Evaluation of Artificial Intelligence-Based Grading of Diabetic Retinopathy in Primary Care
Evaluation of Artificial Intelligence-Based Grading of Diabetic Retinopathy in Primary Care
复制标题
DOI：
10.1001/jamanetworkopen.2018.2665
复制
发表时间：
2018-09-01
影响因子：
13.8
通讯作者：
Mehrotra, Ateev
中科院分区：
文献类型：
--
作者：
Kanagasingam, Yogesan;Xiao, Di;Mehrotra, Ateev
重要性 人们对使用基于人工智能 (AI) 的视网膜图像分级来识别糖尿病视网膜病变产生了广泛的兴趣，但这样的系统从未在临床实践中部署和评估。 目的 描述在初级保健实践中部署的糖尿病视网膜病变人工智能系统的性能。 设计、设置和参与者 与 4 名西方医生一起对初级保健实践中的糖尿病患者进行诊断研究 2016年12月1日至2017年5月31日期间在澳大利亚进行。共有193名患者同意参加这项研究，并拍摄了他们眼睛的视网膜照片。基于 AI 的系统和眼科医生对 386 张图像进行了评估。 主要结果和测量 与眼科医生评估的金标准相比，AI 系统的敏感性和特异性。 结果 在 193 名患者（93 [48%] 女性；平均 [SD] 年龄。55 [17] 岁 [范围，18-87 岁]）中，AI 系统判断 17 人患有糖尿病 视网膜病变严重程度足以需要转诊。系统正确识别了 2 名患有真正疾病的患者，并将 15 名患者错误分类为患有疾病（假阳性）。结果特异性为 92%（95% CI，87%-96%）。阳性预测值为12%（95% CI，8%-18%）。许多假阳性是由图像质量不足（例如脏镜片）和光泽反射造成的。结论和相关性结果证明了在临床实践中使用人工智能系统识别糖尿病视网膜病变的潜力和挑战。主要挑战包括疾病发病率低、假阳性率高以及图像质量差。需要对初级保健中的人工智能系统进行进一步评估。
IMPORTANCE There has been wide interest in using artificial intelligence (AI)-based grading of retinal images to identify diabetic retinopathy, but such a system has never been deployed and evaluated in clinical practice.OBJECTIVE To describe the performance of an AI system for diabetic retinopathy deployed in a primary care practice.DESIGN, SETTING, AND PARTICIPANTS Diagnostic study of patients with diabetes seen at a primary care practice with 4 physicians in Western Australia between December 1, 2016, and May 31, 2017. A total of 193 patients consented for the study and had retinal photographs taken of their eyes. Three hundred eighty-six images were evaluated by both the AI-based system and an ophthalmologist.MAIN OUTCOMES AND MEASURES Sensitivity and specificity of the AI system compared with the gold standard of ophthalmologist evaluation.RESULTS Of the 193 patients (93 [48%] female; mean [SD] age. 55 [17] years [range, 18-87 years]), the AI system judged 17 as having diabetic retinopathy of sufficient severity to require referral. The system correctly identified 2 patients with true disease and misclassified 15 as having disease (false-positives). The resulting specificity was 92% (95% CI, 87%-96%). and the positive predictive value was 12% (95% CI, 8%-18%). Many false-positives were driven by inadequate image quality (eg. dirty lens) and sheen reflections.CONCLUSIONS AND RELEVANCE The results demonstrate both the potential and the challenges of using AI systems to identify diabetic retinopathy in clinical practice. Key challenges include the low incidence rate of disease and the related high false-positive rate as well as poor image quality. Further evaluations of AI systems in primary care are needed.