Fetal ECG Extraction From Single-lead And Multi-lead Abdominal Signal

Fetal ECG is one of primary objective indicators reflecting fetal physiological activities,extraction of pure fetal ECG has important clinical significance on preventing some types of birthdefect. Through observation of fetal ECG, early diagnosis of fetal distress, congenital heart diseaseand fetal hypoxia, etc. is possible and effectively to reduce the incidence of various diseases andneonatal mortality and achieves the purpose of eugenics. Indirect method is often used by locatingelectrodes on maternal body surface to collect the fetal ECG. However, using the noninvasive wayto obtain fetal ECG indirectly is often subject to a lot of noise interference, including maternal ECGinterference whose amplitude is many times the fetal heart and the most of its frequency overlapswith the fetal ECG. In addition, the drift caused by mother’s breathing, EMG interference caused bycontractions, and50Hz power frequency interference and so on affects the fetal ECG extraction.Therefore, fetal ECG extraction has become a hot issue to many scholars.This paper mainly studies the fetal ECG extraction methods at the single-lead and multi-leadenvironment. Based on the analysis of commonly used fetal ECG extraction method, this paperproposes an independent component analysis in wavelet domain method used in multi-leadenvironment and a single-lead extraction method based on singular value decomposition andartificial neural network. Simulated data and real recordings are both used for the methods. Themain work and results are showed as follows：1. Reviews the history and current status of the development of the fetal ECG signal extractiontechnology, the characters of the abdominal signals are also introduced for laying the foundation forfetal ECG extraction.2. Proposes an extraction method used for multi-lead environment based on independentcomponent analysis in wavelet domain. Firstly transform the data into wavelet domain; and thenInformax algorithm is used to extract each independent component, the threshold de-noise methodis also used to preprocess the original data; Study the effects of different wavelet functions anddecomposition levels to the proposed method, optional choice of them are determined; Finally,simulated and clinical data are used to validate the performance of the algorithm. The results showthat the proposed algorithm separates the fetal ECG effectively under multi-lead environment.3. Proposes a method for extracting fetal ECG based on singular value decomposition andartificial neural network for single-lead system. Firstly, the singular value decomposition is used togenerate an estimation of maternal ECG signal component; and then artificial neural network isused to simulate the nonlinear conduction process of maternal ECG; and finally simulated data and clinical data are used to validate the performance of the proposed algorithm. The results show thatthe proposed algorithm separates fetal ECG effectively under single-lead environment.This paper proposes new methods for both multi-lead and single-lead fetal ECG extractionsystem, provides theoretical reference and technique support for fetal ECG extraction R&D.