Evaluation of machine learning classifiers in keratoconus detection from orbscan II examinations

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Keratoconus is a progressive eye condition affecting corneal shape.
• Accurate diagnosis is crucial for timely intervention and management.
• Orbscan II topography maps provide valuable data for corneal analysis.

Purpose of the Study:

• To assess the efficacy of machine learning classifiers in detecting keratoconus.
• To compare the performance of support vector machine (SVM), multi-layer perceptron (MLP), and radial basis function (RBF) neural networks.
• To evaluate these classifiers using Orbscan II topographical data.

Main Methods:

• 11 attributes were extracted from 318 Orbscan II maps (normal, astigmatism, keratoconus, photorefractive keratectomy).
• Ten-fold cross-validation was employed to train and test SVM, MLP, and RBF classifiers.
• Performance metrics included accuracy, sensitivity, specificity, and ROC curve analysis.

Main Results:

• All three classifiers (SVM, MLP, RBF) demonstrated strong performance in keratoconus identification.
• Classifier performance was comparable across SVM, MLP, and RBF.
• Areas under the ROC curves for classifiers significantly exceeded those of individual Orbscan II attributes (p < 0.05).

Conclusions:

• SVM, MLP, and RBF neural networks are effective tools for keratoconus detection.
• These AI-based classifiers, trained on Orbscan II data, offer a promising approach for clinical use.
• The study highlights the potential of machine learning in enhancing ophthalmic diagnostics.