Fig 1.
Overall framework.
Fig 2.
Architecture of ANN.
Fig 3.
Top pupillometry features identified by the Light Gradient Boosting Machine (LightGBM) importance scores for disease classification.
The features include demographic factors such as age, gender and pupillary dynamics such as Hippus cycle time, dark-adapted baseline diameter, constriction and re-dilatation amplitudes, their respective times (t₁, t₂), and velocities of constriction and dilatation (mm/s).
Table 1.
Baseline pupillary dynamics and their association with diabetic retinopathy.
Table 2.
Cut-off values and diagnostic accuracy metrics for pupillary features in diabetic retinopathy prediction.
Table 3.
Comparative performance metrics of Machine Learning Models across Training, Test, and Validation Sets.
Fig 4.
Comparison of Receiver Operating Characteristic (ROC) curves for model performance across training, validation, and testing phases.
Models compared include Decision Tree Classifier, K-Nearest Neighbors (KNN), Logistic Regression, Naïve Bayes Classifier, Neural Network, Random Forest Classifier, and Support Vector Machine (SVM). The Area Under the Curve (AUC) values are reported for each model.
Table 4.
Test set performance metrics of machine learning models for DR prediction (95% CI reported for accuracy and AUC).