Data for "Deep learning approach for steep and flat corneal curvature prediction using fundus photography"
Description
Corneal curvature is one of the important factors consisting of refractive errors and related to several pathological conditions of the cornea. Fundus photography is the widely used tool to examine the eyes in the current healthcare setting worldwide. The aim of this study is to develop a deep learning model based on fundus photography to predict corneal curvature by categorizing the steep, regular, and flat cornea groups. We analyzed the preoperative measurements of healthy participants without previous surgical history. The participants intended to undergo refractive surgery at the B&VIIT Eye Center from January 2020 to March 2020. participants did not wear contact lenses that could affect corneal curvature four days before the visit for accurate preoperative examination. We retrospectively collected preoperative keratometry measurement and FP that were used to develop a deep learning model. FP of the posterior pole was collected using 3-dimensional retinal imaging device (Topcon 3D OCT‐1 Maestro, Tokyo, Japan). The corneal curvature (shape) was evaluated based on corneal tomography from a Pentacam Scheimpflug system (Oculus Optikgeräte GmbH, Wetzlar, Germany). We used mean anterior keratometry (K) value (an average of the flat [K1] and steep [K2] keratometry values) as a representative value for corneal curvature. The K values of 45 D or more were classified as the sharp cornea group. If the K values were less than 42 D, they were classified as the flat cornea group (Figure 3). The rest were classified into the regular cornea group. This dataset included the fundus photograph samples of this study. Fifty images per each group were uploaded.