[go: up one dir, main page]

WO2022161023A1 - Procédé et appareil de débruitage de signal de son cardiaque, et support de stockage - Google Patents

Procédé et appareil de débruitage de signal de son cardiaque, et support de stockage Download PDF

Info

Publication number
WO2022161023A1
WO2022161023A1 PCT/CN2021/139609 CN2021139609W WO2022161023A1 WO 2022161023 A1 WO2022161023 A1 WO 2022161023A1 CN 2021139609 W CN2021139609 W CN 2021139609W WO 2022161023 A1 WO2022161023 A1 WO 2022161023A1
Authority
WO
WIPO (PCT)
Prior art keywords
heart sound
sound signal
neural network
network model
signal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2021/139609
Other languages
English (en)
Chinese (zh)
Inventor
黄弯弯
杨溪
吕文尔
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Microdigit Co Ltd
Original Assignee
Shanghai Microdigit Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Microdigit Co Ltd filed Critical Shanghai Microdigit Co Ltd
Publication of WO2022161023A1 publication Critical patent/WO2022161023A1/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B7/00—Instruments for auscultation
    • A61B7/02—Stethoscopes
    • A61B7/04—Electric stethoscopes
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7225—Details of analogue processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7253—Details of waveform analysis characterised by using transforms
    • A61B5/726—Details of waveform analysis characterised by using transforms using Wavelet transforms
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B7/00—Instruments for auscultation

Definitions

  • the present application relates to the field of signal processing, and in particular, to a heart sound signal denoising method, device and storage medium.
  • Heart sound signal is one of the important signals used to detect cardiac performance and obtain physiological and pathological information.
  • the heart sound signal acquisition process will inevitably be affected by surrounding noise, such as electromagnetic interference, power frequency noise, electrical interference generated by the human body's own breathing or lung sounds, and so on.
  • the noise signal will seriously interfere with the effective heart sound signal, and even cause the loss of the effective heart sound signal, which is extremely unfavorable for extracting the corresponding pathological information.
  • the commonly used heart sound signal denoising methods mainly include hardware denoising and denoising using software algorithms.
  • denoising by hardware alone is not satisfactory, and it is easy to introduce frequency interference.
  • Denoising using only software algorithms mainly involves denoising the heart sound signal using a low-pass filter such as a Butterworth or Chebyshev low-pass filter alone, or The wavelet transform is used to process the noisy heart sound signal.
  • a low-pass filter such as a Butterworth or Chebyshev low-pass filter alone
  • the wavelet transform is used to process the noisy heart sound signal.
  • using only a low pass filter can only remove high frequency noise, not low frequency noise. Wavelet transform can effectively eliminate high-frequency noise and suppress low-frequency noise.
  • the present application provides a heart sound signal denoising method, device and storage medium, so that not only the high frequency noise of the heart sound signal can be removed, but also the low frequency noise of the heart sound signal can be effectively suppressed without causing effective signal details. of loss.
  • a method for denoising a heart sound signal comprising: acquiring a heart sound signal and an environmental noise signal of a subject to be tested; inputting the heart sound signal and the environmental noise signal into a trained A neural network model is used to obtain the preliminarily denoised heart sound signal; use a low-pass filter to perform low-pass filtering on the preliminarily denoised heart sound signal; use the wavelet threshold denoising method to process the low-pass filtered heart sound signal , to obtain the final heart sound signal of the subject to be tested.
  • the heart sound signal is obtained by using the main microphone of a stethoscope to collect toward the chest of the subject to be tested, and the environmental noise signal is obtained by using the stethoscope while collecting the heart sound signal. is collected from the microphone towards the outside air.
  • the trained neural network model is obtained by acquiring a sample heart sound signal and a sample ambient noise signal, wherein the sample heart sound signal is obtained by using the main microphone to collect the pure heart sound signal , and the sample ambient noise signal is obtained by collecting the sample heart sound signal from the microphone toward the outside air while collecting the sample heart sound signal; based on the sample heart sound signal and the sample noise signal, the pre-built A neural network model is trained such that the trained neural network model outputs the pure heart sound signal.
  • the pre-built neural network model adopts a Keras sequence model structure, which includes at least one long short-term memory network layer and at least one fully connected layer.
  • training a pre-built neural network model based on the sample heart sound signal and the sample noise signal includes: performing a short-time Fourier transform on the sample heart sound signal to extract a first SIFT feature matrix , and perform short-time Fourier transform on the sample environmental noise signal to extract a second SIFT feature matrix; perform normalization on the first SIFT feature matrix to obtain a normalized first feature matrix, and performing normalization processing on the second SIFT feature matrix to obtain a normalized second feature matrix; performing the pre-built neural network model based on the first feature matrix and the second feature matrix training such that the trained neural network model outputs the pure heart sound signal.
  • inputting the heart sound signal and the ambient noise signal into a trained neural network model to obtain a preliminarily denoised heart sound signal includes: performing a short-time Fourier transform on the heart sound signal to obtain a Extract the fifth SIFT feature matrix, and perform short-time Fourier transform on the environmental noise signal to extract the sixth SIFT feature matrix; perform normalization processing on the fifth SIFT feature matrix to obtain a normalized a fifth feature matrix, and normalize the sixth SIFT feature matrix to obtain a normalized sixth feature matrix; input the fifth feature matrix and the sixth feature matrix into the trained The neural network model is obtained, and the output of the trained neural network model is obtained as the preliminary denoised heart sound signal.
  • the low-pass filter is a Butterworth low-pass filter or a Chebyshev low-pass filter.
  • the wavelet threshold denoising method includes: performing wavelet transformation on the low-pass filtered heart sound signal to obtain a set of wavelet decomposition coefficients; performing threshold processing on the set of wavelet decomposition coefficients to obtain estimated wavelet coefficients; Inverse wavelet transform is performed on the estimated wavelet coefficients to obtain the final heart sound signal.
  • performing threshold processing on the set of wavelet decomposition coefficients includes: comparing each wavelet decomposition coefficient in the set of wavelet decomposition coefficients with a preset critical threshold; The wavelet decomposition coefficients at the critical threshold are set to zero, and other wavelet decomposition coefficients are reserved, thereby obtaining the estimated wavelet coefficients.
  • a heart sound signal denoising device which is characterized by comprising: a memory, where the memory stores a machine-executable program; and a processor, where the processor executes the machine-executable program.
  • the program executes the machine-executable program.
  • a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a machine, realizes the heart sound signal extraction according to the first aspect of the present application. noise method.
  • FIG. 1 shows a flowchart of a method for denoising a heart sound signal according to an embodiment of the present application.
  • FIG. 2 shows a flowchart of training a neural network model according to an embodiment of the present application
  • FIG. 3 shows an example block diagram of a pre-built neural network model according to an embodiment of the present application
  • FIG. 4 shows a flowchart of an implementation manner of step 202 of FIG. 2 according to an embodiment of the present application
  • FIG. 5 shows a flowchart of detecting a trained neural network model according to an embodiment of the present application
  • FIG. 6 shows a flowchart of an implementation manner of step 102 of FIG. 1 according to an embodiment of the present application
  • FIG. 7 shows a flowchart of an implementation manner of step 104 of FIG. 1 according to an embodiment of the present application
  • FIG. 8 shows a flowchart of an implementation manner of step 702 of FIG. 7 according to an embodiment of the present application
  • FIG. 9 shows example time-domain and frequency-domain signal diagrams of the heart sound signal output by the heart sound signal denoising method at different denoising stages according to an embodiment of the present application
  • FIG. 10 shows a schematic structural block diagram of an apparatus for denoising a heart sound signal according to an embodiment of the present application.
  • a heart sound signal denoising method includes the following steps:
  • Step 101 Acquire a heart sound signal and an environmental noise signal of the subject to be tested.
  • Figures 9(a) and 9(b) show an example of the originally acquired heart sound signal.
  • the heart sound signal is obtained by using the main microphone of the stethoscope to face the chest of the subject to be tested
  • the ambient noise signal is obtained by collecting the heart sound signal from the microphone of the stethoscope towards the outside air at the same time .
  • the subject to be detected may be any subject that needs to detect heart sounds (eg, patients with abnormal hearts, physical examination users who need to detect heart sounds, etc.).
  • Step 102 input the heart sound signal and the environmental noise signal into the trained neural network model to obtain the heart sound signal after preliminary noise reduction.
  • the trained neural network model is obtained through continuous training and learning using the pure heart sound signal, which can be used to retain useful signals in the heart sound signal and remove most of the noise.
  • Fig. 9(c) and Fig. 9(d) respectively show that the original collected heart sound signals shown in Fig. 9(a) and Fig. 9(b) and the corresponding environmental noise are input into the trained neural network model.
  • the time-domain and frequency-domain signal graphs of the initially denoised heart sound signal output by the neural network model It can be seen from the comparison with FIG. 9( a ) and FIG. 9( b ) that the heart sound signal after preliminary noise reduction retains the useful signal in the original heart sound signal, and removes part of the noise.
  • the trained neural network model shown in Figure 2 can be obtained in the following ways:
  • Step 201 obtain a sample heart sound signal and a sample environmental noise signal, wherein the sample heart sound signal is obtained by using the main microphone of the stethoscope to collect the pure heart sound signal, and the sample environmental noise signal is, while collecting the sample heart sound signal, using the stethoscope's Acquired from the microphone towards the outside air.
  • a pure heart sound signal refers to a heart sound signal that does not include any noise.
  • multiple sample heart sound signals are acquired (eg, 80 sample heart sound signals are acquired), and the same number of sample ambient noise signals are acquired (eg, 80 sample ambient noise signals are acquired), where each sample heart sound
  • the signal is acquired by using the main microphone of the stethoscope toward a corresponding pure heart sound signal in a plurality of pre-prepared pure heart sound signals (for example, 80 pre-prepared pure heart sound signals), and each sample ambient noise signal is, It is obtained by using the stethoscope to collect from the microphone toward the outside air while collecting the corresponding sample heart sound signal.
  • the sampling frequency of the main microphone and the slave microphone of the stethoscope for collecting each pure heart sound signal or noise environment can be 3906HZ
  • the sampling time can be 8 seconds
  • the number of sampling points can be up to 30160. The number of points constitutes the corresponding sample heart sound signal or sample ambient noise signal.
  • Step 202 train a pre-built neural network model based on the sample heart sound signal and the sample noise signal, so that the trained neural network model (ie, the trained neural network) outputs a corresponding pure heart sound signal.
  • the pre-built neural network model may be trained based on each sample heart sound signal and the sample noise signal, so that the The trained neural network model can output a corresponding pure heart sound signal for each sample heart sound signal and sample noise signal, thereby ensuring that the trained neural network model can be used for preliminary noise reduction of the heart sound signal.
  • the pre-built neural network model may adopt a Keras Sequential Model (Keras Sequential Model) structure, which includes at least one long short-term memory (Long Short-Term Memory, LSTM) layer and at least one fully connected layer layer (Dense layer).
  • Keras Sequential Model Keras Sequential Model
  • the pre-built neural network model may only include one layer of LSTM layer 301 and one layer of Dense layer 302, the advantages of this model are simple structure, easy implementation, and good scalability .
  • the present application can also adopt the Keras sequence model structure of multi-layer LSTM layers and multi-layer Dense layers without departing from the scope of this application.
  • the loss function of the pre-built neural network model may employ Mean Absolute Error (MAE), and the optimizer may employ an Adaptive Moment Estimation (Adam) optimizer.
  • MAE Mean Absolute Error
  • Adam Adaptive Moment Estimation
  • the pre-built neural network model may also adopt other neural network model structures, such as a linear neural network model, etc., without departing from the scope of the present application.
  • FIG. 4 a flowchart of an implementation manner of step 202 in FIG. 2 according to an embodiment of the present application is shown, which may include:
  • step 401 short-time Fourier transform is performed on the sample heart sound signal (the sample heart sound signal obtained in step 201) to extract the first SIFT (Scale-invariant feature transform) feature matrix, and the sample environment is analyzed.
  • the noise signal (the sample ambient noise signal obtained in step 201 ) is subjected to short-time Fourier transform to extract the second SIFT feature matrix.
  • the length of each SIFT feature matrix to be extracted may be preset as 300, and it is stipulated that if the length of the extracted SIFT feature matrix is less than 300, the length is supplemented with 0.
  • Step 402 perform normalization processing on the first SIFT feature matrix to obtain a normalized first feature matrix, and perform normalization processing on the second SIFT feature matrix to obtain a normalized second feature matrix.
  • the first SIFT eigenmatrix and the second SIFT eigenmatrix may be normalized by means of mean minus variance, and of course other normalization methods may also be used.
  • Step 403 Train the pre-built neural network model based on the first feature matrix and the second feature matrix, so that the trained neural network model outputs pure heart sound signals.
  • the above steps 401-403 may be repeated for each sample heart sound signal and sample ambient noise signal, so that the The trained neural network model can output the corresponding pure heart sound signal for each sample heart sound signal and sample noise signal.
  • the trained neural network model also needs to be tested to verify the noise reduction effect of the trained neural network model.
  • FIG. 5 a flowchart for detecting a trained neural network model is shown, which may include:
  • Step 501 acquiring the detected heart sound signal and the detected environmental noise signal, wherein the detected heart sound signal is obtained by using the main microphone of the stethoscope to collect the pure heart sound signal, and the detected environmental noise signal is, while collecting the detected heart sound signal, using the stethoscope's Acquired from the microphone towards the outside air.
  • multiple detected heart sound signals are acquired (eg, 20 detected heart sound signals are acquired), and the same number of corresponding detected ambient noise signals are acquired (eg, 20 detected ambient noise signals are acquired), where each detected ambient noise signal is acquired
  • the heart sound signal is obtained by using the main microphone of the stethoscope to collect the corresponding pure heart sound signals in a plurality of pre-prepared pure heart sound signals (for example, 20 pre-prepared pure heart sound signals), and each detected environmental noise signal is , while collecting the corresponding detected heart sound signal, using the stethoscope to collect from the microphone towards the outside air.
  • step 501 the implementation method of step 501 is similar to that of step 201, and the detected heart sound signal and the sample heart sound signal can be collected in the same step, but the pure heart sound signal used for collecting the detected heart sound signal should be different from that used for collecting the sample heart sound signal pure heart sound signal.
  • Step 502 perform short-time Fourier transform on the detected heart sound signal to extract a third SIFT feature matrix, and perform short-time Fourier transform on the detected environmental noise signal to extract a fourth SIFT feature matrix.
  • This step may be similar to step 401, and will not be repeated here.
  • Step 503 perform normalization processing on the third SIFT feature matrix to obtain a normalized third feature matrix, and perform normalization processing on the fourth SIFT feature matrix to obtain a normalized fourth feature matrix.
  • This step may be similar to step 402, and details are not described herein again.
  • Step 504 input the third feature matrix and the fourth feature matrix into the trained neural network model, to verify whether the trained neural network model outputs the corresponding pure heart sound signal (that is, for collecting and detecting the pure heart sound signal of the heart sound signal) .
  • the trained neural network model outputs the corresponding pure heart sound signal, it means that the noise reduction effect of the trained neural network model is better, otherwise it means that the trained neural network model needs further training , in order to achieve the expected noise reduction effect.
  • the above steps 502-504 may be repeated for each detected heart sound signal and detected environmental noise signal. And, in this implementation, if all the detected heart sound signals in the multiple detected heart sound signals, the trained neural network model outputs corresponding pure heart sound signals, it means that the trained neural network model is suitable for the collected heart sound signals. Perform preliminary noise reduction, otherwise it indicates that the trained neural network model is not suitable for preliminary noise reduction of the collected heart sound signals.
  • inputting the heart sound signal and the ambient noise signal into a trained neural network model to obtain a preliminarily denoised heart sound signal may include:
  • Step 601 perform short-time Fourier transform on the heart sound signal (heart sound signal obtained in step 101) to extract the fifth SIFT feature matrix, and perform short-time Fourier transformation on the environmental noise signal (the environmental noise signal obtained in step 101). Lie transform to extract the sixth SIFT feature matrix.
  • This step may be similar to step 401, and will not be repeated here.
  • Step 602 perform normalization processing on the fifth SIFT feature matrix to obtain a normalized fifth feature matrix, and perform normalization processing on the sixth SIFT feature matrix to obtain a normalized sixth feature matrix.
  • This step may be similar to step 402, and details are not described herein again.
  • Step 603 input the fifth feature matrix and the sixth feature matrix into the trained neural network model, and obtain the output of the trained neural network model as the preliminarily denoised heart sound signal.
  • low-pass filtering is performed on the preliminarily denoised heart sound signal by using a low-pass filter.
  • the low-pass filter may be a Butterworth low-pass filter, or a Chebyshev low-pass filter.
  • the low-pass filter can be mainly used to remove high-frequency noise in the heart sound signal.
  • Figure 9(e) and Figure 9(f) the preliminary results shown in Figure 9(c) and Figure 9(d) using a 16th-order Butterworth low-pass filter with a cutoff frequency of 500 Hz are shown, respectively.
  • step 103 may include performing a fast Fourier transform on the preliminarily denoised heart sound signal to convert it from the time domain to the frequency domain, and then using a low-pass filter to low-pass the resulting frequency-domain signal filter, and perform inverse Fourier transform on the low-pass filtered frequency signal to convert it from the frequency domain to the time domain, thereby obtaining the low-pass filtered heart sound signal.
  • a Butterworth low-pass filter can use the following formula to low-pass filter the corresponding frequency domain signal:
  • ⁇ is the frequency signal obtained by performing the fast Fourier transform on the preliminarily denoised heart sound signal
  • H( ⁇ ) is the frequency-domain signal obtained after low-pass filtering the obtained frequency-domain signal with a low-pass filter
  • n is the order of the filter
  • ⁇ c is the cut-off frequency, that is, the frequency at which the amplitude drops to -3 dB;
  • ⁇ p is the passband edge frequency
  • Step 104 using the wavelet threshold denoising method to process the low-pass filtered heart sound signal to obtain the final heart sound signal of the subject to be tested.
  • the wavelet threshold denoising method can be used to further remove the noise existing in the heart sound signal, which can not only remove high frequency noise, but also filter out low frequency noise.
  • wavelet denoising of the low-pass filtered heart sound signal shown in Figures (e) and (f) using the db6 wavelet and a critical threshold of 0.07, respectively is shown After that, the time-domain and frequency-domain signal graphs of the final heart sound signal are obtained. Compared with Fig. 9(f), in Fig.
  • the high frequency signal is further suppressed, and the low frequency signal is also denoised, especially the 0-250HZ signal is reduced due to the noise removal;
  • the lines are further thinned, but the details of the effective signal are still clearly visible. It can be seen that after wavelet denoising, the data is cleaner and the noise is better. is small without causing loss of valid signal details. It can be seen that, through the method for denoising the heart sound signal of the present application, the high-frequency noise can be completely filtered, and the low-frequency noise can be better suppressed, and the effective signal details will not be lost, so that the filtering effect can be optimized. .
  • the wavelet threshold denoising method in step 104 may specifically include:
  • Step 701 Perform wavelet transform on the low-pass filtered heart sound signal to obtain a set of wavelet decomposition coefficients.
  • wavelet transform is that after the wavelet basis function is shifted by ⁇ , the inner product is made with the analyzed signal f(t) at different scales a.
  • the low-pass filtered heart sound signal is wavelet transformed using the following formula:
  • f(t) is the low-pass filtered heart sound signal
  • ⁇ ( ) is a wavelet basis function, and a specific wavelet basis function can be selected in this application according to actual needs, for example, Db6 wavelet in the Daubechies wavelet family can be selected only as an example;
  • ⁇ is a translation factor, which can be positive or negative, and its function is to translate the wavelet basis function along the time axis.
  • ⁇ can be preset according to the actual situation
  • WT(a, ⁇ ) is a set of wavelet decomposition coefficients obtained by using the above formula to perform wavelet transform on the low-pass filtered heart sound signal.
  • Step 702 Perform threshold processing on the wavelet decomposition coefficient set to obtain estimated wavelet coefficients.
  • step 702 may include: step 801 , comparing each wavelet decomposition coefficient in the wavelet decomposition coefficient set with a preset critical threshold ( ⁇ ).
  • the critical threshold ⁇ is used to indicate the boundary between the noise and the effective signal, that is, if the wavelet coefficient is less than ⁇ , it is considered that the wavelet coefficient is mainly caused by noise, and if the wavelet coefficient is greater than ⁇ , it is considered that the wavelet coefficient is mainly caused by noise. is caused by a valid signal.
  • the critical threshold ⁇ may be selected to be 0.07.
  • Step 802 Set the wavelet decomposition coefficients smaller than the critical threshold in the wavelet decomposition coefficient set to zero, and retain other wavelet decomposition coefficients, so as to obtain the estimated wavelet coefficients.
  • Step 703 Perform inverse wavelet transform on the estimated wavelet coefficients to obtain a final heart sound signal.
  • the inverse wavelet transform is the inverse transform of the wavelet transform used in step 701 .
  • the present application further provides an apparatus for denoising a heart sound signal.
  • the apparatus for denoising a heart sound signal includes a memory 1001 and a processor 1002 , and a machine executable program is stored in the memory 1001 .
  • the processor 1002 executes the machine-executable program, it implements the heart sound signal denoising method described in the above embodiments.
  • the number of the memory 1001 and the processor 1002 may be one or more.
  • the heart sound signal denoising apparatus may be implemented using electronic equipment intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, Blade servers, mainframe computers, and other suitable computers.
  • Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices.
  • the heart sound signal denoising apparatus may further include a communication interface 1003 for communicating (wired or wireless) with external devices (eg, a master microphone and a slave microphone of a stethoscope) for data interaction therewith.
  • a communication interface 1003 for communicating (wired or wireless) with external devices (eg, a master microphone and a slave microphone of a stethoscope) for data interaction therewith.
  • the memory 1001 may include nonvolatile memory and volatile memory.
  • Non-volatile memory may include, for example, read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, and the like.
  • Volatile memory may include, for example, random access memory (RAM) or external cache memory.
  • RAM random access memory
  • the RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
  • the memory 1001, the processor 1002, and the communication interface 1003 can be connected to each other through a bus and implement mutual communication.
  • the bus can be an industry standard architecture (Industry Standard Architecture, ISA) bus, a peripheral device interconnect (Peripheral Component, PCI) bus or an extended industry standard architecture (Extended Industry Standard Component, EISA) bus and the like.
  • the bus can be divided into an address bus, a data bus, a control bus, and the like. For ease of presentation, only one thick line is used in FIG. 10, but it does not mean that there is only one bus or one type of bus.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Public Health (AREA)
  • Molecular Biology (AREA)
  • Veterinary Medicine (AREA)
  • General Health & Medical Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Surgery (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Signal Processing (AREA)
  • Pathology (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Psychiatry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physiology (AREA)
  • Acoustics & Sound (AREA)
  • Power Engineering (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)

Abstract

La divulgation concerne un procédé et un appareil de débruitage de signal de son cardiaque, et un support de stockage. Le procédé comprend les étapes consistant à : acquérir un signal de son cardiaque d'un sujet devant être soumis à un examen, et un signal de bruit ambiant (101) ; entrer le signal de son cardiaque et le signal de bruit ambiant dans un modèle de réseau neuronal entraîné, de façon à obtenir un signal de son cardiaque qui a été soumis à un débruitage préliminaire (102) ; utiliser un filtre passe-bas pour effectuer un filtrage passe-bas sur le signal de son cardiaque qui a été soumis à un débruitage préliminaire (103) ; et utiliser un procédé de débruitage à valeur seuil par ondelettes pour traiter le signal de son cardiaque qui a été soumis à un filtrage passe-bas, de façon à obtenir un signal de son cardiaque final dudit sujet (104). Sur la base du procédé de débruitage de signal de son cardiaque, non seulement le bruit haute fréquence d'un signal de son cardiaque est éliminé, mais le bruit basse fréquence du signal de son cardiaque peut également être supprimé de manière efficace, de telle sorte qu'aucune perte de détails de signal efficace n'est provoquée.
PCT/CN2021/139609 2021-01-26 2021-12-20 Procédé et appareil de débruitage de signal de son cardiaque, et support de stockage Ceased WO2022161023A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110103809.7A CN114788709A (zh) 2021-01-26 2021-01-26 心音信号去噪方法、装置和存储介质
CN202110103809.7 2021-01-26

Publications (1)

Publication Number Publication Date
WO2022161023A1 true WO2022161023A1 (fr) 2022-08-04

Family

ID=82459327

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/139609 Ceased WO2022161023A1 (fr) 2021-01-26 2021-12-20 Procédé et appareil de débruitage de signal de son cardiaque, et support de stockage

Country Status (2)

Country Link
CN (1) CN114788709A (fr)
WO (1) WO2022161023A1 (fr)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115307715A (zh) * 2022-08-09 2022-11-08 南昌航空大学 一种基于sagnac光纤声传感系统的改进小波去噪方法
CN116563637A (zh) * 2023-05-16 2023-08-08 海南大学 基于自适应阈值小波去噪的乳腺癌病理图像分类系统
CN119014894A (zh) * 2024-08-01 2024-11-26 中国人民解放军总医院第一医学中心 基于心音分析的休克分类方法、装置、设备及介质
CN119049454A (zh) * 2024-10-31 2024-11-29 杭州海康威视数字技术股份有限公司 一种模型训练方法、声音信号处理方法、装置
CN119377646A (zh) * 2024-12-27 2025-01-28 中国航空工业集团公司金城南京机电液压工程研究中心 一种液压泵监测信号自适应滤波处理方法和装置

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115753156A (zh) * 2022-11-29 2023-03-07 成都轨道交通产业技术研究院有限公司 一种列车走行部声音检测及识别方法
CN119586971A (zh) * 2024-10-21 2025-03-11 深圳市欧美亚实业有限公司 生物电信号抗干扰方法、装置、设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009079976A2 (fr) * 2007-12-21 2009-07-02 Guy Leonard Kouemou Procédé et dispositif de surveillance cardiaque, circulatoire et respiratoire au moyen de modèles de markov cachés et de réseaux neuronaux
CN109044396A (zh) * 2018-06-25 2018-12-21 广东工业大学 一种基于双向长短时记忆神经网络的智能心音识别方法
CN110558944A (zh) * 2019-09-09 2019-12-13 成都智能迭迦科技合伙企业(有限合伙) 心音处理方法、装置、电子设备及计算机可读存储介质
US20200046244A1 (en) * 2018-08-08 2020-02-13 Tata Consultancy Services Limited Parallel implementation of deep neural networks for classifying heart sound signals
WO2021004345A1 (fr) * 2019-07-10 2021-01-14 东南大学 Système et procédé d'acquisition et d'analyse de son cardiaque utilisant une architecture en nuage

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN209032422U (zh) * 2017-11-08 2019-06-28 华南师范大学 一种心音信号检测设备
CN111223493B (zh) * 2020-01-08 2022-08-02 北京声加科技有限公司 语音信号降噪处理方法、传声器和电子设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009079976A2 (fr) * 2007-12-21 2009-07-02 Guy Leonard Kouemou Procédé et dispositif de surveillance cardiaque, circulatoire et respiratoire au moyen de modèles de markov cachés et de réseaux neuronaux
CN109044396A (zh) * 2018-06-25 2018-12-21 广东工业大学 一种基于双向长短时记忆神经网络的智能心音识别方法
US20200046244A1 (en) * 2018-08-08 2020-02-13 Tata Consultancy Services Limited Parallel implementation of deep neural networks for classifying heart sound signals
WO2021004345A1 (fr) * 2019-07-10 2021-01-14 东南大学 Système et procédé d'acquisition et d'analyse de son cardiaque utilisant une architecture en nuage
CN110558944A (zh) * 2019-09-09 2019-12-13 成都智能迭迦科技合伙企业(有限合伙) 心音处理方法、装置、电子设备及计算机可读存储介质

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115307715A (zh) * 2022-08-09 2022-11-08 南昌航空大学 一种基于sagnac光纤声传感系统的改进小波去噪方法
CN116563637A (zh) * 2023-05-16 2023-08-08 海南大学 基于自适应阈值小波去噪的乳腺癌病理图像分类系统
CN119014894A (zh) * 2024-08-01 2024-11-26 中国人民解放军总医院第一医学中心 基于心音分析的休克分类方法、装置、设备及介质
CN119049454A (zh) * 2024-10-31 2024-11-29 杭州海康威视数字技术股份有限公司 一种模型训练方法、声音信号处理方法、装置
CN119377646A (zh) * 2024-12-27 2025-01-28 中国航空工业集团公司金城南京机电液压工程研究中心 一种液压泵监测信号自适应滤波处理方法和装置

Also Published As

Publication number Publication date
CN114788709A (zh) 2022-07-26

Similar Documents

Publication Publication Date Title
WO2022161023A1 (fr) Procédé et appareil de débruitage de signal de son cardiaque, et support de stockage
CN110246504B (zh) 鸟类声音识别方法、装置、计算机设备和存储介质
CN110634497B (zh) 降噪方法、装置、终端设备及存储介质
Castillo et al. Noise Suppression in ECG Signals through Efficient One‐Step Wavelet Processing Techniques
CN107403416B (zh) 基于nsct的改进滤波与阈值函数的医学超声图像去噪方法
CN102663695B (zh) 基于小波变换的dr图像去噪方法及系统
CN107184187B (zh) 基于DTCWT-Spline的脉搏波信号去噪处理方法
CN113314147B (zh) 音频处理模型的训练方法及装置、音频处理方法及装置
Misal et al. Denoising of PCG signal by using wavelet transforms
El Bouny et al. ECG signal filtering based on CEEMDAN with hybrid interval thresholding and higher order statistics to select relevant modes
WO2019041772A1 (fr) Procédé et système de surveillance de profondeur d'anesthésie basé sur un signal d'électroencéphalogramme
CN107341769A (zh) 一种心电信号除噪方法及系统
CN107392123B (zh) 一种基于相参积累消噪的射频指纹特征提取和识别方法
CN107463956B (zh) 一种基于非负矩阵分解的心肺音分离的方法及装置
WO2021217873A1 (fr) Procédé, appareil et dispositif d'élimination par filtrage d'une dérive de ligne de base d'un électrocardiosignal, et support de stockage
CN111862978A (zh) 一种基于改进mfcc系数的语音唤醒方法及系统
CN111582138A (zh) 基于频域倒谱系数特征的心电身份识别方法及系统
CN111863008A (zh) 一种音频降噪方法、装置及存储介质
CN108956141A (zh) 基于模平方阈值—FastICA的滚动轴承降噪方法
Fu et al. Improved wavelet thresholding function and adaptive thresholding for noise reduction
CN110930983A (zh) 用于语音信号的小波特征提取方法、系统和存储介质
CN105046654B (zh) 一种基于粒子群优化的心电信号自适应非局部均值降噪方法
CN106651983A (zh) 磁共振图像重建方法及装置
CN118503659A (zh) 基于小波去噪和模态分解的单通道信号分离方法
CN107341519B (zh) 一种基于多分辨率分析的支持向量机识别优化方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21922613

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21922613

Country of ref document: EP

Kind code of ref document: A1