Explainable Machine Learning Model for Glaucoma Diagnosis and Its Interpretation.

Area of Science:

• Ophthalmology
• Medical Informatics
• Machine Learning

Background:

• Glaucoma diagnosis relies on various clinical tests, including visual field testing, RNFL OCT, IOP measurement, and fundus photography.
• Accurate and early diagnosis of glaucoma is crucial for preventing vision loss.
• Existing diagnostic methods can be complex and may benefit from advanced computational approaches.

Purpose of the Study:

• To develop and evaluate machine learning models for glaucoma diagnosis.
• To create an explainable AI system to interpret specific diagnostic predictions.
• To assess the performance of different machine learning algorithms in glaucoma detection.

Main Methods:

• Clinical data from visual field tests, RNFL OCT, IOP measurements, and fundus photography were used.
• Five key features were selected for model development.
• Support vector machine, C5.0, random forest, and XGboost algorithms were trained and evaluated using 10-fold cross-validation.
• Explainable AI techniques, including gauge, radar, and SHAP charts, were employed.

Main Results:

• All four machine learning models demonstrated high diagnostic performance, with accuracy ranging from 0.903 to 0.947.
• The XGboost model achieved the highest accuracy (0.947), sensitivity (0.941), specificity (0.950), and AUC (0.945).
• Three statistical charts were developed to explain the XGboost model's predictions, enhancing interpretability.

Conclusions:

• The XGboost model offers superior diagnostic performance for glaucoma compared to other tested algorithms.
• Explainable AI charts provide valuable insights into the model's decision-making process for specific predictions.
• This study represents a novel application of explainable AI in the diagnosis of eye diseases, potentially improving clinical decision support.