Pre-ejection period, the reason why the electrocardiogram Q-wave is an unreliable indicator of pulse wave initialization

pyPPG: a Python toolbox for comprehensive photoplethysmography signal analysis

Objective.Photoplethysmography is a non-invasive optical technique that measures changes in blood volume within tissues. It is commonly and being increasingly used for a variety of research and clinical applications to assess vascular dynamics and physiological parameters. Yet, contrary to heart rate variability measures, a field which has seen the development of stable standards and advanced toolboxes and software, no such standards and limited open tools exist for continuous photoplethysmogram (PPG) analysis. Consequently, the primary objective of this research was to identify, standardize, implement and validate key digital PPG biomarkers.Approach.This work describes the creation of a standard Python toolbox, denotedpyPPG, for long-term continuous PPG time-series analysis and demonstrates the detection and computation of a high number of fiducial points and digital biomarkers using a standard fingerbased transmission pulse oximeter.Main results.The improved PPG peak detector had an F1-score of 88.19% for the state-of-the-art benchmark when evaluated on 2054 adult polysomnography recordings totaling over 91 million reference beats. The algorithm outperformed the open-source original Matlab implementation by ∼5% when benchmarked on a subset of 100 randomly selected MESA recordings. More than 3000 fiducial points were manually annotated by two annotators in order to validate the fiducial points detector. The detector consistently demonstrated high performance, with a mean absolute error of less than 10 ms for all fiducial points.Significance.Based on these fiducial points,pyPPGengineered a set of 74 PPG biomarkers. Studying PPG time-series variability usingpyPPGcan enhance our understanding of the manifestations and etiology of diseases. This toolbox can also be used for biomarker engineering in training data-driven models.pyPPGis available onhttps://physiozoo.com/.

Optimizing the input feature sets and machine learning algorithms for reliable and accurate estimation of continuous, cuffless blood pressure

The advent of mobile devices, wearables and digital healthcare has unleashed a demand for accurate, reliable, and non-interventional ways to measure continuous blood pressure (BP). Many consumer products claim to measure BP with a cuffless device, but their lack of accuracy and reliability limit clinical adoption. Here, we demonstrate how multimodal feature datasets, comprising: (i) pulse arrival time (PAT); (ii) pulse wave morphology (PWM), and (iii) demographic data, can be combined with optimized Machine Learning (ML) algorithms to estimate Systolic BP (SBP), Diastolic BP (DBP) and Mean Arterial Pressure (MAP) within a 5 mmHg bias of the gold standard Intra-Arterial BP, well within the acceptable limits of the IEC/ANSI 80601-2-30 (2018) standard. Furthermore, DBP's calculated using 126 datasets collected from 31 hemodynamically compromised patients had a standard deviation within 8 mmHg, while SBP's and MAP's exceeded these limits. Using ANOVA and Levene's test for error means and standard deviations, we found significant differences in the various ML algorithms but found no significant differences amongst the multimodal feature datasets. Optimized ML algorithms and key multimodal features obtained from larger real-world data (RWD) sets could enable more reliable and accurate estimation of continuous BP in cuffless devices, accelerating wider clinical adoption.

New Hemodynamic Parameters in Peri-Operative and Critical Care-Challenges in Translation

Hemodynamic monitoring technologies are evolving continuously-a large number of bedside monitoring options are becoming available in the clinic. Methods such as echocardiography, electrical bioimpedance, and calibrated/uncalibrated analysis of pulse contours are becoming increasingly common. This is leading to a decline in the use of highly invasive monitoring and allowing for safer, more accurate, and continuous measurements. The new devices mainly aim to monitor the well-known hemodynamic variables (e.g., novel pulse contour, bioreactance methods are aimed at measuring widely-used variables such as blood pressure, cardiac output). Even though hemodynamic monitoring is now safer and more accurate, a number of issues remain due to the limited amount of information available for diagnosis and treatment. Extensive work is being carried out in order to allow for more hemodynamic parameters to be measured in the clinic. In this review, we identify and discuss the main sensing strategies aimed at obtaining a more complete picture of the hemodynamic status of a patient, namely: (i) measurement of the circulatory system response to a defined stimulus; (ii) measurement of the microcirculation; (iii) technologies for assessing dynamic vascular mechanisms; and (iv) machine learning methods. By analyzing these four main research strategies, we aim to convey the key aspects, challenges, and clinical value of measuring novel hemodynamic parameters in critical care.

Single measurement estimation of central blood pressure using an arterial transfer function

BACKGROUND

Central blood pressure (BP) better reflects the loading conditions on the major organs and is more closely correlated with future cardiovascular events. The increased invasiveness and risk of infection prevents the routine measurement of central BP. Arterial transfer functions can provide central BP estimates from clinically available peripheral measurements. However, current methods are either generalized, potentially lacking the ability to adapt to inter and intra subject variability, or individualized based on additional, clinically unavailable, pulse transit time measurements. This work proposes a novel, self-contained method for individualizing an arterial transfer function from a single peripheral pressure measurement, capable of accurately estimating central BP in a range of hemodynamic conditions.

METHODS

Pulse wave analysis of femoral BP waves was employed to formulate initial approximations of central BP and arterial inlet flow waveforms, to serve as objective functions for the identification of all model parameters. Root mean squared error (RMSE), and systolic and pulse pressure errors were assessed with respect to invasive aortic BP measurements in a seven (7) porcine endotoxin experiments. Systolic and pulse pressure errors were analysed using Bland-Altman analysis. Method accuracy is also compared with an idealized transfer function, derived using the measured aortic-femoral pulse transit time and minimizing the RMSE of model output pressure with respect to reference aortic pressure, a generalized transfer function model, and invasive femoral pressure measurements.

RESULTS

Mean bias and limits of agreement (95% CI) for the proposed method were 1.0(-4.6, 6.7)mmHg and -1.0(-6.6, 4.6)mmHg for systolic and pulse pressure, respectively, compared to 3.6(-0.9, 8.2)mmHg and 2.7(-1.8, 7.3)mmHg for the generalized transfer function model. Mean bias and limits of agreement for femoral pressure measurements were -6.4(-15.0, 2.3)mmHg and -9.4(-18.1, -0.8)mmHg, for systolic and pulse pressure, respectively. The pooled mean and standard deviation of the RMSE produced by the single measurement method, relative to reference aortic pressure, was 4.3(1.1)mmHg, consistent with estimates produced by the idealized transfer function, 3.9(1.2)mmHg, and improving of the generalized transfer function, 4.6(1.4)mmHg.

CONCLUSIONS

The proposed single measurement method provides accurate central BP estimates from routinely available peripheral pressure measurements, and nothing else. The method allows for the individualization of transfer functions on a per patient basis to better capture changes in patient condition during the progression of disease and subsequent treatment, at no additional clinical cost.

Assessing hemodynamics from the photoplethysmogram to gain insights into vascular age: a review from VascAgeNet

The photoplethysmogram (PPG) signal is widely measured by clinical and consumer devices, and it is emerging as a potential tool for assessing vascular age. The shape and timing of the PPG pulse wave are both influenced by normal vascular aging, changes in arterial stiffness and blood pressure, and atherosclerosis. This review summarizes research into assessing vascular age from the PPG. Three categories of approaches are described: 1) those which use a single PPG signal (based on pulse wave analysis), 2) those which use multiple PPG signals (such as pulse transit time measurement), and 3) those which use PPG and other signals (such as pulse arrival time measurement). Evidence is then presented on the performance, repeatability and reproducibility, and clinical utility of PPG-derived parameters of vascular age. Finally, the review outlines key directions for future research to realize the full potential of photoplethysmography for assessing vascular age.

Predicting fluid-response, the heart of hemodynamic management: A model-based solution

BACKGROUND

Intravenous fluid infusions are an important therapy for patients with circulatory shock. However, it is challenging to predict how patients' cardiac stroke volume (SV) will respond, and thus identify how much fluids should be delivered, if any. Model-predicted SV time-profiles of response to fluid infusions could potentially be used to guide fluid therapy.

METHOD

A clinically applicable model-based method predicts SV changes in response to fluid-infusions for a pig trial (N = 6). Validation/calibration SV, SVmea, is from an aortic flow probe. Model parameters are identified in 3 ways: fitting to SVmea from the entire infusion, SVflfit, from the first 200 ml, SVfl200, or from the first 100 ml, SVfl100. RMSE compares error of model-based SV time-profiles for each parameter identification method, and polar plot analysis assesses trending ability. Receiver-operating characteristic (ROC) analysis evaluates ability of model-predicted SVs, SVfl200 and SVfl100, to distinguish non-responsive and responsive infusions, using area-under the curve (AUC), and balanced accuracy as a measure of performance.

RESULTS

RMSE for SVflFit, SVfl200, and SVfl100 was 1.8, 3.2, and 6.5 ml, respectively, and polar plot angular limit of agreement from was 11.6, 28.0, and 68.8°, respectively. For predicting responsive and non-responsive interventions SVfl200, and SVfl100 had ROC AUC of 0.64 and 0.69, respectively, and balanced accuracy was 0.75 in both cases.

CONCLUSIONS

The model-predicted SV time-profiles matched measured SV trends well for SVflFit, SVfl200, but not SVfl100. Thus, the model can fit the observed SV dynamics, and can deliver good SV prediction given a sufficient parameter identification period. This trial is limited by small numbers and provides proof-of-method, with further experimental and clinical investigation needed. Potentially, this method could deliver model-predicted SV time-profiles to guide fluid therapy decisions, or as part of a closed-loop fluid control system.

Pulse arrival time as a surrogate of blood pressure

Various models have been proposed for the estimation of blood pressure (BP) from pulse transit time (PTT). PTT is defined as the time delay of the pressure wave, produced by left ventricular contraction, measured between a proximal and a distal site along the arterial tree. Most researchers, when they measure the time difference between the peak of the R-wave in the electrocardiogram signal (corresponding to left ventricular depolarisation) and a fiducial point in the photoplethysmogram waveform (as measured by a pulse oximeter attached to the fingertip), describe this erroneously as the PTT. In fact, this is the pulse arrival time (PAT), which includes not only PTT, but also the time delay between the electrical depolarisation of the heart's left ventricle and the opening of the aortic valve, known as pre-ejection period (PEP). PEP has been suggested to present a significant limitation to BP estimation using PAT. This work investigates the impact of PEP on PAT, leading to a discussion on the best models for BP estimation using PAT or PTT. We conducted a clinical study involving 30 healthy volunteers (53.3% female, 30.9 ± 9.35 years old, with a body mass index of 22.7 ± 3.2 kg/m[Formula: see text]). Each session lasted on average 27.9 ± 0.6 min and BP was varied by an infusion of phenylephrine (a medication that causes venous and arterial vasoconstriction). We introduced new processing steps for the analysis of PAT and PEP signals. Various population-based models (Poon, Gesche and Fung) and a posteriori models (inverse linear, inverse squared and logarithm) for estimation of BP from PTT or PAT were evaluated. Across the cohort, PEP was found to increase by 5.5 ms ± 4.5 ms from its baseline value. Variations in PTT were significantly larger in amplitude, - 16.8 ms ± 7.5 ms. We suggest, therefore, that for infusions of phenylephrine, the contribution of PEP on PAT can be neglected. All population-based models produced large BP estimation errors, suggesting that they are insufficient for modelling the complex pathways relating changes in PTT or PAT to changes in BP. Although PAT is inversely correlated with systolic blood pressure (SBP), the gradient of this relationship varies significantly from individual to individual, from - 2946 to - 470.64 mmHg/s in our dataset. For the a posteriori inverse squared model, the root mean squared errors (RMSE) for systolic and diastolic blood pressure (DBP) estimation from PAT were 5.49 mmHg and 3.82 mmHg, respectively. The RMSEs for SBP and DBP estimation by PTT were 4.51 mmHg and 3.53 mmHg, respectively. These models take into account individual calibration curves required for accurate blood pressure estimation. The best performing population-based model (Poon) reported error values around double that of the a posteriori inverse squared model, and so the use of population-based models is not justified.

Leveraging the potential of machine learning for assessing vascular ageing: state-of-the-art and future research

Vascular ageing biomarkers have been found to be predictive of cardiovascular risk independently of classical risk factors, yet are not widely used in clinical practice. In this review, we present two basic approaches for using machine learning (ML) to assess vascular age: parameter estimation and risk classification. We then summarize their role in developing new techniques to assess vascular ageing quickly and accurately. We discuss the methods used to validate ML-based markers, the evidence for their clinical utility, and key directions for future research. The review is complemented by case studies of the use of ML in vascular age assessment which can be replicated using freely available data and code.

Carotid-Femoral Pulse Transit Time Variability Predicted Mortality and Improved Risk Stratification in the Elderly

[Figure: see text].

Preload & Frank-Starling curves, from textbook to bedside: Clinically applicable non-additionally invasive model-based estimation in pigs

BACKGROUND

Determining physiological mechanisms leading to circulatory failure can be challenging, contributing to the difficulties in delivering effective hemodynamic management in critical care. Continuous, non-additionally invasive monitoring of preload changes, and assessment of contractility from Frank-Starling curves could potentially make it much easier to diagnose and manage circulatory failure.

METHOD

This study combines non-additionally invasive model-based methods to estimate left ventricle end-diastolic volume (LEDV) and stroke volume (SV) during hemodynamic interventions in a pig trial (N = 6). Agreement of model-based LEDV and measured admittance catheter LEDV is assessed. Model-based LEDV and SV are used to identify response to hemodynamic interventions and create Frank-Starling curves, from which Frank-Starling contractility (FSC) is identified as the gradient.

RESULTS

Model-based LEDV had good agreement with measured admittance catheter LEDV, with Bland-Altman median bias [limits of agreement (2.5th, 97.5th percentile)] of 2.2 ml [-13.8, 22.5]. Model LEDV and SV were used to identify non-responsive interventions with a good area under the receiver-operating characteristic (ROC) curve of 0.83. FSC was identified using model LEDV and SV with Bland-Altman median bias [limits of agreement (2.5th, 97.5th percentile)] of 0.07 [-0.68, 0.56], with FSC from admittance catheter LEDV and aortic flow probe SV used as a reference method.

CONCLUSIONS

This study provides proof-of-concept preload changes and Frank-Starling curves could be non-additionally invasively estimated for critically ill patients, which could potentially enable much clearer insight into cardiovascular function than is currently possible at the patient bedside.

Tube-load model: A clinically applicable pulse contour analysis method for estimation of cardiac stroke volume

BACKGROUND AND OBJECTIVES

Accurate, reproducible, and reliable real-time clinical measurement of stroke volume (SV) is challenging. To accurately estimate arterial mechanics and SV by pulse contour analysis, accounting for wave reflection, such as by a tube-load model, is potentially important. This study tests for the first time whether a dynamically identified tube-load model, given a single peripheral arterial input signal and pulse transit time (PTT), provides accurate SV estimates during hemodynamic instability.

METHODS

The model is tested for 5 pigs during hemodynamic interventions, using either an aortic flow probe or admittance catheter for a validation SV measure. Performance is assessed using Bland-Altman and polar plot analysis for a series of long-term state-change and short-term dynamic events.

RESULTS

The overall median bias and limits of agreement (2.5th, 97.5th percentile) from Bland-Altman analysis were -10% [-49, 36], and -1% [-28,20] for state-change and dynamic events, respectively. The angular limit of agreement (maximum of 2.5th, 97.5th percentile) from polar-plot analysis for state-change and dynamic interventions was 35.6^∘, and 35.2^∘, respectively.

CONCLUSION

SV estimation agreement and trending performance was reasonable given the severity of the interventions. This simple yet robust method has potential to track SV within acceptable limits during hemodynamic instability in critically ill patients, provided a sufficiently accurate PTT measure.

Incorporating pulse wave velocity into model-based pulse contour analysis method for estimation of cardiac stroke volume

BACKGROUND AND OBJECTIVES

Stroke volume (SV) and cardiac output (CO) are important metrics for hemodynamic management of critically ill patients. Clinically available devices to continuously monitor these metrics are invasive, and less invasive methods perform poorly during hemodynamic instability. Pulse wave velocity (PWV) could potentially improve estimation of SV and CO by providing information on changing vascular tone. This study investigates whether using PWV for parameter identification of a model-based pulse contour analysis method improves SV estimation accuracy.

METHODS

Three implementations of a 3-element windkessel pulse contour analysis model are compared: constant-Z, water hammer, and Bramwell-Hill methods. Each implementation identifies the characteristic impedance parameter (Z) differently. The first method identifies Z statically and does not use PWV, and the latter two methods use PWV to dynamically update Z. Accuracy of SV estimation is tested in an animal trial, where interventions induce severe hemodynamic changes in 5 pigs. Model-predicted SV is compared to SV measured using an aortic flow probe.

RESULTS

SV percentage error had median bias and [(IQR); (2.5th, 97.5th percentiles)] of -0.5% [(-6.1%, 4.7%); (-50.3%, +24.1%)] for the constant-Z method, 0.6% [(-4.9%, 6.2%); (-43.4%, +29.3%)] for the water hammer method, and 0.8% [(-6.5, 8.6); (-37.1%, +47.6%)] for the Bramwell-Hill method.

CONCLUSION

Incorporating PWV for dynamic Z parameter identification through either the Bramwell-Hill equation or the water hammer equation does not appreciably improve the 3-element windkessel pulse contour analysis model's prediction of SV during hemodynamic changes compared to the constant-Z method.

Real-Time Cuffless Continuous Blood Pressure Estimation Using Deep Learning Model

Blood pressure monitoring is one avenue to monitor people’s health conditions. Early detection of abnormal blood pressure can help patients to get early treatment and reduce mortality associated with cardiovascular diseases. Therefore, it is very valuable to have a mechanism to perform real-time monitoring for blood pressure changes in patients. In this paper, we propose deep learning regression models using an electrocardiogram (ECG) and photoplethysmogram (PPG) for the real-time estimation of systolic blood pressure (SBP) and diastolic blood pressure (DBP) values. We use a bidirectional layer of long short-term memory (LSTM) as the first layer and add a residual connection inside each of the following layers of the LSTMs. We also perform experiments to compare the performance between the traditional machine learning methods, another existing deep learning model, and the proposed deep learning models using the dataset of Physionet’s multiparameter intelligent monitoring in intensive care II (MIMIC II) as the source of ECG and PPG signals as well as the arterial blood pressure (ABP) signal. The results show that the proposed model outperforms the existing methods and is able to achieve accurate estimation which is promising in order to be applied in clinical practice effectively.

A Comparative Study of Brachial-Ankle Pulse Wave Velocity and Heart-Finger Pulse Wave Velocity in Korean Adults

Arterial stiffness is considered an index of vascular aging. The brachial–ankle pulse wave velocity (baPWV) method is widely used because of its proven effectiveness; and the pulse wave velocity measurement method using both electrocardiogram (ECG) and photoplethysmogram (PPG) is actively being studied due to the convenience of measurement and the possibility of miniaturization. The aim of this study was to evaluate and compare the effects of age and gender in Korean adults using both the baPWV method and the PWV method with ECG and finger PPG (heart–finger PWV). The measurements have been carried out for 185 healthy subjects of Korean adults, and the results showed that the baPWV was highly correlated with age in both genders (r = 0.94 for both males and females). However, the correlation values in heart–finger PWV measurement were significantly lower than those of baPWV (r = 0.37 for males and r = 0.71 for females). Although the heart–finger PWV method is suitable for mobile applications because it can be easily miniaturized while maintaining its signal quality, these results show that the heart–finger PWV method is not as effective as baPWV at evaluating the arterial stiffness.

Clinically applicable model-based method, for physiologically accurate flow waveform and stroke volume estimation

BACKGROUND AND OBJECTIVES

Cardiovascular dysfunction can be more effectively monitored and treated, with accurate, continuous, stroke volume (SV) and/or cardiac output (CO) measurements. Since direct measurements of SV/CO are highly invasive, clinical measures are often discrete, or if continuous, can require recalibration with a discrete SV measurement after hemodynamic instability. This study presents a clinically applicable, non-additionally invasive, physiological model-based, SV and CO measurement method, which does not require recalibration during or after hemodynamic instability.

METHODS AND RESULTS

The model's ability to predict flow profiles and SV is assessed in an animal trial, using endotoxin to induce sepsis in 5 pigs. Mean percentage error between beat-to-beat SV measured from an aortic flow probe and estimated by the model was -2%, while 90% of estimations fell within -24.2% and +27.9% error. Error between estimated and measured changes in mean SV following interventions was less than 30% for 4 out of the 5 pigs. Correlations between model estimated and probe measured flow, for each pig and hemodynamic interventions, was r² = 0.58 - 0.96, with 21 of the 25 pig intervention stages having r²  >  0.80.

CONCLUSION

The results demonstrate the model accurately estimates and tracks changes in flow profiles and resulting SV, without requiring model recalibration.

Deep learning for prediction of cardiac indices from photoplethysmographic waveform: A virtual database approach

Deep learning methods combined with large datasets have recently shown significant progress in solving several medical tasks. However, collecting and annotating large datasets can be a very cumbersome and expensive task. We tackle these problems with a virtual database approach where training data is generated using computer simulations of related phenomena. Specifically, we concentrate on the following problem: can cardiovascular indices such as aortic elasticity, diastolic and systolic blood pressures, and blood flow from heart be predicted continuously using wearable photoplethysmographic sensors? We simulate the blood flow using a haemodynamic model consisting of the entire human circulation. Repeated evaluation of the simulator allows us to create a database of "virtual subjects" with size that is only limited by available computational resources. Using this database, we train neural networks to predict the cardiac indices from photoplethysmographic signal waveform. We consider two approaches: neural networks based on predefined input features and deep convolutional neural networks taking waveform directly as the input. The performance of the methods is demonstrated using numerical examples, thus carrying out a preliminary assessment of the approaches. The results show improvements in accuracy compared with the previous methods. The improvements are especially significant with indices related to aortic elasticity and maximum blood flow. The proposed approach would provide new means to measure cardiovascular health continuously, for example, with a simple wrist device.

Accurate end systole detection in dicrotic notch-less arterial pressure waveforms

Identification of end systole is often necessary when studying events specific to systole or diastole, for example, models that estimate cardiac function and systolic time intervals like left ventricular ejection duration. In proximal arterial pressure waveforms, such as from the aorta, the dicrotic notch marks this transition from systole to diastole. However, distal arterial pressure measures are more common in a clinical setting, typically containing no dicrotic notch. This study defines a new end systole detection algorithm, for dicrotic notch-less arterial waveforms. The new algorithm utilises the beta distribution probability density function as a weighting function, which is adaptive based on previous heartbeats end systole locations. Its accuracy is compared with an existing end systole estimation method, on dicrotic notch-less distal pressure waveforms. Because there are no dicrotic notches defining end systole, validating which method performed better is more difficult. Thus, a validation method is developed using dicrotic notch locations from simultaneously measured aortic pressure, forward projected by pulse transit time (PTT) to the more distal pressure signal. Systolic durations, estimated by each of the end systole estimates, are then compared to the validation systolic duration provided by the PTT based end systole point. Data comes from ten pigs, across two protocols testing the algorithms under different hemodynamic states. The resulting mean difference ± limits of agreement between measured and estimated systolic duration, of [Formula: see text] versus [Formula: see text], for the new and existing algorithms respectively, indicate the new algorithms superiority.

Electrocardiogram-Based R Wave Pulse Wave Index for Assessment of Carotid Atherosclerosis

BACKGROUND Carotid atherosclerosis (CA) is a common disease in middle-aged and elderly people, which is closely related to cardiovascular and cerebrovascular disease. In this study, we investigated the benefits of the electrocardiogram (ECG)-based R wave pulse wave index (ERWVI) for the diagnosis of CA. MATERIAL AND METHODS According to CA examinations by color Doppler ultrasound, patients were assigned to positive and negative groups. The ECG R wave-Pulse wave transit time (ERWPTT) was obtained by synchronously collecting ECG signals (R wave in ECG) and the time variations in maximum finger pulse oxygen (DOP) on the ECG monitor. RESULTS ERPWI was positively correlated with sex, age, BMI, diastolic/systolic blood pressure, fasting blood glucose, uric acid, cholesterol and triglyceride levels, LDL-cholesterol, non-alcoholic fatty liver disease (NAFLD), creatinine, and homocysteine, and was negatively correlated with HDL-cholesterol (P<0.05). With the increase of ERPWI, the incidence of CA significantly increased to various degrees among the subgroups (P<0.05). The binary logistic regression model showed that ERPWI was an independent risk factor for atherosclerosis. The ROC curve showed that when ERPWI was above 0.505, the incidence of CA increased significantly. CONCLUSIONS There is a close relationship between ERPWI and CA. ERPWI is an independent risk factor for CA. ERPWI ≥0.505 can be used as a diagnostic threshold for CA and a reference index for the diagnosis of CA.

Pulse transit time estimation of aortic pulse wave velocity and blood pressure using machine learning and simulated training data

Recent developments in cardiovascular modelling allow us to simulate blood flow in an entire human body. Such model can also be used to create databases of virtual subjects, with sizes limited only by computational resources. In this work, we study if it is possible to estimate cardiovascular health indices using machine learning approaches. In particular, we carry out theoretical assessment of estimating aortic pulse wave velocity, diastolic and systolic blood pressure and stroke volume using pulse transit/arrival timings derived from photopletyshmography signals. For predictions, we train Gaussian process regression using a database of virtual subjects generated with a cardiovascular simulator. Simulated results provides theoretical assessment of accuracy for predictions of the health indices. For instance, aortic pulse wave velocity can be estimated with a high accuracy (r > 0.9) when photopletyshmography is measured from left carotid artery using a combination of foot-to-foot pulse transmit time and peak location derived for the predictions. Similar accuracy can be reached for diastolic blood pressure, but predictions of systolic blood pressure are less accurate (r > 0.75) and the stroke volume predictions are mostly contributed by heart rate.

Recently there has been a strong trend for self-monitoring of your cardiovascular health and new wearable sport trackers and mobile applications are coming to the market everyday. However, such solutions are mostly taking advantage of heart rate measurement. Other health indices such as blood pressure and pulse wave velocity reflecting to the condition of cardiovascular system would also be of great interest, but such solutions for continuous monitoring are barely existing or are at least unreliable. In this paper, we use computational modelling to assess theoretical capabilities of such measurements. We concentrate on predicting health indices using on pulse transmit time type of measurements. Such measurements could be carried out, for example, with photopletyshmography sensor or an optical sensor already found from several wearable sport trackers. We use cardiovascular modelling to create a database of “virtual subjects”, which is applied with machine learning to construct predictors for health indices. Our findings suggest that aortic pulse wave velocity and diastolic blood pressured could be predicted with a high accuracy, but predictions of systolic blood pressure are less accurate.