GB2555574A

GB2555574A - Method for blind extraction of fetal electrocardio based on time-frequency conversion

Info

Publication number: GB2555574A
Application number: GB1618027.5A
Authority: GB
Inventors: Xie Kan; Zhou Guoxu; Xie Shengli; Zhang Haochuan
Original assignee: Guangdong University of Technology
Current assignee: Guangdong University of Technology
Priority date: 2016-10-25
Filing date: 2016-10-25
Publication date: 2018-05-09
Also published as: GB201618027D0

Abstract

A method for extracting foetal ECG from combined maternal and foetal ECG measured using abdominal electrodes uses a separation vector, and is based on the relative sparse characteristic of the source signal in time domain. A plurality of mixed mother-fetus electrocardio signals from different positions on the abdomen of a mother are collected and pre-processed. Two of the mixed signals are selected using maximum SNR, and positions of QRS wave groups of the maternal ECG and fetal ECG are determined to identify relatively sparse periods of the electrocardiograms where the QRS groups of the maternal ECG and FECG do not overlap. The relatively sparse periods are converted to a time-frequency domain using an ambiguity function, and signal items and cross items in each time-frequency distribution are calculated. A control function is constructed based on general Rayleigh quotient to separate the maternal and fetal signals.

Description

(54) Title of the Invention: Method for blind extraction of fetal electrocardio based on time-frequency conversion Abstract Title: Blind extraction of foetal ECG based on time-frequency conversion (57) A method for extracting foetal ECG from combined maternal and foetal ECG measured using abdominal electrodes uses a separation vector, and is based on the relative sparse characteristic of the source signal in time domain. A plurality of mixed mother-fetus electrocardio signals from different positions on the abdomen of a mother are collected and pre-processed. Two of the mixed signals are selected using maximum SNR, and positions of QRS wave groups of the maternal ECG and fetal ECG are determined to identify relatively sparse periods of the electrocardiograms where the QRS groups of the maternal ECG and FECG do not overlap. The relatively sparse periods are converted to a timefrequency domain using an ambiguity function, and signal items and cross items in each time-frequency distribution are calculated. A control function is constructed based on general Rayleigh quotient to separate the maternal and fetal signals.

Fig. 2

1/4

Collect a plurality of mixed mother-fetus electrocard io signals from different positions on abdomen of mother _____i_______

Pre-process the mixed mother-fetus eIectrocardio signals, including correcting baseline drift of the signals, filtering out interference at working frequency of 50Hz, and filtering out high frequency signal [ interference incIuding myoeIectricity

V

Select two of the plurality of pre-processed mixed mother-fetus eIectrocardio signals based on maximum signaI-to-noise ratio

I Locate QRS wave groups of the eIectrocardio of the mother and the eIectrocardio signal of the fetus in the two mixed signal using

R wave locating techniques for adult eIectrocardio and fetal eIectrocardio, respectively _r_____T_____

L Search for relatively sparse periods of the electrocardio of the mother and the electrocardio of the fetus in the mixed s i gnaIs

................▼............

Convert the relatively sparse signals to a time-frequency domain using an ambiguity

I function and construct a control function based on features of the sparse signal in the time-frequency domain for separating the electrocardio of the mother and the _electrocardio of the fetus_

F i g. 1

2/4

Fig. 2

3/4

4/4

Fig. 4

. '1.

Jk.....d ft-j L^*- 4„ Jf J; jAv -- ·· «* v

τ^: λ «ΐ i I

Fig. 5

Fig.6

METHOD FOR BLIND EXTRACTION OF FETAL ELECTROCARDIO BASED ON TIME-FREQUENCY CONVERSION

TECHNICAL FIELD

The present disclosure relates to a method for blind extraction of fetal electrocardio based on time-frequency conversion, which is an improved technique for blind extraction of fetal electrocardio based on time-frequency conversion.

BACKGROUND

Researches have shown that most of pressures suffered by mothers during their pregnancies come from health conditions of their fetuses. There are to approximately 0.8% of newborns have congenital heart defects every year. Hence, it is necessary to monitor fetuses during their perinatal periods, which has direct impacts on safeties of pregnant women and their fetuses, growth of newborns and their long-term intelligence development. As demands on health and safety of mothers and infants become increasingly higher, it has been an important topic to study physiologies and pathologies associated with fetal growth.

At present, a regular clinical examination on a health condition of a fetus during a mother’s pregnancy and delivery is detection of heart rate of the fetus. This is an important parameter to determine the health condition of the fetus. Abnormal fetal heart rate or pattern implies anoxia or other problems, sometimes even dangers that may require C-section. Conventionally, schemes for obtaining fetal heart rate include Doppler ultrasound detection and fetal magnetocardiogram (FMCG). The Doppler ultrasound detection is commonly used in clinic, but has some drawbacks, e.g., it cannot be used for continuous measurement, it may misinterpret uncertain acceleration or deceleration or normal transients as noises, and it may cause radioactive damage to fetuses. The FMCG detects electrophysiological variations of heart by detecting changes in heart magnetic field. It uses very sensitive Super-Conducting Quantum Interference Devices (SCQIDs) as probes arranged at the abdomen of a pregnant woman and is capable of distinguish between the magnetocardio of the fetus and the stronger magnetocardio of the mother. However, the FMCG devices are heavy and expensive and not available in most of hospitals. Further, there are various signal processing methods for fetal heart rate detection. Most of such methods are based on non-invasive detection, i.e., a mixed mother-fetus electrocardio signal collected at the abdomen is processed by means of e.g., matched filtering and adaptive filtering. In the matched filtering, the mother’s electrocardio collected at the chest is subtracted from the amplified mixed mother-fetus electrocardio signal collected at the abdomen, so as to obtain the fetal electrocardio with noise and interference and then determine the fetal heart rate from the obscure fetal electrocardio. In the adaptive filtering, the mother’s electrocardio signal is inputted as a reference for adaptive filtering, and then cancelled to extract the fetal electrocardio signal. However, this method does not have a good effect and can only be used as a method for fetal heart rate detection.

From the perspective of physiology, an electrocardio waveform contains plenty of information on human health. Fetal electrocardiogram (FECG) can record potential variations of a fetal heart occurring in each cardiac cycle activity and its conduction process in the heart. It can be used for multiple detections and dynamic observations. A clear fetal electrocardio waveform can not only detect the fetal heart rate, but also contain significant clinical information. By observing some features of the FECG, such as heart rate, waveform and dynamic fluctuation, it is possible to derive the growth, development, maturity, fetal distress or congenital heart defect of the fetus very conveniently, so as to assist doctors in making proper and emergent decisions during the gestation period and reduce the incidence rate and death rate of perinatal fetuses. Currently, clinical collections and waveform researches on adults’ electrocardio have been relatively mature, but researches on fetal electrocardio are still in the initial stage for many reasons. First, the fetal electrocardio is weak, i.e., weaker than the mother’s electrocardio amplitude by several orders of magnitudes. Meanwhile, there are a significant amount of noises in the signal transmission and collection processes and the fetal signal is typically buried in the noises, such that the Signal-to-Noise Ratio (SNR) of the fetal electrocardio is low. Second, there is no research on the clinical significance of the fetal electrocardio waveform. Third, there is no research database for fetal electrocardio waveforms.

Currently, methods for obtaining a fetal electrocardio signal mainly include invasive and non-invasive electrocardio signal collection methods. The invasive method, i.e., the scalp electrode method, causes traumas and can only be applied after amniorrhexis, and is thus inconvenient. The non-invasive method collects an electrocardio signal at the mother’s abdomen and, after noise reduction and separation by algorithm, extracts a fetal electrocardio signal. It does no hurt to the mother and the fetus. However, due to the characteristics of the electrocardio to signals of the mother and the fetus, it is difficult to extract a clear fetal electrocardio waveform in a simple way. Hence, it is important to select an appropriate and robust algorithm to improve the quality of the fetal electrocardiogram.

Blind signal processing, i.e., Independent Component Analysis (ICA), is a better method for signal processing and can be used to remove the mother’s electrocardio waveform, reduce interferences of myoelectricity, movement and power source, and enhance the QRS wave group of the fetal electrocardio.

SUMMARY

In view of the above problems, it is an object of the present disclosure to provide a method for blind extraction of fetal electrocardio based on time-frequency conversion, which is stable, efficient and simple. The present disclosure is capable of extracting the fetal electrocardio efficiently for clinical detection.

In a solution of the present disclosure, a method for blind extraction of fetal electrocardio based on time-frequency conversion is provided. The method comprises steps of:

1) placing electrodes at different positions on a surface of abdomen of a mother to collect a plurality of mixed mother-fetus electrocardio signals each containing an electrocardio of the mother and an electrocardio of a fetus mixed with each other;

2) pre-processing the plurality of mixed mother-fetus electrocardio signals, said pre-processing comprising correcting baseline drift, filtering out high frequency signal interference including myoelectricity, and filtering out interference at working frequency of 50Hz;

3) selecting two of the pre-processed mixed mother-fetus electrocardio signals based on their signal-to-noise ratios (SNRs), so as to locate the electrocardio of the mother and the electrocardio of the fetus in the two mixed mother-fetus electrocardio signals, respectively, and search for relatively sparse periods from the signals; and

4) converting the signals in the relatively sparse periods from a time domain to a time-frequency domain using an ambiguity function, applying math operations to signal items and cross items to obtain a separation vector for extracting a fetal electrocardio signal from the mixed signals.

In the step 2), the baseline drift is corrected by applying an 8-order Butterworth HR digital high pass filter having a cutoff frequency of 0.03Hz, the interference at working frequency of 50Hz is filtered out by using a digital comb band trap, and the high frequency myoelectricity signal is filtered out by using a 4-order Butterworth HR digital low pass filter having a cutoff frequency of 250Hz.

The step 3) comprises:

31) selecting one of the pre-processed mixed mother-fetus electrocardio signals that has a highest SNR, with the electrocardio of the mother being a signal and other signals being noises;

32) searching for a position of a QRS wave group of the electrocardio signal of the mother using an adult electrocardio R wave locating technique;

33) selecting one of the pre-processed mixed mother-fetus electrocardio signals that has a highest SNR, with the electrocardio of the fetus being a signal and other signals being noises;

34) searching for a position of a QRS wave group of the electrocardio signal of the fetus using a fetal electrocardio R wave locating technique; and

35) expanding the QRS wave groups of the mother and the fetus on one single time axis, wherein overlapping wave groups are discarded, non-overlapping and adjacent QRS wave groups of the mother and the fetus are determined as the relatively sparse periods for the electrocardio of the mother and the electrocardio of the fetus, and the two selected mixed signals in the relatively sparse periods are χ™/ ₌ [χ^ x™/ l^r represented as a vector ^{1 ] J J} , where the electrocardio of the mother in the relatively sparse periods are represented as a vector

-J and the electrocardio of the fetus in the relatively sparse periods are represented as a vector ~¹ i >2 J .

io

The step 4) comprises:

41) for the vector ^x ’ ¹ ’”2 1 obtained in the step 3), where * ¹ and * 2 denote the electrocardio of the mother and the electrocardio of the fetus in the relatively sparse periods in the two selected mixed mother-fetus electrocardio signals, respectively, converting X ‘ to a time-frequency plane with the ambiguity _xmf V'f, function to obtain four time-frequency distributions ’ i , where ^,jM“denote ambiguity function distributions when ^f _Or cross ambiguity function distributions when wherein the signal items are located at coordinate origin and cross items are laterally symmetric with respect to the origin; calculating a maximum real number in the left cross items in each of the four time-frequency distributions, and arranging the four real numbers in accordance with Equation (5) to obtain a matrix A;

42) for the vector

7ft. . fti _ΊΓ obtained in the step 3), where ¹ and ^x2 denote the electrocardio of the mother in the relatively sparse periods in the two selected mixed mother-fetus electrocardio signals, as in the above step,

7?ΐ converting ^x to a time-frequency plane with the ambiguity function; calculating a maximum real number in the signal items located at coordinate origin in each time-frequency plane, and arranging the four real numbers in accordance with Equation (5) to obtain a matrix ®, and, similarly, calculating a matrix by _x/ _ r_x/ _χ/ converting the relative sparse vector ^{1 1 f 2 J} of the electrocardio of the fetus to the time-frequency domain; and

43) constructing a control function ^minJ<^w> =w^rAw/w^rBw _Qr minJ(w)-w Aw/w Cw , _w^_ere w is a column vector in the separation matrix ₅ W for extracting the fetal electrocardio signal, wherein the two selected signals are represented as a vector ^x-[^xi'^xd and the separated fetal electrocardio signal is * 7 .

Compared with the conventional solutions, the present disclosure has the following advantages and effects:

io 1. Based on the relative sparse characteristic of the source signal in time domain, the present disclosure solves the problem that the electrocardio signal of the mother and the electrocardio signal of the fetus overlap in time domain and frequency domain and are difficult to be separated.

2. The method for converting the signal to the time-frequency domain has a high anti-noise ability and can extract the fetal electrocardio signal efficiently and accurately for medical diagnoses.

3. The method according to the present disclosure does not need to estimate high-order statistical characteristics of the source signal and has low computational complexity.

The method for blind extraction of fetal electrocardio based on time-frequency conversion according to the present disclosure has a sophisticated design and a good performance and is convenient and applicable.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a flowchart illustrating a method according to the present disclosure;

Fig. 2 is a flowchart illustrating a process for searching for relatively sparse periods in mixed mother-fetus electrocardio signals in the method according to the present disclosure;

Fig. 3 is a flowchart illustrating a blind extraction method based on 5 time-frequency analysis in the method according to the present disclosure;

Fig. 4 and Fig. 5 are schematic diagrams each showing effects of locating a mother’s electrocardio and a fetus’ electrocardio using the method according to the present disclosure; and

Fig. 6 shows a fetal electrocardio signal separated from the mixed io mother-fetus electrocardio signals using the algorithm in the method according to the present disclosure.

DETAILED DESCRIPTION OF THE EMBODIMENTS

The blind separation technique is to convert a signal to the time-frequency domain using a Cohen Class time-frequency distribution function, so as to represent the features of the signal on a time-frequency plane. However, a quadratic time-frequency conversion will create cross items on the time-frequency plane, such that joint diagonalization of spatial time-frequency distribution matrices will be influenced. Hence, a kernel function is used in advance to remove the interferences from the cross items. However, in the application of blind signal separation of mother-fetus signals based on time-frequency domain conversion, due to the sparseness of the mother-fetus signals, the cross items can be considered as a control function to facilitate the signal separation. This algorithm is simple and fast.

It is assumed here that a mixed mother-fetus electrocardio satisfies a noise-free linear instantaneous mixture model for blind separation:

an observation vector is

mother-fetus electrocardio signal vector and is a 2x2 full-rank mixing matrix.

The object of blind separation is to find a de-mixing matrix AV, which can be considered as a virtual inverse matrix of the mixing matrix -A, such that:

The source signal can be recovered by using the matrix ^w .

₅ ?'(*)- LM*)»Λis an estimate of the source signal. Assuming the mother’s electrocardio signal and the fetus’ electrocardio signal are statistically independent from each other, we have:

From the above equation, ⁼ (*-t/2)]_?/ = L2 , _tgken to in conjunction with ^- A^s(t), _we ή_3νθ·

Fourier transformation is applied to with respect to a spatial

-yf tl ambiguity function spectrum (SAFS) of the observation signal '' ' can be obtained:

(A jT) = I R_m (A τ) exp( -- }2πJt)di = E{S.4FD_M (At) } (4) *Fβ^Ht/2T«(Here, ' ' * is referred to a spatial ambiguity function distribution (SAFD). Given the finiteness of

SAFD (t f} the observation signal, ^M ' can be considered as an estimate of

SAFS^t.f), _where &!F£\_X (t_ff) _{cgn be re}p_resent_ecj _as:

SAfD^i,f)

(5) where

Is referred to as a cross SAFD between and “T. The SAFD of ** is

when

The ambiguity function is a common time-frequency transformation, which transforms the instantaneous correlation function of the signal to a time delay frequency offset plane for representing a correlation, i.e., a correlation domain representation. For a signal having two complex harmonic waves superimposed, its to ambiguity function has the following features: two signal items of the ambiguity function are located together, centered at the origin (0, 0), and two cross items are separated, symmetric with respect to the origin.

The method for blind extraction of fetal electrocardio based on time-frequency conversion according to the present disclosure includes steps of:

In an embodiment, in the above step 2), the baseline drift is corrected by applying an 8-order Butterworth HR digital high pass filter having a cutoff frequency of 0.03Hz, the interference at working frequency of 50Hz is filtered out by using a digital comb band trap, and the high frequency myoelectricity signal is filtered out by using a 4-order Butterworth HR digital low pass filter having a cutoff frequency of 250Hz.

In an embodiment, the above step 3) includes:

35) expanding the QRS wave groups of the mother and the fetus on one single time axis, wherein overlapping wave groups are discarded, non-overlapping and adjacent QRS wave groups of the mother and the fetus are determined as the relatively sparse periods for the electrocardio of the mother and the electrocardio of the fetus, and the two selected mixed signals in the relatively sparse periods are represented as a vector ~ i Ί ’* 2 J where the electrocardio of the mother in the relatively sparse periods are represented as a vector and io the electrocardio of the fetus in the relatively sparse periods are represented as a vector ^χΓ =Ι^χί·^χίΓ.

In an embodiment, the above step 4) includes:

mf _ r ffi/ mf j?

41) for the vector ^x J obtained in the step 3), where -1 and * - denote the electrocardio of the mother and the electrocardio of the fetus in the relatively sparse periods in the two selected mixed mother-fetus electrocardio signals, respectively, converting ^{x J} to a time-frequency plane with the ambiguity function to obtain four time-frequency distributions

AF .,-W.?' j 7 am where ' fli « λ .· “ ‘ denote ambiguity function distributions when _or cross ambiguity function distributions when * wherein the signal items are located at coordinate origin and cross items are laterally symmetric with respect to the origin; calculating a maximum real number in the left cross items in each of the four time-frequency distributions, and arranging the four real numbers in accordance with Equation (5) to obtain a matrix

42) for the vector obtained in the step 3), where and denote the electrocardio of the mother in the relatively sparse periods in the two selected mixed mother-fetus electrocardio signals, as in the above step, converting to a time-frequency plane with the ambiguity function; calculating a maximum real number in the signal items located at coordinate origin in each time-frequency plane, and arranging the four real numbers in accordance with

Equation (5) to obtain a matrix B , and, similarly, calculating a matrix by converting the relative sparse vector H ' 2 J _of the electrocardio of the fetus to the time-frequency domain; and

43) constructing a control function minJ(w)-w'Aw/w^rBw Qr min J(w) = w^r Aw/w^rCw, where w is g co|umn vec(or jn ,_{he separation matrix} , for extracting the fetal electrocardio signal, wherein the two selected signals 11 χ = |χ χΊ⁷ are represented as a vector ^L and the separated fetal electrocardio y , = w¹ X signal is * ⁷

Claims

What is claimed is: 1. A method for blind extraction of fetal electrocardio based on time-frequency conversion, comprising steps of:

4. The method of claim 1, wherein the step 4) comprises:

41) for the vector A >·* i - 4 obtained in the step 3), where

20 and * 2 denote the electrocardio of the mother and the electrocardio of the fetus in the relatively sparse periods in the two selected mixed mother-fetus electrocardio signals, respectively, converting x to a time-frequency plane with the ambiguity function to obtain four time-frequency distributions ’ j , where >- denote ambiguity function distributions when or cross ambiguity

25 function distributions when #, wherein the signal items are located at coordinate origin and cross items are laterally symmetric with respect to the origin; calculating a maximum real number in the left cross items in each of the four time-frequency distributions, and arranging the four real numbers in accordance with Equation (5) to obtain a matrix A;

42) for the vector

X' obtained in the step 3), where xt and ??l 2 denote the electrocardio of the mother in the relatively sparse periods in the two selected mixed mother-fetus electrocardio signals, as in the above step, converting to a time-frequency plane with the ambiguity function; calculating a maximum real number in the signal items located at coordinate origin in each time-frequency plane, and arranging the four real numbers in accordance with

Equation (5) to obtain a matrix ® , and, similarly, calculating a matrix by / _ r f ιΓ converting the relative sparse vector x “ -Xl J of the electrocardio of the fetus to the time-frequency domain; and

43) constructing a control function n«nJ(w) = wAw/w^Bw or min J(w) = wr Aw/wrCw, where „ is θ c0|umn vector in the separation matrix for extracting the fetal electrocardio signal, wherein the two selected signals are represented as a vector x_fxi'x2l and the separated fetal electrocardio signal is is * Λ

Intellectual

Property

Office

Application No: GB1618027.5

2. The method of claim 1, wherein, in the step 2), the baseline drift is corrected by applying an 8-order Butterworth HR digital high pass filter having a cutoff frequency of 0.03Hz, the interference at working frequency of 50Hz is filtered out by using a digital comb band trap, and the high frequency myoelectricity signal is filtered out by using a 4-order Butterworth HR digital low pass filter having a cutoff frequency of 250Hz.

3. The method of claim 1, wherein the step 3) comprises:

35) expanding the QRS wave groups of the mother and the fetus on one single time axis, wherein overlapping wave groups are discarded, non-overlapping and adjacent QRS wave groups of the mother and the fetus are determined as the relatively sparse periods for the electrocardio of the mother and the electrocardio of the fetus, and the two selected mixed signals in the relatively sparse periods are represented as a vector _ r mf ιΓ

Λ — «Λί j where the electrocardio of the mother in the relatively sparse periods are represented as a vector and the electrocardio of the fetus in the relatively sparse periods are represented as a vector