Epithelial remodeling as basis for machine-based identification of keratoconus

Area of Science:

• Ophthalmology
• Medical Imaging
• Biomedical Engineering

Background:

• Keratoconus is a progressive corneal ectasia characterized by thinning and irregular protrusion.
• Accurate differentiation between normal and keratoconus corneas is crucial for timely diagnosis and management.
• Current diagnostic methods often rely on complex imaging and clinical assessment.

Purpose of the Study:

• To develop and validate automated computerized algorithms for distinguishing normal from keratoconus corneas.
• To assess the utility of epithelial and stromal thickness data in keratoconus diagnosis.
• To evaluate the performance of linear discriminant analysis (LDA) and neural network (NN) models.

Main Methods:

• Corneal epithelial and stromal thickness maps were generated using high-frequency ultrasound arc-scans.
• 130 normal and 74 keratoconic corneas were analyzed, with keratoconus severity graded.
• Stepwise LDA and NN analyses were performed using 161 derived features to classify corneas.

Main Results:

• Stepwise LDA achieved 100% area under the curve (AUC), indicating perfect separation.
• Leave-one-out analysis yielded 99.2% specificity and 94.6% sensitivity.
• NN analysis demonstrated high performance with 99.5% specificity and 98.9% sensitivity on test sets.

Conclusions:

• Automated algorithms utilizing corneal thickness data effectively differentiate normal from keratoconus corneas.
• Epithelial remodeling in keratoconus provides an independent marker for diagnosis.
• These findings support the use of thickness data for objective keratoconus classification.