基于多模态的干眼图像描述与分级研究  

Research on Dry Eye Image Description and Grading Based on Multi-mode

作　　者：张婉玉 李秀丽[1] ZHANG Wanyu;LI Xiuli(North China University of Water Resources and Electric Power,Zhengzhou 450046,China)

机构地区：[1]华北水利水电大学,河南郑州450046

出　　处：《现代信息科技》2025年第6期110-115,共6页Modern Information Technology

基　　金：2023年河南省高等教育教学改革研究与实践项目(研究生教育类)(2023SJGLX112Y);2024年度河南省高等教育教学改革研究与实践项目(2024SJGLX0332)。

摘　　要：在干眼诊断与评估中,眼表荧光素染色图像具有重要临床价值,但人工评估费时费力,不同医生间的评分也影响一致性。为提升诊断效率与准确性,文章提出了基于深度学习的自动化模型OFGD-Net(Ocular Fluorescence Grading and Description Network),用于眼表荧光素染色图像的分级与描述。OFGD-Net包括图像编码器、两个解码器及注意力机制,编码器负责特征提取,解码器分别生成图像描述及病变严重程度评分,注意力机制则增强了对关键区域的关注。与现有先进模型的对比分析表明OFGD-Net在主要评估指标上表现优越,在生成文本准确性和相似性方面有明显优势。Ocular surface fluorescein staining images hold significant clinical value in the diagnosis and assessment of dry eye.However,manual evaluation is time-consuming and labor-intensive,with scores among the different physicians affecting consistency.To enhance diagnostic efficiency and accuracy,this paper proposes OFGD-Net(Ocular Fluorescence Grading and Description Network),an automated model based on Deep Learning,which is used for grading and describing ocular surface fluorescein staining images.OFGD-Net comprises an image encoder,two decoders,and an Attention Mechanism.The encoder is responsible for extracting features,the decoders generate image description and lesion severity scores separately,and the Attention Mechanism enhances the attention to key regions.Comparative analysis with current advanced models demonstrates that the performance of OFGD-Net is superior in primary evaluation metrics,and it exhibits significant advantages in the accuracy and similarity of generated text.

关 键 词：深度学习 多模态 干眼 眼表荧光素染色 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]