An intelligent patient bed architecture based on fuzzy cognitive maps, multimodal Internet of Things sensing, and machine learning

Le résumé fourni par la source

Ageing is a prime contributor to the rising demand for safe, uninterrupted care at the bedside. Many existing smart-bed systems are reactive, using individual thresholds instead of reasoning about the patient's state. Pressure injuries, falls, and poor sleep quality are issues for older adults, and mobility issues are also a concern. This study introduces and tests a proof-of-concept intelligent patient bed architecture based on fuzzy cognitive maps (FCMs), multi-modal IoT sensing, machine-learning prediction, and closed-loop actuation towards proactive elderly care. A design science research framework guided four linked phases: FCM construction through literature synthesis and expert elicitation (n = 12), multimodal sensing and machine-learning development, architecture-level evaluation in a digital twin, and a pilot usability study. A controlled laboratory dataset from 50 older participants comprising 210,000 pressure maps and 14,500 annotated movement events was used. A convolutional neural network (CNN)–long short-term memory (LSTM) model classified posture and transitions, a radial basis function–support vector machine (SVM) stratified fall risk, and an 11-node FCM translated predictive outputs into context-sensitive actions. Digital twin testing covered 100 virtual patient-days; usability was explored with 20 stakeholders. The FCM matched expert-defined responses in 87.3% of predefined scenarios. The CNN–LSTM achieved 92.4% test accuracy, and the SVM achieved 87.6% sensitivity for the high fall risk class. Mean sensor-to-action latency was 1.3 seconds, the false-alarm rate was 0.7 per virtual patient-day, and the mean System Usability Scale score was 76.8. With simulated sensor dropout at or below 25%, FCM agreement declined to 82.1%. The findings support the technical plausibility of combining causal FCM reasoning with multimodal sensing and predictive models in an intelligent patient bed. However, evidence remains preliminary due to the small sample size, the controlled laboratory setting in which the data were collected, and the simulation-based system-level testing. Physical prototyping, external model comparison, component ablation, and long-term clinical deployment are required before clinical effectiveness can be established.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.