基于眼底光谱分类网络的青光眼快速筛查研究

Research on Rapid Glaucoma Screening Based on Fundus Spectral Classification Network

摘要: 青光眼是一种全球范围内广泛存在的致盲性眼病，其导致的视神经损伤具有不可逆性，早期诊断与干预对于保护患者视功能至关重要。本研究提出一种基于眼底彩照虚拟光谱技术的青光眼快速筛查方法，通过MST++神经网络将眼底彩色图像转换为高光谱图像，并利用VGG16深度学习模型对视杯/视盘区域的光谱特征进行分类识别。研究采用iChallenge、Messidor等公开数据集，系统验证了该方法的可行性与准确性。结果表明，基于光谱特征的分类模型在青光眼与健康眼分类中准确率达到95%以上，在多疾病分类任务中也表现出良好的区分能力。该方法具有计算复杂度低、设备要求低、操作简便等优势，特别适用于大规模筛查和基层医疗场景，具有重要的临床应用价值和推广前景。

Abstract: Glaucoma is a globally prevalent blinding eye disease, and the optic nerve damage it causes is irreversible. Early diagnosis and intervention are crucial for protecting patients’ visual function. This study proposes a rapid glaucoma screening method based on virtual spectral technology of fundus color photographs. The fundus color images are converted into hyperspectral images using the MST++ neural network, and the spectral features of the optic cup/optic disc region are classified and identified by the VGG16 deep learning model. Public datasets such as iChallenge and Messidor are adopted to systematically verify the feasibility and accuracy of the method. The results show that the classification model based on spectral features achieves an accuracy of over 95% in distinguishing between glaucomatous and healthy eyes, and also exhibits excellent discriminative ability in multi-disease classification tasks. This method has the advantages of low computational complexity, low equipment requirements, and simple operation, making it particularly suitable for large-scale screening and primary medical care scenarios, and possesses significant clinical application value and promotion prospects.