Research On Fetal Electrocardiogram Signal Extraction

The fetal electrocardiogram (FECG) provides important clinical information about the healthy condition of the fetus. By monitoring the FECG during pregnancy, physicians may get evidence concerning the ability of the fetus to adapt to anoxia, arrhythmia and increments in the intra-uterine pressure. The early diagnosis of any cardiac defects before delivery increases the effectiveness of the appropriate treatment. The small potential changes of fetal heart can be accurately distinguished via utilizing the FECG. Its performance is better than B-type ultrasonic inspection, the ultrasonic doppler fetal actograph and fetal heart monitor. The improving of the FECG extraction method and obtaining the clear FECG were important for the FECG clinical application of FECG. But the existing FECG extraction methods have various distances to be used clinically on a large scale. In this paper, the development of the FECG extraction methods was comprehensive presented and the concept of FECG extraction was analysed in detail. It was in-depth researched that the FECG extraction methods utilizing artificial neural networks (ANNs), support vector machines (SVMs), the non-stationary signal processing method, wavelet packet denosing and EMD denoising.The FECG extraction method using ANNs was studied. The concept of FECG extraction based on adaptive noise cancellation (ANC) was introduced in detail. The characteristic of the abdominal composite signal of pregnant woman was analyzed. The key problem of the FECG extraction methods was pointed out. After the shortcoming of ANC based on the adaptive filter algorithm was analysed, ANNs was introduced into the FECG extraction field and the aim and meaning of ANNs was explained. The working principle of radial basis function neural network (RBFNN) and generalized regression neural network (GRNN) was given. The FECG extraction methods based on RBFNN and GRNN was studied and the parameter selection of RBFNN and GRNN was discussed. The visual results from the same electrocardiogram (ECG) signals were used to compare with the power of the RBFNN technique, the GRNN technique, the Kalman filter algorithm and the normalized least means squares (NLMS) technique in FECG extraction. The visual results from the RBFNN technique and the GRNN technique revealed the disadvantage of the two methods in the FECG extraction field.The FECG extraction method using least squares support vector machines (LSSVMs) was studied. The working principle of LSSVMs was introduced and the FECG extraction method by LSSVMs was given. The performance of the FECG extraction method was evaluated and the signal-to-noise (SNR) of the extracted FECG was estimated using eigenvalue analysis and cross-correlation. The LSSVM used in the FECG extraction was optimized via the selection of kernel function, parameter optimization, time-derivations selection of the input signal and the selection of training data number. The visual results and SNR from the same ECG signals were used to compare with the performance of the LSSVM technique, the RBFNN technique, the GRNN technique, the Kalman filter algorithm and the NLMS technique in FECG extraction. Taking account of non-stationary of ECG signal, empirical mode decomposition (EMD) was introduced to preprocess of the maternal ECG (MECG). The FECG extraction method by EMD and LSSVMs was studied and the parameter selection was discussed. The visual results and SNR from the same ECG signals were used to compare with the power of the EMD and LSSVM technique, the RBFNN technique and the LSSVM technique in FECG extraction.The FECG denoising was studied. The characteristic of the noise in the FECG was given and the research status of FECG denoising was indicated. The FECG denoising using the wavelet packet transform was studied. The extracted FECG by the proposed methods in the paper was denoised utlizing the wavelet packet denoising. The concept of EMD denoising was indicated and the FECG denoising by EMD was studied, then the extracted FECG by the proposed methods in the paper was denoised using the EMD technique. Finally, the contrast experiments of FECG denoisng was performed between the EMD denoising and the wavelet packet denoising.