Application of Deep Learning Algorithms Based on the Multilayer Y0L0v8 Neural Network to Identify Fungal Keratitis

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Fungal keratitis is a serious eye infection requiring accurate and timely diagnosis.
• Current diagnostic methods can be time-consuming and may require specialized expertise.
• Deep learning offers a potential solution for automated image analysis in diagnosing ocular conditions.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for diagnosing fungal keratitis using anterior segment eye photographs.
• To compare the diagnostic performance (sensitivity and specificity) of the deep learning model against practicing ophthalmologists.

Main Methods:

• A dataset of 274 anterior segment images (130 fungal keratitis, 144 controls) was curated.
• The YOLOv8 convolutional neural network was trained on pre-processed and annotated images.
• The model's performance was evaluated on a separate test set and compared with diagnoses made by ophthalmologists.

Main Results:

• The deep learning model achieved 56.0% sensitivity, 96.1% specificity, and 76.5% overall accuracy.
• Practicing ophthalmologists achieved 57.7% sensitivity, 41.7% specificity, and 50.0% accuracy.
• The deep learning model demonstrated superior accuracy compared to expert judgment in this study.

Conclusions:

• Deep learning algorithms possess high potential for fungal keratitis diagnosis, surpassing human expert accuracy without metadata.
• Computer vision technologies can serve as a valuable complementary tool in complex cases and telemedicine settings.
• Further research is needed to refine the model, expand datasets, and compare with alternative diagnostic approaches.