ArtEARial: Arterial Pressure Waveform Reconstruction Using Earbud Audio

Noninvasive retrieval of the arterial pressure waveform (APW) enables a variety of cardiovascular monitoring tasks using pulse wave analysis (PWA). This paper demonstrates a novel way of reconstructing the APW using the audio recorded by an active noise-cancelling earbud. Our key observation is that low-frequency audio recorded in the ear canal is morphologically similar to the second derivative of the APW; however, differences in sensing site and modality lead to non-trivial differences in the waveforms. Our system, called ArtEARial, combines signal processing and deep learning to overcome these differences and accurately reconstruct the pulsatile APW. Using a dataset collected from 50 healthy adults, we compare ArtEARial's reconstructed APW against the signals produced by a continuous noninvasive blood pressure monitor. We demonstrate that the signal generated by ArtEARial yields better waveform fidelity, fiducial point localization, and PWA feature extraction compared to photoplethysmography (PPG) signals collected from the finger and earlobe. While ArtEARial is primarily envisioned for use while participants are seated in a quiet space, we also show that it is able to operate in scenarios with moderate background noise.