Noninvasive blood pressure monitoring by PPG based on neural networks

In this study, a hybrid neural network multi-input multi-output blood pressure prediction model based on self-attention mechanism autoencoder (CAE-SA-MIMO-CNN) is expected to output two different predictive values (systolic and diastolic blood pressure) using PPG signal waves. photoplethysmography (PPG) was used for continuous non-invasive blood pressure monitoring. In this study, a CAE-SA-MIMO-CNN hybrid neural network model was constructed. PPG signals from the medical information set of our university, physiological data of the tester and discrete blood pressure values were used to simultaneously record signal data. PPG signal waves were preprocessed and input into the model to process signal data and static features. The final output is two different predicted values (systolic and diastolic blood pressure).When CAE-SA-MIMO-CNN neural network with window length 700 was used, the error between the predicted blood pressure value and the actual blood pressure value was the smallest, and the confidence of the homologous T test was above 95%, and the average blood pressure error was 4.76 mmHg and 9.065 mmHg, respectively. The predicted blood pressure value and the actual blood pressure value had good consistency and correlation. According to the standards of the Association for the Advancement of Medical Instrumentation (AAMI).