Combination of tissue-derived and shape-based parameters for subclinical keratoconus detection.

This study investigates whether combining clinically conventional shape-based tomographic parameters and image-derived descriptors extracted from raw Scheimpflug corneal images improves the detection of subclinical keratoconus (SKC). A dataset of 186 eyes (30 SKC and 156 controls) was analyzed, from which 100 candidate parameters were computed, encompassing both shape-based indices and image-derived metrics based on statistical, textural, and frequency-domain features. Feature selection using bootstrapping identified a subset of five highly discriminative parameters, comprising three image-derived descriptors and two shape-based metrics used to construct a directly interpretable predictive equation. When evaluated with an XGBoost classifier, this reduced 5-feature set achieved an overall accuracy of 96.78% and an area under the ROC curve of 0.993, closely matching the performance obtained using the full feature set. These results demonstrate that integrating image-derived information with traditional morphological parameters enhances SKC detection while maintaining interpretability, providing a practical and clinically translatable framework for subclinical-stage keratoconus screening.