Identifying the severity of diabetic retinopathy by visual function measures using both traditional statistical...

Area of Science:

• Ophthalmology
• Medical Informatics
• Biomedical Engineering

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Accurate staging of DR is crucial for timely intervention.
• Current methods may not fully leverage multimodal data for DR classification.

Purpose of the Study:

• To assess the efficacy of machine learning (ML) in classifying diabetic retinopathy severity.
• To determine the contribution of visual function measures, age, and sex to DR classification.
• To compare ML model performance against traditional logistic regression.

Main Methods:

• Visual function of 1901 eyes was assessed using nine tests.
• Participants were categorized into four groups: no diabetes, diabetes without DR, DR without DMO, and DR with DMO.
• Ensemble ML models and logistic regression were used for classification tasks.

Main Results:

• Ensemble ML models achieved high accuracies (0.92, 1.00, 0.84) for DR classification tasks.
• ML models significantly outperformed logistic regression.
• Key visual function variables included reading index, near visual acuity, and microperimetry.

Conclusions:

• ML models effectively predict diabetic eye disease status using readily available clinical data.
• Interpretable ML aids in understanding visual function profiles associated with DR stages.
• These methods show promise for developing robust prediction models from complex clinical datasets.