Prediction of the Fundus Tessellation Severity With Machine Learning Methods

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• Fundus tessellation (FT) severity assessment is crucial for eye health.
• Population-based studies are essential for understanding ophthalmic conditions.

Purpose of the Study:

• To predict fundus tessellation (FT) severity using machine learning methods.
• Evaluate the performance of various machine learning models for FT severity prediction.

Main Methods:

• A population-based cross-sectional study of 3,468 individuals from the Beijing Eye Study 2011.
• Utilized five machine learning methods: ordinal logistic regression, ordinal probit regression, ordinal log-gamma regression, ordinal forest, and neural network.
• Assessed performance using precision, recall, F1-score, weighted-average F1-score, and AUC value.

Main Results:

• Ordinal forest demonstrated strong in-sample performance (81.28% precision, 0.7249 AUC).
• Ordinal logistic regression showed strong out-of-sample prediction (77.12% precision, 0.7187 AUC).
• Classification accuracy was highest for mild FT and lowest for severe FT.

Conclusions:

• Ordinal forest and ordinal logistic regression models exhibit robust predictive capabilities for FT severity.
• The developed models and their threshold ranges can assist in clinical fundus disease screening.
• These findings support the application of machine learning in FT severity assessment.