幾何セミナー


2019/1/28(Mon)

13:00--14:30 理学部 E404/406/408 大セミナー室

Hau-Tieng Wu

Duke University

Single channel blind source separation via differential geometry -- recover fetal ECG morphology

Abstract: Compared with the snapshot health information, high-frequency physiological signals provide health information from the other dimension, but we encounter new challenges when we handle these signals. In this talk, I propose a novel solution to the single-channel blind source separation challenge when we have only one physiological signal. In addition to introducing the algorithm, the theoretical foundation of the solution based on diffusion geometry and nonlinear-type time-frequency analysis will be provided, particularly when the signal has complicated statistical features, like time varying amplitude, frequency and non-sinusoidal pattern, and non-stationary noise. Its application to extracting fetal ECG morphology from the single channel maternal trans-abdominal ECG signal and a just finished clinical trial for maternal stress prediction will be discussed.