Diabetic retinopathy (DR) is the third leading cause of irreversible blindness worldwide and the leading cause among adults of working age, underscoring the critical need for adequate screening. The quality of the information is a key aspect in any image analysis process and is essential for the correct analysis by automatic processes in pathologies of great importance such as DR. This study aimed to generate a low-cost computational tool that preprocesses fundus images to optimize time in the timely referral of patients. Using two public databases, HDR and DRIMDB, which include both good and poor-quality images, we performed Retinal Image Quality Assessment (RIQA) by analyzing entropy and Peak Signal-to-Noise Ratio (PSNR) under Salt & Pepper and Gaussian Blur noise conditions. Additionally, a vector based on vessel segmentation was created using the Galdran et al. algorithm. The data were evaluated on the WEKA machine learning platform, revealing that neither entropy nor PSNR with Salt & Pepper noise were effective in classifying medically useful images. However, PSNR with Gaussian Blur, especially when combined with vessel segmentation, proved to be a reliable indicator of image quality. These findings suggest that images should be classified as gradable or non-gradable to enhance the efficiency and accuracy of DR screening systems.