Preprint

Novel Blood Pressure Waveform Reconstruction from Photoplethysmography using Cycle Generative Adversarial Networks

Authors:
Preprints and early-stage research may not have been peer reviewed yet.
To read the file of this research, you can request a copy directly from the authors.

Abstract

Continuous monitoring of blood pressure (BP)can help individuals manage their chronic diseases such as hypertension, requiring non-invasive measurement methods in free-living conditions. Recent approaches fuse Photoplethysmograph (PPG) and electrocardiographic (ECG) signals using different machine and deep learning approaches to non-invasively estimate BP; however, they fail to reconstruct the complete signal, leading to less accurate models. In this paper, we propose a cycle generative adversarial network (CycleGAN) based approach to extract a BP signal known as ambulatory blood pressure (ABP) from a clean PPG signal. Our approach uses a cycle generative adversarial network that extends theGAN architecture for domain translation, and outperforms state-of-the-art approaches by up to 2x in BP estimation.

No file available

Request Full-text Paper PDF

To read the file of this research,
you can request a copy directly from the authors.

ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
The objective is to develop a cuffless modelling approach to accurately estimate the blood pressure (BP) waveform and extract important BP features, such as the systolic BP (SBP), diastolic BP (DBP), and mean BP (MAP). Access to the full waveform has significant advantages over previous cuffless BP estimation tools in terms of accuracy and access to additional cardiovascular health markers (e.g., cardiac output), as well as potentially providing arterial stiffness and identifying different cardiovascular diseases. Nonlinear autoregressive models with exogenous input (NARX) are implemented using an artificial neural network to predict the BP waveforms using electrocardiography (ECG), and/or photoplethysmography (PPG) signals as inputs. The efficacy of the model is compared with a pulse arrival time (PAT) model using 15 subjects from the MIMIC II database. Two training modes are considered: training on the first eight minutes of data for each subject (Predictive training) and testing on the rest (up to 5.2 hours); and training on the first and the last eight minutes (Interval training) and testing the model in between. Predictive training and Interval training exhibited similar results initially, while Interval training resulted in higher accuracy over longer periods. The proposed method models the BP as a dynamical system leading to better accuracy in the estimation of SBP, DBP and MAP when compared to the PAT model. Moreover, the NARX model, with its ability to provide the BP waveform, yields more insight into patient health.
Article
Full-text available
Background: The benefits of blood pressure lowering treatment for prevention of cardiovascular disease are well established. However, the extent to which these effects differ by baseline blood pressure, presence of comorbidities, or drug class is less clear. We therefore performed a systematic review and meta-analysis to clarify these differences. Method: For this systematic review and meta-analysis, we searched MEDLINE for large-scale blood pressure lowering trials, published between Jan 1, 1966, and July 7, 2015, and we searched the medical literature to identify trials up to Nov 9, 2015. All randomised controlled trials of blood pressure lowering treatment were eligible for inclusion if they included a minimum of 1000 patient-years of follow-up in each study arm. No trials were excluded because of presence of baseline comorbidities, and trials of antihypertensive drugs for indications other than hypertension were eligible. We extracted summary-level data about study characteristics and the outcomes of major cardiovascular disease events, coronary heart disease, stroke, heart failure, renal failure, and all-cause mortality. We used inverse variance weighted fixed-effects meta-analyses to pool the estimates. Results: We identified 123 studies with 613 815 participants for the tabular meta-analysis. Meta-regression analyses showed relative risk reductions proportional to the magnitude of the blood pressure reductions achieved. Every 10 mm Hg reduction in systolic blood pressure significantly reduced the risk of major cardiovascular disease events (relative risk [RR] 0·80, 95% CI 0·77-0·83), coronary heart disease (0·83, 0·78-0·88), stroke (0·73, 0·68-0·77), and heart failure (0·72, 0·67-0·78), which, in the populations studied, led to a significant 13% reduction in all-cause mortality (0·87, 0·84-0·91). However, the effect on renal failure was not significant (0·95, 0·84-1·07). Similar proportional risk reductions (per 10 mm Hg lower systolic blood pressure) were noted in trials with higher mean baseline systolic blood pressure and trials with lower mean baseline systolic blood pressure (all ptrend>0·05). There was no clear evidence that proportional risk reductions in major cardiovascular disease differed by baseline disease history, except for diabetes and chronic kidney disease, for which smaller, but significant, risk reductions were detected. β blockers were inferior to other drugs for the prevention of major cardiovascular disease events, stroke, and renal failure. Calcium channel blockers were superior to other drugs for the prevention of stroke. For the prevention of heart failure, calcium channel blockers were inferior and diuretics were superior to other drug classes. Risk of bias was judged to be low for 113 trials and unclear for 10 trials. Heterogeneity for outcomes was low to moderate; the I(2) statistic for heterogeneity for major cardiovascular disease events was 41%, for coronary heart disease 25%, for stroke 26%, for heart failure 37%, for renal failure 28%, and for all-cause mortality 35%. Interpretation: Blood pressure lowering significantly reduces vascular risk across various baseline blood pressure levels and comorbidities. Our results provide strong support for lowering blood pressure to systolic blood pressures less than 130 mm Hg and providing blood pressure lowering treatment to individuals with a history of cardiovascular disease, coronary heart disease, stroke, diabetes, heart failure, and chronic kidney disease. Funding: National Institute for Health Research and Oxford Martin School.
Article
Full-text available
A clinical method for monitoring cardiac output (CO) should be continuous, minimally invasive, and accurate. However, none of the conventional CO measurement methods possess all of these characteristics. On the other hand, peripheral arterial blood pressure (ABP) may be measured reliably and continuously with little or no invasiveness. We have developed a novel technique for continuously monitoring changes in CO by mathematical analysis of a peripheral ABP waveform. In contrast to the previous techniques, our technique analyzes the ABP waveform over time scales greater than a cardiac cycle in which the confounding effects of complex wave reflections are attenuated. The technique specifically analyzes 6-min intervals of ABP to estimate the pure exponential pressure decay that would eventually result if pulsatile activity abruptly ceased (i.e., after the high frequency wave reflections vanish). The technique then determines the time constant of this exponential decay, which equals the product of the total peripheral resistance and the nearly constant arterial compliance, and computes proportional CO via Ohm's law. To validate the technique, we performed six acute swine experiments in which peripheral ABP waveforms and aortic flow probe CO were simultaneously measured over a wide physiologic range. We report an overall CO error of 14.6%.
Article
A photoplethysmography (PPG) is an uncomplicated and inexpensive optical technique widely used in the healthcare domain to extract valuable health-related information, e.g., heart rate variability, blood pressure, and respiration rate. PPG signals can easily be collected continuously and remotely using portable wearable devices. However, these measuring devices are vulnerable to motion artifacts caused by daily life activities. The most common ways to eliminate motion artifacts use extra accelerometer sensors, which suffer from two limitations: i) high power consumption and ii) the need to integrate an accelerometer sensor in a wearable device (which is not required in certain wearables). This paper proposes a low-power non-accelerometer-based PPG motion artifacts removal method outperforming the accuracy of the existing methods. We use Cycle Generative Adversarial Network to reconstruct clean PPG signals from noisy PPG signals. Our novel machine-learning-based technique achieves 9.5 times improvement in motion artifact removal compared to the state-of-the-art without using extra sensors such as an accelerometer, which leads to 45% improvement in energy efficiency.
Article
High blood pressure (BP) is a major source of death worldwide as it slows down the flow of blood and oxygen, and leads to various chronic diseases such as chest pain (angina), heart disease and heart failure. Therefore, regular measurement and monitoring of BP is an essential element of the home healthy settings, especially for the elderly. In this study, a novel deep neural network (called MLP-BP, which including gMLP-BP and MLPlstm-BP) adapted from MLP-Mixer is proposed to estimate BP from plethysmography (PPG) and electrocardiograph (ECG) signal. More precisely, a novel multi-filter to multi-channel (MFMC) is presented for preprocessing ECG and PPG signals, namely using various filters and filtering parameters to handle with ECG and PPG, and integrating the filtered bio-signals into multi-channel data. Then, the multi-channel data is fed into the proposed methods to estimate blood pressure directly. The up-to-date concept MLP-Mixer is employed in the proposed frameworks, which are both end-to-end pipelines without any handcrafted feature extraction operation. MLPlstm-BP (gMLP-BP) achieves the predicted diastolic BP with a mean absolute error (MAE) of 2.13 (2.47) mmHg, and with a standard deviation (SD) of 3.07 (3.52) mmHg; those of systolic BP is an MAE of 3.52 (4.18) mmHg, and a SD of 5.10 (5.87) mmHg on the MIMIC II dataset. In addition, the testing results are all meet the highest level of the Association for the Advancement of Medical Instrumentation (AAMI) and British Hypertension Society (BHS). Extensive experiments demonstrate that the proposed methods attain competitive results over state-of-the-art (SOTA) learning-based pipelines.
Conference Paper
Respiratory rate (RR) is a clinical sign representing ventilation. An abnormal change in RR is often the first sign of health deterioration as the body attempts to maintain oxygen delivery to its tissues. There has been a growing interest in remotely monitoring of RR in everyday settings which has made photoplethysmography (PPG) monitoring wearable devices an attractive choice. PPG signals are useful sources for RR extraction due to the presence of respiration-induced modulations in them. The existing PPG-based RR estimation methods mainly rely on hand-crafted rules and manual parameters tuning. An end-to-end deep learning approach was recently proposed, however, despite its automatic nature, the performance of this method is not ideal using the real world data. In this paper, we present an end-to-end and accurate pipeline for RR estimation using Cycle Generative Adversarial Networks (CycleGAN) to reconstruct respiratory signals from raw PPG signals. Our results demonstrate a higher RR estimation accuracy of up to 2× (mean absolute error of 1.9±0.3 using five fold cross validation) compared to the state-of-th-art using a identical publicly available dataset. Our results suggest that CycleGAN can be a valuable method for RR estimation from raw PPG signals.
Conference Paper
The paper proposes accurate Blood Pressure Monitoring (BPM) based on a single-site Photoplethysmographic (PPG) sensor and provides an energy-efficient solution on edge cuffless wearable devices. Continuous PPG signal preprocessed and used as input of the Artificial Neural Network (ANN), and outputs systolic BP (SBP), diastolic BP (DBP), and mean arterial BP (MAP) values for each heartbeat. The improvement of the BPM accuracy is obtained by removing outliers in the preprocessing step and the whole-based inputs compared to parameter-based inputs extracted from the PPG signal. Performance obtained is 3.42 ± 5.42 mmHg (MAE ± RMSD) for SBP, 1.92 ± 3.29 mmHg for DBP, and 2.21 ± 3.50 mmHg for MAP which is competitive compared to other studies. This is the first BPM solution with edge computing artificial intelligence as we have learned so far. Evaluation experiments on real hardware show that the solution takes 42.2 ms, 18.2 KB RAM, and 2.1 mJ average energy per reading.
Conference Paper
Blood Pressure (BP) is one of the four primary vital signs indicating the status of the body's vital (life-sustaining) functions. BP is difficult to continuously monitor using a sphygmomanometer (i.e. a blood pressure cuff), especially in everyday-setting. However, other health signals which can be easily and continuously acquired, such as photoplethysmography (PPG), show some similarities with the Aortic Pressure waveform. Based on these similarities, in recent years several methods were proposed to predict BP from the PPG signal. Building on these results, we propose an advanced personalized data-driven approach that uses a three-layer deep neural network to estimate BP based on PPG signals. Different from previous work, the proposed model analyzes the PPG signal in time-domain and automatically extracts the most critical features for this specific application, then uses a variation of recurrent neural networks (RNN) called Long-Short-Term-Memory (LSTM) to map the extracted features to the BP value associated with that time window. Experimental results on two separate standard hospital datasets, yielded absolute errors mean and absolute error standard deviation for systolic and diastolic BP values outperforming prior works.
Article
Blood pressure monitoring is very important for the prevention of cardiovascular diseases. In this paper, we proposed a multi-type features fusion (MTFF) neural network model for blood pressure (BP) prediction based on photoplethysmography (PPG). The model includes two convolutional neural networks (CNN) which used to train the morphological and frequency spectrum features of PPG signal, and one Bi-directional long short term memory (BLSTM) network which used to train the temporal features of PPG signal. These multi-features were fused through a specific fusion module after training, so more information of PPG signals were obtained and the hidden relationship between the fused features and blood pressure was established. The standard deviation (STD) and mean absolute error (MAE) of the fusion model are 7.25 mmHg and 5.59 mmHg respectively for systolic blood pressure (SBP), 4.48 mmHg and 3.36 mmHg respectively for diastolic blood pressure (DBP). The results are in full compliance with the Advancement of Medical Instrumentation (AAMI) and the British Hypertension Society (BHS) international standards. We conclude that the MTFF neural network proposed in this paper can accurately predict blood pressure. The significant difference from the traditional methods of BP prediction based on manual calculation of features is that our method automatically extracts PPG features through the deep learning model which can easily handle the complicated and tedious calculation. Compared with other similar BP prediction methods based on deep learning, three different features are trained and fused, which further improves the accuracy of BP prediction.
Article
Objective Regulation and inhibition of high blood pressure, known as hypertension are intricate, and it demands a continuous, accurate blood pressure measurement system. All the existing continuous non-invasive techniques own challenges such as exact placement of the sensor, reconstruction of arterial pressure from finger cuff, frequent and subject based calibration. This paper presents an algorithm based on new time-domain features for continuous blood pressure monitoring which is crucial in intensive care units and can be used to predict cardiovascular ailments. Methods Here, we propose the method to estimate BP that extracts informative features like Womersley number (α), QRS, QTc interval, SDI from ECG and PPG signals and regression techniques which are employed to estimate blood pressure continuously. Performance metrics like MAE, RMSE, r, bias & 95% CI are considered to validate the proposed method. To explore the relevance of proposed physiological features with the blood pressure, genetic algorithm(GA) with the random forest model is employed. Results Significant features like alpha, QRS complex, QT interval, SDI, heart rate are acquired. The best optimal feature set from GA reduced the MAE from 13.20 to 9.54 mmHg, 9.91 to 5.48 mmHg, 7.71 to 3.37 mmHg for SBP, DBP and MBP respectively. Conclusion Model build with ECG and PPG time-domain features outperform the model trained with PPG features alone. Results obtained from GA validates the significance of ECG features correlation with BP. Significance Identifying the associations of ECG features with BP helps in selecting relevant features which minimize the computational cost as well as errors in the assessment of BP.
Article
Blood pressure (BP) is a critical vital sign in health, measured millions of times per day worldwide. Current BP measurement requires cumbersome tools, is painful and can be inconvenient. Non-invasive cuffless BP measurement based on pulse arrival time (PAT) techniques allow an alternative way of monitoring BP in healthcare settings with refined wearability and user-friendly features. PAT extraction requires at least two measurements, one as a time reference and another to obtain time delay; there are several approaches to calculate the PAT from various sensors placed on the body. Commonly used signals are electrocardiography (ECG) and photoplethysmography (PPG), which can be recorded from a patients body using more than two separate sensors attachment set-ups. In this work, cuffless BP calculation based on five different PAT readings using Bio-impedance (BImp) at the shoulder as an alternative to PPG, has been investigated. Sensor placement is on the patients chest; which hides them beneath the patient's clothes making them more suitable for ambulatory monitoring systems. Technology performance was assessed using different postures, exercises and Glyceryl Trinitrate (GTN) spray doses; which provided stable, rising and falling BPs for evaluation. Data were collected from 41 participants who were sitting, standing and supine. Twenty-four of 41 participants undertook experiments including a handgrip task (isometric exercise), three periods of cycling on an exercise bike with light, moderate and heavy resistance settings and an observed rest period at the end. The remaining 17 of 41 subjects received GTN spray for predefined times with variable recovery periods afterwards. Different methods of PAT extraction from BImp data were compared for accuracy. Comparisons were made between PAT readings alone and PAT combined with Heart Rate and the combination model performed better when calculating BP. Simultaneously, data were collected using PPG-based PATs compared to BImp-based PATs. BImp-based PATs proved 3% more accurate than PPG-based PATs, demonstrating the potential superiority of BImp-based BP calculations.
Article
Although photoplethysmogram (PPG) and electrocardiogram (ECG) signals can be used to estimate blood pressure (BP) by extracting various features, the changes in morphological contours of both PPG and ECG signals due to various diseases of circulatory system and interaction of other physiological systems make the extraction of such features very difficult. In this work, we propose a waveform-based hierarchical Artificial Neural Network – Long Short Term Memory (ANN-LSTM) model for BP estimation. The model consists of two hierarchy levels, where the lower hierarchy level uses ANNs to extract necessary morphological features from ECG and PPG waveforms and the upper hierarchy level uses LSTM layers to account for the time domain variation of the features extracted by the lower hierarchy level. The proposed model is evaluated on 39 subjects using the Association for the Advancement of Medical Instrumentations (AAMI) standard and the British Hypertension Society (BHS) standard. The method satisfies both the standards in the estimation of systolic blood pressure (SBP) and diastolic blood pressure (DBP). For the proposed network, the mean absolute error (MAE) and the root mean square error (RMSE) for SBP estimation are 1.10 and 1.56 mmHg, respectively, and for DBP estimation are 0.58 and 0.85 mmHg, respectively. The performance of the proposed hierarchical ANN-LSTM model is found to be better than the other feature engineering-based networks. It is shown that the proposed model is able to automatically extract the necessary features and their time domain variations to estimate BP reliably in a noninvasive continuous manner. The method is expected to greatly facilitate the presently available mobile health-care gadgets in cuffless continuous BP estimation.
Article
We consider image transformation problems, where an input image is transformed into an output image. Recent methods for such problems typically train feed-forward convolutional neural networks using a \emph{per-pixel} loss between the output and ground-truth images. Parallel work has shown that high-quality images can be generated by defining and optimizing \emph{perceptual} loss functions based on high-level features extracted from pretrained networks. We combine the benefits of both approaches, and propose the use of perceptual loss functions for training feed-forward networks for image transformation tasks. We show results on image style transfer, where a feed-forward network is trained to solve the optimization problem proposed by Gatys et al in real-time. Compared to the optimization-based method, our network gives similar qualitative results but is three orders of magnitude faster. We also experiment with single-image super-resolution, where replacing a per-pixel loss with a perceptual loss gives visually pleasing results .
Article
This paper proposes a novel wavelet neural network algorithm for the continuous and noninvasive dynamic estimation of blood pressure (BP). Unlike prior algorithms, the proposed algorithm capitalizes on the correlation between photoplethysmography (PPG) and BP. Complete BP waveforms are reconstructed based on PPG signals to extract systolic blood pressure (SBP) and diastolic blood pressure (DBP). To improve the robustness, Daubechies wavelet is implemented as the hidden layer node function for the neural network. An optimized neural network structure is proposed to reduce the computational complexity. Further, this paper investigates an inhomogeneous resilient backpropagation (IRBP) algorithm to calculate the weight of hidden layer nodes. The IRBP improves the convergence speed and reconstruction accuracy. Multiparameter intelligent monitoring in Intensive Care (MIMIC) databases, which contain a variety of physiological parameters captured from patient monitors, are used to validate this algorithm. The standard deviation σ between reconstructed and actual BP signals is 4.4797 mmHg, which satisfies the American National Standards of the Association for the Advancement of Medical Instrumentation. The reconstructed BP waveform can be used to extract the SBP and DBP, whose standard deviations σ are 2.91 mmHg and 2.41 mmHg respectively. © 2015 Science China Press and Springer-Verlag Berlin Heidelberg
Article
Introduction: Pulse palpation is an important diagnostic tool in Traditional Chinese Medicine (TCM) and related Oriental medicine systems. However, mastering pulse diagnosis requires long-term experience and remains subjective up to a certain degree even in an advanced stage of practice. Accordingly, considerable research efforts have been spent to objectively measure the radial pulse and in further consequence automate Oriental pulse diagnosis by means of technological aids. This article gives the first comprehensive review about the current state of the art of this field covering topics such as developed pulse acquisition systems, suggested data preprocessing and feature extraction methods, and proposed classification approaches. Furthermore, persisting problems are pointed out and recommendations for future research directions are given. Methods: A literature search was conducted using scientific databases such as PubMed and ScienceDirect. ResearchGate, Academia, and GoogleScholar were used as additional literature sources. Key search terms were Chinese pulse diagnosis, Oriental pulse diagnosis, wrist pulse, standardization, objectification, automation, technological aids, sensors, data processing, classification, devices, acquisition systems and combinations and synonyms of these terms. Additionally, the reference lists of found articles were scanned to identify further relevant literature. Results: A total of 111 references were included into the review. Four main research themes emerged from the literature: (1) pulse waveform acquisition, (2) pulse waveform preprocessing, (3) pulse waveform feature extraction, (4) classifiers and classification objectives. Conclusion: Automating TCM pulse diagnosis by means of technological aids is a very active research field. Nevertheless, much remains to be done in terms of both technological developments and standardization of procedures.
Conference Paper
There is a relation, not always linear, between the blood pressure and the pulse duration, obtained from photoplethysmography (PPG) signal. In order to estimate the blood pressure from the PPG signal, in this paper the Artificial Neural Networks (ANNs) are used. Training data were extracted from the Multiparameter Intelligent Monitoring in Intensive Care waveform database for better representation of possible pulse and pressure variation. In total there were analyzed more than 15000 heartbeats and 21 parameters were extracted from each of them that define the input vector for the ANN. The comparison between estimated and reference values shows better accuracy than the linear regression method and satisfy the American National Standards of the Association for the Advancement of Medical Instrumentation.
Arterial hypertension
  • S Brouwers
S. Brouwers et al., "Arterial hypertension," 2021.
Generative adversarial networks
  • I J Goodfellow
I. J. Goodfellow et al., "Generative adversarial networks," arXiv preprint arXiv:1406.2661, 2014.