[3Xin4] Poster session 1

[3Xin4-62] Blood Pressure Measurement by Cuff Pressure Pulse Wave:Feature Modeling Cuff Pressure Pulse Waveform Inspired by Audio Source Separation

〇Shihori Uchida¹, Yukiko Yanagawa¹, Yuki Sasaki¹, Hirotaka Wada¹, Masayuki Koizumi¹, Hiroshi Nakajima¹ (1.OMRON Corporation)

Keywords:Deep Learning, Blood Pressure Estimation, Time Series Data, Sound Source Separation, Oscillometric

The purpose of the study is to improve the conventional model of blood pressure(BP) measurement. The oscillometric method is most widely used to measure BP by employing a cuff pressure sensor. However, the method is affected by fluctuations in waveform shapes due to body movements, etc. Previous studies have not addressed the variabilities of waveform shapes by an appropriate model structure with the waveform characteristics. Inspired by the similarity between the cuff pressure pulse waveforms and the voice ones, we developed a novel BP measurement method with robustness against variabilities in waveform shapes. The model employs the WAVE-U-NET which has the natures of capturing both global and local features. The model is extended by considering the input channels, the size of the model, and the loss function. We conducted the experiments to investigate the performance of the developed using our own database and proved that it outperformed the oscillometric method.