Features Extraction for Cuffless Blood Pressure Estimation by Autoencoder from Photoplethysmography

• Published in: Conference proceedings (IEEE Engineering in Medicine and Biology Society. Conf.) Vol. 2018; pp. 2857 - 2860
• Main Authors: Shimazaki, Shota, Bhuiyan, Shoaib, Kawanaka, Haruki, Oguri, Koji
• Format: Conference Proceeding Journal Article
• Language: English
• Published: United States IEEE 01.07.2018
• Subjects: Artificial neural networks; Biomedical monitoring; Blood pressure; Correlation coefficient; Estimation; Feature extraction; Neurons
• ISSN: 1557-170X; 1558-4615
• DOI: 10.1109/EMBC.2018.8512829

Several studies have been proposed to estimate blood pressure (BP) with cuffless devices using only a Photoplethysmograph (PPG) sensor on the basis of the physiological knowledge that the PPG changes depend on the state of the cardiovascular system. In these studies, machine learning algorithms were used to extract various features from the wave height and the elapsed time from the rising point of the pulse wave to feature points have been used to estimate the BP. However, the accuracy is still not adequate to be used as medical equipment because their features cannot express fully information of the pulse waveform which changes according to the BP. And, no other effective knowledge about the pulse waveform for estimating BP has been found yet. Therefore, in this study, we focus on the autoencoder which can extract complex features and can add new features of the pulse waveform for estimating the BP. By using autoencoder, we extracted 100 features from the coupling signal of the pulse wave and from its first-order differentiation and second-order differentiation. The result of examination with 1363 test subjects show that the correlation coefficients and the standard deviation of the difference between the measured BP and the estimated BP got improved from R = 0.67, SD = 13.97 without autoencoder to R = 0.78, SD = 11.86 with autoencoder.