TWI809825B - System for diagnosing and monitoring abnormal lung rales as well as establishing method of the system - Google Patents

Abstract

A system for diagnosing and monitoring abnormal lung rales includes a lung sound acquisition device and a processor. The lung sound acquisition device acquires a lung sound of a person at a sampling frequency, and convert the lung sound into a lung sound signal. The processor receives the lung sound signal and convert the lung sound signal to a frequency domain, retrieves a power spectral density curve within a range of frequency bands, selects a specific frequency band from the range of frequency bands, and calculates a specific band power and an all band power which correspond, respectively, to the specific frequency band and the range of frequency bands according to the power spectral density curve. Furthermore, a determining parameter is acquired by substituting a specific band power percentage, which corresponds to the specific frequency band, and the all band power into a logarithmic determining function. When the determining parameter falls in a normal range, the processor determines that the lung sound of the person is normal; when the determining parameter falls in an abnormal range, the processor determines that the lung sound of the person is abnormal.

Description

Abnormal pulmonary rales diagnosis and monitoring system and its establishment method

The present invention relates to a diagnostic monitoring system for abnormal pulmonary rales and a method for establishing a diagnostic monitoring system for abnormal pulmonary rales, especially to a diagnostic monitoring system for abnormal pulmonary rales that converts lung sound signals in the frequency domain for lung sound judgment and its Build method.

Generally, thoracic physicians usually use non-invasive auscultation to diagnose the patient's condition first. Physicians listen to the patient's lung sounds through a stethoscope. Lung sounds can bring doctors a lot of judgment information, among which lung rales ), or crackle, is a discontinuous abnormal lung sound during lung auscultation, and is an important physical examination feature that doctors rely on to diagnose diseases such as pneumonia, pulmonary edema, and acute lung injury. With the advancement of medical treatment and technology, more and more medical instruments have been developed to assist in diagnosing the condition of the human body to assist doctors in judging symptoms. For example, the Republic of China Patent Application No. 098109075 "Sputum Sound Detection, Identification and Health Education System". The sound sensor, the sound sensor is set in the mechanical auscultation sound collection plate, and is used to convert the collected breath sound into the breath sound analog signal; a conversion circuit includes an analog circuit, an analog multiplexer and a digital circuit, used to To convert the breath sound analog signal into a breath sound digital signal; an analysis device for storing and analyzing the breath sound digital signal; a phlegm sound identification module for identifying the phlegm sound signal of the breath sound digital signal; a computing unit, It is electrically connected with the sputum sound recognition module, using To receive and process the sputum sound signal, and generate a processing result; a warning device, electrically connected with the computing unit, is used for not acting or generating a corresponding warning message according to the processing result of the computing unit. The above-mentioned patent is to judge the sputum sound by extracting breath sounds.

Therefore, how to extract meaningful judgment information for diagnosing diseases from the stethoscope is the goal of continuous research in this field.

Therefore, the present inventor proposes a diagnostic monitoring system for abnormal pulmonary rales and a diagnostic monitoring system for establishing abnormal pulmonary rales in order to extract the conditions that can monitor and assist in the diagnosis of pulmonary rales from the obtained information of lung sounds. systematic approach.

The abnormal pulmonary rales diagnosis and monitoring system includes a lung sound acquisition device and a processor. The lung sound capture device captures the lung sound of a human body with a sampling frequency, and converts it into a lung sound signal; the processor includes a filter unit, and the processor signal is connected to the lung sound capture device, the processing The device receives the lung sound signal, converts the lung sound signal in the frequency domain through the filter unit, and extracts a power spectral density curve in a frequency range; the processor selects a special frequency band from the frequency range , and calculate a special frequency band power and a full frequency band power respectively defined by the power spectral density curve in the special frequency band and the frequency range; the processor is based on a special frequency band power percentage corresponding to the full frequency band power , and the full-band power is substituted into at least one judging function of the logarithmic relationship to obtain at least one judging parameter; when the at least one judging parameter falls within a normal range, the processor judges that the human body is a normal lung sound; when the at least one judging parameter When falling into an abnormal range, the processor judges that the human body has abnormal pulmonary rales.

Further, the processor also includes a lung sound judgment database, the lung sound judgment database includes a reference normal range, a reference abnormal range and a cut-off point range, the cut-off point range is within the reference Between the reference normal range and the reference abnormal range, the processor is controlled to set two cut-off points within the cut-off point range, and will approach the reference normal range between the cut-off point and the reference normal range The range and the reference normal range are set as the normal range, and the range between the cutting point approaching the reference abnormal range and the reference abnormal range and the reference abnormal range are set as the abnormal range, and the reference normal The range, the reference abnormal range, and the boundary of the cut point range are obtained by analyzing the receiver operating characteristic curve and Youden index.

Further, the processor also includes a lung sound judgment database, the lung sound judgment database includes a reference normal range, a reference abnormal range and a cut point range, the cut point range is between the reference normal range and the reference abnormal Between ranges, the processor is controlled to set a cut point within the range of the cut point, and set the range between the cut point to the reference normal range and the reference normal range as the normal range, and set the The range between the cutting point and the reference abnormal range and the reference abnormal range are set as the abnormal range, and the reference normal range, the reference abnormal range, and the boundaries of the cutting point range are determined by the receiver operating characteristic curve and approximately Obtained from Deng index analysis.

Further, the frequency range is from 50 Hz to 1000 Hz, the special frequency band is from 100 Hz to 150 Hz, and the number of the judgment functions is two, one of which is a base 10, and the power percentage of the special frequency band is logarithmic, and the other The judgment function is based on 10, and the logarithm of the full-band power is taken. The reference normal range is that the output value of one of the judgment functions is less than -1.51, and the output value of the other judgment function is less than 12.3. The reference abnormal range The output value of one of the judgment functions is greater than -1.31, the output value of the other judgment function is greater than 12.6, and the range of each cut point is that the output value of one of the judgment functions is greater than or equal to -1.51 to less than or equal to -1.31 Between, the output value of the other judging function is greater than or equal to 12.3 to less than or equal to 12.6.

Further, the frequency range is from 50 Hz to 1000 Hz, the special frequency band is from 100 Hz to 150 Hz, and the number of the judgment functions is two, one of which is the result of taking the logarithm of the power percentage of the special frequency band with base 10. The other judgment function is twice the result of taking the logarithm of the full-band power with base 10. The reference normal range is that the sum of the output values of these judgment functions is less than 16.5. The reference abnormal range is that of these judgment functions The sum of the output values of these judging functions is greater than or equal to 16.5 to less than or equal to 17.3.

Further, the abnormal pulmonary rales diagnosis and monitoring system also includes a lung sound judgment database, the signal of the lung sound judgment database is connected to the processor, the lung sound judgment database includes the normal range and the abnormal range, the normal range and The abnormal range is based on the standard lung sound signal identified by thoracic medical experts, and the at least one judging function is set as a feature vector to be classified through a supervised machine learning algorithm.

Further, the frequency range is from 50 Hz to 1000 Hz, the special frequency band is from 100 Hz to 150 Hz, and the number of the judgment functions is two, one of which is a base 10, and the power percentage of the special frequency band is logarithmic, and the other The judgment function is to take the logarithm of the full-band power with base 10.

Further, the frequency range is from 50 Hz to 1000 Hz, the special frequency band is from 100 Hz to 150 Hz, and the number of the judgment functions is two, one of which is the result of taking the logarithm of the power percentage of the special frequency band with base 10. times, another judgment function is twice the result of taking the logarithm of the full-band power with base 10.

The method for establishing a diagnostic monitoring system for abnormal pulmonary rales includes a step of setting a range, and the step of setting a range includes the following steps.

A processor includes a filter unit, the filter unit performs frequency domain conversion on a plurality of reference lung sound signals, and the processor extracts a frequency between 50Hz and 1000Hz from each reference lung sound signal through the filter unit a reference power spectral density curve for a segment; the processor selects a specific frequency band between 100 Hz and 150 Hz from the frequency segment, and calculates each reference power spectral density curve in the specific frequency band and the frequency segment A reference special frequency band power and a reference full frequency band power respectively defined; the processor compares the reference special frequency band power of each reference power spectral density curve to a reference special frequency band power percentage of the reference full frequency band power, and the reference The full-band power is respectively substituted into the two judgment functions of the logarithmic relationship, one of which is to take the logarithm of the power percentage of the special frequency band with a base of 10, and the other judgment function is to take the logarithm of the full-band power with a base of 10, thus Two reference judgment parameters of each reference power spectral density curve are obtained.

Further, the method for establishing a diagnostic monitoring system for abnormal pulmonary rales also includes a step of establishing a lung sound judgment database for monitoring, the step of establishing a lung sound judgment database for monitoring is: the processor converts all reference power spectral density curves The at least one reference judgment parameter and the diagnostic results of the actual lung sounds of the human body are analyzed to obtain a reference normal range, a reference abnormal range, and the boundaries of the cut-off point range through receiver operating characteristic curve and Youden index analysis; wherein, the The reference normal range is that the output value of one of the judgment functions is less than -1.51 and the output value of the other judgment function is less than 12.3, which is defined as normal lung sound; the reference abnormal range is that the output value of one of the judgment functions is greater than -1.31 And the output value of the other judging function is greater than 12.6, which is defined as abnormal pulmonary rales; The output value of the judging function is greater than or equal to 12.3 to less than or equal to 12.6, so the range of the segmentation point is between the reference normal range and the reference abnormal range; the processor is controlled within the reference normal range, the reference abnormal range scope And at least a cut point is set in the cut point range, and a normal range and an abnormal range are established according to the set at least cut point, the reference normal range, and the reference abnormal range.

The method for establishing a diagnosis and monitoring system for abnormal lung rales also includes a step of establishing a lung sound judgment database for auxiliary diagnosis. The step of establishing a lung sound judgment database for auxiliary diagnosis is: The lung sound signal is set as a feature vector by at least one judgment function, and is classified through a supervised machine learning algorithm to obtain a normal range and an abnormal range of a prediction model, wherein the normal range is normal lung sound, and the abnormal The range is abnormal pulmonary rales.

According to the above-mentioned technical features, the following effects can be achieved:

1. By converting the lung sound signal converted into a spectrum, and then calculating the specific frequency band power percentage and the full frequency band power, substituting the at least one judgment function of the logarithmic relationship to obtain the at least one judgment parameter, the at least one Judgment parameters are meaningful information for judging whether the lung sounds of the human body are abnormal, and then judging whether the lung sounds of the human body are abnormal by at least one judgment parameter falling into the normal range or the abnormal range, so as to achieve the purpose of assisting doctors in diagnosing symptoms or monitoring .

2. The normal range, the abnormal range, and the boundaries of the cut-off point range can be obtained objectively through the receiver operating characteristic curve and Youden index analysis. Users can adjust the cut-off point by themselves to improve the diagnostic accuracy, The purpose of improving diagnostic specificity or improving diagnostic sensitivity.

3. According to the standard lung sound signal identified by thoracic medical experts, and the above judgment functions are set as feature vectors, the lung sound judgment database established by the supervised machine learning algorithm can also have an accuracy rate of 80%. critical result.

1: Lung sound capture device

2: output device

3: Processor

31: Filter unit

4: Lung sound judgment database

S1: Steps for setting judgment parameters

S11: frequency domain conversion sub-step

S12: Calculating power sub-step

S13: Calculation parameter sub-step

S2: Steps for establishing a lung sound judgment database for monitoring

S21: Find out the scope substep

S22: adjust range sub-step

S3: Steps for establishing a lung sound judgment database for auxiliary diagnosis

[The first figure] is a block diagram illustrating the circuit block of the abnormal pulmonary rale diagnosis and monitoring system of the present invention.

[The second figure] is a flowchart illustrating a first embodiment of the method of the present invention for establishing a diagnosis and monitoring system for abnormal pulmonary rales.

[The third figure] is a schematic diagram of a power spectrum, illustrating a reference power spectral density curve.

[Figure 4] is a schematic diagram illustrating a reference normal range, a reference abnormal range, and cut point ranges of each judgment function of the first embodiment.

[Fifth Figure A] is a flowchart illustrating the monitoring process of the abnormal pulmonary rales diagnostic monitoring system established by the first embodiment.

[Fifth Figure B] is a flowchart illustrating the process of implementing the abnormal pulmonary rales diagnosis and monitoring system established by the first embodiment for auxiliary diagnosis.

[Figure 6] is a flow chart illustrating a second embodiment of the method for establishing a diagnostic monitoring system for abnormal pulmonary rales of the present invention.

[Figure 7] is a schematic diagram illustrating the reference normal range, the reference abnormal range and the cut point range of the addition judgment function of the second embodiment.

[Eighth Figure A] is a flowchart illustrating the monitoring process of the abnormal pulmonary rales diagnostic monitoring system established by the second embodiment.

[Eighth Figure B] is a flowchart illustrating the process of implementing the abnormal pulmonary rales diagnosis and monitoring system established by the second embodiment for auxiliary diagnosis.

Based on the above technical features, the main functions of the abnormal pulmonary rale diagnostic monitoring system and the method for establishing the abnormal pulmonary rale diagnostic monitoring system of the present invention will be clearly presented in the following embodiments.

Referring to the first figure and the second figure, a first embodiment of the abnormal pulmonary rales diagnostic monitoring system of the present invention, the abnormal pulmonary rales diagnostic monitoring system is suitable for a test subject, and the lung sound of the test subject is captured and Determine whether the lung sounds are normal or abnormal, and assist in diagnosing whether the subject has pneumonia, pulmonary edema, acute lung injury and other diseases. The abnormal pulmonary rales diagnosis and monitoring system includes a lung sound acquisition device 1 , an output device 2 , a processor 3 , and a lung sound judgment database 4 .

The lung sound acquisition device 1 acquires the lung sound of the subject's human body with a sampling frequency, and converts it into a lung sound signal accordingly. The main structure of the lung sound capture device 1 is, for example, to use a stethoscope and a microphone to capture lung sounds, and then process and convert them into lung sound signals through a sound card. The lung sound capture device 1 is, for example, an electronic stethoscope. In this example, the sampling frequency is 8kHz, which is based on the Nyquist sampling theorem. The lung sound file of the lung sound signal is in Waveform Audio File Format (WAVE), and the bit depth of the lung sound file is 16 bits.

The output device 2 is used for outputting a diagnosis result. The output device 2 is, for example, a screen or a voice player. The diagnosis result is displayed on the screen or played by the voice player, without limitation.

The processor 3 is signally connected to the lung sound capture device 1 and the output device 2 . The processor 3 includes a filtering unit 31 . The spectral range of abnormal lung sounds caused by symptoms is mostly between 50Hz and 1000Hz. Therefore, the filter unit 31 is configured to retain signals within the spectrum range from 50 Hz to 1000 Hz, and filter out signals below 50 Hz and signals above 1000 Hz. The processor 3 stores at least one judgment function.

The lung sound judgment database 4 is connected to the processor 3 through signals. The lung sound judgment database 4 includes a reference normal range, a reference abnormal range and cut-off point range for monitoring. The cut-off point range is between the reference normal range and the reference abnormal range. The reference normal range, the reference abnormal range and the cut-off point range are output results corresponding to the at least one judgment function, wherein the reference normal range is normal lung sounds, and the reference abnormal range is abnormal lung rales. The above-mentioned reference normal range, the reference difference The normal range and the boundary of the cut point range are obtained by analyzing the receiver operating characteristic curve and Youden index.

The lung sound judgment database 4 further includes a normal range and an abnormal range of a prediction model for auxiliary diagnosis. The normal range and the abnormal range are output results corresponding to the at least one judgment function, wherein the normal range is normal lung sounds, and the abnormal range is abnormal lung rales. The normal range and the abnormal range of the prediction model are obtained by a supervised machine learning algorithm.

Next, the method for establishing the diagnosis and monitoring system for abnormal pulmonary rales will be described firstly. By means of the method for establishing the diagnosis and monitoring system for abnormal pulmonary rales, the lung sound judgment database 4 is set to store the reference normal range and the reference abnormal range for monitoring. and the cut-off point range, as well as the normal range and the abnormal range for auxiliary diagnosis. The method for establishing a diagnosis and monitoring system for abnormal pulmonary rales includes a step S1 of setting judgment parameters, a step S2 of establishing a lung sound judgment database for monitoring, and a step S3 of establishing a lung sound judgment database for auxiliary diagnosis.

The step S1 of setting judgment parameters includes a frequency domain conversion sub-step S11 , a power calculation sub-step S12 , and a parameter calculation sub-step S13 .

The frequency domain conversion sub-step S11 is for the filter unit 31 to perform passband filtering of a plurality of reference lung sound signals in a frequency range, and the processor 3 performs frequency domain conversion on the filtered reference lung sound signals respectively to extract Take multiple reference power spectral density curves in the frequency range. Wherein, the lung sound files of the reference lung sound signals are also in Waveform Audio File Format (WAVE), and the bit depth of the lung sound files is 16 bits. These reference lung sound signals can be converted by capturing the lung sounds of a plurality of human bodies through the lung sound acquisition device 1 in advance. When the reference lung sound signals of more human bodies are captured, the abnormal pulmonary rales diagnosis and monitoring system will be established. more accurate. This frequency range is the frequency band most likely to cause abnormal lung sounds caused by symptoms, and it is between 50Hz and 1000Hz.

With reference to the third figure and the fourth figure, the calculation power sub-step S12 selects a special frequency band from the frequency range for the processor 3, and calculates each reference power spectral density curve in the special frequency band and the frequency range A reference special frequency band power and a reference full frequency band power are respectively defined. For convenience of description, the frequency range is divided into three frequency bands, namely, a first frequency band ranging from 50 Hz to 100 Hz, a second frequency band ranging from 100 Hz to 150 Hz, and a third frequency band ranging from 150 Hz to 1000 Hz. The frequency band power corresponding to the first frequency band is a first frequency band power P1, and the percentage of the first frequency band power P1 corresponding to the reference full frequency band power is a first frequency band power percentage R1, and the frequency band power corresponding to the second frequency band is A second frequency band power P2, the percentage of the second frequency band power P2 corresponding to the reference full frequency band power is a second frequency band power percentage R2, the frequency band power corresponding to the third frequency band is a third frequency band power P3, the third The percentage of the frequency band power P3 corresponding to the reference full frequency band power is a third frequency band power percentage R3. The special frequency band of each reference power spectral density curve in this example is the second frequency band, and the corresponding reference special frequency band power is the second frequency band power P2; the reference full frequency band power is expressed as (P1+P2+P3) .

The parameter calculation sub-step S13 is for the processor 3 to use the reference special frequency band power of each reference power spectral density curve to correspond to a reference special frequency band power percentage of the reference full frequency band power, and substitute the reference full frequency band power into the logarithmic relationship at least A judgment function obtains at least one reference judgment parameter for each reference power spectral density curve. In this example, the number of the judging functions is two. One of the judging functions is to take the logarithm of the power percentage of the reference special frequency band with base 10, and express it as logR2, wherein R2=P2/(P1+P2+P3). Another judging function is to take the logarithm of the reference full-band power with base 10, and express it as log(P1+P2+P3). Therefore, two reference judgment parameters can be obtained from each reference power spectral density curve.

The step S2 of establishing a lung sound judgment database for monitoring includes a finding range sub-step S21 and an adjusting range sub-step S22.

In the sub-step S21 of finding out the range, the processor 3 combines the two reference judgment parameters of all reference power spectral density curves with the corresponding diagnosis results of the actual lungs of the human body through the receiver-operating-characteristic curve. , ROC curve) and Youden index (Youden index) analysis to obtain the reference normal range, the reference abnormal range, the boundary of the cut-off point range and the best cut-off point. In this example, each of the reference normal ranges is that the output value of one of the judgment functions is less than -1.51, and the output value of the other judgment function is less than 12.3, that is, logR2<-1.51, log(P1+P2+P3)<12.3 , each of the reference abnormal ranges is that the output value of one of the judgment functions is greater than -1.31, and the output value of the other judgment function is greater than 12.6, that is, logR2>-1.31, log(P1+P2+P3)>12.6, each should be cut The subpoint range is that the output value of one of the judging functions is greater than or equal to -1.51 and less than or equal to -1.31, and the output value of the other judging function is greater than or equal to 12.3 and less than or equal to 12.6. The optimal cut point is at the midpoint of the cut point range.

The adjusting range step S22 is that the processor 3 is controlled to set at least a cut-off point within the range of the reference normal range, the reference abnormal range and the cut-off point, and according to the set at least cut-off point, the reference normal range, the cut-off point The normal range and the abnormal range for monitoring are established with reference to the abnormal range. Through the step S22 of adjusting the range, the user can adjust the monitored lung sounds toward the purpose of improving the diagnostic accuracy, improving the diagnostic specificity, or improving the diagnostic sensitivity. Two adjustment methods are described below.

The first adjustment method is that the user can select two cut-off points within the range of the cut-off point, and set the range between the cut-off point that is close to the reference normal range and the reference normal range and the reference normal range. is the normal range, and the range between the cut-off point close to the reference abnormal range and the reference abnormal range and the reference abnormal range are set as the abnormal range, by adjusting the cut-off points Narrowing down the range of the cut-off point will increase the correct rate of diagnosis. Alternatively, the user can adjust one of the cutting points to be within the reference normal range, and the other cutting point to be within the reference abnormal range, and the reference normal range narrowed by the one of the cutting points is set as the normal range, by The reference abnormal range narrowed by the other cut-off point is set as the abnormal range, and the range of the cut-off point is substantially enlarged by adjusting the cut-off points.

The second adjustment method is, for example, the user can select the optimal segmentation point found by the finding range sub-step S21 as the segmentation point, that is, at the midpoint of the segmentation point range, the segmentation point The range between the point and the reference normal range and the reference normal range are set as the normal range, the range between the cutting point and the reference abnormal range and the reference abnormal range are set as the abnormal range, then through the normal range , The abnormal range judgment can get the best accuracy of more than 70%.

Or the user can adjust the cut-off point within the range of the cut-off point to approach the reference normal range, and also set the range between the cut-off point to the reference normal range and the reference normal range as the normal range , the range between the cut-off point and the reference abnormal range and the reference abnormal range are set as the abnormal range. By adjusting the cut-off point to approach the reference normal range, the diagnostic sensitivity can be increased. If the diagnostic sensitivity is higher , that is to increase the probability of suffering from rales and the system diagnoses them as abnormal, that is, it is more likely to increase the proportion of normal people who are judged as abnormal, and not to miss any abnormal people.

Or the user can adjust the cut-off point within the range of the cut-off point to approach the reference abnormal range, and also set the range between the cut-off point to the reference normal range and the reference normal range as the normal range , the range between the cut-off point and the reference abnormal range and the reference abnormal range are set as the abnormal range. By adjusting the cut-off point to approach the reference abnormal range, the diagnostic specificity can be increased. If the diagnostic specificity is higher , that is, the probability of not suffering from rales and the system diagnosis as normal increases, that is, the proportion of normal people who are more likely not to be judged as abnormal increases, but it is also easy to miss abnormal people and not detect them.

Step S3 of establishing a lung sound judgment database for auxiliary diagnosis. It should be noted that there is no sequence between the step S2 of establishing the lung sound judgment database for monitoring and the step S3 of establishing the lung sound judgment database for auxiliary diagnosis.

The step S3 of establishing a lung sound judgment database for auxiliary diagnosis is the standard lung sound signal identified by the processor 3 according to chest medical experts, and the above judgment functions, i.e. logR2, log(P1+P2+P3) are set as The feature vector is classified through a supervised machine learning algorithm to obtain the normal range and the abnormal range of a prediction model for auxiliary diagnosis. The supervised machine learning algorithm uses one of Random Forest, Support Vector Machines (SVM), and K Nearest Neighbor (KNN). The normal range is normal lung sounds, and the abnormal range is abnormal lung rales.

If the reference lung sound signals fall within the normal range, the corresponding prediction model is normal lung sound, and if the reference lung sound signals fall within the abnormal range, the corresponding prediction model is abnormal lung sound.

The accuracy of the lung sound judgment database 4 established by analyzing the supervised machine learning algorithm is as high as 80% on average, and can reach 90% at the highest.

Referring to the first figure and the fifth figure A, the medical personnel use the abnormal pulmonary rales diagnosis and monitoring system to monitor whether the subject's lung sounds are normal or abnormal, and the process is as follows.

The medical personnel use the lung sound capture device 1 to capture the lung sound of the test subject at the sampling frequency of 8 kHz, and convert the lung sound into the lung sound signal.

The processor 3 receives the lung sound signal, passes through the filter unit 31 to perform passband filtering in the frequency range, and the processor 3 performs frequency domain conversion to extract a power spectral density in the frequency range Curve, that is, the power spectral density curve in the 50Hz to 1000Hz frequency band, the power spectral density curve is similar to a reference power spectral density curve, as shown in the third figure.

The processor 3 selects the special frequency band 100 Hz to 150 Hz from the frequency range, and calculates the special frequency band power and the full frequency band power defined by the power spectral density curve in the special frequency band and the frequency range respectively. The special frequency band power is the above-mentioned second frequency band power P2, and the full-band power is the sum of the above-mentioned first frequency band power P1, the second frequency band power P2, and the third frequency band power P3.

The processor 3 obtains two judging parameters by substituting the percentage of the special frequency band power corresponding to the special frequency band power and the full frequency generation power into the two judging functions of the logarithmic relationship. Wherein, the power percentage of the special frequency band is the power percentage R2 of the second frequency band, and the two judging functions are the above-mentioned logR2, log(P1+P2+P3).

Afterwards, the medical staff can select the cut-off point within the range of the cut-off point. When the cut-off point approaches the reference normal range, the sensitivity of the processor 3 to judge that the human body is abnormal pulmonary rales can be improved. When the segmentation point approaches the reference abnormal range, the specificity of the processor 3 in judging that the human body has abnormal pulmonary rales can be improved. After selecting the segmentation point, the processor 3 sets the normal range and the abnormal range for monitoring.

When the two judging parameters fall within the normal range for monitoring, the processor 3 judges that the subject has normal lung sounds. When the two judging parameters fall within the abnormal range for monitoring, the processor 3 judges that the human body has abnormal pulmonary rales. The processor 3 judges the lung sound state of the subject, and controls the output device 2 to output the judgment result.

Referring to Figure 1 and Figure 5B, the medical personnel use the abnormal pulmonary rales diagnosis and monitoring system to assist in the process of diagnosing whether the subject's lung sounds are normal or abnormal, and the above-mentioned process of monitoring the subject The process of whether the lung sound is normal or abnormal is similar. When the two judgment parameters fall into the normal range for auxiliary diagnosis, the processor 3 judges that the subject is normal lung sound; When the abnormal range is within the abnormal range, the processor determines that the subject has abnormal pulmonary rales. The processor 3 judges the lung sound state of the subject, and controls the output device 2 to output the judgment result.

Refer to the first figure, the third figure, the sixth figure and the seventh figure, a second embodiment of the abnormal pulmonary rales diagnosis and monitoring system of the present invention, the second embodiment is similar to the first embodiment, the difference lies in the The at least one judgment function stored in the processor 3, and the reference normal range, the reference abnormal range, and the cut-off point range for monitoring stored in the lung sound judgment database 4, as well as the normal range and the cut-off point range for auxiliary diagnosis. exception range. In the second embodiment, the special frequency band selected by the processor 3 from the frequency range is also the second frequency band, the number of the judgment functions is two, and one of the judgment functions is based on 10 The base number is 4 times the result of taking the logarithm of the power percentage of the special frequency band, which is expressed as 4*logR2, and the other judgment function is 2 times the result of taking the logarithm of the full frequency band power with the base 10, expressed as 2*log( P1+P2+P3). Therefore, the method for establishing the abnormal pulmonary rales diagnosis and monitoring system in this example will be different from the first step in the finding range sub-step S21 and the establishment of the lung sound judgment database step S3 for auxiliary diagnosis due to the difference in the judgment function. Examples vary.

In the sub-step S21 of finding out the range, the processor 3 adds the respective two reference judgment parameters of all reference power spectral density curves, i.e. 4*logR2+2*log(P1+P2+P3), and the corresponding The diagnostic results of the actual lungs of the human body are analyzed through the receiver operating characteristic curve and Youden index to obtain the reference normal range, the reference abnormal range, the boundary of the cut-off point range and the best cut-off point. In this example, the reference normal range is that the sum of the output values of these judgment functions is less than 16.5, that is, 4*logR2+2*log(P1+P2+P3)<16.5, and the reference abnormal range is the sum of the output values of these judgment functions Output value phase If it is greater than 17.3, that is, 4*logR2+2*log(P1+P2+P3)>17.3, the range of the cutting point is that the sum of the output values of these judgment functions is greater than or equal to 16.5 to less than or equal to 17.3.

In step S3 of establishing a lung sound judgment database for auxiliary diagnosis, the processor 3 is based on the standard lung sound signal identified by thoracic medicine experts, and is set with two judgment functions of 4*logR2 and 2*log(P1+P2+P3). is a feature vector, and is classified through the supervised machine learning algorithm to obtain the normal range and the abnormal range for auxiliary diagnosis.

The normal range and the abnormal range of the prediction model are obtained through the supervised machine learning algorithm. Therefore, if the reference lung sound signals fall within the normal range, the corresponding prediction model is normal lung sound. If the If the reference lung sound signal falls within the abnormal range, the corresponding prediction model is abnormal pulmonary rales.

Referring to the first figure and the eighth figure A, in this example, the process of the medical staff using the abnormal pulmonary rales diagnostic monitoring system to monitor whether the subject's lung sounds are normal or abnormal is similar to the first embodiment, first Also, the medical personnel use the lung sound capture device 1 to capture the lung sound of the test subject at the sampling frequency of 8 kHz, and convert the lung sound into the lung sound signal. The processor 3 receives the lung sound signal, first passes through the filter unit 31 to perform passband filtering in the frequency range, and then the processor 3 performs frequency domain conversion on the filtered lung sound signal, and extracts the The power spectral density curve of the segment. The processor 3 selects the special frequency band from the frequency range, and calculates the special frequency band power defined by the power spectral density curve in the special frequency band and the frequency range, which is the above-mentioned second frequency band power P2 and the full-band power. The processor 3 obtains two judging parameters by substituting the specific frequency band power percentage of the special frequency band power and the full frequency band power into the judging functions of the logarithmic relationship respectively.

Similar to the way of use in the first embodiment, the medical staff can select the cut-off point within the range of the cut-off point to set the normal range and the abnormal range, and then the processor 3 2. Judging whether the parameter falls into the normal range or the abnormal range for monitoring and monitoring the lung sound status of the subject. Through repeated sampling verification, the lung sound status of the subject can be judged by the abnormal pulmonary rales diagnostic monitoring system in this example, and the accuracy of the established lung sound judgment database 4 is over 80%.

Referring to Figure 1 and Figure 8B, the medical personnel use the abnormal pulmonary rales diagnosis and monitoring system to assist in the process of diagnosing whether the subject's lung sounds are normal or abnormal, and the above-mentioned monitoring of whether the subject's lung sounds are normal or abnormal process is similar, when the two judgment parameters fall into the normal range for auxiliary diagnosis, the processor 3 judges that the subject to be tested is normal lung sound; when the two judgment parameters fall into the abnormal range for auxiliary diagnosis , the processor 3 judges that the subject has abnormal pulmonary rales. After cross-validation, the lung sound status of the subject can be judged by this method, and the accuracy is as high as 90%.

Based on the description of the above-mentioned embodiments, it is possible to fully understand the operation of the present invention, use and the effect that the present invention produces, but the above-mentioned embodiments are only preferred embodiments of the present invention, and should not be used to limit the implementation of the present invention. The scope, that is, the simple equivalent changes and modifications made according to the patent scope of the present invention and the content of the description of the invention, all fall within the scope of the present invention.

1: Lung sound capture device

2: output device

3: Processor

31: Filter unit

4: Lung sound judgment database

Claims (6)

A diagnostic monitoring system for abnormal pulmonary rales, comprising: a lung sound acquisition device, which acquires a human lung sound at a sampling frequency, and converts it into a lung sound signal; and a processor, including a filter unit, The processor signal is connected to the lung sound acquisition device, after the processor receives the lung sound signal, the filter unit passes the lung sound signal to a frequency range between 50 Hz and 1000 Hz for passband filtering, and performing frequency domain conversion on the filtered lung sound signal, and extracting a power spectral density curve in the frequency range; the processor selects a special frequency band between 100Hz and 150Hz from the frequency range, And calculate a special frequency band power and a full frequency band power respectively defined by the power spectral density curve in the special frequency band and the frequency range; a lung sound judgment database, the signal is connected to the processor, and the lung sound judgment database includes A reference normal range, a reference abnormal range, and a cutpoint range, the cutpoint range is between the reference normal range and the reference abnormal range, and the reference normal range, the reference abnormal range, and the cutpoint range The boundary is obtained by analyzing the receiver operating characteristic curve and Youden index; the processor is controlled to set the cutting point within the range of the cutting point, and the range between the cutting point and the reference normal range and the The reference normal range is set as a normal range, and the range between the cut point and the reference abnormal range and the reference abnormal range are set as an abnormal range; or, the processor is controlled to be set within the cut point range Two cut-off points, and set the range between the cut-off point close to the reference normal range to the reference normal range and the reference normal range as the normal range, and set the range close to the reference abnormal range The range between the segmentation point and the reference abnormal range and the reference abnormal range are set as the abnormal range; The processor corresponds to a special frequency band power percentage of the full frequency band power corresponding to the special frequency band power, and substitutes the full frequency band power into two judging functions of logarithmic relationship, wherein one judging function is base 10, the percentage of the special frequency band power Take the logarithm, and another judgment function is to take the logarithm of the full-band power with 10 as the base number, so as to obtain two judgment parameters; the reference normal range is that the output value of one of the judgment functions is less than -1.51 and the output value of the other judgment function If the output value is less than 12.3, the reference abnormal range is that the output value of one of the judgment functions is greater than -1.31 and the output value of the other judgment function is greater than 12.6, and the cutoff point range is that the output value of one of the judgment functions is greater than or equal to Between -1.51 and less than or equal to -1.31 and the output value of the other judgment function is between greater than or equal to 12.3 and less than or equal to 12.6; when the at least one judgment parameter falls within the normal range, the processor judges that the human body is normal Lung sounds: when the at least one determination parameter falls within the abnormal range, the processor determines that the human body has abnormal pulmonary rales. The abnormal pulmonary rales diagnosis and monitoring system as described in Claim 1, wherein, the frequency range is 50Hz to 1000Hz, the special frequency band is 100Hz to 150Hz, and the number of the judgment functions is two, and one of the judgment functions is The base 10 is 4 times the result of taking the logarithm of the power percentage of the special frequency band, and another judgment function is 2 times the result of taking the logarithm of the full frequency band power with the base 10. The reference normal range is the output of these judgment functions The sum of values is less than 16.5, the reference abnormal range is that the sum of the output values of these judgment functions is greater than 17.3, and the range of the cut-off point is that the sum of the output values of these judgment functions is greater than or equal to 16.5 to less than or equal to 17.3. The abnormal pulmonary rales diagnosis and monitoring system as described in Claim 1, the normal range and the abnormal range are based on standard lung sound signals identified by thoracic medicine experts, at least one judgment function is set as a feature vector, and a supervised machine learning is performed algorithms for classification. A method for establishing a diagnosis and monitoring system for abnormal pulmonary rales, comprising a step of setting judgment parameters, the step of setting judgment parameters includes: a processor includes a filter unit, the filter unit performs frequency domain conversion on a plurality of reference lung sound signals, and The processor extracts each reference lung sound signal to a reference power spectral density curve in a frequency range between 50 Hz and 1000 Hz through the filtering unit; the processor selects a frequency range between 100 Hz and A special frequency band between 150 Hz, and calculate a reference special frequency band power and a reference full-band power defined by each reference power spectral density curve in the special frequency band and the frequency range; and the processor converts each reference The reference special frequency band power of the power spectral density curve corresponds to a reference special frequency band power percentage of the reference full frequency band power, and the reference full frequency band power is respectively substituted into two judgment functions of the logarithmic relationship, wherein one judgment function is based on 10. The power percentage of the special frequency band is logarithmic, and another judgment function is to take the logarithm of the full frequency band power with base 10, so as to obtain two reference judgment parameters of each reference power spectral density curve. The method for establishing a diagnostic monitoring system for abnormal pulmonary rales as described in claim item 4 also includes a step of establishing a lung sound judgment database for monitoring. The step of establishing a lung sound judgment database for monitoring is: the processor collects all The at least one reference judgment parameter of the reference power spectral density curve and the diagnosis results of the actual lung sounds of the human body are analyzed to obtain a reference normal range, a reference abnormal range and a cut-off point range through receiver operating characteristic curve and Youden index analysis. Boundary; wherein, the reference normal range is that the output value of one of the judgment functions is less than -1.51 and the output value of the other judgment function is less than 12.3, which is defined as normal lung sound; the reference abnormal range is the value of one of the judgment functions If the output value is greater than -1.31 and the output value of the other judgment function is greater than 12.6, it is defined as abnormal pulmonary rales; and the output value of the other judgment function is in greater than or equal to 12.3 and less than or equal to 12.6, so the cut point range is between the reference normal range and the reference abnormal range; and the processor is controlled within the reference normal range, the reference abnormal range and the cut At least one cut-off point is set in the point range, and a normal range and an abnormal range are established according to the set at least one cut-off point, the reference normal range, and the reference abnormal range. The method for establishing a diagnosis and monitoring system for abnormal pulmonary rales as described in claim 4 further includes a step of establishing a lung sound judgment database for auxiliary diagnosis, the step of establishing a lung sound judgment database for auxiliary diagnosis is: the processor According to the standard lung sound signal identified by thoracic medical experts, at least one judgment function is set as a feature vector, and a supervised machine learning algorithm is used to classify to obtain a normal range and an abnormal range of a prediction model, wherein the normal range It is normal lung sounds, and the abnormal range is abnormal lung rales.