Machine Learning-Based Detection of Diabetic Retinopathy in Female Patients Aged 30–50 Years

Diabetic Retinopathy (DR) is an eye disease that can cause vision damage and blindness if it is not detected at initial stages. This study proposes a deep learning-based approach for predicting Diabetic Retinopathy from retinal fundus images, only focusing on female patients aged 30–50 years. The ODIR-5K dataset was used as the source of retinal images. After filtering the dataset according to gender and age a final dataset of 202 patients and 404 fundus images was considered, consisting of 103 DR patients and 99 No-DR patients. A deep learning model called EfficientNetB0 was used to classify the retinal images. The data was divided into training and testing sets at the patient level. The model achieved 56.10% accuracy, 56.82% precision, 59.52% sensitivity, 52.50% specificity, 58.14% F1-score, and 55.00% ROC-AUC. An interactive dashboard was also developed to demonstrate the model's predictions using the selected images. The results show that deep learning can be used for retinal image analysis. The model needs further improvement and testing with larger datasets before it can be considered for real-world medical use.