Effective automatic classification methods via deep learning for multi-type infectious keratitis diagnosis.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Infectious keratitis (IK) is a major cause of corneal blindness, often due to microbial infections.
• Current diagnostic methods like corneal cultures are slow and may be inaccurate, necessitating automated solutions.
• Early detection and treatment are vital for preventing vision loss from IK.

Purpose of the Study:

• To develop and evaluate a deep learning system for automated classification of infectious keratitis using slit-lamp images.
• To compare the performance of various deep learning models in identifying corneal infections.

Main Methods:

• A dataset of 1,065 diffuse pattern slit-lamp images was collected between March 2018 and November 2023.
• Five deep learning models (EfficientNet_B0, EfficientNet_V2_S, ResNet50, Vision Transformer, DeepIK) were trained for corneal infection classification.
• Performance was assessed using accuracy, precision, recall, F1-score, Cohen's Kappa, and ROC analysis.

Main Results:

• EfficientNet_B0 demonstrated superior performance with 75.2% accuracy, 74.9% sensitivity, 93.8% specificity, and an AUC of 0.943.
• The model achieved a Kappa value of 0.689, indicating good agreement.
• All key performance metrics favored the EfficientNet_B0 model over others.

Conclusions:

• The EfficientNet_B0 deep learning model effectively distinguishes between normal eyes and four types of infectious keratitis.
• This AI approach shows significant potential for improving the diagnosis of keratitis.
• Larger datasets are recommended for future research to further enhance diagnostic accuracy and patient care.