Automated detection of diabetic retinopathy using machine learning classifiers.

Area of Science:

• Ophthalmology
• Medical Informatics
• Machine Learning

Background:

• Diabetic Retinopathy (DR) is a severe complication of diabetes, necessitating annual screening.
• Current screening methods face challenges due to high patient volume, resource limitations, and cost.
• Machine learning (ML) offers a promising approach for assisted diagnosis in medical science.

Purpose of the Study:

• To classify Diabetic Retinopathy (DR) in a Saudi population using various ML methods.
• To identify the most accurate ML model for DR classification.
• To pinpoint discriminative socio-demographic and clinical features associated with DR.

Main Methods:

• A cross-sectional study involving 327 diabetic patients in Saudi Arabia.
• Data collection included socio-demographic and clinical information via systematic random sampling.
• ML classifiers (LDA, SVM, KNN, Random Forest, Ranger Random Forest) were employed with cross-validation.

Main Results:

• Ranger Random Forest achieved the highest accuracy (86%) in classifying DR patients.
• HbA1c and diabetes duration were the most significant predictors of DR (p<0.001).
• Other influential factors included BMI, age of onset, age, systolic blood pressure, and medication use.

Conclusions:

• Integrating ML with ophthalmology can significantly improve DR diagnosis and clinical decision-making.
• ML serves as a valuable adjunct tool for clinicians, not a replacement.
• Future research will explore more advanced ML methods for multi-class DR data classification.