Blood Pressure Estimation from Wavelet Scalogram of PPG signals using Convolutional Neural Networks

Abstract

Introduction In this work, calibration-free blood pressure estimation using wavelet scalograms of PPG signals using Convolutional Neural Network (CNN) has been proposed. The PPG signal, easily obtained from a subject, serves as a reliable indicator for predicting blood pressure (BP). Methods The proposed methodology involves employing Continuous Wavelet Transform (CWT) scalograms of the PPG signal as inputs for the CNN. Two distinct architectures for BP estimation are explored: one employing regression with a fully connected neural network and another utilizing CNN with Support Vector Regression (SVR). Results The results demonstrate superior BP estimation with the CNN-SVR architecture. With the CNN-SVR model, the Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP) are estimated with a Root Mean Square Error (RMSE) of 6.7 mmHg and 8.9 mmHg, respectively. Conclusion The proposed CNN-SVR model gives 52% better estimation error performance in SBP estimation compared to a machine learning model reported in a previous work.