Foundation Model in Radiology

In recent years, AI has undergone a significant paradigm shift. The emergence of large foundation models, including large language models (LLMs), large vision models (LVMs), and large multimodal foundation models (LMMs), has revolutionized the field, showcasing unprecedented capabilities and delivering impressive performance across various natural language understanding, computer vision, and multimodal tasks. One particularly promising application area of foundation models is in radiology. Radiologists are now confronted with pressures and challenges driven by the increasing volume and complexity of medical imaging studies. The exceptional capabilities of foundation models hold immense potential to support radiologists in numerous tasks, such as triaging cases, generating reports, identifying abnormalities, and enhancing diagnostic confidence. Therefore, exploring how to effectively unlock the potential benefits of foundation models in radiology and establish radiology-specific foundation models is a crucial and valuable research topic.

The goal of this research topic is to highlight and advance research on the application of foundation models in radiology. This topic aims to bring together cutting-edge studies that explore the development, adaptation, and implementation of large language models (LLMs), large vision models (LVMs), and large multimodal foundation models (LMMs) in the radiological domain. Through innovative research, we aim to foster a deeper understanding of how these models can be tailored to meet the unique challenges of radiology, such as how to harness increasing imaging volumes and address complexity of diagnostic tasks. This topic seeks contributions that demonstrate the potential of these models in enhancing clinical workflows, improving diagnostic accuracy, and ultimately advancing patient care. Through this focused collection of research, we aspire to pave the way for the integration of advanced AI technologies into radiological practice, thus supporting radiologists and finally improving healthcare outcomes.

We invite original research articles and systematic reviews that showcase the latest breakthroughs, challenges, and future directions in this emerging field of radiology-specific foundation models. We welcome submissions from scholars working at the intersection of artificial intelligence, machine learning, and domain applications.

Topics of interest include, but are not limited to:
• New neural network architecture design for radiology-specific foundation models.
• New training, fine-tuning, and adaptation methods for radiology-specific foundation models.
• New multimodality alignment algorithms for radiology-specific foundation models.
• New paradigm for incorporating human feedback for aligning radiology-specific foundation models.
• New methods for optimizing, accelerating, and compressing radiology-specific foundation models.
• New frameworks for fusing and integrating multiple foundation models and their applications in radiology.
• Privacy-preserving training of foundation models their applications in radiology.
• Interpretability and explainability of radiology-specific foundation models.
• Trustworthiness of radiology-specific foundation models, such as ethical considerations and bias mitigation.
• Evaluation metrics, open datasets, and benchmarking for radiology-specific foundation models.

Keywords: foundation models, radiology, radiology-specific foundation models, the latest breakthroughs, challenges, future directions

Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.