부산대학교 도서관

Cuff-less blood pressure (BP) is estimated via the pulse wave velocity (PWV) between two sensors located on the superficial artery of the human body, such as the index fingers. However, the pulse transmission distance between these two sensors is considered as a constant for any individual, which can lead to incorrect BP estimation. In this research, we proposed a BP estimation system, which is based on the symmetrical photoplethysmography (PPG) signals captured by two sensors placed at a fixed distance. We designed a high integration, low cost, and wearable device on the wrist for PPG signal collection. The device integrates two photodetectors and a light source to achieve precise bi-channel PPG signal collection over short distances. To improve the robustness of the BP estimation, we use an attention-based convolutional neural network with bi-directional long short-term memory (CNN-biLSTM) architecture that combines morphological and computational features extracted from symmetric PPG signals to estimate diastolic BP (DBP) and systolic BP (SBP). We compared our system’s BP measurement with that of an electronic sphygmomanometer, indicating that the mean absolute error (MAE) and standard deviation (STD) of DBP and SBP are ${1}.{65}\pm {1}.{91}$ mmHg and ${2}.{16}\pm {2}.{39}$ mmHg, respectively, which outperforms the state-of-the-art methods. Our system performance complies with the requirements of the association for the advancement of medical instrumentation (AAMI) device standards and has achieved a Grade A rating from the British hypertension society (BHS).