Intelligent fetal growth and development detection method and system

Abstract

The invention relates to the technical field of biomedical signal processing, and particularly discloses a fetal growth and development detection method and system based on intellectualization, and the method comprises the steps: obtaining a mixed physiological signal and a fetal movement signal, obtaining a fetal movement interference coefficient and an interference peak period through cross-correlation analysis, constructing an artifact monitoring vector, and predicting a signal deterioration probability; artifact separation is carried out; distinguishing physiological and non-physiological types of the artifact signal, and extracting a signal degradation event; and if the degradation event exists, analyzing a conduction relationship between the event and equipment sampling parameters through a rule engine, generating a main conduction path, constructing an artifact strategy mapping model, and outputting a parameter correction rule. The system comprises a fetal movement analysis module, an artifact separation module, a degradation analysis module and an equipment correction module. According to the method, the multi-dimensional feature analysis and the intelligent model are combined, so that the fetal physiological signal purity and the anomaly detection accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of biomedical signal processing, and in particular to an intelligent fetal growth and development detection method and system. Background Art

[0002] In the field of fetal growth and development monitoring, acquiring fetal autonomic nervous system signals is crucial for assessing fetal health. However, the mixed physiological signals collected by abdominal electrodes are easily interfered with by factors such as fetal movement, poor electrode contact, and environmental noise. This can cause artifacts to be mixed with the true physiological signals, seriously affecting the accuracy of subsequent analysis.

[0003] Existing technologies use fixed filtering or single adaptive filtering, which makes it difficult to dynamically match the time-varying characteristics of fetal movement artifacts and has limited effect on suppressing non-physiological artifacts, resulting in a large number of artifacts remaining in the separated neural signals. Existing artifact classification methods mostly rely on dominant frequency or time domain statistics, and do not combine electrode impedance fluctuations and signal correlation characteristics. It is difficult to distinguish between physiological artifacts (fetal movement related) and non-physiological artifacts (equipment abnormalities), resulting in an increased misjudgment rate of abnormal events.

[0004] To this end, the present invention provides an intelligent fetal growth and development detection method and system. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent fetal growth and development detection method and system to solve the above-mentioned background problems.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] The intelligent fetal growth and development detection method includes the following steps:

[0008] Acquire mixed physiological signals and motion signals, perform fetal movement interference analysis on the bioelectrical signals and motion data, and obtain the fetal movement interference coefficient and interference peak period;

[0009] Extract the monitoring features of the fetal movement interference sequence within the interference peak period, establish an artifact monitoring vector for signal deterioration analysis, trigger the artifact separation algorithm based on the deterioration analysis results, perform artifact separation on the mixed physiological signal, and output neural feature signals and artifact signals;

[0010] Differentiate and process artifact signals to obtain physiological artifact signals and non-physiological artifact signals, capture abnormal neural responses, and correlate non-physiological artifact signals with abnormal neural responses to extract signal degradation events;

[0011] If there is a signal degradation event, it is used to analyze the transmission relationship between the abnormal event and the current device sampling parameters, obtain the event transmission path and build an artifact strategy mapping model, and output the correction rules for the sampling device.

[0012] As a further solution of the present invention: the method of performing fetal movement interference analysis on the bioelectrical signal and movement data is:

[0013] Obtaining the fetal movement interference coefficients in all sliding windows, and constructing a fetal movement interference sequence including the fetal movement interference coefficients in all sliding windows;

[0014] Based on the fetal movement interference sequence, the gradient of the fetal movement interference coefficient in adjacent sliding windows is calculated using the local maximum monitoring algorithm.

[0015] Obtain the time when the gradient of the fetal movement interference coefficient changes from positive to negative as the interference peak moment;

[0016] Obtain the interference peak moment of the fetal movement event, calculate the interval time between the interference peak moments of adjacent fetal movement events, and use the interval time as the interference peak period.

[0017] As a further solution of the present invention: the method of obtaining the fetal movement interference coefficient is:

[0018] Collect fetal ECG signals, maternal ECG signals, and myoelectric signals during the monitoring period as mixed physiological signals;

[0019] The acceleration of the body surface movement caused by the fetal movement is collected by an acceleration sensor to obtain a fetal movement signal;

[0020] Calculate the mutual correlation coefficients between different types of mixed physiological signals and fetal movement signals within the sliding window;

[0021] Obtain the cross-correlation coefficients of the three types of mixed physiological signals and fetal movement signals under different sliding windows, and draw the cross-correlation curves of the three types of mixed physiological signals;

[0022] From each cross-correlation curve, obtain the time delay corresponding to the point with the maximum cross-correlation coefficient;

[0023] Based on the time delay corresponding to the maximum value of the cross-correlation coefficient, a numerical analysis was performed on the variance of the cross-correlation coefficients of different types of mixed physiological signals and the variance of the fetal movement signal to obtain the fetal movement interference coefficient.

[0024] As a further solution of the present invention, the method of performing artifact separation on mixed physiological signals is as follows:

[0025] Extract the monitoring features of the fetal movement interference sequence within the interference peak period to establish an artifact monitoring vector, perform signal deterioration analysis on the artifact monitoring vector, and obtain the signal deterioration probability;

[0026] If the signal deterioration probability is higher than or equal to the preset signal deterioration threshold, an artifact separation model is constructed to perform artifact separation on the mixed physiological signal and output neural feature signals and artifact signals.

[0027] As a further solution of the present invention, the method of performing signal degradation analysis on the artifact monitoring vector is:

[0028] Obtain the mean value of the fetal movement interference coefficient within the interference peak period, as well as the trend value and the maximum value of the mutual correlation coefficient of the fetal movement interference coefficient in the fetal movement interference sequence;

[0029] Extract the standard deviation of fetal heart rate variability from mixed physiological signals;

[0030] Construct an artifact monitoring vector including the mean value of the fetal movement interference coefficient within the interference peak period, the trend value of the fetal movement interference coefficient in the fetal movement interference sequence, the maximum value of the mutual correlation coefficient, and the standard deviation of the fetal heart rate variability;

[0031] The artifact monitoring vector is input into the binary classifier algorithm, which outputs the probability of signal deterioration in the future sliding window.

[0032] As a further solution of the present invention: the method of constructing the artifact separation model is:

[0033] Determining an artifact separation mode based on the probability of signal degradation;

[0034] Generate periodic energy coding based on interference peak period to determine artifact separation direction;

[0035] Determining different artifact separation algorithms based on the artifact separation mode;

[0036] A multi-loss function optimization model is established based on the artifact monitoring vector to achieve artifact separation.

[0037] As a further solution of the present invention, the non-physiological artifact signal is associated with the abnormal neural response in the following manner:

[0038] Extract HRV sequence from neural feature signals and obtain time domain and frequency domain features of HRV sequence;

[0039] Based on the time and frequency domain characteristics of the HRV sequence, a Boolean decision function is constructed to determine abnormal neural responses.

[0040] If the neural response is abnormal, the associated time window and neural response window are set;

[0041] If non-physiological artifacts are detected within the artifact monitoring window and there are neural abnormalities within the time window, a signal degradation event is determined to have occurred.

[0042] As a further solution of the present invention, the method of distinguishing and processing the artifact signal is as follows:

[0043] The cross-correlation coefficient between the artifact signal and the fetal movement signal within the sliding window, the dominant frequency of the artifact signal in the frequency domain within different sliding windows, and the electrode impedance fluctuation characteristics of the acquisition device within the sliding window are obtained for spatiotemporal alignment processing, and the sliding window corresponding to the fetal movement interference sequence is correspondingly spliced ​​to establish a signal discrimination vector;

[0044] The signal discrimination vector is input into the classifier, and the classifier distinguishes the artifact signal to obtain physiological artifact signals and non-physiological artifact signals.

[0045] As a further solution of the present invention, the correction rule of the sampling device is outputted in the following manner:

[0046] Analyze the degree of parameter deviation based on the event conduction path and calculate the event conduction path weight;

[0047] The event conduction path with the largest event conduction path weight value is selected as the main conduction path;

[0048] Based on the main conduction path, an artifact strategy mapping model is constructed to output correction rules for the main conduction path sampling device.

[0049] The intelligent fetal growth and development detection system includes the following modules:

[0050] Fetal movement analysis module: used to obtain mixed physiological signals and movement signals, perform fetal movement interference analysis on bioelectric signals and movement data, and obtain fetal movement interference coefficient and interference peak period;

[0051] Artifact separation module: used to extract the monitoring features of the fetal movement interference sequence within the interference peak period, establish the artifact monitoring vector for signal deterioration analysis, trigger the artifact separation algorithm based on the deterioration analysis results, perform artifact separation on the mixed physiological signal, and output the neural feature signal and artifact signal;

[0052] Degradation analysis module: used to distinguish and process artifact signals to obtain physiological artifact signals and non-physiological artifact signals, capture neural response abnormalities, and correlate non-physiological artifact signals with neural response abnormalities to extract signal degradation events;

[0053] Equipment correction module: If there is a signal degradation event, it is used to analyze the transmission relationship between the abnormal event and the current equipment sampling parameters, obtain the event transmission path and build an artifact strategy mapping model, and output the correction rules for the sampling equipment.

[0054] Beneficial effects of the present invention:

[0055] (1) By combining a classifier with a deep learning model, adaptive artifact separation of mixed physiological signals is achieved. An artifact monitoring vector is constructed based on the fetal movement interference sequence characteristics, and through a multi-loss function, the separated neural feature signals are made consistent with the physiological distribution, the deviation rate of historical artifact-free data is reduced, and the purity of the fetal autonomic neural signal is improved.

[0056] (2) Using the frequency concentration coefficient of the dominant frequency in the frequency domain and the electrode impedance fluctuation ratio, a multidimensional signal discrimination vector is constructed. The classifier is used to automatically identify the artifact type, which helps to reduce the interference of artifacts on fetal neural signal analysis. The monitoring window and neural response window are dynamically set based on the time when the artifact occurs. The fetal autonomic neural response delay is introduced to capture the impact of artifacts on HRV indicators. The abnormal neural response is quantified by the Boolean decision function, which improves the real-time monitoring capability of abnormal events.

[0057] (3) Combining frequency domain features, time domain features and device parameters, a multi-level feature analysis system is constructed, which is conducive to maintaining the reliability of neural signal analysis; the conduction path of abnormal events and sampling parameters is analyzed through the rule engine, and dynamic correction rules are generated based on the weight of the main conduction path, which is conducive to improving the adaptability of the device to complex physiological environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The present invention will be further described below with reference to the accompanying drawings.

[0059] Figure 1 This is a flow chart of the intelligent fetal growth and development detection method;

[0060] Figure 2 This is a flow chart of the determination method for performing artifact separation in the present invention;

[0061] Figure 3 It is a module diagram of the intelligent fetal growth and development detection system of the present invention. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0063] Example 1

[0064] See also Figure 1 As shown, the present invention is an intelligent fetal growth and development detection method, comprising the following steps:

[0065] S1. Acquire mixed physiological signals and motion signals, perform fetal movement interference analysis on the bioelectrical signals and motion data, and obtain the fetal movement interference coefficient and interference peak period;

[0066] Among them, the method of fetal movement interference analysis of bioelectric signals and movement data is as follows:

[0067] Preferably, the fetal electrocardiogram signal FECG and the maternal electrocardiogram signal ECG and electromyography signal EMG are collected as mixed physiological signals S during the monitoring period through the flexible electrode array on the abdomen. l (t);

[0068] Among them, l corresponds to the category of mixed physiological signals, l∈{FECG,RCG,EMG};

[0069] The acceleration sensor is used to collect the acceleration of the body surface caused by fetal movement to obtain the fetal movement signal S p (t);

[0070] It needs to be explained that the multi-axis surface motion acceleration is converted into a scalar total acceleration as the fetal movement signal through vector synthesis;

[0071] Calculate the cross-correlation coefficient CCF of type l mixed physiological signal and fetal movement signal in the sliding window l (τ);

[0072] Preferably, by the cross-correlation equation: Get the cross-correlation coefficient CCF of the lth category l (τ);

[0073] Where τ is the time delay and T is the length of the sliding window;

[0074] Obtain the cross-correlation coefficients of the three types of mixed physiological signals and fetal movement signals under different sliding windows, and draw the cross-correlation curves of the three types of mixed physiological signals;

[0075] From each cross-correlation curve, obtain the time delay τ corresponding to the maximum cross-correlation coefficient point peak,l ;

[0076] It is understandable that τ peak,l The value represents the interference delay of fetal movement on bioelectrical signals, reflecting the strongest correlation moment between mixed physiological signals and fetal movement signals; τ peak,l >0 means that the motion signal changes before the bioelectric signal, which corresponds to the interference of body surface movement caused by fetal movement on the electrode signal;

[0077] Based on the mutual correlation coefficient, the formula: Obtain fetal movement interference coefficient G l ;

[0078] It can be understood that the fetal movement interference coefficient quantifies the correlation and interference intensity between the surface movement caused by fetal movement and the bioelectric signal. The fetal movement interference coefficient is used to construct a fetal movement interference sequence, which provides dynamic interference features for the subsequent artifact separation algorithm, thereby achieving the suppression of fetal movement artifacts in mixed physiological signals and the purification of neural feature signals.

[0079] Among them, Var[S l (t)]、Var[S p (t)] represent the variance of the cross-correlation coefficient of the lth type of mixed physiological signal and the variance of the fetal movement signal in the sliding window respectively;

[0080] Obtain the fetal movement interference coefficients in all sliding windows, and construct a fetal movement interference sequence Z(t) containing the fetal movement interference coefficients in all sliding windows;

[0081] Based on the fetal movement interference sequence Z(t), the gradient of the fetal movement interference coefficient in adjacent sliding windows is calculated using the local maximum monitoring algorithm;

[0082] Obtain the time t when the gradient of the fetal movement interference coefficient turns from positive to negative as the interference peak moment;

[0083] Obtaining the interference peak moment of the fetal movement event, calculating the interval between the interference peak moments of adjacent fetal movement events, and using the interval time as the interference peak period;

[0084] S2. Extract monitoring features of the fetal movement interference sequence within the interference peak period, establish an artifact monitoring vector to perform signal deterioration analysis, trigger an artifact separation algorithm based on the deterioration analysis results, perform artifact separation on the mixed physiological signal, and output neural feature signals and artifact signals;

[0085] The method for extracting the monitoring features of the fetal movement interference sequence within the interference peak period is as follows:

[0086] Preferably, the mean value of the fetal movement interference coefficient within the interference peak period, as well as the trend value and the maximum value of the mutual correlation coefficient of the fetal movement interference coefficient in the fetal movement interference sequence Z(t) are obtained;

[0087] Extracting the standard deviation of fetal heart rate variability (HRV) from mixed physiological signals using SDNN;

[0088] It should be noted that the trend value of the fetal movement interference coefficient is obtained by linear regression fitting of the fetal movement interference coefficient sequence within the sliding window, and is used to characterize the dynamic change trend of the fetal movement interference intensity;

[0089] When extracting fetal heart rate variability (HRV) from mixed physiological signals, the time interval between two adjacent R waves (i.e., RR interval) is first extracted from the fetal electrocardiogram (FECG) signal. Then, the standard deviation (SDNN) of the time domain characteristics of the fetal electrocardiogram (FECG) signal during the RR interval is calculated to quantify the fluctuation characteristics of the fetal heart rate.

[0090] Construct an artifact monitoring vector including the mean value of the fetal movement interference coefficient within the interference peak period, the trend value of the fetal movement interference coefficient in the fetal movement interference sequence Z(t), the maximum value of the mutual correlation coefficient, and the standard deviation of the fetal heart rate variability;

[0091] The method of establishing an artifact monitoring vector for signal degradation analysis is as follows:

[0092] The artifact monitoring vector is input into the LightGBM binary classifier algorithm, and the probability of signal deterioration in the future sliding window is output;

[0093] Those skilled in the art will understand that before the artifact monitoring vector is input as a feature into the LightGBM binary classifier, the feature must first be standardized and the training set and test set must be divided into time series. By setting the LightGBM tree depth, learning rate, and minimum number of leaf node samples to optimize the model, the signal deterioration probability ≥ threshold is used as the historical signal deterioration label for the supervision signal training model, and finally the trained model is used to infer the current artifact monitoring vector and output a signal deterioration probability value between 0 and 1, which represents the risk probability of signal quality degradation due to artifacts in the future sliding window;

[0094] like Figure 2 As shown, if the signal deterioration probability is higher than or equal to the preset signal deterioration threshold, an artifact separation model is constructed to perform artifact separation on the mixed physiological signal, and a neural feature signal and an artifact signal are output;

[0095] If the signal deterioration probability is lower than the preset signal deterioration threshold, the separation is skipped and the original mixed physiological signal is output;

[0096] The mixed physiological signal is separated from the artifacts and the neural feature signal and the artifact signal are outputted as follows:

[0097] S201, determining an artifact separation mode based on a signal deterioration probability;

[0098] Preferably, by the formula: Get the separation strength coefficient λ;

[0099] Among them, P deg 、P th are signal deterioration probability and signal deterioration threshold respectively;

[0100] It can be understood that the separation mode is selected according to the separation strength coefficient λ:

[0101] 0<λ<0.5: Mild separation mode, i.e. fast filtering;

[0102] 0.5≤λ≤0.8: standard separation mode, for spatiotemporal feature modeling;

[0103] λ>0.8: Deep separation mode and enhanced adversarial learning;

[0104] If the signal deterioration probability is lower than the preset signal deterioration threshold, the separation is skipped and the original mixed physiological signal is output;

[0105] S202, generating a periodic energy code based on the interference peak period, and determining an artifact separation direction;

[0106] Preferably, the time delay τ corresponding to the maximum point of the mutual correlation coefficient is peak,l , generate periodic energy mask M p (t);

[0107] Mapping the artifact monitoring vector into a conditional vector, wherein the dimension of the conditional vector matches the mixed physiological signal feature;

[0108] Perform SIFT transformation on the mixed physiological signal to generate the time-frequency matrix X TF , expand the fetal movement interference sequence Z(t) into time-frequency domain coding;

[0109] The spatial correlation matrix of the mixed physiological signal is combined with the conditional vector to generate the channel attention weight;

[0110] It can be understood that by marking the periodic time points of fetal movement interference, a mask matrix representing the interference energy distribution is formed in the time domain to highlight the temporal characteristics of fetal movement interference;

[0111] When mapping the artifact monitoring vector to the conditional vector, it is necessary to make the dimension of the conditional vector consistent with the characteristic dimension of the mixed physiological signal through dimension matching processing (such as linear transformation), so as to provide constraints for the subsequent separation model; SIFT transformation is performed on the mixed physiological signal to convert the signal from the time domain into a time-frequency two-dimensional matrix X TF , intuitively presenting the change of energy of each frequency band over time;

[0112] The fetal movement interference sequence is expanded into time-frequency domain coding, which is to map the one-dimensional time series signal to the two-dimensional time-frequency space through Fourier transform or time-frequency analysis to form a TF A dimensionally matched interference feature matrix; when the spatial correlation matrix of the mixed physiological signal is combined with the conditional vector, a channel attention weight is generated through matrix dot multiplication or weighted operation. This weight is used to quantify the degree of interference of each channel, thereby suppressing high-interference channels and enhancing effective signal channels in artifact separation;

[0113] S203, determining different artifact separation algorithms based on the artifact separation mode;

[0114] Preferably, if the artifact separation mode is mild separation, the cutoff frequency is adjusted based on the trend value of the fetal movement interference coefficient, and the adaptive filter coefficient is calculated by the LMS algorithm with the motion signal as a reference to estimate the artifact;

[0115] If the artifact separation mode is the standard separation mode, the temporal features ht of the fetal movement interference sequence Z(t) are extracted through BiGRU, and the temporal features are concatenated with the signal features to guide the artifact separation direction;

[0116] At the same time, the time-frequency matrix X is transformed by the two-dimensional CNN algorithm TF Perform spatial feature extraction and combine channel attention weights to suppress high-interference channels;

[0117] If the artifact separation mode is the depth separation mode, the adversarial learning algorithm is introduced based on the standard separation mode. The generator and discriminator of the adversarial learning algorithm are used to conduct a game to make the separation signal conform to the physiological distribution.

[0118] It can be understood that if the artifact separation mode is mild separation, the cutoff frequency of the low-pass filter is dynamically adjusted based on the trend value of the fetal movement interference coefficient. For example, when the trend value increases, the cutoff frequency is lowered to suppress high-frequency artifacts. At the same time, the motion signal is used as the reference input, and the optimal filter coefficient is iteratively calculated through the LMS adaptive filtering algorithm to achieve real-time estimation and cancellation of the artifact signal.

[0119] If the standard separation mode is used, the BiGRU network is used to extract the time interval regularity of the interference peak of the fetal movement interference sequence, and the time interval regularity of the interference peak is concatenated with the feature vector of the mixed physiological signal as the separation direction guide. At the same time, the two-dimensional CNN is used to analyze the time-frequency matrix X. TF Perform spatial feature extraction to identify the spatial distribution of high-interference frequency bands, and combine channel attention weights to suppress channels with poor signal quality;

[0120] In the deep separation mode, a generative adversarial network is introduced on the basis of standard separation. The generator fits the distribution characteristics of the real neural signal, and the discriminator distinguishes the authenticity of the separated signal from the historical artifact-free signal. Through the game between the two sides, the separated neural feature signal is made more consistent with the physiological distribution characteristics of the fetal autonomic neural signal.

[0121] S204, establishing a multi-loss function optimization model based on the artifact monitoring vector to achieve artifact separation;

[0122] Preferably, by formula 1: Establish interference residual loss function La;

[0123] Among them, Sl (t) is the mixed physiological signal, E(t) is the neural feature signal predicted by the artifact separation model output, and A(t) is the artifact signal predicted by the artifact separation model output;

[0124] Through formula 2: Establish HRV constraint function;

[0125] Among them, SDNN(E) is the SDNN (standard deviation normal to normal interval) index for calculating the heart rate variability (HRV) of the predicted signal E(t), E * is the reference signal, which is derived from historical artifact-free data;

[0126] Through formula 3: Ls=KL[PSD(E),PSD(E * )] Establish spectrum constraint function Ls;

[0127] Among them, PSD(E), PSD(E * ) are used to estimate the power spectrum density of the prediction signal E(t) and the reference signal E * Perform power spectral density estimation;

[0128] Through formula 4: Lad = -log[D(E)]-log[1-D(E * )] Establish the adversarial loss constraint function Lad for the depth separation model;

[0129] Among them, D(E) is the authenticity score of the discriminator for the predicted signal E(t), D(E * ) is the discriminator's response to the reference signal E * authenticity rating;

[0130] Those skilled in the art will appreciate that the interference residual loss, HRV constraint loss, spectrum constraint loss, and adversarial loss optimize the artifact separation model from multiple dimensions.

[0131] The interference residual loss calculates the signal reconstruction error by weighting the artifact monitoring vector to improve the artifact suppression effect; the HRV constraint loss compares the deviation of indicators such as SDNN of the separated signal and historical artifact-free data to maintain the physiological characteristics of fetal heart rate variability;

[0132] The spectrum constraint loss matches the normal signal spectrum distribution through the KL divergence of the power spectral density to reduce signal distortion during the separation process; the adversarial loss makes the separated signal conform to the physiological distribution of the real neural signal through the game between the generator and the discriminator;

[0133] By constructing an artifact separation model that integrates signal processing and machine learning, the model first extracts features from the mixed physiological signal to generate an artifact monitoring vector containing parameters such as the trend value of the fetal movement interference coefficient and the maximum value of the cross-correlation coefficient. Based on this, the model dynamically selects mild, standard, or deep separation modes: mild separation uses LMS adaptive filtering combined with dynamic adjustment of the cutoff frequency to suppress artifacts; standard separation uses BiGRU+2D CNN to extract temporal and spatial features and combines channel attention weights to guide the separation direction; deep separation introduces adversarial learning to make the signal conform to the physiological distribution. Simultaneously, a multi-loss function system consisting of interference residual loss, HRV constraint loss, spectrum constraint loss, and adversarial loss is used as the optimization target. The separation effect is simultaneously optimized under the constraints of the artifact monitoring vector, removing the artifact signal A from the mixed physiological signal and outputting a neural feature signal E that retains the time domain waveform (such as the fetal ECG R wave sequence) and HRV sequence (time domain indicators such as SDNN and RMSSD and LF / HF frequency domain features), achieving effective separation of physiological signals from artifacts.

[0134] Based on the constructed artifact separation model, the neural feature signal E including the time domain waveform and HRV sequence, as well as the artifact signal A, is output;

[0135] It is understandable that the purpose of extracting neural feature signals and artifact signals is:

[0136] Function 1: Provide original separation samples for physiological and non-physiological classification of artifact signals, identify artifact types by analyzing artifact signal characteristics, and associate neural response abnormalities with HRV indicators of neural feature signals to extract signal degradation events;

[0137] Function 2: Provide data support for device sampling parameter correction. By analyzing the purity of neural characteristic signals and the characteristic distribution of artifact signals, the conduction relationship between abnormal events and device parameters is analyzed to generate the main conduction path and parameter correction rules.

[0138] Function 3: Provide a quantitative basis for the assessment of fetal autonomic nervous function. The HRV sequence in the neural characteristic signal directly reflects the fetal heart rate variability and is used to monitor the growth and development status of the fetus. The separation results of the artifact signal can optimize the signal acquisition process and improve the accuracy of clinical monitoring.

[0139] Example 2

[0140] like Figure 1 As shown, the intelligent fetal growth and development detection method also includes the following steps:

[0141] S3. Differentiating and processing the artifact signals to obtain physiological artifact signals and non-physiological artifact signals, capturing neural response abnormalities, and correlating the non-physiological artifact signals with the neural response abnormalities to extract signal degradation events;

[0142] The method of distinguishing and processing the artifact signal to obtain the physiological artifact signal and the non-physiological artifact signal is as follows:

[0143] Calculate the cross-correlation coefficient between the artifact signal and the fetal movement signal within the sliding window;

[0144] Extract the dominant frequency of the artifact signal in the frequency domain within different sliding windows, as well as the electrode impedance fluctuation characteristics of the acquisition device within the sliding window;

[0145] Preferably, the method of extracting the frequency domain dominant frequencies of the artifact signal in different sliding windows is:

[0146] Performing fast Fourier transform on the artifact signal in each sliding window to convert the artifact signal in the time domain into the artifact signal in the frequency domain;

[0147] Obtain all peak values ​​of the artifact signal in the frequency domain and extract the half-width and peak frequency of each peak;

[0148] The frequency concentration factor is obtained by performing ratio processing on the half width and peak frequency of each peak;

[0149] determining a frequency domain dominant frequency within a sliding window based on a frequency concentration factor;

[0150] Those skilled in the art will appreciate that, after the artifact signal is segmented into sliding windows, an FFT transform is performed on the signal in each window to obtain a frequency domain amplitude spectrum, and all significant peaks are obtained through a peak detection algorithm; the full width at half maximum (FWHM) is calculated for each peak and compared with the corresponding peak frequency to obtain a frequency concentration coefficient that reflects the degree of concentration of the frequency component, where a smaller frequency concentration coefficient indicates a more concentrated frequency; the frequency concentration coefficients of all peaks are traversed, and the peak frequency with the smallest coefficient is selected as the dominant frequency in the frequency domain within the sliding window. By quantizing the frequency concentration, wide-range noise peaks are suppressed, thereby improving the dominant frequency's ability to characterize the true artifact characteristics;

[0151] The method for collecting the electrode impedance fluctuation characteristics of the device within the sliding window is:

[0152] Obtaining the electrode impedance of the mixed physiological signal acquisition device, obtaining the ratio of the standard deviation and the mean of the electrode impedance within the sliding window, and obtaining the impedance fluctuation ratio;

[0153] The impedance fluctuation ratio is used as the electrode impedance fluctuation characteristic;

[0154] The cross-correlation coefficient between the artifact signal and the fetal movement signal calculated in the sliding window, the dominant frequency of the artifact signal in the frequency domain in different sliding windows, and the electrode impedance fluctuation characteristics of the acquisition device in the sliding window are temporally and spatially aligned, and then correspondingly spliced ​​with the sliding window corresponding to the fetal movement interference sequence to establish a signal discrimination vector;

[0155] The signal discrimination vector is input into the LightGBM classifier, which distinguishes the artifact signals and obtains physiological artifact signals and non-physiological artifact signals;

[0156] It can be understood that the cross-correlation coefficient of the artifact signal and the fetal movement signal calculated in the sliding window, the dominant frequency of the artifact signal in the frequency domain, the electrode impedance fluctuation ratio and other features are spatiotemporally aligned and spliced ​​to form a multidimensional signal distinction vector; the vector is standardized (after which it is input into the pre-trained LightGBM classifier, and the GOSS gradient sampling strategy is used to reduce the computational complexity. The tree depth is set to 6-8 layers and the learning rate is 0.1-0.3. The model is trained using historical annotated data (physiological artifacts are marked as 0 and non-physiological artifacts are marked as 1). The feature splitting threshold is optimized by the gradient boosting algorithm. Finally, the artifact type probability value is output based on the input signal distinction vector, and the automatic distinction between physiological artifact signals and non-physiological artifact signals is achieved by setting the probability threshold.

[0157] The purpose of obtaining physiological artifact signals and non-physiological artifact signals is:

[0158] Objective 1: To correlate non-physiological artifact signals with abnormal neural responses, determine neural abnormalities through HRV time and frequency domain characteristics, and extract signal degradation events to provide a basis for monitoring fetal neural signal abnormalities.

[0159] Objective 2: Analyze the conduction relationship between abnormal events and device sampling parameters (sampling rate, filter cutoff frequency) based on artifact signal types, provide feature input for the device correction module, and generate the main conduction path and parameter correction rules;

[0160] Objective 3: To optimize the artifact separation algorithm strategy by distinguishing artifact signal types. For physiological artifacts (fetal movement related), the filtering parameters can be dynamically adjusted in combination with the fetal movement interference sequence characteristics. For non-physiological artifacts (device abnormalities), the adversarial learning separation mode can be strengthened to improve the purity of the neural feature signal.

[0161] Extract HRV sequence from neural feature signals and obtain time domain and frequency domain features of HRV sequence;

[0162] The method for obtaining HRV frequency domain features is:

[0163] Perform fast Fourier transform on the HRV sequence to obtain the low-frequency and high-frequency power of the HRV sequence, and calculate the ratio of low-frequency to high-frequency power as the frequency domain feature;

[0164] The method of obtaining the time domain features of the HRV sequence is as follows: obtaining the standard deviation of the HRV sequence and the root mean square of adjacent RR intervals in each sliding window;

[0165] Based on the time and frequency domain characteristics of the HRV sequence, a Boolean decision function is constructed to determine abnormal neural responses.

[0166] Among them, the way to construct the Boolean judgment function is:

[0167] The Boolean decision function At of time domain anomaly is constructed by formula 1: At = I(|SDNN-μ1|>k1*δ1∪|RMSSD-μ2|>k2*δ2);

[0168] To indicate the function, convert the logical condition into a numerical value (0 or 1);

[0169] Wherein, SDNN is the standard deviation of the normal RR interval in the fetal ECG signal, RMSSD is the root mean square of the difference between adjacent RR intervals, μ1 and μ2 are the normal reference means of SDNN and RMSSD, respectively;

[0170] δ1: normal reference standard deviation of SDNN (the degree of dispersion of healthy fetal SDNN data in clinical statistics, used to quantify the normal fluctuation range;

[0171] k1, k2: are the preset abnormality determination coefficients;

[0172] δ2: Normal reference standard deviation of RMSSD

[0173] Through formula 2: Construct the frequency domain anomaly Boolean decision function Af;

[0174] Among them, Rl is the frequency domain feature, The normal reference mean, that is, the average value of the frequency domain characteristics of a healthy fetus based on clinical statistics, represents the baseline level of neural activity;

[0175] k3 is the preset frequency domain anomaly determination coefficient;

[0176] Determine the abnormal neural response based on the Boolean determination functions At and Af to obtain a determination result of the abnormal neural response;

[0177] It should be explained that the time domain features and frequency domain features of the neural characteristic signal are first extracted through HRV analysis, and the time domain features and frequency domain features are substituted into the Boolean judgment function At to obtain the Boolean result of time domain anomaly; at the same time, RI is substituted into the Boolean judgment function Af to obtain the Boolean result of frequency domain anomaly;

[0178] According to the clinical judgment rules for abnormal neural response, the Boolean values ​​of the Boolean judgment function At and the Boolean judgment function Af are logically operated, and the judgment result of whether the neural response is abnormal is finally output (true is 1, false is 0), thereby realizing multi-dimensional monitoring of abnormal fetal autonomic neural signals in the time domain and frequency domain;

[0179] Among them, the method of correlating non-physiological artifacts with abnormal neural responses and marking events with signal degradation is as follows:

[0180] If the neural response is abnormal, the time when the artifact signal is generated is set as t0, and the associated time window Tg is set to extract the artifact features;

[0181] Construct an artifact monitoring window [t0, t0+Tg] based on the artifact signal generation time t0 and the set associated time window Tg;

[0182] Setting the neural response window: [t0, t0+Tg+t1] is used to detect neural signal abnormalities;

[0183] Wherein, Tg is the length of the associated time window, and t1 is the preset fetal autonomic nervous response delay;

[0184] If a non-physiological artifact is detected in the artifact monitoring window and there is a neural abnormality in the neural response window time window [t0, t0+Tg+t1], it is determined that a signal degradation event exists.

[0185] S4. If a signal degradation event occurs, analyze the transmission relationship between the abnormal event and the current device sampling parameters, obtain the event transmission path, build an artifact strategy mapping model, and output the correction rules for the sampling device;

[0186] Among them, the method of analyzing the transmission relationship between abnormal events and current device sampling parameters to obtain the event transmission path is as follows:

[0187] Obtain the sampling rate, filter cutoff frequency, frequency domain dominant frequency, impedance fluctuation ratio and fetal movement interference coefficient of the acquisition device, HRV as an abnormal event and the conduction analysis characteristics of the current device sampling parameters;

[0188] Build a rule engine, input the conduction analysis features into the rule engine, and output the conduction relationship between abnormal events and current device sampling parameters;

[0189] Construct an event transmission path based on the transmission relationship between abnormal events and current devices;

[0190] It is understandable that by sorting out the correlation between the sampling rate, filter cutoff frequency, device sampling parameters, artifact characteristics and neural signal abnormalities in historical data, we can extract the if-then rule set: "If [sampling rate < threshold] and [high-frequency non-physiological artifacts are detected], then it is determined that the sampling rate is insufficient → artifact conduction";

[0191] The knowledge base of the rule engine is constructed based on a set of if-then rules. When real-time conduction analysis features are input, the rule engine quickly retrieves matching rules through a pattern matching algorithm and outputs a causal relationship conclusion between abnormal events (such as abnormal neural HRV) and device parameters with too low a current sampling rate. For example, the conclusion of insufficient sampling rate - signal aliasing - non-physiological artifacts - abnormal neural time domain characteristics. Based on this conclusion, the timing dependency of the causal nodes is further analyzed, and the logical hierarchy is analyzed with device parameters as the cause, artifacts as the intermediate medium, and neural abnormalities as the effect. Through directed graph modeling and connecting nodes in series, a visual event conduction path from device parameter defects to neural signal abnormalities is formed, presenting a complete link of abnormal transmission. At the same time, the rule base is supported to dynamically iterate with new data to continuously optimize the determination accuracy of the conduction relationship.

[0192] Analyze the degree of parameter deviation based on the event conduction path and calculate the event conduction path weight;

[0193] For example, when the sampling rate of the device is reduced from the standard value of 250 Hz to 150 Hz, an event conduction path of "insufficient sampling rate → generation of high-frequency non-physiological artifacts → abnormal time domain characteristics (SDNN) of fetal ECG signals" is constructed;

[0194] The degree of parameter deviation was calculated. The sampling rate deviation was (250-150) / 250=40%, and the corresponding deviation weight coefficient was 0.4 after linear mapping according to the deviation percentage. In the artifact node, the clinical impact factor of high-frequency non-physiological artifacts was 0.3 (based on the probability of signal degradation caused by this artifact type in historical data). In the neural abnormality node, the SDNN limit exceeded 20% and the corresponding weight was 0.3 (referring to the severity of SDNN abnormality in the clinical diagnostic criteria). The total path weight was the weighted product of the weights of each node (0.4×0.3×0.3=0.036). If there was a path weight of 0.028: the filter cutoff frequency was set too low → low-frequency physiological artifact accumulation → HRV frequency domain characteristics (LF / HF) abnormality, the former with the largest weight was selected as the main conduction path, and then based on the correction rule of "the output sampling rate of this path was increased to 250Hz, and the low-pass filter cutoff frequency was maintained at 60Hz";

[0195] The event conduction path with the largest event conduction path weight value is selected as the main conduction path;

[0196] The method for constructing the artifact strategy mapping model and outputting the correction rules for the sampling device is as follows:

[0197] Based on the main conduction path, an artifact strategy mapping model is constructed to output correction rules for the main conduction path sampling device;

[0198] Those skilled in the art will appreciate that the causal chain structure of the primary conduction pathway is analyzed and key node features on the chain are extracted, including: device parameter deviation (e.g., sampling rate 30% lower than the standard value), type, artifact attributes in the frequency domain interval, and neural abnormality type;

[0199] Based on the historical correction case library, samples that fully match the main conduction pathway nodes are screened, and data such as parameter correction amount, filtering strategy, and correction signal improvement are extracted. A mapping model is constructed with the main conduction pathway characteristics as input and the optimal correction strategy as output.

[0200] The mapping model uses decision trees or association rule mining to learn the association patterns between node features and correction schemes. When the feature vector of the real-time main conduction pathway is input (such as a sampling rate deviation of 25%, high-frequency non-physiological artifacts, and SDNN overlimit of 20%), the model outputs targeted correction rules (such as increasing the sampling rate from 200Hz to 300Hz and adjusting the low-pass filter cutoff frequency from 60Hz to 80Hz) through feature matching and rule reasoning. At the same time, combined with the high-weight attributes of the main conduction pathway, the rules prioritize the most core abnormal conduction links, ultimately forming instructions that can directly drive device parameter corrections.

[0201] Example 3

[0202] like Figure 3 As shown, the intelligent fetal growth and development detection system also includes the following modules:

[0203] Fetal movement analysis module: used to obtain mixed physiological signals and movement signals, perform fetal movement interference analysis on bioelectric signals and movement data, and obtain fetal movement interference coefficient and interference peak period;

[0204] Artifact separation module: used to extract the monitoring features of the fetal movement interference sequence within the interference peak period, establish the artifact monitoring vector for signal deterioration analysis, trigger the artifact separation algorithm based on the deterioration analysis results, perform artifact separation on the mixed physiological signal, and output the neural feature signal and artifact signal;

[0205] Degradation analysis module: used to distinguish and process artifact signals to obtain physiological artifact signals and non-physiological artifact signals, capture neural response abnormalities, and correlate non-physiological artifact signals with neural response abnormalities to extract signal degradation events;

[0206] Equipment correction module: If there is a signal degradation event, it is used to analyze the transmission relationship between the abnormal event and the current equipment sampling parameters, obtain the event transmission path and build an artifact strategy mapping model, and output the correction rules for the sampling equipment.

[0207] The above is a detailed description of an embodiment of the present invention. However, the content is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. An intelligent fetal growth and development detection method, characterized by: The steps include: Acquire mixed physiological signals and motion signals, perform fetal movement interference analysis on the bioelectrical signals and motion data, and obtain the fetal movement interference coefficient and interference peak period; Extract the monitoring features of the fetal movement interference sequence within the interference peak period, establish an artifact monitoring vector for signal deterioration analysis, trigger the artifact separation algorithm based on the deterioration analysis results, perform artifact separation on the mixed physiological signal, and output neural feature signals and artifact signals; Differentiate and process artifact signals to obtain physiological artifact signals and non-physiological artifact signals, capture abnormal neural responses, and correlate non-physiological artifact signals with abnormal neural responses to extract signal degradation events; If there is a signal degradation event, it is used to analyze the transmission relationship between the abnormal event and the current device sampling parameters, obtain the event transmission path and build an artifact strategy mapping model, and output the correction rules for the sampling device.

2. The intelligent fetal growth and development detection method according to claim 1, characterized in that: The method of performing fetal movement interference analysis on the bioelectrical signal and movement data is as follows: Obtaining the fetal movement interference coefficients in all sliding windows, and constructing a fetal movement interference sequence including the fetal movement interference coefficients in all sliding windows; Based on the fetal movement interference sequence, the gradient of the fetal movement interference coefficient in adjacent sliding windows is calculated using the local maximum monitoring algorithm. Obtain the time when the gradient of the fetal movement interference coefficient changes from positive to negative as the interference peak moment; Obtain the interference peak moment of the fetal movement event, calculate the interval time between the interference peak moments of adjacent fetal movement events, and use the interval time as the interference peak period.

3. The intelligent fetal growth and development detection method according to claim 2, characterized in that: The method for obtaining the fetal movement interference coefficient is: Collect fetal ECG signals, maternal ECG signals, and myoelectric signals during the monitoring period as mixed physiological signals; The acceleration of the body surface movement caused by the fetal movement is collected by an acceleration sensor to obtain a fetal movement signal; Calculate the mutual correlation coefficients between different types of mixed physiological signals and fetal movement signals within the sliding window; Obtain the cross-correlation coefficients of the three types of mixed physiological signals and fetal movement signals under different sliding windows, and draw the cross-correlation curves of the three types of mixed physiological signals; From each cross-correlation curve, obtain the time delay corresponding to the point with the maximum cross-correlation coefficient; Based on the time delay corresponding to the maximum value of the cross-correlation coefficient, a numerical analysis was performed on the variance of the cross-correlation coefficients of different types of mixed physiological signals and the variance of the fetal movement signal to obtain the fetal movement interference coefficient.

4. The intelligent fetal growth and development detection method according to claim 1, characterized in that: The method for separating artifacts from mixed physiological signals is: Extract the monitoring features of the fetal movement interference sequence within the interference peak period to establish an artifact monitoring vector, perform signal deterioration analysis on the artifact monitoring vector, and obtain the signal deterioration probability; If the signal deterioration probability is higher than or equal to the preset signal deterioration threshold, an artifact separation model is constructed to perform artifact separation on the mixed physiological signal and output neural feature signals and artifact signals.

5. The intelligent fetal growth and development detection method according to claim 4, characterized in that: The signal degradation analysis method for the artifact monitoring vector is as follows: Obtain the mean value of the fetal movement interference coefficient within the interference peak period, as well as the trend value and the maximum value of the mutual correlation coefficient of the fetal movement interference coefficient in the fetal movement interference sequence; Extract the standard deviation of fetal heart rate variability from mixed physiological signals; Construct an artifact monitoring vector including the mean value of the fetal movement interference coefficient within the interference peak period, the trend value of the fetal movement interference coefficient in the fetal movement interference sequence, the maximum value of the mutual correlation coefficient, and the standard deviation of the fetal heart rate variability; The artifact monitoring vector is input into the binary classifier algorithm, which outputs the probability of signal deterioration in the future sliding window.

6. The intelligent fetal growth and development detection method according to claim 4, characterized in that: The artifact separation model is constructed as follows: Determining an artifact separation mode based on the probability of signal degradation; Generate periodic energy coding based on interference peak period to determine artifact separation direction; Determining different artifact separation algorithms based on the artifact separation mode; A multi-loss function optimization model is established based on the artifact monitoring vector to achieve artifact separation.

7. The intelligent fetal growth and development detection method according to claim 1, characterized in that: The method of associating the non-physiological artifact signal with the abnormal neural response is as follows: Extract HRV sequence from neural feature signals and obtain time domain and frequency domain features of HRV sequence; Based on the time and frequency domain characteristics of the HRV sequence, a Boolean decision function is constructed to determine abnormal neural responses. If the neural response is abnormal, the associated time window and neural response window are set; If non-physiological artifacts are detected within the artifact monitoring window and there are neural abnormalities within the time window, a signal degradation event is determined to have occurred.

8. The intelligent fetal growth and development detection method according to claim 1, characterized in that: The method for distinguishing and processing artifact signals is as follows: The cross-correlation coefficient between the artifact signal and the fetal movement signal within the sliding window, the dominant frequency of the artifact signal in the frequency domain within different sliding windows, and the electrode impedance fluctuation characteristics of the acquisition device within the sliding window are obtained for spatiotemporal alignment processing, and the sliding window corresponding to the fetal movement interference sequence is correspondingly spliced ​​to establish a signal discrimination vector; The signal discrimination vector is input into the classifier, and the classifier distinguishes the artifact signal to obtain physiological artifact signals and non-physiological artifact signals.

9. The intelligent fetal growth and development detection method according to claim 1, characterized in that: The correction rules for the sampling device are output in the following manner: Analyze the degree of parameter deviation based on the event conduction path and calculate the event conduction path weight; The event conduction path with the largest event conduction path weight value is selected as the main conduction path; Based on the main conduction path, an artifact strategy mapping model is constructed to output correction rules for the main conduction path sampling device.

10. An intelligent fetal growth and development detection system for implementing any one of the intelligent fetal growth and development detection methods according to claims 1-9, characterized in that: Includes the following modules: Fetal movement analysis module: used to obtain mixed physiological signals and movement signals, perform fetal movement interference analysis on bioelectric signals and movement data, and obtain fetal movement interference coefficient and interference peak period; Artifact separation module: used to extract the monitoring features of the fetal movement interference sequence within the interference peak period, establish the artifact monitoring vector for signal deterioration analysis, trigger the artifact separation algorithm based on the deterioration analysis results, perform artifact separation on the mixed physiological signal, and output the neural feature signal and artifact signal; Degradation analysis module: used to distinguish and process artifact signals to obtain physiological artifact signals and non-physiological artifact signals, capture neural response abnormalities, and correlate non-physiological artifact signals with neural response abnormalities to extract signal degradation events; Equipment correction module: If there is a signal degradation event, it is used to analyze the transmission relationship between the abnormal event and the current equipment sampling parameters, obtain the event transmission path and build an artifact strategy mapping model, and output the correction rules for the sampling equipment.

Images