The application of convolutional neural networks (CNN) for feature extraction and categorization of medical images has advanced to an impressive level. With better architectures and improved parameters, their efficiency is improving day by day. Using optical coherence tomography (OCT) images, this research proposes a CNN-based automated diagnosis system for diabetic macular edema (DME). DME is a significant cause of visual impairment in diabetic patients globally. As a result, it is critical to creating an automated diagnostic system capable of detecting DME symptoms as early as possible. The proposed research compares the performance of three CNNs (a proposed CNN, VGG-16, and DenseNet) for DME classification in two categories: normal and DME. The proposed CNN outperformed VGG-16 & DenseNet with a lower system loss and exceeded significantly in terms of classifier accuracy, precision, f1-score, and receiver operating characteristics (ROC). The proposed CNN obtained classifier accuracy of 90.26%, while VGG-16 & DenseNet obtained 87.41% & 82.66% respectively. The area under ROC for the proposed CNN, VGG-16 & DenseNet is 0.96, 0.96 &0.92 respectively.