Long Short Term Memory (LSTM)-based Cuffless Continuous Blood Pressure Monitoring

Continuously monitoring blood pressure (BP) is crucial for individuals at high risk of cardiac diseases. However, existing BP measurement techniques lack the ability to provide non-invasive and continuous monitoring. To address this challenge, researchers have recently explored accelerometer-based systems for BP estimation. These systems rely on signal processing algorithms that often necessitate extensive feature engineering, making updates and calibration difficult. In this paper, we propose a novel device for non-invasive continuous BP monitoring. The device consists of a patch with two inertial measurement units (IMUs) attached to the skin in the user's neck region, specifically along the carotid artery. A control unit connected to the patch receives sensor data from these IMU units. It employs a machine learning (ML) model based on Long Short Term Memory (LSTM) to estimate BP using the sensor data. The model undergoes two general ML processes. The first ML process involves training the analysis model using a training set comprised of data from various individuals. Subsequently, the second machine learning process re-trains a portion of the analysis model using an individualized training set gathered from the specific user. This approach enhances the accuracy and personalization of the BP estimation, providing a promising solution for continuous monitoring of BP in high-risk individuals. The BP monitoring device's analysis model is also trained and tested on 11 volunteers. The BP monitoring device can measure an individual's BP every 5 seconds. Additionally, the best root mean squared error (RMSE) loss obtained with our model is less than 2.932 mmHg for systolic BP and 2.231 mmHg for diastolic BP.