PMPred-AE: a computational model for the detection and interpretation of pathological myopia based on artificial...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Pathological myopia (PM) is a significant cause of irreversible vision loss worldwide.
• Current diagnostic methods for PM require timely and accurate detection for effective management.
• There is a growing need for automated and interpretable technologies to diagnose PM.

Purpose of the Study:

• To develop an automated and interpretable computational model for pathological myopia detection.
• To enhance the accuracy and efficiency of PM diagnosis using artificial intelligence.

Main Methods:

• A computational model, PMPred-AE, was developed using EfficientNetV2-L with attention mechanism optimization.
• Gradient-weighted class activation mapping (Grad-CAM) was employed for visual interpretation of the model's decisions.

Main Results:

• PMPred-AE achieved high detection accuracies: 98.50% (training), 98.25% (validation), and 97.25% (testing).
• The model demonstrated the ability to focus on specific image regions relevant to PM detection.
• Grad-CAM provided intuitive visual explanations for the AI's diagnostic process.

Conclusions:

• The PMPred-AE model offers a reliable and accurate tool for automated pathological myopia detection.
• The integration of Grad-CAM enhances the interpretability of AI-driven diagnostic decisions in ophthalmology.
• This interpretable AI approach can serve as a valuable tool for healthcare professionals in managing PM.