Multimodal Artificial Intelligence for Personalised Disease Diagnosis and Prognosis in Smart Healthcare Systems: A Critical Survey of Methods and Emerging Trends

Abstract

Limitations of a single diagnostic approach are being addressed by advanced solutions in smart healthcare thanks to multimodal Artificial Intelligence (AI). Heterogeneous data from medical imaging, electronic health records, genomic profiles, and different types of output from wearable sensors enable more accurate and personalised diagnosis and illness prediction. However, there are concerns for clinical translation concerning the interpretability, heterogeneity of data, and the safeguarding of privacy. A systematic literature search and critical synthesis have been performed based on PRISMA guidelines. The studies were sourced from IEEE Xplore, Scopus, PubMed, SpringerLink, and Web of Science. There were three types of multimodal AI approaches: Early, Late, and Hybrid Fusion. Compare and contrast classical neural models and basis, foundation, and evaluation. The evidence synthesis comprised thematic analysis to assess the robustness, scalability, and clinical applicability of the evidence, and performance measures of accuracy, AUC-ROC, precision, recall, and F1-score. In multimodal medical tasks, transformer-based architectures and deep fusion often outperform traditional machine learning approaches in prediction. Data sets are heterogeneous. Despite these developments, explainable AI algorithms are not sufficiently developed to explain in a clinical context. Although there are still computational efficiency issues with federated learning, it could be a privacy-preserving system. Key gaps include missing information, an imbalanced protocol, and the absence. Multimodal artificial intelligence greatly enhances personalised healthcare by improving diagnostic accuracy and the reliability of prognoses. However, issues of interpretability, privacy, and standardisation remain to be solved in practice. It supports a systematic approach to building transparent, scalable, and clinically deployable Intelligent Healthcare Systems.