Comparison of Artificial Intelligence-Based Machine Learning Classifiers for Early Detection of Keratoconus

Area of Science:

• Ophthalmology
• Medical Technology
• Artificial Intelligence

Background:

• Corneal ectatic conditions, such as keratoconus (KCN) and subclinical keratoconus (SKCN), require accurate diagnosis for effective management.
• Early detection of KCN is crucial to prevent disease progression and vision loss.

Purpose of the Study:

• To compare the diagnostic agreement between artificial intelligence (AI)-based classifiers and expert ophthalmologists in identifying normal corneas versus ectatic conditions.
• To evaluate the performance of specific AI classifiers against clinical expertise in diagnosing KCN and SKCN.

Main Methods:

• A prospective diagnostic test study involved 212 eyes categorized by expert examiners into normal, SKCN, and KCN groups.
• Four AI classifiers (Pentacam BADD, TKC, Sirius Phoenix, OPD-Scan III Corneal Navigator) were used to categorize the same cases.
• Agreement was assessed using sensitivity, specificity, and the Kappa index (κ).

Main Results:

• Sirius Phoenix showed the highest agreement with clinical diagnosis for SKCN (κ=0.70) and KCN (κ=0.79).
• Phoenix also achieved the highest accuracy in differentiating KCN (91.24%) and SKCN (88.68%) compared to other AI classifiers.
• While AI classifiers demonstrated strong performance, their agreement with expert opinion varied.

Conclusions:

• AI-based classifiers, particularly Sirius Phoenix, are valuable tools for detecting early keratoconus.
• AI classifiers cannot entirely substitute for the expertise of clinical specialists in diagnosing corneal ectasia and guiding treatment decisions, especially before refractive surgery.
• The study acknowledges potential variability in both AI performance and expert clinical judgment.