Measuring multi-site pulse transit time with an AI-enabled mmWave radar

TL;DR

The first AI-enabled mmWave system for contactless multi-site PTT measurement using a single radar using a single radar is presented, suggesting that the proposed system has the potential to provide a non-invasive measure of cardiovascular health across multiple regions of the body.

Abstract

Pulse Transit Time (PTT) is a measure of arterial stiffness and a physiological marker associated with cardiovascular function, with an inverse relationship to diastolic blood pressure (DBP). We present the first AI-enabled mmWave system for contactless multi-site PTT measurement using a single radar. By leveraging radar beamforming and deep learning algorithms our system simultaneously measures PTT and estimates diastolic blood pressure at multiple sites. The system was evaluated across three physiological pathways - heart-to-radial artery, heart-to-carotid artery, and mastoid area-to-radial artery -- achieving correlation coefficients of 0.73-0.89 compared to contact-based reference sensors for measuring PTT. Furthermore, the system demonstrated correlation coefficients of 0.90-0.92 for estimating DBP, and achieved a mean error of -1.00-0.62 mmHg and standard deviation of 4.97-5.70 mmHg, meeting the FDA's AAMI guidelines for non-invasive blood pressure monitors. These results suggest that our proposed system has the potential to provide a non-invasive measure of cardiovascular health across multiple regions of the body.

Figures (11)

• Figure 1: Overview of the mmWave radar system for multi-site pulse transit time (PTT) measurement.a, Rhythmic contractions of the heart generate pulse waves that propagate across the arterial system. (Image source: https://smart.servier.com/category/anatomy-and-the-human-body/cardiovascular-system) b, The mmWave radar detects minute surface displacements by measuring phase variations in reflected signals over time. c, The mmWave radar's beamforming algorithm targets four key physiological sites to measure PTT: the apex of the heart (green) and the mastoid area (yellow) as proximal reference points, and the radial artery (blue) and carotid artery (red) as distal arterial sites. The waveforms represent the phase signals from these sites, captured by the mmWave radar positioned at a fixed distance beneath the subject. d, Our system estimates pulse transit time across three pairs: $PTT_{SCG \rightarrow wrist}, PTT_{SCG \rightarrow neck}, PTT_{BCG \rightarrow wrist}$. This is achieved by first identifying key cardiac waveforms features: the aortic opening in the SCG signal, the J peak in the BCG signal, and the foot of the carotid and radial arterial pulse waveform. PTT is then computed as the time difference between these signal features at the proximal reference points and those at distal arterial sites.
• Figure 1: Placement of contact-based reference sensors at different anatomical sites. (Image source: https://www.template.net/design-templates/print/free-body-diagram/)
• Figure 2: Signal processing and deep neural network pipeline to estimate PTT from mmWave radar reflections along multiple physiological pathways.a, The system spatially decomposes the radar input channels into radar range and angular bins using FFT and beamforming. For each bin, phase and magnitude information are extracted. An adaptive bin ranking algorithm is applied to identify the bins with the strongest cardiac signal features, followed by a DNN model for PTT estimation. b, Beamformed radar signal power is visualized across azimuth, range, and time axes, regions associated with higher signal power are brighter. Key physiological areas of interest, including the heart, neck, and wrist, are marked with bounding boxes that correspond to pre-defined search areas for a cardiac signal. The radar phase and magnitude from the key physiological sites show periodic signals corresponding to the cardiac cycle, while signals from non-physiological regions lack recognizable cardiac patterns. c, Processing pipeline of the adaptive bin ranking algorithm. We leverage the pre-defined geometric constraints of the physiological sites, periodicity of the signal, and multi-step filtering to prioritize signal bins containing cardiac pulse features. d, DNN architecture for multi-site PTT estimation. Our DNN architecture for PTT estimation employs a multi-branch design that jointly processes waveforms from four physiological sites (heart, neck, wrist, head) using three key components: Spatial Pooling Blocks for dimensionality reduction, Convolutional Encoder-Decoders with bi-directional LSTM for modeling the temporal relationship between cardiac waveform features, and Cross-region Fusion Modules that fuses correlated cardiac waveform features across different physiological sites.
• Figure 2: Experiment protocol timeline. Ten participants took part in the preliminary protocol, while 27 took part in the full experimental protocol.
• Figure 3: Comparison of PTT estimates from the mmWave radar system against contact-based reference sensors.a, c, e Correlation plots of PTT estimates and b, d, f Cumulative Distribution Function (CDF) plots of PTT errors for $PTT_{SCG \rightarrow wrist}$, $PTT_{SCG \rightarrow neck}$, and $PTT_{BCG \rightarrow wrist}$. Data is shown for 13 participants (visualized in different colors) across 158 sessions in total.