Friday 21 March 2025
The quest for a non-invasive, cuffless blood pressure monitor has been an ongoing challenge in the field of medical technology. Researchers have been exploring various approaches, including photoplethysmography (PPG) signals, to estimate blood pressure without the need for cumbersome cuffs. A recent study published in a leading scientific journal presents a novel approach that combines physiological modeling with deep learning to achieve accurate blood pressure estimation.
The researchers designed a physiological model-based neural network (PMB-NN) framework that incorporates two key components: a neural network and a composed loss function. The neural network is trained on PPG signals collected from the finger, which are processed using various techniques to enhance signal quality and remove noise. The composed loss function combines data fitting terms with physiological constraint terms derived from established equations describing blood pressure dynamics.
The PMB-NN framework was tested on data collected from a single healthy participant performing three different activities: resting, cycling at 50 watts, and cycling at 100 watts. The results show that the model achieved promising performance in estimating systolic and diastolic blood pressures during rest and varying activity intensities. Moreover, the estimated total peripheral resistance (TPR) and arterial compliance (AC) exhibited physiologically consistent trends, with TPR decreasing and AC increasing as activity intensity increased.
The study’s findings are significant because they demonstrate the potential of PMB-NN for mobile health applications, particularly in high-risk populations requiring continuous blood pressure monitoring. The ability to estimate R and C alongside BP provides additional metrics for cardiovascular risk assessment, enabling a more comprehensive understanding of patient health.
While the results are encouraging, there are limitations to the study that need to be addressed. For instance, the model’s performance was affected by higher errors in systolic pressure estimation compared to diastolic pressure. This discrepancy may be attributed to the complexity and variability of systolic pressure due to transient cardiovascular dynamics during activity. Additionally, the estimated compliance parameter value (C) further limits the model’s reliability.
To overcome these limitations, future studies should focus on incorporating additional signal features, such as waveform morphology, which may enhance the model’s sensitivity to systolic fluctuations. Expanding the dataset by including data from more participants will also help validate the PMB-NN framework and pave the way for its potential application in mobile health monitoring systems.
The development of a non-invasive, cuffless blood pressure monitor has significant implications for patients with hypertension or other cardiovascular conditions.
Cite this article: “Advances in Cuffless Blood Pressure Monitoring Using Physiological Modeling and Deep Learning”, The Science Archive, 2025.
Blood Pressure Monitoring, Non-Invasive, Cuffless, Photoplethysmography, Physiological Modeling, Deep Learning, Neural Network, Ppg Signals, Mobile Health, Cardiovascular Risk Assessment.
