Diagnosability of Keratoconus Using Deep Learning With Placido Disk-Based Corneal Topography

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Placido disk-based corneal topography is a widely used diagnostic tool.
• Accurate diagnosis and staging of keratoconus are crucial for effective patient management.

Purpose of the Study:

• To evaluate the diagnostic performance of deep learning applied to color-coded maps from Placido disk corneal topography for keratoconus detection.
• To assess the ability of this AI approach to accurately stage keratoconus severity.

Main Methods:

• Retrospective analysis of 179 keratoconic eyes (Grades 1-4) and 170 healthy eyes using Placido disk corneal topography data.
• Application of deep learning algorithms to analyze color-coded corneal topography maps.
• Evaluation of diagnostic accuracy, sensitivity, and specificity for keratoconus screening and staging.

Main Results:

• Deep learning achieved high accuracy (0.966) in distinguishing keratoconus from normal eyes (sensitivity 0.988, specificity 0.944).
• The model demonstrated strong performance in classifying keratoconus stages, with accuracies ranging from 0.785 to 0.920 across different grades.
• Area under the curve values indicated excellent discrimination across all grades.

Conclusions:

• Deep learning analysis of color-coded maps from conventional corneal topography is effective for keratoconus diagnosis.
• This AI-driven method aids in both screening and staging of keratoconus, offering potential as a clinical decision support tool.