Related Experiment Video
Updated: Jul 12, 2025
04:11
Author Spotlight: Developing a Tool for Using Inverted Confocal Microscopes for In Vivo Intravital Imaging
Published on: June 30, 2023
1.6K
Monitoring the Progression of Clinically Suspected Microbial Keratitis Using Convolutional Neural Networks
Ming-Tse Kuo1,2,3, Benny Wei-Yun Hsu4, Yi Sheng Lin4
1Department of Ophthalmology, Kaohsiung Chang Gung Memorial Hospital and Chang Gung University College of Medicine, Kaohsiung City, Taiwan.
Translational Vision Science & Technology
|November 1, 2023
Summary
A deep learning model using convolutional neural networks (CNNs) can monitor microbial keratitis (MK) progression by analyzing eye images. This AI approach aids in managing MK treatment effectiveness through image pair comparison.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
- Deep Learning
Background:
- Microbial keratitis (MK) requires careful monitoring for effective treatment.
- External eye photography is a common method for tracking ocular conditions.
- The application of advanced AI, specifically deep learning, in ocular diagnostics is an emerging field.
Purpose of the Study:
- To evaluate a convolutional neural network (CNN)-based method for monitoring microbial keratitis (MK) progression.
- To assess the feasibility of using a CNN feature extractor and identifier for managing suspected MK.
- To determine if AI can effectively track disease changes using serial external eye photography.
Main Methods:
- A multicenter longitudinal cohort study involving patients with suspected MK.
- Serial external eye photography was conducted over a 20-year period.
- A CNN-based model, including a feature extractor and identifier, was trained and evaluated using F1 score, accuracy, and AUROC metrics.
Main Results:
- Full training of the CNN model (feature extractor and identifier) yielded significantly higher accuracy compared to training only the identifier.
- The EfficientNet b3-based CNN model achieved a 90.2% F1 score for disease improvement and 82.1% for worsening.
- The model demonstrated 87.3% accuracy and a 94.2% AUROC for classifying disease progression/regression in test image pairs.
Conclusions:
- A CNN-based deep learning approach is effective in monitoring the progression and regression of suspected microbial keratitis (MK).
- Comparing serial external eye image pairs with AI enables effective management of MK during treatment.
- This study validates the utility of AI in ocular image analysis for clinical decision-making in suspected MK cases.
