VeinCluster: Unsupervised Segmentation of Retinal Vessels of Glaucoma

Retinal blood flow is a critical biomarker for the onset and progression of glaucoma. The identification of major blood vessels in non-invasive Optical Coherence Tomography Angiography (OCTA) scans is essential for predicting blood flow patterns. However, this task is highly challenging due to the complexity of vessel structures, frequent intersections, and dense capillary networks. Additionally, annotating such intricate datasets is both labor-intensive and costly. Most existing Artificial Intelligence (AI) models rely on supervised learning, requiring large volumes of high-precision labeled data and significant computational resources to achieve effective training. To address these challenges, we introduce VeinCluster, a novel unsupervised segmentation algorithm designed to extract major blood vessels and their nodes from deep retinal OCTA images. VeinCluster leverages the pixel distribution characteristics of OCTA images to achieve vessel segmentation and vascular node labeling through pixel density grading. Unlike conventional AI models, VeinCluster operates without the need for high-performance GPUs, extensive training times, or large labeled datasets, significantly enhancing its efficiency and accessibility. Moreover, it demonstrates superior interpretability and has outperformed state-of-the-art (SOTA) methods in terms of segmentation accuracy. Beyond vessel segmentation, VeinCluster provides valuable insights for blood flow prediction and glaucoma progression analysis. Future research will focus on extending this work toward 3D reconstruction and dataset development, further advancing its potential for glaucoma diagnosis and treatment.