Using Artificial Intelligence for Analyzing Retinal Images (OCT) in People with Diabetes: Detecting Diabetic Macular Edema Using Deep Learning Approach

Abstract

Medical imaging evolved rapidly to play a vital role in diagnosis and treatment of a disease.  Automate analysis of medical image analysis has increased effectively through the use of deep learning techniques to obtain much quicker classifications once trained and learn relevant features for specific tasks, shown to be assessable in clinical practice and valuable tool to support decision making in medical field. Within Opthalmology, Optical Coherence Tomography (OCT) is a volumetric imaging procedure that uses in the diagnosis, monitoring and measuring response to treatment in eyes. Early detection of eyes diseases including Diabetic Macular Edema (DME) is vital process to avoid complications such as blindness. This work employed a deep convolutional neural network (CNN) based method for DME classification task. To demonstrate the impact of convolutional, five models with different Convolutional layers built then the best one selected based on evaluation metrics. The accuracy of model improved while increasing the number of Convolutional Layers and achieved 82% by 5-Convolutional Layer,  Precision and Recall of CNN model per DME class was 87%% and 74%, respectively. These results highlighted the potential of deep learning in assisting decision-making in patients with DME.