Intelligent Real-Time Healthcare and Biomedical Monitoring Systems: A Narrative Review of AI, IoT, and Emerging Technologies
Background: Chronic disease management, an ageing global population, and the aftermath of the COVID-19 pandemic have pushed real-time, continuous health monitoring from a research curiosity toward routine clinical practice. Artificial intelligence (AI), the Internet of Medical Things (IoMT), edge computing, and next-generation wireless networks are converging to enable systems that sense, interpret, and act on physiological data outside the traditional hospital setting. Methods: This is a narrative review, not a PRISMA-guided systematic or scoping review; it synthesises 75 sources (a mixture of primary studies, systematic/scoping reviews, and meta-analyses), individually verified against their published record. The topics span IoMT system architecture and security, wearable and implantable sensing, deep learning for electrocardiogram (ECG), fall-related and human-activity signal analysis, edge and TinyML deployment, federated learning, blockchain-based health-record security, medical imaging diagnostics, explainable AI (XAI), continuous glucose monitoring, 5G/6G-enabled telemonitoring, digital twins, consumer-grade and contactless cardiac sensing, neurological and mental health monitoring, and the materials, regulatory, and acute care infrastructure surrounding real-time deployment. Studies were organised into a five-layer architectural taxonomy spanning perception, edge, network, cloud, and application layers. Results: Reported accuracies for deep learning models on ECG arrhythmia classification range from 91% to 99.5% across the reviewed studies, but the figures come from different datasets, class definitions, and validation protocols and are therefore not directly comparable; within this heterogeneous evidence, edge-deployed models report accuracies in the 85–96% range at substantially reduced power budgets. Deep-learning-based fall detection and chest radiograph classification are each reported, in the individual studies reviewed, to outperform threshold-based or classical alternatives, though this has not been established through head-to-head comparison across the full evidence base. Federated learning and blockchain are discussed as technical mechanisms that can contribute to data privacy and record integrity; neither constitutes regulatory compliance with frameworks such as HIPAA or the GDPR on its own. Persistent obstacles identified across the reviewed literature include dataset heterogeneity, limited external clinical validation, energy-constrained edge hardware, low clinician trust in opaque models, and fragmented interoperability standards. Conclusions: The evidence reviewed here is consistent with, but does not by itself establish, a layered, privacy-preserving, and explainable architecture that couples lightweight on-device inference with federated or blockchain-secured cloud learning as a design direction for future real-time healthcare monitoring systems. Future work should prioritise standardised benchmarking, prospective clinical validation, and regulatory-aligned data–governance frameworks.