinal Images Tree Extraction (RITE) dataset demonstrates proposed satisfactory performance, particularly structures. Furthermore, comparison with state art shows our achieves competitive classification, while significantly improving multi-segmentation detect more better structures, achieving performance. Also, terms, outperforms approaches various reference works. Moreover, contrast crossings, well preserving continuity at complex locations.