A Deep Learning Model for Detecting the Eyes Receiving Glaucoma Medications Using Anterior Segment Images

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma medications can cause ocular side effects.
• Detecting medication use from anterior segment images is challenging.
• Objective assessment of treatment effects is needed.

Purpose of the Study:

• To develop and evaluate a deep learning model for detecting glaucoma medication use from anterior segment images.
• To visualize the anatomical areas the model focuses on for classification.
• To compare the model's diagnostic performance with human recognition.

Main Methods:

• A deep learning model was trained on 20,000 augmented anterior segment images.
• The model was tested on 200 anterior segment images (100 medication, 100 no medication).
• Performance was measured by Area Under the Receiver Operating Characteristic Curve (AROC); Gradient-Weighted Class Activation Mapping (Grad-CAM) was used for visualization.

Main Results:

• The deep learning model achieved significantly higher accuracy (AROC 0.90) than human recognition (AROC 0.75).
• Performance was consistent across varying conjunctival hyperemia, prostaglandin analog use, and illumination conditions.
• Grad-CAM highlighted the periocular area more frequently in eyes using glaucoma medication (P < 0.01).

Conclusions:

• Deep learning models can objectively detect glaucoma medication use from anterior segment images.
• Saliency mapping indicates the model identifies subtle periocular changes related to treatment.
• This technology aids in assessing medication side effects and developing improved glaucoma treatments.