Study On Microvolt T-wave Alternans Multi-channel Fusion Detection And Estimation

T-wave alternans(TWA)is a cardiac phenomenon,which has great significance for cognition and prediction of sudden cardiac death(SCD)in clinical practice.In general,the TWA with clinical detection value is at the microvolt level,which is also called microvolt TWA.It features nonuniform distribution among channels,short-time beat-to-beat high-dynamic changing and interference from complex noises,resulting in difficulties of detection and estimation TWA in routine electrocardiogram,which severely limits its clinical applications in SCD assessment and prediction.The existing clinical TWA analysis methods are usually conducted within a single channel,resulting in limited available information and a high sensitivity to non-Gaussian noise.And suffers from the conflict information between channels,nonuniform distribution among channels and beat-to-beat high-dynamic changing of TWA,the existing multi-channel TWA analysis methods have problems of low information utilization,low detection reliability and difficulties of estimation for short-time high-dynamic TWA.From the idea of multi-sensor information fusion and based on the homology of TWA among channels,this thesis studies microvolt TWA multi-channel fusion detection and estimation,which aims to improve the reliability of detection and estimation by integrated utilizing the information from different channels.Within the framework of the belief function theory,the microvolt TWA detection method based on conflict information redistribution is studied to deal with the problem of conflict information between channels,which is cased by low amplitude,complex noises and nonuniform distribution of TWA among channels.Within the framework of the tensor theory,the microvolt TWA estimation method based on canonical polyadic decomposition(CPD)and reconstruction is studied to deal with the difficulties in estimating short-time high-dynamic TWA.The main contents of this thesis are listed as follows.(1)Since microvolt TWA is a low-amplitude signal and interfered by complex noises,the decision information from different channels is highly in conflict with each other,which is the main reason the most clinical TWA detection methods are conducted within a single channel.In this thesis,from the idea of multi-sensor information fusion and within the framework of Dezert-Smarandache theory(DSmT),the electro-physiological and morphological features of TWA are studied and a novel multi-channel fusion detection method based on proportional conflict redistribution(PCR)is proposed to improve the detection reliability in a complex noisy environment.The experimental results show that the proposed multi-channel fusion detection method base on belief function theory has a higher sensitivity to microvolt TWA and a higher rejection capability to non-Gaussian noises like muscle artifact and electrode motion artifact.(2)Since microvolt TWA is non-uniformly distributed among channels in clinical practice,the information from different channels takes different degrees of importance,which makes it difficult for equally-weighted fusion method to avoid interferences from a few of channels with high false detection rate.In this thesis,within the framework of weighted multi-sensor information fusion and combined with the discounting method of the evidence,the relationship between background T-wave and TWA is utilized to dynamically evaluate the weight of each channel,and then a novel weighted multi-channel fusion detection method is proposed to improve the detection ability for clinical non-uniformly distributed TWA.The experimental results show that the proposed weighted multi-channel fusion detection method has higher reliability and stronger robustness to real ECG records from clinical patients,and a higher detection rate of patients with myocardial infarction.(3)Since microvolt TWA is highly dynamic between beats,it is difficult to estimate the beat-to-beat waveform of microvolt TWA.In this thesis,from the idea of multi-channel fusion estimation and within the framework of the tensor theory,the features of short-time high-dynamic TWA in the time,space and beatquency domain are combined to build a high-order tensor model,and then a novel multi-channel fusion estimation method based on CPD is proposed to improve the waveform estimation capability to short-time high-dynamic TWA.The experimental results show that the proposed multi-channel fusion estimation method based on CPD has a better dynamic tracking capability to short-time TWA.