Artificial Intelligence Platform Architecture for Hospital Systems: Systematic Review.

Background: The construction of artificial intelligence (AI) platforms in hospitals is the backbone of the revolution in health care. While traditional hospital information systems have facilitated digitalization, they are still limited by data silos, fragmented workflows, and insufficient clinical intelligence that impede organizations from realizing the promise of data-led decision-making. Objective: This study aimed to derive a hospital-specific 5-layer architecture (infrastructure, data, algorithm, application, and security and compliance) and to systematically review the evidence mapped onto the 5-layer framework to assess its applicability. Methods: A systematic literature search was performed in Web of Science, Embase, PubMed, and Scopus from inception to May 2025. The review followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Studies were screened and selected for full-text review by two independent reviewers. We included peer-reviewed empirical studies describing hospital-based AI implementations across clinical domains. Reviews, commentaries, and purely technical bench studies without hospital context and non-English literature were excluded. Quality assessment of the identified papers was conducted using the Critical Appraisal Skills Programme tool. Using a 0 to 5 point ordinal maturity scale of 5 layers, we conducted a structured mapping with quantitative mapping, weighted co-occurrence analysis, weighted Jaccard similarity, and thematic synthesis with examples. Results: In total, 29 studies met the eligibility criteria and included work specifically in emergency, radiology, routine imaging, chronic disease, and multihospital platform work, conducted in 11 countries. On average, the application (mean 3.17, SD 0.85) and data (mean 3.00, SD 0.76) layers demonstrated the highest maturity, followed by algorithm (mean 2.79, SD 0.77) and infrastructure (mean 2.79, SD 1.70). The security and compliance layer showed the lowest and most variable maturity (mean 1.69, SD 1.89). Weighted co-occurrence and Jaccard analyses revealed strong interconnections among data, algorithm, and application (Jaccard=0.80-0.89), forming a technical core, whereas security and compliance exhibited weak alignment (0.43-0.46). Conclusions: Our review excluded non-English and gray literature, which may limit comprehensiveness. The ordinal maturity scoring may still simplify the contextual complexity of hospital AI implementations. Our synthesis validates a 5-layer hospital AI platform architecture, grounded in both theoretical frameworks and empirical evidence. The findings highlight that while clinical feasibility is achievable, sustainable hospital-wide AI requires stronger investment in infrastructure, data governance, and compliance.