PyGlaucoMetrics: A Stacked Weight-Based Machine Learning Approach for Glaucoma Detection Using Visual Field Data.

Area of Science:

• Ophthalmology
• Machine Learning
• Data Science

Background:

• Glaucoma classification is critical for early intervention but faces challenges with current diagnostic models and International Classification of Diseases (ICD) codes.
• Existing methods often lack predictive power and exhibit inconsistencies in clinical labeling.

Purpose of the Study:

• To enhance glaucoma (GL) classification accuracy by developing stacked weight-based machine learning models.
• To improve upon the limitations of stand-alone diagnostic models and ICD codes for glaucoma detection.

Main Methods:

• Utilized a dataset of 33,636 participants and 340,444 visual fields (VFs) from the Mass Eye and Ear (MEE) dataset.
• Trained two multi-layer perceptron (MLP) models using visual field data and clinical variables to extract weights from five base glaucoma detection models.
• Employed logistic regression (LR), extreme gradient boosting (XGB), and MLP as meta-learners to classify glaucoma cases.

Main Results:

• The MLP meta-learner achieved the highest accuracy (96.43%), F-score (96.01%), and AUC (97.96%) with minimal prediction uncertainty.
• XGB and LR meta-learners also demonstrated strong performance, with accuracies of 92.86% and 89.29%, respectively.
• The superior temporal sector in VFs and patient age were identified as the most influential features for glaucoma classification.

Conclusions:

• Stacked weight-based meta-learner models significantly outperform stand-alone models in glaucoma classification.
• Achieved an 8.92% accuracy improvement over the best stand-alone model, demonstrating potential for automated glaucoma detection.
• The developed approach offers a robust and accurate tool for clinical application in identifying glaucoma.