Estimation of blood pressure waveform from facial video using a deep U-shaped network and the wavelet representation of imaging photoplethysmographic signals

Assessing heart rate and blood pressure estimation from image photoplethysmography using a digital blood pressure meter

“…The extraction of BP features from PPG signals using waveform morphology is a complex and intricate process, and a CWT was therefore used to represent the continuous time frequency of the PPG signal [ 21 , 25 , 32 ]. Different segment lengths were considered in a study investigating BP through the use of a CWT with PPG signals.…”

“…Liang et al previously reported the successful classification of hypertension by analyzing 5 s segments of cPPG using a CWT [ 30 ], whereas Wu et al found that CWTs performed optimally with segment lengths of 2.0 and 2.4 s for cPPG [ 21 ]. To represent iPPG signals, Bousefsaf et al demonstrated that a segment length of 2.56 s was adequate for predicting the CWT for BP [ 25 ]. In our study, we found that using a segment length of 6 s for iPPG led to better performance of the model compared to 3 s and 9 s. Although our finding contradicts the results reported by Bousefsaf et al, this emphasizes the effectiveness of CWT as a method for representing iPPG signals in predicting BP.…”

“…However, the results of their study may not be generalizable to a larger population or different experimental conditions. The same research group [ 25 ] has also reported a model for predicting the continuous wavelet transform (CWT) representation of a BP signal from the CWT of an iPPG signal, in which the BP time series was recovered based on an inverse CWT. The results showed that the use of a wavelet transform could represent the continuous time frequency of the iPPG signal, although only videos with clear iPPG signals were included in the dataset.…”

Remote Blood Pressure Estimation via the Spatiotemporal Mapping of Facial Videos

“…Secondly, PTT estimation based on rPPG technology requires multiple cameras to take videos of different human body parts, so the different light intensities on the different body parts will decrease the accuracy of PTT. On the other hand, some studies used machine learning methods to estimate BP values based on the features of the BVP signals extracted from videos [ 25 , 26 , 27 ]. Schrumpf et al [ 28 ] used contact PPG signals to train an artificial neural network and then used transfer learning to estimate BP based on the extracted BVP signals.…”

Non‐invasive measurement of blood pressure waveform based on spatiotemporal multi‐dimensional pulse waves

“…A webcam under ambient light was utilized to collect videos of face, based on which the pulse waves were extracted, and then, the features were extracted from the pulse wave so as to facilitate machine learning and estimate BP. Reference [12] obtained BP waveform based on IPPG signal using continuous wavelet transform (CWT) and U‐shaped deep neural network (U‐Net). The measurement processes of these methods are more convenient and less disturbed by weak motion, but non‐contact visual measurement is easily affected by factors such as ambient light and stray light.…”