Diagnosis of microbial keratitis using smartphone-captured images; a deep-learning model

Area of Science:

• Ophthalmology
• Medical Diagnostics
• Artificial Intelligence

Background:

• Microbial keratitis (MK) is a leading cause of corneal blindness, necessitating rapid diagnosis and treatment.
• Current diagnostic methods face challenges in low-resource settings, hindering timely intervention.
• Developing accessible and efficient diagnostic tools for MK is crucial.

Purpose of the Study:

• To investigate the efficacy of deep learning (DL) for diagnosing and differentiating subtypes of microbial keratitis.
• To assess the utility of smartphone-captured images for MK diagnosis using DL.

Main Methods:

• A dataset of 889 microbial keratitis cases (bacterial keratitis (BK), fungal keratitis (FK), and acanthamoeba keratitis (AK)) was compiled.
• A convolutional neural network (CNN) model was developed and trained for image classification.

Main Results:

• The DL model achieved an overall classification accuracy of 83.8% for MK subtypes.
• Specific accuracies included 81.2% for AK, 82.3% for BK, and 86.6% for FK.
• The model demonstrated strong performance with an AUC of 0.92 for ROC curves.

Conclusions:

• Deep learning models show promise for diagnosing microbial keratitis using smartphone images.
• The approach offers a practical and accessible diagnostic solution, particularly for resource-limited environments.
• Smartphone-based AI diagnostics can improve timely and accurate MK management globally.