Bridging the islands of innovation: A machine-assisted Semantic-Bibliometric review and conceptual roadmap for closed-loop digital dementia care.

BACKGROUND: Artificial intelligence (AI), extended reality (XR), and socially assistive robotics (SAR) are each advancing Alzheimer's disease (AD) research and care at a rapid pace. Yet despite substantial progress within each domain, clinical implementation remains weakly integrated across diagnostic, therapeutic, and care functions, producing islands of innovation rather than coordinated care systems. METHODS: We conducted a machine-assisted semantic-bibliometric synthesis of 2,636 publications on AI-, immersive technology/VR-, and SAR-enabled approaches to AD diagnosis, intervention, and care published between 2021 and 2025. Available title-abstract metadata were encoded using SBERT embeddings, projected via UMAP, and clustered using K-Means to characterize the functional topology of the field. From this mapped corpus, we selected a semantically central subset of 50 studies for high-fidelity full-text synthesis, preserving cross-domain representativeness while maintaining interpretive tractability. RESULTS: The mapped landscape suggests a three-part pattern of architectural separation. Precision neuroimaging (Cluster C4, 22%) functions primarily as a state-oriented diagnostic domain. Immersive therapeutics (Cluster C2, 48%), the largest cluster, increasingly incorporate adaptive personalization but remain weakly connected to biomarker-based stratification. Embodied robotic care (Cluster C5, 10%) addresses behavioral stabilization with little longitudinal coupling to upstream sensing. Across all three domains, high component-level sophistication coexists with limited evidence of cross-layer coordination. CONCLUSION: Contemporary digital solutions for AD remain predominantly siloed and state-oriented rather than longitudinally integrated. We synthesize these observations into a conceptual Closed-Loop Architecture, proposed as a roadmap for future integrated digital dementia care. Advancing this agenda will depend on interoperable infrastructure, longitudinal modeling, and prospective cross-layer evaluation-not solely on further isolated gains in classification accuracy.