Fig. 1: Overview of IMU-Track.
A Overview of how a non-zero hydrostatic pressure Ph contributes to transmural pressure Pt (i.e., by adding to the arterial pressure Pa, the quantity of interest). For example, for a BP sensor placed on an arm, lowering the arm increases Pt and PWV, and decreases the PTT measured compared to the case of a BP sensor at heart level. PWV and PTT relationships to arm levels below heart level, near heart level, and above heart level are illustrated. B Illustration of a commercial method for correcting for Ph, based on an initial calibration procedure where the reference sensors (at heart and fingers) are placed at the same vertical level, followed by continued hydrostatic pressure correction using a fluid-filled tube relative pressure sensor that connects heart level to the finger. C Overview of approach for Ph tracking. The approach uses wrist-based inertial sensors and a deep learning model to infer parameterized arm orientation, which is then used to calculate Ph to correct errors that result from height differences between BP sensors using an analytical biomechanics wave model. D Diagram illustrating changes in Ph along the arm relative to the heart. The arrows represent the pulse wave traveling down the arm, with arrow length corresponding to PWV magnitude. E PTT calculation from ECG and PPG waveforms. F Block diagram of the deep learning-assisted model and BP prediction pipeline. Arm pose is estimated from a wrist-based IMU and a deep learning model based on measurements from IMU and a parametrized arm-pose coordinate system; this arm pose information is used to calculate hydrostatic pressure (Ph). PTT is measured using ECG and PPG. A prediction of BP is made using an analytical pressure wave propagation model with inputs of PTT and Ph following person-specific calibration to calculate fitting coefficients. Pink denotes IMU for estimating Ph, and blue denotes devices for measuring PTT. G The devices used in this study were one lead ECG, finger PPG, and wrist-mounted IMU.
