Deep Learning-Based Diabetic Retinopathy Detection Using ResNet34 Model

Using the APTOS 2019 dataset and ResNet-34 architecture, this study presents an approach utilizing deep learning for classification of retinal images. Diabetes-related retinopathy and age-related macular degeneration are among the leading causes of blindness, and early detection is crucial for effective treatment. Convolutional neural networks (CNNs) have been shown to extract meaningful features from retinal images and accurately classify them by severity. In this study, the model was trained and validated using high-quality retinal images with varying disease severity from the APTOS 2019 dataset. ResNet-34, a 34-layer deep learning model renowned for image recognition, was used in the experiment. The model was trained on Google Colab using the PyTorch deep learning framework, with a batch size of 32 and 20 epochs. The introduced method demonstrated robust performance across various evaluation metrics, such as exactitude, recall, harmonized metric, Area Under the Receiver Operating Characteristic Curve (AUC ROC), and correct classification ability. Results showed that the proposed method accurately predicted the target variable. The study also presents loss and accuracy plots, confusion matrices, and AUC ROC plots, which provide a comprehensive analysis of the performance of the model which can be taken as a suggestion for improvement.