<b>Fundus Image-Based Automatic Segmentation</b>

In this study, we formulate the task of retinal disease diagnosis as a supervised multiclass classification problem. The goal is to automatically assign each fundus image to one of four predefined diagnostic categories: HR, DR, papilledema, or normal. This classification is based on quantitative features derived from vessel segmentation maps, including both radiomic descriptors and the AVR. By framing the task in this way, the study aims to develop interpretable and generalizable models that support clinical decision-making across a broad spectrum of retinal pathologies.

本研究将视网膜疾病诊断任务构建为监督多分类问题。研究目标为自动将每张眼底图像归类至四个预设诊断类别之一：高血压性视网膜病变（HR，Hypertensive Retinopathy）、糖尿病视网膜病变（DR，Diabetic Retinopathy）、视乳头水肿（papilledema）或正常眼底。该分类基于从血管分割图中提取的定量特征，包含影像组学描述符与动静脉比值（AVR，Arteriovenous Ratio）两类。通过采用该任务框架，本研究旨在开发可解释且泛化性良好的模型，以支撑针对广泛视网膜病变谱的临床决策工作。