Cuffless Blood Pressure Estimation from Six Wearable Sensor Modalities in Multi-Motion-State Scenarios

TL;DR

Hypertension poses a major health burden and cuffless BP monitoring remains challenging in real-life motion. The authors introduce a six-modal wearable framework combining ECG, multi-channel PPG, attachment pressure, sensor temperature, and inertial signals, processed by six modality-specific encoders, a contrastive cross-modal fusion module, and a Mixture-of-Experts regression head. On the Pulse Transit Time PPG Dataset, the approach achieves SBP MAE 3.60 mmHg and DBP MAE 3.01 mmHg, with British Hypertension Society Grade A and compliant AAMI numerical metrics (though limited by sample size), demonstrating improved robustness across running, walking, and sitting. These results highlight the value of richer multimodal sensing and adaptive fusion for practical cuffless BP monitoring in everyday wearables, while noting the need for larger, more diverse validation cohorts.

Abstract

Cardiovascular disease (CVD) is a leading cause of morbidity and mortality worldwide, and sustained hypertension is an often silent risk factor, making cuffless continuous blood pressure (BP) monitoring with wearable devices important for early screening and long-term management. Most existing cuffless BP estimation methods use only photoplethysmography (PPG) and electrocardiography (ECG) signals, alone or in combination. These models are typically developed under resting or quasi-static conditions and struggle to maintain robust accuracy in multi-motion-state scenarios. In this study, we propose a six-modal BP estimation framework that jointly leverages ECG, multi-channel PPG, attachment pressure, sensor temperature, and triaxial acceleration and angular velocity. Each modality is processed by a lightweight branch encoder, contrastive learning enforces cross-modal semantic alignment, and a mixture-of-experts (MoE) regression head adaptively maps the fused features to BP across motion states. Comprehensive experiments on the public Pulse Transit Time PPG Dataset, which includes running, walking, and sitting data from 22 subjects, show that the proposed method achieves mean absolute errors (MAE) of 3.60 mmHg for systolic BP (SBP) and 3.01 mmHg for diastolic BP (DBP). From a clinical perspective, it attains Grade A for SBP, DBP, and mean arterial pressure (MAP) according to the British Hypertension Society (BHS) protocol and meets the numerical criteria of the Association for the Advancement of Medical Instrumentation (AAMI) standard for mean error (ME) and standard deviation of error (SDE).

Figures (8)

• Figure 1: Preprocessing pipeline used to construct the six-modal multi-motion-state dataset from the Pulse Transit Time PPG recordings, including slicing, down-sampling, standardization and VMD–WT denoising, concatenation, labeling, and partitioning.
• Figure 2: Illustrative example of the denoising pipeline applied to ECG and PPG signals, comparing raw waveforms with the corresponding filtered and denoised signals.
• Figure 3: Overall architecture of the proposed six-modal model, including six modality-specific encoders, a contrastive-learning-based cross-modal semantic alignment module, and a Mixture-of-Experts (MoE) regression head for joint systolic and diastolic blood pressure estimation.
• Figure 4: Structure of the one-dimensional squeeze-and-excitation (SE) block used in the CACNN encoder to implement channel-wise attention for multi-channel sensor inputs.
• Figure 5: Regression plots comparing reference and predicted blood pressure values on the test set: (a) systolic blood pressure (SBP) and (b) diastolic blood pressure (DBP). The solid line denotes the identity line, and the dashed line shows the fitted regression.