Bridging the islands of innovation: A machine-assisted Semantic–Bibliometric review and conceptual roadmap for closed-loop digital dementia care
Hairong Wang, Xingyu ZhangBackground
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.