Modelling ECG and PPG signals in hypertension and HFpEF: a database for in silico evaluation of health assessment algorithms

Lead Supervisor
Dr Jordi Alastruey
Senior Lecturer
Biomedical Engineering, School of Biomedical Engineering & Imaging Sciences
King’s College London

Prof Phil Chowienczyk, Prof Steven Niederer

Project Details

Approaches to identify individuals at risk of cardiovascular disease (CVD) could help reduce mortality and morbidity by prompting effective strategies that have been identified to reduce cardiovascular (CV) risk and improve patient outcomes [1]. The proliferation of smart wearables which measure the electrocardiogram (ECG) and photoplethysmogram (PPG; an optical measure of the arterial pulse wave) provide continuous signals that are strongly influenced by the heart and blood vessels. This gives the opportunity to monitor CV health in daily life and are emerging as potential tools for early detection of CVD, guiding of therapies, and measuring outcomes. In particular, the combination of ECG and PPG signals acquired by wearable devices may be used, in daily life, for (i) blood pressure monitoring (where the ECG is used to estimate the time of ventricular contraction and to measure pulse arrival time from the PPG) [2]; (ii) vascular age assessment which may have particular utility for CV-risk prediction [3]; (iii) detection of atrial fibrillation (where PPG-based devices could prompt ECG-based assessment when possible atrial fibrillation is detected) [4]; (iv) early detection of viral infections such as COVID-19 [5]; (v) improving mental stress assessment [6]; and (vi) accurate estimation of breathing rate [7]. If these techniques could be implemented in smart wearables, then they could facilitate early identification of at-risk individuals, in contrast to current practice where CVD screening mostly requires direct contact with patients. However, algorithms need developing and testing, and that requires large datasets to perform clinical trials.

Acquiring comprehensive datasets in humans for assessing the performance of these algorithms is usually a complex task: it can be difficult to measure reference variables precisely (e.g. cardiac output); it is challenging to study the influence of individual CV properties on the ECG and PPG in vivo since other properties may change concurrently; it can be complex to measure pulse waves at all the sites of interest (particularly central arteries); clinical trials are expensive and time-consuming; and in vivo measurements are subject to experimental error. While there are also disadvantages to in silico studies (e.g., reliance on modelling hypotheses), they can provide additional hemodynamic and mechanistic insights, which would be difficult to obtain in vivo, and can be used for preliminary design and assessment of signal analysis techniques across a wide range of CV conditions in a relatively quick and inexpensive manner. Furthermore, the results of in silico studies can be used to inform the design of in vivo studies, facilitating next-generation clinical trials, and to confirm the findings of in vivo studies [8,9].

The aim of this project is to (i) produce and test a dataset of in silico ECG and PPG signals representative of a real population of individuals aged 25 to 75 years old with and without hypertension and heart failure with preserved ejection fraction (HFpEF), and (ii) use it for pre-clinical testing of algorithms for early detection of hypertension and HFpEF. The following tasks will be carried out to fulfil this aim:

Task 1 – Cardiac-Vascular Coupling (18 months): Our existing computational models of cardiac electrophysiology [10] and vascular blood flow [11], which enable simulation of the ECG and PPG signals, respectively, will be coupled into a single biophysical model capable of simultaneously simulating the ECG and PPG signals during healthy aging and with hypertension and HFpEF. The new model will be tested by comparison with in vivo data.

Task 2 – Dataset of ECG and PPG signals for thousands of virtual subjects (6 months): A database of ECG and PPG signals representative of a sample from a healthy adult population will be created and verified through comparison with in vivo data. It will be produced building on our expertise for simulating datasets of pulse wave signals (including the PPG signal) for thousands of virtual subjects [11].

Task 3 – Diseased virtual subjects (6 months): Two pathologies will be modelled for each virtual subject of the dataset created in Task 2: HFpEF and hypertension. A qualitative verification, using clinical data, will be carried out to verify that the dataset faithfully reproduces the changes in ECG and PPG wave shape observed in real subjects for both pathologies.

Task 4 – CV determinants of the PPG (6 months): The determinants of the ECG have been widely investigated, but not those of the PPG. We have shown that individual features of the PPG can be influenced by multiple CV properties [11], which highlights a potential challenge to using the PPG pulse wave to monitor cardiovascular health: algorithms for estimating parameters from the PPG must be robust to simultaneous changes in other parameters. For instance, algorithms to estimate blood pressure from the PPG must be carefully designed to remain accurate in the presence of changes in arterial stiffness and peripheral compliance, as these properties influenced pulse wave shape in a similar manner to blood pressure. Our proposed dataset of in silico ECG and PPG signals will enable controlled experiments to provide valuable insight into the cardiac and vascular determinants of ECG and PPG wave shapes.

Task 5 – Algorithms for estimating physiological parameters (6 months): Algorithms will be developed for estimating CV parameters from the ECG and PPG signals and initially tested under a range of normal and pathological CV conditions, using the dataset of in silico pulse waves created in Tasks 3 and 4. For each virtual subject, the in silico dataset will provide precise reference cardiac and vascular parameters that have been shown to be key markers of cardiac and vascular function in HFpEF and hypertension (such as ejection fraction and aortic stiffness). Moreover, in silico signals will be free of measurement error, enabling investigation of the effect of measurement errors on algorithms’ estimates, by adding controlled noise to the simulated ECG and PPG signals. This will enable (i) identification of the algorithm/s that have the potential to perform sufficiently well for clinical use, including the measurement site/s where the PPG should be acquired (e.g. finger, wrist), and (ii) design of a controlled experiment at the clinical research facility (CRF) at Guy’s and St Thomas’ (~10 volunteers), for which ECG and PPG signals will be acquired simultaneously at baseline and with non-invasive interventions such as breathing maneuverers and arithmetic tests that have been proven to alter cardiac and vascular properties.

The potential role of ECG-PPG-based approaches is likely to be in the early detection of CV disease and ubiquitous measurement of risk factors, which could trigger further clinical assessment. The proposed databases and accompanying codes will be made freely available to support further research on (i) learning from big data for health knowledge and (ii) inform the next-generation clinical trials, to realise the potential of wearable ECG-PPG for cardiovascular monitoring and for aiding clinical decision making.


  1. J. Stewart, G. Manmathan, and P. Wilkinson, Primary prevention of cardiovascular disease: A review of contemporary guidance and literature. JRSM Cardiovascular Disease 6:1–9, 2017.
  2. K. Welykholowa et al., Multimodal photoplethysmography-based approaches for improved detection of hypertension. Journal of Clinical Medicine 9(4):1203, 2020.
  3. S. C. Millasseau et al., The vascular impact of aging and vasoactive drugs: Comparison of two digital volume pulse measurements. American Journal of Hypertension 16(6):467–72, 2003.
  4. M. Manninger et al., Role of wearable rhythm recordings in clinical decision making The wEHRAbles project. Clinical Cardiology 43(9):1032–39, 2020.
  5. J. M. Radin et al., Harnessing wearable device data to improve state-level real-time surveillance of influenza-like illness in the USA: a population-based study. The Lancet Digital Health 2(2):e85–e93, 2020.
  6. H. G. Kim et al., Stress and heart rate variability: A meta-analysis and review of the literature. Psychiatry Investigation 15(3):235–45, 2018.
  7. P. H. Charlton et al., Breathing rate estimation from the electrocardiogram and photoplethysmogram: a review. IEEE Reviews in Biomedical Engineering 11:2–20, 2018.
  8. Y. Li et al., Forward and backward pressure waveform morphology in hypertension. Hypertension 69:375–81, 2017.
  9. S. Vennin et al., Identifying hemodynamic determinants of pulse pressure: a combined numerical and physiological approach. Hypertension 70:1176–82, 2017.
  10. A. Neic et al., Efficient computation of electrograms and ECGs in human whole heart simulations using a reaction-eikonal model. J Comput Phys 346:191-211, 2017.
  11. P. H. Charlton et al., Modeling arterial pulse waves in healthy aging: a database for in silico evaluation of hemodynamics and pulse wave indexes. American Journal of Physiology-Heart and Circulatory Physiology, 317(5):H1062–85, 2019.


Existing datasets from the clinical research facility (CRF) at Guy’s and St Thomas’ which Professor Chowienczyk is currently directing. ECG and PPG signals are also available in external datasets such as UK Biobank and MIMIC.


Learning from big data for health knowledge; In silico clinical trials; ECG; Pulse wave; Haemodynamics; Electrophysiology; Signal processing