Machine-learning for age-related macular degeneration using multimodal fundus data.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Age-related macular degeneration (AMD) is a leading cause of vision loss.
• Predicting the progression of AMD from early to late stages is crucial for timely intervention.
• Current methods may not fully capture the complex risk factors involved in AMD progression.

Purpose of the Study:

• To develop an integrated model combining multimodal fundus image features.
• To accurately predict the individualized risk of progression from early to late stages of AMD.
• To establish a new paradigm for clinical individualized precision medicine in AMD management.

Main Methods:

• Retrospective analysis of 324 AMD patients, divided into training (n=227) and validation (n=97) sets.
• Identification of independent risk factors (pigmentary abnormalities, drusen area, ellipsoid zone) and protective factors (choroidal thickness, blood flow density) using logistic regression.
• Construction and evaluation of prediction models (Random Forest, SVM, XGBoost, KNN) using Python, with performance assessed by AUC.

Main Results:

• Multivariate logistic regression identified pigmentary abnormalities, drusen area, and ellipsoid zone as independent risk factors for AMD progression.
• Subfoveal choroidal thickness and choroidal capillary blood flow density were identified as independent protective factors.
• The Random Forest model demonstrated superior predictive performance (AUC: 0.779 training, 0.700 validation) compared to other algorithms.

Conclusions:

• A robust AMD progression prediction model was successfully developed using multimodal fundus images.
• The Random Forest model effectively identifies patients at high risk of AMD progression.
• This approach offers a novel strategy for personalized precision medicine in managing age-related macular degeneration.