Detection of Diabetic Macular Edema in Optical Coherence Tomography Image Using Convolutional Neural Network

Diabetic macular edema (DME) is known as the main cause of patients with Diabetic Retinopathy loss their vision. The vision loss can be prevented if DME could be detected and diagnosed in the early stage. The purpose of this research is to detect DME from retinal-Optical Coherence Tomography (OCT) images using Convolutional Neural Network (CNN). In this research, 2 pre-trained models using Transfer Learning namely MobileNet and VGG-16 and 1 custom CNN models were used to classify the retinal-OCT images. The dataset of the retinal-OCT images used in this research has been obtained from Kaggle website. The dataset is organized into 3 folders (train, validation, and test) and contains subfolders for DME image category and normal image category. There are 37,663 retinal-OCT images used in the training dataset, 484 images in the validation dataset, and 16 images in the testing dataset. In this research, the custom 5 layer CNN model was compared with the 2 pre-trained models to estimate the performance of DME detection. The results show both the 2 pre-trained models using Transfer Learning and the custom 5 layer CNN model could detect DME from retinal-OCT images. Compared with the 2 pre-trained models, the custom 5 layer CNN model distinguishing DME images from normal images achieved the highest accuracy of 96%.