Early identification is essential to prevent serious visual impairment in diabetic patients as Diabetic Retinopathy (DR) is a main reason for blindness. In this paper, an optimal Convolutional Neural Network (CNN) model is used to propose an automated approach for categorizing the stages of DR. The pre-trained VGG16 model uses deep feature extraction for the given retinal pictures, which uses scaling and feature selection heuristics by the Grey Wolf Optimizer. Those selected features from GWO belonging to the most relevant features would also give a better boost to the classification performance along with the optimization of both hyperparameters. The proposed model will perform better than the traditional techniques based on the Precision, Recall, and F1 scores’ experimental results. It has an accuracy of 99.31% on the DR dataset. The inclusion of GWO in CNN, models hold tremendous potential for use in the analysis of medical images and yields Optimized CNN, an efficient technique that is effective in improving healthcare diagnosis.