Noninvasive Blood Pressure Prediction Based on Dual Encoder U-Net

Cardiovascular disease is one of the most serious causes of death, causing serious loss of life in the world every year. In this paper, we present a deep learning approach to estimate the continuous arterial blood pressure(ABP) waveform using the photoplethysmogram(PPG) signals. We propose a U-Net model structure of dual channel encoder, which extracts the semantic information roughly and finely through two encodings. In view of the strong periodicity and continuity of PPG signal, information distillation model with improved attention mechanism block is added to the U-Net network encoder, which can not only improve the ability of U-Net network to mine deep-seated semantic information, but also suppress unimportant channels and effectively reduce information redundancy. In addition, a multi-scale information fusion module is designed to fuse semantic information at different scales and improve the prediction accuracy of the model. Independent test on MIMIC-II dataset provides an MAE of 3.80mmHg and 1.81mmHg for SBP and DBP. The results met British Hypertension Society(BHS) Grade A and surpassed the studies from the current literature.