Diabetic Retinopathy (DR) is a progressive eye disease caused by high blood glucose levels that leads to retinal lesions and significant vision impairment. It is a major contributor to blindness, particularly among the working-age population in developed countries. Due to its irreversible nature, early detection of DR is essential to preserve vision. However, manual diagnosis is often time-consuming and costly and relies on a detailed analysis of images of the retinal fundus by trained ophthalmologists. To address these challenges, machine learning-based medical image analysis has emerged as an effective solution, with deep learning algorithms that enable an early and accurate diagnosis of DR. This paper explores various machine learning approaches, focusing on advanced deep learning techniques, including supervised learning, self-supervised learning, and vision transformers, for image analysis of the retinal fundus. The study investigates key tasks in the detection, classification, and segmentation of blood vessels with DR, providing a comprehensive evaluation of methodologies and their potential to improve diagnostic efficiency and accuracy.