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

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.

To compare the reproducibility of clinic and ambulatory blood pressure measurements in elderly hypertensive subjects. Twenty-two untreated elderly hypertensives, with a clinic systolic blood pressure (SBP) > 160 mmHg and/or diastolic blood pressure (DBP) > 95 mmHg, and a mean age of 76 years (range 66-86). Following three supine clinic blood pressure readings the subjects underwent 24-h non-invasive ambulatory blood pressure monitoring, measurements being taken at 20-min intervals from 0700 to 2200 h and at 30-min intervals from 2200 to 0700 h. Measurements were repeated during a further visit at a median interval of 10 weeks (range 1-10 months). Daytime ambulatory SBP levels were 20 mmHg (95% confidence interval 14-27 mmHg, P < 0.001) lower than clinic SBP, although DBP values were similar. The mean 24-h ambulatory SBP and DBP reproducibility [assessed by the standard deviation of differences (SDD) between visits] was significantly better than that for mean clinic blood pressure (SBP 6.3 versus 17.4 mmHg, P < 0.001; DBP 4.8 versus 7.0 mmHg, respectively, P < 0.05). With daytime defined as 1000-1959 h, the SDD between visits was 12.4 mmHg for SBP and 8.3 mmHg for DBP, but with daytime defined as 0700-2159 h, the SDD fell to 6.0 mmHg for SBP and 4.8 mmHg for DBP, values almost identical to those obtained with full 24-h blood pressure monitoring. There was no difference in night-time blood pressure reproducibility, whether night-time was defined as 2400-0559 h or 2200-0659 h. Both 24-h and daytime ambulatory blood pressure monitoring significantly improve the reproducibility of blood pressure measurements compared with clinic blood pressure readings in elderly hypertensive subjects. Increasing the number of daytime blood pressure readings by 50% (from 30 to 45) reduced the variability of blood pressure measurement by 50%. Twenty-four-hour ambulatory blood pressure monitoring is of value in obtaining reproducible blood pressure measurements in elderly hypertensive subjects. However, more than 30 readings are needed during a daytime recording to significantly reduce variability compared with repeated clinic measurements, although night-time variability is not significantly affected if the number of readings is reduced to 12 over a 6-h period.

PurposeTo validate the Dinamap ProCare 200 blood pressure (BP) monitor against a mercury sphygmomanometer in children 7 to 18 years old in accordance with the 2010 International Protocol of European Society of Hypertension (ESH-IP2) and the British Hypertension Society (BHS) protocol.MethodsForty-five children were recruited for the study. A validation procedure was performed following the protocol based on the ESH-IP2 and BHS protocols for children and adolescents. Each subject underwent 7 sequential BP measurements alternatively with a mercury sphygmomanometer and the test device by trained nurses. The results were analyzed according to the validation criteria of ESH-IP2.ResultsThe mean (±SD) difference in the absolute BP values between test device and mercury sphygmomanometer readings was 1.85±1.65 mmHg for systolic BP (SBP) and 4.41±3.53 mmHg for diastolic BP (DBP). These results fulfilled the Association for the Advancement of Medical Instrumentation criterion of a mean±SD below 5±8 mmHg for both SBP and DBP. The percentages of test device-observer mercury sphygmomanometer BP differences within 5, 10, and 15 mmHg were 96%, 100%, and 100% for SBP, and 69%, 92%, and 100% for DBP, respectively, in the part 1 analysis; both SBP and DBP passed the part 1 criteria. In the part 2 analysis, SBP passed the criteria but DBP failed.ConclusionAlthough the Dinamap ProCare 200 BP monitor failed an adapted ESH-IP2, SBP passed. When comparing BP readings measured by oscillometers and mercury sphygmomanometers, one has to consider the differences between them, particularly in DBP, because DBP can be underestimated.

Continuous and non-invasive blood pressure (BP) monitoring is crucial for effective health management in patients with arrhythmia. Cuffless BP measurement methods, utilizing electrocardiogram (ECG) and photoplethysmogram (PPG) signals combined with deep learning, have shown great promise. However, existing deep learning algorithms face challenges in representing physiological data due to the position-sensitivity of such data and the need for global context to accurately interpret it, resulting in limited accuracy. This study introduces BM-BPW, a novel deep learning-based model for continuous BP waveform estimation from ECG and PPG signals for arrhythmia patients. The bidirectional Mamba architecture, combined with position embeddings, is employed in the BM-BPW model to enable global context modeling and locally-aware understanding. A multi-scale convolutional neural network (CNN) feature extractor is also employed to capture the local features. The model was evaluated using a dataset of 36,537 beats collected from 48 arrhythmia patients, with invasive arterial waveforms as the reference. Experimental results indicate that BM-BPW effectively manages BP waveform variability caused by arrhythmia, achieving estimation errors of 0.04 ± 5.03 mmHg and -0.05 3.20 mmHg for systolic BP (SBP), and diastolic BP (DBP),±respectively. These results outperform state-of-the-art methods. This study provides an accurate and efficient solution for cuffless continuous BP waveform monitoring for arrhythmia patients, offering substantial potential for improving clinical management in this population.

Background Hypertension is a major risk factor for heart failure (HF). Randomized trials have shown that treatment to a systolic blood pressure (SBP) &amp;lt;130 mmHg decreases the risk of HF. Less is known about the relationship between blood pressure (BP) level and new-onset HF in patients with treated hypertension under real-world conditions. Purpose To determine the risk of new-onset HF in different BP categories, as defined by SBP and diastolic blood pressure (DBP) in HTN guidelines, and at continuous SBP levels in a real-world population of patients with hypertension in primary care. Methods Observational cohort study of patients with hypertension from a large Swedish primary care-register with data linked from national and regional registers. Patients with known HF or loop diuretics were excluded. The index date was the day of the first BP measurement at least a year after a diagnosis of hypertension and patients were included from 2011 to 2017. The risk of new-onset HF, defined as the occurrence of a HF diagnosis after the index date, was estimated with Cox regression models, for BP categories as well as for continuous SBP. The BP categories were Optimal (&amp;lt;120 mmHg SBP and &amp;lt;80 mmHg DBP), Normal (120 – 129 mmHg SBP and/or 80 – 84 mmHg DBP), High normal (130 – 139 mmHgSBP and/or 85 – 89 mmHg DBP), Grade 1 hypertension (140 – 159 mmHg SBP and/or 90 – 99 mmHg DBP), Grade 2 hypertension (160 – 179 mmHg SBP and/or 100 – 109 mmHg DBP) and Grade 3 hypertension (≥180 mmHg SBP or ≥110 mmHg DBP). Normal BP was the reference in the categorical analysis. The models were adjusted for age, sex and cardiovascular comorbidities. A restricted cubic spline with four knots was used to plot the hazard ratios (HR) from the analysis of SBP. Results In all, 211 624 patients with hypertension were included (mean age 66.9 years, 50.7% female). The largest proportion of patients had Grade 1 hypertension (36.9%) and the mean SBP was 139.4 mmHg. During a median follow-up of 5.0 years (interquartile range 3.0 – 6.1), 6 520 patients were diagnosed with HF. The cumulative incidence of HF was highest in patients with Optimal BP (3.7%) and lowest in those with High normal BP (2.6%). The adjusted hazard ratio for HF was highest in the patients with Grade 3 hypertension (HR 1.42, 95% confidence interval (CI) 1.23 – 1.64). Only patients in the High normal category had a lower risk of HF (HR 0.89, 95% CI 0.82 – 0.97) than the reference Normal category. See figure 1. There was a U-shaped relationship between HF and continuous SBP, with a nadir in risk around 130 mmHg. See figure 2. Conclusions The lowest risk of HF in treated hypertension was observed in patients in the High normal category (130 – 139 mmHg SBP and/or 85-89 mmHg DBP), and at continuous SBP at around 130 – 135 mmHg. Patients with lower BP had higher risk of new-onset HF. This contrasts to findings from randomized trials, where treatment to SBP levels lower than 130 mmHg was associated with lower risk of HF.Risk of heart failure by BP category Risk of heart failure by continuous SBP

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.

To test a prototype hydraulic, non-invasive, continuous finger blood pressure monitor based on the volume-clamp principle for procedure-related factors likely to influence precision. The influence of these factors was determined by repeatability of finger blood pressure measurement and the relationship to contralateral arm-cuff blood pressure. Repeated blood pressure measurements from three different fingers were made in 60 subjects following re-initialization of the device and re-insertion of the finger into the cuff. Repeatability was assessed in relation to simultaneous arm-cuff readings. Drift in arm-finger discrepancy was measured over a 1-h period. Finger diameter, drug therapy and presence of peripheral vascular disease were correlated with arm-finger blood pressure difference. Repeatability coefficients (twice the SD of the arm-finger difference) across device re-initialization were large, but similar to parallel repeated arm blood pressure determinations: 17.6 and 17.1 mmHg for systolic blood pressure (SBP) and 13.9 and 13.6 mmHg for diastolic blood pressure (DBP), respectively. Withdrawing and re-inserting the finger reduced repeatability substantially, with a 50% increase in repeatability coefficient. A trend towards a progressive 9-mmHg increase was observed in overestimation of SBP over the 1-h period. Mean +/- SD pooled arm-finger blood pressure differences were -10.8 +/- 14.6 mmHg for SBP and 4.5 +/- 9.4 mmHg for DBP. Blood pressure measured in different fingers was similar on average, with repeatability no poorer than for re-insertion of the same finger. The presence of peripheral vascular disease in 15 subjects correlated with a smaller arm-finger difference for DBP. Variations in positioning of the finger within the cuff influences blood pressure measurement during volume-clamp plethysmography, reducing its precision. Finger SBP exceeds brachial auscultatory readings and has similar precision.

IntroductionThe intra-arterial catheter (invasive method) is the clinical and reference standard for continuous blood pressure (BP) monitoring. Continuous noninvasive blood pressure (CNBP) monitoring methods, for example applanation tonometry, volume clamp, and cuffless BP monitoring devices are gaining popularity. This review clarified the evidence on the accuracy of CNBP monitoring devices, which could potentially replace invasive monitoring in clinical care.MethodsA systematic search was carried out to look for systematic reviews with the following elements:Population: patients needing continuous BP monitoring;Intervention: CNBP monitoring devices;Comparator: intra-arterial BP monitoring; andOutcomes: accuracy of BP monitoring.The databases searched included PubMed (MEDLINE), Epistemonikos, and the Cochrane Database of Systematic Reviews. Two reviewers independently reviewed the search results and shortlisted relevant articles for retrieval of full texts. Included reviews were critically appraised with the AMSTAR 2 instrument and the findings were summarized in a narrative synthesis.ResultsThree systematic reviews with meta-analyses of fairly good quality were included. The included primary studies were conducted in perioperative or critical care settings. Subgroup analyses by CNBP monitoring method were included in each meta-analysis. The findings from the systematic reviews were consistent. On average, CNBP devices consistently measured lower systolic BPs than the invasive method (mean difference <0) and measured higher diastolic BPs and mean arterial pressures than the invasive method (mean difference >0), with wide ranging 95 percent limits of agreement. It was evident from subgroup analyses that the measurements obtained from different CNBP methods varied significantly.ConclusionsThis rapid review found the accuracy of CNBP monitoring devices to be suboptimal in comparison with invasive monitoring. CNBP devices should not be routinely used but may be considered when there is difficult arterial access, or in patients who do not require arterial puncture for other purposes.

Accurate and convenient monitoring of blood pressure (BP) is challenging and relies on cuff-based devices or in the postoperative/intensive care settings, on invasive measurements. The aim of this study was to prospectively evaluate the accuracy of BP measurements obtained from a novel, commercially available cuffless, non-invasive photoplethysmography (PPG)-based chest patch monitor in patients after cardiac surgery. This single center prospective preliminary validation study enrolled adults who underwent cardiac surgery. Data generated by the PPG-based device was compared to those of a standard invasive arterial pressure (IAP). Bland-Altman plots and Pearson's correlations were used to assess the accuracy of the PPG-based device. Stability and BP changes were not formally evaluated. Ninety-six patients consented for the study. Mean age was 63.2 ± 12.2 years (range 24-84), and 32 (33%) were women. Average monitoring was 25.6 ± 17.2 h. In total, we evaluated 78,659 readings for systolic BP (SBP), 78,818 for diastolic BP (DBP), and 92,544 for heart rate (HR). The correlation coefficients were r = 0.959, 0.973, 0.966, and 0.962 for SBP, DBP, mean arterial pressure (MAP), and (HR), respectively. The bias ± SD was 0.1 ± 4.8 mmHg for SBP; 0.4 ± 2.1 mmHg for DBP; 0.26 ± 2.6 mmHg for MAP, and 0.15 ± 3.6 beats per minutes for HR. 95% of SBP, and 99.9% of DBP measurements were within 10 mmHg of the reference measurement. In conclusion, the tested cuffless device offers acceptable accuracy and is a promising novel noninvasive tool for continuous BP monitoring. Further studies are needed to validate these findings according to the most updated validation protocols for pulseless devices.ClinicalTrials.gov ID NCT04071015. Demonstrating (A) patient's flow chart; (B) Photoplethysmography sensor principals of action; (C) Study design, parallel measurement of IAP and cuffless device for patients in ICCU; and (D) the main results of the study.Blood pressure (BP), Systolic blood pressure (SBP), Diastolic blood pressure (DBP), Mean arterial pressure (MAP), and Heart rate (HR).

Objective: Continuous cuffless blood pressure (BP) monitoring research has emerged as blood pressure is one of the dynamic parameters which reflects cardiac arrhythmias and rheological disorders without the drawbacks of current techniques. All the existing measurement techniques are cuff-based with drawbacks such as being discontinuous in nature, being uncomfortable for the patient, etc. Therefore, the goal is to develop an algorithm to estimate BP accurately using the pulse transit time (PTT), photoplethysmogram intensity ratio (PIR) and Womersley number (α) in a noninvasive way. Approach: The PTT technique holds the promise of real-time, cuffless, continuous BP monitoring in clinical settings. However, the non-Newtonian fluid nature of blood is considered insignificant in the conventional PTT model-based BP modeling. In the proposed work, α, representing viscous effects, along with PTT and PIR, is included in the algorithm since it is also one of the parameters that affects the BP. Main results: The proposed algorithm is evaluated with 42 healthy and 39 diseased subjects and compared with the other conventional techniques to evaluate the importance of viscous effects in BP, using performance metrics like mean ± standard deviation, and 95% confidence interval for bias and limits of agreement; a better estimate is achieved with the proposed algorithm. Significance: The proposed algorithm reflects the influences of vasomotor tone and baroflex as well, which makes it suitable for the assessment of BP in both healthy and diseased subjects, with improved accuracy than the current measurement techniques.

Because it has been suggested that food rich in γ-aminobutyric acid (GABA) or angiotensin-converting enzyme inhibitor (ACEI) peptides have beneficial effects on blood pressure (BP) and other cardiovascular risk factors, we tested the effects of low-sodium bread, but rich in potassium, GABA, and ACEI peptides on 24-hour BP, glucose metabolism, and endothelial function.A randomized, double-blind, crossover trial was conducted in 30 patients with pre or mild-to-moderate hypertension, comparing three 4-week nutritional interventions separated by 2-week washout periods. Patients were randomly assigned to consume 120 g/day of 1 of the 3 types of bread for each nutritional intervention: conventional wheat bread (CB), low-sodium wheat bread enriched in potassium (LSB), and low-sodium wheat bread rich in potassium, GABA, and ACEI peptides (LSB + G). For each period, 24-hour BP measurements, in vivo endothelial function, and biochemical samples were obtained.After LSB + G consumption, 24-hour ambulatory BP underwent a nonsignificant greater reduction than after the consumption of CB and LSB (0.26 mm Hg in systolic BP and -0.63 mm Hg in diastolic BP for CB; -0.71 mm Hg in systolic BP and -1.08 mm Hg in diastolic BP for LSB; and -0.75 mm Hg in systolic BP and -2.12 mm Hg in diastolic BP for LSB + G, respectively). Diastolic BP at rest decreased significantly during the LSB + G intervention, although there were no significant differences in changes between interventions. There were no significant differences between interventions in terms of changes in in vivo endothelial function, glucose metabolism, and peripheral inflammatory parameters.Compared with the consumption of CB or LSB, no greater beneficial effects on 24-hour BP, endothelial function, or glucose metabolism were demonstrated after the consumption of LSB + G in a population with pre or mild-to-moderate hypertension. Further studies are warranted to clarify the effect of GABA on BP, preferably using a specific design for noninferiority trials and ambulatory BP monitoring as a measure of BP.This study was registered at Current Controlled Trials as ISRCTN31436822.

Continuous blood pressure monitoring is of great significance for the prevention and early diagnosis of cardiovascular diseases. However, the existing continuous blood pressure monitoring methods, especially the sleeveless blood pressure monitoring methods, are complex and computationally heavy. In this paper, we propose a method, using plethysmography (PPG) signals alone, to estimate continuous blood pressure by extracting multiple PPG features related to intravascular blood flow characteristics. The performance of our method was evaluated using ten minutes synchronized PPG signals and continuous blood pressure from 21 healthy volunteers and 19 patients with hypertension and diabetes. The test results have shown that the absolute mean errors and standard deviation errors between the estimated and referenced blood pressure are 3.22±0.66 mmHg for systolic blood pressure and 2.11±1.11 mmHg for diastolic blood pressure, which meet AAMI (the association for the advancement of medical instrumentation) error acceptance.

Continuous and non-invasive monitoring of blood pressure (BP) is important to prevent hypertensive patients from stroke and heart attack. However, most of the prevalent BP devices can provide solely intermittent measurements. In this study, a novel parameter RS <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2 </sub> , defined as the time interval from the R wave of electrocardiographic (ECG) signal to the peak of second heart sound of phonocardiographic (PCG) signal, was used for BP estimation. Experiment was conducted on 66 normal and 19 abnormal subjects during resting condition. The results of this study showed that estimation error of systolic BP (SBP) and diastolic BP (DBP) was 2.0plusmn7.5 mmHg and 0.9plusmn5.1 mmHg respectively. 94.7% of the difference between measured and estimated SBPs was within plusmn2 standard deviations in the Bland Altman plot. 93.4% of the difference between measured and estimated DBFs was within plusmn2 standard deviations in the Bland Altman plot. The preliminary results illustrated that the RS <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> is a promising parameter for continuous and non-invasive estimation of BP in homecare and mobile healthcare

Previous research shows serum 25-hydroxyvitamin D (25(OH)D) and parathyroid hormone (PTH) are each associated with blood pressure (BP), but it is unclear whether these associations are independent. Cross-sectional data from the US National Health and Nutrition Examination Surveys (NHANES) during 2003-2006. Analyses were restricted to 7,561 participants aged ≥20 years with measurements of 25(OH)D, PTH, BP, BP treatment, smoking, physical activity, serum calcium, and creatinine. Results were adjusted for these plus demographic variables. Serum 25(OH)D was more strongly associated (inversely) with systolic than diastolic BP. Adjusted mean (standard error) difference in BP for the lowest 25(OH)D quintile (≤13 ng/ml) was 3.5 (0.7) mm Hg for systolic BP and 1.8 (0.6) mm Hg for diastolic BP, compared with the highest quintile (≥30 ng/ml). In contrast, PTH was positively associated with both systolic and diastolic BP (P < 0.0001). Adjusted mean (standard error) difference in BP for the highest PTH quintile (≥59 ng/l) was 5.9 (0.8) mm Hg for systolic BP and 4.5 (0.5) mm Hg for diastolic BP, compared with the lowest quintile (≤27 ng/l). When both 25(OH)D and PTH were included in the same model, the associations of PTH with systolic and diastolic BP were unchanged. However, the associations between 25(OH)D and BP were attenuated, with mean (standard error) difference between the highest and lowest quintiles being 2.2 (0.6) mm Hg for systolic BP (P < 0.01) and 0.8 (0.6) mm Hg for diastolic BP. PTH may mediate most of the association between 25(OH)D and BP, which was not significant when also adjusting for body mass index.

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.