Computer aided diagnosis for suspect keratoconus detection.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Keratoconus is a progressive eye condition affecting corneal shape.
• Early detection is crucial for effective management and preventing vision loss.
• Current diagnostic methods have limitations in accuracy and efficiency.

Purpose of the Study:

• To develop a stable and cost-effective computer-aided diagnosis (CAD) system for early keratoconus detection.
• To improve the accuracy and efficiency of keratoconus screening in clinical settings.

Main Methods:

• A custom mathematical model combined with a feedforward neural network (FFN) and Grossberg-Runge Kutta architecture was developed.
• The CAD system was trained and validated on retrospective data from 851 subjects with varying degrees of keratoconus.
• Validation employed 10-cross-validation, holdout validation, and Receiver Operating Characteristic (ROC) curves.

Main Results:

• The CAD system achieved 96.56% accuracy in detecting suspect keratoconus, outperforming existing methods like Belin/Ambrosio Deviation (BADD) and Topographical Keratoconus Classification (TKC).
• For mild to moderate keratoconus, the CAD system demonstrated comparable accuracy to established methods (99.50% vs. 99.46% for BADD).
• The algorithm reduced computation time by 70% and enhanced stability and convergence compared to traditional machine learning techniques.

Conclusions:

• The proposed CAD algorithm offers a highly accurate and stable screening platform for early keratoconus detection.
• This framework can assist ophthalmologists in clinical decision-making for keratoconus diagnosis.
• The system's adaptability suggests potential integration with various Scheimpflug tomography systems.