A Novel Chaos-inspired Artificial Intelligence-based Model for Keratoconus Prediction

To develop a novel chaos-inspired artificial intelligence (AI)-based system for keratoconus prediction.

We constructed, trained, and tested our prediction model using corneal tomography data obtained from Pentacam and Sirius imaging systems. The study included 65 healthy individuals and 48 patients with keratoconus recruited from a private ophthalmology clinic and Bina Eye Hospital in Tehran, Iran. Model development included stratified partitioning of the data into training and testing sets, followed by performance evaluation using standard classification metrics. Scheimpflug-based tomographic parameters were preprocessed, normalized, and analyzed using a chaos-inspired tensor decomposition framework.

Using Pentacam data, the model achieved sensitivity and specificity values of 88% and 90%, respectively. When using Sirius data, the sensitivity and specificity increased to 92% and 96%, respectively. Performance metrics were reproducible across repeated random train-test splits, indicating model stability. The chaos-inspired AI-based model demonstrated balanced sensitivity and specificity across both imaging systems, supporting robust discrimination between keratoconus and normal eyes.

Given the complex and nonlinear behavior of keratoconus progression, influenced by biomechanical, genetic, and environmental interactions, we developed a novel chaos-inspired AI-based system (Iran Model) capable of accurately detecting keratoconus. Although the findings are promising, further validation in larger cohorts is required.