Duncan D D, Shukla O B, West S K, Schein O D
Johns Hopkins University Applied Physics Laboratory, Laurel, Maryland 20723-6099, USA.
J Opt Soc Am A Opt Image Sci Vis. 1997 Jun;14(6):1197-204. doi: 10.1364/josaa.14.001197.
We have developed an autonomous objective classification scheme for degree of nuclear opacification. The algorithm was developed by using a series of color 35-mm slides acquired with a Topcon photo slit-lamp microscope and use of standard camera settings. The photographs were digitized, and first, and second-order gray-level statistics were extracted from within circular regions of the nucleus. Classifications of severity were performed by using these features as input to a neural network. Training versus classification performance was tested by using photographs of different eyes, and test/retest classification reproducibility was evaluated by using paired photographs of the same eyes. We demonstrate good performance of the classifier against subjective assessments rendered by the Wilmer grading system [Invest. Ophthalmol. Visual Sci. 29, 73 (1988)] and markedly better test/retest reproducibility.
我们已经开发出一种用于核混浊程度的自主客观分类方案。该算法是通过使用一系列用拓普康照相裂隙灯显微镜拍摄并采用标准相机设置获取的35毫米彩色幻灯片开发而成的。照片被数字化，然后从细胞核的圆形区域内提取一阶和二阶灰度统计量。通过将这些特征作为神经网络的输入来进行严重程度分类。使用不同眼睛的照片测试训练与分类性能，并使用同一只眼睛的配对照片评估测试/重测分类的可重复性。我们证明该分类器相对于威尔默分级系统[《Invest. Ophthalmol. Visual Sci.》29, 73 (1988)]给出的主观评估具有良好性能，并且测试/重测可重复性明显更好。
