Enhancing few-shot personalized cuffless blood pressure estimation with self-supervised learning

Accurate cuffless blood pressure (BP) estimation remains challenging, particularly under dynamic conditions with significant intra-individual BP variations. This study introduces UTransBPNet, a novel, calibration-free model for cuffless BP estimation. It integrates a squeeze-and-excitation-enhanced Unet architecture for short-range feature extraction with a transformer and cross attention module to capture long-range dependencies from high-resolution, multi-channel physiological signals, further refined through an optimized fine-tuning scheme. Comprehensive validations were conducted across multiple dynamic datasets—Dataset_Drink, Dataset_Exercise, and Dataset_MIMIC—in both scenario-specific and cross-scenario settings. Results demonstrate that UTransBPNet outperformed existing models in tracking BP variations under dynamic conditions, achieving individual Pearson’s correlation coefficients of 0.61 ± 0.17 and 0.62 ± 0.13 for systolic BP (SBP) and diastolic BP (DBP) in Dataset_Drink, 0.82 ± 0.11 and 0.72 ± 0.18 in Dataset_Exercise, and low mean absolute differences (MADs) of 4.38 and 2.25 mmHg in Dataset_MIMIC. The analysis also highlights the impact of dataset characteristics on model performance, such as distribution shift, distribution imbalance and individual BP variability, highlighting the need for well-curated data to ensure generalizability. This study advances the development of robust, cuffless BP estimation models for real-world applications.

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.

Precision medicine requires cuffless blood pressure (BP) estimation technologies to achieve high accuracy levels suitable for healthcare monitoring. Recent advances, notably in personalized BP estimation methods utilizing transfer learning, have partially met these requirements. However, traditional personalized fine-tuning pipelines often ignore key physiological factors across age groups, thus limiting their effectiveness. This study identifies an important fine-tuning principle: target subjects benefit most when fine-tuned using a source model pretrained on an age-matched group. We introduce a novel two-level ensemble learning method, ATTE, which integrates multiple BP models for cuffless BP estimation based on pulse waves. It first performs feature-level ensemble learning (FEL) to combine age-group-specific fine-tuned models and enrich representations, then applies an optimized Bayesian model averaging (BMA) method that adaptively weights model outputs to boost accuracy and reduce model-selection uncertainty. Using only 50 transfer samples, ATTE improved systolic and diastolic BP estimation accuracy by 6.55% and 5.58% on a public dataset, compared to the traditional fine-tuning paradigm. It also achieved improvements of 5.10% and 8.30% on a self-collected dataset, which has been preliminarily validated in an elderly care scenario. Ablation and interpretability analyses validated the age-matched fine-tuning principle, the effectiveness of ATTE, and its model-agnostic generalizability. This work represents a pioneering step in optimizing personalized BP estimation and underscores its significant potential for real-world healthcare applications.

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.

Objective. Photoplethysmography (PPG) is a promising wearable technology that detects volumetric changes in microcirculation using a light source and a sensor on the skin’s surface. PPG has been shown to be useful for non-invasive blood pressure (BP) measurement. Deep learning-based BP measurements are now gaining popularity. However, almost all methods focus on 1D PPG. We aimed to design an end-to-end approach for estimating BP using image encodings from a 2D perspective. Approach. In this paper, we present a BP estimation approach based on an image encoding and fusion (BP-IEF) technique. We convert the PPG into five image encodings and use them as input. The proposed BP-IEF consists of two parts: an encoder and a decoder. In addition, three kinds of well-known neural networks are taken as the fundamental architecture of the encoder. The decoder is a hybrid architecture that consists of convolutional and fully connected layers, which are used to fuse features from the encoder. Main results. The performance of the proposed BP-IEF is evaluated on the UCI database in both non-mixed and mixed manners. On the non-mixed dataset, the root mean square error and mean absolute error for systolic BP (SBP) are 13.031 mmHg and 9.187 mmHg respectively, while for diastolic BP (DBP) they are 5.049 mmHg and 3.810 mmHg. On the mixed dataset, the corresponding values for SBP are 4.623 mmHg and 3.058 mmHg, while for DBP the values are 2.350 mmHg and 1.608 mmHg. In addition, both SBP and DBP estimation on the mixed dataset achieved grade A compared to the British Hypertension Society standard. The DBP estimation on the non-mixed dataset also achieved grade A. Significance. The results indicate that the proposed approach has the potential to improve on the current mobile healthcare for cuffless BP measurement.

A novel cuffless blood pressure (BP) measurement technology is proposed in this paper. Systolic BP (SBP) and diastolic BP (DBP) can be estimated by measuring pulse transit time (PTT) between radial and carotid artery. The pulse signals of artery are measured by split-ring resonator microwave sensors with high sensitivity in a non-contact manner. Instead of measuring resonance frequency deviation, the measurements can be performed by detecting amplitude variation for not only simplified circuits but also increasing sampling rate and sensitivity. In addition, cascading a notch filter can steepen the transition response of sensor to further improves 135 times sensitivity compared to traditional frequency-based method. From experiment results, PTT and BP show the significant negative correlation, and the correlation coefficient is more than -0.79. Based on the relational function of PTT and BP, SBP and DBP can be estimated accurately, and the average error is lower than 8.4%.

The key to cuffless blood pressure (BP) estimation lies in identifying the efficient noninvasive features or signals that can indicate BP changes. Most these features are identified based on their correlation with BP. However, the correlation does not imply causation, as there might be confounders impacting the relationship between the features and BP. In this study, we introduce causal inference to identify the confounders that can possibly affect the accuracy of the commonly studied pulse transit time (PTT) based cuffless BP estimation. We further propose a hierarchical regression model to address the impact of the confounder thereby improving the estimation performance. We first build the causal graph of BP and common features extracted from electrocardiogram (ECG) and photoplethysmogram (PPG) signals with the fast causal inference algorithm and causal generative neural network. We found heart rate is the main confounder that impacting the relationship between PTT and BP. To consider the effect of the confounder, we then developed a hierarchical regression model for cuffless BP estimation. We validated the proposed method on 17 subjects, and the results show that the mean absolute difference (MAD) of the estimated systolic BP and diastolic BP were 6.29 mmHg and 5.03 mmHg respectively, which significantly outperformed the linear regression model without the hierarchical strategy.

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

Pulse transit time (PTT) has been identified as a potential indicator to estimate blood pressure (BP) continuously and cufflessly in the previous studies, but it is still unsatisfactory for clinical uses. In order to improve the BP estimation accuracy, another candidate signal, photoplethysmogram (PPG) has been used to estimate continuous BP. In this study, a new PPG indicator, the intensity ratio of the first derivative wave of PPG (1st dPIR), was proposed and a novel BP estimation algorithm by using both PTT and 1st dPIR was developed with an attempt to further increase the BP estimation accuracy. The performance of the proposed algorithm was examined and validated using the ECG and PPG recordings from 22 healthy subjects when they performed both mental arithmetic stress and Valsalva's manoeuvre tests that could induce their BP fluctuations. The continuous Finapres® BP were measured with a finger cuff as a referenced BP value. Our results showed that using the proposed BP estimation algorithm, the grand absolute mean and standard deviation (SD) errors between the estimated and referenced BP were 3.22±8.02 mmHg for systolic BP (SBP) and 3.13±4.82mmHg for diastolic BP (DBP). It could achieve a decrease of 0.34±0.27 mmHg in SBP and 0.33±0.44 mmHg in DBP on the grand absolute mean and SD estimation errors in comparison to the state-of-the-art PPT-based method. Findings from this study suggest that the newly proposed PPG indicator would be promising for improving the accuracy of current PTT-based continuous and cuffless BP estimation algorithms.

Deep learning models for cuffless blood pressure monitoring from PPG signals using attention mechanism

To develop a cuffless method for estimating blood pressure (BP) from fingertip strain plethysmography (SPG) recordings. A custom-built micro-electromechanical systems (MEMS) strain sensor is employed to record heartbeat-induced vibrations at the fingertip. An XGboost regressor is then trained to relate SPG recordings to beat-to-beat systolic BP (SBP), diastolic BP (DBP), mean arterial pressure (MAP) values. For this purpose, each SPG segment in this setup is represented by a feature vector consisting of cardiac time interval, amplitude features, statistical properties, and demographic information of the subjects. In addition, a novel concept, coined geometric features, are introduced and incorporated into the feature space to further encode the dynamics in SPG recordings. The performance of the regressor is assessed on 32 healthy subjects through 5-fold cross-validation (5-CV) and leave-subject-out cross validation (LSOCV). Mean absolute errors (MAEs) of 3.88 mmHg and 5.45 mmHg were achieved for DBP and SBP estimations, respectively, in the 5-CV setting. LSOCV yielded MAEs of 8.16 mmHg for DBP and 16.81 mmHg for SBP. Through feature importance analysis, 3 geometric and 26 integral-related features introduced in this work were identified as primary contributors to BP estimation. The method exhibited robustness against variations in blood pressure level (normal to critical) and body mass index (underweight to obese), with MAE ranges of [1.28, 4.28] mmHg and [2.64, 7.52] mmHg, respectively. The findings suggest high potential for SPG-based BP estimation at the fingertip. This study presents a fundamental step towards the augmentation of optical sensors that are susceptible to dark skin tones.

Continuous blood pressure (BP) monitoring in daily life is needed to enable early detection of hypertension and improve its control. Although pulse transit time (PTT)-based BP estimation represents a promising approach, it still lacks of performance in systolic blood pressure (SBP) estimation and its use in daily life is limited owing to the requirement of bulky systems for measurement of PTT. This study aims at developing a wearable system providing continuous PTT measurement and enhanced SBP estimation for continuous BP monitoring. A single chest-worn device was developed, which measures the photoplethysmogram and seismocardiogram simultaneously, and thereby obtains PTT by time difference between two signals. A multivariate model using the seismocardiogram amplitude (SA) in conjunction with PTT was proposed for SBP estimation, and validated against 30 healthy males (31.47 ± 7.23 years old). Performance of the proposed model was compared with that of conventional univariate models using PTT or pulse arrival time in two types of BP interventions, and for the verification of real use in daily life, performance assessment with calibration and BP monitoring in daily life were conducted. The results suggested that the proposed model (1) outperformed the conventional models, (2) showed potential to be generalized with just a simple calibration, and (3) demonstrated the potential of continuous BP monitoring in daily life. In conclusion, the presented system provides an improved performance of continuous BP monitoring by using a combination of PTT and SA with a convenient and compact single chest-worn device, and thus, it can contribute to mobile healthcare services.

Continuous blood pressure (BP) provides essential information for monitoring one's health condition. However, BP is currently monitored using uncomfortable cuff-based devices, which does not support continuous BP monitoring. This paper aims to introduce a blood pressure monitoring algorithm based on only photoplethysmography (PPG) signals using the deep neural network (DNN). The PPG signals are obtained from 125 unique subjects with 218 records and filtered using signal processing algorithms to reduce the effects of noise, such as baseline wandering, and motion artifacts. The proposed algorithm is based on pulse wave analysis of PPG signals, extracted various domain features from PPG signals, and mapped them to BP values. Four feature selection methods are applied and yielded four feature subsets. Therefore, an ensemble feature selection technique is proposed to obtain the optimal feature set based on major voting scores from four feature subsets. DNN models, along with the ensemble feature selection technique, outperformed in estimating the systolic blood pressure (SBP) and diastolic blood pressure (DBP) compared to previously reported approaches that rely only on the PPG signal. The coefficient of determination ( ) and mean absolute error (MAE) of the proposed algorithm are 0.962 and 2.480 mmHg, respectively, for SBP and 0.955 and 1.499 mmHg, respectively, for DBP. The proposed approach meets the Advancement of Medical Instrumentation standard for SBP and DBP estimations. Additionally, according to the British Hypertension Society standard, the results attained Grade A for both SBP and DBP estimations. It concludes that BP can be estimated more accurately using the optimal feature set and DNN models. The proposed algorithm has the potential ability to facilitate mobile healthcare devices to monitor continuous BP.

Introduction Blood pressure (BP) has been a potential risk factor for cardiovascular diseases. BP measurement is one of the most useful parameters for early diagnosis, prevention, and treatment of cardiovascular diseases. At present, BP measurement mainly relies on cuff-based techniques that cause inconvenience and discomfort to users. Although some of the present prototype cuffless BP measurement techniques are able to reach overall acceptable accuracies, they require an electrocardiogram (ECG) and a photoplethysmograph (PPG) that make them unsuitable for true wearable applications. Therefore, developing a single PPG-based cuffless BP estimation algorithm with enough accuracy would be clinically and practically useful. Methods The University of Queensland vital sign dataset (online database) was accessed to extract raw PPG signals and its corresponding reference BPs (systolic BP and diastolic BP). The online database consisted of PPG waveforms of 32 cases from whom 8133 (good quality) signal segments (5 s for each) were extracted, preprocessed, and normalised in both width and amplitude. Three most significant pulse features (pulse area, pulse rising time, and width 25%) with their corresponding reference BPs were used to train and test three machine learning algorithms (regression tree, multiple linear regression (MLR), and support vector machine (SVM)). A 10-fold cross-validation was applied to obtain overall BP estimation accuracy, separately for the three machine learning algorithms. Their estimation accuracies were further analysed separately for three clinical BP categories (normotensive, hypertensive, and hypotensive). Finally, they were compared with the ISO standard for noninvasive BP device validation (average difference no greater than 5 mmHg and SD no greater than 8 mmHg). Results In terms of overall estimation accuracy, the regression tree achieved the best overall accuracy for SBP (mean and SD of difference: −0.1 ± 6.5 mmHg) and DBP (mean and SD of difference: −0.6 ± 5.2 mmHg). MLR and SVM achieved the overall mean difference less than 5 mmHg for both SBP and DBP, but their SD of difference was >8 mmHg. Regarding the estimation accuracy in each BP categories, only the regression tree achieved acceptable ISO standard for SBP (−1.1 ± 5.7 mmHg) and DBP (−0.03 ± 5.6 mmHg) in the normotensive category. MLR and SVM did not achieve acceptable accuracies in any BP categories. Conclusion This study developed and compared three machine learning algorithms to estimate BPs using PPG only and revealed that the regression tree algorithm was the best approach with overall acceptable accuracy to ISO standard for BP device validation. Furthermore, this study demonstrated that the regression tree algorithm achieved acceptable measurement accuracy only in the normotensive category, suggesting that future algorithm development for BP estimation should be more specific for different BP categories.

Arterial blood pressure is one of the most important physiological parameters for the noninvasive diagnosis and monitoring of cardiovascular diseases. This study examines the relationships between arterial blood pressure and certain features of the photoplethysmographic (PPG) signals obtained from 15 young healthy subjects. Width of 1/2 pulse amplitude, width of 2/3 pulse amplitude, systolic upstroke time and diastolic time of the pulse were selected as features of the PPG signals. The experiment involved three sessions: rest, step-climbing exercise, and recovery from exercise. It was found that the diastolic time has higher correlation with systolic blood pressure (SBP) and diastolic blood pressure (DBF) than other features. The mean differences between the estimated and the measured blood pressure using a standard device are 0.21 mm Hg for SBP and 0.02 mm Hg for DBF. The corresponding standard deviations are 7.32 mm Hg for SBP and 4.39 mm Hg for DBF. The preliminary results indicate that it is possible to use the photoplethysmography only for cuffless and continuous estimation of arterial blood pressure.