Diabetic retinopathy (DR) is a major cause of visual impairment among working-age populations. In recent years, artificial intelligence (AI) has demonstrated significant application value in DR diagnosis, leveraging core advantages such as high efficiency and low error rates. Currently, the technical system of AI in DR image diagnosis mainly includes links like image preprocessing, feature extraction, diverse algorithmic models, and dataset construction. In practical applications, AI models can achieve automated screening and grading diagnosis of DR images, enhance diagnostic efficiency by integrating multimodal technologies, and have been successfully applied to mobile devices; meanwhile, the development of explainable AI has further boosted the credibility of AI models. Currently, this field still faces challenges, including insufficient data quality and scale, limited model interpretability, inadequate clinical validation, ethical and privacy risks, and a lack of unified technical standards. In the future, with continuous technological breakthroughs and the establishment of standardized evaluation systems, the reliability and accessibility of AI in DR diagnosis will be further enhanced.