A mixed attention-gated U-Net for continuous cuffless blood pressure estimation

Blood pressure (BP) is an important vital sign of the human body. The traditional cuff measurement methods are mainly intermittent, which cannot meet the clinical practice well. Continuous measurements of BP are of great significance for monitoring vital signs of patients. In order to achieve BP estimation in a continuous, cuffless and non-invasive way, this paper proposes a BP estimation model based on mixed attention gating U-Net (MAGU), which can effectively improve the accuracy and efficiency of BP estimation. The photoplethysmography signal is fed into the improved U-Net to extract features. A mixed attention gating mechanism is added between up-sampling and down-sampling, as well as the residual blocks are added in down-sampling to prevent the vanishing gradient, so as to improve the feature extraction efficiency in the deep network. The performance of the MAGU is validated against those state-of-the-art results on the MIMIC-II public dataset. The mean absolute error and standard deviation of systolic blood pressure predicted by the proposed method are 3.49 mmHg and 4.13 mmHg respectively, and those of diastolic blood pressure (DBP) are 2.11 mmHg and 2.49 mmHg. The comparison shows that the proposed method outperforms those state-of-the-art methods.