This paper investigates the impact of fundus image enhancement techniques on improving diabetic retinopathy (DR) prescreening accuracy using machine learning. DR, a prevalent complication of diabetes and a leading cause of vision impairment and blindness, necessitates precise and early detection to prevent severe outcomes. We explore various image enhancement methods, including histogram equalization, contrast stretching, contrast-limited adaptive histogram equalization (CLAHE), and noise reduction, to optimize the quality of retinal images used for analysis. Enhanced images are evaluated using a convolutional neural network (CNN) tailored for DR classification, assessing the impact of these techniques on model accuracy, sensitivity, and specificity. Our findings reveal a significant improvement in prescreening performance, with enhanced imaging techniques leading to more reliable and earlier diagnosis of DR, achieving 90% accuracy by CLAHE with CNN-LSTM. This research highlights the critical role of high-quality image input in machine learning algorithms and the potential integration of these techniques into automated screening systems. Future research will focus on the real-time application of these techniques in diagnostic frameworks and their effectiveness in other retinal conditions, ultimately contributing to better screening strategies and improved patient care in managing DR.