知镜 Cognispect · 医学文献与研究工作台

Impact of mydriasis on image gradability and automated diabetic retinopathy screening with a handheld camera. A real-world setting evaluation.

Costa IDA, Restrepo D, Ribeiro LZ, Aragaki AK, Malerbi FK, Regatieri CS, Nakayama LF.
Arquivos brasileiros de oftalmologia · 2026-02-02 · PMID 41637371

文献摘要

Purpose Diabetic retinopathy screening in lowand middle-income countries is limited by restricted access to specialized care. Portable retinal cameras offer a practical alternative; however, image quality - affected by mydriasis - directly influences the performance of artificial intelligence models. This study evaluated the effect of mydriasis on image gradability and AI-based diabetic retinopathy detection in real-world, resource-limited settings. Methods The proportions of gradable images were compared between mydriatic and non-mydriatic groups. Generalized estimating equations were used to identify factors associated with image gradability, including age, sex, race, diabetes duration, and systemic hypertension. A ResNet-200d model was trained on the mobile Brazilian Ophthalmological dataset and externally validated on both mydriatic and non-mydriatic images. Model performance was evaluated using accuracy, F1 score, area under the curve, and confusion matrix metrics. Sensitivity differences were assessed using the McNemar test, and area under the curves were compared using DeLong's test. The Youden index was used to determine optimal classification thresholds. Agreement between maculaand disc-centered images was analyzed using Cohen's κ. Results The mydriatic group demonstrated a higher proportion of gradable images compared with the non-mydriatic group (82.1% vs. 55.6%; p Conclusion Mydriasis significantly improves image gradability and enhances AI performance in diabetic retinopathy screening. Nonetheless, in lowand middle-income countries where pharmacologic dilation may be impractical, optimizing model calibration and thresholding for non-mydriatic images is essential to ensure effective AI implementation in real-world clinical environments.

在知镜精读原文来源DOI 原文

知镜提供医学文献检索、阅读笔记、人体图谱、统计绘图与研究复现工具。