Artificial Intelligence-Assisted Digital Twin Architecture for Personalized Healthcare Monitoring and Predictive Disease Management in Smart Hospitals

Abstract

Continuous integration of physiological, clinical and contextual data is required in the personalized healthcare monitoring in smart hospitals. The conventional monitoring systems have disadvantages: fixed thresholds, limited personalization and they may miss detection of deterioration. In this work, artificial intelligence-enabled digital twin architecture for personalized monitoring of adolescent health and predictive disease management is proposed. Five-minute physiological observations were used to simulate a synthetic cohort of 5,000 adolescents ages 10-19 years for 30 days. Each patient-specific twin was continuously synchronized in a two-layer gated recurrent unit with attention to estimate the risk of the metabolic, respiratory, cardiovascular and overall deterioration. The proposed model successfully simulated the accuracy of 94.3%, F1-score of 92.8%, and an area under the receiver operating characteristic curve of 0.967. Prediction latency of mean was 42 ms, and early-warning time was 7.4 h. The architecture illustrates the scalability of decision support; however, validation with clinical data from adolescents is required to implement the architecture.