Automating Radiology Report Generation: A Systematic Review of Deep Learning Methods

\textbf{Background:} Deep learning has revolutionized various fields of medical imaging, including radiology report generation. Automating radiology report generation can help alleviate radiologists' workload, reduce reporting inconsistencies, and enhance diagnostic accuracy. However, challenges such as data scarcity, model limitations, and clinical validation remain significant barriers to real-world implementation.\textbf{Objective:} This systematic review aims to synthesize existing research on deep learning-based radiology report generation, analyzing commonly used datasets, model architectures, evaluation metrics, and emerging trends.\textbf{Methods:} We conducted a comprehensive literature search across major scientific databases, selecting studies that applied deep learning techniques to generate radiology reports from medical images. Studies were categorized based on their methodologies, datasets, and evaluation approaches.\textbf{Results:} Our review of \textbf{356} studies reveals a shift from traditional CNN-RNN architectures to Transformer-based and multimodal models that incorporate both image and textual features. The most frequently used datasets include MIMIC-CXR and IU X-ray, while evaluation remains largely dependent on NLP metrics such as BLEU, ROUGE, and METEOR. Despite advancements, challenges persist in clinical accuracy, model interpretability, and real-world adoption.\textbf{Conclusion:} While deep learning has significantly advanced radiology report generation, critical issues such as data availability, evaluation standardization, and clinical integration must be addressed before widespread deployment. Future research should focus on developing knowledge-enhanced models, explainable AI techniques, and clinician-in-the-loop frameworks to ensure reliable and trustworthy AI-assisted radiology reporting.