Cuffless blood pressure monitoring from a wristband with calibration-free algorithms for sensing location based on bio-impedance sensor array and autoencoder

Continuous monitoring of blood pressure (BP) is essential for the prediction and the prevention of cardiovascular diseases. Cuffless BP methods based on non-invasive sensors integrated into wearable devices can translate blood pulsatile activity into continuous BP data. However, local blood pulsatile sensors from wearable devices suffer from inaccurate pulsatile activity measurement based on superficial capillaries, large form-factor devices and BP variation with sensor location which degrade the accuracy of BP estimation and the device wearability. This study presents a cuffless BP monitoring method based on a novel bio-impedance (Bio-Z) sensor array built in a flexible wristband with small-form factor that provides a robust blood pulsatile sensing and BP estimation without calibration methods for the sensing location. We use a convolutional neural network (CNN) autoencoder that reconstructs an accurate estimate of the arterial pulse signal independent of sensing location from a group of six Bio-Z sensors within the sensor array. We rely on an Adaptive Boosting regression model which maps the features of the estimated arterial pulse signal to systolic and diastolic BP readings. BP was accurately estimated with average error and correlation coefficient of 0.5 +/- 5.0 mmHg and 0.80 for diastolic BP, and 0.2 +/- 6.5 mmHg and 0.79 for systolic BP, respectively.

Automated summary

Continuous monitoring of blood pressure is crucial for predicting and preventing cardiovascular diseases. Cuffless blood pressure methods using non-invasive sensors in wearable devices can provide continuous blood pressure data. However, current wearable sensors have issues with accuracy, large size, and variation in sensor location, leading to reduced accuracy of blood pressure estimation. This study presents a cuffless blood pressure monitoring method using a novel bio-impedance sensor array in a small-form factor wristband, providing robust blood pulsatile sensing and blood pressure estimation without calibration. The method utilizes a convolutional neural network autoencoder to accurately estimate arterial pulse signals independent of sensor location and an Adaptive Boosting regression model to map the features of the estimated pulse signals to systolic and diastolic blood pressure readings. The results show accurate estimation of blood pressure with small average errors and high correlation coefficients.