Diabetic Retinopathy Detection Using Convolutional Neural Network

Abstract

In persons with diabetes, diabetic retinopathy (DR), a serious eye condition, can cause blindness. Improved patient outcomes and the prevention of vision loss are possible with early DR identification. Convolutional neural networks (CNN), for example, have demonstrated considerable potential in automating the identification of DR. This paper's goal is to investigate CNN's effectiveness in DR detection. The suggested methodology entails using a publicly accessible collection of retinal pictures to train a CNN model. In order to categorize images as having no DR, mild DR, moderate DR, or severe DR, the model is trained. A number of metrics, including sensitivity, specificity, and the area under the receiver operating characteristic (ROC) curve, are used to assess the model's performance. The study's findings demonstrated that the CNN model had a DR detection accuracy of 84.26%.. The model had a sensitivity and a specificity of 81.34% and 78.67%, respectively. The ROC curve's area under it showed good performance in DR detection at 0.82. The performance of the CNN model was also evaluated in comparison to that of other established machine learning techniques like Random Forest and Support Vector Machines. The CNN model performed more accurately and sensitively than these conventional techniques. The findings of this study show that CNN has a lot of potential for automated DR detection. With the purpose of increasing the precision and effectiveness of DR diagnosis, the proposed methodology can be included into current clinical workflows. The study also emphasizes the significance of enhancing medical imaging's capacity for disease identification and diagnosis through the use of deep learning techniques.