Deep learning approaches for continuous blood pressure estimation from photoplethysmography signal

Monitoring continuous Blood Pressure (BP) signals is essential as BP can vary rapidly. However, current Photoplethysmography (PPG)-based methods for estimating BP need to be more accurate and provide predictions for Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP). Full cycle of PPG waveform is considered to estimate SBP and DBP values. This study recommends deep learning techniques, including Temporal Convolutional Network (TCN), Long-Short Term Memory (LSTM), TCN-LSTM, and Autoencoder-LSTM, to estimate SBP and DBP. According to the outcomes, the proposed framework estimates continuous BP precisely utilising PPG signals. Specifically, the Autoencoder-LSTM algorithm achieved a Mean Average Error (MAE) of 1.05 and 0.92 for SBP and DBP and a Standard Deviation (SD) of 1.89 and 1.05 for SBP and DBP, respectively, indicating that the model is suitable for estimating these values from PPG signals. The Autoencoder-LSTM approach produced a Mean Average Error (MAE) of 1.05 and 0.92 for SBP and DBP, respectively, as well as a Standard Deviation (SD) of 1.89 and 1.05, demonstrating that the model can estimate these values using PPG signals. This paper evaluates an algorithm that estimates BP continuously using the PPG signal. Autoencoder-LSTM is suitable for estimating continuous BP values since MAE and SD values are low for SBP and DBP.