Artificial intelligence for corneal ectasia screening: a review and insights from data preprocessing

Abstract

Keratoconus is a progressive corneal ectatic disorder that may cause irregular astigmatism, visual impairment and iatrogenic ectasia after refractive surgery. Early detection remains challenging, as conventional indices often fail to identify fruste or subclinical cases. Advances in optical coherence tomography (OCT) and biomechanical assessments have provided valuable biomarkers, yet sensitivity and generalizability are still limited. This review summarizes recent Artificial Intelligence (AI) applications for corneal ectasia screening published between 2020 and 2025. From 894 initial records, 35 studies were included, covering both Machine Learning (ML) and Deep Learning (DL) methods. Ensemble ML models, such as Random Forest and Gradient Boosting, consistently outperformed alternatives, while convolutional neural networks and transfer learning dominated DL approaches. Future work will focus on analyzing real clinical data available within our group, including OCT-derived metrics and epithelial thickness maps, to explore supervised and unsupervised ML strategies and assess their feasibility for clinical translation.