Preliminary Long ECG Data Abnormal Segment Screening And Abdominal ECG Data Enhancement

Holter continuously records electrocardiogram(ECG)for at least 24 hours.So computer aid is required to perform automatic abnormal screening on the recorded electrocardiogram.A reasonable strategy is: first screen out the abnormal segments in the recorded long data without any losing,that is,preliminary screening,and then further analyze the abnormal segments.Fetal ECG monitoring needs to record the abdominal ECG of a pregnant woman and separate it.The first key step involved in the separation is to enhance the maternal ECG component in the abdominal ECG to ensure correct detection of maternal R waves and accurate estimation of maternal ECG component.This paper focuses on the above two issues,long ECG data abnormal segment screening and abdominal ECG data enhancement.The main research work is as follows:(1)For long-term sequences in Holter,a preliminary screening method for abnormal segments based on LSTM is proposed.This method does not require prior knowledge of abnormal signals or R wave detection.This method utilizes the LSTM model to learn the pseudo-periodic pattern of normal ECG signals,and uses the kernel density estimation method to calculate the probability density of prediction errors of normal ECG signals,and then formulates a confidence threshold that can distinguish positive and abnormal signals.In practice,when the abnormal ECG is happened,the model trained with normal data will produce a large prediction error,which exceeds the confidence threshold.Experiments show that under the condition of ensuring that no abnormal signals are missed,normal segments which are average accounted for 53.89%of the total number of normal segments,can be effectively excluded.(2)A preliminary screening method for abnormal segments based on sparse representation is proposed.This method also does not require a priori knowledge of abnormal signals.The K-SVD algorithm is used to train a dictionary model of normal signals,and according to the different performance of the reconstruction errors of normal signals and abnormal signals after the sparse decomposition in the same dictionary,we can distinguish between normal and abnormal segments.On condition that no abnormal signals are missed,experiments show that the proportion of normal segments excluded by the method is 60.15%.(3)For the single-channel abdominal ECG signals,an enhancement and detection method of the maternal component based on sparse representation is proposed.The KSVD algorithm is utilized to learn the maternal dictionary components.Combined with the fixed fetal Gaussian dictionary scales,they construct an adaptive complete dictionary.After sparse decomposition,enhanced maternal component is obtained,and finally R wave positions of the maternal component is determined according to the selection rules.Experiments show that the method has better robustness and accuracy than the wavelet enhancement and the fixed dictionary-based enhancement method.