Development and Validation of a Deep Learning-based Myopia and Myopic Maculopathy Detection and Prediction System

Myopia has become a global public health issue. Myopia affects the psychological health of children and adolescents and poses a financial burden. Furthermore, as myopia progresses it increases the risk of ocular complications such as myopic macular degeneration, leading to irreversible visual impairment or even blindness. According to the World Health Organization , more than 1 billion people worldwide are living with vision impairment caused by myopia, hyperopia, and other problems due to late detection. Therefore, early detection and prediction of children at a high risk of myopia development and progression are critical for precise and effective interventions.

In this study, we developed a deep learning system DeepMyopia, based on fundus images with the following objectives: 1) to predict myopia onset and progression; 2) To detect myopic macular degeneration for AI-assisted diagnosis; 3) To predict the development of myopic macular degeneration; 4) evaluate its cost-effectiveness.

Inclusion criteria

• Subjects with fundus images in the Shanghai Child and Adolescent Large-scale Eye Study (SCALE) ;
• Subjects with fundus images in the Shanghai Time Outside to Reduce Myopia [STORM] trial;
• Subjects with fundus images in the High Myopia Registration Study [SCALE-HM]
• Subjects with fundus images in the Shanghai Myopia Screening (SMS) Study;
• Subjects with fundus images in the Beijing Children Eye Study
• Subjects with fundus images in the First Affiliated Hospital of Kunming Medical University;
• Subjects with fundus images at the Ophthalmology Department of the First Affiliated Hospital of Xinjiang Medical University;
• Subjects with fundus images at the Ophthalmology Department of the Affiliated Hospital of Inner Mongolia Medical University;
• Subjects with fundus images at Zhongshan Eye Centre, Sun Yat-sen University;
• Subjects with fundus images in the Hong Kong Children Eye Study;

Exclusion criteria

• Participants with poor-quality fundus images