Knowledge Discovery (KDD) And Algorithm Designing For High-Quality Fetal ECG Extraction

Fetal monitoring is quite important for better bearing and rearing children,especially for some high-risk pregnant women.Compared with the current monitoring methods based on Doppler ultrasound,fetal monitoring based on abdominal electrocardiogram(aECG)has advantages as follows:absolutely non-invasive and suitable for long-time monitoring.So it gains lots of attention from all over the world in these years.However,it is still a challenge to stably and robustly extract high-quality fetal electrocardiogram(fECG)from aECG because of the individual differences,too weak fECG on body surface etc.The existing monitoring devices based on aECG cannot satisfy the clinical needs for now.Therefore,it is still a hot research field about how to extract high-quality fECG.Looking throughout the various fECG extraction algorithms,the author finds that all of the algorithms are designed based on prior knowledge.The performance of an extraction algorithm depends greatly on the rightness,precision and integrity of the used prior knowledge.The more precise prior knowledge is used,the better the performance.Some prior knowledges used in the present algorithms need to be confirmed,and some new prior knowledges need to be discovered for achieving a more robust and high-quality fetal ECG extraction algorithm.Therefore,the dissertation did a lot of related research work,and its research contents and highlights are shown as follows:(1)Investigate the prior knowledge involved in applying Independent Component Analysis(ICA)to fetal ECG extraction and propose a corresponding maternal ECG component identification algorithm.ICA is one of the most important and frequently used methods in fetal ECG extraction algorithms.The first dilemma in it is the automatical identification of the separated maternal ECG component or other components.For this task,the author studied the QRS amplitude,heart rate and morphology characteristics of separated maternal ECGs and then proposed an automatic identification algorithm for the separated maternal ECG by ICA,called PRH,based on these prior knowledges.The success rate is as high as 99%superior to the other existed algorithms.Experiment shows that the PRH algorithm also can be used to identify the maternal ECG component after applying Principal Component Analysis(PCA)to abdominal signals.The second problem in applying ICA to fetal ECG extraction is that it has negative effect on real-time performance of the monitors.As to this problem,through lots of experiments the author finds an important knowledge,that is,the optimal maternal ECG combination vector(mECV)and optimal fetal combination vector(fECV)decided by prior data segment are applicable to the current or later data.That means mECV and fECV have some time-invariant characteristics.The knowledge can be used to exclude the bad influence of ICA on real-time performance.(2)Research and ascertain the prior knowledge on the periodic characteristics of ECG and then propose an improved algorithm for maternal ECG estimation.In fetal ECG extraction,most of the algorithms take advantage of the periodic characteristics of an ECG signal.But different algorithms have different understandings about the periodic character,which may even conflict with each other.It is urgent to ascertain which understanding is correct.Through the experiments,the author finds that for normal sinus ECG signal the Heart Beat Span stays invariant and the diastolic period changes with the heart beat interval.According to this knowledge,we designed an improved maternal ECG estimation algorithm,called Partial RR interval Resampling comb filter(PRR).The algorithm can estimate the maternal ECG more accurately compared to other algorithms.(3)Research some other prior knowledge and propose a complete scheme for fetal ECG extraction.The knowledge includes:(a)whether the instantaneous fetal heart rate changes periodically or not,whether the amplitude of fetal QRS changes periodically or not,and if so,the ratio between the period of instantaneous fetal heart rate and the period of instantaneous maternal heart rate and the ratio between the period of fetal QRS amplitude and the period of maternal QRS amplitude have some regularity or not;(b)the distribution regularity of the standard derivation and range of instantaneous fetal heart rates for 10-second data,the distribution regularity of the average value,standard derivation and range of fetal QRS amplitudes for 10-second data;(c)the coincidence possibility of maternal QRS wave and fetal QRS wave.At last,according to all of the discovered or ascetained knowledge in this dissertation,the author proposed a new complete scheme for fetal ECG extraction.Compared it with several public algorithms,the proposed complete algorithm displays promising results.The above research contents are arranged in the second,third and fourth chapters of this dissertation respectively.The fifth chapter gives the summary and prospects.