image: Advanced deep learning (DL) and large language models (LLMs) decode multimodal imaging-omics associations across modalities, scales, organs, and diseases. The framework enables interpretable analysis for generating disease atlases, predicting progression trajectories, and optimizing therapies, ultimately bridging biological mechanisms to clinical practice.
Credit: Created with http://BioRender.com
Despite significant advancements in human genomics, the phenotypic and clinical relevance of genomic features remains largely unknown. Imaging genomics, also known as radiogenomics, was proposed to find the associations between clinical image data and human genetic data, facilitating a better understanding of the molecular characteristics of diseases.
In recent years, imaging genomics has been extended to the analysis of multi-modal data, where the imaging data includes CT, MRI, X-rays, and ultrasonography and the molecular data includes genomic, transcriptomic, proteomic data. The applications of imaging genomics have extended to different kinds of diseases, including cancers and cardiovascular diseases.
In recent years, with the evolution of computing technology, experimental technology, and application, the concept of image genomics has been constantly updated. This evolution provides guidance for defining imaging genomics over the next decade. In this regard, this perspective points out two major directions that should be broken through in imaging genomics in the next decade: the development of cross-modal foundation models and the realization of precision medicine.
The development of large-scale model technology has shown great potential for multimodal data analysis, especially in single-cell and medical image analysis. However, the current research lacks a foundation model for integrating image and omics data, and the key difficulties include the integration of cross-scale image and omics data, cross-modal interpretable analysis, and computing power requirements.
Imaging genomics effectively complements the limitations of medical electronic medical record data in phenotypic characterization, providing a new perspective for precision medicine. However, the current translation of image genomics into clinical practice has certain limitations. On the one hand, existing studies lack analysis of longitudinal data. Second, these analyses are often limited to identifying existing diagnostic or therapeutic methods and have not yet uncovered image biomarkers by correlating image data with omics data to improve the potential of current diagnosis and treatment. In addition, recent breakthroughs in cross-modal translation from image data to omics data suggest a future shift in imaging genomics for treatment optimization. Systematic analysis of cross-organ or cross-disease associations is another important aspect of imaging genomics-guided precision medicine. Incorporating existing theories of cross-organ connectivity or cross-disease linkages into imaging genomics analysis will enable a systematic understanding of disease development, thereby facilitating accurate and early diagnosis of disease.
The perspective pointed out the roadmap for imaging genomics in the next decade.
“In the next decade, imaging genomics will shift from retrospective validation to systems biology modeling, promoting the discovery of new biomarkers and therapeutic targets, and providing novel tools for precision medicine.” Xiao Ping Cen, PhD, College of Life Sciences, University of Chinese Academy of Sciences, and first author of the article, said.
The potential of imaging genomics is huge, and with the development of large language models and the maturity of global data collaboration networks, imaging genomics may be potentially applied to precise clinical treatment decisions for every patient.