Diabetic retinopathy (DR) can lead to blindness without timely detection and treatment. Traditional diagnosis, through ophthalmologic examinations, is laborious, expensive, and limited in remote areas. Early detection of DR presents challenges as it involves manual review of retinal images, making it a slow and costly process. Machine learning (ML) emerges as a promising alternative. As a branch of artificial intelligence, it allows computers to learn tasks without explicit programming. In detecting DR, ML can train models to identify signs of the disease in retinal images. This article reviews advances in ML models to detect DR, demonstrating efficacy in early detection. Models based on convolutional neural networks (CNN), achieve accuracy comparable to that of physicians or optometrists with greater precision by overcoming human subjectivity, speed in image analysis, and utility in remote environments. Although further studies are needed to confirm results and address challenges, ML emerges as a valuable tool in the fight against blindness caused by DR.