Multimodal generative artificial intelligence (AI) has emerged as a transformative technology in clinical diagnostics, integrating diverse data sources—medical imaging, genomic profiles, clinical narratives, and electronic health records—to significantly enhance diagnostic accuracy, clinical decision-making, and personalized patient care. This review systematically explores the landscape of multimodal AI across key medical specialties, including radiology, pathology, dermatology, ophthalmology, neurology, and oncology, highlighting recent methodological advancements, performance evaluations, and practical clinical implementations. Technical strategies such as tool-use, grafting, and unified multimodal architectures are critically assessed, identifying their strengths and limitations concerning clinical applicability, interpretability, and computational efficiency. Synthetic multimodal data generation methodologies—Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), diffusion models, and large language models (LLMs)—are evaluated for addressing data scarcity in rare disease research, enhancing international collaboration, and mitigating privacy concerns. Additionally, this review addresses pivotal ethical, regulatory, and liability challenges, emphasizing fairness, transparency, and accountability in AI-driven clinical diagnostics. Strategic priorities for future research are identified, including rigorous prospective clinical validation, development of standardized multimodal datasets, enhanced model interpretability, and robust regulatory frameworks. Ultimately, realizing the transformative potential of multimodal generative AI in clinical practice will require interdisciplinary collaboration among clinicians, researchers, ethicists, regulators, and patient advocacy groups, ensuring these powerful tools effectively augment human expertise, improve healthcare delivery, and advance precision medicine.
Maleki et al. (Mon,) studied this question.