|
Cuffless and continuous estimation of blood pressure can be achieved by using photoplethysmography (PPG) signals and neural networks. This approach offers an improvement over the inconvenience of operating commonly used cuffed electronic sphygmomanometers which are unable to provide continuous readings. The research aims to identify more suitable network models for the task of blood pressure estimation using PPG signals. Four network models for time-series signals were selected for this study, including recurrent neural network, long and short term memory (LSTM), and gated recurrent unit, which have previously been used for blood pressure estimation with PPG signals in existing studies, as well as a newer network structure, a temporal convolutional network (TCN). PPG signals and blood pressure data obtained from the intensive care medical information set were used as a sample set for evaluating the network's performance. The 61 features with the highest correlation from the PPG waveform feature set were selected as input feature vectors. The findings indicate that LSTM and TCN exhibit high accuracy and can serve as the foundational model for measuring blood pressure. TCN obtains equivalent precision to LSTM, despite a substantially smaller network size. Additionally, the former technique has a greater probability of attaining ongoing blood pressure readings without the need for cuffs.
|
