This paper presents an automated approach for analyzing retinal vascular networks using advanced graph-based techniques. The proposed method integrates deep learning models for the segmentation of retinal vessels, generation of surface meshes, and network construction to quantify and cluster retinal features. A key component of the framework is the transformation of retinal vessels into a biomedical network, where topological and semantic graphs are constructed to capture detailed vascular structures. Network metrics, including clustering coefficients, centrality measures, and routing efficiency, are extracted to characterize vascular networks. Then, Graph Neural Networks (GNN) models are applied to encode each retinal vascular network as a vector of numbers; a similar vector means a similar structure. Finally, clustering algorithms group analogous patterns and identify potential anomalies. Preliminary results, based on a dataset of 1221 retinal augmented surface meshes, demonstrate the effectiveness of the approach in distinguishing groups with similar characteristics, highlighting the potential for the early detection of diseases such as diabetes, hypertension, and cardiovascular conditions.