Wavelet scattering transform and long short-term memory network-based noninvasive blood pressure estimation from photoplethysmograph signals

Measuring blood pressure from photoplethysmograph (PPG) signals is gaining popularity as the PPG devices are inexpensive, convenient to use and much portable. The advent of wearable PPG devices, machine learning and signal processing has motivated in the development of cuffless blood pressure calculation from PPG signals captured from fingertip. The conventional pulse transit time-based method of measuring blood pressure from PPG is inconvenient as it requires electrocardiogram signals and PPG signals or PPG signals captured simultaneously from two different sites of the body. The proposed system uses the PPG signals alone to estimate blood pressure (BP). A signal analysis method called wavelet scattering transform is applied on the preprocessed PPG signals to extract features. Predictor model that estimates BP are derived by training the support vector regression model and long short term memory prediction model. The derived models are evaluated with testing dataset and the results are compared with ground truth values. The results show that the accuracy of the proposed method achieves grade B for the estimation of the diastolic blood pressure and grade C for the mean arterial pressure under the standard British Hypertension Society protocol. On comparing the results of the proposed system with the benchmark machine learning algorithms, it is observed that the proposed model outperforms others by a considerable margin. A comparative analysis with prior studies shows that the results obtained from proposed work are comparable with existing works in the literature.