Electronic stethoscope signal acquisition system based on anomaly recognition

Abstract

The invention provides an electronic stethoscope signal acquisition system based on anomaly recognition, and relates to the technical field of data intelligent recognition, and the method comprises the steps: carrying out the anomaly loss analysis of a plurality of stethoscope signal data sets based on a signal twinning network, and generating a plurality of abnormal feature twinning databases; inputting the plurality of abnormal feature twin databases into a deep learning network for training to obtain a plurality of abnormal recognition networks; global training is carried out, and a global anomaly recognition network is constructed; performing anomaly recognition on the target signal flow based on a global anomaly recognition network, and outputting a target anomaly feature result; and performing signal clustering acquisition control based on a target abnormal feature result. The technical problems that due to the fact that the refining degree of signal data collection of the electronic stethoscope is low, the data analysis dimension and complexity are large, and the data recognition and analysis efficiency and accuracy are low can be solved, the fineness and accuracy of signal collection can be improved, and therefore the efficiency and accuracy of signal recognition and analysis are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data intelligent identification, and more particularly, to an electronic stethoscope signal collection system based on abnormality identification. BACKGROUND

[0002] A stethoscope is an instrument for detecting sound phenomena or the activity of a listening organ. The common stethoscope at present is a soft tube type stethoscope which can be used by both ears and is equipped with a bow piece. They all have an open bell or a component enclosed by a diaphragm as a listening head. In addition to the mechanical soft tube type stethoscope, electronic stethoscopes are also available on the market. The listening head of the stethoscope refers to a listening tube which can be placed on the body of a patient to be listened to. The listening tube is connected to a chest piece, and two bow pieces lead to earplugs from the chest piece.

[0003] For the soft tube type stethoscope, the sound signal is directly transmitted from the listening tube on the side of the listening object, through the chest piece and the two bow pieces, to the ears of the doctor being examined, while for the electronic stethoscope, the sound signal is received by a microphone installed in the listening head, converted into an electrical signal, transmitted to the earpiece on the side of the ear, and emitted after being amplified at that place.

[0004] The existing electronic stethoscope usually stores the collected signal data centrally when collecting signals, so that the signal data is complex and chaotic, and there may be a certain amount of abnormal data such as noise data and redundant data, resulting in a large signal data analysis dimension and complexity, which requires a large amount of time for data analysis, and ultimately causes low signal data recognition analysis efficiency and poor recognition quality. SUMMARY

[0005] Therefore, in order to solve the above technical problems, the technical scheme adopted by the embodiments of the present application is as follows: a signal acquisition system of an electronic stethoscope based on abnormality identification, comprising: a stethoscope signal dataset acquisition module, the stethoscope signal dataset acquisition module is used for acquiring a plurality of stethoscope signal datasets of a plurality of electronic stethoscopes, the plurality of stethoscope signal datasets include a plurality of normal stethoscope signal datasets, a plurality of abnormal stethoscope signal datasets and signal data abnormal factors, wherein the normal stethoscope signal data refers to signal data within a normal data range, wherein the normal data range can be set according to the actual scene and the body state; the abnormal stethoscope signal data refers to signal data not within the normal data range; the signal data abnormal factors include missing values, repeated values, inconsistent data, noise data and deviated data; an abnormal loss analysis module, the abnormal loss analysis module is used for performing abnormal loss analysis on the plurality of stethoscope signal datasets based on a signal twin network, generating a plurality of stethoscope signal abnormal feature twin databases, the abnormal feature twin database includes constructing a stethoscope signal feature classifier, performing feature classification on the plurality of stethoscope signal datasets through the stethoscope signal feature classifier, acquiring a plurality of stethoscope signal data feature information sets; performing feature label annotation on the plurality of stethoscope signal datasets based on the plurality of stethoscope signal data feature information sets, obtaining a plurality of stethoscope signal feature label data sets, clustering and dividing the plurality of stethoscope signal feature label data sets, determining a plurality of normal feature label signal data sets and a plurality of abnormal feature label signal data sets, performing abnormal loss analysis on the plurality of normal feature label signal data sets and the plurality of abnormal feature label signal data sets based on the signal twin network, generating the plurality of stethoscope signal abnormal feature twin databases, the abnormal feature twin database includes building the signal twin network, the signal twin network includes a normal twin subnetwork and an abnormal twin subnetwork, and the normal twin subnetwork and the abnormal twin subnetwork are shared weight networks, the plurality of normal feature label signal data sets and the plurality of abnormal feature label signal data sets are respectively input into the normal twin subnetwork and the abnormal twin subnetwork, a plurality of normal stethoscope feature sets and a plurality of abnormal stethoscope feature sets are output, feature loss analysis is performed on the plurality of normal stethoscope feature sets and the plurality of abnormal stethoscope feature sets, a plurality of stethoscope signal feature loss data sets are obtained, the plurality of stethoscope signal feature loss data sets are integrated and labeled to generate the plurality of stethoscope signal abnormal feature twin databases; a stethoscope signal abnormality identification network obtaining module, the stethoscope signal abnormality identification network obtaining module is used for inputting the plurality of stethoscope signal abnormal feature twin databases as local sample data into a deep learning network structure for identification training to obtain a plurality of stethoscope signal abnormality identification networks.The global stethoscope signal anomaly recognition network construction module is configured to encrypt and transmit model parameters of the plurality of stethoscope signal anomaly recognition networks to a cloud center platform for global training, and to construct a global stethoscope signal anomaly recognition network. The target stethoscope anomaly feature result output module is configured to monitor and acquire a target electronic stethoscope signal stream, perform anomaly recognition on the target electronic stethoscope signal stream based on the global stethoscope signal anomaly recognition network, and output a target stethoscope anomaly feature result. The signal clustering collection control module is configured to perform signal clustering collection control on the target electronic stethoscope signal stream based on the target stethoscope anomaly feature result.

[0006] Thanks to the above technical methods, the present application has the following technical progress compared with the prior art: The technical problems of low signal data collection refinement, large signal data analysis dimension and complexity, and low signal data recognition analysis efficiency and accuracy of the existing electronic stethoscope signal collection method can be solved. Firstly, a plurality of electronic stethoscope signal data sets are collected and acquired, including a plurality of normal stethoscope signal data sets and a plurality of abnormal stethoscope signal data sets. Then, the plurality of stethoscope signal data sets are analyzed for anomaly loss based on a signal twin network to generate a plurality of stethoscope signal anomaly feature twin databases. The plurality of stethoscope signal anomaly feature twin databases are input into a deep learning network structure as local sample data for recognition training to obtain a plurality of stethoscope signal anomaly recognition networks. The model parameters of the plurality of stethoscope signal anomaly recognition networks are encrypted and transmitted to a cloud center platform for global training to construct a global stethoscope signal anomaly recognition network. A target electronic stethoscope signal stream is monitored and acquired, and the target electronic stethoscope signal stream is recognized for anomaly based on the global stethoscope signal anomaly recognition network to output a target stethoscope anomaly feature result. Finally, the target electronic stethoscope signal stream is controlled for signal clustering collection based on the target stethoscope anomaly feature result. The above method can improve the accuracy and accuracy of electronic stethoscope signal collection, reduce the dimension and complexity of signal data analysis, facilitate the extraction of valuable information from signal data, and improve the efficiency and accuracy of signal data recognition analysis. BRIEF DESCRIPTION OF DRAWINGS

[0007] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the description of the embodiments will be briefly introduced.

[0008] Figure 1A flowchart of an electronic stethoscope signal acquisition method based on abnormality identification is provided for the present application. Figure 2 A flowchart of acquiring multiple stethoscope signal data sets of multiple electronic stethoscopes in an electronic stethoscope signal acquisition method based on abnormality identification is provided for the present application. Figure 3 A structural diagram of an electronic stethoscope signal acquisition system based on abnormality identification is provided for the present application.

[0009] Marked for illustration: Stethoscope signal data set acquisition module 01, abnormal loss analysis module 02, stethoscope signal abnormality identification network obtaining module 03, global stethoscope signal abnormality identification network construction module 04, target stethoscope abnormality feature result output module 05, signal clustering acquisition control module 06. DETAILED DESCRIPTION

[0010] The technical solutions in the embodiments of the present application will be clearly described below in combination with the drawings in the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts fall within the scope of protection of the present application. EMBODIMENT

[0011] Based on the above description, as Figure 1 indicated, the present application provides an electronic stethoscope signal acquisition system and method based on abnormality identification, which comprises: An electronic stethoscope is a medical device that uses electronic technology to amplify and process body sounds. Compared with traditional acoustic stethoscopes, electronic stethoscopes have higher sensitivity and more accurate sound transmission capability.

[0012] The method provided by the present application is used to optimize the existing electronic stethoscope signal acquisition method by combining the technical means of data abnormality identification, so as to improve the fineness and accuracy of electronic stethoscope signal acquisition, reduce the dimension and complexity of signal data analysis, and achieve the technical effects of improving the efficiency and accuracy of signal data identification analysis. The method is specifically implemented in an electronic stethoscope signal acquisition system based on abnormality identification.

[0013] Acquire multiple stethoscope signal data sets of multiple electronic stethoscopes, the multiple stethoscope signal data sets include multiple normal stethoscope signal data sets and multiple abnormal stethoscope signal data sets; In the embodiments of the present application, first, the stethoscope signal management system is connected to obtain a plurality of stethoscope signal data sets of a plurality of electronic stethoscopes, wherein the plurality of electronic stethoscopes refer to different types of electronic stethoscopes or electronic stethoscopes of different manufacturers of the same type, and the types of electronic stethoscopes include capacitive electronic stethoscopes, piezoelectric electronic stethoscopes, etc. The stethoscope signal data set is a historical stethoscope signal data set of the electronic stethoscope.

[0014] The plurality of stethoscope signal data sets include a plurality of normal stethoscope signal data sets and a plurality of abnormal stethoscope signal data sets, wherein the normal stethoscope signal data refers to signal data within a normal data range, and the normal data range can be set according to the actual scene and the body state, for example, the normal heart rate range is 60 to 100 times; and the abnormal stethoscope signal data refers to signal data not within the normal data range, for example, abnormal heart sound data, including tachycardia, bradycardia, dull heart sound, etc., which indicates that there may be a heart disease; abnormal breath sound data, including wheezing sound, phlegm sound, rapid breathing, etc., which indicates that there may be a respiratory disease.

[0015] By obtaining a plurality of stethoscope signal data sets, data support is provided for the next step of signal abnormal loss analysis.

[0016] As shown in Figure 2 In one embodiment, the method further comprises: obtaining signal data abnormal factors, including missing values, repeated values, inconsistent data, noise data, and deviated data; collecting a plurality of electronic stethoscope source signal data sets, identifying abnormal values in the plurality of electronic stethoscope source signal data sets based on the signal data abnormal factors, and obtaining abnormal stethoscope signal data information; determining an abnormal data correlation cleaning strategy according to the signal data abnormal factors; performing standardized processing on the abnormal stethoscope signal data information based on the abnormal data correlation cleaning strategy, and obtaining the plurality of stethoscope signal data sets.

[0017] In the embodiments of the present application, the method for collecting a plurality of stethoscope signal data sets of a plurality of electronic stethoscopes is as follows: first, obtaining signal data abnormal factors, including missing values, repeated values, inconsistent data, noise data, and deviated data, wherein the noise data refers to signal components mixed in normal signals that interfere with the identification and analysis of normal signals, such as environmental noise and electromagnetic interference; and the deviated data refers to data that exceeds the measurable threshold range, for example, the upper limit of the measurable threshold of heart rate is 200 times per minute, and heart rate data greater than 200 times per minute is deviated data.

[0018] First, connect the stethoscope signal management system, collect and obtain a plurality of electronic stethoscope source signal data sets, wherein the electronic stethoscope source signal data refers to the normal storage and data processing of the electronic stethoscope signal data. Then, according to the signal data abnormal factor, the abnormal value of the plurality of electronic stethoscope source signal data sets is identified, and the abnormal stethoscope signal data information is obtained, wherein the abnormal stethoscope signal data information refers to the electronic stethoscope source signal data satisfying the signal data abnormal factor.

[0019] In turn, each factor index in the signal data abnormal factor is analyzed by the abnormal data cleaning strategy. First, select the first factor index in the signal data abnormal factor, wherein the first factor index is any one of the missing value, repeated value, inconsistent data, noise data and deviation data. Then, based on the first factor index, the abnormal data cleaning strategy is analyzed, wherein the abnormal data cleaning strategy refers to the specific data processing method for the first factor index, for example: assuming that the first factor index is a missing value, the corresponding abnormal data cleaning strategy is to supplement the missing data; assuming that the first factor index is a repeated value, the corresponding abnormal data cleaning strategy is to delete the repeated value; assuming that the first factor index is noise data, the corresponding abnormal data cleaning strategy is data filtering processing. The person skilled in the art can set the abnormal data cleaning strategy according to the actual situation, and obtain the first abnormal data cleaning strategy corresponding to the first factor index.

[0020] Using the same method as obtaining the first abnormal data cleaning strategy, a plurality of abnormal data cleaning strategies corresponding to a plurality of factor indexes are obtained in turn, and then based on the mapping association of the factor index and the abnormal data cleaning strategy, the factor index is taken as a subnode, and the abnormal data cleaning strategy is taken as a leaf node of the subnode. The abnormal data association cleaning strategy is constructed, wherein the abnormal data association cleaning strategy contains the association standardization data processing mode corresponding to a plurality of factor indexes.

[0021] Based on the abnormal data association cleaning strategy, the cleaning strategy matching of the abnormal stethoscope signal data information is performed, and the abnormal stethoscope signal data information is standardized according to the cleaning strategy matching result, to obtain a plurality of stethoscope signal data sets. By constructing the abnormal data association cleaning strategy, the standardization processing of the abnormal stethoscope signal data can improve the accuracy and accuracy of the abnormal stethoscope signal data processing, thereby improving the accuracy of the stethoscope signal data set.

[0022] Based on the signal twin network, the abnormal loss analysis of the plurality of stethoscope signal data sets is performed, and a plurality of stethoscope signal abnormal feature twin databases are generated; In the embodiment of the present application, a signal twin network is constructed, and the plurality of stethoscope signal data sets are subjected to abnormal loss analysis through the signal twin network, wherein the abnormal loss analysis refers to obtaining loss features of abnormal signals, and a plurality of stethoscope signal abnormal feature twin databases are generated.

[0023] In one embodiment, the method further comprises: A stethoscope signal feature classifier is constructed, and the plurality of stethoscope signal data sets are subjected to feature classification through the stethoscope signal feature classifier, and a plurality of stethoscope signal data feature information sets are obtained. In the embodiment of the present application, the method for generating a plurality of stethoscope signal abnormal feature twin databases is as follows: first, a stethoscope signal feature classifier is constructed, and then the plurality of stethoscope signal data sets are subjected to feature classification through the stethoscope signal feature classifier, wherein the feature classification refers to dividing the plurality of stethoscope signal data sets into different types of feature data sets, wherein the attribute standards of different types of stethoscope signals and the abnormal feature distinguish different, and a plurality of stethoscope signal data feature information sets are obtained.

[0024] In one embodiment, the method further comprises: Obtaining stethoscope signal influencing factors, wherein the stethoscope signal influencing factors include physiological signal types, physiological feature information, and detection environment time periods. Each of the influencing factors in the stethoscope signal influencing factors is subjected to attribute content filling, and a signal influencing factor attribute content set is obtained. Each attribute content in the signal influencing factor attribute content set is sequentially encoded as a signal feature classification node, and node feature encoding information is determined. The signal influencing factor attribute content set is recursively constructed based on the node feature encoding information, and the stethoscope signal feature classifier is obtained.

[0025] In the embodiment of the present application, the method for constructing a stethoscope signal feature classifier is as follows: first, stethoscope signal influencing factors are obtained, wherein the stethoscope signal influencing factors include physiological signal types, physiological feature information, and detection environment time periods, wherein the physiological signal types include heartbeat, respiration, blood pressure, etc.; the physiological feature information includes age, gender, occupation, etc. of the detected person, and different physiological feature information persons correspond to different detection standards, for example, the detection standard of an athlete is more stringent than that of an ordinary person; wherein the detection environment time period refers to different scenes during detection, wherein different detection scenes correspond to different detection standards, for example, for the same person, the detection standard in motion is different from the detection standard during sleep.

[0026] The attribute content of each of the stethoscope signal influencing factors is filled respectively, where the attribute content filling refers to supplementing the specific attribute content corresponding to each influencing factor, for example, the attributes corresponding to the physiological signal type include heartbeat, respiration, blood pressure, etc. The attribute content can be set by the person skilled in the art according to the actual situation to obtain a signal influencing factor attribute content set, wherein each signal influencing factor contains one or more attribute contents.

[0027] Then, each attribute content in the signal influencing factor attribute content set is sequentially coded as a signal feature classification node, that is, each attribute content is taken as a classification node corresponding to a signal feature, and node feature coding information is determined, wherein each signal feature in the node feature coding information contains one or more corresponding classification nodes. According to the node feature coding information, the signal influencing factor attribute content set is recursively constructed, that is, a plurality of signal feature classification nodes corresponding to a plurality of signal features are sequentially obtained, and based on the mapping relationship between the signal features and the signal feature classification nodes, a stethoscope signal feature classifier is constructed according to the plurality of signal features and the plurality of signal feature classification nodes.

[0028] By performing signal feature node division and constructing a stethoscope signal feature classifier based on the mapping relationship between the signal features and the classification nodes, the accuracy and precision of feature classification can be improved, thereby improving the accuracy of obtaining the stethoscope signal data feature information set.

[0029] Based on the plurality of stethoscope signal data feature information sets, the plurality of stethoscope signal data sets are labeled with feature labels to obtain a plurality of stethoscope signal feature label data sets; The plurality of stethoscope signal feature label data sets are clustered and divided to determine a plurality of normal feature label signal data sets and a plurality of abnormal feature label signal data sets; Based on the signal twin network, the plurality of normal feature label signal data sets and the plurality of abnormal feature label signal data sets are analyzed for abnormal loss to generate the plurality of stethoscope signal abnormal feature twin databases.

[0030] In the embodiments of the present application, the plurality of stethoscope signal data sets are labeled with feature labels according to the plurality of stethoscope signal data feature information sets, where the feature label labeling refers to labeling the stethoscope signal data with feature information, and labeling the types of abnormal data and normal data with type labels, for example, labeling the stethoscope signal data as male, professional athlete, arrhythmia, abnormal data, etc. to obtain a plurality of stethoscope signal feature label data sets.

[0031] Then the multiple stethoscope signal feature label data sets are clustered and divided, that is, the stethoscope signal feature labels are clustered according to normal data labels and abnormal data labels, to obtain multiple normal feature label signal data sets and multiple abnormal feature label signal data sets.

[0032] Based on the signal twin network, the multiple normal feature label signal data sets and the multiple abnormal feature label signal data sets are subjected to abnormal loss analysis, and multiple stethoscope signal abnormal feature twin databases are generated according to the abnormal loss analysis results.

[0033] In an embodiment, the method further comprises: The signal twin network is built, the signal twin network comprising a normal twin sub-network and an abnormal twin sub-network, and the normal twin sub-network and the abnormal twin sub-network being a shared weight network; In the embodiment of the application, the method for generating multiple stethoscope signal abnormal feature twin databases comprises the following steps. First, a signal twin network is built, wherein the signal twin network comprises a normal twin sub-network and an abnormal twin sub-network, and the normal twin sub-network and the abnormal twin sub-network are a shared weight network. The shared weight network refers to that in a neural network, the normal twin sub-network and the abnormal twin sub-network use the same weight parameters. Setting the same weight parameters can extract similar features.

[0034] In an embodiment, the method further comprises: Signal feature extraction elements are acquired, the signal feature extraction elements comprising spectral features, time domain features, and energy features; A wavelet decomposition function is constructed, and based on the wavelet decomposition function, signal multi-scale decomposition is performed according to the signal feature extraction elements to generate a signal feature extraction layer; A signal feature coding rule is set, and a signal feature coding identification layer is created according to the signal feature coding rule; The signal feature extraction layer and the signal feature coding identification layer are functionally connected and fused to build the signal twin network.

[0035] In the embodiment of the application, the method for building the signal twin network comprises the following steps. First, signal feature extraction elements are acquired, the signal feature extraction elements comprising spectral features, time domain features, and energy features. The spectral features refer to the performance state of a signal in a frequency domain, including the amplitude and phase information of each frequency component. The time domain features refer to the performance state of a signal in a time domain, including the duration, interval time, and waveform of the signal. The energy features refer to the energy performance state of a signal, including the amplitude and power of the signal.

[0036] Construct a wavelet decomposition function, wherein the wavelet decomposition function is a tool for signal processing, which can decompose signals into components of different frequencies and different time scales to better understand and analyze signal characteristics, wherein the wavelet decomposition function includes a wavelet basis function and a decomposition layer number, wherein the wavelet basis function includes Haar wavelet, Daubechies wavelet, Morlet wavelet, etc., and a person skilled in the art can select a suitable wavelet basis function according to the actual application scenario; wherein the decomposition layer number refers to how many wavelet coefficients it is decomposed into, which can be determined according to signal characteristics and processing requirements.

[0037] Based on the wavelet decomposition function, perform signal multiscale decomposition according to the signal feature extraction elements, wherein signal multiscale decomposition refers to extracting more information by analyzing signals at different scales to generate a signal feature extraction layer.

[0038] Set a signal feature coding rule, wherein common signal feature coding rules include differential coding, prediction coding, transform coding, etc., which can be set according to actual conditions, and then create a signal feature coding identification layer according to the signal feature coding rule, wherein the signal feature coding identification layer is used to encode the extracted signal features, i.e., to convert the signal features into a processable data format, facilitating subsequent feature analysis and classification.

[0039] Finally, the signal feature extraction layer and the signal feature coding identification layer are functionally connected and fused to construct a signal twin network. By constructing a signal twin network, the efficiency and accuracy of signal feature extraction can be improved, and by encoding and identifying signal features, support is provided for comparison and analysis of signal features.

[0040] The plurality of normal feature label signal data sets and the plurality of abnormal feature label signal data sets are respectively input into the normal twin subnetwork and the abnormal twin subnetwork, and a plurality of normal stethoscope feature sets and a plurality of abnormal stethoscope feature sets are output; Perform feature loss analysis on the plurality of normal stethoscope feature sets and the plurality of abnormal stethoscope feature sets to obtain a plurality of stethoscope signal feature loss data sets; Integrate the labels based on the plurality of stethoscope signal feature loss data sets to generate the plurality of stethoscope signal abnormal feature twin databases.

[0041] In the embodiments of the present application, the plurality of normal feature label signal data sets are input into the normal twin subnetwork in the signal twin network, and the plurality of abnormal feature label signal data sets are input into the abnormal twin subnetwork in the signal twin network for analysis to obtain a plurality of normal stethoscope feature sets and a plurality of abnormal stethoscope feature sets.

[0042] Then, the feature loss analysis is performed on the plurality of normal stethoscope feature sets and the plurality of abnormal stethoscope feature sets in sequence, that is, the feature deviation values of the normal stethoscope features and the corresponding abnormal stethoscope features are calculated, and the feature deviation values are taken as feature loss data to obtain a plurality of stethoscope signal feature loss data sets. Then, the plurality of stethoscope signal feature loss data sets are labeled according to the abnormal feature labels and integrated by feature classification to obtain a plurality of stethoscope signal abnormal feature twin databases.

[0043] The plurality of stethoscope signal abnormal feature twin databases are respectively input into a deep learning network structure as local sample data for identification training to obtain a plurality of stethoscope signal abnormal identification networks. In the embodiments of the present application, the plurality of stethoscope signal abnormal feature twin databases are respectively input into a deep learning network structure as local sample data for identification training, wherein the input data of the deep learning network structure is the abnormal stethoscope feature, and the output data is the stethoscope signal feature loss data; a preset training constraint is obtained, wherein the preset training constraint can be set by a person skilled in the art according to actual needs, for example: setting the preset training constraint as the output result accuracy rate of 95%, and until the current training output result meets the preset training constraint, the plurality of stethoscope signal abnormal identification networks trained are obtained.

[0044] The model parameters of the plurality of stethoscope signal abnormal identification networks are encrypted and transmitted to a cloud center platform for global training to construct a global stethoscope signal abnormal identification network. In the embodiments of the present application, the model parameters of the plurality of stethoscope signal abnormal identification networks are encrypted and transmitted to a cloud center platform for global training to construct a global stethoscope signal abnormal identification network.

[0045] Through the integrated training of the model parameters based on the cloud center and the federated learning platform, the security of the model training and the accuracy of the model obtained can be improved.

[0046] The target electronic stethoscope signal stream is monitored and obtained, and the global stethoscope signal abnormal identification network is used for abnormal identification of the target electronic stethoscope signal stream to output a target stethoscope abnormal feature result. In the embodiment of the present application, the target electronic stethoscope signal stream of the target electronic stethoscope is monitored and acquired, and then the target electronic stethoscope signal stream is abnormally identified through the global stethoscope signal abnormality identification network to obtain a target stethoscope abnormality feature result.

[0047] The target electronic stethoscope signal stream is signal cluster collected and controlled based on the target stethoscope abnormality feature result.

[0048] In the embodiment of the present application, finally, the target electronic stethoscope signal stream is signal clustered and stored in zones according to the target stethoscope abnormality feature result. Through the above method, the technical problem that the existing electronic stethoscope signal collection method has low signal data collection refinement degree, leads to large signal data analysis dimension and complexity, and causes low signal data recognition analysis efficiency and accuracy can be solved, the refinement and accuracy of electronic stethoscope signal collection can be improved, the dimension and complexity of signal data analysis can be reduced, and valuable information in signal data can be easily extracted, thereby improving the efficiency and accuracy of signal data recognition analysis.

[0049] In one embodiment, the method further comprises: According to the plurality of abnormal feature label signal data sets, a stethoscope collected signal storage block set is constructed; Based on the target stethoscope abnormality feature result and the stethoscope collected signal storage block, feature block mapping is performed to determine a target collected signal storage block; Based on the target collected signal storage block, the target electronic stethoscope signal stream is distributedly stored in clusters.

[0050] In the embodiment of the present application, first, signal storage blocks are set according to the plurality of abnormal feature label signal data sets, wherein the abnormal feature label signal data set and the signal storage block have a one-to-one correspondence relationship, a plurality of signal storage blocks corresponding to the plurality of abnormal feature label signal data sets are obtained, and a stethoscope collected signal storage block set is constructed based on the plurality of signal storage blocks corresponding to the plurality of abnormal feature label signal data sets.

[0051] The target stethoscope abnormality feature result is input into the stethoscope collected signal storage block set for feature block mapping matching to determine a target collected signal storage block, and the target electronic stethoscope signal stream is stored in the target collected signal storage block, thereby completing the distributed storage in clusters of the target electronic stethoscope signal stream.

[0052] By constructing the stethoscope collected signal storage block set, the stethoscope signals of different types of features can be collected and stored in clusters, thereby improving the precision and accuracy of stethoscope signal collection and storage. Embodiments

[0053] Based on the same inventive concept as the electronic stethoscope signal acquisition method based on anomaly recognition in Embodiment 1 above, such as Figure 3 As shown, this application also provides an electronic stethoscope signal acquisition system based on anomaly recognition, including: a stethoscope signal dataset acquisition module 01, an anomaly loss analysis module 02, a stethoscope signal anomaly recognition network acquisition module 03, a global stethoscope signal anomaly recognition network construction module 04, a target stethoscope anomaly feature output module 05, and a signal clustering acquisition and control module 06, wherein: Stethoscope signal dataset acquisition module 01 is used to acquire multiple stethoscope signal datasets from multiple electronic stethoscopes. The multiple stethoscope signal datasets include multiple normal stethoscope signal datasets and multiple abnormal stethoscope signal datasets. Anomaly loss analysis module 02 is used to perform anomaly loss analysis on the multiple stethoscope signal datasets based on a signal twin network, and generate multiple stethoscope signal anomaly feature twin databases. Stethoscope signal anomaly recognition network module 03 is used to input the multiple stethoscope signal anomaly feature twin databases as local sample data into the deep learning network structure for recognition training, thereby obtaining multiple stethoscope signal anomaly recognition networks. Global stethoscope signal anomaly recognition network construction module 04 is used to encrypt and transmit the model parameters of the multiple stethoscope signal anomaly recognition networks to the cloud central platform for global training, thereby constructing a global stethoscope signal anomaly recognition network. The target stethoscope abnormal feature output module 05 is used to monitor and acquire the signal stream of the target electronic stethoscope, perform abnormal identification on the signal stream of the target electronic stethoscope based on the global stethoscope signal abnormality identification network, and output the target stethoscope abnormal feature result. The signal clustering acquisition and control module 06 is used to perform signal clustering acquisition and control on the signal stream of the target electronic stethoscope based on the abnormal feature results of the target stethoscope.

[0054] In one embodiment, the system further includes: A signal data anomaly acquisition module is used to acquire signal data anomaly factors, including missing values, duplicate values, inconsistent data, noisy data, and deviation data. An outlier identification module, configured to collect a plurality of electronic stethoscope source signal data sets, identify outliers in the plurality of electronic stethoscope source signal data sets based on signal data abnormal factors, and obtain abnormal stethoscope signal data information; An abnormal data correlation cleaning strategy determination module, configured to determine an abnormal data correlation cleaning strategy according to the signal data abnormal factors; A standardization processing module, configured to perform standardization processing on the abnormal stethoscope signal data information based on the abnormal data correlation cleaning strategy, and obtain the plurality of stethoscope signal data sets.

[0055] In one embodiment, the system further comprises: A stethoscope signal feature classifier construction module, configured to construct a stethoscope signal feature classifier, perform feature classification on the plurality of stethoscope signal data sets respectively through the stethoscope signal feature classifier, and obtain a plurality of stethoscope signal data feature information sets; A feature label annotation module, configured to perform feature label annotation on the plurality of stethoscope signal data sets based on the plurality of stethoscope signal data feature information sets, and obtain a plurality of stethoscope signal feature label data sets; A clustering division module, configured to perform clustering division on the plurality of stethoscope signal feature label data sets, and determine a plurality of normal feature label signal data sets and a plurality of abnormal feature label signal data sets; An abnormal loss analysis module, configured to perform abnormal loss analysis on the plurality of normal feature label signal data sets and the plurality of abnormal feature label signal data sets based on a signal twin network, and generate a plurality of stethoscope signal abnormal feature twin databases.

[0056] In one embodiment, the system further comprises: A signal twin network building module, configured to build the signal twin network, wherein the signal twin network comprises a normal twin sub-network and an abnormal twin sub-network, and the normal twin sub-network and the abnormal twin sub-network are shared weight networks; A stethoscope feature set output module, configured to input the plurality of normal feature label signal data sets and the plurality of abnormal feature label signal data sets into the normal twin sub-network and the abnormal twin sub-network respectively, and output a plurality of normal stethoscope feature sets and a plurality of abnormal stethoscope feature sets; a feature loss analysis module configured to perform feature loss analysis on the plurality of normal stethoscope feature sets and the plurality of abnormal stethoscope feature sets to obtain a plurality of stethoscope signal feature loss data sets; a label integration module configured to integrate labels based on the plurality of stethoscope signal feature loss data sets to generate the plurality of stethoscope signal abnormal feature twin databases.

[0057] In one embodiment, the system further comprises: a signal feature extraction element acquisition module configured to acquire signal feature extraction elements, the signal feature extraction elements including spectral features, time domain features, and energy features; a signal multi-scale decomposition module configured to construct a wavelet decomposition function, perform signal multi-scale decomposition according to the signal feature extraction elements based on the wavelet decomposition function, and generate a signal feature extraction layer; a signal feature encoding identification layer creation module configured to set a signal feature encoding rule and create a signal feature encoding identification layer according to the signal feature encoding rule; a signal twin network building module configured to functionally connect and fuse the signal feature extraction layer and the signal feature encoding identification layer to build the signal twin network.

[0058] In one embodiment, the system further comprises: a stethoscope signal influencing factor acquisition module configured to acquire stethoscope signal influencing factors, the stethoscope signal influencing factors including physiological signal types, physiological feature information, and detection environment time periods; an attribute content filling module configured to fill attribute contents for each of the stethoscope signal influencing factors to obtain a signal influencing factor attribute content set; a node feature encoding information determination module configured to sequentially encode each attribute content in the signal influencing factor attribute content set as a signal feature classification node to determine node feature encoding information; a recursive construction module configured to recursively construct the signal influencing factor attribute content set based on the node feature encoding information to obtain the stethoscope signal feature classifier.

[0059] In one embodiment, the system further comprises: The stethoscope collection signal storage block set construction module is configured to construct a stethoscope collection signal storage block set according to the plurality of abnormal feature label signal data sets. The feature block mapping module is configured to determine a target collection signal storage block by performing feature block mapping between the target stethoscope abnormal feature result and the stethoscope collection signal storage block. The clustered distributed storage module is configured to perform clustered distributed storage of the target electronic stethoscope signal stream based on the target collection signal storage block.

[0060] In summary, compared with the prior art, the embodiments of the present application have the following technical effects: (1) By performing clustered collection control of the electronic stethoscope signal based on the abnormal feature analysis result of the electronic stethoscope signal, the precision and accuracy of the electronic stethoscope signal collection can be improved, the dimension and complexity of the signal data analysis can be reduced, and the valuable information in the signal data can be extracted, thereby improving the efficiency and accuracy of the signal data recognition analysis.

[0061] (2) By performing signal feature node division and constructing a stethoscope signal feature classifier based on the mapping relationship between the signal features and the classification nodes, the precision and accuracy of the feature classification can be improved, thereby improving the accuracy of the stethoscope signal data feature information set acquisition.

[0062] (3) By constructing a signal twin network, the efficiency and accuracy of the signal feature extraction can be improved, and by encoding and identifying the signal features, support is provided for the comparison and analysis of the signal features.

[0063] The above-described embodiments only express several implementation manners of the present application, but should not be interpreted as limiting the scope of the patent protection of the present application. Therefore, various types of replacements, modifications and changes made by those skilled in the art without departing from the scope of the concept of the present application as defined by the appended claims are all within the scope of protection of the present application.

Claims

1. A signal acquisition system for an electronic stethoscope based on anomaly recognition, characterized in that, The system includes: The stethoscope signal dataset acquisition module is used to collect multiple stethoscope signal datasets from multiple electronic stethoscopes. These datasets include multiple normal stethoscope signal datasets, multiple abnormal stethoscope signal datasets, and factors contributing to signal data anomalies. Normal stethoscope signal data refers to signal data within the normal data range, which can be set according to the actual scenario and the patient's condition. Abnormal stethoscope signal data refers to signal data outside the normal data range. Factors contributing to signal data anomalies include missing values, duplicate values, inconsistent data, noisy data, and data deviations. An anomaly loss analysis module is used to perform anomaly loss analysis on the multiple stethoscope signal datasets based on a signal twin network, generating multiple stethoscope signal anomaly feature twin databases. The anomaly feature twin databases include: constructing a stethoscope signal feature classifier; classifying the features of the multiple stethoscope signal datasets using the stethoscope signal feature classifier to obtain multiple stethoscope signal data feature information sets; labeling the multiple stethoscope signal datasets with feature labels based on the multiple stethoscope signal data feature information sets to obtain multiple stethoscope signal feature label datasets; clustering the multiple stethoscope signal feature label datasets to determine multiple normal feature label signal datasets and multiple abnormal feature label signal datasets; and using a signal twin network to analyze the multiple normal feature label signal datasets and multiple abnormal feature label signal datasets. Anomaly loss analysis is performed on the normal feature label signal dataset to generate a series of stethoscope signal anomaly feature twin databases. The anomaly feature twin databases include the construction of the signal twin network, which includes a normal twin subnetwork and an anomaly twin subnetwork, and the normal twin subnetwork and the anomaly twin subnetwork are shared weight networks. The series of normal feature label signal datasets and the series of anomaly feature label signal datasets are respectively input into the normal twin subnetwork and the anomaly twin subnetwork, and the output consists of a series of normal stethoscope feature sets and a series of anomaly stethoscope feature sets. Feature loss analysis is performed on the series of normal stethoscope feature sets and the series of anomaly stethoscope feature sets to obtain a series of stethoscope signal feature loss datasets. Based on the series of stethoscope signal feature loss datasets, the labels are integrated to generate the series of stethoscope signal anomaly feature twin databases. A stethoscope signal anomaly recognition network acquisition module is used to input the multiple stethoscope signal anomaly feature twin databases as local sample data into a deep learning network structure for recognition training, thereby obtaining multiple stethoscope signal anomaly recognition networks. A global stethoscope signal anomaly recognition network construction module is used to encrypt and transmit the model parameters of the multiple stethoscope signal anomaly recognition networks to the cloud central platform for global training, thereby constructing a global stethoscope signal anomaly recognition network. The target stethoscope abnormal feature output module is used to monitor and acquire the signal stream of the target electronic stethoscope, perform abnormal identification on the signal stream of the target electronic stethoscope based on the global stethoscope signal abnormality identification network, and output the target stethoscope abnormal feature results. The signal clustering acquisition and control module is used to perform signal clustering acquisition and control on the signal stream of the target electronic stethoscope based on the abnormal feature results of the target stethoscope.

2. The electronic stethoscope signal acquisition system based on anomaly recognition according to claim 1, characterized in that, The construction of the signal twin network includes: acquiring signal feature extraction elements, including spectral features, time-domain features, and energy features; constructing a wavelet decomposition function, performing multi-scale decomposition of the signal based on the wavelet decomposition function according to the signal feature extraction elements, and generating a signal feature extraction layer; setting signal feature coding rules, creating a signal feature coding identifier layer according to the signal feature coding rules; and connecting and fusing the signal feature extraction layer and the signal feature coding identifier layer at the functional layer level to construct the signal twin network.

3. The electronic stethoscope signal acquisition system based on anomaly recognition according to claim 2, characterized in that, The factors influencing the stethoscope signal are obtained, including physiological signal type, physiological characteristic information, and detection environment time period. The attribute content of each factor influencing the stethoscope signal is filled to obtain a set of attribute content of signal influencing factors. The attribute content of each attribute content in the set of attribute content of signal influencing factors is used as a signal feature classification node and sorted in order.

4. The electronic stethoscope signal acquisition system based on anomaly recognition according to claim 3, characterized in that, Based on the multiple abnormal feature label signal datasets, a set of storage blocks for stethoscope-acquired signals is constructed. Based on the abnormal feature results of the target stethoscope and the storage block of the stethoscope acquired signal, feature block mapping is performed to determine the storage block of the target acquired signal; The target electronic stethoscope signal stream is clustered and distributed for storage based on the target acquisition signal storage block.

Images