Automated radiology report generation (ARRG) has emerged as a critical research frontier, bridging advancements in computer vision, natural language processing (NLP), and clinical intelligence. This survey comprehensively reviews the most recent studies that propose methods for generating diagnostic reports from medical images. The reviewed approaches are organized by first examining the visual backbones adopted for feature extraction. Report generation strategies are then categorized into encoder-decoder models based on recurrent neural networks (RNN), transformer-based frameworks, knowledge-enhanced methods, and prompt-driven integrations of large language models (LLMs). Vision-language integration techniques, datasets, and evaluation metrics are also systematically analyzed. Although notable progress has been achieved in improving language fluency and partial clinical correctness, persistent challenges remain related to hallucination control, multi-modal generalization, and fine-grained visual grounding. Limitations of current datasets and evaluation methods are identified, highlighting the gap between research advancements and real-world clinical deployment. This survey concludes by outlining future directions aimed at enhancing factual consistency, interpretability, and clinical robustness in medical report generation systems.