PaperlayerView exact revision on arXiv

Source-linked AI summary

Networking Architecture and Key Supporting Technologies for Human Digital Twin in Personalized Healthcare: A Comprehensive Survey

Jiayuan Chen, Changyan Yi, Samuel D. Okegbile, Jun Cai, Xuemin, Shen

arXiv:2301.03930v3cs.NI

TL;DR

HDT applies digital-twin concepts to human bodies for personalized healthcare, but its networking architecture and supporting technologies remain insufficiently synthesized. This survey organizes HDT requirements, architecture, and enabling technologies, covering five networking layers and associated healthcare applications. It identifies practical challenges including data quality, AI interpretability, semantic-communication metrics, latency, security, and sustainability.

  • Problem

    HDT for personalized healthcare faces substantial research challenges, while existing surveys do not comprehensively cover its networking architecture and supporting technologies.

  • Method

    The survey compares HDT with conventional digital twins, analyzes requirements and challenges, and reviews a five-layer networking architecture with supporting technologies.

  • Results

    The survey synthesizes technologies for data acquisition, communication, computation, data management, and data analysis and decision making in HDT for personalized healthcare.

  • Takeaways & Limitations

    HDT implementation requires sophisticated multisource, multimodal, real-time data, explainable AI, suitable semantic-communication metrics, low-latency security, and energy-aware computing.

Abstract

from arXiv · show

Digital twin (DT), refers to a promising technique to digitally and accurately represent actual physical entities. One typical advantage of DT is that it can be used to not only virtually replicate a system's detailed operations but also analyze the current condition, predict future behaviour, and refine the control optimization. Although DT has been widely implemented in various fields, such as smart manufacturing and transportation, its conventional paradigm is limited to embody non-living entities, e.g., robots and vehicles. When adopted in human-centric systems, a novel concept, called human digital twin (HDT) has thus been proposed. Particularly, HDT allows in silico representation of individual human body with the ability to dynamically reflect molecular status, physiological status, emotional and psychological status, as well as lifestyle evolutions. These prompt the expected application of HDT in personalized healthcare (PH), which can facilitate remote monitoring, diagnosis, prescription, surgery and rehabilitation. However, despite the large potential, HDT faces substantial research challenges in different aspects, and becomes an increasingly popular topic recently. In this survey, with a specific focus on the networking architecture and key technologies for HDT in PH applications, we first discuss the differences between HDT and conventional DTs, followed by the universal framework and essential functions of HDT. We then analyze its design requirements and challenges in PH applications. After that, we provide an overview of the networking architecture of HDT, including data acquisition layer, data communication layer, computation layer, data management layer and data analysis and decision making layer. Besides reviewing the key technologies for implementing such networking architecture in detail, we conclude this survey by presenting future research directions of HDT.

I. INTRODUCTION

Healthcare resource shortages and rising costs motivate personalized healthcare, while AI, DT, and HDT offer data-driven ways to improve individualized care. This survey addresses a gap in prior work by organizing HDT networking architecture and its supporting technologies for personalized healthcare.

  • Background and Motivation: Traditional healthcare faces insufficient resources and growing costs, highlighted by pandemic-driven demand exceeding medical-system capacity.The passage cites shortages of facilities and substantial COVID-19 impacts as examples.
  • Background and Motivation: Personalized healthcare uses individual medical records, genes, and values to provide targeted treatments rather than one-size-fits-all care.The passage also links this approach to reduced unnecessary side effects and costs.
  • Background and Motivation: AI-based personalized healthcare is constrained by the costly, error-prone, and time-consuming acquisition of individual training datasets.Existing regression- and classification-based approaches are also described as limited to specific diseases, diagnoses, or small populations.
  • Digital Twin Motivation: Digital twins can provide virtual testbeds that generate diverse individual synthetic datasets for more efficient AI training without harming the human body.DTs are presented as digital replicas combining technologies including big data, AI, communications, computation, visualization, and cybersecurity.
  • HDT Benefits: HDT is associated with continuous monitoring, early detection, future-health prediction, safer individualized therapies, remote access, and improved diagnostic precision.The listed benefits include monitoring viral infections, supporting clinical trials, recommending therapies, identifying genomic events, and reducing geographic barriers.
  • Related Work and Survey Scope: Prior surveys largely addressed conventional industrial DTs or high-level HDT technologies, leaving HDT networking architecture and its enabling technologies insufficiently covered.This survey responds by examining the end-to-end networking architecture and related technologies for HDT in personalized healthcare.

D. Contributions

The survey examines HDT in personalized healthcare through its networking architecture and supporting technologies. It distinguishes HDT from conventional DTs, analyzes networking-oriented requirements and challenges, surveys a five-layer architecture, and outlines future research directions.

  • The survey comprehensively investigates networking architecture and key technologies for HDT in personalized healthcare.
  • It surveys HDT’s differences from conventional DTs, universal framework, and essential functions.
  • A networking perspective is used to analyze HDT design requirements and challenges for ubiquitous, timely, secure, and accurate healthcare applications.
  • The survey presents a five-layer end-to-end architecture covering data acquisition, communication, computation, data management, and data analysis and decision making.
  • It concludes by outlining research directions intended to encourage future studies in HDT networking and supporting technologies.

B. Electronic Health Record

The survey is organized around its treatment of complex human systems, whose behavioral processes are more uncertain and difficult to abstract than those of machines.

  • Figure 1 presents the organization of the survey.
  • Humans are particularly complicated systems, with more uncertainty than machines because machine behavioral rules are similar and predetermined.
  • The abstract process of humans is significantly more difficult than that of machines.

I. Green HDT

The supplied passages describe HDT as a human-focused extension of conventional healthcare DTs, supported by heterogeneous data, virtual modeling, communications, computation, management, and analysis components. They also identify ethical concerns and the need for reliable data acquisition and high-fidelity virtual replicas.

  • HDT raises ethical questions about healthcare inequality and patient authority over access to actual health status.
  • HDT requires heterogeneous physiological, psychological, medical, environmental, and social data to construct high-fidelity virtual twins.
  • HDT represents human beings through a more holistic concept than conventional DTs, whose healthcare applications are evolving toward HDT.
  • A universal HDT framework comprises data acquisition, digital modelling and virtualization, communication, computation, data management, and data analysis and decision making.
  • A virtual twin is an in-silico replica of a physical twin that co-evolutes through reliable connections and uses real-world data to represent real-time human status.

3) Communication and Computation:

HDT requires coordinated communication, computation, data management, and analytics to synchronize physical and virtual twins and support personalized healthcare. These functions face stringent latency, data-quality, privacy, and reliability requirements.

  • Communication connects physical and virtual twins, while computation extracts, processes, securely transmits, and executes data through AI-driven techniques.
  • HDT manages heterogeneous, multisource, multiscale, noisy data to construct and evolve the virtual twin.
  • Data analytics extracts information and knowledge from massive received datasets to enhance personalized healthcare services.
  • Virtual twins support immersive surgical planning and virtual testing of potential prescriptions before physical-world interventions.
  • HDT feedback requires efficient network-resource optimization because multimodal interactions must achieve ultra-low round-trip times.
  • HDT applications impose stringent requirements for high-quality real-time data, synchronization, xURLLC, privacy, and secure communications.

5) Data Storage:

HDT needs substantial network-side storage and computation to handle continuous, high-volume data and time-sensitive analytics. Its networking architecture coordinates acquisition, communication, management, computation, and analysis across heterogeneous technologies.

  • Each HDT application can generate up to a few gigabytes of data daily, requiring mechanisms for storage, virtual-twin updates, and analytics.
  • Synchronization, model evolution, and analytics are computation-intensive, motivating network-side deployment and optimal resource scheduling.
  • The five-layer architecture collects data, communicates it, manages it, computes over it, and performs analysis and decision making before reversing feedback toward the physical space.
  • HDT uses heterogeneous acquisition nodes and communication paradigms to gather sophisticated data and meet diverse transmission requirements.
  • Key technologies include pervasive sensing, data cleaning, reduction and fusion, Bluetooth, ZigBee, molecular communication, tactile Internet, semantic communication, multi-access edge computing, and edge-cloud collaboration.

C. Lessons Learned

HDT data acquisition combines wearable, implantable, and software-based sensing to capture physiological and psychological information. These sources support personalized representations but differ in placement, signals, and sensing capabilities.

  • Personalized healthcare requires biomedical sensors to capture individual variability and help construct digital representations that prevent false positives.
  • Wearable devices are organized by head, torso, and limb placement and sample diverse physiological data for HDT construction.
  • Wearables support applications including ECG, glucose, glaucoma, temperature, heart-rate, respiratory, sleep, posture, activity, and gait monitoring.
  • Implantable nanobiosensors enable precision drug delivery, sensing, micro procedures, biomarker detection, and vital-signal monitoring inside the body.

3) Social Networks Sensing:

HDT integrates psychological, medical, and physiological information through social networks, EHRs, and body communications. These inputs support real-time human modeling but remain constrained by sensing, synchronization, bandwidth, storage, and security challenges.

  • Because humans have consciousness, emotions, and psychology, high-fidelity virtual twins must synchronize physiological and emotional states in real time.
  • Social-network content can provide real-time information about individuals’ thoughts, feelings, and psychological states.
  • EHRs contain imaging, histories, allergies, diagnoses, and treatments, supporting virtual-twin construction, diagnosis accuracy, and patient outcomes.
  • EHR-based virtual-twin models can accurately forecast disease progression and support tailored treatment and prevention personalization.
  • Current sensing and networking remain limited by fine-grained measurement, continuous collection, heterogeneous-data integration, storage, privacy, jamming, and in-body communication constraints.
  • HDT communications span on-body sensor-to-gateway links and beyond-body gateway-to-remote-server links.
  • Beyond-body HDT communication must convey massive multimodal and haptic information, motivating tactile and semantic communication approaches.
  • On-body sensors use WBANs with star or two-level topologies, while low-power devices commonly rely on BLE or ZigBee.

3) Molecular Communication:

The survey presents molecular communication as a bio-inspired option for transmitting information within the body, while semantic communication reduces HDT traffic by exchanging task-relevant meaning rather than complete source data.

  • Molecular Communication: Molecular communication transmits information using molecules or lipid vesicles released into aqueous media such as blood.It mimics communication mechanisms found in living cells.
  • Communication Requirements: HDT communication requires bidirectional, real-time PT-VT synchronization despite complex, massive, heterogeneous, multiscale, and noisy data.These requirements make communication delay-sensitive and data-driven.
  • Tactile Internet: Tactile Internet can transmit audio-visual-haptic feedback between real and virtual environments to support timely HDT interactions.The survey identifies tactile Internet as a promising response to the need for xURLLC services.
  • Semantic Communication: Semantic communication extracts task-relevant information before transmission, reducing bandwidth, computing latency, downlink pressure, and device energy use.It can also support efficient verification of PT-VT synchronization.
  • Semantic Communication: Semantic communication may improve security because communicating parties exchange semantic information rather than complete source data, while shared knowledge bases hinder interpretation by eavesdroppers.Its performance evaluation remains difficult because suitable metrics depend on message type.
  • Open Issues: Semantic communication systems face a critical lack of appropriate performance metrics, with proposed measures differing for sentence and image transmission.Examples include sentence similarity and peak signal-to-noise ratio.

C. Lessons Learned

The survey links HDT communication to on-body and beyond-body networks, then identifies interference, resource allocation, security, privacy, and computation as continuing system-level challenges.

  • Communication Scope: On-body communication uses BLE, ZigBee, and molecular communication, whereas beyond-body communication connects PTs and VTs through tactile and semantic communication.Wireless body area networks support on-body links, while beyond-body links carry massive multimodal data.
  • Open Communication Issues: Reliable HDT communication requires managing interference among simultaneous wireless technologies, optimizing resources, and protecting healthcare data during transmission.The survey mentions encryption, secure protocols, digital signatures, and secure timestamps.
  • Future Directions: Future communication research should target interference management, resource optimization, and security and privacy protection for reliable HDT deployment.These priorities are presented as necessary for successful implementation.
  • Computing Requirements: Centralized cloud computing may not satisfy HDT’s fast-responsive and computation-intensive requirements, motivating edge computing.HDT applications require substantial computing power for real-time rendering and rapid data analysis.
  • Mobile Edge Computing: Mobile edge computing brings computation near users and can address HDT mobility while reducing response time compared with cloud computing alone.The survey also describes a cardio twin implemented on a smartphone for responsive, mobile operation.

3) Achieving the Extremely High Quality-of-Service (QoS):

The survey describes MEC and edge-cloud collaboration as complementary approaches for meeting HDT’s stringent QoS and scalability requirements, while preserving privacy through distributed processing and federated learning.

  • Mobile Edge Computing: Offloading computation-intensive HDT applications such as augmented reality to edge nodes can reduce latency and mobile-device computation burden.The cited approach jointly considers privacy, mobility, motion-to-photon latency, energy computation, and edge-node load balancing.
  • Mobile Edge Computing: MEC faces practical challenges including high energy consumption, strained radio resources, heavy edge-server computation, and data privacy concerns.These challenges motivate mechanisms for joint offloading and transmission scheduling.
  • Edge-Cloud Collaboration: Edge-cloud collaboration combines edge and cloud resources by allowing edge servers to process tasks locally or cooperate with cloud servers.This addresses the limited resources of either edge or cloud computing alone.
  • Privacy-Preserving Computing: Distributing data across multiple edges can reduce privacy-leakage risk, while federated learning shares model updates instead of raw data.Local training can occur at the edge, with aggregation performed through the federated-learning server.
  • Open Issues: HDT computing still requires resource-allocation schemes for massive tasks and incentive mechanisms that attract participating servers while addressing privacy risks.These are identified as open issues for MEC-, edge-cloud-, and federated-learning-based HDT.

C. Data Security and Privacy

The survey frames HDT data management as a pipeline spanning preprocessing, storage, and security, with preprocessing addressing noisy, incomplete, heterogeneous, and redundant data from multiple sources.

  • Data Management Scope: HDT data management requires preprocessing, database storage, cybersecurity, privacy-preserving mechanisms, and distributed ledger technologies.The survey organizes data management into preprocessing, storage, and security and privacy perspectives.
  • Data Preprocessing: Data preprocessing handles heterogeneity, multiscale structure, high noise, missing data, redundancy, conflicts, and errors through cleaning, reduction, and fusion.These operations are intended to make physical data usable for subsequent HDT processing.
  • Data Cleaning: Data cleaning includes noise identification and removal, outlier detection and elimination, and data imputation.Reported approaches include denoising autoencoders, filtering, clustering, and missing-data replacement.
  • Data Cleaning: Data imputation replaces missing values using the underlying dataset structure, with KNN imputation and deep-learning imputation discussed for HDT data.Deep-learning imputation is described as suitable for discrete data because of its nonlinear approximation capability.
  • Data Reduction: Data reduction lowers dimensionality while preserving informative content through feature selection and feature extraction.Feature selection methods include filters, wrappers, and embedded methods; feature extraction creates new features from original ones.
  • Data Reduction: PCA and CNNs are reviewed as feature-extraction techniques for reducing dimensionality and identifying informative features in healthcare datasets.The survey cites applications involving HDT, ECG, medical, and tumor data.

3) Data Fusion:

HDT data fusion combines heterogeneous sources to support comprehensive analysis, while its storage and security infrastructure must handle massive, sensitive, and continuously exchanged health data. The survey reviews fusion levels, distributed storage, and cybersecurity technologies for these requirements.

  • Data Fusion: Data fusion is necessary in HDT because heterogeneous sources produce uneven data distributions and comprehensive analysis requires more than one data type.The survey identifies data fusion as a way to improve data collection and transmission from multisource HDT environments.
  • Data Fusion: Data-level fusion combines raw inputs from multiple sources and can be performed at edge devices after rearranging them into a unified matrix.It is described as the simplest fusion level and is widely adopted in HDT frameworks.
  • Data Fusion: Feature-level fusion extracts relevant features, removes non-informative data, and combines features for applications such as heart-disease prediction.One reviewed framework fused Framingham risk factors with IoMT data, selected features using information gain, and trained an ensemble deep-learning classifier.
  • Data Management: HDT requires distributed storage for demographics, clinical history, real-time health data, genetic information, diagnostic trajectories, analysis results, and mobility data.The survey discusses MySQL, HBase, NoSQL databases, HDFS, and OpenStack Swift as storage technologies for large-scale HDT data.
  • Cybersecurity: HDT security must protect confidentiality, access control, integrity, and authentication against threats including ransomware, eavesdropping, replay, denial-of-service, and battery-exhaustion attacks.The survey reviews intrusion detection and cyber-resilience technologies, including machine-learning and federated-learning-based systems.
  • Cybersecurity: Traditional intrusion-detection solutions can have high false-positive rates and require manual modification, limiting scalability in HDT.Federated-learning-based and mobile-agent-driven IDSs are reviewed as more flexible approaches for HDT healthcare networks.

2) Privacy-Preserving Mechanism:

HDT privacy-preserving mechanisms address threats from exposed communication channels, compromised storage, and honest-but-curious service providers. The survey covers cryptography, anonymization, differential privacy, and federated learning as complementary safeguards.

  • Privacy Threats: HDT privacy is threatened by eavesdropping, repeated-key attacks, compromised storage servers, and honest-but-curious cloud or computing providers.These threats can expose plaintext transmissions, stored health information, or query requests.
  • Mechanisms: The survey groups HDT privacy-preserving mechanisms into cryptographic techniques, anonymization, differential privacy, and federated learning.These mechanisms are reviewed as approaches for protecting privacy-sensitive health data across HDT functions.
  • Cryptography: Cryptography protects personal health data across HDT segments through symmetric-key, asymmetric-key, homomorphic, and quantum cryptography.Symmetric cryptography uses one key for encryption and decryption, whereas asymmetric cryptography uses public and private keys.
  • Cryptography: Homomorphic encryption enables computation on encrypted data without first decrypting it, supporting privacy protection in communication, computation, and data management layers.The survey also discusses post-quantum cryptography for protecting exchanged healthcare data against future quantum threats.
  • Anonymization: Anonymization removes identity information from health data to reduce the risk of identifying the source or owner of health records.The reviewed scheme targets identity disclosure even when background knowledge about quasi-identifiers is available.
  • Federated Learning: Federated learning preserves privacy by keeping participants’ raw data local and transmitting gradients to a central global model instead.This avoids transmitting sensitive health data through public networks to a potentially honest-but-curious cloud server.

3) Distributed Ledger Technology:

Distributed ledger technology, especially blockchain, provides decentralized and tamper-resistant infrastructure for HDT health-data storage and sharing. The survey connects blockchain with cryptographic identity, consensus, access control, and broader AI-enabled HDT functions.

  • Distributed Ledger Technology: Distributed ledger technology stores data across multiple network participants rather than under a single centralized controller.Blockchain is presented as a prominent DLT that supports secure and transparent recording across nodes.
  • Blockchain: Blockchain validates transactions through consensus among validators and makes stored information immutable after it is appended to the chain.These properties make blockchain suitable for enhancing security in HDT-enabled personalized healthcare.
  • Blockchain: Blockchain-enabled HDT data sharing uses private keys to sign data and public keys to provide unique identities and support encryption.End users such as smartphones or medical institutions sign collected data before sharing it through the blockchain system.
  • Health-Data Storage: Blockchain can provide immutable, secure, and reliable health-data storage through consensus, digital signatures, and hash-chain techniques.The survey describes a blockchain health application for physiological signals collected through body sensor networks.
  • Data Access Control: Blockchain can enforce differentiated access to encrypted electronic health records by storing indexes for complete and shared data on separate blockchains.Users request different data portions from different blockchains according to their attributes and permissions.
  • AI-Enabled HDT: HDT must process and exchange a sheer volume of data while applying AI to support analysis and decisions across monitoring, diagnosis, prescription, surgery, and rehabilitation.The survey identifies supervised, unsupervised, and reinforcement learning as major AI categories for HDT data analysis.
  • Diagnosis: In AAA diagnosis, a customized CNN achieved 99.91% accuracy for detection and 97.79% for severity classification.The reported results were higher than those of traditional clinics.
  • Prescription: For treatment decisions in head and neck cancer, deep Q-learning achieved mean and median accuracies of 87.09% and 90.85%, while survival increased by 3.73%.The reviewed system used survival and toxicity metrics to predict treatment outcomes.

C. Surgery

AI-enabled HDT supports personalized surgical planning, robotic assistance, and rehabilitation while addressing broader implementation challenges. The survey emphasizes improved precision and quantitative decision support but notes explainability, computational, data, and mobility constraints.

  • Surgery: AI-enabled HDT can support preoperative, intraoperative, and postoperative management to improve surgical accuracy and precision.The motivation is linked to surgical mortality, adverse reactions, limited technology precision, and insufficient surgeon experience.
  • Preoperative Planning: AI-assisted preoperative planning uses medical records and imaging for anatomical classification, detection, segmentation, and registration.The survey also reviews an HDT remote surgical rehearsal platform combining an rAC-GAN prediction model with mixed-reality navigation images.
  • Robotic Assistance: Learning from demonstration allows surgical robots to perform learned tasks or motions without requiring tedious manual programming.The survey presents this paradigm as useful for surgical procedures.
  • Robotic Assistance: AI-supported human-robot interaction enables surgeons to control surgical robots through touchless manipulation, including eye-gaze commands.A dense CNN-based model was used for 9-direction and 36-direction gaze estimation, while another system used an HMM for real-time gaze-gesture recognition.
  • Rehabilitation: AI can predict patient movements for limb rehabilitation and use predicted joint angles to control an exoskeleton trajectory synchronously.The reviewed upper-limb approach used a multi-stream LSTM duelling-based motion-prediction model.
  • Rehabilitation: AI provides therapists with quantitative rehabilitation analysis and can significantly increase decision-making accuracy through collaboration with domain experts.One system used reinforcement learning to assess motion quality and generate patient-specific explanatory analyses.
  • Open Issues: Full-fledged healthcare explainability remains difficult, while HDT AI training can incur high computing costs that motivate low-complexity models without accuracy loss.These are identified as open issues for AI-enabled HDT implementation.
  • Mobile HDT: Patient mobility can interrupt PT-VT connectivity and services, requiring VT migration while leaving handover, energy, security, privacy, and latency challenges unresolved.The survey identifies migration frameworks and optimization parameters as open problems for reliable HDT implementation.

D. Interoperable Management of Subsystems in HDT

Interoperable HDT management must mirror the human body’s coupled subsystems while enabling their digital twins to share information across semantic and data levels. The networking architecture also requires secure, explainable, scalable, and sustainable technologies for reliable personalized healthcare.

  • Subsystem interoperability: Each major human subsystem can have a corresponding digital twin whose data management supports mutual information sharing and interoperability.The proposed approach includes ontology-based semantic interoperability and model-transformation methods.
  • Subsystem interoperability: Respiratory and circulatory digital twins must interoperate to represent exercise-related feedback between oxygen delivery and blood circulation.Without data-level interoperability, the HDT cannot accurately support the coupled behavior described in the use case.
  • Interface design: HDT interfaces should be human-centered, multimodal, secure, efficient, scalable, interactive, and standardized across acquisition through decision-making layers.Design considerations include patient comfort, multimodal inputs and outputs, and handling large, complex healthcare workloads.
  • Security and privacy: Blockchain can protect HDT security and privacy, but complicated consensus mechanisms create latency that conflicts with ultra-low-latency healthcare interactions.The survey calls for intelligent blockchain implemented across physical and virtual environments, including interactions among virtual twins.
  • Explainable analysis: Explainable AI is required because black-box decisions can undermine trust and confidence in HDT recommendations for healthcare professionals.The paper links full-fledged explanations with safer and more reliable personalized healthcare decisions.
  • AI learning: HDT AI must learn effectively from largely unlabelled health data while supporting real-time, reliable decisions without developing a separate model for every task.The survey identifies unlabeled-data training and general artificial intelligence as directions for reducing task-specific development effort.
  • Sustainability: Green HDT is needed because cloud-edge communication and computation increase energy consumption and environmental concerns as HDT scales.Proposed directions include sustainable architectures, energy-efficient resource allocation, and integration with other green technologies.
  • Networking architecture: The survey organizes HDT networking around data acquisition, communication, computing, management, analysis, and decision making, and reviews technologies across the end-to-end data stream.Its coverage includes sensing, communications, edge computing, data management, artificial intelligence, and future research directions.
Loading 2301.03930v3…