Generative Digital Twins for Continuous and Reliable IoMT Monitoring in Medical Metaverse Environments

Abstract

The rapid expansion of the Internet of Medical Things (IoMT) has enabled continuous multi-sensor patient monitoring. However, the limited computational capabilities of IoMT devices make them vulnerable to data poisoning, denial-of-service attacks, and other disruptions that lead to missing or unreliable sensor readings. Such irregularities undermine clinical decision-making and significantly reduce the effectiveness of Digital Twin (DT)–enabled monitoring frameworks envisioned for the emerging medical Industrial Metaverse (IMV). To address these challenges, we propose a smart bidirectional DT architecture capable of ensuring real-time data continuity by generating realistic physiological signals during periods of data loss or corruption. The DT leverages a hybrid LSTM-VAE generative model trained on normal physiological patterns to synthesize high-fidelity temporal sequences while providing confidence and uncertainty estimates essential for transparent and medically accountable operation. Experimental evaluations on DHT11, EMG, and GSR sensor datasets show that the model effectively reconstructs physiological signals with low mean absolute error 0.0928, 0.6992, and 0.2719, respectively and produces stable confidence and uncertainty estimates. By ensuring uninterrupted and trustworthy data streams, the proposed framework enhances the resilience, reliability, and clinical utility of IoMT systems and strengthens the foundation for robust DT-driven healthcare monitoring in IMV environments.