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

Area of Science:

• Ophthalmology
• Medical Imaging
• Biomedical Engineering

Background:

• Subclinical keratoconus (SKC) diagnosis remains challenging.
• Early detection is crucial to prevent vision loss.
• Conventional methods may lack sensitivity for early-stage disease.

Purpose of the Study:

• To evaluate if combining shape-based tomographic parameters and image-derived descriptors from Scheimpflug images enhances SKC detection.
• To develop a clinically translatable framework for SKC screening.

Main Methods:

• Analysis of 186 eyes (30 SKC, 156 controls) using Scheimpflug imaging.
• Computation of 100 parameters: shape-based indices and image-derived metrics (statistical, textural, frequency-domain).
• Feature selection via bootstrapping identified a 5-parameter subset (3 image-derived, 2 shape-based) for a predictive equation; XGBoost classifier used for evaluation.

Main Results:

• A 5-feature set achieved 96.78% accuracy and an AUC of 0.993 with an XGBoost classifier.
• Performance closely matched the full feature set, demonstrating efficiency.
• The selected features provided a directly interpretable predictive equation.

Conclusions:

• Integrating image-derived information with traditional morphological parameters significantly improves SKC detection.
• This approach offers enhanced sensitivity and interpretability for clinical application.
• The study provides a practical and translatable framework for subclinical keratoconus screening.