Tallinn University of Technology, Tallinn, Harjumaa, Estonia.
East Tallinn Central Hospital, Tallinn, Harjumaa, Estonia.
Eur J Ophthalmol. 2022 Jan;32(1):508-513. doi: 10.1177/1120672120972027. Epub 2020 Nov 8.
To validate mobile-based funduscope for diabetic retinopathy screening in Estonia.
Quality validation comparison of HEINE iC2 funduscope and Zeiss Visucam camera with image scoring and diagnostic test accuracy measurement by sensitivity and specificity. Study took place from January 2020 until March 2020 in East-Tallinn Central Hospital's eye clinic.
Based on 90 patients, the Zeiss Visucam showed 35.6% DR prevalence while iC2 had 18.9% for images and 17.8% for videos. The average Likert score was 4.7 for Zeiss Visucam and 2.4 for both iC2 images and iC2 videos. The sensitivity of iC2 images was 72.7% (95%CI 49.6-88.4) for grader 1 and 61.9% (95%CI 38.7-81.0) for grader 2, iC2 video sensitivity was 57.1% (95%CI 37.4-75.0) and 65.4% (95%CI 44.4-82.1), respectively. The grader-based specificity for iC2 images was 96.7% (95%CI 80.9-99.8) and 93.5% (95%CI 77.2-98.9). iC2 videos had a 100% (95%CI 91.7-1.0; 92.0-1.0) specificity by both graders. Cohen's kappa agreement was 0.82 and 0.96 for images and videos.
Mobile-based funduscope iC2 is not valid for DR screening with non-dilated pupils and thus not suitable for clinics that do not have experienced specialist present. Moreover, the screening specialist needs to be experienced fundus photographer with extra multiple day training for funduscope use. As main resolution, mobile-based funduscope was not validated for DR screening in Estonia based on pre-set study criteria. Additional research and development of funduscope algorithm for image stripping from videos is needed for validation as iC2 benefits do not offset the gold standard at the moment.
验证基于移动设备的眼底镜在爱沙尼亚用于糖尿病视网膜病变筛查的效果。
通过图像评分和敏感度及特异度的诊断测试准确性测量，对 HEINE iC2 眼底镜和 Zeiss Visucam 相机进行质量验证比较。研究于 2020 年 1 月至 3 月在东塔林中央医院的眼科诊所进行。
基于 90 名患者，Zeiss Visucam 显示出 35.6%的 DR 患病率，而 iC2 的图像患病率为 18.9%，视频患病率为 17.8%。Zeiss Visucam 的平均李克特评分是 4.7，而 iC2 图像和 iC2 视频的评分均为 2.4。iC2 图像的敏感度分别为 1 级 72.7%（95%CI 49.6-88.4）和 2 级 61.9%（95%CI 38.7-81.0），iC2 视频的敏感度分别为 57.1%（95%CI 37.4-75.0）和 65.4%（95%CI 44.4-82.1）。基于分级的 iC2 图像特异性为 96.7%（95%CI 80.9-99.8）和 93.5%（95%CI 77.2-98.9）。两位分级者均认为 iC2 视频的特异性为 100%（95%CI 91.7-1.0；92.0-1.0）。Cohen's kappa 一致性为 0.82 和 0.96，分别用于图像和视频。
基于移动设备的眼底镜 iC2 不适用于未散瞳瞳孔的糖尿病视网膜病变筛查，因此不适合没有经验丰富的专科医生的诊所。此外，眼底镜筛查专家需要是有经验的眼底摄影师，并额外接受多日的眼底镜使用培训。根据预设的研究标准，基于移动设备的眼底镜未在爱沙尼亚验证用于糖尿病视网膜病变筛查。需要对眼底镜算法进行额外的研究和开发，以用于从视频中提取图像的验证，因为目前 iC2 的优势还无法替代金标准。
