Multimodal and fair artificial intelligence in paediatric rheumatology: challenges and opportunities for a holistic and equitable future

Multimodal artificial intelligence (AI) is an emerging domain comprising a set of tools with potential clinical relevance in paediatric rheumatology, a field characterised by rare, heterogeneous diseases and diagnostic delays. This narrative review synthesises current evidence on how multimodal and fair AI could reshape diagnosis, monitoring, and treatment for children with rheumatic diseases, while examining the risks of exacerbating existing inequities. Complex architectures fuse clinical data, laboratory parameters, imaging, omics profiles, and wearable signals into dynamic ‘digital phenotypes’ capable of predicting flares, treatment response, and long-term outcomes. Explainable AI approaches render high-capacity models more actionable and trustworthy. Large language models can translate complex clinical information into family-centred plain language; however, a human-in-the-loop approach remains essential to limit hallucinations, omissions, and confidentiality breaches. Bias accumulation across the AI lifecycle demands equity-centred design. Emerging ethical, methodological, and regulatory frameworks should guide responsible adoption, outlining a roadmap toward a more equitable future for children with rheumatic diseases worldwide.