Implementation and Application of an Intelligent Pterygium Diagnosis System Based on Deep Learning

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Pterygium is a common eye condition requiring accurate diagnosis.
• Early detection and severity classification are crucial for effective management.
• Existing diagnostic methods may have limitations in accessibility and consistency.

Purpose of the Study:

• To develop and evaluate a deep learning-based intelligent diagnostic system for pterygium.
• To assess the system's ability to differentiate between normal eyes, pterygium observation cases, and pterygium operation cases.
• To compare the system's diagnostic performance against expert ophthalmologists.

Main Methods:

• A dataset of 1,220 anterior segment photographs was used for training and testing.
• A deep learning model was trained to classify images into three categories: normal, observation, and operation.
• Diagnostic performance was evaluated using accuracy, sensitivity, specificity, kappa value, AUC, and F1-score.

Main Results:

• The system achieved an overall accuracy of 94.68% on test images.
• High diagnostic consistency was observed, with kappa values above 85% for all groups.
• Excellent performance metrics were reported, including high sensitivity, specificity, and AUC values, indicating robust diagnostic capabilities.

Conclusions:

• The deep learning system effectively diagnoses pterygium and classifies its severity from anterior segment photographs.
• This intelligent system shows potential as a novel screening tool for pterygium.
• The technology could significantly benefit patients in regions with limited access to specialized medical resources.