Classification of Color-Coded Scheimpflug Camera Corneal Tomography Images Using Deep Learning.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Corneal tomography provides detailed maps of the cornea.
• Accurate detection of conditions like keratoconus is crucial for refractive surgery.
• Deep learning offers potential for automated image analysis.

Purpose of the Study:

• To evaluate deep learning for classifying corneal maps from Scheimpflug camera images.
• To assess the diagnostic performance of a convolutional neural network (CNN) for corneal imaging.

Main Methods:

• A domain-specific CNN was developed for image classification.
• Performance was evaluated using accuracy metrics and error analysis.
• Network activation maps were used to interpret model decisions.

Main Results:

• The CNN achieved high classification accuracy (0.983 training, 0.958 test).
• Activation maps indicated the model focused on clinically relevant corneal regions.
• The model demonstrated strong performance in distinguishing normal, subclinical, and keratoconus images.

Conclusions:

• Deep learning effectively classifies Scheimpflug corneal images.
• This approach shows potential for clinical use in screening refractive surgery candidates.
• AI can assist in identifying keratoconus from corneal tomography maps.