RetDense: A Fine-tuned DenseNet121 Model for Retinal Eye Disease Detection

Retinal diseases, which cover a variety of eye disorders, are one of the biggest problems threatening human vision. As diabetic retinopathy, age-related macular degeneration and glaucoma rise in incidence, more urgent than ever are effective diagnostic tools. This abstract summarizes the latest achievements in retinal disease detection, with a focus on emerging AI and image analysis technologies. The application of deep learning algorithms for automated retinal abnormalities detection and classification has achieved unprecedented success. These models show great sensitivity and specificity, better than traditional diagnostic methods for the earlier identification of disease. Furthermore, the use of large-scale datasets and advanced imaging methods including optical coherence tomography (OCT) as well as fundus photography has greatly increased diagnostic systems 'precision. In the proposed work, a deep learning-based DenseNet121 model has been used and fine-tuned to predict retinal eye disease. The model has been trained with Adam and SGD optimizer with different values of learning rates. The highest accuracy has been achieved while implemented with a learning rate value of 0.01. The training accuracy of 93.68% and testing accuracy of 95.87% have been identified, whereas, the training loss and testing loss of 0.101 and 0.048 have been resulting at epoch 50 with a learning rate of 50. The model shows an optimal performance with the Adam optimizer, therefore, this work can also be extended for other applications of healthcare departments.