Development of an Improved Stacked U-Net Model for Cuffless Blood Pressure Estimation Based on PPG Signals

Cuffless blood pressure (BP) monitoring is a transformative technology that enables continuous noninvasive measurement of BP, which is critical for early diagnosis and prevention of cardiovascular diseases (CVD). However, current predictive models exhibit performance limitations across underrepresented demographics. This limitation is particularly pronounced in hypertensive populations, where the large BP volatility challenges accuracy of the model for BP estimation. Addressing these demographic disparities is crucial to ensure the objectivity and fairness of model for cuffless BP estimation. This paper proposes a self-organizing mixture of experts (SO-MoE) framework for BP estimation, which implements automated expert allocation mechanisms based on patient-specific characteristics. This specialized design enables precise expert selection for individualized BP estimation. We evaluated this system on the CAS-BP dataset of over seven hundred subjects and compared it with several deep learning models. The proposed system demonstrated promising performance for BP estimation with mean absolute errors (MAE) of 5.59 mmHg and 4.07 mmHg for systolic BP (SBP) and diastolic BP (DBP) respectively. Besides, the SO-MoE yielded estimation errors of -0.24 ± 7.85 mmHg for SBP and 0.33 ± 5.49 mmHg for DBP in the overall population. Key innovations of the proposed framework include reduced computational overhead, sustained temporal stability, and consistent performance across diverse demographic groups. These results establish a foundation for equitable and clinically accurate BP estimation, addressing the representation challenges inherent in cuffless BP monitoring.

A substantial barrier to the clinical adoption of cuffless blood pressure (BP) monitoring techniques is the lack of unified error standards and methods of estimating measurement uncertainty. This study proposes a fusion approach to improve accuracy and estimate prediction interval (PI) as a proxy for uncertainty for cuffless blood BP monitoring. BP was estimated during activities of daily living using three model architectures: nonlinear autoregressive models with exogenous inputs, feedforward neural network models, and pulse arrival time models. Multiple one-class support vector machine (OCSVM) models were trained to cluster data in terms of the percentage of outliers. New BP estimates were then assigned to a cluster using the OCSVMs hyperplanes, and the PIs were estimated using the BP error standard deviation associated with different clusters. The OCSVM was used to estimate the PI for the three BP models. The three BP estimations from the models were fused using the covariance intersection fusion algorithm, which improved BP and PI estimates in comparison with individual model precision by up to 24%. The employed model fusion shows promise in estimating BP and PI for potential clinical uses. The PI indicates that about 71%, 64%, and 29% of the data collected from sitting, standing, and walking can result in high-quality BP estimates. Our PI estimator offers an effective uncertainty metric to quantify the quality of BP estimates and can minimize the risk of false diagnosis.

Precise blood pressure (BP) estimation is vital for diagnosing arterial hypertension and other cardiovascular ailments. The photoplethysmogram (PPG)-based cuff-less BP measurement is an alternative to the traditional cuff-based systems. The morphological, temporal, and frequency-domain-based features have been used for BP estimation in the PPG-based BP measurement systems. However, dealing with varying signal morphology and feature dependency on the fiducial points in PPG contours remains challenging and limits the performance of the existing BP estimation algorithms, especially in wearable devices. This work presents a novel approach that considers the nonlinear features of PPG signals evaluated using higher order derivatives. In particular, the PPG signal’s third and fourth derivative contours are used to extract features, such as fractal dimension, bubble entropy (BE), Lyapunov exponent, and moving slope. Machine learning algorithms such as random forest (RF), extreme gradient boosting (XGBoost), and support vector regression (SVR) models are used for BP estimation using the nonlinear features of PPG signals. The estimated BP values are further categorized based on five broad classes based on the BP stratification criteria such as hypotension, normal, prehypertension, stage-I, and stage-II hypertension, respectively. The performance of the suggested approach is evaluated using PPG signals from three publicly available databases (multiparameter intelligent monitoring in intensive care (MIMIC)-I, II, and III). The proposed estimation approach has outperformed the recent existing algorithms and achieved a minimum value of mean absolute error (MAE) ± standard deviation (STD) in (systolic BP (SBP) and diastolic BP (DBP) values) as <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${0}.{74}\pm {2}.{42}$ </tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${0}.{35} \pm {1}.{06}$ </tex-math></inline-formula> , respectively, for the MIMIC-I database. The suggested approach has also achieved grade-A on the British Hypertension Society (BHS) standard.

ObjectiveContinuous blood pressure (BP) provides valuable information for the disease management of patients with arrhythmias. The traditional intra-arterial method is too invasive for routine healthcare settings, whereas cuff-based devices are inferior in reliability and comfortable for long-term BP monitoring during arrhythmias. The study aimed to investigate an indirect method for continuous and cuff-less BP estimation based on electrocardiogram (ECG) and photoplethysmogram (PPG) signals during arrhythmias and to test its reliability for the determination of BP using invasive BP (IBP) as reference.MethodsThirty-five clinically stable patients (15 with ventricular arrhythmias and 20 with supraventricular arrhythmias) who had undergone radiofrequency ablation were enrolled in this study. Their ECG, PPG, and femoral arterial IBP signals were simultaneously recorded with a multi-parameter monitoring system. Fifteen features that have the potential ability in indicating beat-to-beat BP changes during arrhythmias were extracted from the ECG and PPG signals. Four machine learning algorithms, decision tree regression (DTR), support vector machine regression (SVR), adaptive boosting regression (AdaboostR), and random forest regression (RFR), were then implemented to develop the BP models.ResultsThe results showed that the mean value ± standard deviation of root mean square error for the estimated systolic BP (SBP), diastolic BP (DBP) with the RFR model against the reference in all patients were 5.87 ± 3.13 and 3.52 ± 1.38 mmHg, respectively, which achieved the best performance among all the models. Furthermore, the mean error ± standard deviation of error between the estimated SBP and DBP with the RFR model against the reference in all patients were −0.04 ± 6.11 and 0.11 ± 3.62 mmHg, respectively, which complied with the Association for the Advancement of Medical Instrumentation and the British Hypertension Society (Grade A) standards.ConclusionThe results indicated that the utilization of ECG and PPG signals has the potential to enable cuff-less and continuous BP estimation in an indirect way for patients with arrhythmias.

This paper presents a new solution that enables the use of transfer learning for cuff-less blood pressure (BP) monitoring via short duration of photoplethysmogram (PPG). The proposed method estimates BP with low computational budget by 1) creating images from segments of PPG via visibility graph (VG), hence, preserving the temporal information of the PPG waveform, 2) using pre-trained deep convolutional neural network (CNN) to extract feature vectors from VG images, and 3) solving for the weights and bias between the feature vectors and the reference BPs with ridge regression. Using the University of California Irvine (UCI) database consisting of 348 records, the proposed method achieves a best error performance of 0.00±8.46 mmHg for systolic blood pressure (SBP), and -0.04±5.36 mmHg for diastolic blood pressure (DBP), respectively, in terms of the mean error (ME) and the standard deviation (SD) of error, ranking grade B for SBP and grade A for DBP under the British Hypertension Society (BHS) protocol. Our novel data-driven method offers a computationally-efficient end-to-end solution for rapid and user-friendly cuff-less PPG-based BP estimation.

Blood pressure measurement is a significant part of preventive healthcare and has been widely used in clinical risk and disease management. However, conventional measurement does not provide continuous monitoring and sometimes is inconvenient with a cuff. In addition to the traditional cuff-based blood pressure measurement methods, some researchers have developed various cuff-less and noninvasive blood pressure monitoring methods based on Pulse Transit Time (PTT). Some emerging methods have employed features of either photoplethysmogram (PPG) or electrocardiogram (ECG) signals, although no studies to our knowledge have employed the combined features from both PPG and ECG signals. Therefore this study aims to investigate the performance of a predictive, machine learning blood pressure monitoring system using both PPG and ECG signals. It validates that the employment of the combination of PPG and ECG signals has improved the accuracy of the blood pressure estimation, compared with previously reported results based on PPG signal only. © 2018 Institution of Engineering and Technology. All rights reserved.

Cuffless blood pressure estimation from PPG signals and its derivatives using deep learning models

Continuous cuff-less blood pressure (BP) monitoring using pulse transit time (PTT) or pulse arrival time (PAT) has shown great prospect in the estimation of systolic blood pressure (SBP) and diastolic blood pressure (DBP) due to the convenience. To further improve the accuracy and efficiency, in this paper, we present a novel model for continuous cuff-less BP estimation, which is derived from empirical formulas based on physical model of arteries. We then apply the proposed model on 26 subjects with a cuff based model as a reference and compare the correlation coefficients for validation. Numerical results suggest that the estimation method based on the proposed model outperforms traditional approaches by achieving correlation coefficients for SBP and DBP as 0.953 and 0.872 respectively using PAT index, and as 0.932 and 0.874 respectively using PTT index. Besides, as British Hypertension Society (BHS) protocol implied, the results are consistent with the highest grade of A in the estimation of both SBP and DBP, which demonstrates the effectiveness of the proposed model for continuous BP estimation.

Cuffless blood pressure estimation using chaotic features of photoplethysmograms and parallel convolutional neural network

In this study, we conducted research on a deep learning-based blood pressure (BP) estimation model suitable for wearable environments. To measure BP while wearing a wearable watch, it needs to be considered that computing power for signal processing is limited and the input signals are subject to noise interference. Therefore, we employed a convolutional neural network (CNN) as the BP estimation model and utilized time-series electrocardiogram (ECG) and photoplethysmogram (PPG) signals, which are quantifiable in a wearable context. We generated periodic input signals and used differential and thresholding methods to decrease noise in the preprocessing step. We then applied a max-pooling technique with filter sizes of 2 × 1 and 5 × 1 within a 3-layer convolutional neural network to estimate BP. Our method was trained, validated, and tested using 2.4 million data samples from 49 patients in the intensive care unit. These samples, totaling 3.1 GB were obtained from the publicly accessible MIMIC database. As a result of a test with 480,000 data samples, the average root mean square error in BP estimation was 3.41, 5.80, and 2.78 mm Hg in the prediction of pulse pressure, systolic BP (SBP), and diastolic BP (DBP), respectively. The cumulative error percentage less than 5 mm Hg was 68% and 93% for SBP and DBP, respectively. In addition, the cumulative error percentage less than 15 mm Hg was 98% and 99% for SBP and DBP. Subsequently, we evaluated the impact of changes in input signal length (1 cycle vs. 30 s) and the introduction of noise on BP estimation results. The experimental results revealed that the length of the input signal did not significantly affect the performance of CNN-based analysis. When estimating BP using noise-added ECG signals, the mean absolute error (MAE) for SBP and DBP was 9.72 and 6.67 mm Hg, respectively. Meanwhile, when using noise-added PPG signals, the MAE for SBP and DBP was 26.85 and 14.00 mm Hg, respectively. Therefore, this study confirmed that using ECG signals rather than PPG signals is advantageous for noise reduction in a wearable environment. Besides, short sampling frames without calibration can be effective as input signals. Furthermore, it demonstrated that using a model suitable for information extraction rather than a specialized deep learning model for sequential data can yield satisfactory results in BP estimation.

This study introduces a novel dual-path deep learning framework using Photoplethysmogram (PPG) signals to address key challenges in continuous, non-invasive cuffless Blood Pressure (BP) monitoring. To this end, we introduce -for the first time- the use of two novel deep neural network architectures: Conformer-Transformer and 1D Swin Transformer. These architectures are adapted here to model both the morphological structure and rhythmic dynamics of PPG signals. This cross-domain transfer enables Arterial Blood Pressure (ABP) waveform reconstruction and significantly improves the accuracy and physiological consistency of Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP) estimation. Extensive experiments on two public datasets demonstrate that our methods consistently outperform mainstream baselines across multiple key metrics. Specifically, the Conformer-Transformer achieved the lowest Mean Absolute Error (MAE) of 2.979 mmHg for systolic and 1.603 mmHg for diastolic BP, improving upon previous studies by 9.6% and 8.4%, respectively, while delivering the best waveform reconstruction performance too. The Swin Transformer achieved a systolic MAE of 3.034 mmHg and a diastolic MAE of 1.714 mmHg. All experimental results conform to the British Hypertension Society (BHS) grade A and Association for the Advancement of Medical Instrumentation (AAMI) standards.

Cuffless blood pressure (BP) estimation models have been extensively studied in recent years. However, due to aleatoric and epistemic uncertainty, these methods make it difficult to provide reliable and accurate BP estimations meeting the clinical requirement. In this study, we propose a novel method to quantify the uncertainty of the cuffless BP estimation model and combine epistemic uncertainty with conformal prediction to generate a statistically rigorous uncertainty interval (UI). First, we develop a deep evidential regression (DER) model to estimate BP with five features extracted from a noninvasive photoplethysmogram (PPG) signal. The DER model predicts a distribution of the target BP instead of a point estimation, so it can offer an analytical solution for the estimation uncertainty. We then utilize conformal prediction to generate a UI that covers the reference BP with a defined probability. We validate the proposed method on 37 subjects with continuous Finpres BP as a reference. The results show that the mean absolute difference (MAD) of systolic BP (SBP) and diastolic BP (DBP) estimations with the proposed method are 5.56 and 3.18 mmHg, respectively. The estimated UI can capture the reference SBP and DBP with a coverage rate of 94.8% and 95.9%, respectively. The findings indicate that the proposed method has the potential to empower more reliable and accurate cuffless BP measurement.

Due to the growing public awareness of cardiovascular disease (CVD), blood pressure (BP) estimation models have been developed based on physiological parameters extracted from both electrocardiograms (ECGs) and photoplethysmograms (PPGs). Still, in order to enhance the usability as well as reduce the sensor cost, researchers endeavor to establish a generalized BP estimation model using only PPG signals. In this paper, we propose a deep neural network model capable of extracting 32 features exclusively from PPG signals for BP estimation. The effectiveness and accuracy of our proposed model was evaluated by the root mean square error (RMSE), mean absolute error (MAE), the Association for the Advancement of Medical Instrumentation (AAMI) standard and the British Hypertension Society (BHS) standard. Experimental results showed that the RMSEs in systolic blood pressure (SBP) and diastolic blood pressure (DBP) are 4.643 mmHg and 3.307 mmHg, respectively, across 9000 subjects, with 80.63% of absolute errors among estimated SBP records lower than 5 mmHg and 90.19% of absolute errors among estimated DBP records lower than 5 mmHg. We demonstrated that our proposed model has remarkably high accuracy on the largest BP database found in the literature, which shows its effectiveness compared to some prior works.

Blood pressure measurements are one of the most routinely performed medical tests globally. Blood pressure is an important metric since it provides information that can be used to diagnose several vascular diseases. Conventional blood pressure measurement systems use cuff-based devices to measure the blood pressure, which may be uncomfortable and sometimes burdensome to the subjects. Therefore, in this study, we propose a cuffless blood pressure estimation model based on Monte Carlo simulation (MCS). We propose a heterogeneous finger model for the MCS at wavelengths of 905 nm and 940 nm. After recording the photon intensities from the MCS over a certain range of blood pressure values, the actual photoplethysmography (PPG) signals were used to estimate blood pressure. We used both publicly available and self-made datasets to evaluate the performance of the proposed model. In case of the publicly available dataset for transmission-type MCS, the mean absolute errors are 3.32 ± 6.03 mmHg for systolic blood pressure (SBP), 2.02 ± 2.64 mmHg for diastolic blood pressure (DBP), and 1.76 ± 2.8 mmHg for mean arterial pressure (MAP). The self-made dataset is used for both transmission- and reflection-type MCSs; its mean absolute errors are 2.54 ± 4.24 mmHg for SBP, 1.49 ± 2.82 mmHg for DBP, and 1.51 ± 2.41 mmHg for MAP in the transmission-type case as well as 3.35 ± 5.06 mmHg for SBP, 2.07 ± 2.83 mmHg for DBP, and 2.12 ± 2.83 mmHg for MAP in the reflection-type case. The estimated results of the SBP and DBP satisfy the requirements of the Association for the Advancement of Medical Instrumentation (AAMI) standards and are within Grade A according to the British Hypertension Society (BHS) standards. These results show that the proposed model is efficient for estimating blood pressures using fingertip PPG signals.

Objective: The accuracy of cuffless and continuous blood pressure (BP) estimation has been improved, but it is still unsatisfactory for clinical uses. This study was designed to further increase BP estimation accuracy. Approach: In this study, a number of new indicators were extracted from photoplethysmogram (PPG) recordings and a linear regression method was used to construct BP estimation models based on the PPG indicators and pulse transit time (PTT). The performance of the BP estimation models was evaluated by the PPG recordings from 22 subjects when they performed mental arithmetic stress and Valsalva’s manoeuvre tasks that could induce BP fluctuations. Main results: Our results showed that the best PPG-based BP estimation model could achieve a decrease of 0.31 ± 0.08 mmHg in systolic BP (SBP) and 0.33 ± 0.01 mmHg in diastolic BP (DBP) on estimation errors of grand absolute mean (GAM) and standard deviation (GSD) in comparison to the previously reported PPG-based methods. The best estimation model based on the combination of PPG and PPT could achieve a decrease (GAM & GSD) of 0.81 ± 0.95 mmHg in SBP and 0.75 ± 0.54 mmHg in DBP in comparison to the PPT-based methods. Significance: The findings suggest that the newly proposed PPG indicators would be promising for improving the accuracy of continuous and cuffless BP estimation.