ArterialNet: Reconstructing Arterial Blood Pressure Waveform with Wearable Pulsatile Signals, a Cohort-Aware Approach

TL;DR

ArterialNet is a novel wearable signal processing approach that accurately reconstructs arterial blood pressure waveforms, reducing individual variability and enabling remote health monitoring.

Contribution

It introduces a hybrid signal translation and personalized feature extraction framework with regularizations, improving ABP waveform reconstruction accuracy over existing methods.

Findings

Achieved RMSE of 5.41 mmHg on MIMIC-III dataset

Reduced subject variance by 58% compared to prior techniques

Demonstrated robustness in remote health scenarios

Abstract

Goal: Continuous arterial blood pressure (ABP) waveform is invasive but essential for hemodynamic monitoring. Current non-invasive techniques reconstruct ABP waveforms with pulsatile signals but derived inaccurate systolic and diastolic blood pressure (SBP/DBP) and were sensitive to individual variability. Methods: ArterialNet integrates generalized pulsatile-to-ABP signal translation and personalized feature extraction using hybrid loss functions and regularizations. Results: ArterialNet achieved a root mean square error (RMSE) of 5.41 -+ 1.35 mmHg on MIMIC-III, achieving 58% lower standard deviation than existing signal translation techniques. ArterialNet also reconstructed ABP with RMSE of 7.99 -+ 1.91 mmHg in remote health scenario. Conclusion: ArterialNet achieved superior performance in ABP reconstruction and SBP/DBP estimations with significantly reduced subject variance, demonstrating its potential in remote health settings. We also ablated ArterialNet's architecture to investigate contributions of each component and evaluated ArterialNet's translational impact and robustness by conducting a series of ablations on data quality and availability.