Supervised Machine Learning Models for Ocular Sagittal Height Prediction Incorporating Corneoscleral Profile Data

OBJECTIVE: This study aimed to develop supervised machine learning (ML) models to predict ocular sagittal height (OC-SAG), using anterior eye data derived from topography and tomography. METHOD: ology: A retrospective cohort of 100 eyes provided data such as keratometry, eccentricity, white-to-white distance, and corneoscleral junction (CSJ) metrics through anterior segment optical coherence tomography (CASIA 2). Four supervised ML models were developed to predict OC-SAG at 10 mm and 14 mm chord lengths, using either topographical data alone or both topographical and tomographical data. RESULTS: The model with the strongest predictive accuracy for the validation sample was the Random Forest, both for the 10-mm chord (r=0.88; MAE=28.16±26.81 μm) and the 14-mm chord (r=0.77; MAE=73.56±61.31 μm; flat corneal meridian, topography-based model). The inclusion of tomographic data, such as measurements related to the CSJ, did not significantly improve the performance of the models compared with using topographic data alone. CONCLUSIONS: Adding tomographic predictors did not significantly improve model performance. Despite reduced performance for longer chords, the observed error could be considered clinically acceptable for soft contact lens fitting. The study demonstrated the potential of ML in predicting OC-SAG, providing an accessible estimation method for all clinicians to support decision-making.