Development and validation of a machine learning model to predict recurrence in polypoidal choroidal vasculopathy: a...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Polypoidal choroidal vasculopathy (PCV) recurrence poses challenges for long-term vision preservation.
• Accurate prediction of PCV recurrence is crucial for timely and personalized treatment adjustments.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting PCV recurrence.
• To utilize optical coherence tomography (OCT) and OCT angiography (OCTA) biomarkers for predictive modeling.

Main Methods:

• A multicentre prospective study involving 204 eyes over 1-year follow-up.
• Feature selection using LASSO regression and development of five ML classifiers.
• Evaluation of models using AUC, accuracy, sensitivity, and specificity; SHAP analysis for interpretation.

Main Results:

• The XGBoost model achieved the highest Area Under the Curve (AUC) of 0.861 in the external validation set.
• Key predictors included changes in polyp height, branching neovascular network area, and subfoveal choroidal thickness.
• SHAP analysis provided insights into feature importance for recurrence prediction.

Conclusions:

• The developed XGBoost model offers a reliable tool for predicting PCV recurrence.
• This predictive capability can support personalized treatment strategies and optimize clinical resource allocation.