Transfer Learning-Based Keratoconus Detection from Scheimpflug Images

Resumen

Preclinical detection of keratoconus is crucial to avoid irreversible corneal damage associated to refractive surgery for laser myopia correction. This work proposes, for the first time, a deep learning-based approach for preclinical keratoconus diagnosis using corneal images. Our model, trained in 22,750 images, achieved an overall accuracy of 90.70%, specificity of 94.29% and an AUC of 0.96, outperforming state of the art clinical standards.