A method for blood pressure hydrostatic pressure correction using wearable inertial sensors and deep learning.

Abstract

Cuffless noninvasive blood pressure (BP) measurement could enable early unobtrusive detection of abnormal BP patterns, but when the sensor is placed on a location away from heart level (such as the arm), its accuracy is compromised by variations in the position of the sensor relative to heart level; such positional variations produce hydrostatic pressure changes that can cause swings in tens of mmHg in the measured BP if uncorrected. A standard method to correct for changes in hydrostatic pressure makes use of a bulky fluid-filled tube connecting heart level to the sensor. Here, we present an alternative method to correct for variations in hydrostatic pressure using unobtrusive wearable inertial sensors. This method, called IMU-Track, analyzes motion information with a deep learning model; for sensors placed on the arm, IMU-Track calculates parameterized arm-pose coordinates, which are then used to correct the measured BP. We demonstrated IMU-Track for BP measurements derived from pulse transit time, acquired using electrocardiography and finger photoplethysmography, with validation data collected across 20 participants. Across these participants, for the hand heights of 25 cm below or above the heart, mean absolute errors were reduced for systolic BP from 13.5 ± 1.1 and 9.6 ± 1.1 to 5.9 ± 0.7 and 5.9 ± 0.5 mmHg, respectively, and were reduced for diastolic BP from 15.0 ± 1.0 and 11.5 ± 1.5 to 6.8 ± 0.5 and 7.8 ± 0.8, respectively. On a commercial smartphone, the arm-tracking inference time was ~134 ms, sufficiently fast for real-time hydrostatic pressure correction. This method for correcting hydrostatic pressure may enable accurate passive cuffless BP monitors placed at positions away from heart level that accommodate everyday movements.