Generative AI-Driven Human Digital Twin in IoT-Healthcare: A Comprehensive Survey

TL;DR

This paper surveys the integration of Generative AI with Human Digital Twin (HDT) in IoT-healthcare, defining HDT, contrasting it with conventional digital twins, and outlining a framework that combines five HDT components with GAI techniques. It details concrete implementations across data acquisition, communication, data management, digital modeling, and data analysis, providing examples such as SynSigGAN, scGPT, ScarGAN, SNF-CVAE, MADEGAN, and pose-guided diffusion. The survey then maps GAI-driven HDT to IoT-healthcare applications—personalized health monitoring and diagnosis, personalized prescription, and rehabilitation—while discussing ethical, privacy, and integration challenges. Finally, it highlights open issues and future directions, including energy efficiency, human-centric metrics, real-time performance, and seamless integration with HL7/FHIR-based healthcare IT ecosystems, to guide research and practice in this nascent field.

Abstract

The Internet of things (IoT) can significantly enhance the quality of human life, specifically in healthcare, attracting extensive attentions to IoT-healthcare services. Meanwhile, the human digital twin (HDT) is proposed as an innovative paradigm that can comprehensively characterize the replication of the individual human body in the digital world and reflect its physical status in real time. Naturally, HDT is envisioned to empower IoT-healthcare beyond the application of healthcare monitoring by acting as a versatile and vivid human digital testbed, simulating the outcomes and guiding the practical treatments. However, successfully establishing HDT requires high-fidelity virtual modeling and strong information interactions but possibly with scarce, biased and noisy data. Fortunately, a recent popular technology called generative artificial intelligence (GAI) may be a promising solution because it can leverage advanced AI algorithms to automatically create, manipulate, and modify valuable while diverse data. This survey particularly focuses on the implementation of GAI-driven HDT in IoT-healthcare. We start by introducing the background of IoT-healthcare and the potential of GAI-driven HDT. Then, we delve into the fundamental techniques and present the overall framework of GAI-driven HDT. After that, we explore the realization of GAI-driven HDT in detail, including GAI-enabled data acquisition, communication, data management, digital modeling, and data analysis. Besides, we discuss typical IoT-healthcare applications that can be revolutionized by GAI-driven HDT, namely personalized health monitoring and diagnosis, personalized prescription, and personalized rehabilitation. Finally, we conclude this survey by highlighting some future research directions.

Figures (13)

• Figure 1: The workflow of recent primary trends of GAI models, including generative adversarial network, variational autoencoder, transformer and diffusion model.
• Figure 2: The framework of GAI-driven HDT. It includes the GAI-enabled data acquisition, data management, data modeling and data analysis. With the implementation of them, the GAI-driven HDT can be applied in IoT-healthcare, including personalized health monitoring and diagnosis, personalized prescription, and personalized rehabilitation.
• Figure 3: Overview of the GAN-based biomedical signal synthesis, SynsigGAN, proposed in [71]. The collected signals proceed through the preprocessing stage, eliminating noise and refining the signals using discrete wavelet transform, thresholding, and inverse discrete wavelet transform. After preprocessing, the signals are forwarded to the segmentation stage that uses the Z-score to solve the amplitude scaling problem and eliminate offset. Next is the GAN, which takes in the segmented signals and generates synthetic biomedical signals using bidirectional grid long short-term memory for generator network and convolutional neural network for the discriminator. Finally, SynsigGAN outputs the synthesized biomedical signals.
• Figure 4: Overview of the diffusion model-based histopathology image synthesis approach proposed in [76]. The real histopathology images are extracted the genotype information firstly, then it is used as a conditional input to the diffusion probabilistic model, which generates synthetic histopathology images that are tailored to specific genotypes.
• Figure 5: Overview of AGRoL proposed in [81]. It takes the orientations of the head and hands from HMDs and hand controllers as the input. These input processed by AGRoL. The architecture of AGRoL is presented in the middle of this figure, where $t$ is the noising step. $x_{t}^{1:N}$ denotes the motion sequence of length $N$ at step $t$, which is pure Gaussian noises when $t=0$. $p^{1:N}$ denotes the sparse upper body signals of length $N$. $\hat{x}_{t}^{1:N}$ denotes the denoised motion sequence at step $t$. The output is the synthesized full body motion.