A noise optimization processing method for fetal heart rate signal

Abstract

The present invention provides a noise optimization processing method for fetal heart signal, which relates to the field of signal processing technology, including: dual-channel acquisition of maternal electrocardiogram signals and fetal electrocardiogram signals, obtaining maternal electrocardiogram signal sequences and fetal electrocardiogram signal sequences, and acquiring maternal characteristic information of the mother; respectively predicting maternal electrocardiogram noise according to multiple characteristic information, and obtaining multiple maternal electrocardiogram noise sequences; acquiring maternal activity image sequences, respectively predicting myoelectric noise signals according to multiple characteristic information, and obtaining multiple maternal myoelectric noise sequences; performing noise optimization processing on the fetal electrocardiogram signal sequence, obtaining the optimal noise-filtered fetal electrocardiogram signal sequence, performing random noise filtering on the fetal electrocardiogram signal sequence, calculating the similarity between the noise signal sequence and multiple maternal electrocardiogram noise sequences and multiple maternal myoelectric noise sequences, and optimizing. The present invention solves the technical problem of poor noise processing effect of fetal heart signal in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a noise optimization processing method for fetal heart rate signals. Background Art

[0002] During the acquisition of fetal heart rate signals (FECG), they are often interfered by maternal electrocardiogram signals (MECG) and electromyogram signals (EMG). Since the maternal ECG and fetal heart rate signals have similar frequencies and the maternal electromyogram signals are non-stationary and random, conventional filtering methods are difficult to accurately separate the fetal heart rate signals. The maternal ECG signals often mask the characteristics of the fetal heart rate signals due to their large amplitude, while the maternal electromyogram signals are derived from muscle activities such as the abdomen, have high frequencies and vary with the intensity of maternal activities, and are not easy to process using fixed threshold filtering.

[0003] The prior art has the technical problem of poor fetal heart signal noise processing effect. Summary of the invention

[0004] The present application provides a noise optimization processing method for fetal heart signal, which is used to solve the technical problem of poor noise processing effect of fetal heart signal in the prior art. In view of the above problem, the present application provides a noise optimization processing method for fetal heart signal.

[0005] In a first aspect, the present application provides a noise optimization processing method for a fetal heart signal, the method comprising:

[0006] Performing dual-channel acquisition of maternal ECG signals and fetal ECG signals to obtain maternal ECG signal sequences and fetal ECG signal sequences, and acquiring maternal characteristic information of the mother, wherein the maternal characteristic information includes characteristic information of multiple characteristic categories;

[0007] According to the plurality of feature information, respectively, maternal ECG noise is predicted according to the maternal ECG signal sequence to obtain a plurality of maternal ECG noise sequences;

[0008] Collecting a maternal activity image sequence, predicting the electromyographic noise signal according to a plurality of feature information, and obtaining a plurality of maternal electromyographic noise sequences;

[0009] The fetal ECG signal sequence is subjected to noise optimization processing to obtain an optimal noise-filtered fetal ECG signal sequence, wherein, during the noise optimization processing, the fetal ECG signal sequence is subjected to random noise filtering, and the similarity between the noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences is calculated for optimization.

[0010] In a second aspect, the present application provides a noise optimization processing system for fetal heart rate signals, the system comprising:

[0011] A signal acquisition module, used to perform dual-channel acquisition of maternal ECG signals and fetal ECG signals, obtain maternal ECG signal sequences and fetal ECG signal sequences, and acquire maternal characteristic information of the mother, wherein the maternal characteristic information includes characteristic information of multiple characteristic categories;

[0012] An ECG noise prediction module, used for predicting maternal ECG noise according to the maternal ECG signal sequence according to a plurality of feature information, to obtain a plurality of maternal ECG noise sequences;

[0013] The myoelectric noise prediction module is used to collect the maternal activity image sequence, and predict the myoelectric noise signal according to multiple feature information to obtain multiple maternal myoelectric noise sequences;

[0014] The noise optimization processing module is used to perform noise optimization processing on the fetal ECG signal sequence to obtain the optimal noise-filtered fetal ECG signal sequence, wherein, during the noise optimization processing, the fetal ECG signal sequence is randomly filtered, and the similarity between the noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences is calculated for optimization.

[0015] In a third aspect, the present invention further provides an electronic device, comprising:

[0016] At least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor so that the at least one processor can perform the steps of the method in the first aspect above.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0018] The embodiment of the present application provides a noise optimization processing method for fetal heart signal. By collecting maternal ECG signals and fetal ECG signals through dual channels, it can effectively distinguish the ECG signal sources of the mother and the fetus, and obtain multi-dimensional maternal feature information, and then perform multi-angle prediction of maternal ECG noise and myoelectric noise respectively. The feature information of multiple feature categories can assist in generating multiple noise sequences, realizing the accurate identification of maternal noise and multi-dimensional interference separation. This scheme collects the active image sequence of the mother and combines the maternal feature information to predict myoelectric noise, thereby generating multiple maternal myoelectric noise sequences, making the removal of myoelectric noise more accurate. Through random noise filtering and similarity analysis, the similarity between the noise signal sequence and the maternal noise sequence is evaluated, and the feature information is weighted by similarity to optimize the noise filtering process and output the optimal noise-filtered fetal ECG signal sequence. The scheme of the present application improves the effectiveness of removing noise interference in the fetal heart signal, thereby improving the accuracy, clarity and stability of fetal ECG signal extraction, and providing high-quality signal support for fetal heart signal monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic flow chart of a noise optimization processing method for fetal heart rate signals is provided for the present application.

[0020] Figure 2 A schematic diagram of the structure of a noise optimization processing system for fetal heart rate signals is provided for this application.

[0021] Figure 3 This is a schematic structural diagram of an electronic device provided by the present invention.

[0022] Explanation of reference numerals: signal acquisition module 11 , electrocardiographic noise prediction module 12 , electromyographic noise prediction module 13 , noise optimization processing module 14 , electronic device 500 , memory 510 , processor 520 , first computer program 511 . DETAILED DESCRIPTION

[0023] The present application provides a noise optimization processing method for fetal heart signal, which is used to solve the technical problem of poor fetal heart signal noise processing effect in the prior art.

[0024] Embodiment 1: Figure 1 As shown, the present application provides a noise optimization processing method for fetal heart signal, the method comprising:

[0025] S100: Perform dual-channel acquisition of maternal ECG signals and fetal ECG signals to obtain maternal ECG signal sequences and fetal ECG signal sequences, and acquire maternal characteristic information of the mother, wherein the maternal characteristic information includes characteristic information of multiple characteristic categories.

[0026] In the embodiment of the present application, firstly, dual-channel acquisition is performed on maternal electrocardiogram (MECG) and fetal electrocardiogram (FECG) to obtain independent maternal electrocardiogram sequence and fetal electrocardiogram sequence.

[0027] Dual-channel acquisition means placing sensors on different channels to collect the ECG signals of the mother and fetus respectively, which helps to distinguish the ECG activities from the mother and fetus and minimize the superposition interference of the two signals, thereby improving the acquisition accuracy of the fetal ECG signal.

[0028] Further collect various characteristic information of the mother to support subsequent noise prediction and processing.

[0029] By collecting maternal and fetal ECG signals through dual channels, it is possible to subsequently assist in distinguishing the maternal and fetal ECG signals, and perform maternal noise signal analysis based on maternal characteristic information, which can improve the accuracy of subsequent fetal ECG signal noise processing.

[0030] Step S100 in the method provided in the embodiment of the present application includes:

[0031] Perform dual-channel acquisition of maternal ECG signals and fetal ECG signals to obtain maternal ECG signal sequences and fetal ECG signal sequences;

[0032] According to multiple feature categories, multiple feature information of the mother is collected and integrated to obtain the mother feature information.

[0033] In an embodiment of the present application, firstly, a dual-channel acquisition is performed on the maternal ECG signal (MECG) and the fetal ECG signal (FECG), and the maternal ECG signal sequence and the fetal ECG signal sequence are obtained respectively. Among them, sensors are installed respectively using two independent signal channels, for example, ECG signal monitoring equipment is used respectively to collect the maternal and fetal ECG signals. In order to separate the maternal and fetal ECG signals. In this way, the maternal ECG signal can be used as a reference while acquiring the fetal ECG signal, thereby effectively reducing the noise in the fetal signal caused by maternal interference.

[0034] At the same time, the physiological characteristic information of the mother is collected according to multiple feature categories. Multiple feature categories include age, physical signs, heart rate and blood pressure. For example, multiple feature information of the mother includes parameters such as the mother's age of 28 years old, weight of 63kg, heart rate of 83 times / minute, blood pressure of systolic pressure of 115 mmHg, and diastolic pressure of 75 mmHg. These feature information will affect the noise effect of the mother on the fetal heart signal. For example, the weight and age of the mother affect the propagation of the ECG signal in the body, and then affect the noise of the maternal ECG signal on the fetal heart signal. In this way, multiple feature information of the mother is collected and obtained, and the maternal feature information is integrated to obtain a data set.

[0035] S200: Predicting maternal ECG noise according to the maternal ECG signal sequence according to the plurality of feature information to obtain a plurality of maternal ECG noise sequences.

[0036] In an embodiment of the present application, noise prediction is performed on the maternal ECG signal sequence based on multi-category feature information of the mother, and the noise signal formed by the maternal ECG signal sequence in the fetal heart signal is predicted to obtain multiple maternal ECG noise sequences.

[0037] Among them, the ECG signals of mothers with different characteristic information have different noise impact signals on the fetal heart signal. For example, the ECG signal transmission of a mother with a larger weight is weaker, and the noise impact signal on the fetal heart signal is weaker, and vice versa. Therefore, the noise signal is predicted in combination with the maternal characteristic information to improve the prediction accuracy of the noise signal formed by the maternal ECG signal sequence in the fetal heart signal.

[0038] Step S200 in the method provided in the embodiment of the present application includes:

[0039] According to the plurality of feature categories, training an ECG noise predictor including a plurality of ECG noise prediction branches;

[0040] The plurality of feature information and the maternal ECG signal sequence are respectively combined and input into a plurality of ECG noise prediction branches in the ECG noise predictor, and a plurality of maternal ECG noise sequences are obtained by prediction output.

[0041] In an embodiment of the present application, an ECG noise predictor including multiple ECG noise prediction branches is trained according to multiple feature categories to predict the noise signal formed by the maternal ECG signal sequence in the fetal heart signal according to the maternal characteristic information and the collected maternal ECG signal sequence.

[0042] In the embodiment of the present application, according to the multiple feature categories, training an ECG noise predictor including multiple ECG noise prediction branches includes:

[0043] According to the multiple feature categories, in the maternal ECG noise detection data within the historical time, a plurality of sample feature information sets, a sample maternal ECG signal sequence set, and a sample maternal ECG noise sequence set are respectively collected;

[0044] The plurality of sample feature information sets, the sample parent ECG signal sequence sets and the sample parent ECG noise sequence sets are respectively used to train a plurality of ECG noise prediction branches;

[0045] The multiple ECG noise prediction branches are combined to obtain an ECG noise predictor.

[0046] In an embodiment of the present application, an ECG noise predictor including multiple ECG noise prediction branches is trained according to multiple feature categories to accurately predict the maternal ECG noise signal formed in the fetal heart signal according to maternal characteristic information and the maternal ECG signal.

[0047] In an embodiment of the present application, according to the multiple feature categories, in the fetal heart signal detection data within the historical time, for example, in the fetal heart signal detection record database in the hospital, feature information of multiple feature categories of multiple mothers in the previous fetal heart detection process is collected and obtained to obtain multiple sample feature information sets. Then, a set of sample maternal ECG signal sequences of multiple mothers is collected, such as a maternal ECG signal sequence previously obtained by dual-channel signal collection. And, the noise formed in the fetal heart signal by the maternal ECG signal after fetal heart signal detection and artificial noise signal separation, that is, the maternal ECG noise, is collected to obtain a set of sample maternal ECG noise sequences.

[0048] Exemplarily, in the fetal heart signal previously detected, based on the R wave peak position, QRS complex and other ECG signal characteristics of the sample maternal ECG signal sequence, the signal characteristics of these maternal ECG signals are found in the previously monitored fetal heart signal sequence, and manual annotation is performed to obtain the noise of the maternal ECG signal in the fetal heart signal, that is, the maternal ECG noise, and obtain a set of sample maternal ECG noise sequences.

[0049] Furthermore, multiple sample feature information sets are used respectively, combined with a sample maternal ECG signal sequence set and a sample maternal ECG noise sequence set, to train multiple ECG noise prediction branches, so as to predict the maternal ECG noise signal formed in the fetal heart signal according to the sample feature information of different feature categories and combined with the maternal ECG signal.

[0050] Exemplarily, a first sample feature information set among multiple sample feature information sets is used, for example, a first sample feature information set whose feature category is maternal age. A sample maternal ECG signal sequence set is used as input training data, and a sample maternal ECG noise sequence set is used as output training data. Based on a feedforward neural network, a first ECG noise prediction branch among multiple ECG noise prediction branches is trained.

[0051] The first ECG noise prediction branch is trained by a feedforward neural network. For example, the input training data and the output training data are divided into a training set, a validation set and a test set, with a ratio of 80%, 10% and 10%. The training set is used for model learning, the validation set is used for parameter adjustment, and the test set is used to evaluate the generalization performance of the model. The first ECG noise prediction branch is constructed based on the feedforward neural network, including an input layer, an output layer and multiple hidden layers, and the hidden layer uses a ReLU activation function. Each group of input training data is input into the first ECG noise prediction branch, and the predicted maternal ECG noise sequence is output. The mean square error loss function is used to calculate the output loss in combination with the sample maternal ECG noise sequence in the corresponding output training data, and then the stochastic gradient descent (SGD) is used to adjust the network parameters in the first ECG noise prediction branch, such as weights, etc., so that the loss is reduced until it is less than the requirement (for example, the signal amplitude difference is less than 5%). In this way, the training set is used for multiple training until convergence, for example, the accuracy is greater than 95%, and then the validation set and the test set are used for verification and testing. If the accuracy meets the requirements, the training is completed, otherwise, the training continues.

[0052] In this way, the first ECG noise prediction branch is trained. Continue to use other multiple sample feature information sets, combine the sample mother ECG signal sequence set and the sample mother ECG noise sequence set, and train other multiple ECG noise prediction branches. The training process is the same as the first ECG noise prediction branch, and will not be repeated.

[0053] The multiple ECG noise prediction branches that have been trained are combined to obtain an ECG noise predictor. Each ECG noise prediction branch can predict the maternal ECG noise signal formed in the fetal heart signal based on a type of characteristic information of the mother combined with the maternal ECG signal.

[0054] Furthermore, the multiple characteristic information of the mother currently collected is combined with the maternal ECG signal sequence as multiple input data, and respectively input into the corresponding multiple ECG noise prediction branches in the ECG noise predictor, and the prediction output obtains multiple maternal ECG noise sequences, that is, the maternal ECG noise signal formed by the maternal ECG signal in the fetal heart signal.

[0055] By predicting the maternal ECG noise signal formed by the maternal ECG signal sequence in the fetal heart signal based on different feature information, the accuracy and comprehensiveness of the analysis of the impact of the maternal ECG signal on the fetal heart signal noise can be improved, thereby improving the accuracy of the fetal heart signal noise processing.

[0056] S300: collecting a maternal activity image sequence, and performing myoelectric noise signal prediction according to a plurality of feature information to obtain a plurality of maternal myoelectric noise sequences;

[0057] In the embodiment of the present application, while collecting the maternal ECG signal and the fetal ECG signal, the maternal activity image sequence is collected. The maternal activity will cause the maternal myoelectric signal to change, and then form different maternal myoelectric noise signals in the fetal heart signal. Therefore, based on the maternal activity image sequence, the noise formed by the maternal myoelectric signal in the fetal heart signal can be identified and predicted.

[0058] The same principle as step S200, the characteristic information of the mother will affect the noise formed by the mother's electromyographic signal in the fetal heart signal. Therefore, combining multiple characteristic information, the electromyographic noise signal formed by the mother's electromyographic signal in the fetal heart signal is predicted according to the activity image sequence of the mother, and multiple maternal electromyographic noise sequences are obtained. In this way, the noise filtering accuracy caused by the mother's electromyographic signal in the fetal heart signal can be improved in the subsequent fetal heart signal noise processing process.

[0059] Step S300 in the method provided in the embodiment of the present application includes:

[0060] Collecting a sequence of maternal moving images;

[0061] According to the multiple feature categories, in the maternal electromyographic noise detection data within the historical time, a plurality of sample feature information sets, a sample activity image sequence set, and a sample maternal electromyographic noise sequence set are collected respectively;

[0062] Using the multiple sample feature information sets, sample activity image sequence sets and sample parent electrocardiogram noise sequence sets, based on a convolutional neural network, multiple electromyographic noise prediction branches are trained to obtain an electromyographic noise predictor in combination;

[0063] The plurality of feature information and the activity image sequence are combined and respectively input into a plurality of myoelectric noise prediction branches in the myoelectric noise predictor, and a plurality of myoelectric noise sequences are obtained by prediction output.

[0064] In the embodiment of the present application, the myoelectric noise signal is predicted by collecting a sequence of maternal activity images, so as to more accurately identify the impact of maternal myoelectric interference on the fetal heart signal.

[0065] Collect the mother's active image sequence, use the bed's camera or other visual sensors to record the mother's body movement status in real time, including information such as body position, movement amplitude and posture changes, and obtain the active image sequence.

[0066] According to the historical data of previous fetal heart signal detection, the maternal feature information of the previous fetal heart signal detection of multiple maternal bodies, that is, the feature information of multiple feature categories, is collected to obtain multiple sample feature information sets of multiple feature categories. Further, the activity image sequences collected when the previous multiple maternal bodies performed fetal heart signals are collected to obtain a sample activity image sequence set. And the noise formed by the maternal electromyographic signal in the fetal heart signal sequence obtained by the detection is collected and separated to obtain a sample maternal electromyographic noise sequence set.

[0067] Exemplarily, in the fetal heart signal previously detected, the mother's electromyographic signal is collected, and based on the high frequency in the mother's electromyographic signal, the noise signal with obvious high-frequency and random fluctuation parts is separated in the fetal heart signal. These parts are most likely maternal electromyographic noise, and then manually labeled as maternal electromyographic noise sequence to obtain a set of sample maternal electromyographic noise sequences.

[0068] Furthermore, multiple sample feature information sets and sample activity image sequence sets are used as input training data, and a sample maternal electrocardiogram noise sequence set is used as output training data. A convolutional neural network is used to train multiple electromyographic noise prediction branches that predict the electromyographic noise formed by the maternal electromyographic signal in the fetal heart signal based on the feature information and the activity image sequence. The training process of the multiple electromyographic noise prediction branches and the multiple electrocardiogram noise prediction branches are the same, but the training data are different.

[0069] Exemplarily, a convolutional neural network is used to train the EMG noise prediction branch. The EMG noise prediction branch is constructed based on the convolutional neural network, including an input layer, multiple convolutional layers, multiple pooling layers, a fully connected layer, and an output layer. The convolutional layer and the pooling layer are used to extract image features in the active image sequence, and the fully connected layer uses the ReLU activation function. Each set of input training data is input into the EMG noise prediction branch, and the predicted parent EMG noise sequence is output. The mean square error loss function is used to calculate the output loss in combination with the sample parent EMG noise sequence in the corresponding output training data, and then the stochastic gradient descent (SGD) is used to adjust the network parameters in the EMG noise prediction branch, such as weights, etc., so that the loss is reduced until it is less than the requirement (for example, the frequency difference of the noise signal is less than 5%). In this way, multiple rounds of training are performed until convergence, for example, the accuracy rate is greater than 95%.

[0070] The multiple EMG noise prediction branches that have been trained are combined to obtain an EMG noise predictor.

[0071] The multiple feature information currently collected is combined with the active image sequence to obtain multiple groups of input data, which are respectively input into multiple myoelectric noise prediction branches in the myoelectric noise predictor, and multiple myoelectric noise sequences are obtained by prediction output. That is, the myoelectric noise signal formed by the myoelectric signal under the multiple feature information of the mother and the active image is formed in the fetal heart signal.

[0072] The multiple myoelectric noise sequences finally obtained will be used to optimize the fetal heart signal processing, so as to more effectively filter out the myoelectric noise and realize the clear extraction of the fetal heart signal.

[0073] S400: performing noise optimization processing on the fetal ECG signal sequence to obtain an optimal noise-filtered fetal ECG signal sequence, wherein, during the noise optimization processing, the fetal ECG signal sequence is randomly noise-filtered, and the similarity between the noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences is calculated for optimization.

[0074] In the embodiment of the present application, in order to obtain the optimal noise-filtered fetal ECG signal sequence, the fetal ECG signal sequence is subjected to multi-stage noise optimization processing. The processing obtains a noise signal sequence by random noise filtering, calculates the similarity between the noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences, optimizes and obtains the noise signal sequence with the greatest similarity, and then effectively separates the maternal ECG and myoelectric noise, thereby improving the clarity and accuracy of the fetal ECG signal.

[0075] Step S400 in the method provided in the embodiment of the present application includes:

[0076] Performing random noise filtering on the fetal electrocardiogram signal sequence to obtain a first noise filtered fetal electrocardiogram signal sequence and a first noise signal sequence;

[0077] Analyze and calculate the similarity between the first noise signal sequence and the multiple maternal electrocardiogram noise sequences and the multiple maternal myoelectric noise sequences to obtain a first noise filtering similarity;

[0078] The fetal electrocardiogram signal sequence is subjected to noise optimization processing until the optimization processing converges, and a noise-filtered fetal electrocardiogram signal sequence with the maximum noise-filtered similarity is output to obtain an optimal noise-filtered fetal electrocardiogram signal sequence.

[0079] In an embodiment of the present application, in order to obtain a clear fetal ECG signal sequence, the original fetal ECG signal sequence is subjected to random noise filtering. Random noise filtering refers to randomly filtering the fetal ECG signal sequence through random different filtering parameters (such as the bandwidth of the bandpass filter, the cutoff frequency of the low-pass filter, etc.) to generate random noise filtering results in the noise filtering process. After the noise filtering is completed, a first noise-filtered fetal ECG signal sequence and a first noise signal sequence are obtained, the first noise-filtered fetal ECG signal sequence includes the fetal ECG signal sequence after the noise is filtered out, and the first noise signal sequence includes the noise signal removed from the fetal ECG signal.

[0080] Since maternal ECG noise and myoelectric noise cannot be directly and automatically filtered out, random noise filtering is used here to achieve automated fetal heart signal noise processing on the one hand, and to improve the accuracy of noise filtering through similarity analysis and optimization between the noise signal sequence and the maternal ECG noise sequence and maternal myoelectric noise sequence on the other hand.

[0081] Furthermore, the similarity between the first noise signal sequence and multiple maternal ECG noise sequences and multiple maternal myoelectric noise sequences is analyzed and calculated, and the first noise filtering similarity is calculated. Among them, the greater the similarity between the first noise signal sequence and multiple maternal ECG noise sequences and multiple maternal myoelectric noise sequences, the better the current random noise filtering effect on the fetal ECG signal sequence, and the closer the filtered noise signal is to the noise signal formed by the maternal ECG signal and the myoelectric signal in the fetal heart signal. In this way, automated fetal heart signal noise filtering can be achieved and the accuracy of noise processing can be improved.

[0082] Similarity calculation can use measurement methods such as cosine similarity or Pearson correlation coefficient to compare the similarity between signal sequences. The higher the value, the closer the characteristics of the two are.

[0083] In the embodiment of the present application, analyzing and calculating the similarity between the first noise signal sequence and the multiple maternal electrocardiogram noise sequences and the multiple maternal myoelectric noise sequences to obtain the first noise filtering similarity includes:

[0084] Respectively analyzing the similarities between the first noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences to obtain multiple first ECG noise similarities and multiple first myoelectric noise similarities;

[0085] Acquire multiple feature information sets of the multiple feature categories, and assign multiple feature weights according to the appearance ratios of the multiple feature information in the multiple feature information sets;

[0086] The plurality of feature weights are used to perform weighted calculation on the plurality of first electrocardiogram noise similarities and the plurality of first electromyography noise similarities to obtain a first weighted electrocardiogram noise similarity and a first weighted electromyography noise similarity, and the mean is calculated to obtain a first noise-filtered similarity.

[0087] In the embodiment of the present application, firstly, the similarities between the first noise signal sequence and multiple maternal ECG noise sequences and multiple maternal EMG noise sequences are analyzed respectively to obtain multiple first ECG noise similarities and multiple first EMG noise similarities.

[0088] In the embodiment of the present application, machine learning is used to analyze the similarity between the first noise signal sequence and multiple maternal electrocardiographic noise sequences and multiple maternal myoelectric noise sequences.

[0089] In the embodiment of the present application, the similarities between the first noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences are analyzed respectively to obtain multiple first ECG noise similarities and multiple first myoelectric noise similarities, including:

[0090] Collecting a set of sample signal sequence combinations, and marking the similarity of two sample signal sequences in each sample signal sequence combination to obtain a sample similarity set;

[0091] Using the sample signal sequence combination set and the sample similarity set to train a noise signal similarity identifier;

[0092] The first noise signal sequence is combined with the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences respectively, and input into the noise signal similarity identifier, and output to obtain multiple first ECG noise similarities and multiple first myoelectric noise similarities.

[0093] In the embodiment of the present application, based on the historical data of fetal heart signal processing in the previous time, a set of sample signal sequence combinations is collected, and each sample signal sequence combination includes a noise signal sequence in the fetal heart signal sequence and a maternal ECG noise sequence or a maternal myoelectric noise sequence. The noise signal sequence, the maternal ECG noise sequence or the maternal myoelectric noise sequence can be obtained by processing according to the method in the aforementioned content, or by signal acquisition and separation processing.

[0094] The two sample signal sequences in each sample signal sequence combination are annotated with similarity, and the two signals in each sample signal sequence combination are annotated with similarity. Exemplarily, the similarity of the two sample signal sequences is calculated by mathematical metrics (such as cosine similarity or Pearson correlation coefficient), and the similarity value is added to the sample combination to form a sample similarity set. Alternatively, the signal frequency similarity, waveform similarity or peak similarity is calculated for the two sample signal sequences in each sample signal sequence combination as sample similarity to obtain a sample similarity set. For example, the ratio of the frequency of a signal sequence with a smaller frequency to the frequency of another signal sequence is calculated as signal frequency similarity, and then as sample similarity to obtain a sample similarity set.

[0095] Further, the sample signal sequence combination set is used as input training data, the sample similarity set is used as output training data, and machine learning is used to train the noise signal similarity identifier. Exemplarily, a feedforward neural network is used to train the noise signal similarity identifier, and the training process is the same as the training process of the electrostatic noise prediction branch in the above content.

[0096] Based on the trained noise signal similarity identifier, the current first noise signal sequence is combined with multiple maternal ECG noise sequences and multiple maternal EMG noise sequences respectively, and the multiple combinations are input into the noise signal similarity identifier, and the similarities between the first noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal EMG noise sequences are obtained through identification output, that is, multiple first ECG noise similarities and multiple first EMG noise similarities.

[0097] Furthermore, multiple sample feature information sets of multiple feature categories are obtained. For example, all feature information of multiple feature categories of all mothers is obtained based on a database of all mothers in the hospital to obtain multiple feature information sets.

[0098] Using multiple feature information of the current mother, the same feature information is retrieved in multiple feature information sets to obtain the appearance ratio of multiple feature information in multiple feature information sets. For example, the appearance ratio of the current mother's weight of 62 kg in the feature information set of the weight feature category is 5%. For example, the appearance ratio of multiple other feature information is 4%, 8%, etc. The larger the appearance ratio, the greater the similarity between a certain feature information of the current mother and other mothers, the stronger the generalization, and the more accurate the maternal ECG noise sequence and maternal myoelectric noise sequence predicted based on the feature information.

[0099] Furthermore, according to the multiple occurrence ratios of the multiple feature information in the multiple feature information sets, multiple feature weights of the current mother are allocated, for example, the ratio of each occurrence ratio to the sum of the multiple occurrence ratios is calculated as the feature weight, and multiple feature weights are obtained. For example, 5% / 5%+4%+8%=0.29 is calculated as the feature weight of the current mother's weight feature information.

[0100] A plurality of first ECG noise similarities are weightedly calculated using a plurality of feature weights to obtain a first weighted ECG noise similarity. The first weighted ECG noise similarity reflects the comprehensive similarity between the first noise signal sequence and the plurality of maternal ECG noise sequences, wherein the greater the proportion of feature information appearance and the higher the accuracy of the maternal ECG noise sequence, the greater the proportion of ECG noise similarity.

[0101] And, using multiple feature weights, multiple first electromyographic noise similarities are weightedly calculated to obtain a first weighted electromyographic noise similarity. The first weighted electromyographic noise similarity reflects the comprehensive similarity between the first noise signal sequence and multiple maternal electromyographic noise sequences. Among them, the greater the proportion of feature information occurrence and the higher the accuracy of the maternal electromyographic noise sequence, the greater the proportion of electromyographic noise similarity.

[0102] Further, the mean of the first weighted ECG noise similarity and the first weighted myoelectric noise similarity is calculated to obtain the first noise filtering similarity. The first noise filtering similarity reflects the similarity between the first noise signal sequence and multiple maternal ECG noise sequences and multiple maternal myoelectric noise sequences. The greater the first noise filtering similarity, the more similar the current random noise filtering first noise signal sequence is to the noise signal caused by the maternal ECG signal and the myoelectric signal to the fetal heart signal, and the better the noise processing effect of the first noise filtering fetal ECG signal sequence.

[0103] In the embodiment of the present application, based on the processing method of the first noise-filtered fetal ECG signal sequence, the first noise signal sequence and the first noise-filtered similarity, the noise optimization processing of the fetal ECG signal sequence is continued. For example, random noise filtering is continued to obtain a second noise-filtered fetal ECG signal sequence and a second noise signal sequence, and the second noise-filtered similarity is obtained by processing.

[0104] The noise optimization process is continued until the optimization process converges, for example, a preset number of optimization processes, for example, 100 times, is reached and the optimization converges. The noise-filtered fetal ECG signal sequence with the largest noise-filtered similarity during the optimization process is output as the optimal noise-filtered fetal ECG signal sequence.

[0105] The present application provides a noise optimization processing method for fetal heart rate signals, which has the following technical effects:

[0106] The embodiment of the present application provides a noise optimization processing method for fetal heart signal. By collecting maternal ECG signals and fetal ECG signals through dual channels, it can effectively distinguish the ECG signal sources of the mother and the fetus, and obtain multi-dimensional maternal feature information, and then perform multi-angle prediction of maternal ECG noise and myoelectric noise respectively. The feature information of multiple feature categories can assist in generating multiple noise sequences, realizing the accurate identification of maternal noise and multi-dimensional interference separation. This scheme collects the active image sequence of the mother and combines the maternal feature information to predict myoelectric noise, thereby generating multiple maternal myoelectric noise sequences, making the removal of myoelectric noise more accurate. Through random noise filtering and similarity analysis, the similarity between the noise signal sequence and the maternal noise sequence is evaluated, and the feature information is weighted by similarity to optimize the noise filtering process and output the optimal noise-filtered fetal ECG signal sequence. The scheme of the present application improves the effectiveness of removing noise interference in the fetal heart signal, thereby improving the accuracy, clarity and stability of fetal ECG signal extraction, and providing high-quality signal support for fetal heart signal monitoring.

[0107] Embodiment 2: Based on the same inventive concept as the noise optimization processing method for fetal heart signal in the above embodiment, Figure 2 As shown, the present application provides a noise optimization processing system for fetal heart signal, the system comprising:

[0108] The signal acquisition module 11 is used to perform dual-channel acquisition of maternal ECG signals and fetal ECG signals, obtain maternal ECG signal sequences and fetal ECG signal sequences, and acquire maternal characteristic information of the mother, wherein the maternal characteristic information includes characteristic information of multiple characteristic categories;

[0109] The ECG noise prediction module 12 is used to predict the maternal ECG noise according to the maternal ECG signal sequence according to the plurality of feature information, and obtain a plurality of maternal ECG noise sequences;

[0110] The myoelectric noise prediction module 13 is used to collect the maternal activity image sequence, and perform myoelectric noise signal prediction according to multiple feature information to obtain multiple maternal myoelectric noise sequences;

[0111] The noise optimization processing module 14 is used to perform noise optimization processing on the fetal ECG signal sequence to obtain the optimal noise-filtered fetal ECG signal sequence. During the noise optimization processing, the fetal ECG signal sequence is randomly noise-filtered, and the similarity between the noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences is calculated for optimization.

[0112] The signal acquisition module 11 is further configured to perform the following steps:

[0113] Perform dual-channel acquisition of maternal ECG signals and fetal ECG signals to obtain maternal ECG signal sequences and fetal ECG signal sequences;

[0114] According to multiple feature categories, multiple feature information of the mother is collected and integrated to obtain the mother feature information.

[0115] The ECG noise prediction module 12 is further configured to perform the following steps:

[0116] According to the plurality of feature categories, training an ECG noise predictor including a plurality of ECG noise prediction branches;

[0117] The plurality of feature information and the maternal ECG signal sequence are respectively combined and input into a plurality of ECG noise prediction branches in the ECG noise predictor, and a plurality of maternal ECG noise sequences are obtained by prediction output.

[0118] According to the multiple feature categories, training an ECG noise predictor including multiple ECG noise prediction branches includes:

[0119] According to the multiple feature categories, in the maternal ECG noise detection data within the historical time, a plurality of sample feature information sets, a sample maternal ECG signal sequence set, and a sample maternal ECG noise sequence set are respectively collected;

[0120] The plurality of sample feature information sets, the sample parent ECG signal sequence sets and the sample parent ECG noise sequence sets are respectively used to train a plurality of ECG noise prediction branches;

[0121] The multiple ECG noise prediction branches are combined to obtain an ECG noise predictor.

[0122] The myoelectric noise prediction module 13 is further configured to perform the following steps:

[0123] Collecting a sequence of maternal moving images;

[0124] According to the multiple feature categories, in the maternal electromyographic noise detection data within the historical time, a plurality of sample feature information sets, a sample activity image sequence set, and a sample maternal electromyographic noise sequence set are collected respectively;

[0125] Using the multiple sample feature information sets, sample activity image sequence sets and sample parent electrocardiogram noise sequence sets, based on a convolutional neural network, multiple electromyographic noise prediction branches are trained to obtain an electromyographic noise predictor in combination;

[0126] The plurality of feature information and the activity image sequence are combined and respectively input into a plurality of myoelectric noise prediction branches in the myoelectric noise predictor, and a plurality of myoelectric noise sequences are obtained by prediction output.

[0127] The noise optimization processing module 14 is further configured to perform the following steps:

[0128] Performing random noise filtering on the fetal electrocardiogram signal sequence to obtain a first noise filtered fetal electrocardiogram signal sequence and a first noise signal sequence;

[0129] Analyze and calculate the similarity between the first noise signal sequence and the multiple maternal electrocardiogram noise sequences and the multiple maternal myoelectric noise sequences to obtain a first noise filtering similarity;

[0130] The fetal electrocardiogram signal sequence is subjected to noise optimization processing until the optimization processing converges, and a noise-filtered fetal electrocardiogram signal sequence with the maximum noise-filtered similarity is output to obtain an optimal noise-filtered fetal electrocardiogram signal sequence.

[0131] The step of analyzing and calculating the similarity between the first noise signal sequence and the plurality of maternal electrocardiogram noise sequences and the plurality of maternal myoelectric noise sequences to obtain a first noise filtering similarity includes:

[0132] Respectively analyzing the similarities between the first noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences to obtain multiple first ECG noise similarities and multiple first myoelectric noise similarities;

[0133] Acquire multiple feature information sets of the multiple feature categories, and assign multiple feature weights according to the appearance ratios of the multiple feature information in the multiple feature information sets;

[0134] The plurality of feature weights are used to perform weighted calculation on the plurality of first electrocardiogram noise similarities and the plurality of first electromyography noise similarities to obtain a first weighted electrocardiogram noise similarity and a first weighted electromyography noise similarity, and the mean is calculated to obtain a first noise-filtered similarity.

[0135] The similarities between the first noise signal sequence and the multiple maternal electrocardiogram noise sequences and the multiple maternal electromyogram noise sequences are analyzed respectively to obtain multiple first electrocardiogram noise similarities and multiple first electromyogram noise similarities, including:

[0136] Collecting a set of sample signal sequence combinations, and marking the similarity of two sample signal sequences in each sample signal sequence combination to obtain a sample similarity set;

[0137] Using the sample signal sequence combination set and the sample similarity set to train a noise signal similarity identifier;

[0138] The first noise signal sequence is combined with the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences respectively, and input into the noise signal similarity identifier, and output to obtain multiple first ECG noise similarities and multiple first myoelectric noise similarities.

[0139] Through the above detailed description of a noise optimization processing method for fetal heart signal in this specification, those skilled in the art can clearly know the noise optimization processing method for fetal heart signal in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0140] For example 3, please refer to Figure 3 , Figure 3 Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 520, and a first computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the first computer program 511, the steps of the method in Embodiment 1 are implemented.

[0141] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A noise optimization processing method for fetal heart rate signal, characterized in that: The method comprises: Performing dual-channel acquisition of maternal ECG signals and fetal ECG signals to obtain maternal ECG signal sequences and fetal ECG signal sequences, and acquiring maternal characteristic information of the mother, wherein the maternal characteristic information includes characteristic information of multiple characteristic categories, and the multiple characteristic categories include age, physical signs, heart rate and blood pressure; According to the plurality of feature information, respectively, maternal ECG noise is predicted according to the maternal ECG signal sequence to obtain a plurality of maternal ECG noise sequences; Collecting a maternal activity image sequence, predicting the electromyographic noise signal according to a plurality of feature information, and obtaining a plurality of maternal electromyographic noise sequences; The fetal ECG signal sequence is subjected to noise optimization processing to obtain an optimal denoised fetal ECG signal sequence, wherein, during the noise optimization processing, the fetal ECG signal sequence is subjected to random denoising to obtain a denoised fetal ECG signal sequence and a noise signal sequence, and the similarity between the noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences is calculated and optimized, and the random denoising includes randomly filtering the fetal ECG signal sequence through random different filtering parameters.

2. The noise optimization processing method for fetal heart signal according to claim 1, characterized in that: Perform dual-channel acquisition of maternal ECG signals and fetal ECG signals to obtain maternal ECG signal sequences and fetal ECG signal sequences, and collect maternal characteristic information of the mother, including: Perform dual-channel acquisition of maternal ECG signals and fetal ECG signals to obtain maternal ECG signal sequences and fetal ECG signal sequences; According to multiple feature categories, multiple feature information of the mother is collected and integrated to obtain the mother feature information.

3. The noise optimization processing method for fetal heart signal according to claim 1, characterized in that: According to the plurality of feature information, maternal ECG noise is predicted according to the maternal ECG signal sequence to obtain a plurality of maternal ECG noise sequences, including: According to the plurality of feature categories, training an ECG noise predictor including a plurality of ECG noise prediction branches; The plurality of feature information and the maternal ECG signal sequence are respectively combined and input into a plurality of ECG noise prediction branches in the ECG noise predictor, and a plurality of maternal ECG noise sequences are obtained by prediction output.

4. The noise optimization processing method for fetal heart signal according to claim 3, characterized in that: According to the multiple feature categories, training an ECG noise predictor including multiple ECG noise prediction branches comprises: According to the multiple feature categories, in the maternal ECG noise detection data within the historical time, a plurality of sample feature information sets, a sample maternal ECG signal sequence set, and a sample maternal ECG noise sequence set are respectively collected; The plurality of sample feature information sets, the sample parent ECG signal sequence sets and the sample parent ECG noise sequence sets are respectively used to train a plurality of ECG noise prediction branches; The multiple ECG noise prediction branches are combined to obtain an ECG noise predictor.

5. The noise optimization processing method for fetal heart signal according to claim 1, characterized in that: The activity image sequence of the mother is collected, and the electromyographic noise signal is predicted according to multiple feature information to obtain multiple electromyographic noise sequences of the mother, including: Collecting a sequence of maternal moving images; According to the multiple feature categories, in the maternal electromyographic noise detection data within the historical time, a plurality of sample feature information sets, a sample activity image sequence set, and a sample maternal electromyographic noise sequence set are collected respectively; Using the multiple sample feature information sets, sample activity image sequence sets and sample parent electrocardiogram noise sequence sets, based on a convolutional neural network, multiple electromyographic noise prediction branches are trained to obtain an electromyographic noise predictor in combination; The plurality of feature information and the activity image sequence are combined and respectively input into a plurality of myoelectric noise prediction branches in the myoelectric noise predictor, and a plurality of myoelectric noise sequences are obtained by prediction output.

6. The noise optimization processing method for fetal heart signal according to claim 1, characterized in that: Performing noise optimization processing on the fetal electrocardiogram signal sequence to obtain an optimal noise-filtered fetal electrocardiogram signal sequence includes: Performing random noise filtering on the fetal electrocardiogram signal sequence to obtain a first noise filtered fetal electrocardiogram signal sequence and a first noise signal sequence; Analyze and calculate the similarity between the first noise signal sequence and the multiple maternal electrocardiogram noise sequences and the multiple maternal myoelectric noise sequences to obtain a first noise filtering similarity; The fetal electrocardiogram signal sequence is subjected to noise optimization processing until the optimization processing converges, and a noise-filtered fetal electrocardiogram signal sequence with the maximum noise-filtered similarity is output to obtain an optimal noise-filtered fetal electrocardiogram signal sequence.

7. The noise optimization processing method for fetal heart signal according to claim 6, characterized in that: Analyzing and calculating the similarity between the first noise signal sequence and the multiple maternal electrocardiogram noise sequences and the multiple maternal myoelectric noise sequences to obtain a first noise filtering similarity includes: Respectively analyzing the similarities between the first noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences to obtain multiple first ECG noise similarities and multiple first myoelectric noise similarities; Acquire multiple feature information sets of the multiple feature categories, and assign multiple feature weights according to the appearance ratios of the multiple feature information in the multiple feature information sets; The plurality of feature weights are used to perform weighted calculation on the plurality of first electrocardiogram noise similarities and the plurality of first electromyography noise similarities to obtain a first weighted electrocardiogram noise similarity and a first weighted electromyography noise similarity, and the mean is calculated to obtain a first noise-filtered similarity.

8. The noise optimization processing method for fetal heart signal according to claim 7, characterized in that: Respectively analyzing the similarities between the first noise signal sequence and the multiple maternal electrocardiogram noise sequences and the multiple maternal electromyogram noise sequences to obtain multiple first electrocardiogram noise similarities and multiple first electromyogram noise similarities, including: Collecting a set of sample signal sequence combinations, and marking the similarity of two sample signal sequences in each sample signal sequence combination to obtain a sample similarity set; Using the sample signal sequence combination set and the sample similarity set to train a noise signal similarity identifier; The first noise signal sequence is combined with the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences respectively, and input into the noise signal similarity identifier, and output to obtain multiple first ECG noise similarities and multiple first myoelectric noise similarities.

9. A noise optimization processing system for fetal heart rate signals, characterized in that: A method for optimizing the processing of noise for a fetal heart signal according to any one of claims 1 to 8, the system comprising: A signal acquisition module, used to perform dual-channel acquisition of maternal ECG signals and fetal ECG signals, obtain maternal ECG signal sequences and fetal ECG signal sequences, and acquire maternal characteristic information of the mother, wherein the maternal characteristic information includes characteristic information of multiple characteristic categories, and the multiple characteristic categories include age, physical signs, heart rate and blood pressure; An ECG noise prediction module, used for predicting maternal ECG noise according to the maternal ECG signal sequence according to a plurality of feature information, to obtain a plurality of maternal ECG noise sequences; The myoelectric noise prediction module is used to collect the maternal activity image sequence, and predict the myoelectric noise signal according to multiple feature information to obtain multiple maternal myoelectric noise sequences; The noise optimization processing module is used to perform noise optimization processing on the fetal ECG signal sequence to obtain the optimal noise-filtered fetal ECG signal sequence, wherein, during the noise optimization processing, the fetal ECG signal sequence is randomly denoised to obtain a noise-filtered fetal ECG signal sequence and a noise signal sequence, and the similarity between the noise signal sequence and the multiple maternal ECG noise sequences and the multiple maternal myoelectric noise sequences is calculated and optimized, and the random denoising includes randomly filtering the fetal ECG signal sequence through random different filtering parameters.

10. An electronic device, characterized in that: include: Memory for storing computer software programs; A processor is used to read and execute the computer software program, thereby implementing the steps of a noise optimization processing method for fetal heart signal as described in any one of claims 1 to 8.

Images