A Hybrid Deep Learning Construct for Detecting Keratoconus From Corneal Maps

• DOI: 10.1167/tvst.10.14.16
• Publication Date: 2021-12-16T16:31:45Z
• Authors: Ali H. Al-Timemy; Zahraa M. Mosa; Zaid Alyasseri; Alexandru Lavric; Marcelo M. Lui; Rossen M. Hazarba...; Siamak Yousefi
• Keywords: Confusion

To develop and assess the accuracy of a hybrid deep learning construct for detecting keratoconus (KCN) based on corneal topographic maps.We collected 3794 images from 542 eyes 280 subjects developed seven models anterior posterior eccentricity, elevation, sagittal curvature, thickness maps to extract features. An independent subset with 1050 150 85 separate center was used validate models. We model detect KCN. visualized features parameters quality subjectively computed area under receiver operating characteristic curve (AUC), confusion matrices, accuracy, F1 score evaluate objectively.In development dataset, 204 were normal, 123 suspected KCN, 215 had In validation 50 Images annotated by three specialists. The AUC two-class three-class problems set 0.99 0.93, respectively.The achieved high in identifying KCN provided time-efficient framework low computational complexity.Deep can non-invasive suggesting potential application research clinical practice identify