Fetal ECG Extraction Using Tensor Decomposition

Please enable javascript in your browser.

Akbari, Hassan | 2015

2360 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 48156 (05)
University: Sharif University of Technology
Department: Electrical Engineering
Advisor(s): Shamsollahi, Mohammad Bagher
Abstract:
In this work, we evaluate differernt tensor decomposition methods in application of fECG extraction from abdominal ECG recordings. After selecting proper tensor decomposition tool (Tucker decomposition) we propose a linear source separation algorithm based on a measure of quasi-periodicity. The quasi-periodicity is attained through the use of a constraint on a matrix factorization problem. In practice, we form a three dimensional ”tensor” by stacking the observation matrix and rough estimates obtained by both linear and non-linear subspace reconstruction methods. The method is applied to a database of electrocardiography (ECG) recordings, where rough subspace estimates of maternal and fetal ECG are used to enhance the estimation of the fetal ECG components. Validation is carried out on a simulated dataset with varying signal to interference plus noise ratios, where results show us that the proposed method successfully outperforms existing methods in literature such as ICA, CA, MBSS, Deflation, ICA+EKS, SVD and Kalman filter
Keywords:
Fetal Electrocardiagram (ECG)Signals ; Blind Sources Separation (BSS) ; Tensor Decomposition ; Tucker Decomposition ; FECG Extraction