A machine learning approach to predict the glaucoma filtration surgery outcome.

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Machine Learning

Background:

• Glaucoma filtration surgery (FS) success rates vary, necessitating predictive models.
• Accurate prediction of FS outcomes can optimize patient management and surgical planning.
• Current methods for predicting FS outcomes lack sufficient accuracy.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting filtration surgery (FS) outcomes in glaucoma patients.
• To identify key pre-operative parameters that influence FS success.
• To utilize a classification tree (CT) algorithm for outcome prediction.

Main Methods:

• 102 glaucoma patients undergoing FS were analyzed.
• Ocular surface clinical tests (OSCTs), surgical site-related biometric parameters (SSPs), and conjunctival parameters were assessed.
• Anterior segment optical coherence tomography (AS-OCT) measured conjunctival epithelial and stromal thickness (CET, CST) and reflectivity (ECR, SCR).
• A classification tree (CT) ML algorithm was employed for data analysis.

Main Results:

• Filtration surgery success was 60.8% at 12 months.
• Conjunctival stroma thickness (CST) and reflectivity (SCR), along with age, were the most important predictors of FS outcome.
• Cut-off values for CST, SCR, and age were identified for predicting success/failure.
• The ML model achieved an AUC of 0.784, indicating good predictive accuracy.

Conclusions:

• Machine learning, specifically CT analysis, can accurately predict filtration surgery outcomes.
• Pre-operative conjunctival stroma thickness and reflectivity, along with patient age, are significant predictors of FS success.
• A thicker, hyper-reflective conjunctival stroma and younger age are associated with an increased risk of FS failure.