Continuous Blood Pressure Measurement based on Optical Sensing and Modeling Analysis of Tonoarteriogram Imaging

Abstract

Cardiovascular disease (CVD) remains the leading cause of mortality worldwide, with hypertension as a major risk factor. According to the World Health Organization (WHO), an estimated 1.4 billion people globally suffer from hypertension, yet only 14% have it under control. This alarming gap underscores the urgent need for a paradigm shift from traditional reactive medical care to proactive healthcare, focusing on early prediction, prevention, diagnosis, and intervention. Effective control and management of hypertension is crucial for reducing the risk of CVD and other related diseases.

Continuous, cuffless, and wearable blood pressure (BP) measurement technologies can offer a transformative solution by enabling cost-effective and real-time hemodynamic monitoring during daily activities, thereby improving hypertension control and management, hence reducing CVD risk. However, current cuffless BP devices constrained by unimodal configurations, one-dimensional (1D) sensing, and single-site measurements are suffering from their limited ability to capture regional BP variations and estimate central BP accurately.

To overcome these limitations, this thesis proposes the development of novel multi-modal systems beyond 1D sensing for the design of wearable cuffless BP devices. These systems are capable of detecting and processing physiological signals from optical, electrical, and mechanical domains to image local BP for the very first time, estimate systolic BP (SBP) and diastolic BP (DBP) values, and reconstruct tonoarteriogram (TAG) waveforms, enabling advanced and reliable BP measurements.

The specific objectives of this thesis include: (1) to enhance 1D BP measurement with a compact multi-modal configuration; (2) to extend BP measurement from 1D to two-dimensional (2D) local BP imaging with an optical sensor array-based TAG system; (3) to mitigate contact force (CF) effects with a dual-modal TAG imaging system combining photoplethysmography (PPG) and pressure sensors; (4) to advance TAG imaging into a multi-modal system with a photo-electro-mechano sensing array; (5) to investigate the multi-sensor sties effect in a localized area on central BP estimation using a machine learning (ML)-driven model; and (6) to conduct studies on three-dimensional (3D) BP measurement with a custom multi-wavelength PPG (MWPPG) sensor array.

To achieve these objectives, we carried out the research activities as follows:
Firstly, to enhance 1D BP measurement through multi-modal configurations by integrating diverse physiological signals, we designed a compact, integrated hardware system incorporating PPG, electrocardiography (ECG), ultrasound, temperature, bioimpedance (BIM), pressure, and accelerometers (ACC) sensors. Subsequently, we developed and evaluated two dual-modal prototypes: BIM and MWPPG, as well as Pressure and ACC. Tested on 10 subjects, the BIM+MWPPG prototype achieved a mean PWV value of 5.95 m/s at the wrist, consistent with reported physiological ranges, while the Pressure+ACC prototype demonstrated TAG signal changes corresponding to hydrostatic variations. These results highlighted the potential of combining modalities for more accurate cuffless BP monitoring.

Secondly, to address the limitation of current 1D sensing-based BP measurement techniques, which are confined to providing SBP, DBP values, or TAG signals at a single site and thus fail to capture regional BP variations across different anatomical locations, we developed a wearable TAG imaging system based on a PPG sensor array (515 nm), extending BP measurement from 1D to 2D. The system simultaneously captures nine PPG signals from distinct wrist locations, enabling 2D visualization of local PTT and BP using PTT-based models for the first time. All BP estimations in this work were referenced against continuous BP (cBP) acquired from a BIOPAC system, calibrated using the Omron HEM-7132 sphygmomanometer. Testing on 22 subjects demonstrated the system's capability to map PPG, PTT, and BP distributions across multiple locations within a localized area. Notably, the calculated differences between the reference BP and estimated BP across locations showed that SBP ranged from -4.94 mmHg to 13.04 mmHg, while DBP ranged from -1.92 mmHg to 1.40 mmHg. By providing location-specific BP information, the TAG imaging system has the potential to enhance central BP estimation for improved cardiovascular health monitoring.

Thirdly, to mitigate the impact of contact force (CF) on PPG signal quality, we developed a dual-modal TAG imaging system by integrating a 2×2 pressure sensor array with the green PPG array at a wavelength of 515 nm, featuring detection sensors with dimensions of 36mm^2, positioned on the wrist area. Wrist extension experiments revealed inconsistencies in PPG amplitude (PPGA), and b/a ratios under varying CF conditions, with a correlation coefficient of 0.65069 observed between normalized PPG and flipped CF, underscoring the impact of posture on PPG measurements. These findings highlighted the need for mitigating the effects of CF on PPG signals for the platform’ design and measurement setting, thereby to improve potentially the accuracy of PPG-based BP monitoring.

Fourthly, to improve brachial BP estimation on 2D TAG imaging system through multi-modal signal integration, we developed a multi-modal photo-electro-mechano TAG imaging system that integrated a single-lead ECG into the 3×3 PPG (515 nm), and 2×2 pressure sensor arrays. This multi-modal sensing patch enabled the simultaneous acquisition of 15-channel physiological signals at the wrist. Test on 20 subjects, the system revealed significant local PTT and BP variations across positions, specifically, PTT variations ranged from 4.9 ms to 32.0 ms, while BP variations ranged from 0.8 mmHg to 13.1 mmHg for SBP and 0.3 mmHg to 2.0 mmHg for DBP. Additionally, the system demonstrated improved brachial BP estimation using a ML-driven multichannel PPG-based approach. Comparisons between single-channel and three-channel PPG signals showed that the three-channel approach reduced estimation errors: for SBP, Mean Absolute Error (MAE)±Standard Deviation (SD) decreased from 6.47±3.18 mmHg to 5.88±2.56 mmHg, and for DBP, from 4.09±1.89 mmHg to 3.79±1.74 mmHg. Beyond providing 2D visualization of local PTT and BP and their variations, the system demonstrated its potential to enhanced central BP estimation.

Fifthly, to investigate the impact of multi-sensor sites in a localized area on BP estimation, we first compared 9 different PPG features across channels, revealing notable site-specific variations, with amplitude-based features exhibiting greater sensitivity to location than temporal features. Building on these findings, we developed the 3M-BPNet, which utilizes multi-channel PPG signals detected from multiple sensor locations, processed through various channel combinations for BP prediction. The network incorporated signal trimming and Bayesian Optimization for channel weight allocation. The weighted features were input into a Random Forest Regression (RFR) model with personalized calibration for BP estimation. Validated on 121 subjects, the 3M-BPNet significantly improved accuracy over the standard RFR model, achieving MAEs of 4.84 mmHg for SBP and 3.28 mmHg for DBP. Notably, BP estimation accuracy using 3M-BPNet improved as the channel count increased from 1 to 5, then declined beyond 5. The optimal 5-channel combination (C5-C6-C7-C8-C9), located near the radial artery, achieved MAEs of 1.33 mmHg for SBP and 1.16 mmHg for DBP, representing relative improvements of 72.6% and 73.2%, respectively, compared to the single-channel setup (P < 0.001 for both). Moreover, our smart channel selection strategy achieved a 71.9% reduction in SBP MAE and a 51.3% reduction in DBP MAE compared to the best previously reported results in PPG-based BP estimation literature. This study demonstrated the critical role of sensor location in PPG-based BP estimation, suggesting that optimized sensor positioning could enhance accuracy.

Sixthly, to address the depth resolution of skin blood pulsation in BP assessment, MWPPG offers a more practical and versatile solution compared to the above proposed ultrasound sensor arrays. Its ability to provide layer-specific insights at lower cost and energy consumption, along with its compatibility with wearable devices, makes it a preferred choice for next-generation BP monitoring systems. Building on this, we developed a 3D TAG imaging system incorporating a 2×2 MWPPG sensor array. Each MWPPG sensor consists of one photodetector and four LEDs with emission wavelengths of 470 nm, 490 nm, 590 nm, and 940 nm, and a detection area of 24 mm². The system enabled the simultaneous acquisition of 16-channel MWPPG signals from four distinct wrist positions, and a wavelength-based light-tissue interaction model was developed to reconstruct TAG signals. Tested on 10 subjects, the system achieves 3D visualization of local and depth-specific PTT and BP variations by employing PTT-based BP models. Notably, analysis of the MAE for positional SBP and DBP reveals a variability of 6.79 mmHg (SD: 2.10 mmHg) for SBP and 3.59 mmHg (SD: 1.14 mmHg) for DBP, revealing the influence of position on estimation accuracy and the greater sensitivity of SBP to positional factors compared to DBP. This preliminary approach underscores the feasibility and necessity of multi-location and multi-depth measurements, introducing a promising new dimension to cardiovascular assessment.

In summary, this thesis work extended BP monitoring for the first time from the conventional 1D to 2D and preliminarily 3D methods, establishing a critical foundation for next-generation high-dimensional and high-resolution BP devices. These devices should enable continuous, cuffless hemodynamic multi-dimensional monitoring, advancing hypertension management and peripheral blood circulation studies. Looking forward, ongoing refinement of optical sensing technologies and methodologies, including advancements in PPG, MWPPG, and light-tissue interaction modeling, will drive the development of precise and reliable wearable systems. Future research will focus on a "5M" framework to enhance BP monitoring precision and versatility by integrating multi-location signals for spatial BP variations, multi-wavelength insights for depth-specific vascular assessment, multi-parameter metrics like Heart Rate (HR), Heart Rate Variability (HRV), PPG Intensity Ratio (PIR), and BP, multi-velocity visualization of PWV for micro-circulation studies, and multi-dimensional data for spatial, temporal, and depth-resolved hemodynamic monitoring. These efforts promise to redefine BP measurement, paving the way for a new era of proactive, personalized healthcare and transformative advancements in reducing the global burden of CVD.

Date of Award	8 Jul 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Xinge YU (Supervisor) & Yuanting Zhang (External Co-Supervisor)