Table 1 Digital twin-related applications in cardiovascular diseases.

Auricchio et al. 2013 Italy⁵¹

AAA repair technique in a poor candidate for open surgery.

Compared pre-operative patient-specific simulation of the implant of a custom-made endograft prediction with post-operative outcomes.

Numerical analyses; FEA

Proof of concept;

Precursor of a DT

Biancolini et al. 2020 Italy⁵²

High-fidelity surgical planning tool for thoracic aortic aneurysm repair to visualise, interactively and almost in real-time, the effect of various bulge shape parameters.

ROM framework to overcome the computing costs required in CFD techniques that are needed for blood flow prediction.

ROM, RBF, CFD

Proof of concept;

Precursor of a DT

Chakshu et al. 2020 UK¹⁹

Detection of AAA and severity classification using a virtual patient database.

Applies an inverse analysis system to blood flow prediction; and recurrent neural networks to classify AAA severity.

Deep learning and neural networks, waveform calculation/vessel dynamics, inverse analysis

Model validation;

Precursor concept for an active DT

Hemmler et al. 2019 Germany⁵³

A DT for pre-operative selection of stent-graft size and material to overcome late complications of infrarenal endovascular repair versus open-surgical AAA repair.

Use of patient-specific pre-operative data and a morphing algorithm to predict post-operative graft configuration and wall stress; mechanical modelling of the graft and the geometry of aneurysms.

CFD, FEA

Model validation

Larrabide et al. 2012 Spain⁵⁴

To improve selection, safety, and accuracy of intracranial stent implantation for intracranial aneurysm using a novel virtual stent deployment.

Use of a ‘phantom’ and a digital replica to compare in vitro experiments with computational analysis of stent configurations within patient-specific anatomy and aneurysm geometry.

CFD, deformation models

Computational model

Martinez-Velazquez et al. 2019 Canada²⁶

Use of ‘edge computing’ means, e.g., body sensors, Bluetooth, and 5G networks, to detect and aggregate bio-signals into a DT interface for detecting dysrhythmias caused by a myocardial infarction.

Multilayer platform proposed in which a pipeline of AI-based analyses of ECG and biodata from the real twin (the IHD patient) in real-time builds a DT rendering of the heart. PTB Diagnostic ECG Database⁵⁵ was used to train and test the CNN model.

Edge computing, AI, neural networks

Proof of concept;

Precursor concept for an active DT

Mazumder et al. 2019 India²⁷

Training machine-learning algorithms with conventional mathematically-derived synthetic bulk data requires an alternative approach to improve the accuracy of simulated ‘what if’ scenarios for CAD with better pathophysiological interpretability.

The DT is modelled with a two-chambered heart and baroreflex-based blood pressure control to generate synthetic physiological data in healthy and atherosclerotic conditions. The MIMIC-II database⁵⁶ was used to develop the PPG signal algorithm.

ROM of haemodynamics/ flow resistance; synthetic PPG signal data generation for training machine-learning algorithms

Model validation

Naplekov et al. 2018 Russia²³

A DT of coronary vessels can give a visual representation of the wearing process and progression of heart disease but requires haemodynamic and shear stress modelling.

Numerical simulation of the mechanical characteristics of the coronary vessel system, such as laminar and turbulent blood flow, and the impact of thrombus-induced vortex flow on load, blood pressure, and valves.

CFD

Computational model

Semakova et al. 2018 Russia²⁵

Data-driven DT profiles of real hypertensive patients can be used to facilitate virtual clinical trials that predict blood pressure variability and the effect of treatment.

Modelling of the annual average blood pressure variability and treatment effectiveness of antihypertensive drugs, based on diverse variables obtained from actual EMR data (n = 4521).

Probabilistic modelling/stochastic methods

Clusters are precursors of a larger dynamic population model

Chakshu et al. 2019 UK⁵⁷

Detection of carotid stenosis severity from a video of a human face.

In vivo head vibrations are compared against virtual vibration data generated from a coupled computational blood flow and head vibration model.

Principal component analysis

Model validation;

Semi-active DT model

Jones et al. 2021 UK⁵⁸

Applies machine learning for the detection of stenoses and aneurysms, adopting algorithms that learn patterns and biomarkers from a labelled dataset.

Presents the ML methodology and metrics used for quantification of arterial disease classification accuracies using only pressure and flow-rate measurements at select locations in the arterial network. A freely available virtual patient database⁵⁹ was used to train the algorithms.

ML methods: Naive Bayes, logistic regression, support vector machine, multilayer perceptron, random forests, and gradient boosting

Proof of concept

Sharma et al. 2020 USA⁶⁰

DT benefits are discussed in a hierarchy of AI applications in diagnostic and prognostic imaging, e.g., apparent superior diagnostic accuracy of coronary stenosis by machine-learning-based CT-FFR over CTA alone.

n/a

n/a

Hirschvogel et al. 2019 Germany⁶¹

DT model to demonstrate a personalised model of the failing heart, vascular system, and BiVAD implant design.

Increasing ventricular augmentation is applied and the effect on patient-specific ventricular wall mechanics and geometry is modelled.

0-D and 3D geometry/echocardiography; deformation elastodynamics

Proof of concept in vivo porcine model (n = 11)

Pagani et al. 2021 Italy⁶²

Reviews issues with integrating imaging, rhythm, and other clinical data into numerical models for patient-specific prediction in cardiac EP.

n/a

n/a

Gillette et al. 2021 Austria⁶³

Generating high-fidelity cardiac digital twins comprises both anatomical (from tomographic data) and functional (inferred from ECG) twinning stages. This study addresses limitations for both stages that impede efficiency and accuracy for clinical utility.

Describes and demonstrates methodologies (parameter vector and fast-forward ECG model), to improve the value of a biophysically-detailed digital twin replicating ventricular EP.

Finite element analysis

Proof of concept

Camps et al. 2021 UK⁶⁴

Investigates new computational techniques for the efficient quantification of subject-specific ventricular activation properties using CMR-based modelling and simulation and non-invasive electrocardiographic data.

Describes a sequential Monte Carlo approximate Bayesian algorithm to conduct the simultaneous inference of endocardial and myocardial conduction speeds and the root nodes; quantified the accuracy of recovering these activation properties in a cohort of twenty virtual subjects.

Bayesian computation-based inference method

Statistical method;

Precursor concept for a DT

Gerach et al. 2021 Germany⁶⁵

Bidirectional coupling or strong coupling is required to simulate physiological behaviour of the heart including mechano-electric feedback; adaptation of this framework allows personalisation from ion channels to the organ level enabling digital twin modelling.

Provides parameterisations of a fully coupled multi-scale model of the human heart, including electrophysiology, mechanics, and a closed-loop model of circulation; demonstrates model validity using a simulation on personalised heart geometry created from MRI data of a healthy volunteer.

Mathematical framework for geometry and deformation

Model validation; Precursor concept for a DT

Bende et al. 2020 India⁵⁰

Machine-learning algorithms can be trained using data from implanted devices, e.g., pacemakers, to create an updateable virtual organ using simulation software.

Demonstrates the simulation method to create a DT of the heart and tests the accuracy of the decision tree obtained for classifying disease severity.

Finite element analysis; machine learning

Statistical method

Lamata P. 2018 UK⁴⁰

Challenges with the use of machine learning to reason from data within statistical models for CVD prediction.

n/a

n/a

Lamata P. 2020 UK²⁸

Risks and benefits for the cardiac DT of mechanistic and statistical models; strategies to improve how the latter use patterns in big data for CVD prediction.

n/a

n/a

Niederer et al. 2019 UK⁶⁶

Describes biophysical models in cardiology and prediction models for dysrhythmia and heart failure therapies; outlines translational barriers to personalisation and uptake into clinical decision-making.

n/a

n/a

Niederer et al. 2020 UK³⁷

Describes patient-specific cardiac models and how virtual patient cohort models are developed and validated, and how model uncertainty is quantified; also, potential and future applications of virtual cohorts.

n/a

n/a

Hose et al. 2019 UK⁴²

Processes for cardiovascular models for clinical decision support and uptake of DT-related disciplines and sciences, such as AI.

n/a

n/a

Corral-Acero et al. 2020 UK¹³

Discussion of DT concepts and applications in precision cardiovascular medicine.

n/a

n/a

Peirlinck et al. 2021 USA⁶⁷

Historical development of cardiac modelling; future roles; challenges for precision medicine.

n/a

n/a