TNSBCS-SOC, Regional Institute of Ophthalmology, Chennai, India.
SC Johnson College of Business, Cornell University, Ithaca NY, USA.
Indian J Ophthalmol. 2023 Aug;71(8):2984-2989. doi: 10.4103/IJO.IJO_3372_22.
To assess the accuracy of e-Paarvai, an artificial intelligence-based smartphone application (app) that detects and grades cataracts using images taken with a smartphone by comparing with slit lamp-based diagnoses by trained ophthalmologists.
In this prospective diagnostic study conducted between January and April 2022 at a large tertiary-care eye hospital in South India, two screeners were trained to use the app. Patients aged >40 years and with a best-corrected visual acuity <20/40 were recruited for the study. The app is intended to determine whether the eye has immature cataract, mature cataract, posterior chamber intra-ocular lens, or no cataract. The diagnosis of the app was compared with that of trained ophthalmologists based on slit-lamp examinations, the gold standard, and a receiver operating characteristic (ROC) curve was estimated. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were computed.
The two screeners used the app to screen 2,619 eyes of 1,407 patients. In detecting cataracts, the app showed high sensitivity (96%) but low specificity (25%), an overall accuracy of 88%, a PPV of 92.3%, and an NPV of 57.8%. In terms of cataract grading, the accuracy of the app was high in detecting immature cataracts (1,875 eyes, 94.2%), but its accuracy was poor in detecting mature cataracts (73 eyes, 22%), posterior chamber intra-ocular lenses (55 eyes, 29.3%), and clear lenses (2 eyes, 2%). We found that the area under the curve in predicting ophthalmologists' cataract diagnosis could potentially be improved beyond the app's diagnosis based on using images only by incorporating information about patient sex and age (P < 0.0001) and best-corrected visual acuity (P < 0.0001).
Although there is room for improvement, e-Paarvai app is a promising approach for diagnosing cataracts in difficult-to-reach populations. Integrating this with existing outreach programs can enhance the case detection rate.
评估基于人工智能的智能手机应用程序 e-Paarvai 的准确性，该应用程序通过比较智能手机拍摄的图像与训练有素的眼科医生的裂隙灯诊断，来检测和分级白内障。
本项在印度南部一家大型三级护理眼科医院进行的前瞻性诊断研究于 2022 年 1 月至 4 月进行，两名筛查员接受了使用该应用程序的培训。招募年龄>40 岁且最佳矫正视力<20/40 的患者参加研究。该应用程序旨在确定眼睛是否患有未成熟白内障、成熟白内障、后房人工晶状体或无白内障。将应用程序的诊断与基于裂隙灯检查的、作为金标准的训练有素的眼科医生的诊断进行比较，并绘制了接收者操作特征（ROC）曲线。计算了敏感性、特异性、阳性预测值（PPV）和阴性预测值（NPV）。
两名筛查员使用该应用程序筛查了 1407 名患者的 2619 只眼。在检测白内障方面，该应用程序具有较高的敏感性（96%）但较低的特异性（25%），总准确率为 88%，PPV 为 92.3%，NPV 为 57.8%。在白内障分级方面，该应用程序在检测未成熟白内障方面的准确性较高（1875 只眼，94.2%），但在检测成熟白内障（73 只眼，22%）、后房人工晶状体（55 只眼，29.3%）和正常晶状体（2 只眼，2%）方面的准确性较差。我们发现，通过仅使用图像并结合患者性别和年龄（P<0.0001）和最佳矫正视力（P<0.0001）信息，预测眼科医生白内障诊断的曲线下面积有可能超过应用程序的诊断。
尽管还有改进的空间，但 e-Paarvai 应用程序是一种有前途的方法，可用于诊断难以到达的人群中的白内障。将其与现有的外展计划相结合，可以提高病例检出率。
