Machine Learning for Early Detection of Diabetic Retinopathy: Leveraging DenseNet CNNs

The research overcomes the limitations of existing diabetic retinopathy detection systems by proposing an innovative system based on DenseNet CNNs. Existing systems have limitations, such as manual examination, which can result in inaccuracies and time-consuming procedures. The proposed system improves efficiency and accuracy using the cutting-edge machine learning technique DenseNet CNNs. Comprehensive datasets, preprocessing, DenseNet architectural modification, and training processes are utilized. In comparison to the existing system, the results show increased performance in accuracy ($98 \%$), sensitivity $(98.7 \%)$, specificity $(96 \%)$, and F1-Score $(93 \%)$. The system’s limitations include low accuracy (97%) and specificity $(94.5 \%)$. The proposed system’s autonomous feature extraction with DenseNet CNNs eliminates the need for user intervention while enhancing adaptability. These enhancements not only improve diagnosis accuracy but also give practical benefits by automating activities and lowering the workload of healthcare workers. The development has the potential to transform diabetic retinopathy identification, allowing for more rapid treatments and better patient care in clinical settings.