Self-supervised learning-enhanced deep learning method for identifying myopic maculopathy in high myopia patients.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• High myopia presents diagnostic challenges, necessitating improved methods for detecting myopic maculopathy (MM).
• Current diagnostic techniques for MM require enhancement for large-scale screening efficiency.

Purpose of the Study:

• To develop and validate a deep learning (DL) system augmented with self-supervised learning (SSL) for accurate MM diagnosis.
• To assess the performance of the SSL-enhanced DL system against conventional methods and expert diagnosis.

Main Methods:

• Utilized a large dataset (7,906 images) from the Shanghai High Myopia Screening Project and a public validation set (1,391 images).
• Developed a deep learning model incorporating a self-supervised learning strategy for enhanced feature extraction.
• Evaluated the model's diagnostic accuracy, sensitivity, specificity, and AUC values on both internal and external datasets.

Main Results:

• The SSL-enhanced DL system achieved high internal performance (96.8% accuracy, 83.1% sensitivity, 95.6% specificity, AUCs up to 0.999).
• External validation demonstrated robust performance (89.0% accuracy, 71.7% sensitivity, 87.8% specificity, AUCs up to 0.978).
• The model's Cohen's kappa values exceeded 0.8, indicating substantial agreement with retinal experts.

Conclusions:

• The self-supervised learning-enhanced deep learning approach significantly improves the automatic diagnosis of myopic maculopathy.
• This AI system shows great potential for improving the efficiency and accuracy of large-scale myopia screenings.
• The findings highlight the broader significance of advanced AI in early detection and treatment of MM.