Mother and fetus monitoring system

Abstract

The invention discloses a maternal and fetal monitoring system which comprises an intelligent sensing device composed of a multi-channel flexible electrode and an accelerometer, the electrode is used for collecting maternal abdomen bio-electricity signals, and the accelerometer is used for collecting maternal acceleration signals; the signal processing and analyzing module is used for carrying out continuous wavelet transformation on the bio-electricity signals according to the collected bio-electricity signals to obtain time-frequency characteristics, recognizing the classification of each fragment in the bio-electricity signals by adopting a deep neural network according to the time-frequency characteristics, extracting maternal electrocardio and fetal electrocardio according to the classified fragments, and sending the extracted maternal electrocardio and fetal electrocardio to a server; according to the collected acceleration signal of the parent body, the sleeping posture of the parent body is detected, and an abnormal alarm is given out; the closed-loop regulation and control module is used for evaluating the maternal-fetal state according to the fetal electrocardiogram, the maternal electrocardiogram and the sleeping posture and adaptively generating an intervention scheme; and the audio intervention module is used for playing music according to the audio intervention strategy. The method is high in accuracy and can realize simultaneous monitoring of the maternal body and the fetus.

Description

Technical Field

[0001] The present invention relates to the technical field of physiological signal processing, and in particular to a maternal-fetal monitoring system. Background Art

[0002] In obstetric clinical practice, electronic fetal monitoring (EFM) is an important tool to ensure the safety of the fetus in utero and the good outcome of the newborn. It mainly relies on ultrasound technology to detect changes in the fetal heart rate and is often used to monitor the fetal condition before and during delivery. However, the physical characteristics of ultrasound signals make it difficult to achieve continuous and long-term monitoring. Especially in clinical practice, the ultrasound probe needs to be fixed on the mother's abdomen, which is greatly affected by the movement of the mother and the fetus, and it is easy to cause signal loss. In addition, the operation process of EFM is relatively complicated and the fetal heart is difficult to locate. The probe of fetal heart monitoring needs to be placed close to the back of the fetus. It is not only difficult to find the fetal heart in the abdomen of the pregnant woman, but the position of the fetal heart is easy to shift with the intrauterine movement of the fetus. Medical staff often need to manually adjust the equipment or position to ensure the stability of the signal. Another significant problem is that the sensitivity and specificity of EFM are not ideal, which brings many troubles to actual clinical work. Some pregnant women who are identified as false positives by fetal heart monitoring are given emergency cesarean sections, which increases unnecessary surgery; while another part of pregnant women with false negatives do not identify fetal distress in time, resulting in intrauterine fetal death or permanent damage to the fetus / neonatal. Since EFM is based on Doppler ultrasound, it is extremely disadvantageous for fetal monitoring in obese pregnant women (especially those with more abdominal fat). In view of the current limitations of EFM, it is urgent to find a fetal intrauterine monitoring system that is easy to operate and highly accurate.

[0003] During long-term fetal ECG monitoring, the fetal ECG signal is easily affected by noise, maternal motion artifacts, and myoelectric interference. Existing methods lack a targeted screening mechanism, which affects the accuracy of fetal ECG analysis. In addition, existing systems focus more on fetal health monitoring, ignore the comprehensive assessment of maternal health status, and cannot provide comprehensive health management support for pregnant women. The sleeping position of pregnant women may also cause health problems for the fetus, and existing monitoring systems lack effective assessment of sleeping position. Existing maternal-fetal monitoring systems mainly rely on fixed rules for intervention, and fail to perform dynamic adaptive adjustments based on the physiological state of the mother and fetus, which affects the personalization and effectiveness of intervention. Summary of the invention

[0004] In view of the problems existing in the prior art, an object of the present invention is to provide a maternal-fetal monitoring system which is more accurate and can monitor the fetus and the mother at the same time.

[0005] In order to achieve the above-mentioned object of the invention, the present invention provides a maternal-fetal monitoring system, which comprises:

[0006] An intelligent sensing device, comprising a multi-channel flexible electrode and an accelerometer, wherein the multi-channel flexible electrode is used to collect bioelectric signals from the abdomen of the mother, and the accelerometer is used to collect acceleration signals from the mother;

[0007] The signal processing and analysis module is used to perform continuous wavelet transform on the collected bioelectric signals to obtain time-frequency features, and use a deep neural network to identify the classification of each segment in the bioelectric signal based on the time-frequency features, extract the maternal electrocardiogram and fetal electrocardiogram based on the classified segments, and detect the maternal sleeping posture based on the collected maternal acceleration signal, and issue an abnormal alarm;

[0008] Closed-loop control module, used to evaluate maternal and fetal status based on fetal ECG, maternal ECG, and sleeping posture, and adaptively generate intervention plans including audio intervention strategy and maternal regulation strategy;

[0009] The audio intervention module is used to play music according to the audio intervention strategy.

[0010] Furthermore, the fetal monitoring system further comprises:

[0011] Mobile terminals, used for data visualization, monitoring result display, user interaction and abnormal alarm notification;

[0012] The communication module uses wired and / or wireless methods to achieve data transmission between the intelligent sensing sensor device and the mobile terminal.

[0013] Furthermore, the signal processing and analysis module specifically includes:

[0014] A preprocessing unit, used for preprocessing the collected abdominal bioelectric signals, including removing power frequency interference, removing baseline drift, and removing high-frequency noise;

[0015] The signal detection and screening unit is used to divide the bioelectric signal into several signal segments, and perform continuous wavelet transform on each segment to obtain time-frequency characteristics. A deep neural network is used to identify whether each segment belongs to a maternal-fetal ECG mixed segment, a maternal ECG segment, or other types of segments according to the time-frequency characteristics, and finally select high-quality signal segments;

[0016] A maternal ECG extraction unit is used to separate the maternal ECG signal from the maternal ECG segment when the screened high-quality signal segment contains the maternal ECG segment, and to obtain the maternal ECG signal from the maternal-fetal ECG mixed segment by using an independent component analysis method when the screened high-quality signal segment does not contain the maternal ECG segment, and calculate the maternal heart rate and heart rate variability;

[0017] A fetal ECG extraction unit is used to extract a fetal ECG signal from a maternal-fetal ECG mixed segment based on the maternal ECG signal obtained by the maternal ECG extraction unit, and calculate the fetal heart rate and heart rate variability;

[0018] The sleeping posture monitoring unit is used to detect the sleeping posture of the mother according to the collected acceleration signal of the mother. If it is detected that the mother is sleeping in a supine position and exceeds a preset time, an abnormal alarm is issued.

[0019] Furthermore, the signal detection and screening unit specifically includes:

[0020] The time-frequency coefficient calculation subunit is used to divide the bioelectric model into several signal segments, and use continuous wavelet transform to transform the signal segments from the time domain to the time-frequency domain to obtain the time-frequency coefficients. The formula is as follows:

[0021]

[0022] Among them, W(a,b) is the time-frequency coefficient after wavelet transformation, x(t) is the signal segment of the bioelectric signal, ψ(t) is the mother wavelet function, a is the scale factor, which controls the width of the wavelet, and b is the translation factor, which controls the time domain position of the wavelet. By adjusting the scale factors a and b to different values, the time-frequency coefficients of the signal segments in different frequency bands and at different times are obtained;

[0023] A time-frequency feature acquisition subunit is used to construct a time-frequency feature M = {W (a, b)} using time-frequency coefficients;

[0024] A segment classification subunit is used to input the time-frequency feature M into a trained deep neural network, output the probability that the signal segment is of each category, and take the category with the maximum probability as the category of the signal segment, wherein the deep neural network includes a convolutional layer, a pooling layer and a fully connected layer connected in sequence, and the categories include maternal-fetal ECG mixed segments, maternal ECG segments, fetal movement interference segments, uterine myoelectric segments, and noise segments;

[0025] The screening subunit is used to calculate the quality score of each signal segment according to the following formula, and select several signal segments with the largest quality scores as high-quality signal segments:

[0026] ω class =α 1 P FECG +α 2 P MECG +α 3 P UEMG +α 4 P Move +α 5 P Noise

[0027] In the formula, ω class is the mass fraction, P FECG , P MECG , P UEMG , P Move , PNoise are the probabilities of the maternal-fetal ECG mixed fragment, maternal ECG fragment, fetal movement interference fragment, uterine myoelectric fragment, and noise fragment, respectively, 1 , α 2 , α 3 , α 4 , α 5 are the weighting coefficients for each category.

[0028] Furthermore, the maternal ECG extraction unit specifically includes:

[0029] The multi-scale analysis subunit is used to perform multi-scale analysis on the maternal ECG fragments by using discrete wavelet transform when there are maternal ECG fragments in the screened signal fragments, and decompose the maternal ECG fragments into sub-band signals D of n scales. 1 , D 2 ...D n ,satisfy Where D i represents the subband signal of the i-th scale, x MECG (t) represents the maternal ECG segment;

[0030] The db wavelet decomposition subunit is used to select the db wavelet to decompose the subband signal of each scale to obtain the approximation signal A j and detail signal D j ;

[0031] Denoising subunit, used to decompose the detail signal D obtained by the db wavelet decomposition subunit j , the soft threshold method is used to remove high-frequency noise, specifically:

[0032] D thresh,j (t) = sign (D j )·max(|D j |-λ,0)

[0033] In the formula, λ is the threshold, D thresh,j (t) is the detail signal after denoising;

[0034] The reconstruction subunit is used to perform inverse wavelet transform (IDWT) on the denoised signal to reconstruct the maternal ECG signal, specifically:

[0035]

[0036] Where M(t) is the reconstructed maternal ECG signal, A J is the approximation signal of the last level, J is the db wavelet decomposition series;

[0037] An independent component analysis subunit is used to extract a number of independent components from the maternal-fetal ECG mixed segment using an independent component analysis method when there is no maternal ECG segment in the screened signal segment;

[0038] A power spectral density calculation subunit, used to calculate the power spectral density of each independent component;

[0039] The kurtosis calculation subunit is used to calculate the kurtosis of each independent component by the following formula:

[0040]

[0041] In the formula, K is the kurtosis, μ is the mean, E represents the expected value, x FECG,ic Represents an independent component of the maternal-fetal ECG mixed fragment;

[0042] The peak average spacing determination subunit is used to obtain all peaks in the maternal-fetal ECG mixed segment and calculate the peak average spacing based on the peak spacing:

[0043]

[0044] D s,s+1 =t s+1 -t s

[0045] Where D avg represents the average peak spacing, N represents the number of peaks, and t s+1 ,t s Respectively represent the time of the s+1th and sth peaks, D s,s+1 represents the s+1th and sth peak intervals;

[0046] The peak spacing standard deviation determination subunit is used to calculate the peak spacing standard deviation according to the following formula:

[0047]

[0048] In the formula, σ D represents the standard deviation of peak spacing;

[0049] A maternal ECG component identification subunit is used to identify maternal ECG signals based on power spectrum density, kurtosis, and peak spacing standard deviation, using high-order statistical analysis and correlation analysis;

[0050] The maternal characteristic calculation subunit is used to calculate the maternal heart rate and heart rate variability based on the identified maternal electrocardiogram signal.

[0051] Furthermore, the fetal electrocardiogram extraction unit specifically includes:

[0052] The adaptive filtering subunit is used to adopt the least mean square (LMS) adaptive filtering algorithm, use the maternal ECG signal extracted by the maternal ECG extraction unit as the reference signal for adaptive filtering, and calculate the fetal ECG signal f from the maternal-fetal ECG mixed segment. fecg (t);

[0053] The fetal ECG feature calculation subunit is used to extract features from the extracted fetal ECG signal and calculate the time domain features, frequency domain features, nonlinear features and R peak position respectively;

[0054] A classification subunit, used for determining whether the extracted fetal electrocardiogram signal is valid by using a support vector machine to extract the features;

[0055] The fetal ECG calculation subunit is used to calculate the fetal heart rate and heart rate variability using the fetal ECG signal judged to be valid.

[0056] Furthermore, the sleeping posture monitoring unit specifically includes:

[0057] The preprocessing subunit is used to remove high-frequency noise by low-pass filtering the three-axis raw acceleration data obtained by the accelerometer using a Butterworth filter;

[0058] The normalization unit is used to normalize the three-axis acceleration using the following formula:

[0059]

[0060] In the formula, A″ x , A″ y , A z ″ represents the normalized acceleration of X, Y, and Z axes, respectively, and A′ x , A′ x , A′ z Respectively represent the original acceleration of the X, Y, and Z axes, A mag is the acceleration modulus;

[0061] The feature calculation subunit is used to calculate the acceleration feature according to the following formula:

[0062]

[0063] In the formula, θ represents the pitch angle, φ represents the roll angle, φ represents the azimuth angle, and F represents the acceleration characteristic;

[0064] The clustering subunit is used to perform K-means clustering according to the acceleration features to obtain the categories of sleeping postures. The number of categories of K-means clustering is 5, corresponding to the five postures of supine, left side, right side, prone, and standing. When clustering, the cluster center of each type is first set, and then the Euclidean distance from all acceleration features to the cluster center is calculated, so as to redistribute the categories to which the acceleration features belong according to the Euclidean distance, and iteratively update the cluster center until convergence, complete clustering, and obtain the category of the current acceleration feature;

[0065] The warning subunit is used to obtain the category of sleeping posture obtained by the clustering subunit. If the category of sleeping posture is obtained, the continuous duration T is calculated. s , when T s >T tp , T tp If the preset time threshold is exceeded, an abnormal alarm will be issued.

[0066] Furthermore, the sleeping posture monitoring unit also includes:

[0067] The maternal state calculation subunit is used to extract the mother's resting stage based on the acceleration signal and calculate the sliding standard deviation through the formula:

[0068]

[0069] In the formula, σ m represents the sliding standard deviation, m represents the current time, A mag.u is the acceleration signal at time u, is the average resultant acceleration within the preset time window, and W is the preset time window length, when σ m >θ activity If the current time m is determined to be an active state, it will still be determined to be an active state within the next 15 minutes.

[0070] Furthermore, the closed-loop control module includes:

[0071] The status monitoring unit is used to evaluate the health status of the mother and fetus in real time based on the maternal and fetal ECG extracted by the signal processing and analysis module;

[0072] The adaptive intervention strategy generation unit is used to select or combine different intervention methods according to the analysis results of the state monitoring unit, and adaptively adjust the intervention intensity. When the mother or fetus is in an abnormal state, an audio intervention strategy is generated and sent to the audio intervention module, and a maternal adjustment strategy for suggesting what kind of adjustment the mother should take is generated to the mobile terminal; and after the audio intervention module intervenes, the maternal electrocardiogram and fetal electrocardiogram are obtained, and the intervention strategy is adaptively adjusted based on the changes in the maternal electrocardiogram and fetal electrocardiogram to achieve closed-loop optimization;

[0073] The interactive unit is used to provide the user with maternal regulation strategies through the mobile terminal when an abnormality is detected or intervention is required, and the user can make an independent choice.

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] The present invention constructs a maternal-fetal monitoring system that integrates high-precision signal extraction, intelligent state assessment, and adaptive intervention. The system covers intelligent sensing sensors, signal processing and analysis modules, audio intervention modules, and mobile terminal applications, and is combined with a closed-loop control mechanism. The system uses continuous wavelet transform for time-frequency analysis, and combines neural networks for signal detection and screening to identify and extract the optimal number of mixed signal fragments containing fetal electrocardiograms. Through a combination of wavelet threshold denoising, independent component analysis (ICA), and adaptive filtering, the fetal electrocardiogram signal is accurately extracted, and a machine learning model is used to evaluate the signal quality to improve the reliability of signal extraction. Through maternal electrocardiogram, maternal electrocardiogram is extracted, and combined with parameters such as fatigue, sleep, and stress, a comprehensive assessment of maternal health is provided to achieve stress management during pregnancy, sleep quality assessment, and emotional fluctuation warning. Through fetal electrocardiogram and maternal sleeping posture analysis, the health status of the fetus is assessed to reduce the risk of fetal intrauterine hypoxia and premature birth. The maternal and fetal status is adjusted through the audio intervention module (such as playing prenatal education music and soothing music), and combined with the adaptive adjustment strategy, based on the maternal HRV, fetal heart rate and PQRST waveform characteristics, the intervention effect is optimized to achieve precise and personalized intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a system block diagram of a maternal-fetal monitoring system provided by the present invention;

[0077] Figure 2 It is a schematic diagram of the closed-loop control mechanism of the maternal-fetal monitoring system of the present invention. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0079] The embodiment of the present invention provides a maternal-fetal monitoring system, such as Figure 1As shown, the system includes:

[0080] An intelligent sensing device, comprising a multi-channel flexible electrode and an accelerometer, wherein the multi-channel flexible electrode is used to collect bioelectric signals from the abdomen of the mother, and the accelerometer is used to collect acceleration signals from the mother;

[0081] The signal processing and analysis module is used to perform continuous wavelet transform on the collected bioelectric signals to obtain time-frequency features, and use a deep neural network to identify the classification of each segment in the bioelectric signal based on the time-frequency features, extract the maternal electrocardiogram and fetal electrocardiogram based on the classified segments, and detect the maternal sleeping posture based on the collected maternal acceleration signal, and issue an abnormal alarm;

[0082] Closed-loop control module, used to evaluate maternal and fetal status based on fetal ECG, maternal ECG, and sleeping position, and adaptively generate intervention plans including audio intervention strategy and maternal regulation strategy; according to fetal heart rate, maternal stress level, emotional state, uterine contraction activity and sleep status, adapt different types of intervention plans to optimize maternal and fetal health status; maternal and fetal status includes but is not limited to: fetal heart rate and heart rate variability analysis and warning, fetal movement monitoring, maternal heart rate and heart rate variability analysis, sleep quality assessment, sleeping position detection, sleep apnea screening, stress detection and emotional fluctuation detection and warning, uterine contraction detection and premature birth warning;

[0083] An audio intervention module is used to play music through a speaker according to an audio intervention strategy; for example, prenatal education music, soft soothing music, or other intervention audio to regulate the mother's stress and emotions and improve the fetal intrauterine environment;

[0084] Mobile terminals, used for data visualization, monitoring result display, user interaction and abnormal alarm notification;

[0085] The communication module uses wired and / or wireless methods to achieve data transmission between the intelligent sensing sensor device and the mobile terminal.

[0086] Each module is described in detail below.

[0087] The intelligent sensing device consists of a flexible electrode array, a right leg drive electrode (DRL), a base electrode and an acceleration sensor module. It adopts an ergonomic design and is worn on the abdomen of pregnant women through a belly support belt to ensure comfort and signal quality for long-term monitoring. Among them, the flexible electrode array includes multiple biological electrodes, which are arranged in a ring layout, with densely arranged electrodes in the core area, mainly covering the middle and lower abdomen area, to enhance the signal acquisition capability of fetal electrocardiogram (FECG). The accelerometer is arranged in the central area of ​​the belt to monitor the position, movement status and sleeping posture changes of pregnant women. This module can detect whether the pregnant woman is in a supine position, and combine the intelligent algorithm of the signal processing and analysis module for risk warning. In the specific experiment, considering the problems of high sampling rate and multi-channel, the wireless communication module ESP32WIFI ​​module can be used for data transmission. The intelligent sensing device transmits the real-time collected physiological signals to the signal processing and analysis module through the wireless communication module to ensure efficient data transmission and remote monitoring capabilities.

[0088] The signal processing and analysis module specifically includes:

[0089] The preprocessing unit is used to preprocess the collected abdominal bioelectric signals, including using a comb filter to remove power frequency interference (such as 50Hz power supply noise), and then applying a median filter to remove baseline drift to ensure signal stability. Finally, a low-pass filter (cutoff frequency 90Hz) is used to remove high-frequency noise, thereby improving the clarity and quality of the signal;

[0090] The signal detection and screening unit is used to divide the bioelectric signal into several signal segments, and perform continuous wavelet transform on each segment to obtain time-frequency characteristics. A deep neural network is used to identify whether each segment belongs to a maternal-fetal ECG mixed segment, a maternal ECG segment, or other types of segments according to the time-frequency characteristics, and finally select high-quality signal segments;

[0091] A maternal ECG extraction unit is used to separate the maternal ECG signal from the maternal ECG segment when the screened high-quality signal segment contains the maternal ECG segment, and to obtain the maternal ECG signal from the maternal-fetal ECG mixed segment by using an independent component analysis method when the screened high-quality signal segment does not contain the maternal ECG segment, and calculate the maternal heart rate and heart rate variability;

[0092] A fetal ECG extraction unit is used to extract a fetal ECG signal from a maternal-fetal ECG mixed segment based on the maternal ECG signal obtained by the maternal ECG extraction unit, and calculate the fetal heart rate and heart rate variability;

[0093] The sleeping posture monitoring unit is used to detect the sleeping posture of the mother according to the collected acceleration signal of the mother. If it is detected that the mother is sleeping in a supine position and exceeds a preset time, an abnormal alarm is issued.

[0094] The signal detection and screening unit specifically includes:

[0095] The time-frequency coefficient calculation subunit is used to divide the bioelectric model into several signal segments, and use continuous wavelet transform to transform the signal segments from the time domain to the time-frequency domain to obtain the time-frequency coefficients. The formula is as follows:

[0096]

[0097] Among them, W(a,b) is the time-frequency coefficient after wavelet transformation, x(t) is the signal segment of the bioelectric signal, ψ(t) is the mother wavelet function, a is the scale factor, which controls the width of the wavelet, and b is the translation factor, which controls the time domain position of the wavelet. By adjusting the scale factors a and b to different values, the time-frequency coefficients of the signal segments in different frequency bands and at different times are obtained;

[0098] The time-frequency feature acquisition subunit is used to construct the time-frequency feature M = {W (a, b)} using the time-frequency coefficients, where each column of the matrix M represents the feature of the signal at a time point, and each row represents the feature of the signal at that time in different frequency bands;

[0099] The segment classification subunit is used to input the time-frequency feature M into a trained deep neural network, output the probability that the signal segment is of each category, and take the category with the maximum probability as the category of the signal segment, wherein the deep neural network includes:

[0100] 1) Convolution layer: extract local feature maps in the time-frequency matrix through convolution operation, and use convolution kernel to operate. The formula is Y = f conv (M*K), where K is the trainable convolution kernel, Y is the feature map after the convolution operation, and f conv is the convolution operation;

[0101] 2) Pooling layer: Downsample the local feature map through the pooling operation to retain the main features. The formula is Z = pool (Y), where Z is the feature map after pooling and pool () is the pooling operation;

[0102] 3) Fully connected layer: After several convolution and pooling layers, the fully connected layer is used to fuse high-dimensional features, and finally the classification probability of the signal segment is obtained through the softmax activation function. The formula is: P class =softmax(WZ+b), where P class is the probability that the signal segment belongs to a certain class, W is the weight matrix, and b is the bias term;

[0103] The categories are specifically:

[0104] 1) Fetal and maternal ECG mixed segments: segments with fetal ECG (FECG) and maternal ECG (MECG) mixed as the main characteristic signal. Such segments contain significant waveform features of fetal ECG and maternal ECG, and have a good signal-to-noise ratio;

[0105] 2) Maternal ECG segments: segments with only maternal ECG (MECG) as the dominant component;

[0106] 3) Fetal movement interference segments: segments affected by fetal movement and may contain motion artifacts or non-cardiogenic signals. Such segments have more complex non-periodic waveforms and large signal frequency changes;

[0107] 4) Uterine myoelectric segment: a segment mainly composed of uterine myoelectric (UEMG) signals caused by uterine contraction activities, usually showing high-frequency, short-term, strong fluctuations, with the frequency band concentrated in the high frequency band;

[0108] 5) High noise segments: segments where the signal-to-noise ratio is lower than the preset threshold due to poor electrode contact, external interference or environmental noise, appearing as irregular signals without clear periodicity;

[0109] The screening subunit is used to calculate the quality score of each signal segment according to the following formula, and select several signal segments with the largest quality scores as high-quality signal segments:

[0110] ω class =α 1 P FECG +α 2 P MECG +α 3 P UEMG +α 4 P Move +α 5 P Noise

[0111] In the formula, ω class is the mass fraction, P FECG , P MECG , P UEMH , P Move , P Noise are the probabilities of the maternal-fetal ECG mixed fragment, maternal ECG fragment, fetal movement interference fragment, uterine myoelectric fragment, and noise fragment, respectively, 1 , α 2 , α 3 , α 4 , α 5 are the weighting coefficients for each category.

[0112] The maternal ECG extraction unit specifically comprises:

[0113] The multi-scale analysis subunit is used to perform multi-scale analysis on the maternal ECG fragments by using discrete wavelet transform when there are maternal ECG fragments in the screened signal fragments, and decompose the maternal ECG fragments into sub-band signals D of n scales. 1 , D 2 ...D n ,satisfy Where D i represents the subband signal of the i-th scale, x MeCG (t) represents the maternal ECG segment;

[0114] The db wavelet decomposition subunit is used to select the db wavelet to decompose the subband signal of each scale to obtain the approximation signal A j and detail signal D j ; Select the db wavelet basis, and the selection of the basis function should be optimized according to the characteristics of the signal and the requirements of processing accuracy, and set an accurate threshold to remove the noise component to improve the denoising effect;

[0115] Denoising subunit, used to decompose the detail signal D obtained by the db wavelet decomposition subunit j , the soft threshold method is used to remove high-frequency noise. The soft threshold method can remove high-frequency noise in the signal and maintain the main components of the maternal ECG signal. By controlling the selection of the threshold, the signal is smoothed and the noise interference in the frequency domain is eliminated. Specifically:

[0116] D thresh,h (t) = sign (D j )·max(|D j |-λ,0)

[0117] In the formula, λ is the threshold, D thresh,j (t) is the detail signal after denoising;

[0118] The reconstruction subunit performs inverse wavelet transform (IDWT) on the denoised signal to reconstruct the maternal ECG signal, specifically:

[0119]

[0120] Where M(t) is the reconstructed maternal ECG signal, A J is the approximation signal of the last level, and J is the db wavelet decomposition series.

[0121] An independent component analysis subunit is used to extract a number of independent components from the maternal-fetal ECG mixed segment using an independent component analysis method when there is no maternal ECG segment in the screened signal segment;

[0122] A power spectral density calculation subunit, used to calculate the power spectral density of each independent component;

[0123]

[0124] X(f) is the frequency domain representation of the independent components of the maternal-fetal ECG mixed segment, T is the time window, and PSD(f) is the power spectral density of the independent components;

[0125] The kurtosis calculation subunit is used to calculate the kurtosis of each independent component by the following formula:

[0126]

[0127] In the formula, K is the kurtosis, μ is the mean, E represents the expected value, x FECG,ic Represents an independent component of the maternal-fetal ECG mixed fragment;

[0128] The peak average spacing determination subunit is used to obtain all peaks in the maternal-fetal ECG mixed segment and calculate the peak average spacing based on the peak spacing:

[0129]

[0130] D s,s+1 =t s+1 -t s

[0131] Where D avg represents the average peak spacing, N represents the number of peaks, and t s+1 ,t s Respectively represent the time of the s+1th and sth peaks, D s,s+1 represents the s+1th and sth peak intervals;

[0132] The peak spacing standard deviation determination subunit is used to calculate the peak spacing standard deviation according to the following formula:

[0133]

[0134] In the formula, σ D represents the standard deviation of peak spacing;

[0135] A maternal ECG component identification subunit is used to identify maternal ECG signals based on power spectrum density, kurtosis, and peak spacing standard deviation, using high-order statistical analysis and correlation analysis;

[0136] The maternal characteristic calculation subunit is used to calculate the maternal heart rate and heart rate variability based on the identified maternal electrocardiogram signal.

[0137] The fetal electrocardiogram extraction unit specifically comprises:

[0138] The adaptive filtering subunit is used to use the least mean square (LMS) algorithm to perform the following steps, using the maternal ECG signal M(t) extracted by the maternal ECG extraction unit as the reference signal for adaptive filtering, and calculating the fetal ECG signal f from the maternal-fetal ECG mixed segment. fecg (t), specifically

[0139] 1) Select the maternal-fetal ECG mixed signal s obtained by the segment classification subunit in the signal detection and screening unit mix As the main input signal, the maternal ECG signal M(t) is used as the reference input:

[0140] 2) Maternal ECG signal estimated by adaptive filtering

[0141] Among them, ω i (n) is the weight coefficient of the filter adaptive adjustment, and N is the filter order;

[0142] 3) Calculate the error signal

[0143] 4) Filter weight update ω i (n+1)=ω i (n)+μe(n)M(ni), where μ is the step size parameter;

[0144] 5) Final fetal ECG signal f fecg (n) = e(n);

[0145] The fetal ECG feature calculation subunit is used to extract features from the extracted fetal ECG signal and calculate the time domain features, frequency domain features, nonlinear features and R peak position respectively;

[0146] The classification subunit is used to use the support vector machine to determine whether the extracted fetal ECG signal is valid; and adjust the signal processing parameters according to the classification results to improve the quality and reliability of the final signal;

[0147] The fetal ECG calculation subunit is used to calculate the fetal heart rate and heart rate variability using the fetal ECG signal judged to be valid.

[0148] The sleeping posture monitoring unit specifically includes:

[0149] The preprocessing subunit is used to remove high-frequency noise by low-pass filtering the three-axis raw acceleration data obtained by the accelerometer using a Butterworth filter;

[0150] The normalization unit is used to normalize the three-axis acceleration using the following formula:

[0151]

[0152] In the formula, A″ x , A″ y , A z ″ represents the normalized acceleration of X, Y, and Z axes, respectively, and A′ x , A′ x , A′ z Respectively represent the original acceleration of the X, Y, and Z axes, A mag is the acceleration modulus;

[0153] The feature calculation subunit is used to calculate the acceleration feature according to the following formula:

[0154]

[0155] In the formula, θ represents the pitch angle, φ represents the roll angle, φ represents the azimuth angle, and F represents the acceleration characteristic;

[0156] The clustering subunit is used to perform K-means clustering according to the acceleration features to obtain the categories of sleeping postures. The number of categories of K-means clustering is 5, corresponding to the five postures of supine, left side, right side, prone, and standing. When clustering, the cluster center of each type is first set, and then the Euclidean distance from all acceleration features to the cluster center is calculated, so as to redistribute the categories to which the acceleration features belong according to the Euclidean distance, and iteratively update the cluster center until convergence, complete clustering, and obtain the category of the current acceleration feature;

[0157] The warning subunit is used to obtain the category of sleeping posture obtained by the clustering subunit. If the category of sleeping posture is obtained, the continuous duration T is calculated. s , when T s >T tp , T tp If the preset time threshold is exceeded, an abnormal alarm is issued and sent to the mobile terminal through the wireless communication module to remind the pregnant woman to adjust her sleeping position.

[0158] In a specific implementation, the sleeping posture monitoring unit may further include:

[0159] The maternal state calculation subunit is used to extract the mother's resting stage based on the acceleration signal and calculate the sliding standard deviation through the formula:

[0160]

[0161] In the formula, σ m represents the sliding standard deviation, m represents the current time, A mag.u is the acceleration signal at time u, is the average resultant acceleration within the preset time window, and W is the preset time window length, when σ m >θactivity If the current time m is determined to be an active state, it will still be determined to be an active state within the next 15 minutes.

[0162] like Figure 2 As shown, the closed-loop control module includes:

[0163] The status monitoring unit is used to evaluate the health status of the mother and the fetus in real time based on the maternal ECG and fetal ECG extracted by the signal processing and analysis module. Specifically, the fetal heart health index and the fetal intrauterine hypoxia index are further calculated from the extracted fetal ECG signal, and the maternal sleep index, maternal stress index and maternal emotional index are further calculated from the extracted maternal ECG signal, and compared with the norm;

[0164] The adaptive intervention strategy generation unit is used to select or combine different intervention methods according to the analysis results of the state monitoring unit, and adaptively adjust the intervention intensity. When the mother or fetus is in an abnormal state, an audio intervention strategy is generated and sent to the audio intervention module, and a maternal adjustment strategy for suggesting what kind of adjustment the mother should perform (for example, guiding the pregnant woman to perform appropriate body position adjustment or relaxation exercises) is generated to the mobile terminal; and after the audio intervention module performs intervention, the maternal electrocardiogram and fetal electrocardiogram are obtained, and the intervention strategy is adaptively adjusted based on the changes in the maternal electrocardiogram and fetal electrocardiogram to achieve closed-loop optimization;

[0165] The interactive unit is used to provide the user with maternal regulation strategies through the mobile terminal when an abnormality is detected or intervention is required, and the user can make an independent choice.

[0166] Specifically, the state monitoring unit compares each characteristic parameter (indicator) with the benchmark parameter, which is derived from the group norm verified by large-scale clinical practice, and dynamically optimizes it through a dual-mode architecture of standard parameters based on group data and individualized parameter calibration. Among them, the standard parameters of group data include multiple maternal parameter norms and fetal parameter norms verified by extensive clinical cases; individualized calibration is achieved by analyzing the above indicators of the state monitoring unit in a resting state. The personalized calibration of the norm can be dynamically updated through the following formula μ i (t+1)=αμ i (t)+(1-α)x i (t). where μ i (t) is the individual benchmark value of the i-th indicator at time t, x i(t) is the original observation value of the i-th characteristic parameter at time t, and α is the memory factor of the benchmark value. In daily monitoring, the real-time characteristic parameters are continuously compared with the dynamic benchmark, and a three-level abnormality assessment model is constructed based on the degree of deviation. The first-level abnormality is: less than 3 characteristic parameters exceed the ±1.5σ range of the individual calibration library; the second-level abnormality is: 3-5 characteristic parameters exceed the ±2σ range or 1 key parameter exceeds the standard continuously for more than 5 minutes. The third-level abnormality is: more than 5 characteristic parameters exceed the ±2σ range or 3-5 key parameters exceed the standard continuously for more than 7 minutes. The adaptive intervention strategy generation unit generates a multi-level intervention plan through an intelligent algorithm to form an adaptive control strategy that links the intervention intensity with the abnormal level: when the intervention response is effective, the system will perform a step-by-step downgrade adjustment according to the preset logic to achieve closed-loop optimization and form a closed-loop control mechanism; on the contrary, if the abnormal level is continuously upgraded, the control fails, or exceeds the preset safety threshold, the multi-level early warning mechanism will be triggered and sent to the interaction unit.

[0167] It is worth noting that in the above embodiments, the various units and modules included are only divided according to functional logic, but are not limited to the above divisions, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0168] The embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, or of course, it can be implemented only by hardware, as long as the function or effect can be achieved.

[0169] It should be understood that the above embodiments and descriptions only describe the principles, main features and advantages of the present invention. Without departing from the spirit and scope of the present invention, the present invention may be subject to various changes and improvements, and these changes and improvements all fall within the scope of protection of the present invention.

Claims

1. A maternal-fetal monitoring system, characterized in that: include: An intelligent sensing device, comprising a multi-channel flexible electrode and an accelerometer, wherein the multi-channel flexible electrode is used to collect bioelectric signals from the abdomen of the mother, and the accelerometer is used to collect acceleration signals from the mother; The signal processing and analysis module is used to perform continuous wavelet transform on the collected bioelectric signals to obtain time-frequency features, and use a deep neural network to identify the classification of each segment in the bioelectric signal based on the time-frequency features, extract the maternal electrocardiogram and fetal electrocardiogram based on the classified segments, and detect the maternal sleeping posture based on the collected maternal acceleration signal, and issue an abnormal alarm; Closed-loop control module, used to evaluate maternal and fetal status based on fetal ECG, maternal ECG, and sleeping posture, and adaptively generate intervention plans including audio intervention strategy and maternal regulation strategy; The audio intervention module is used to play music according to the audio intervention strategy.

2. The fetal monitoring system according to claim 1, characterized in that: The fetal monitoring system also includes: Mobile terminals, used for data visualization, monitoring result display, user interaction and abnormal alarm notification; The communication module uses wired and / or wireless methods to achieve data transmission between the intelligent sensing sensor device and the mobile terminal.

3. The fetal monitoring system according to claim 1, characterized in that: The signal processing and analysis module specifically includes: A preprocessing unit, used for preprocessing the collected abdominal bioelectric signals, including removing power frequency interference, removing baseline drift, and removing high-frequency noise; The signal detection and screening unit is used to divide the bioelectric signal into several signal segments, and perform continuous wavelet transform on each segment to obtain time-frequency characteristics. A deep neural network is used to identify whether each segment belongs to a maternal-fetal ECG mixed segment, a maternal ECG segment, or other types of segments according to the time-frequency characteristics, and finally select high-quality signal segments; A maternal ECG extraction unit is used to separate the maternal ECG signal from the maternal ECG segment when the screened high-quality signal segment contains the maternal ECG segment, and to obtain the maternal ECG signal from the maternal-fetal ECG mixed segment by using an independent component analysis method when the screened high-quality signal segment does not contain the maternal ECG segment, and calculate the maternal heart rate and heart rate variability; A fetal ECG extraction unit is used to extract a fetal ECG signal from a maternal-fetal ECG mixed segment based on the maternal ECG signal obtained by the maternal ECG extraction unit, and calculate the fetal heart rate and heart rate variability; The sleeping posture monitoring unit is used to detect the sleeping posture of the mother according to the collected acceleration signal of the mother. If it is detected that the mother is sleeping in a supine position and exceeds a preset time, an abnormal alarm is issued.

4. The fetal monitoring system according to claim 3, characterized in that: The signal detection and screening unit specifically includes: The time-frequency coefficient calculation subunit is used to divide the bioelectric model into several signal segments, and use continuous wavelet transform to transform the signal segments from the time domain to the time-frequency domain to obtain the time-frequency coefficients. The formula is as follows: Among them, W(a,b) is the time-frequency coefficient after wavelet transformation, x(t) is the signal segment of the bioelectric signal, ψ(t) is the mother wavelet function, a is the scale factor, which controls the width of the wavelet, and b is the translation factor, which controls the time domain position of the wavelet. By adjusting the scale factors a and b to different values, the time-frequency coefficients of the signal segments in different frequency bands and at different times are obtained; A time-frequency feature acquisition subunit is used to construct a time-frequency feature M = {W (a, b)} using time-frequency coefficients; A segment classification subunit is used to input the time-frequency feature M into a trained deep neural network, output the probability that the signal segment is of each category, and take the category with the maximum probability as the category of the signal segment, wherein the deep neural network includes a convolutional layer, a pooling layer and a fully connected layer connected in sequence, and the categories include maternal-fetal ECG mixed segments, maternal ECG segments, fetal movement interference segments, uterine myoelectric segments, and noise segments; The screening subunit is used to calculate the quality score of each signal segment according to the following formula, and select several signal segments with the largest quality scores as high-quality signal segments: oh class =α1p FECG +α2P MECG +α3P UEMG +α4P Move +α5P Noise In the formula, ω class is the mass fraction, P FECG , P MECG , P UEMG , P Move , P Noise are the probabilities of the maternal-fetal ECG mixed segment, maternal ECG segment, fetal movement interference segment, uterine myoelectric segment, and noise segment, respectively. α1, α2, α3, α4, and α5 are the weighting coefficients of each category, respectively.

5. The fetal monitoring system according to claim 3, characterized in that: The maternal ECG extraction unit specifically comprises: The multi-scale analysis subunit is used to perform multi-scale analysis on the maternal ECG fragments by using discrete wavelet transform when there are maternal ECG fragments in the screened signal fragments, and decompose the maternal ECG fragments into n-scale sub-band signals D1, D2...D n ,satisfy Where D i represents the subband signal of the i-th scale, x MECG (t) represents the maternal ECG segment; The db wavelet decomposition subunit is used to select the db wavelet to decompose the subband signal of each scale to obtain the approximation signal A j and detail signal D j ; Denoising subunit, used to decompose the detail signal D obtained by the db wavelet decomposition subunit j , the soft threshold method is used to remove high-frequency noise, specifically: D thresh,j (t)=sign(D j )·max(|D j |-λ,0) In the formula, λ is the threshold, D thresh,j (t) is the detail signal after denoising; The reconstruction subunit is used to perform inverse wavelet transform (IDWT) on the denoised signal to reconstruct the maternal ECG signal, specifically: Where M(t) is the reconstructed maternal ECG signal, A J is the approximation signal of the last level, J is the db wavelet decomposition series; An independent component analysis subunit is used to extract a number of independent components from the maternal-fetal ECG mixed segment using an independent component analysis method when there is no maternal ECG segment in the screened signal segment; A power spectral density calculation subunit, used to calculate the power spectral density of each independent component; The kurtosis calculation subunit is used to calculate the kurtosis of each independent component by the following formula: In the formula, K is the kurtosis, μ is the mean, E represents the expected value, x FECG,ic Represents an independent component of the maternal-fetal ECG mixed fragment; The peak average spacing determination subunit is used to obtain all peaks in the maternal-fetal ECG mixed segment and calculate the peak average spacing based on the peak spacing: D s,s+1 =t s+1 -t s Where D avg represents the average peak spacing, N represents the number of peaks, and t s+1 ,t s Respectively represent the time of the s+1th and sth peaks, D s,s+1 represents the s+1th and sth peak intervals; The peak spacing standard deviation determination subunit is used to calculate the peak spacing standard deviation according to the following formula: In the formula, σ D represents the standard deviation of peak spacing; A maternal ECG component identification subunit is used to identify maternal ECG signals based on power spectrum density, kurtosis, and peak spacing standard deviation, using high-order statistical analysis and correlation analysis; The maternal characteristic calculation subunit is used to calculate the maternal heart rate and heart rate variability based on the identified maternal electrocardiogram signal.

6. The fetal monitoring system according to claim 3, characterized in that: The fetal electrocardiogram extraction unit specifically comprises: The adaptive filtering subunit is used to adopt the least mean square (LMS) adaptive filtering algorithm, use the maternal ECG signal extracted by the maternal ECG extraction unit as the reference signal for adaptive filtering, and calculate the fetal ECG signal f from the maternal-fetal ECG mixed segment. fecg (t); The fetal ECG feature calculation subunit is used to extract features from the extracted fetal ECG signal and calculate the time domain features, frequency domain features, nonlinear features and R peak position respectively; A classification subunit, used for determining whether the extracted fetal electrocardiogram signal is valid by using a support vector machine to extract the features; The fetal ECG calculation subunit is used to calculate the fetal heart rate and heart rate variability using the fetal ECG signal judged to be valid.

7. The fetal monitoring system according to claim 3, characterized in that: The sleeping posture monitoring unit specifically includes: The preprocessing subunit is used to remove high-frequency noise by low-pass filtering the three-axis raw acceleration data obtained by the accelerometer using a Butterworth filter; The normalization unit is used to normalize the three-axis acceleration using the following formula: In the formula, A″ x , A″ y , A z ″ represents the normalized acceleration of X, Y, and Z axes, respectively, and A′ x , A′ x , A′ z Respectively represent the original acceleration of the X, Y, and Z axes, A mag is the acceleration modulus; The feature calculation subunit is used to calculate the acceleration feature according to the following formula: In the formula, θ represents the pitch angle, φ represents the roll angle, φ represents the azimuth angle, and F represents the acceleration characteristic; The clustering subunit is used to perform K-means clustering according to the acceleration features to obtain the categories of sleeping postures. The number of categories of K-means clustering is 5, corresponding to the five postures of supine, left side, right side, prone, and standing. When clustering, the cluster center of each type is first set, and then the Euclidean distance from all acceleration features to the cluster center is calculated, so as to redistribute the categories to which the acceleration features belong according to the Euclidean distance, and iteratively update the cluster center until convergence, complete clustering, and obtain the category of the current acceleration feature; The warning subunit is used to obtain the category of sleeping posture obtained by the clustering subunit. If the category of sleeping posture is obtained, the continuous duration T is calculated. s , when T s >T tp , T tp If the preset time threshold is exceeded, an abnormal alarm will be issued.

8. The fetal monitoring system according to claim 7, characterized in that: The sleeping posture monitoring unit also includes: The maternal state calculation subunit is used to extract the mother's resting stage based on the acceleration signal and calculate the sliding standard deviation through the formula: In the formula, σ m represents the sliding standard deviation, m represents the current time, A mag.u is the acceleration signal at time u, is the average resultant acceleration within the preset time window, and W is the preset time window length, when σ m >θ activity It is judged as active state when it is in the state of sleep, otherwise it is judged as resting state.

9. The fetal monitoring system according to claim 8, characterized in that: If the current time m is determined to be in an active state, it will still be determined to be in an active state within the next 15 minutes.

10. The fetal monitoring system according to claim 2, characterized in that: The closed-loop control module includes: The status monitoring unit is used to evaluate the health status of the mother and fetus in real time based on the maternal and fetal ECG extracted by the signal processing and analysis module; The adaptive intervention strategy generation unit is used to select or combine different intervention methods according to the analysis results of the state monitoring unit, and adaptively adjust the intervention intensity. When the mother or fetus is in an abnormal state, an audio intervention strategy is generated and sent to the audio intervention module, and a maternal adjustment strategy for suggesting what kind of adjustment the mother should take is generated to the mobile terminal; and after the audio intervention module intervenes, the maternal electrocardiogram and fetal electrocardiogram are obtained, and the intervention strategy is adaptively adjusted based on the changes in the maternal electrocardiogram and fetal electrocardiogram to achieve closed-loop optimization; The interactive unit is used to provide the user with maternal regulation strategies through the mobile terminal when an abnormality is detected or intervention is required, and the user can make an independent choice.

Images