Multimodal Artificial Intelligence for Predicting 3- and 5-Year Risks of Myopic Choroidal Neovascularization in High...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• High myopia presents a significant risk for developing myopic macular neovascularization (MNV).
• Accurate risk prediction is crucial for timely intervention and management of MNV in high myopia.
• Current risk assessment methods may not fully capture the complexities of MNV development.

Purpose of the Study:

• To develop and validate artificial intelligence (AI)-based models for estimating the risk of MNV in highly myopic eyes.
• To assess the predictive performance of image-based deep learning (DL) models and a multimodal survival model.
• To explore the clinical utility and interpretability of AI models for MNV risk stratification.

Main Methods:

• Retrospective cohort study involving 4235 eyes from 2501 high myopia patients.
• Development of three DL models using fundus photographs (DenseNet-121) for 1-, 3-, and 5-year MNV risk prediction.
• Creation of a multimodal DeepSurv model integrating image features and clinical variables for time-to-event analysis.

Main Results:

• DL models achieved area under the receiver operating characteristic curves (AUROCs) ranging from 0.729 to 0.798 for different prediction horizons.
• The multimodal DeepSurv model demonstrated a concordance index (C-index) of 0.683, enabling effective risk stratification.
• SHAP analysis highlighted axial length and pathologic myopia category as key predictors in the multimodal model.

Conclusions:

• AI-based models provide accurate fixed-horizon and time-to-event risk estimation for MNV in high myopia.
• These models offer clinically meaningful risk stratification, potentially supporting tailored long-term monitoring.
• Further external validation is recommended to support the clinical implementation of these AI approaches for high myopia management.