This research offers a complete investigation into the effectiveness of various neural networks and Deep learning algorithms for retinal image class, which is essential for Diabetic Retinopathy(DR)detection. To Facilitate the robust assessment, the proposed study leverages five different retinal image benchmark datasets – DRIVE, DIARETDB0, CHASEDB, STARE and MESSIDOR. The four different algorithms- Maximum Principal Curvature(MPC), Multilayer Perceptron(MLP), Dense Convolutional Neural Network(DCNN) and Discriminative CNN with 121 layers(Dis-CNN) are employed to examine the overall performance through different metrics, namely Accuracy, Sensitivity and Specificity. In this study, every algorithm is subjected to rigorous evaluation based on the unique characteristics of the dataset.The primary goal is to determine the algorithm's overall efficacy and identify the nuanced performance outcome of retinal images in the different datasets. This study outlines the strengths and differences of every set of algorithms in dealing with the facts of the retinal images. The selected algorithms include a range of approaches, from conventional techniques like Maximum Principal Curvature (MPC) to better neural network architectures along with MLP, Dense CNN, and a Discriminative CNN with 121 layers. The variety in algorithms and datasets ensures a comprehensive evaluation, providing precious insights into their suitability for real-world applications in Diabetic Retinopathy.Discriminative CNN with 121 layers is a better alternative algorithm across all datasets with accuracies ranging from 95.01% to 99.83%. This outstanding performance underscores the robustness of the Discriminative CNN, affirming its potential as a highly effective tool for the accurate DR classification and diagnosing abnormalities