PPG2BP: An End-to-End Cuffless Blood Pressure Estimation Model Using Single Channel Photoplethysmography

Cardiovascular disease (CVD) is one of the leading causes of death globally, and regular blood pressure (BP) monitoring is essential to provide one of the most critical indicators of CVD prevention. Although current cuffless continuous BP measurement methods can achieve high accuracy, most require manual feature extraction influenced by prior knowledge and waveform quality. To address the above problems, this work proposes an end-to-end deep learning model, PPG2BP, which introduces the U-Net for image segmentation into one-dimensional BP estimation. PPG2BP utilizes a convolutional neural network to extract features adaptively and can perform BP estimation using only photoplethysmography (PPG) signals. In addition, PPG2BP includes an attention mechanism and a long short-term memory neural network in the decoder part of the U-Net. According to the British Hypertension Society standard, both diastolic blood pressure (DBP) and systolic blood pressure (SBP) estimation of PPG2BP reaches Grade A. PPG2BP also meets the standard of the Association for the Advancement of Medical Instrumentation in DBP estimation. The mean absolute error of SBP and DBP estimation are 4.62 mmHg and 2.74 mmHg, respectively.