Detecting flash artifacts in fundus imagery.
Detecting flash artifacts in fundus imagery.
复制标题
检测眼底图像中的闪光伪影。
DOI：
10.1109/embc.2012.6346211
复制
发表时间：
2012
期刊：
影响因子：
--
通讯作者：
Chaum,Edward
中科院分区：
文献类型：
--
作者：
Paquit,VincentC;Karnowski,ThomasP;Aykac,Deniz;Li,Yaqin;TobinJr,KennethW;Chaum,Edward
在视网膜病变筛查的远程医疗环境中，需要对初始输入图像进行质量检查，以确保足够的清晰度以进行正确的诊断。无论系统使用人工筛选还是自动化软件进行诊断，这都是正确的。我们提出了一种用于检测视网膜图像中发现的闪光伪影的方法。我们收集了一组视网膜眼底图像从2009年2月至2011年8月从几个诊所在美国中南部地区的远程医疗项目的一部分。这些图像已经过质量检查，有时会忽略特定的闪光伪影，这可能不利于视网膜异常的自动检测。结合特征色度信息和形态学模式匹配，提出了一种多步检测视网膜中心区域闪光伪影的方法。闪光检测是在代表人群的5218张图像的数据集上测试的。该系统对闪光伪影检测的灵敏度为96.54%，特异性为70.16%。闪光伪影检测可以作为远程医疗网络中视网膜图像质量筛选的有用工具。通过为这些图像提供特殊处理并结合闪光缓解或去除方法，可以预期该检测将改进自动检测。
In a telemedicine environment for retinopathy screening, a quality check is needed on initial input images to ensure sufficient clarity for proper diagnosis. This is true whether the system uses human screeners or automated software for diagnosis. We present a method for the detection of flash artifacts found in retina images. We have collected a set of retina fundus imagery from February 2009 to August 2011 from several clinics in the mid-South region of the USA as part of a telemedical project. These images have been screened with a quality check that sometimes omits specific flash artifacts, which can be detrimental for automated detection of retina anomalies. A multi-step method for detecting flash artifacts in the center area of the retina was created by combining characteristic colorimetric information and morphological pattern matching. The flash detection was tested on a dataset of 5218 images representative of the population. The system achieved a sensitivity of 96.54% and specificity of 70.16% for the detection of the flash artifacts. The flash artifact detection can serve as a useful tool in quality screening of retina images in a telemedicine network. The detection can be expected to improve automated detection by either providing special handling for these images in combination with a flash mitigation or removal method.