Issarti Ikram, Consejo Alejandra, Jiménez-García Marta, Kreps Elke O, Koppen Carina, Rozema Jos J
Department of Ophthalmology, Antwerp University Hospital (UZA), Edegem, Belgium; Department of Medicine and Health Sciences, University of Antwerp, Antwerp, Belgium.
Institute of Physical Chemistry, Polish Academy of Sciences, Warsaw, Poland.
Comput Biol Med. 2020 Jul;122:103809. doi: 10.1016/j.compbiomed.2020.103809. Epub 2020 May 16.
To develop an objective severity scoring system for keratoconus for the use in clinical practice.
Corneal elevation and minimum thickness data of 812 subjects were retrospectively collected and divided into two groups: one control group with normal topography in both eyes (304 eyes), and one keratoconus group (508 eyes). Keratoconus cases ranged from suspect to moderate and had at least 1 examination in 1 of 2 recruiting centres. The elevation data were fitted to Zernike polynomial functions up to 8th order. An adapted machine learning algorithm was then applied to derive a platform-independent severity scoring and identification system for keratoconus.
The resulting logistic index for keratoconus (Logik) provided consistent and progressing scoring that reflected keratoconus severity. Moreover, the system provided an accurate classification of suspect keratoconus versus normal (sensitivity of 85.2%, specificity of 70.0%) when compared with Belin/Ambrosio Display Deviation (BAD_D) (sensitivity of 75.0%, specificity of 74.4%) and the Pentacam Topographical Keratoconus Classification (TKC) (sensitivity of 9.3%, specificity of 97.0%). Logik also showed better accuracy for grading keratoconus stages with an average accuracy of 99.9% versus (98.2%, 94.7%) with BAD_D and TKC respectively.
Logik is a reliable index to identify suspect keratoconus and to score the severity of the disease. It shows an agreement with existing approaches while achieving better performance.
开发一种用于圆锥角膜的客观严重程度评分系统，以供临床实践使用。
回顾性收集812名受试者的角膜高度和最小厚度数据，并将其分为两组：一组为双眼地形图正常的对照组（304只眼），另一组为圆锥角膜组（508只眼）。圆锥角膜病例范围从疑似到中度，且至少在两个招募中心之一进行过1次检查。将高度数据拟合到八阶泽尼克多项式函数。然后应用一种改进的机器学习算法来推导一个与平台无关的圆锥角膜严重程度评分和识别系统。
所得的圆锥角膜逻辑指数（Logik）提供了一致且渐进的评分，反映了圆锥角膜的严重程度。此外，与贝林/安布罗西显示偏差（BAD_D）（灵敏度75.0%，特异性74.4%）和Pentacam地形图圆锥角膜分类（TKC）（灵敏度9.3%，特异性97.0%）相比，该系统对疑似圆锥角膜与正常情况提供了准确的分类（灵敏度85.2%，特异性70.0%）。Logik在圆锥角膜分期分级方面也显示出更高的准确性，平均准确率为99.9%，而BAD_D和TKC分别为98.2%、94.7%。
Logik是识别疑似圆锥角膜和对疾病严重程度进行评分的可靠指标。它与现有方法一致，同时表现更优。
