Artificial Intelligence Powered Triple Decision Tree Algorithm For Identifying High-Risk Corneas: A Novel Approach To Prevent Post-Laser Ectasia

To investigate a new machine learning based algorithm in the detection of corneal at risk of ectasia before laser vision correction

Refractive Surgery Unit, Ophthalmology Department, Hadassah Medical Center

A total of 372 eyes of 197 patients were enrolled: 177 normal eyes of 95 subjects, 47 eyes of 47 patients with forme fruste keratoconus and 148 eyes of 102 patients with keratoconus

All eyes were imaged with a Dual Scheimpflug Analyzer System (Galilei G6). Fifty-five parameters derived from anterior and posterior corneal measurements were analyzed for each eye. A set of 3 synergistic decisions trees were built to improve the performance of detection of subclinical keratoconus.

The discriminating rules generated with the automated decision tree classifier allowed to discriminate between normal and keratoconus with 100% sensitivity and 99.5% specificity and between normal and FFKC with 97.6% sensitivity and 99.2% specificity. The first 2 decisions tree are based on anterior asymmetry-based indices and enable 80% sensitivity and 91% specificity, whereas the additional third decision tree incorporates posterior and pachymetry based indices and enable the further improvement of the sensitivity (97.6) and specificity (99.2)

This new machine learning classifier synergistic method showed a very good performance for discriminating between normal corneas and subclinical keratoconus and provided a tool that is closer to an automated medical reasoning.