Deep Learning for Blood Pressure Estimation: an Approach using Local Measure of Arterial Dual Diameter Waveforms

In this work, we present a novel approach for ubiquitous blood pressure (BP) measurement that involves a deep learning technique based on the extraction of the inherent features that are indicative of arterial pressure–diameter and pulse transit relationships. The proposed artificial neural network (ANN) architecture is the first to use the combined features of local arterial dimensions and blood pulse propagation characteristics for continuous, cuffless BP estimation. A dual-channel A-mode ultrasound system and a probe with a pair of single-element ultrasound transducers were developed for simultaneous measurement of luminal diameter waveforms from small arterial segments. In our present system, the probe design was optimized to capture local vessel dynamics from the common carotid artery. The reference continuous BP corresponding to individual cardiac cycles was acquired from the same arterial segment using a tonometer synchronized with the diameter measurement system. The data required to train and validate the developed ANN-based BP estimation model was recorded by conducting an in-vivo study on 20 young subjects. Thirty-seven unique features derived from the dual diameter waveform were extracted for beat-by-beat measurement of BP parameters from the carotid site, and hence to construct the carotid pressure waveform. Experimental results showed that the proposed approach exhibited an acceptable accuracy, with a root-mean-square-error of 4 mmHg and 6 mmHg for DBP and SBP respectively. In conclusion, the proposed approach provides a potentially novel cuffless BP technique for continuous measurement of arterial pressure waveform.