Applications of Computer Vision for Infectious Keratitis: A Systematic Review

Area of Science:

• Ophthalmology
• Medical Diagnostics
• Artificial Intelligence

Background:

• Corneal ulcers cause over 2 million cases of preventable blindness annually, disproportionately affecting low- and middle-income countries.
• Accurate and rapid pathogen identification is crucial for effective antimicrobial treatment of infectious keratitis, but current methods are often inaccessible due to cost, speed, and expertise requirements.

Purpose of the Study:

• To systematically review the literature on artificial intelligence (AI) models developed for pathogen detection and classification in infectious keratitis.
• To analyze the methodologies, datasets, and validation practices of AI models in this field.

Main Methods:

• A systematic literature review was conducted for studies published between 2017 and 2024.
• 37 studies developing or validating AI models for infectious keratitis pathogen detection were analyzed.
• Analysis focused on model types, input data, datasets, ground truth determination, and validation strategies.

Main Results:

• AI models demonstrated high accuracy in pathogen detection, particularly through image interpretation.
• Key limitations identified include poor generalizability, limited dataset diversity, lack of multilabel classification, and inconsistent ground truth standards.
• The majority of studies utilized single-center retrospective datasets, hindering real-world clinical application.

Conclusions:

• AI holds significant potential to enhance diagnostic accuracy and global accessibility for infectious keratitis pathogen detection.
• Future research must prioritize diverse datasets, multilabel classification, prospective and multicenter validation, and standardized ground truth definitions to overcome current limitations.