文献速递眼底病与葡萄膜炎

评估基于眼底照片的人工智能软件在深肤色人群中自动筛查糖尿病视网膜病变

ASSESSMENT OF ARTIFICIAL INTELLIGENCE SOFTWARE FOR AUTOMATIC SCREENING OF DIABETIC RETINOPATHY BASED ON FUNDUS PHOTOGRAPHS IN MELANODERM SUBJECTS.
PubMed 文献 · Retina · 2025-01-01

该前瞻性研究于2021年3月1日至2022年9月30日在塞内加尔达喀尔Abass NDAO医院眼科开展,纳入305只眼,以美国眼科学会临床分级为参照,评估Gaiha Prio Retino +™人工智能软件检测糖尿病视网膜病变的表现。结果显示,该软件筛查需转诊糖尿病视网膜病变的敏感度为92.31%、特异度为99%、曲线下面积0.989;作者认为其是筛查需转诊糖尿病视网膜病变的有效工具。

要点

  • 305只眼中,眼科医生判定需转诊糖尿病视网膜病变104只眼,人工智能判定96只眼。
  • 人工智能筛查需转诊糖尿病视网膜病变的敏感度92.31%、特异度99%、曲线下面积0.989。
  • 作者认为Gaiha Prio Retino +™是筛查需转诊糖尿病视网膜病变的有效工具。

核心结论

  • 作者认为,基于上述结果,Gaiha Prio Retino +™是筛查需转诊糖尿病视网膜病变的有效工具。

研究设计与样本

  • 前瞻性研究,于2021年3月1日至2022年9月30日在塞内加尔达喀尔Abass NDAO医院眼科开展,共纳入305只眼。
  • 糖尿病视网膜病变的临床分类基于美国眼科学会标准,并将临床结果与Gaiha Prio Retino +™自动判读视网膜照片的结果进行比较。

主要发现

  • 需转诊糖尿病视网膜病变:眼科医生判定104只眼,人工智能判定96只眼。
  • 威胁视力的糖尿病视网膜病变:眼科医生检出102只眼,人工智能检出94只眼,对应敏感度92.16%、特异度99.01%、曲线下面积0.975。
  • 黄斑病变:眼科医生识别93只眼,人工智能识别89只眼,对应敏感度95.7%、特异度97.17%、曲线下面积0.988。

关键数值与推荐

  • 需转诊糖尿病视网膜病变:敏感度92.31%、特异度99%、曲线下面积0.989。
  • 威胁视力的糖尿病视网膜病变:敏感度92.16%、特异度99.01%、曲线下面积0.975。
  • 黄斑病变:敏感度95.7%、特异度97.17%、曲线下面积0.988。

临床怎么用

  • 原文未据此给出推荐。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

PURPOSE: To assess the Gaiha Prio Retino +™ Artificial Intelligence (AI) software for detecting diabetic retinopathy (DR). METHODS: This prospective study was conducted from March 1, 2021, to September 30, 2022, in the Ophthalmology Department of the Abass NDAO Hospital (Dakar, Senegal). The clinical classification of DR was based on American Academy of Ophthalmology. The clinical results were compared with those obtained from the automated reading of retinophotos taken using Gaiha Prio Retino +™, a software designed to detect DR. RESULTS: The study covered 305 eyes. Referable DR was observed in 104 eyes by the ophthalmologist and in 96 eyes by AI, corresponding with a sensitivity of 92.31%, a specificity of 99%, and an area under the curve of 0.989. Vision-threatening DR was detected in 102 eyes by the ophthalmologist and in 94 eyes by AI, with a corresponding sensitivity of 92.16%, specificity of 99.01%, and an area under the curve of 0.975. Maculopathy was identified in 93 eyes by the ophthalmologist and in 89 eyes by AI, with a corresponding sensitivity of 95.7%, specificity of 97.17%, and an area under the curve of 0.988. CONCLUSION: Considering these results, the authors may conclude that Gaiha Prio Retino +™ is an effective tool for screening referable DR.

来源:Retina,查看原文。版权归原作者所有。

引用

  1. Mbaye S, Aw A, Sy EHM, et al. ASSESSMENT OF ARTIFICIAL INTELLIGENCE SOFTWARE FOR AUTOMATIC SCREENING OF DIABETIC RETINOPATHY BASED ON FUNDUS PHOTOGRAPHS IN MELANODERM SUBJECTS. Retina. 2025. PMID: 39437373. doi:10.1097/IAE.0000000000004310

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。