A review of multimodal large models in the field of major diseases

Abstract:

The prevention and control of major diseases, such as cancer, cardiovascular and cerebrovascular diseases, and neurodegenerative disorders, remain core challenges in modern medicine. Their precise diagnosis and treatment critically rely on the integrative analysis of heterogeneous multi-source data, including medical imaging, electronic health records, and genomics. Traditional unimodal approaches, however, suffer from information silos and struggle to comprehensively characterize the complex biological mechanisms and clinical phenotypes of diseases. In response to this challenge, this paper systematically reviews the progress of multimodal large models (MLM) in major disease prevention and control. First, we summarize the transformer-centered technical paradigm, elucidating the underlying architecture and synergistic mechanisms that enable fusion of multimodal medical data. Second, we systematically survey applications across core clinical scenarios-early diagnosis, precise subtyping, and prognostic prediction, while deeply analyzing its technical potential and empirical value. Furthermore, we summarize common challenges encountered in practice, including data heterogeneity, the model “black box” problem, and ethical, legal, and data security issues. Finally, we outlook future development trends and propose key breakthrough directions, emphasizing clinically task-oriented model optimization, causal reasoning and enhanced interpretability, federated learning and privacy-preserving computation, and human-AI collaborative intelligent diagnostics. This review aims to provide a systematic reference for researchers, clinicians, and policymakers, promoting the clinical translation of multimodal large models in the prevention and treatment of major diseases, thereby empowering the high-quality development of precision medicine.