Large language models in radiogenomics: a comprehensive survey of applications from imaging to genetics

Large language models are now widely used to better understand and process biomedical data in precision medicine to improve disease diagnosis and treatment responses. This survey provides a comprehensive overview of the current landscape of Large Language Models (LLMs) by examining key technologies, their evolution, and real-world applications in medical imaging and radiogenomics. LLMs can interpret biological sequences and integrate multi-omics and imaging data to picture clinically meaningful insights, which is fundamental to radiogenomics research. LLMs can read and interpret DNA and RNA sequences. They can also find important regulatory elements and help discover new biomarkers, supporting critical applications in radiogenomics such as linking genetic variations to imaging phenotypes. This makes them very useful for biomedical research and radiogenomics. Because they are flexible, LLMs can be applied to numerous genomic and proteomic tasks with little extra training, supporting seamless integration into radiogenomics pipelines. We begin by outlining the foundational principles of LLMs and their growing significance in healthcare, particularly in tasks highly relevant to radiogenomics workflows, including diagnostic support, image captioning, multimodal data fusion, and automated reporting. Together with a discussion on the many advantages that LLMs bring to radiogenomics, we also inspect their present limitations. Key challenges, such as limited interpretability, data sparsity, significant computational demands, and issues of generalization, are discussed in detail, particularly in the context of applying LLMs to high-dimensional radiogenomics datasets, in this survey. The survey also reveals how AI research is evolving, particularly in areas involving sensitive genetic and clinical imaging data, toward creating systems that are easier to understand, protecting user privacy, and enhancing performance factors crucial for radiogenomics. This survey outlines future directions and openings for revolution, focusing on radiogenomics as a promising application domain at the crossroads of LLMs, imaging, and genetic sciences.