WO2018176919A1 - 爆裂音识别方法及系统 - Google Patents

爆裂音识别方法及系统 Download PDF

Info

Publication number
WO2018176919A1
WO2018176919A1 PCT/CN2017/116491 CN2017116491W WO2018176919A1 WO 2018176919 A1 WO2018176919 A1 WO 2018176919A1 CN 2017116491 W CN2017116491 W CN 2017116491W WO 2018176919 A1 WO2018176919 A1 WO 2018176919A1
Authority
WO
WIPO (PCT)
Prior art keywords
sound
component
wet
preset frequency
frequency bands
Prior art date
Application number
PCT/CN2017/116491
Other languages
English (en)
French (fr)
Inventor
陈雪
张勇
盖淑萍
Original Assignee
京东方科技集团股份有限公司
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 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Priority to US16/081,880 priority Critical patent/US11660062B2/en
Publication of WO2018176919A1 publication Critical patent/WO2018176919A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B7/00—Instruments for auscultation
    • A61B7/003—Detecting lung or respiration noise
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/08—Detecting, measuring or recording devices for evaluating the respiratory organs
    • 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
    • A61B7/02—Stethoscopes
    • A61B7/04—Electric stethoscopes
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2562/00—Details of sensors; Constructional details of sensor housings or probes; Accessories for sensors
    • A61B2562/02—Details of sensors specially adapted for in-vivo measurements
    • A61B2562/0204—Acoustic sensors
    • 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 analog 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
    • A61B5/7253—Details of waveform analysis characterised by using transforms
    • A61B5/726—Details of waveform analysis characterised by using transforms using Wavelet transforms
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/63—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for local operation
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems

Definitions

  • the present disclosure relates to a popping sound recognition method and system.
  • Pulmonary auscultation sounds are closely related to the pathological condition of the lungs.
  • the popping sound is used as a special wet rales as the primary clinical clue to suggest interstitial lung disease.
  • the hearing sensitivities of different human ears in different frequency bands are different, and the intensity of wet rales is low, the duration is short, and the bandwidth is wide, which may result in inconsistent diagnosis between different doctors.
  • digital auscultation is usually used for assisted diagnosis, which can make up for the subjectivity and limitations of doctors' auscultation.
  • accuracy of existing digital auscultation techniques for identifying popping sound is not high.
  • the present disclosure provides a pop sound recognition method comprising the following steps:
  • Calculating a power spectrum of the wet rales component and performing at least one of calculating: calculating a ratio of power of each of the plurality of preset frequency bands to total power of all of the preset frequency bands based on the power spectrum And the total power of all the preset frequency bands, at least one is selected as the frequency domain parameter; and the ratio of the number of wet rales in the late inspiratory phase to the total number of occurrences of the entire inspiratory wet rales and the entire inhalation wet The maximum amplitude of the sound, at least one is selected as the time domain parameter;
  • the feature is input into the classification model for classification recognition to identify the popping sound.
  • the step of processing the collected lung sound signal to extract a wet rales component of a breathing cycle comprising:
  • a breathing cycle is determined based on the breath sound component, and the wet rales of the breathing cycle are extracted.
  • the wavelet decomposition comprises coif2 wavelet decomposition.
  • the number of decomposition layers is N ⁇ 6.
  • An represents the approximate component of the nth layer of wavelet decomposition, where n is an integer and 0 ⁇ n ⁇ 10.
  • the step of determining a breathing cycle according to the breath sound component comprises:
  • a smooth breathing cycle is determined based on the respiratory gas phase information.
  • the ratio corresponding to the two frequency bands with the largest difference is selected as the frequency domain parameter.
  • the preset frequency band includes 50 Hz to 200 Hz and 500 Hz to 1000 Hz.
  • the inspiratory phase is 1/2 of the entire inspiratory phase.
  • the step of processing the acquired lung sound signal includes:
  • the lung tone signal is filtered using a bandpass filter.
  • the method further comprises acquiring a lung sound signal using the collector.
  • the present disclosure also provides a popping sound recognition system, including:
  • a processor configured to process the acquired lung sound signal to extract a wet rales component of a breathing cycle
  • a calculator configured to perform at least one of the following calculations: calculating the wet rales a power spectrum of the component, based on the power spectrum, calculating a ratio of power of each of the preset frequency bands to a total power of all the preset frequency bands and a total power of all the preset frequency bands, and selecting at least one of Frequency domain parameters; and the ratio of the number of wet-sounds in the late inspiratory phase of the inspiratory cycle to the total number of occurrences of the total inspiratory wet rales and the maximum amplitude of the entire inspiratory wet rales, and at least one is selected as the time domain parameter. ;
  • the identifier is configured to input the obtained frequency domain parameter and/or the time domain parameter as a parameter feature into the classification model for classification and identification to identify a popping sound.
  • the processor includes:
  • the wavelet decomposition circuit is configured to perform wavelet decomposition processing on the collected lung sound signals to obtain wet rales and breath sound components
  • An extraction circuit configured to determine a breathing cycle based on the breath sound component and extract a wet rake of the breathing cycle.
  • the wavelet decomposition comprises coif2 wavelet decomposition.
  • the extracting circuit is configured to obtain an average power curve of the breath sound component in a preset frequency range; the peak point of the average power curve is identified as an inspiratory gas apex, and the valley point is a respiratory gas phase switching point, Thereby obtaining respiratory gas phase information; determining a smooth breathing cycle based on the respiratory gas phase information.
  • the calculating unit is configured to select, according to a ratio of the power of each preset frequency band to the total power of all the preset frequency bands, a ratio corresponding to the two frequency bands with the largest difference as the frequency domain parameter.
  • the preset frequency band includes 50 Hz to 200 Hz and 500 Hz to 1000 Hz.
  • the inspiratory phase is 1/2 of the entire inspiratory phase.
  • the popping sound recognition system further comprises a band pass filter for filtering the acquired lung sound signal.
  • the popping sound recognition system further includes a collector configured to acquire a lung sound signal.
  • the collector comprises a digital stethoscope.
  • the present disclosure also provides a pop sound recognition system comprising:
  • a processor configured to perform the following steps:
  • Calculating a power spectrum of the wet rales component and performing at least one of calculating: calculating a ratio of power of each of the plurality of preset frequency bands to total power of all of the preset frequency bands based on the power spectrum And the total power of all the preset frequency bands, at least one is selected as the frequency domain parameter; and the ratio of the number of wet rales in the late inspiratory phase to the total number of occurrences of the entire inspiratory wet rales and the entire inhalation wet The maximum amplitude of the sound, at least one is selected as the time domain parameter;
  • At least one of the obtained frequency domain parameter and the time domain parameter is input as a parameter feature into the classification model for classification recognition to identify a popping sound.
  • FIG. 1 is a flowchart of a popping sound recognition method according to Embodiment 1 of the present disclosure
  • FIG. 2 is a schematic diagram showing the relationship between a lung sound signal, a wet rales component, and a breath sound component in the present disclosure
  • FIG. 3 is a flow chart of step S2 of Figure 1;
  • FIG. 4 is a flow chart of step S22 of Figure 3;
  • Figure 5 is a graph showing the average power of the breath sound component
  • FIG. 6 is a schematic block diagram of a popping sound recognition system according to Embodiment 2 of the present disclosure.
  • Figure 7 is a schematic block diagram of the processing unit of Figure 6.
  • FIG. 1 is a flowchart of a method for identifying a popping sound according to Embodiment 1 of the present disclosure; referring to FIG. 1, the method for identifying a popping sound provided by the embodiment includes:
  • the present disclosure may use a power ratio (PR) of the power of each preset frequency band as a total frequency of all preset frequency bands as a frequency domain parameter; or, the total power PR total of all preset frequency bands is taken as The frequency domain parameter; or, the ratio of the power of each preset frequency band to the total power of all the preset frequency bands (Power Ratio, PR) and the total power PR total of all the preset frequency bands are used as the frequency domain parameters.
  • PR power ratio
  • the ratio of the number of wet rales in the inspiratory phase of the respiratory cycle to the total number of occurrences of the entire inspiratory wet rales is used as a time domain parameter; or, the maximum amplitude of the entire inspiratory wet rales is used as a time domain parameter.
  • the ratio of the number of wet rhythms in the inspiratory phase of the respiratory cycle to the total number of occurrences of the wetness of the inspiratory phase and the maximum amplitude of the entire inspiratory wet rales are taken as time domain parameters.
  • the classification model is pre-trained and has a function of identifying a popping sound according to frequency domain parameters and/or time domain parameters.
  • the recognition result can be directly obtained.
  • the training classification model can be a support vector machine model (SVM classification model).
  • step S1 may be omitted, that is, the present invention is not limited to processing the directly collected lung sound signal, and the method of the present invention may process the lung sound signal obtained by other means.
  • the present invention can also process lung tone signals that are acquired in advance and stored in a computer readable medium.
  • an inspiratory phase includes an inspiratory phase and an inspiratory phase.
  • the wet rales can be classified into an inspiratory pre-wet rales and an inspiratory late-wet rales.
  • the present disclosure since the pitch of the cracking sound is higher, stronger, and more frequently in the late phase of the inspiratory phase according to the clinical experience, and the amplitude and frequency of the high, strong, and wet rales of the pitch are correlated, the present disclosure passes the wet rales. At least one of a power ratio in different frequency bands and a total power of all frequency bands as a frequency domain parameter; and/or, a maximum amplitude of wet rales and a late inspiratory phase wet At least one of the rhythm appearance ratios is used as a time domain parameter to improve the recognition accuracy of the popping sound.
  • the acquired lung sound signal is filtered by using a band pass filter to remove noise such as heart sound signals and power frequency interference.
  • the band pass filter includes, but is not limited to, an 8th order Butterworth band pass filter, and the pass band of the band pass filter may be 50 Hz to 2000 Hz.
  • the acquisition time can be, but not limited to, 10s.
  • the sampling frequency fs can be but not limited to 10000Hz. .
  • step S2 can be implemented by processing the collected lung sound signals to extract a wet rales component of a breathing cycle by the following steps S21 and S22:
  • the wavelet decomposition method is used in the present disclosure to process the collected lung sound signals, so that the processing process is simpler, and thus the burst sound recognition efficiency can be improved.
  • the wavelet is decomposed into a Coiflet wavelet (coif N) decomposition.
  • the coif2 wavelet decomposition method is adopted in the present disclosure because the coif2 wavelet has better symmetry, and can avoid waveform distortion of the reconstructed wet rales and breath sound components, thereby further improving the recognition of popping sound. Accuracy.
  • Wavelet decomposition process the original signal S is first divided into two components, the high-frequency component is also called the detail component d1, the low-frequency component is also called the approximate component a1, and then the low-frequency component is decomposed for the second time, divided into two components, the detail component d2 and Approximate component a2, and so on, until it is decomposed to a predetermined number of layers N.
  • the number of decomposition layers of the coif2 wavelet decomposition is N ⁇ 6, and the waveform distortion of the wet rales component can be well avoided.
  • the number of decomposition layers is 9, the d1 to d6 wavelet components are selected to obtain the wet rales component, and the d7 to d9 and a9 wavelet components are cumulatively obtained to obtain the breath component, wherein dn represents the wavelet decomposition of the nth layer of the detail component. , an represents the approximate component of the nth layer of wavelet decomposition, n is an integer, and 0 ⁇ n ⁇ 10.
  • the frequency range corresponding to the high frequency band of the nth layer component is fs/2 n+1 to fs/2 n , 1 ⁇ n ⁇ N, although the main frequency band of the burst sound distribution is 200 Hz. ⁇ 1000Hz, however, the burst sound tone is higher than other wet rales, and the wet rales overall distribution frequency band is 50Hz ⁇ 2500Hz. Therefore, the detail component d1 ⁇ d6 wavelet component is selected to obtain the wet rales component, the detail component d7 ⁇ d9 and the approximation. The component a9 wavelet component is accumulated to obtain the breath sound component, which can well avoid waveform distortion of the wet rales component and the breath component.
  • FIG. 2 is a schematic diagram showing the relationship between the lung sound signal, the wet rales component and the breath sound component in the present disclosure. It can be directly seen from FIG. 2: the wet rales component in the lung sound signal. It can be clearly seen that, as indicated by the black arrow, waveform distortion of the wet rales and breath sound components can be well avoided.
  • wavelet decomposition can also adopt other decomposition methods.
  • the step of determining a breathing cycle according to the breath sound component in the step S22 may be implemented by the following steps S221 to S223:
  • S221 Obtain an average power curve of the breath sound component in a preset frequency range.
  • the preset frequency range may be, but not limited to, 150 to 450 Hz.
  • the frequency range of 150 Hz to 450 Hz is selected because the energy difference of the respiratory gas phase in this frequency band is most significant.
  • the breath sound component may be first divided into a window of 100 ms duration, the window is shifted by 75 ms, and the respiratory sound signal in each window is calculated by the following formula (1).
  • x(m) is the mth window of the breath sound component
  • t n and f k are time and frequency, respectively
  • w(t n Dm) is a moving window function of duration T and window shift to D.
  • the present disclosure uses a Hanning window, although other window functions can of course be used.
  • f high is 450 Hz and f low is 150 Hz.
  • the start time of the waveform at the apex of the inspiratory gas to the end of the waveform of the respiratory gas phase switching point is a breathing cycle, as shown in FIG. 5 T1-T3; the so-called stable breathing cycle refers to a relatively stable breathing cycle of the power curve.
  • T1 is relatively flat compared to T2, so T2 is chosen as a smooth breathing cycle.
  • the reason for determining a smooth breathing cycle in the present disclosure is that the patient's lung condition can be better reflected by a smooth breathing cycle, thereby facilitating the improvement of the accuracy of identifying the popping sound.
  • the ratio corresponding to the two frequency bands with the largest difference is selected as the frequency domain parameter, which is equivalent to removing some similarities.
  • the interference band factor can improve the accuracy of identifying the popping sound.
  • the preset frequency band includes 50 Hz to 200 Hz and 500 Hz to 1000 Hz, because the ratio of the power of the two frequency bands to the total power of all the preset frequency bands is the largest.
  • the preset frequency band may include, but is not limited to, four preset frequency bands of 50 Hz to 200 Hz, 200 Hz to 500 Hz, 500 Hz to 1000 Hz, and 1000 Hz to 1500 Hz.
  • the inspiratory phase in the present disclosure is 1/2 of the entire inspiratory phase, that is, the inspiratory phase is equally divided into an inspiratory phase and an inspiratory phase, so that it can be ensured as much as possible.
  • the popping sound is located in the late phase of the inspiratory phase, which improves the accuracy of identifying the popping sound.
  • the power spectrum of one respiratory cycle wet rales component may be calculated, but not limited to, using a multi-order autoregressive model, for example, in this embodiment, The 14th-order autoregressive model calculates the power spectrum of a respiratory cycle wet rales.
  • FIG. 6 is a schematic block diagram of a popping sound recognition system according to Embodiment 2 of the present disclosure.
  • the popping sound recognition system includes an acquisition unit 10, a processing unit 11, a calculation unit 12, and an identification unit 13. among them
  • the acquisition unit 10 is configured to acquire a lung sound signal.
  • the processing unit 11 is configured to process the acquired lung sound signal to extract a wet rales component of a breathing cycle.
  • the calculating unit 12 is configured to calculate a power spectrum of the wet rales component, and calculate, based on the power spectrum, a ratio of the power of each of the preset frequency bands to the total power of all the preset frequency bands and all the pre- Set the total power of the frequency band, select at least one as the frequency domain parameter, and/or calculate the proportion of the late inspiratory wet rales of the respiratory cycle to the total number of occurrences of the entire inspiratory wet rales and the entire inspiratory phase. For the maximum amplitude of the wet rales, select at least one as the time domain parameter.
  • the identification unit 13 is configured to input the obtained frequency domain parameters and/or the time domain parameters as parameter features into the classification model for classification recognition to identify popping sounds.
  • the acquisition unit 10 may be omitted, that is, the present invention is not limited to processing the directly collected lung sound signals, and the burst sound recognition system of the present invention may acquire lung sound signals obtained by other means. Process it.
  • the present invention can also process lung tone signals that are acquired in advance and stored in a computer readable medium.
  • acquisition unit 10 may include a collector (eg, an audio collector) for acquiring lung sound signals. More specifically, the collector may include a digital stethoscope, although the invention is not limited thereto.
  • the processing unit can include a processor configured to process the lung sound signal.
  • computing unit 12 may include a calculator configured to perform further calculations on the wet rales component. The processor and the calculator may be specifically configured by logic circuits, integrated circuits, dedicated processors, general purpose processors, etc., but the present invention is not limited thereto.
  • the present disclosure is based on wetness due to the fact that the pitch of the popping sound is higher, stronger and more likely to occur in the late phase of the inspiratory phase, and the amplitude and frequency of the pitch of the high, strong and wet rales are related according to clinical experience.
  • Power ratio of Luoyin in different frequency bands, total power of all bands At least one of the rates is used as the frequency domain parameter, and/or at least one of the maximum amplitude of the wet rales and the appearance ratio of the late inspiratory wet rales is used as the time domain parameter, and the recognition accuracy of the popping sound can be improved.
  • the processing unit 11 includes: a wavelet decomposition module 111 and an extraction module 112. among them
  • the wavelet decomposition module 111 is configured to perform wavelet decomposition processing on the collected lung sound signals to obtain wet rales and breath sound components.
  • the extraction module 112 is configured to determine a breathing cycle based on the breath sound component and extract a wet rake of the breathing cycle.
  • the above modules may be constituted by circuits that perform respective functions.
  • the wavelet decomposition module 111 may be constructed of a wavelet decomposition circuit configured to perform wavelet decomposition on the signal.
  • the extraction module 112 can be comprised of an extraction circuit configured to extract wet rales. More specifically, the wavelet decomposition circuit and the extraction circuit may be specifically configured by a logic circuit, an integrated circuit, a dedicated processor, a general purpose processor, etc., but the present invention is not limited thereto.
  • the collected lung sound signal is processed by using wavelet decomposition method, so that the processing process is simpler, and thus the recognition efficiency of the pop sound can be improved.
  • the wavelet decomposition comprises coif2 wavelet decomposition.
  • the coif2 wavelet decomposition method is adopted in the present disclosure because the coif2 wavelet has better symmetry, and can avoid waveform distortion of the reconstructed wet rales and breath sound components, thereby further improving the recognition of popping sound. Accuracy.
  • the number of decomposition layers is N ⁇ 6.
  • the component, an, represents the approximate component of the nth layer of wavelet decomposition, where n is an integer and 0 ⁇ n ⁇ 10.
  • the extraction module 112 is configured to obtain an average power curve of the breath sound component in a preset frequency range; the peak point of the average power curve is identified as an inspiratory gas apex, and the valley point is a respiratory gas phase switching point, Thereby obtaining respiratory gas phase information; determining a smooth breathing cycle based on the respiratory gas phase information. Determining a smooth breathing cycle in the present disclosure can better reflect the patient's lung condition, thereby facilitating the identification of popping sounds. Accuracy.
  • the calculating unit 12 is configured to select, as a frequency domain parameter, a ratio corresponding to two frequency bands with the largest difference among the ratios of the power of each preset frequency band to the total power of all the preset frequency bands.
  • the ratio corresponding to the two frequency bands with the largest difference is selected as the frequency domain parameter, and some factors of the adjacent interference frequency band can be removed, thereby improving the accuracy of identifying the popping sound.
  • the preset frequency band includes 50 Hz to 200 Hz and 500 Hz to 1000 Hz, because the ratio of the power of the two frequency bands to the total power of all the preset frequency bands is the largest.
  • the late phase of the inspiratory phase is 1/2 of the entire inspiratory phase, so that the burst sound can be ensured as far as possible in the late phase of the inspiratory phase, thereby improving the accuracy of identifying the popping sound.
  • the popping sound recognition system further includes a band pass filter configured to filter the lung sound signal.
  • burst sound recognition system provided in this embodiment is a product embodiment corresponding to the burst sound recognition method provided in the first embodiment, and the crack sound recognition method has been described in detail in the above-described first embodiment, For related features of the cracking sound recognition system and the identification method, please refer to the above embodiment 1, and details are not described herein again.
  • a popping sound recognition system including:
  • a processor configured to perform a popping sound recognition method according to the aforementioned embodiment of the present invention.
  • the pitch of the popping sound is higher, stronger and more likely to occur in the late phase of the inspiratory phase, while the pitch, frequency and frequency of the high, strong and wet rales of the pitch are related. Therefore, the present disclosure is based on the wet rales in different frequency bands. At least one of the power ratio above, the total power of all the bands is used as the frequency domain parameter; and/or at least one of the maximum amplitude of the wet rales and the appearance ratio of the late inspiratory wet rales is used as the time domain parameter, which can be improved. The accuracy of the recognition of popping sounds.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Molecular Biology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Veterinary Medicine (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Pulmonology (AREA)
  • Pathology (AREA)
  • Artificial Intelligence (AREA)
  • Physiology (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Databases & Information Systems (AREA)
  • Psychiatry (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Signal Processing (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

一种爆裂音识别方法，包括：对获取到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分(S2)；计算湿罗音成分的功率谱，基于功率谱计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例和所有预设频带的总功率中的至少一种作为频域参数；和/或计算呼吸周期的吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅中的至少一个作为时域参数(S3)；将频域参数和/或时域参数作为参数特征输入训练得到的分类模型中进行识别(S4)。还提供一种爆裂音识别系统。该爆裂音识别方法和系统，可提高爆裂音的识别准确率。

Description

爆裂音识别方法及系统
相关申请的交叉引用
本申请要求于2017年03月31日递交的中国专利申请第201710210176.3号的优先权，在此全文引用上述中国专利申请公开的内容以作为本申请的一部分。
技术领域
本公开涉及一种爆裂音识别方法及系统。
背景技术
肺部听诊音与肺部的病理状况密切相关，爆裂音作为一种特殊的湿罗音被作为提示间质性肺病的首要临床线索。但是，不同人耳在不同频带的听觉灵敏度不同，且湿罗音强度低、持续时间短、带宽广泛，导致不同医生之间诊断不一致的情况时有发生。
目前，通常采用数字化听诊来进行辅助诊断，可以弥补医生听诊的主观性和局限性，然而，现有的数字化听诊技术识别爆裂音的准确率不高。
发明内容
本公开提供了一种爆裂音识别方法，包括以下步骤：
对获取到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分；
计算所述湿罗音成分的功率谱，基于所述功率谱执行以下计算中的至少一种：计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例和所有的预设频带的总功率，选取至少一种作为频域参数；以及计算吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅，选取至少一个作为时域参数；
将获得的所述频域参数和所述时域参数中的至少一个作为参数 特征输入到分类模型中进行分类识别，以识别爆裂音。
可选地，所述对采集到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分的步骤，包括：
对采集到的肺音信号进行小波分解处理，获得湿罗音成分和呼吸音成分；
根据所述呼吸音成分确定一个呼吸周期，并提取该呼吸周期的湿罗音。
可选地，所述小波分解包括coif2小波分解。
可选地，分解层数N≥6。
可选地，分解层数N＝9；选取d1～d6小波分量累加获得湿罗音成分，d7～d9及a9小波分量累加获得呼吸音成分；其中，dn表示小波分解第n层的细节分量，an表示小波分解第n层的近似分量，n为整数，且0<n<10。
可选地，所述根据所述呼吸音成分确定一个呼吸周期的步骤，包括：
获得呼吸音成分在预设频率范围内的平均功率曲线；
识别所述平均功率曲线的峰值点为吸气相顶点，谷值点为呼吸气相切换点，从而获得呼吸气相信息；
根据所述呼吸气相信息确定一个平稳的呼吸周期。
可选地，在每个预设频带的功率占所有的预设频带的总功率的比例中，选取差别最大的两个频带对应的比例作为频域参数。
可选地，所述预设频带包括50Hz～200Hz和500Hz～1000Hz。
可选地，所述吸气相后期为整个吸气相周期的1/2。
可选地，所述对获取到的肺音信号进行处理的步骤，包括:
使用带通滤波器对肺音信号进行滤波。
可选地，所述方法还包括利用采集器获取肺音信号。
本公开还一种爆裂音识别系统，包括：
处理器，配置为对获取到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分；
计算器，配置为执行以下计算中的至少一种：计算所述湿罗音 成分的功率谱，基于所述功率谱计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例和所有的预设频带的总功率，选取至少一种作为频域参数；以及计算呼吸周期的吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅，选取至少一个作为时域参数；
识别器，配置为将获得的所述频域参数和/或所述时域参数作为参数特征输入到分类模型中进行分类识别，以识别爆裂音。
可选地，所述处理器包括：
小波分解电路，配置为对采集到的肺音信号进行小波分解处理，获得湿罗音成分和呼吸音成分
提取电路，配置为根据所述呼吸音成分确定一个呼吸周期，并提取该呼吸周期的湿罗音。
可选地，所述小波分解包括coif2小波分解。
可选地，所述提取电路，配置为获得呼吸音成分在预设频率范围内的平均功率曲线；识别所述平均功率曲线的峰值点为吸气相顶点，谷值点为呼吸气相切换点，从而获得呼吸气相信息；根据所述呼吸气相信息确定一个平稳的呼吸周期。
可选地，所述计算单元，配置为在每个预设频带的功率占所有的预设频带的总功率的比例中，选取差别最大的两个频带对应的比例作为频域参数。
可选地，所述预设频带包括50Hz～200Hz和500Hz～1000Hz。
可选地，所述吸气相后期为整个吸气相周期的1/2。
可选地，所述爆裂音识别系统还包括带通滤波器，对获取到的肺音信号进行滤波。
可选地，所述爆裂音识别系统还包括采集器，配置为获取肺音信号。
可选地，所述采集器包括数字听诊器。
本公开还提供一种爆裂音识别系统，包含：
存储器，存储肺音信号；
处理器，所述处理器配置为执行以下步骤：
对所述肺音信号进行处理以提取一个呼吸周期的湿罗音成分；
计算所述湿罗音成分的功率谱，基于所述功率谱执行以下计算中的至少一种：计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例和所有的预设频带的总功率，选取至少一种作为频域参数；以及计算吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅，选取至少一个作为时域参数；
将获得的所述频域参数和所述时域参数中的至少一个作为参数特征输入到分类模型中进行分类识别，以识别爆裂音。
附图说明
图1为本公开实施例1提供的爆裂音识别方法的流程图；
图2为本公开中的肺音信号、湿罗音成分与呼吸音成分之间的关系示意图；
图3为图1中步骤S2的流程图；
图4为图3中步骤S22的流程图；
图5为呼吸音成分的平均功率曲线图；
图6为本公开实施例2提供的爆裂音识别系统的原理框图；
图7为图6中处理单元的原理框图。
具体实施方式
为使本领域的技术人员更好地理解本公开的技术方案，下面结合附图来对本公开提供的面对爆裂音识别方法及系统进行详细描述。
图1为本公开实施例1提供的爆裂音识别方法的流程图；请参阅图1，本实施例提供的爆裂音识别方法包括：
S1，采集肺音信号。
S2，对采集到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分。
S3，计算所述湿罗音成分的功率谱，基于所述功率谱计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例 和所有的预设频带的总功率，选取至少一种作为频域参数。具体地，本公开可以将每个预设频带的功率占所有的预设频带的总功率的比例(Power Ratio，PR)作为频域参数；或，将所有的预设频带的总功率PRtotal作为频域参数；或，将每个预设频带的功率占所有的预设频带的总功率的比例(Power Ratio，PR)和所有的预设频带的总功率PRtotal均作为频域参数。
和/或，计算呼吸周期的吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅，选取至少一个作为时域参数。具体地，呼吸周期的吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例作为时域参数；或，将整个吸气相湿罗音的最大振幅作为时域参数；或，将呼吸周期的吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅均作为时域参数
S4，将获得的所述频域参数和/或所述时域参数作为参数特征输入到分类模型中进行分类识别，以识别爆裂音。
其中，分类模型为预先训练得到的且具有根据频域参数和/或时域参数识别爆裂音的功能。当将获得的所述频域参数和/或所述时域参数作为参数特征输入训练得到的分类模型时，可以直接获得识别结果。其中，训练得到的分类模型可以为支持向量机模型(SVM分类模型)。
应当注意的是，在本实施例中，步骤S1可以省略，即，本发明不限于对直接采集的肺音信号进行处理，本发明的方法可以对通过其他途径获取的肺音信号进行处理。例如，本发明也可以对通过预先获取并存储到计算机可读介质中的肺音信号进行处理。
本公开中，一个吸气相周期包含吸气相前期和吸气相后期，对应的，湿罗音可分为吸气相前期湿罗音和吸气相后期湿罗音。
本公开中，由于根据临床经验爆裂音的音调较高、较强且多出现在吸气相后期，而音调的高、强和湿罗音的振幅、频率相关，因此，本公开通过湿罗音在不同频带上的功率比例、全部频带总功率中的至少一个作为频域参数；和/或，将湿罗音的最大振幅和吸气相后期湿 罗音出现比例中的至少一个作为时域参数，可以提高爆裂音的识别准确率。
可选地，步骤S2中，使用带通滤波器对获取到的肺音信号进行滤波，以去除心音信号、工频干扰等噪音。进一步地，带通滤波器包括但不限于8阶巴特沃斯带通滤波器，该带通滤波器的通带可以为50Hz～2000Hz。
另外，在实际应用中，采集肺音信号时，可以选择肺音听诊最明显的部位采集单导连肺音的数字信号，采集时间可以为但不限于10s，采样频率fs可以为但不限于10000Hz。
可选地，如图3所示，步骤S2在对采集到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分时，可以通过如下步骤S21和S22来实现：
S21，对采集到的肺音信号进行小波分解处理，获得湿罗音成分和呼吸音成分。
S22，根据所述呼吸音成分确定一个呼吸周期，并提取该呼吸周期的湿罗音。
本公开中采用小波分解方式对采集到的肺音信号进行处理，使得处理过程更加简单，因而可以提高爆裂音识别效率。
可选地，所述小波分解为Coiflet小波(简称coif N)分解。本公开中之所以采用该coif2小波分解方式，是因为coif2小波具有更好的对称性，可以避免重构的湿罗音成分和呼吸音成分有波形失真，因此，可以进一步地提高识别爆裂音的准确度。小波分解过程：对原始信号S先分成两个分量，高频分量也叫细节分量d1，低频分量也叫近似分量a1，接着对低频分量进行第二次分解，分成两个分量，细节分量d2和近似分量a2，以此类推，直至分解到预先设定的层数N。
进一步可选地，coif2小波分解的分解层数N≥6，可以较好地避免湿罗音成分波形失真。
更进一步可选地，分解层数为9，选择d1～d6小波分量累加获得湿罗音成分，d7～d9及a9小波分量累加获得呼吸音成分，其中，dn 表示小波分解第n层的细节分量，an表示小波分解第n层的近似分量，n为整数，且0<n<10。这是因为：根据coif2小波分解的相关内容，第n层分量高频段对应的频率范围为fs/2n+1～fs/2n，1≤n≤N，虽然爆裂音分布的主要频带在200Hz～1000Hz，但是，爆裂音音调较其它湿罗音高，湿罗音整体分布频带在50Hz～2500Hz，因此，选择细节分量d1～d6小波分量累加获得湿罗音成分，细节分量d7～d9及近似分量a9小波分量累加获得呼吸音成分，可以很好地避免湿罗音成分和呼吸音成分有波形失真。
具体地，请参阅图2，图2为通过本公开中肺音信号、湿罗音成分与呼吸音成分之间的关系示意图，从图2可以直接看出：肺音信号中的湿罗音成分可以很清楚地看到，如黑色箭头所示，能够很好地避免湿罗音成分和呼吸音成分有波形失真。
在此说明的是，在实际应用中，小波分解还可以采用其他分解方式。
可选地，如图4所示，所述步骤S22中根据呼吸音成分确定一个呼吸周期的步骤，可以通过如下步骤S221至步骤S223来实现：
S221，获得呼吸音成分在预设频率范围内的平均功率曲线。
其中，所述预设频率范围可以为但不限于150～450Hz。选择150Hz～450Hz频带范围是因为呼吸气相在这个频带范围内的能量差异最显著。
具体地，为获得呼吸音成分在预设频率范围内的平均功率曲线，可以首先将呼吸音成分分为100ms时长的窗口，窗移75ms，通过如下公式(1)计算每一个窗口内呼吸音信号的傅里叶变换并取其模的平方，得到呼吸音信号的短时功率谱为：
Figure PCTCN2017116491-appb-000001
其中x(m)为呼吸音成分的第m个窗，tn、fk分别为时间和频率，w(tnD-m)是时长为T、窗移为D的移动窗函数。其中，本公开使用的是汉宁窗，当然也可以采用其它窗函数。
接着，按照如下公式(2)计算呼吸音成分在150Hz～450Hz频 带内的平均功率曲线，如图5所示。
Figure PCTCN2017116491-appb-000002
其中，fhigh为450Hz，flow为150Hz。
S222，识别平均功率曲线的峰值点为吸气相顶点(如图6中的A1、A2和A3)，平均功率曲线的谷值点为呼吸气相切换点(如图6中的B1和B2)，从而获得呼吸气相信息。
S223，根据呼吸气相信息确定一个平稳的呼吸周期。
其中，吸气相顶点所在波形的起始时刻至呼吸气相切换点所在波形的终止时刻为一个呼吸周期，如图5中的T1-T3；所谓平稳的呼吸周期是指功率曲线相对平稳的呼吸周期，T1和T2相比，T2相对较平缓，因此，选取T2作为一个平稳的呼吸周期。
本公开中之所以要确定一个平稳的呼吸周期，是因为通过一个平稳的呼吸周期可以更好地反应患者的肺部情况，从而有利于提高识别爆裂音的准确度。
可选地，本公开可以在每个预设频带的功率占所有的预设频带的总功率的比例中，选取差别最大的两个频带对应的比例作为频域参数，这相当于去除掉一些相近的干扰频带的因素，从而可以提高识别爆裂音的准确度。
进一步可选地，预设频带包括50Hz～200Hz和500Hz～1000Hz，这是因为这两个频带的功率占所有的预设频带的总功率的比例的差别最大。在此情况下，具体地，预设频带可以包括但不限于：50Hz～200Hz、200Hz～500Hz、500Hz～1000Hz和1000Hz～1500Hz四个预设频带。
可选地，本公开中的所述吸气相后期为整个吸气相周期的1/2，也即将吸气相平均分为吸气相前期和吸气相后期，这样，可以尽可能地确保爆裂音位于吸气相后期，从而可提高识别爆裂音的准确度。
具体地，在步骤S3中可以但不限于使用多阶自回归模型计算一个呼吸周期湿罗音成分的功率谱，例如，在本实施例中，可以采用 14阶自回归模型计算一个呼吸周期湿罗音成分的功率谱。
实施例2
图6为本公开实施例2提供的爆裂音识别系统的原理框图，请参阅图6，该爆裂音识别系统，包括采集单元10、处理单元11、计算单元12、识别单元13。其中
采集单元10配置为采集肺音信号。
处理单元11配置为对采集到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分。
计算单元12配置为计算所述湿罗音成分的功率谱，基于所述功率谱计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例和所有的预设频带的总功率，选取至少一种作为频域参数，和/或，计算呼吸周期的吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅，选取至少一个作为时域参数。
识别单元13配置为将获得的所述频域参数和/或所述时域参数作为参数特征输入到分类模型中进行分类识别，以识别爆裂音。
应当注意的是，在本实施例中，采集单元10可以省略，即，本发明不限于对直接采集的肺音信号进行处理，本发明的爆裂音识别系统可以对通过其他途径获取的肺音信号进行处理。例如，本发明也可以对通过预先获取并存储到计算机可读介质中的肺音信号进行处理。
本公开的上述各单元可以由相应的硬件实现。例如，采集单元10可以包括采集器(例如音频采集器)，以用于采集肺音信号。更具体地，采集器可以包括数字听诊器，然而本发明不限于此。另外，处理单元可以包括处理器，其配置为对肺音信号进行处理。另外，计算单元12可以包括计算器，其配置为对湿罗音成分进行进一步的计算。处理器和计算器可以由逻辑电路，集成电路，专用处理器，通用处理器等具体地配置，然而本发明不限于此。
本公开中，由于根据临床经验，爆裂音的音调较高、较强且多出现在吸气相后期，而音调的高、强和湿罗音的振幅、频率相关，因此，本公开通过基于湿罗音在不同频带上的功率比例、全部频带总功 率中的至少一个作为频域参数，和/或，将湿罗音的最大振幅和吸气相后期湿罗音出现比例中的至少一个作为时域参数，可以提高爆裂音的识别准确率。
可选地，如图7所示，所述处理单元11包括：小波分解模块111和提取模块112。其中
小波分解模块111配置为对采集到的肺音信号进行小波分解处理，获得湿罗音成分和呼吸音成分
提取模块112配置为根据所述呼吸音成分确定一个呼吸周期，并提取该呼吸周期的湿罗音。
在本发明中，上述模块可以由执行相应功能的电路构成。例如，小波分解模块111可以由小波分解电路构成，其配置为对信号进行小波分解。另外，提取模块112可以由提取电路构成，其配置为提取湿罗音。更具体地说，小波分解电路和提取电路可以由逻辑电路，集成电路，专用处理器，通用处理器等具体地配置，然而本发明不限于此。
本公开中采用小波分解方式对采集到的肺音信号进行处理，使得处理过程更加简单，因而可以提高爆裂音的识别效率。
可选地，所述小波分解包括coif2小波分解。本公开中之所以采用该coif2小波分解方式，是因为coif2小波具有更好的对称性，可以避免重构的湿罗音成分和呼吸音成分有波形失真，因此，可以进一步地提高识别爆裂音的准确度。
进一步可选地，分解层数N≥6。
更进一步可选地，分解层数N＝9；选取d1～d6小波分量累加获得湿罗音成分，d7～d9及a9小波分量累加获得呼吸音成分；其中，dn表示小波分解第n层的细节分量，an表示小波分解第n层的近似分量，n为整数，且0<n<10。
可选地，所述提取模块112配置为获得呼吸音成分在预设频率范围内的平均功率曲线；识别所述平均功率曲线的峰值点为吸气相顶点，谷值点为呼吸气相切换点，从而获得呼吸气相信息；根据所述呼吸气相信息确定一个平稳的呼吸周期。本公开中确定一个平稳的呼吸周期可以更好地反应患者的肺部情况，从而有利于提高识别爆裂音的 准确度。
可选地，所述计算单元12配置为在每个预设频带的功率占所有的预设频带的总功率的比例中选取差别最大的两个频带对应的比例作为频域参数。本公开中选取差别最大的两个频带对应的比例作为频域参数，可以去除掉一些相近的干扰频带的因素，从而可以提高识别爆裂音的准确度。
其中，所述预设频带包括50Hz～200Hz和500Hz～1000Hz，这是因为这两个频带的功率占所有的预设频带的总功率的比例的差别最大。
可选地，所述吸气相后期为整个吸气相周期的1/2，这样，可以尽可能地保证爆裂音位于吸气相后期，从而可提高识别爆裂音的准确度。
可选地，根据本实施例的爆裂音识别系统还包括带通滤波器，配置为对肺音信号进行滤波。
需要说明的是，本实施例提供的爆裂音识别系统为上述实施例1提供的爆裂音识别方法相对应的产品实施例，在上述实施例1中已经对爆裂音识别方法进行的详细描述，因此，有关爆裂音识别系统与识别方法的相关特征请参见上述实施例1，在此不再赘述。
另外，本发明的另一方面还提供一种爆裂音识别系统，包含：
存储器，存储肺音信号；
处理器，所述处理器配置为执行根据本发明前述实施例的爆裂音识别方法。
本公开具有以下有益效果：
根据临床经验，爆裂音的音调较高、较强且多出现在吸气相后期，而音调的高、强和湿罗音的振幅、频率相关，因此，本公开通过基于湿罗音在不同频带上的功率比例、全部频带总功率中的至少一个作为频域参数；和/或，将湿罗音的最大振幅和吸气相后期湿罗音出现比例中的至少一个作为时域参数，可以提高爆裂音的识别准确率。
可以理解的是，以上实施方式仅仅是为了说明本公开的原理而采用的示例性实施方式，然而本公开并不局限于此。对于本领域内的 普通技术人员而言，在不脱离本公开的精神和实质的情况下，可以做出各种变型和改进，这些变型和改进也视为本公开的保护范围。

Claims (23)

  1. 一种爆裂音识别方法，包括以下步骤：
    对获取到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分；
    计算所述湿罗音成分的功率谱，基于所述功率谱执行以下计算中的至少一种：计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例和所有的预设频带的总功率，选取至少一种作为频域参数；以及计算吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅，选取至少一个作为时域参数；
    将获得的所述频域参数和所述时域参数中的至少一个作为参数特征输入到分类模型中进行分类识别，以识别爆裂音。
  2. 根据权利要求1所述的爆裂音识别方法，其中，所述对采集到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分的步骤，包括：
    对采集到的肺音信号进行小波分解处理，获得湿罗音成分和呼吸音成分；
    根据所述呼吸音成分确定一个呼吸周期，并提取该呼吸周期的湿罗音。
  3. 根据权利要求2所述的爆裂音识别方法，其中，所述小波分解包括coif2小波分解。
  4. 根据权利要求3所述的爆裂音识别方法，其中，分解层数N≥6。
  5. 根据权利要求4所述的爆裂音识别方法，其中，分解层数N＝9；
    选取d1～d6小波分量累加获得湿罗音成分，d7～d9及a9小波分 量累加获得呼吸音成分；
    其中，dn表示小波分解第n层的细节分量，an表示小波分解第n层的近似分量，n为整数，且0<n<10。
  6. 根据权利要求2所述的爆裂音识别方法，其中，所述根据所述呼吸音成分确定一个呼吸周期的步骤，包括：
    获得呼吸音成分在预设频率范围内的平均功率曲线；
    识别所述平均功率曲线的峰值点为吸气相顶点，谷值点为呼吸气相切换点，从而获得呼吸气相信息；
    根据所述呼吸气相信息确定一个平稳的呼吸周期。
  7. 根据权利要求1-6任一项所述的爆裂音识别方法，其中，在每个预设频带的功率占所有的预设频带的总功率的比例中，选取差别最大的两个频带对应的比例作为频域参数。
  8. 根据权利要求1-7任一项所述的爆裂音识别方法，其中，所述预设频带包括50Hz～200Hz和500Hz～1000Hz。
  9. 根据权利要求1-8任一项所述的爆裂音识别方法，其中，所述吸气相后期为整个吸气相周期的1/2。
  10. 根据权利要求1-9任一项所述的爆裂音识别方法，其中，所述对获取到的肺音信号进行处理的步骤，包括:
    使用带通滤波器对肺音信号进行滤波。
  11. 根据权利要求1所述的爆裂音识别方法，其中，所述分类模型是通过预先训练得到的，所述分类模型被配置为根据频域参数和时域参数中的至少一个识别爆裂音。
  12. 根据权利要求1所述的爆裂音识别方法，还包括：
    利用采集器获取肺音信号。
  13. 一种爆裂音识别系统，包括：
    处理器，配置为对获取到的肺音信号进行处理以提取一个呼吸周期的湿罗音成分；
    计算器，配置为执行以下计算中的至少一种：计算所述湿罗音成分的功率谱，基于所述功率谱计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例和所有的预设频带的总功率，选取至少一种作为频域参数；以及计算呼吸周期的吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅，选取至少一个作为时域参数；
    识别器，配置为将获得的所述频域参数和/或所述时域参数作为参数特征输入到分类模型中进行分类识别，以识别爆裂音。
  14. 根据权利要求13所述的爆裂音识别系统，其中，所述处理器包括：
    小波分解电路，配置为对采集到的肺音信号进行小波分解处理，获得湿罗音成分和呼吸音成分；
    提取电路，配置为根据所述呼吸音成分确定一个呼吸周期，并提取该呼吸周期的湿罗音。
  15. 根据权利要求14所述的爆裂音识别系统，其中，所述小波分解包括coif2小波分解。
  16. 根据权利要求14所述的爆裂音识别系统，其中，所述提取电路，配置为获得呼吸音成分在预设频率范围内的平均功率曲线；识别所述平均功率曲线的峰值点为吸气相顶点，谷值点为呼吸气相切换点，从而获得呼吸气相信息；根据所述呼吸气相信息确定一个平稳的呼吸周期。
  17. 根据权利要求13所述的爆裂音识别系统，其中，所述计算器，配置为在每个预设频带的功率占所有的预设频带的总功率的比例中，选取差别最大的两个频带对应的比例作为频域参数。
  18. 根据权利要求17所述的爆裂音识别系统，其中，所述预设频带包括50Hz～200Hz和500Hz～1000Hz。
  19. 根据权利要求13所述的爆裂音识别系统，其中，所述吸气相后期为整个吸气相周期的1/2。
  20. 根据权利要求13所述的爆裂音识别系统，还包括带通滤波器，对获取到的肺音信号进行滤波。
  21. 根据权利要求13所述的爆裂音识别系统，还包括：
    采集器，配置为获取肺音信号。
  22. 根据权利要求21所述的爆裂音识别系统，其中所述采集器包括数字听诊器。
  23. 一种爆裂音识别系统，包含：
    存储器，存储肺音信号；
    处理器，所述处理器配置为执行以下步骤：
    对所述肺音信号进行处理以提取一个呼吸周期的湿罗音成分；
    计算所述湿罗音成分的功率谱，基于所述功率谱执行以下计算中的至少一种：计算多个预设频带中每个预设频带的功率占所有的预设频带的总功率的比例和所有的预设频带的总功率，选取至少一种作为频域参数；以及计算吸气相后期湿罗音出现个数占整个吸气相湿罗音出现总数的比例和整个吸气相湿罗音的最大振幅，选取至少一个作为时域参数；
    将获得的所述频域参数和所述时域参数中的至少一个作为参数特征输入到分类模型中进行分类识别，以识别爆裂音。
PCT/CN2017/116491 2017-03-31 2017-12-15 爆裂音识别方法及系统 WO2018176919A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/081,880 US11660062B2 (en) 2017-03-31 2017-12-15 Method and system for recognizing crackles

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710210176.3 2017-03-31
CN201710210176.3A CN108652658A (zh) 2017-03-31 2017-03-31 爆裂音识别方法及系统

Publications (1)

Publication Number Publication Date
WO2018176919A1 true WO2018176919A1 (zh) 2018-10-04

Family

ID=63674157

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/116491 WO2018176919A1 (zh) 2017-03-31 2017-12-15 爆裂音识别方法及系统

Country Status (3)

Country Link
US (1) US11660062B2 (zh)
CN (1) CN108652658A (zh)
WO (1) WO2018176919A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111931728B (zh) * 2020-09-23 2021-01-12 杭州百世伽信息科技有限公司 自动提取湿罗音特征的方法
CN112668556B (zh) * 2021-01-21 2024-06-07 广东白云学院 一种呼吸音识别方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2011155048A1 (ja) * 2010-06-10 2011-12-15 富士通株式会社 音声処理装置および呼吸検出方法
CN103932733A (zh) * 2014-04-11 2014-07-23 中国人民解放军第三军医大学第三附属医院 一种基于肺音的肺间质纤维化的数字化检测分析方法
CN105662454A (zh) * 2016-01-08 2016-06-15 中国科学院声学研究所 一种罗音检测的方法和装置
CN105913066A (zh) * 2016-04-13 2016-08-31 刘国栋 一种基于相关向量机的数字化肺音特征降维方法
CN106251880A (zh) * 2015-06-03 2016-12-21 创心医电股份有限公司 识别生理声音的方法以及系统

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6171258B1 (en) * 1998-10-08 2001-01-09 Sleep Solutions, Inc. Multi-channel self-contained apparatus and method for diagnosis of sleep disorders
WO2002071947A1 (en) * 2001-03-09 2002-09-19 Biomedical Acoustic Research, Inc. Acoustic detection of gastric motility dysfunction
US20080058607A1 (en) * 2006-08-08 2008-03-06 Zargis Medical Corp Categorizing automatically generated physiological data based on industry guidelines
US9492096B2 (en) * 2009-11-03 2016-11-15 Vivaquant Llc ECG sensing apparatuses, systems and methods
US8758262B2 (en) * 2009-11-25 2014-06-24 University Of Rochester Respiratory disease monitoring system
US8702628B2 (en) * 2010-03-18 2014-04-22 Panasonic Corporation Physiological sound examination device
WO2012042611A1 (ja) * 2010-09-29 2012-04-05 富士通株式会社 呼吸検出装置および呼吸検出方法
US10595813B2 (en) * 2011-09-01 2020-03-24 Medtronic, Inc. Method and apparatus for monitoring cardiac and respiratory conditions using acoustic sounds
US10786164B2 (en) * 2015-12-12 2020-09-29 Verily Life Sciences Llc Method for improving heart rate estimates by combining multiple measurement modalities

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2011155048A1 (ja) * 2010-06-10 2011-12-15 富士通株式会社 音声処理装置および呼吸検出方法
CN103932733A (zh) * 2014-04-11 2014-07-23 中国人民解放军第三军医大学第三附属医院 一种基于肺音的肺间质纤维化的数字化检测分析方法
CN106251880A (zh) * 2015-06-03 2016-12-21 创心医电股份有限公司 识别生理声音的方法以及系统
CN105662454A (zh) * 2016-01-08 2016-06-15 中国科学院声学研究所 一种罗音检测的方法和装置
CN105913066A (zh) * 2016-04-13 2016-08-31 刘国栋 一种基于相关向量机的数字化肺音特征降维方法

Also Published As

Publication number Publication date
US20210007703A1 (en) 2021-01-14
CN108652658A (zh) 2018-10-16
US11660062B2 (en) 2023-05-30

Similar Documents

Publication Publication Date Title
CN110570880B (zh) 一种鼾声信号识别方法
CN106037655B (zh) 一种睡眠周期信号的分离及周期计算方法及装置
US11304662B2 (en) Lung-sound signal processing method, processing device, and readable storage medium
CN110970042B (zh) 一种电子听诊器的肺部啰音人工智能实时分类方法、系统、装置及可读存储介质
CN102283670A (zh) 一种心音信号量化分析方法及装置
CN111640439A (zh) 一种基于深度学习的呼吸音分类方法
Zakaria et al. Three resnet deep learning architectures applied in pulmonary pathologies classification
CN104473660B (zh) 一种基于子带能量包络自相关特征的异常心音识别方法
Lin et al. Wheeze recognition based on 2D bilateral filtering of spectrogram
WO2020151169A1 (zh) 一种电子听诊器的摩擦音干扰自动消除方法
WO2018176919A1 (zh) 爆裂音识别方法及系统
CN110942784A (zh) 基于支持向量机的鼾声分类系统
Abdollahpur et al. Cycle selection and neuro-voting system for classifying heart sound recordings
Rizal et al. Comparison of discrete wavelet transform and wavelet packet decomposition for the lung sound classification
CN112949639A (zh) 一种先天性心脏病心音智能分割分类算法、装置及存储介质
Singh et al. An Artificial Intelligence based mobile solution for early detection of valvular heart diseases
CN113476074A (zh) 基于智能穿戴系统的肺音实时监测方法
Huang et al. ECG baseline wander correction based on ensemble empirical mode decomposition with complementary adaptive noise
KR20160147591A (ko) 천식 진단을 위한 청진음 신호의 전처리 방법 및 장치
CN109036472B (zh) 一种改进的病理嗓音基音频率提取方法
CN112908291A (zh) 一种电子听诊器的毛刺噪声消除方法
Ihza et al. Study of Denoising Method to Detect Valvular Heart Disease Using Phonocardiogram (PCG)
CN115040135A (zh) 一种基于心电心音的儿童心血管疾病诊断方法
Shah et al. Separation of cardiorespiratory sounds using time-frequency masking and sparsity
Torre-Cruz et al. Wheezing sound separation based on constrained non-negative matrix factorization

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: 17904230

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: 17904230

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 16.03.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17904230

Country of ref document: EP

Kind code of ref document: A1