Machine learning in radiology: current applications, challenges, and future directions—a narrative review

Background and Objective: Machine learning (ML) has significantly advanced the field of radiology through enhanced image interpretation, workflow optimization, and prediction modelling. While the earlier reviews focused on task-oriented deep learning techniques, the recent developments in foundation models, generative artificial intelligence (AI), and large language models (LLMs) have opened new avenues for radiological innovations. The objective of this narrative review is to critically analyze the current applications of ML in radiology, evaluate the quality of existing evidence, and discuss emerging trends influencing radiological innovation.

Methods: Using electronic databases such as PubMed, Scopus, Web of Science, and IEEE Xplore, a systematic search of the literature was conducted between January 1, 2015, and July 31, 2025. The final search was conducted on July 31, 2025. Included were pertinent English-language peer-reviewed articles on ML in diagnostic radiology.

Key Content and Findings: A total of 80 studies met the inclusion criteria, encompassing applications in image acquisition and reconstruction, disease detection, segmentation, prognostic modelling, and workflow optimization. Convolutional neural networks (CNNs) demonstrate promising diagnostic performance, as seen in the context of mammography and lung nodule detection. Federated and self-supervised learning techniques are also being used to improve the availability of data, whereas foundation models and generative AI are used for multimodal data analysis. Additionally, LLMs are used for reporting and decision support, although many of the studies are still retrospective and lack strong external validation, indicating challenges in reproducibility and implementation.

Conclusions: ML is evolving toward integrated, multimodal systems in radiology. Sustained clinical impact requires rigorous validation, ethical oversight, explainability, and interdisciplinary collaboration.