Predicting Corneal Thickness After Collagen Cross-Linking
Published 2025 - 43rd Congress of the ESCRS
Reference: PP14.04 | Type: Poster | DOI: 10.82333/0v50-dq20
Authors: Pavel Beliakouski* 1 , Mikalai Pazniak 1 , Aleh Pazniak 1 , Dmitri Abelski 1 , Tahra AlMahmoud 2 , Elena Likhorad 1 , Yauhen Statsenko 3
1Ophthalmology,Eye Microsurgery Center “Voka”,Minsk,Belarus, 2Surgery Department,College of Medicine and Health Sciences, United Arab Emirates University,Al Ain,United Arab Emirates, 3Radiology Department,College of Medicine and Health Sciences, United Arab Emirates University,Al Ain,United Arab Emirates
Purpose
To develop a predictive model for corneal structural changes after CXL using key preoperative diagnostic parameters and machine learning methods.
Setting
This study was conducted at a specialized ophthalmology center, where patients with keratoconus underwent corneal collagen cross-linking (CXL). Machine learning (ML) models were developed using a retrospective dataset to predict structural corneal changes, providing a data-driven approach for improving risk assessment and clinical decision-making.
Methods
Retrospective analysis of 107 patients 131 eyes who underwent CXL
Dataset included 796 preoperative and postoperative measurements from multiple diagnostic modalities
ML models were applied to correlate preoperative parameters with postoperative corneal thickness changes
Results
MCT(minimal corneal thickness) is a more significant indicator of corneal remodeling post-CXL than CCT(central corneal thickness). Corneal recovery potential decreases in advanced keratoconus, emphasizing early intervention. Polynomial regression models effectively describe the corneal remodeling process, including thinning, stabilization, and partial recovery. Preoperative pachymetry, BAD index, and topography indices strongly correlate with postoperative outcomes, while keratometry and refractometry show moderate associations with post-CXL corneal thickness.ML models integrating multiple diagnostic parameters clinical data and time-dependent factors provide the most precise predictions for CXL effectiveness R2=0.71 and RMSE=22.71
Conclusions
Multimodal preoperative diagnostics enhance risk stratification and predictability of CXL outcomes
The development of an ML-based stratification system can optimize patient selection and improve personalized treatment strategies for keratoconus management