Wearable devices and advanced signal processing techniques, based on deep learning methods, are both disclosing an unprecedented interest in the non-invasive (cuff-less) and continuous estimation of the systemic and pulmonary blood pressure (BP). For systemic BP, deep neural networks may automatically extract fiducial time points from raw physiological signals to compute intermediate measures like heart rate, heart sound interval, heart sound ratio, pulse transit time (PTT), and systolic end-phase, to cite the most relevant ones. These quantities are then related to the arterial BP by means of regression models, which must be calibrated on a subject- or group-specific base.
In this speech, we revise integrated wearable sensors, like chest patches and smartwatches, PTT-BP models, like the Moens-Korteweg equation, and state-of-the-art methods for event detection, including deep networks applied for automatic extraction of fiducial time points and clinical events.
In addition to systemic BP, wearable devices can also be used to provide a non-invasive, continuous, and timely assessment of the pulmonary artery pressure (PAP), thus representing a simple and cost-efficient alternative to invasive cardiac catheterization or echo-Doppler analysis. Various research efforts in this sense have originated from the observation that variations of the PAP can be associated with changes in the waveforms of the phonocardiogram (PCG) and the seismocardiogram (SCG).
In this speech, we will revise different machine learning techniques that have been used to estimate the PAP from PCG/SCG recordings. A special focus will be devoted to the interplay between machine learning algorithms and signal processing approaches, in particular, source separation algorithms able to automatically separate vibration components generated by the aortic and pulmonary valves.