No abstract

Cardiovascular disease (CVD) is one of the world's most serious diseases threatening human health. Among them, arterial blood pressure (ABP) waveform has a great relationship with cardiovascular diseases. It contains a lot of cardiovascular information, which plays an important role in diagnosing and preventing cardiovascular diseases. This paper proposes a deep learning model of ABP-MultiNet3+, which can convert photoplethysmogram (PPG) signals into ABP waveforms containing cardiovascular physiological information. The PPG signal is obtained by monitoring the human body with sensors, and its working principle ensures non-invasiveness and universality. To ensure the quality of the predicted ABP waveform, this paper carefully designs the network structure, input signal, loss function, and structural parameters. A fully convolutional neural network (CNN) MultiResUNet3+ is used as the core architecture of ABP-MultiNet3+. In addition to performing Kalman filtering on the original PPG signal, its first-order derivative and second-order derivative signals are used as ABP-MultiNet3+ enter. The model’s loss function uses a combination of mean absolute error (MAE) and means square error (MSE) loss to ensure that the predicted ABP waveform matches the reference waveform. The proposed ABP-MultiNet3+ model was tested in a subject-dependent manner on the public MIMIC II database, and the MAE of the predicted waveform from the reference waveform was 1.88 mmHg in the subject experiment, indicating a small error in the performance of the model better. In the method of this paper, the MAP and DBP reached the A level in the AAMI and BHS standards, and the Bland-Altman analysis and regression analysis confirmed the statistical significance of the experimental results.