Diabetic retinopathy (DR) is a leading cause of vision loss and blindness, affecting millions of people globally. The study presents a method for detecting DR in retinal images using a deep convolutional neural network (CNN) combined with a multi-dimensional fuzzy entropy thresholding approach. The goal is to accurately detect DR and determine its severity levels. A comprehensive retinal image method was developed to account for variations in retinal and image parameters, generating a dataset of 42,000 images representing a wide range of DR scenarios. Feasibility and consistency analyses were conducted using a Monte Carlo statistical method to evaluate the accuracy and reliability of the automated diagnosis system. The study identified specific conditions that ensure accurate DR detection: a minimum 30% difference between background and blood vessels, 15% between blood vessels and dark lesions for mild DR, 40% between background and blood vessels, 20% between blood vessels and dark lesions for moderate DR, and 30% between background and blood vessels, 15% between blood vessels and dark lesions, and 55% between blood vessels and bright lesions for severe DR. These conditions were verified against real retinal images, confirming their validity as reliable indicators of DR detection success. The study contributes to the advancement and practical application of automated diagnosis systems, providing healthcare professionals with a valuable tool for efficient and accurate detection and assessment of DR. Early detection and treatment are critical in preventing blindness from DR, emphasizing the importance of regular eye screenings for individuals with diabetes. The study highlights the urgent need for effective strategies to address the global challenges of diabetes and its associated eye complications.Diabetic retinopathy (DR) is a leading cause of vision loss and blindness, affecting millions of people globally. The study presents a method for detecting DR in retinal images using a deep convolutional neural network (CNN) combined with a multi-dimensional fuzzy entropy thresholding approach. The goal is to accurately detect DR and determine its severity levels. A comprehensive retinal image method was developed to account for variations in retinal and image parameters, generating a dataset of 42,000 images representing a wide range of DR scenarios. Feasibility and consistency analyses were conducted using a Monte Carlo statistical method to evaluate the accuracy and reliability of the automated diagnosis system. The study identified specific conditions that ensure accurate DR detection: a minimum 30% difference between background and blood vessels, 15% between blood vessels and dark lesions for mild DR, 40% between background and blood vessels, 20% between blood vessels and dark lesions for moderate DR, and 30% between background and blood vessels, 15% between blood vessels and dark lesions, and 55% between blood vessels and bright lesions for severe DR. These conditions were verified against real retinal images, confirming their validity as reliable indicators of DR detection success. The study contributes to the advancement and practical application of automated diagnosis systems, providing healthcare professionals with a valuable tool for efficient and accurate detection and assessment of DR. Early detection and treatment are critical in preventing blindness from DR, emphasizing the importance of regular eye screenings for individuals with diabetes. The study highlights the urgent need for effective strategies to address the global challenges of diabetes and its associated eye complications.