부산대학교 도서관

Medical image classification plays a crucial role in monitoring and detecting diseases. This paper presents deep learning methods to distinguish images taken by the Optical Coherence Tomography technique from the normal eye and three eye-related diseases named Diabetic Macular Edema (DME), Choroidal neovascularization (CNV), and DURSEN. To achieve this aim, the images undergo a patch extraction process; then, the extracted patches are treated as sequences, and Recurrent Neural Networks are implemented to classify the images. Four pre-trained models, including VGG16, ResNet152V2, NasnetMobile, and Densnet169, and a vision transformer model are also applied and compared. Based on the results, The proposed model has achieved 99.38% test accuracy, higher than other models.