Development Of Machine Learning Models For Predicting Vault In Implantable Collamer Lens Surgery Based On Implant Orientation
Published 2025 - 43rd Congress of the ESCRS
Reference: FP30.11 | Type: Free paper | DOI: 10.82333/8c1e-mm04
Authors: Amber Ahmad Khattak 1 , Mohammed Muhtaseb* 2
1iLase Private clinic,Cardiff,United Kingdom, 2Ophthalmology,NHS,Cardiff,United Kingdom;Ophthalmology,University of Sulaimani Medical College,Sulimaniyah,Iraq;Ophthalmology,Hebron University Medical College,HebronHe,Palestinian, State of
Purpose
This study aimed to develop a machine learning-based predictive model to estimate postoperative vault and determine the optimal implantable collamer lens (ICL) size, incorporating implant orientation for the first time in a Caucasian population.
Setting
The research was conducted at Arruzafa Ophthalmological Hospital (Córdoba, Spain) and Barraquer Ophthalmology Center (Barcelona, Spain).
Methods
Biometric data from 235 eyes of patients who underwent ICL implantation were collected using anterior segment optical coherence tomography (AS-OCT) CASIA II. Five advanced machine learning regression models were trained and validated. External validation was performed on a dataset of 45 cases.
Results
The Pearson correlation coefficient between predicted and actual vault values was comparable across all five models, with LASSO regression achieving the highest correlation (r = 0.62, p < 0.001), followed by random forest regression (r = 0.60, p < 0.001) and backward stepwise regression (r = 0.58, p < 0.001). The models provided more accurate vault predictions than Nakamura formulas, with approximately 70% of predictions falling within a 150 µm error margin.
Conclusions
Machine learning-based regression models effectively predict the optimal ICL size, enhancing surgical accuracy and personalization by accounting for implant orientation.