Lung Sound Analysis System

Abstract

This lung sound analysis system is provided with: an acquisition means for acquiring time-series acoustic signals including the lung sounds of a heart failure patient; a detection means for detecting abnormal lung sounds from the acquired time-series acoustic signals; an opinion obtaining means for transmitting, to a medical specialist's terminal device, analysis target lung sound information associated with the acquired time-series acoustic signals and the detection result, and for receiving, from the terminal device, analysis target lung sound information to which has been added the medical specialist's opinion about the time-series acoustic signals; and a generation means for generating, on the basis of the analysis target lung sound information to which the medical specialist's opinion has been added, learning data for detecting abnormal lung sounds.

Description

[Technical field]

[0001] The present invention relates to a lung sound analysis system, a lung sound analysis method, and a recording medium for assisting in the diagnosis of heart failure. [Background technology]

[0002] Heart failure is a clinical syndrome in which some kind of cardiac dysfunction, i.e., organic and / or functional abnormalities in the heart, cause the cardiac pump function to fail to compensate, resulting in dyspnea, fatigue, and edema, and a corresponding decrease in exercise tolerance. Patients with heart failure are always at risk of exacerbation, even if they achieve remission through treatment. If a patient experiences acute exacerbation due to excessive fluid and salt intake, forgetting to take medication, or excessive exercise, they will be forced to be re-admitted to hospital. Therefore, it is important to prevent acute exacerbation by early detection and treatment of exacerbation of heart failure in discharged patients.

[0003] One method for diagnosing heart failure is to check lung sounds by auscultation. This is a safe and easy way to check the health of the lungs, and therefore heart failure. However, it is difficult for a doctor who is not a trained specialist to obtain detailed and accurate diagnostic results. For this reason, detailed diagnoses cannot be made by general nurses or caregivers who make rounds or visit home caregivers.

[0004] To address this problem, a system has been proposed that automatically determines whether or not there is an abnormal sound called an adventitious sound in lung sounds collected by an electronic stethoscope (see, for example, Patent Documents 1 to 6). For example, Patent Document 10 detects adventitious sounds from respiratory sound data to be analyzed using abnormal learning data created from multiple respiratory sound data including different types of adventitious sounds and normal learning data created from multiple respiratory sound data not including adventitious sounds. For example, Patent Document 11 describes that a correct answer is given to each respiratory sound sample as a teacher signal indicating that the respiratory sound is "normal" or "abnormal" and that parameters of a model with the highest performance of distinguishing between normality and abnormality of breathing are generated using an optimization method such as the steepest descent method or the Newton method. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2014-4018 A [Patent Document 2] Special Publication No. 2002-538921 [Patent Document 3] Special Publication No. 2017-536905 [Patent Document 4] WO2010 / 044452 [Patent Document 5] JP 2008-113936 A [Patent Document 6] Patent No. 4849424 [Patent Document 7] WO2019 / 220609 [Patent Document 8] WO2019 / 220620 [Patent Document 9] JP 2007-190081 A [Patent Document 10] JP 2005-66044 A [Patent Document 11] Patent No. 5585428 Summary of the Invention [Problem to be solved by the invention]

[0006] In order to improve the accuracy of detecting abnormal lung sounds, it is necessary to collect a large amount of abnormal lung sound data to be used as learning data. However, it is difficult for anyone other than a trained specialist to accurately determine whether or not lung sound data is abnormal. Therefore, it has been difficult to efficiently collect abnormal lung sound data in clinics or patients' homes in towns where there are no specialists.

[0007] The present invention provides a lung sound analysis system that solves the above-mentioned problems. [Means for solving the problem]

[0008] A lung sound analysis system according to one embodiment of the present invention comprises: An acquisition means for acquiring a time-series acoustic signal including lung sounds of a patient with heart failure; a detection means for detecting abnormal lung sounds from the acquired time-series acoustic signals; a finding acquisition means for transmitting analysis target lung sound information in which the acquired time-series acoustic signal is associated with a result of the detection to a terminal device of a specialist, and receiving analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device; a generating means for generating learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added; The device is configured to include:

[0009] In addition, a lung sound analysis method according to another aspect of the present invention includes the steps of: Acquire a time-series acoustic signal including lung sounds of a patient with heart failure; Detecting abnormal lung sounds from the acquired time-series acoustic signals; transmitting analysis target lung sound information in which the acquired time-series acoustic signal and the result of the detection are associated with each other to a terminal device of a specialist, and receiving analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device; Based on the analysis target lung sound information to which the specialist's findings have been added, learning data is generated for detecting abnormal lung sounds. It is structured as follows.

[0010] In addition, a computer-readable recording medium according to another aspect of the present invention includes: On the computer, acquiring a time-series acoustic signal including lung sounds of a patient with heart failure; detecting abnormal lung sounds from the acquired time-series acoustic signals; transmitting analysis target lung sound information in which the acquired time-series acoustic signal and the result of the detection are associated with each other to a terminal device of a specialist, and receiving the analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device; A process of generating learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added; The recording medium is configured to record a program for causing the recording medium to perform the above steps. Effect of the Invention

[0011] By having the above-described configuration, the present invention can efficiently collect learning data for detecting abnormal lung sounds in clinics in towns where there are no specialists, in patients' homes, and the like. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram of a lung sound analyzer according to a first embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram showing an example of the configuration of a lung sound record stored in the lung sound analysis device according to the first embodiment of the present invention. [Diagram 3] FIG. 1 is an explanatory diagram of auscultation positions (1) to (12) for auscultation with an electronic stethoscope in a lung sound analyzer according to a first embodiment of the present invention. [Figure 4] FIG. 2 is a diagram showing an example of the configuration of analysis target lung sound information stored in the lung sound analysis device according to the first embodiment of the present invention. [Diagram 5]FIG. 2 is a diagram showing an example of the configuration of target personal information set in advance in the lung sound analysis device according to the first embodiment of the present invention. [Figure 6] FIG. 2 is a diagram showing an example of the configuration of analysis target lung sound information in which auscultatory findings by a specialist are recorded and stored in the lung sound analysis device according to the first embodiment of the present invention. [Figure 7] 5 is a flowchart showing an example of a pre-operation of the lung sound analyzer according to the first embodiment of the present invention. [Figure 8] FIG. 2 is an explanatory diagram of learning using learning data in the lung sound analyzer according to the first embodiment of the present invention. [Figure 9] 4 is a flowchart showing an example of an analysis operation of the lung sound analyzer according to the first embodiment of the present invention. [Figure 10] 4 is a flowchart showing details of an analysis operation of the lung sound analyzer according to the first embodiment of the present invention. [Figure 11] FIG. 2 is a diagram showing an example of an abnormality frequency for each auscultation position of a patient calculated by the lung sound analyzer according to the first embodiment of the present invention and an auscultation order determined based on the abnormality frequency. [Figure 12] FIG. 2 is a schematic diagram showing a waveform of a time-series acoustic signal including lung sounds output from the electronic stethoscope in the first embodiment of the present invention. [Figure 13] FIG. 2 is a diagram showing an example of a determination table for determining the severity of heart failure from the analysis results of lung sound data for each auscultation position by the lung sound analysis device according to the first embodiment of the present invention. [Figure 14] FIG. 5 is a block diagram of a lung sound analysis system according to a second embodiment of the present invention. [Figure 15] FIG. 11 is a block diagram of a lung sound analysis system according to a third embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Next, an embodiment of the present invention will be described in detail with reference to the drawings. [First embodiment]

[0014] FIG. 1 is a block diagram of a lung sound analyzer 10 according to a first embodiment of the present invention. The lung sound analyzer 10 is an information processing device that acquires and analyzes lung sounds from a patient with heart failure. The lung sound analyzer 10 may be, but is not limited to, a smartphone, a tablet terminal, a PDA (Personal Digital Assistant), a notebook computer, or the like. Hereinafter, a patient whose lung sounds are analyzed using the lung sound analyzer 10 is referred to as Patient A.

[0015] The lung sound analyzer 10 includes an electronic stethoscope 11 , a communication I / F unit 12 , an operation input unit 13 , a screen display unit 14 , a memory unit 15 , and a calculation processing unit 16 .

[0016] The electronic stethoscope 11 is configured to convert lung sounds of patient A into digital signals by placing the chest piece of the stethoscope against the chest or back of patient A, and transmit the digital signals to the calculation processing unit 16 wirelessly or via wires.

[0017] The communication I / F unit 12 is configured, for example, by a dedicated data communication circuit, and is configured to perform data communication with various devices such as a server device connected via wire or wirelessly.

[0018] The operation input unit 13 is composed of operation input devices such as a keyboard and a mouse, and is configured to detect an operation by an operator and output it to the calculation processing unit 16. The operator is a person who performs the task of acquiring the lung sounds of patient A using the lung sound analysis device 10. The operator may be, for example, a doctor at a local clinic, a medical professional such as a nurse, a care worker such as a care worker, or a family member of patient A.

[0019] The screen display unit 14 is composed of a screen display device such as an LCD (Liquid Crystal Display) or a PDP (Plasma Display Panel), and is configured to display various information such as analysis results on the screen in response to instructions from the calculation processing unit 16.

[0020] The storage unit 15 is composed of a storage device such as a hard disk or a memory, and is configured to store processing information and programs 151 required for various processes in the arithmetic processing unit 16 .

[0021] The program 151 is a program that realizes various processing units by being read into and executed by the arithmetic processing unit 16. The program 151 is read in advance from an external device (not shown) or a storage medium (not shown) via a data input / output function such as the communication I / F unit 12, and is stored in the storage unit 15.

[0022] Main processed information stored in storage unit 15 includes lung sound records 152, lung sound information to be analyzed 153, and a learning data DB (database) 154.

[0023] Lung sound record 152 is a record of the lung sounds of patient A. Lung sound record 152 is a record of medical procedures, including auscultation, performed on patient A for the treatment of heart failure, for example during hospitalization. FIG. 2 shows an example of the configuration of lung sound record 152. In this example, lung sound record 152 is composed of the following items: patient ID 1521, one or more pieces of auscultation information 1527, personal information 1525, and contact email address 1526. An ID that uniquely identifies patient A is recorded in the patient ID 1521 item.

[0024] The auscultation information 1527 items are composed of auscultation date and time 1522, attending physician 1523, and lung sound information 1524. The auscultation date and time 1522 item records the date and time when the diagnosis, including auscultation, was made. One or more auscultation information 1527 items are arranged in ascending order of auscultation date and time 1522. The bottommost auscultation information 1527 (auscultation information immediately before personal information 1525) is the most recent. The attending physician 1523 item records the name of the physician who made the diagnosis.

[0025] An item of lung sound information 1524 is provided for each auscultation position. An auscultation position is a location on the patient's body where the chest piece of the stethoscope is placed to auscultate the lung sounds. In other words, an auscultation position is a part where lung sounds are acquired. In the example of FIG. 2, a total of 12 auscultation positions are set from auscultation position (1) to auscultation position (12) (in FIG. 2, auscultation positions (2) to (11) are omitted). FIG. 3 is a schematic diagram for explaining auscultation positions (1) to (12).

[0026] Referring to FIG. 3, auscultation positions (1) and (2) are set on the left and right of the upper lung field in the anterior chest. Auscultation positions (3) and (4) are set on the left and right of the middle lung field in the anterior chest. Auscultation positions (5) and (6) are set on the left and right of the lower lung field in the anterior chest. Auscultation positions (7) and (8) are set on the left and right of the upper lung field in the back. Auscultation positions (9) and (10) are set on the left and right of the middle lung field in the back. Auscultation positions (11) and (12) are set on the left and right of the lower lung field in the back. The number and locations of auscultation positions are not limited to the above. For example, auscultation positions may be set not only in the anterior chest and back but also in the upper lung field, middle lung field, and lower lung field in the left and right lateral chest, for a total of 18 positions. Alternatively, some of the above auscultation positions may be excluded. For example, auscultation positions (3) to (6), (9), and (10) may be excluded, and the auscultation positions may be limited to a total of six positions: (1), (2), (7), (8), (11), and (12).

[0027] Referring again to FIG. 2, the items of the lung sound information 1524 for each auscultation position include one or more pairs of lung sound data items and auscultation findings items. In the lung sound data items, digital time-series acoustic signals including lung sounds acquired by an electronic stethoscope from the auscultation position of patient A are recorded. The patient's posture during auscultation is broadly classified into a supine position and a sitting position, but anterior chest and back auscultation is usually performed in a sitting position. The signal length of one lung sound data (e.g., lung sound data 1) is arbitrary. For example, one lung sound data may be a signal for N consecutive breaths of patient A, where N is a positive integer equal to or greater than 1. Furthermore, the lung sound data may be a signal obtained by performing processing such as removing the time-series acoustic signal during the resting phase, removing noise, and adding breathing timing to the time-series acoustic signal acquired from the electronic stethoscope.

[0028] The auscultation findings item records the specialist's auscultation findings for the lung sound data. The auscultation findings include the presence or absence of abnormal lung sounds, and if there are abnormal sounds, the type of abnormal sound (such as rales), etc.

[0029] The personal information 1525 item records information such as Patient A's gender, age, weight, body type (BMI), and medical history.

[0030] One or more email addresses of recipients to whom the analysis results are to be sent are recorded in the contact email address 1526 item. The contact email address may be, for example, the email address of the hospital where Patient A was hospitalized, a heart failure specialist, Patient A's family doctor, etc. The method for sending the analysis results is not limited to email, and may be other communication methods such as a groupware messaging function or business chat.

[0031] 1 again, lung sound information acquired from patient A using electronic stethoscope 11 and the analysis results thereof are recorded in analyzed lung sound information 153. Fig. 4 shows an example of the configuration of analyzed lung sound information 153. In this example, analyzed lung sound information 153 is composed of the following items: patient ID 1531, analysis date and time 1532, person in charge 1533, lung sound information 1534, urgency 1535, analysis contact items 1536, and consent information 1537.

[0032] The patient ID 1531 item records an ID that uniquely identifies patient A recorded in the patient ID 1521 item of the lung sound record 152. The analysis date and time 1532 item records the date and time when the lung sounds of patient A were acquired and analyzed. The person in charge 1533 item records an ID that uniquely identifies the operator who performed the task of acquiring the lung sounds of patient A.

[0033] The items of the lung sound information 1534 are provided for each auscultation position. In the example of FIG. 4, a total of 12 auscultation positions are set from auscultation position (1) to auscultation position (12) described with reference to FIG. 3 (in FIG. 4, auscultation positions (2) to (11) are omitted). The items of the lung sound information 1534 for each auscultation position include one or more pairs of items each consisting of a lung sound data item and an analysis result item. In the lung sound data item, a digital time-series acoustic signal including the lung sound acquired by the electronic stethoscope 11 from the corresponding auscultation position of the patient A is recorded. The signal length of one piece of lung sound data (for example, the lung sound data 1) is arbitrary. For example, one piece of lung sound data may be a signal for N consecutive breaths of the patient A. Here, N is a positive integer of 1 or more. The lung sound data may be a signal obtained by performing processing such as removing the time-series acoustic signal during the rest phase, removing noise, and adding breathing timing to the time-series acoustic signal acquired by the electronic stethoscope 11.

[0034] The analysis result field records the results of mechanical analysis of the lung sound data. The analysis result records a numerical value indicating whether the lung sound data is abnormal or not. For example, the analysis result field may record a binary value of 0 indicating a normal lung sound and 1 indicating an abnormal lung sound. Alternatively, the analysis result field may record a numerical value indicating the degree of abnormality of the lung sound data. An abnormality level equal to or lower than a preset threshold indicates that the lung sound data is normal, and an abnormality level exceeding the threshold indicates that the lung sound data is abnormal. Alternatively, the analysis result field may record details of the detected abnormal sound (e.g., type, characteristics, timing of hearing, etc.).

[0035] The item Urgency 1535 records the urgency calculated by comprehensively judging the analysis results of the auscultation positions (1) to (12). The urgency is an index showing how urgent the patient's condition is. In other words, the urgency is an index showing the degree of time available to avoid or reduce the risk of re-hospitalization due to acute exacerbation by performing appropriate heart failure treatment within a certain time. By including such urgency in the analysis results, medical professionals who recognize the analysis results can take action according to the urgency.

[0036] The analysis report 1536 records the condition of patient A on the day of analysis, etc. Examples of patient A's condition include weight, blood pressure, pulse rate, subjective symptoms (shortness of breath when going out, swelling, coughing, loss of appetite, etc.), medication status, and amount of water intake.

[0037] The consent information 1537 item records information as to whether or not consent has been obtained from patient A for all or a part of the lung sound data of patient A recorded in the lung sound information to be analyzed 153 to be used as learning data.

[0038] Referring again to FIG. 1, the learning data DB 154 records learning data for detecting abnormal lung sounds from lung sound data. The learning data is composed of, for example, lung sound data and correct answer labels, and is also called teacher data. The learning data DB 154 records a plurality of learning data for each auscultation position. Furthermore, the plurality of learning data for one auscultation position includes, for example, a plurality of learning data including normal lung sound data and a plurality of learning data including abnormal lung sound data. The label of each piece of learning data indicates whether the corresponding lung sound data is an abnormal lung sound or a normal lung sound. The label indicating that the lung sound is abnormal may further include information regarding the abnormal lung sound (for example, type, characteristics, degree of abnormality, etc.).

[0039] Processing unit 16 has a microprocessor such as a CPU and its peripheral circuits, and is configured to read and execute program 151 from storage unit 15, thereby realizing various processing functions by causing the above hardware to cooperate with program 151. Main processing units realized by processing unit 16 include lung sound record acquisition means 161, analysis target lung sound acquisition means 162, lung sound abnormality detection means 163, analysis result output means 164, learning data generation means 165, and learning means 166.

[0040] The lung sound record acquisition means 161 is configured to acquire lung sound records 152 relating to patient A from an external device (not shown) or a storage medium (not shown) via a data input / output function such as the communication I / F unit 12, and record the lung sound records 152 in the storage unit 15.

[0041] The analysis target lung sound acquisition means 162 is configured to acquire a digital time series acoustic signal including the lung sounds of patient A and other information. The analysis target lung sound acquisition means 162 acquires the digital time series acoustic signal including the lung sounds of patient A from the electronic stethoscope 11 according to an instruction of the operator inputted from the operation input unit 13 or the like. The analysis target lung sound acquisition means 162 also acquires, as other information, a patient ID, analysis date and time, a person in charge, analysis contact information, and consent information from the operator via the operation input unit 13 or from the lung sound record 152 stored in the memory unit 15. The analysis target lung sound acquisition means 162 also generates analysis target lung sound information 153 from the acquired digital time series acoustic signal and other information, and stores it in the memory unit 15. The analysis target lung sound information 153 stored in the memory unit 15 by the analysis target lung sound acquisition means 162 is configured in a format such as that shown in FIG. 4, for example. At the time of saving by the analysis target lung sound acquisition means 162, each analysis result item of the lung sound information 1534 and the urgency level 1535 item are NULL values.

[0042] The lung sound abnormality detection means 163 is configured to detect whether the lung sound data is an abnormal lung sound. There are various methods for detecting lung sound abnormalities. In this embodiment, the lung sound abnormality detection means 163 uses an abnormality detection method using supervised learning. That is, the lung sound abnormality detection means 163 learns in advance abnormal sounds such as coarse crackles and fine crackles, which are intermittent rales, and wheezes and rhonchi, which are continuous rales, and detects abnormal sounds based on the learning results.

[0043] For example, the lung sound abnormality detection means 163 may perform supervised learning by using deep learning to create a model that learns the features and discrimination criteria of input sound data (input data) for example, using learning data that collects abnormal sounds, and may perform detection by checking whether the input data matches the model. For example, the lung sound abnormality detection means 163 may use a spectrogram for learning and input data, in which sound is subjected to FFT (fast Fourier transform) or log-FFT for a certain interval and arranged in chronological order, and may use an RNN (recurrent neural network) or CNN (convolutive neural network) for deep learning.

[0044] Furthermore, the lung sound abnormality detection means 163 may use a method of converting the lung sound waveforms of the learning and input data into short-time features such as zero crossing coefficients and MFCC (Mel Frequency Cepstrum Coefficients) and detecting abnormal sounds by machine learning. For example, the lung sound abnormality detection means 163 may model the data using GMM (Gaussian Mixture) when learning using the learning data, and check whether the input data matches the corresponding model when detecting. The lung sound abnormality detection means 163 may also use the learning data to learn the discrimination surface of a classifier such as SVM (support vector machine) and use the discrimination surface to discriminate whether the input data corresponds to an abnormal sound. In addition to the above-mentioned method of directly obtaining such features, the lung sound abnormality detection means 163 may generate features using the data itself, such as NMF (non-negative matrix factorization) and PCA (principal component analysis).

[0045] Furthermore, the lung sound abnormality detection means 163 may detect abnormal sounds by a decision tree or the like using statistical features of the input waveform, such as the long-term power distribution of the input signal or the distribution of component amounts and component ratios in a specific frequency bin range. In this case, the lung sound abnormality detection means 163 may use, as the items of the decision tree, not only direct values ​​(for example, when the power exceeds 20 mW for three consecutive frames) but also statistical features (for example, when a processing frame greater than 3σ occurs using Gaussian approximation). Furthermore, the lung sound abnormality detection means 163 may model the input signal, rather than the input signal itself, using an AR (autoregressive) process or the like, and detect abnormal sounds by some of the model parameters exceeding a threshold. Although these methods may not include a learning process, they are included in supervised learning for convenience, since the construction of the decision tree and the determination of the threshold include the observation of the abnormal sound, which is the target signal.

[0046] The lung sound abnormality detection means 163 uses the anomaly detection method using supervised learning as described above to analyze the lung sound data for each auscultation position of patient A recorded in the lung sound information to be analyzed 153, and records the analysis results in the analysis result item of the lung sound record for each auscultation position 152. The lung sound abnormality detection means 163 also calculates the urgency based on the analysis result of the lung sound data for each auscultation position, and records it in the urgency item 1535.

[0047] The analysis result output means 164 is configured to output the analyzed lung sound information 153 for the purpose of notifying relevant persons of the heart failure state of patient A. For example, the analysis result output means 164 is configured to read out the analyzed lung sound information 153 from the memory unit 15 and display the analyzed lung sound information 153 on the screen display unit 14. The analysis result output means 164 is also configured to send an email to which the analyzed lung sound information 153 read out from the memory unit 15 is attached as a file via the communication I / F unit 12 to the contact email address 1526 of the lung sound record 152, in accordance with an instruction from the operation input unit 13 or automatically. At this time, the analysis result output means 164 may determine the destination of the analyzed lung sound information 153 based on the urgency 1535 of the analyzed lung sound information 153.

[0048] Furthermore, the analysis result output means 164 is configured to transmit the lung sound information to be analyzed 153 to a terminal device of a heart failure specialist via the communication I / F unit 12 in order to obtain the heart failure specialist's opinion on the lung sound data obtained from patient A and create learning data.

[0049] At this time, the analysis result output means 164 may determine whether or not to transmit the analyzed lung sound information 153 to the heart failure specialist based on the analysis result for each auscultation position. For example, the analyzed lung sound information 153 may not be transmitted if the lung sound data at all auscultation positions is normal, but may transmit the analyzed lung sound information 153 if the lung sound data at one or more auscultation positions is abnormal. In general, learning data including normal lung sound data can be easily obtained from many normal people, whereas learning data including abnormal lung sound data can be obtained only from a limited number of people. Therefore, by setting only the analyzed lung sound information 153 including abnormal lung sound data as the transmission target, the burden on the heart failure specialist can be reduced. In addition, the analyzed lung sound information 153 may be transmitted to the heart failure specialist only if the type of abnormal lung sound recorded in the analysis result matches a predetermined type. The predetermined type of abnormal lung sound may be a type of abnormal lung sound for which learning data is insufficient. This makes it possible to create learning data for the types of abnormal lung sounds that are lacking, without unnecessarily increasing the burden on heart failure specialists.

[0050] Furthermore, the analysis result output means 164 may determine whether or not to transmit the analyzed lung sound information 153 to a heart failure specialist based on the consent information 1537. For example, the analyzed lung sound information 153 may be transmitted to a heart failure specialist only when the patient's consent has been obtained to use the lung sound data for learning. This is to avoid unnecessarily increasing the burden on heart failure specialists, since lung sound data for which consent has not been obtained cannot be used as learning data.

[0051] Furthermore, the analysis result output means 164 may determine whether or not to transmit the analysis target lung sound information 153 to a heart failure specialist based on the personal information of patient A. For example, the analysis result output means 164 may transmit to a heart failure specialist only the analysis target lung sound information of a patient whose personal information matches the predefined target personal information (gender, age group, BMI range, medical history, etc.). The predefined target personal information may be at least one of, or a combination of two or more of, gender, age group, BMI range, medical history, etc., for which learning data is insufficient. FIG. 5 shows an example of the configuration of target personal information. In this example, patients whose gender is female, whose age group is 40 to 60 years old, whose BMI is 20 or less, and whose medical history is high blood pressure are targeted. By narrowing down the target in this way, only the insufficient learning data can be collected efficiently.

[0052] The analyzed lung sound information 153 transmitted to the heart failure specialist is analyzed by the heart failure specialist. For example, the specialist plays back the lung sound data for each auscultation position recorded in the analyzed lung sound information 153 using a personal computer or the like, and diagnoses whether or not any accessory noises such as rales are heard from the lung sounds of patient A. The specialist then creates auscultation findings for the lung sound data for each auscultation position, and records them in the analyzed lung sound information 153. The analyzed lung sound information 153 with the specialist's auscultation findings recorded in this way is returned to the lung sound analysis device 10 that sent it by a communication means such as email. Hereinafter, the analyzed lung sound information with the specialist's auscultation findings recorded will be referred to as analyzed lung sound information with auscultation findings. FIG. 6 shows an example of the configuration of the analyzed lung sound information with auscultation findings 153.

[0053] When the learning data generating means 165 receives the analysis target lung sound information with auscultation findings 153 from a heart failure specialist via the communication I / F unit 12 of the lung sound analyzer 10, it generates learning data based on the received analysis target lung sound information with auscultation findings 153. In addition, the learning data generating means 165 records the generated learning data in the learning data DB 154.

[0054] The learning means 166 is configured to learn a model for detecting abnormal lung sounds, using the learning data recorded in the learning data DB 154. Moreover, the learning means 166 is configured to re-learn the model for detecting abnormal lung sounds after new learning data is added to the learning data DB 154.

[0055] Next, we will explain the operation of the lung sound analyzer 10. The operation of the lung sound analyzer 10 is roughly divided into a preparatory operation, and a subsequent analysis operation, learning data creation, and learning operation.

[0056] First, the pre-operation will be described. Fig. 7 is a flow chart showing an example of the pre-operation. The pre-operation is started, for example, by operating a pre-operation start button displayed on the screen display unit 14 to start the lung sound record acquisition means 161.

[0057] 7, when lung sound record acquisition means 161 is started, it acquires lung sound record 152 of patient A from an external device (not shown) or storage medium (not shown) via a data input / output function such as communication I / F unit 12, and records it in storage unit 15 (step S1). Fig. 2 shows an example of the configuration of lung sound record 152 acquired in this manner. Lung sound record 152 includes at least the past lung sound data and auscultation findings history of patient A.

[0058] When the above operation of lung sound record acquisition means 161 is completed, learning means 166 is started automatically or according to an instruction from operation input unit 13. Learning means 166 reads learning data DB 154 from storage unit 15, learns a model for detecting lung sound abnormalities based on the learning data recorded in learning data DB 154, and stores the learned model in lung sound abnormality detection means 163 (step S2).

[0059] FIG. 8 is an explanatory diagram of model learning. Referring to FIG. 8, the learning means 166 reads out the learning data of the auscultation position (1) from the learning data DB 154, and creates a model 171-1 that detects abnormality in the lung sound data at the auscultation position (1) using the read learning data. Similarly, the learning means 166 generates models 171-2 to 171-12 that detect abnormality in the lung sound data at the auscultation positions (2) to (12) based on the learning data of the auscultation positions (2) to (12) recorded in the learning data DB 154. In the above, the learning means 166 learned a model that detects abnormality in the lung sound data for each auscultation position. However, the learning means 166 may learn one common model for detecting abnormality in the lung sound data at a plurality of auscultation positions. Furthermore, when learning a model common to multiple auscultation positions, the threshold set when using that model to detect anomalies (for example, the threshold set when detecting anomalies using a decision tree) may be set commonly for multiple auscultation positions, or may be set for each individual auscultation position.

[0060] Next, the analysis operation will be described. Fig. 9 is a flow chart showing an example of the analysis operation. The analysis operation is carried out at a place other than a specialized hospital, such as a local clinic or the home of patient A. However, the analysis operation may be used to assist a doctor's diagnosis at a specialized hospital or the like. The analysis operation is started, for example, by operating an analysis operation start button displayed on the screen display unit 14 to start the analysis target lung sound acquisition means 162.

[0061] 9, when the analysis target lung sound acquisition means 162 is started, it creates analysis target lung sound information 153 in which necessary information is entered in each of the fields of patient ID 1531, analysis date and time 1532, person in charge 1533, analysis contact information 1535, and consent information 1537, and other fields are set to NULL values, and records the analysis target lung sound information 153 in the storage unit 15 (step S11). For example, the analysis target lung sound acquisition means 162 acquires the patient ID 1531 from the patient ID 1521 of the lung sound record 152 stored in the storage unit 15. The analysis target lung sound acquisition means 162 also acquires the analysis date and time 1532, person in charge 1533, analysis contact information 1536, and consent information 1537 from the operator via the operation input unit 13. The consent information 1537 may include an electronic signature indicating that patient A has agreed to the use of his or her lung sound data for learning.

[0062] Next, the analysis target lung sound acquisition means 162 acquires a digital time series acoustic signal including the lung sounds at each auscultation position of the patient A from the electronic stethoscope 11, and records the acquired digital time series acoustic signal in the analysis target lung sound information 153 in association with the auscultation position (step S12). Any method may be used to acquire the lung sounds at each auscultation position of the patient by the electronic stethoscope and record the acquired lung sounds in association with the auscultation position. For example, any method may be used, such as a method in which a guidance screen for guiding the operator using the electronic stethoscope 11 to the auscultation position is displayed on the screen display unit 14, as described in Patent Documents 1, 4, or 6. In step S12, the lung sound abnormality detection means 163 reads out the analysis target lung sound information 153 from the storage unit 15, analyzes the lung sound data at each auscultation position of the patient A recorded in the lung sound information 1534 of the analysis target lung sound information 153 using a model created in advance, and records the analysis result in the item of the analysis result for each auscultation position of the lung sound information 1534. Furthermore, in step S12, the analysis result output means 164 appropriately displays the analysis results of the lung sound abnormality detection means 163 on the screen display unit 14.

[0063] Next, the lung sound abnormality detection means 163 calculates the urgency 1535 based on the analysis result of the lung sound data for each auscultation position, and records it in the urgency 1535 item of the analysis target lung sound information 153 (step S13). Next, the analysis result output means 164 reads out the analysis target lung sound information 153 from the storage unit 15, displays the analysis target lung sound information 153 on the screen display unit 14, and also sends an email to which the analysis target lung sound information 153 is attached as a file via the communication I / F unit 12 to the contact email address 1526 of the lung sound record 152 (step S14). In this step S14, as described above, the analysis result output means 164 outputs the analysis target lung sound information 153 for the purpose of notifying relevant parties of the heart failure state of patient A, and also sends the analysis target lung sound information 153 to the terminal device of the heart failure specialist for the purpose of using the lung sound data acquired from patient A as learning data.

[0064] Next, details of step S12 of acquiring lung sounds to be analyzed and detecting abnormalities will be described with reference to the flowchart of Fig. 10. Fig. 10 is a flowchart showing an example of the detailed procedure of step S12 of Fig. 9.

[0065] 10, the analysis target lung sound acquisition means 162 calculates the frequency of abnormal sounds at each auscultation position based on the presence or absence of abnormal sounds recorded in the auscultation findings at each auscultation position in one or more pieces of auscultation information 1527 recorded in the lung sound record 152 of patient A (step S21). Specifically, first, the analysis target lung sound acquisition means 162 initializes the frequency counters for each auscultation position (1) to (12) to 0. Next, the analysis target lung sound acquisition means 162 focuses on the auscultation information 1527 with the most recent auscultation date and time recorded in the lung sound record 152. Next, if there is at least one auscultation finding indicating the presence of an abnormal sound among the one or more auscultation findings recorded at the auscultation position (1), the analysis target lung sound acquisition means 162 increments the frequency count corresponding to the auscultation position (1) by 1. Similarly, if there is at least one auscultation finding indicating the presence of an abnormal sound among one or more auscultation findings recorded at auscultation positions (2) to (12), the analyzed lung sound acquisition means 162 increments the frequency counts corresponding to the auscultation positions (2) to (12) by 1. Next, the analyzed lung sound acquisition means 162 focuses on the auscultation information 1527 with the auscultation date and time immediately preceding the latest one, and performs the same operation as that performed using the latest auscultation information 1527 on the frequency counts for each of the auscultation positions (1) to (12). Similarly, the analyzed lung sound acquisition means 162 repeats the above operation until it has completed processing up to a predetermined number of auscultation information 1527 or has completed processing up to the oldest auscultation information 1527 recorded in the lung sound record 152, whichever occurs first. Then, the analyzed lung sound acquisition means 162 determines the frequency count values ​​for each of the auscultation positions (1)-(12) as the occurrence frequencies of abnormal sounds at the auscultation positions (1)-(12).

[0066] Next, the analysis target lung sound acquisition means 162 determines the order (order) of auscultation positions for auscultating lung sounds from patient A based on the frequency of abnormality at each of the auscultation positions (1) to (12) of patient A (step S22). The fact that there is a difference in the frequency of occurrence of abnormal sounds such as rales among the auscultation positions (1) to (12) of patient A indicates that there are auscultation positions on patient A where abnormal sounds are relatively likely to occur and auscultation positions where they are not. Therefore, by performing auscultation according to the order of auscultation positions determined based on the past frequency of abnormality at each of the auscultation positions (1) to (12) of patient A, the probability of overlooking worsening heart failure can be reduced even if auscultation is interrupted midway for some reason such as patient A's convenience and the state of heart failure of patient A is judged based on the analysis results of the lung sound data at some of the auscultation positions that have been auscultated up to that point.

[0067] The analysis target lung sound acquisition means 162 may determine the order of the auscultation positions based only on the frequency of abnormality at each auscultation position of the patient A. In this case, the analysis target lung sound acquisition means 162 may determine the order of the auscultation positions based on, for example, the result of sorting the auscultation positions in descending order of abnormality frequency (from most to least). For example, when the frequency of abnormality at each auscultation position (1) to (12) of the patient A is shown in FIG. 11, an example of the order of the auscultation positions based on the result of sorting the auscultation positions in descending order of abnormality frequency is shown in auscultation order 1 in FIG. 11. In auscultation order 1, auscultation position (11) with the highest abnormality frequency of 4 is auscultated first. The auscultation positions with the next highest abnormality frequency are auscultation positions (6) and (12) with abnormality frequency of 3. Since there is no difference in the frequency of abnormality, in auscultation order 1, auscultation position (12) located on the back like the first position is the second position, and auscultation position (6) located on the anterior chest is the third position. Similarly, the order of auscultation positions is (5), (9), (10), (7), (1), (2), (3), (4), and (8).

[0068] By determining the order of auscultation positions based only on the frequency of abnormalities at each position on the patient as described above, lung sound data can be acquired in order from the auscultation positions with the highest probability of abnormal lung sounds. However, depending on the distribution of abnormality frequencies, it may be necessary to switch between auscultation of the back and the anterior chest several times, which places a heavy burden on the patient and the operator.

[0069] Therefore, the order of auscultation positions may be determined taking into consideration not only the frequency of abnormalities at each auscultation position on the patient but also the reduction of the burden on the patient and the operator. For example, the analysis target lung sound acquisition means 162 determines the side of the anterior chest or the back where the auscultation position with the highest abnormality frequency is located as the first site to be auscultated, and the opposite site to that site as the next site to be auscultated. In addition, the analysis target lung sound acquisition means 162 sorts all auscultation positions for each site in descending order of their abnormality frequency and determines the order of auscultation positions for that site as the result. An example of an auscultation order based on this determination method is shown in auscultation order 2 in FIG. 11.

[0070] In auscultation order 2, the back, where auscultation position (11) with the highest abnormality frequency of 4 is located, is determined as the first region to be auscultated, and the order of auscultation positions (7) to (12) on the back is sorted in descending order of abnormality frequency, and the order is determined as follows: (11), (12), (9), (10), (7), (8). In addition, in auscultation order 2, after auscultation at all auscultation positions on the back is completed, the auscultation is switched to the anterior chest, and the order of auscultation positions (1) to (6) on the anterior chest is sorted in descending order of abnormality frequency, and the order is determined as follows: (6), (5), (1), (2), (3), (4).

[0071] 10 again, the analysis target lung sound acquisition means 162 focuses on the first auscultation position in the sequence (step S23). Next, the analysis target lung sound acquisition means 162 acquires a digital time-series acoustic signal including the lung sound at the auscultation position being focused on from the electronic stethoscope 11 (step S24). At this time, the analysis target lung sound acquisition means 162 may display a guidance screen on the screen display unit 14 to provide guidance to the operator using the electronic stethoscope 11 on the auscultation position being focused on, thereby supporting the acquisition of the lung sound. The analysis target lung sound acquisition means 162 may also play a guidance sound from a speaker (not shown) to provide guidance to the operator using the electronic stethoscope 11 on the auscultation position being focused on, thereby supporting the acquisition of the lung sound. In this way, the analysis target lung sound acquisition means 162 uses images or audio to guide the operator using the electronic stethoscope 11 as to the auscultation position at which to place the electronic stethoscope 11, based on the order of auscultation positions for auscultating the lung sounds determined for patient A, and acquires from the electronic stethoscope 11 a digital time-series acoustic signal including the lung sounds at the auscultation position being guided.

[0072] Next, the analysis target lung sound acquisition means 162 measures the quality of the acquired lung sound (step S25). Usually, the time-series sound signal output from the electronic stethoscope 11 contains the lung sound of the patient A in a frequency band of 100 Hz to about 2 kHz, and also contains background noise (stationary noise) in the same frequency band. For example, environmental sounds, human voices, metallic sounds, etc. that enter from the outside through the body of the patient A or through the gap between the skin of the patient A and the chest piece are examples of background noise. If the intensity of the lung sound in the time-series sound signal is low and the intensity of the background noise is high, it becomes difficult to detect lung sound abnormalities. Therefore, the analysis target lung sound acquisition means 162 first uses a band-pass filter to extract a time-series sound signal in a frequency band of 100 Hz to about 2 kHz from the time-series sound signal output from the electronic stethoscope 11. Next, the analysis target lung sound acquisition means 162 calculates the intensity of the lung sound and the intensity of the background noise in the extracted time-series sound signal, and calculates the difference between them as an index value of the quality of the lung sound. A method for calculating an index value of lung sound quality will now be described.

[0073] FIG. 12 is a schematic diagram showing an example of the waveform of a time-series acoustic signal including lung sounds output from the electronic stethoscope 11. Generally, lung sounds include tracheal breath sounds, bronchovesicular breath sounds, bronchial breath sounds, and alveolar breath sounds. The lung sound shown in FIG. 12 is a schematic diagram showing an example of alveolar breath sounds that can be heard on most of the chest wall, and thus at all auscultation positions (1) to (12). Referring to FIG. 12, the amplitude of the time-series acoustic signal including lung sounds changes significantly at the start of inspiration. Also, at the start of expiration, although not as much as at the start of inspiration, the amplitude still changes significantly. Therefore, the analysis target lung sound acquisition means 162 compares the time-series acoustic signal with a threshold value T1 that can discriminate the amplitude change at the start of inspiration, and detects the time point when the amplitude of the time-series acoustic signal becomes greater than the threshold value T1 as the start of inspiration. Further, the analysis target lung sound acquisition means 162 sets the interval from the start of one inspiration to the start of the next inspiration as one cycle of respiration, and compares the amplitude of the time-series acoustic signal within that interval with a threshold value T2 (<T1) that can discriminate the amplitude change at the start of expiration, and detects the time point when the amplitude of the time-series signal becomes greater than the threshold value T2 as the start of expiration. Here, if only the distinction between the quiescent phase and other phases is required, it is sufficient to detect only the start of inspiration. However, in the present embodiment, the start of expiration is also detected in order to further divide the phases other than the quiescent phase into an inspiration phase and an expiration phase.

[0074] It is also known that human breathing generally consists of an inhalation phase of about 1 second, an exhalation phase of about 1 second, and a pause phase of about 1 to 1.5 seconds until the next inhalation. That is, there is a pause phase in which neither inhalation nor exhalation occurs immediately before the start of inhalation. The analysis target lung sound acquisition means 162 detects a predetermined period (for example, 1 second) immediately before the start of the detected inhalation as the pause phase. Then, the analysis target lung sound acquisition means 162 calculates the intensity of the time-series acoustic signal in the pause phase as the intensity of the background noise. The intensity of the time-series acoustic signal can be, for example, the root mean square of the amplitude value, but is not limited thereto, and may be the amplitude or the like. The analysis target lung sound acquisition means 162 calculates the intensity of the lung sound by subtracting the intensity of the background noise from the intensity of the time-series acoustic signal in the inhalation phase and / or the exhalation phase. Then, the analysis target lung sound acquisition means 162 sets the ratio of the intensity of the lung sound to the calculated intensity of the background noise as an index value of the quality of the lung sound. The index value of the quality of the lung sounds is not limited to the above, and the S / N ratio calculated from the intensity of the lung sounds and the intensity of the background noise may be used as the index value.

[0075] In the above example, a method for detecting a pause phase has been described using alveolar breath sounds as an example, but bronchoalveolar breath sounds are heard along with alveolar breath sounds at the auscultation positions in the middle and upper lung fields. However, since the amplitude of bronchoalveolar breath sounds during inspiration is equal to or greater than the amplitude of expiration, even when bronchoalveolar breath sounds are heard along with alveolar breath sounds, the start timing of inspiration and expiration can be detected by the method described in FIG. 12. However, when bronchoalveolar breath sounds are close to tracheal breath sounds, the amplitude during expiration may be larger than that during inspiration. Therefore, when bronchoalveolar breath sounds are close to tracheal breath sounds, inspiration and expiration may be reversed in the method described in FIG. 12. Specifically, for example, the following may be performed. First, the frequency at which the amplitude of the frequency spectrum of the auscultated lung sound is maximum is compared with a preset threshold frequency. Next, if the frequency at which the amplitude of the frequency spectrum of the auscultated lung sound is maximum is equal to or greater than the threshold frequency, it is determined that the bronchoalveolar breath sounds contained in the lung sound are close to the tracheal breath sounds, and the timings at which inspiration and expiration start are detected by reversing the inhalation and exhalation in the method described in FIG. 12. On the other hand, if the frequency at which the amplitude of the frequency spectrum of the auscultated lung sound is maximum is less than the threshold frequency, it is determined that the bronchoalveolar breath sounds contained in the lung sound are not close to the tracheal breath sounds, and the timings at which inspiration and expiration start are detected in the method described in FIG. 12. The threshold frequency is a threshold that can determine whether the bronchoalveolar breath sounds contained in the lung sound are close to the tracheal breath sounds or not, and can be determined in advance from a frequency band between the frequency at which the amplitude of the frequency spectrum of the alveolar breath sounds is maximum and the frequency at which the amplitude of the frequency spectrum of the higher tracheal breath sounds is maximum. In addition, instead of the above-mentioned "frequency at which the amplitude of the frequency spectrum is maximum," a "spectral center of gravity" used as a measure for expressing the shape of the spectrum may be used. In the above example, the start time of inspiration and expiration is detected from the time-series acoustic signal output from the electronic stethoscope 11, and a predetermined period immediately before the detected start time of inspiration is detected as a resting phase. However, the method of detecting the inspiration phase, expiration phase, and resting phase is not limited to the above. For example, the analysis target lung sound acquisition means 162 may be configured to input the time-series acoustic signal including the lung sounds of the patient A to a trained learning model that has been trained by machine learning to estimate which section of the time-series acoustic signal including the lung sounds output from the electronic stethoscope is the inspiration phase, expiration phase, or resting phase, and to acquire the estimated probability of the inspiration phase, expiration phase, and resting phase for each section from the training model. The training model can be generated in advance by machine learning using a machine learning algorithm such as a neural network using, for example, a time-series acoustic signal including various lung sounds as teacher data. In addition, the analysis target lung sound acquisition means 162 may detect the timing of breathing, such as the start of inspiration and expiration, of the patient A from other than the time-series acoustic signal output from the electronic stethoscope. For example, the analysis target lung sound acquisition means 162 may detect the timing of breathing of patient A using a respiratory volume sensor such as a pulmonary tachograph, or a respiratory band that uses a sensor to detect changes in the shape of the chest or abdomen due to respiratory activity.

[0076] Next, the analysis target lung sound acquisition means 162 compares the index value of the quality of the lung sound with a preset quality threshold value (step S26). If the index value of the quality of the lung sound is smaller than the threshold value, the analysis target lung sound acquisition means 162 displays an alarm on the screen display unit 14 to the effect that the quality of the lung sound at the auscultation position auscultated by the electronic stethoscope 11 is poor (step S27). The operator who recognizes this alarm takes measures to reduce background noise and / or measures to increase the lung sound, and then performs the task of acquiring the lung sound at the auscultation position of interest again with the electronic stethoscope 11 (step S28). Possible measures to reduce background noise include closing the window to make the room quieter, and pressing the chest piece tightly against the skin of the patient A so that environmental sounds do not enter through the gap between the patient A's skin and the chest piece. Also, possible measures to increase the lung sound include instructing the patient A to breathe more deeply. At this time, for example, the timing of breathing may be instructed to the patient A by a method such as that described in Patent Document 9. Then, the analyzed lung sound acquisition means 162 returns to the process of step S25, and repeats the same process as described above.

[0077] On the other hand, if the index value of the quality of the lung sounds is equal to or greater than the threshold value, the analysis target lung sound acquisition means 162 removes the pause phase period and background noise from the digital time-series acoustic signal including the lung sounds at the auscultation position of interest, and records the digital time-series acoustic signal after the pause phase period and background noise have been removed in association with the auscultation position of interest in the analysis target lung sound information 153 (step S29). The pause phase period and background noise are removed as follows.

[0078] First, the analysis target lung sound acquisition means 162 divides the digital time series acoustic signal including the lung sound at the auscultation position of interest into two sections: an inhalation phase and an exhalation phase immediately thereafter (hereinafter referred to as the inhalation-exhalation section), and a resting phase (hereinafter referred to as the resting section). Next, the analysis target lung sound acquisition means 162 performs a fast Fourier transform (FFT) on the digital time series acoustic signals of the inhalation-exhalation section and the resting section, respectively, to calculate the frequency spectrum of the inhalation-exhalation section and the resting section. Next, the analysis target lung sound acquisition means 162 subtracts the frequency spectrum of the resting section from the frequency spectrum of the inhalation-exhalation section. This subtraction suppresses background noise contained in the inhalation phase and the exhalation phase. Next, the analysis target lung sound acquisition means 162 performs an inverse frequency transform on the frequency spectrum of the inhalation-exhalation section after the subtraction, thereby generating a digital time series acoustic signal after noise removal of the inhalation-exhalation section. Then, the analyzed lung sound acquiring means 162 records the generated digital time-series acoustic signal after noise removal for the inhalation / exhalation period in association with the auscultation position of interest in the analyzed lung sound information 153. The analyzed lung sound acquiring means 162 may remove the pause phase period from the digital time-series acoustic signal including the lung sounds at the auscultation position of interest, without removing background noise. In this case, the analyzed lung sound acquiring means 162 divides the digital time-series acoustic signal including the lung sounds at the auscultation position of interest into an inhalation / exhalation period and a pause period, and records the digital time-series acoustic signal for the inhalation / exhalation period in association with the auscultation position of interest in the analyzed lung sound information 153.

[0079] Next, the lung sound abnormality detection means 163 detects lung sound abnormalities from the lung sound data recorded in the analyzed lung sound information 153 in association with the auscultation position of interest, and records the detection result in the analyzed lung sound information 153 in association with the auscultation position of interest (step S30). Lung sound abnormalities are detected by the anomaly detection method using supervised learning as described above.

[0080] Each time the lung sound abnormality detection means 163 detects an abnormality in the lung sound data at the auscultation position of interest, the analysis result output means 164 displays the abnormality detection result on the screen display unit 14 (step S31). This allows the operator to immediately recognize whether the lung sound data at the auscultation position is an abnormal lung sound during auscultation.

[0081] When the acquisition and analysis of lung sound data for the currently selected auscultation position is completed, the analysis target lung sound acquisition means 162 determines whether the acquisition and analysis of lung sound data for all auscultation positions is completed (step S32). If there are any auscultation positions for which acquisition has not been completed, the analysis target lung sound acquisition means 162 shifts its attention to the next auscultation position (step S33), returns to step S24, and repeats the same process as described above.

[0082] Furthermore, when the acquisition and analysis of lung sound data at all auscultation positions has been completed, the analysis target lung sound acquisition means 162 ends the processing in Fig. 10. Furthermore, the analysis target lung sound acquisition means 162 may end the processing in Fig. 10 before the acquisition and analysis of lung sound data at all auscultation positions has been completed due to the convenience of patient A, etc. If the processing in Fig. 10 ends midway, lung sound information 1534 in the analysis target lung sound information 153 corresponding to the auscultation positions where lung sound data has not been acquired and analyzed remains as a NULL value.

[0083] Next, step S13 in FIG. 9 for calculating the urgency level 1535 will be described in detail.

[0084] The lung sound abnormality detection means 163 determines the severity of heart failure of patient A based on the analysis results of the lung sound data for each auscultation position, and calculates the urgency level 1535 based on the determined severity. In determining the severity of heart failure, the lung sound abnormality detection means 163 determines the severity of heart failure by referring to a determination table for determining the severity of heart failure from the analysis results of the lung sound data for each auscultation position.

[0085] Fig. 13 is a diagram showing an example of the above-mentioned judgment table. The judgment table shown in Fig. 13 has columns that correspond one-to-one to the auscultation positions (1) to (12) and rows that correspond one-to-one to the severity levels, and a + symbol indicating the presence of abnormal lung sounds and a - symbol indicating the absence of abnormal lung sounds are set at the intersections of the rows and columns. Referring to Fig. 13, in the judgment table, if there is no abnormal lung sounds at any of the auscultation positions, the severity level is judged to be 0. Also, in the judgment table, if there is an abnormal lung sound at at least one of the auscultation positions (11) and (12) set in the lower lung field on the back, and there is no abnormal lung sound at the other auscultation positions (1) to (10), the severity level is judged to be 1. In the judgment table, when there is an abnormality in lung sounds at both auscultation positions (11) and (12), there is an abnormality in lung sounds at only one of auscultation positions (5) and (6) set in the lower lung field of the anterior chest, and there is no abnormality in lung sounds at the other auscultation positions (1) to (4) and (7) to (10), the severity is judged to be 2. The severity N set in the last row is considered to be an abnormality in lung sounds at all auscultation positions (1) to (12). In FIG. 13, the severity levels 1 and above between severity levels 2 and N are omitted, but auscultation positions where there is an abnormality in lung sounds and auscultation positions where there is no abnormality in lung sounds are set for these severity levels as well. For severity levels 1 and above between severity levels 2 and N, the number of auscultation positions where there is an abnormality in lung sounds is 4 or more and less than 12, and the number increases as the severity level approaches N.

[0086] In the judgment table shown in FIG. 13, the severity of heart failure is classified into N+1 classes from severity 0 to severity N according to the combination of the presence or absence of abnormal lung sounds at auscultation positions (1) to (12). Here, severity 0 is a state in which no abnormal lung sounds are heard, and therefore it can be said that heart failure is in remission. Severity 1 is a state in which abnormal lung sounds can be heard only in the lower lung field in the back, and therefore it cannot be said that heart failure is in remission, but it is a mild state, and there are patients who are discharged in this state. Severity 2 is a state in which abnormal lung sounds are heard not only in the lower lung field in the back but also in one of the lower lung fields in the anterior chest, and therefore it can be said that it is more severe than severity 1. However, since it is still a mild state, it can be said that there is a high probability that readmission can be prevented if appropriate measures are taken at this point.

[0087] The judgment table for judging the severity of heart failure from the analysis results of the auscultation position is not limited to the one shown in Fig. 13. For example, it is known that rales are mild when they are heard only at the end of inspiration, and severe when they are heard immediately after inspiration begins. Therefore, in addition to the presence or absence of abnormal lung sounds for each auscultation position, the judgment table may set the type of abnormal lung sounds and the timing at which the abnormal lung sounds are heard, and the severity of heart failure may be judged based on the combination of the auscultation position, the presence or absence of abnormal lung sounds, the type of abnormal lung sounds, and the timing at which the abnormal lung sounds are heard.

[0088] Furthermore, the abnormal lung sound detection means 163 may determine the severity of heart failure of patient A from the number of auscultation positions where abnormal lung sounds were detected, regardless of the auscultation position. For example, the abnormal lung sound detection means 163 may determine the severity as 0, 1, 2, 3, or 4 (maximum) when the number of auscultation positions where abnormal lung sounds were detected is 0, 1 to 2, 3 to 4, 5 to 8, or 9 or more, respectively.

[0089] Furthermore, when at least a part of the analysis results of the lung sound information 1534 for each auscultation position is a NULL value due to the processing shown in FIG. 10 being terminated midway due to the convenience of patient A, etc., the lung sound abnormality detection means 163 performs the following processing. First, the lung sound abnormality detection means 163 judges whether or not the number of auscultation positions where the analysis results are NULL values, that is, where no lung sound data has been acquired and no analysis has been performed as to whether the lung sound is abnormal, is less than a first threshold value set in advance. In other words, the lung sound abnormality detection means 163 judges whether or not the number of auscultation positions where lung sound data has been acquired and analysis has been performed as to whether the lung sound is abnormal is equal to or greater than a second threshold value set in advance. Here, the first threshold value and the second threshold value may be fixed values ​​or variable values ​​according to the latest condition of the patient. When the values ​​are fixed, for example, the first threshold value may be 4 or less, and the second threshold value may be 8 or more. Furthermore, when using variable values, for example, the first threshold value may be 10 or less and the second threshold value may be 2 or more for a patient who has no abnormal lung sounds at any auscultation position in the latest state, and for other patients, the values ​​may be the same as fixed values. If the above conditions are not satisfied, the lung sound abnormality detection means 163 does not calculate the severity (and therefore does not calculate the urgency either), ends the current lung sound analysis with an error, and displays a message to that effect on the screen display unit 14. This is to prevent erroneous information from being given to the operator, etc.

[0090] On the other hand, if the above conditions are satisfied, the lung sound abnormality detection means 163 calculates the severity on the assumption that no abnormal lung sound was detected at the auscultation positions where no analysis of whether or not there is abnormal lung sound has been performed. The lung sound abnormality detection means 163 then holds the calculated severity as the most optimistic value. That is, if the calculated severity is severity 1, it is held as "severity 1 or higher" or "at least severity 1" rather than "severity 1". For example, suppose that for a patient who has no abnormal lung sounds at any auscultation position in the latest state, acquisition and analysis of lung sound data is performed only at two auscultation positions (11) and (12), and as a result, abnormal lung sounds are detected at at least one of the auscultation positions. In this case, the lung sound abnormality detection means 163 assumes that no abnormal lung sound was detected at the other auscultation positions (1) to (10), and judges the severity to be 1 based on the judgment table in FIG. 13, and based on this, judges the severity to be "severity 1 or higher".

[0091] After judging the severity of heart failure from the analysis results of the lung sound data for each auscultation position as described above, the lung sound abnormality detection means 163 determines the urgency 1535 from the determined severity. For example, the lung sound abnormality detection means 163 may determine the urgency 1535 based only on the severity of heart failure, 0 to N. In other words, the lung sound abnormality detection means 163 may set the possible range of urgency 1535 to N+1 classes, from urgency 0 to urgency N, and determine the urgency i that corresponds one-to-one to the determined severity i of heart failure (i=0 to N).

[0092] Furthermore, the lung sound abnormality detection means 163 may determine the urgency 1535 based on the severity of heart failure 0 to N and the condition of patient A. For example, the condition of patient A may be whether or not the weight has increased by a certain amount in a unit period (for example, increased by 3 kg or more in one week), whether or not there are subjective symptoms such as swelling, coughing, and loss of appetite, and whether or not the pulse rate exceeds a predetermined number. Then, the lung sound abnormality detection means 163 may correct the urgency determined based on the severity of heart failure to a higher value according to the condition of patient A, and use this as the final urgency. For example, even if the urgency determined based on the severity of heart failure is 0 or 1, the lung sound abnormality detection means 163 may increase the urgency to 1 or 2 if there is weight gain. However, the upper limit of the urgency after correction is N.

[0093] Next, the operation after the analysis result output means 164 transmits the email to which the analysis target lung sound information 153 is attached as a file to the terminal device of the heart failure specialist will be described.

[0094] In the terminal device of the heart failure specialist who received the email, the analysis target lung sound information 153 stored in the attached file is analyzed by the heart failure specialist. The analysis target lung sound information 153 may be shared with the heart failure specialist not only by file attachment but also by a SaaS format such as by writing a link. For example, the specialist plays back the lung sound data for each auscultation position recorded in the analysis target lung sound information 153 using a personal computer or the like, and diagnoses whether or not any accessory noise such as rales is heard from the lung sounds of patient A. Then, the specialist creates auscultation findings for the lung sound data for each auscultation position, and records them in the analysis target lung sound information 153 as shown in FIG. 6. The analysis target lung sound information 153 with the specialist's auscultation findings recorded in this way is returned to the lung sound analysis device 10 that sent it by communication means such as email.

[0095] The learning data generating means 165 updates the original analysis target lung sound information 153 recorded in the storage unit 15 with the analysis target lung sound information 153 with auscultation findings received through the communication I / F unit 12 of the lung sound analysis device 10. Next, the learning data generating means 165 creates learning data for each pair of lung sound data and auscultation findings recorded corresponding to each auscultation position of the analysis target lung sound information 153. For example, the learning data generating means 165 creates learning data including a label indicating that the lung sound is abnormal and the lung sound data from a pair of auscultation findings indicating that the lung sound is abnormal and the lung sound data. Also, the learning data generating means 165 creates learning data including a label indicating that the lung sound is normal and the lung sound data from a pair of auscultation findings indicating that the lung sound is normal and the lung sound data. At this time, personal information of the patient may be added to the label. Also, the learning data generating means 165 records the created learning data in the learning data DB 154 in association with the auscultation position. At this time, the learning data generating means 165 may record the learning data in a manner that distinguishes it from other learning data already recorded in the learning data DB 154 by attaching a time stamp such as the date and time of recording to the learning data.

[0096] The learning data added to the learning data DB 154 in the above manner is used the next time the learning means 166 re-learns the model for detecting lung sound abnormalities. In this manner, the accuracy of the model for detecting lung sound abnormalities can be gradually improved.

[0097] As described above, according to this embodiment, learning data for detecting abnormal lung sounds can be efficiently collected in clinics in towns without specialists, patients' homes, etc. The reason for this is that the lung sound analysis device 10 acquires a time-series acoustic signal including the lung sounds of a patient with heart failure, detects abnormal lung sounds from this acquired time-series acoustic signal, transmits analysis target lung sound information in which the acquired time-series acoustic signal and the detection result are associated with each other to a terminal device of a specialist, and upon receiving analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added, generates learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added.

[0098] [Second embodiment] Fig. 14 is a block diagram of a lung sound analysis system 20 according to a second embodiment of the present invention. Referring to Fig. 14, the lung sound analysis system 20 is composed of a plurality of lung sound analysis devices 21, a server device 22, and a terminal device 24. The plurality of lung sound analysis devices 21, the server device 22, and the terminal device 24 are connected to each other so as to be able to communicate with each other via a network such as the Internet.

[0099] The lung sound analyzer 21 is an information processing device that acquires and analyzes lung sounds from a patient who has been treated for heart failure and has been discharged from the hospital. The terminal device 24 is a terminal device used by a heart failure specialist. The lung sound analyzer 21 and the terminal device 24 may be, but are not limited to, a smartphone, a tablet terminal, a PDA, a notebook computer, or the like. The lung sound analyzer 21 includes an electronic stethoscope, a communication I / F unit, an operation input unit, a screen display unit, a memory unit, and a calculation processing unit, all of which are not shown. The terminal device 24 also includes a communication I / F unit, an operation input unit, a screen display unit, a memory unit, and a calculation processing unit, all of which are not shown.

[0100] Server device 22 is a computer that provides various services required for lung sound analysis to multiple lung sound analysis devices 21 via network 23. For example, server device 22 stores at least a portion of lung sound records 152, analysis target lung sound information 153, learning data DB 154, and program 151 shown in Fig. 1 and provides them to lung sound analysis devices 21 via network 23. Therefore, compared to lung sound analysis device 10 of Fig. 1, lung sound analysis device 21 does not need to store at least a portion of lung sound records 152, analysis target lung sound information 153, learning data DB 154, and program 151 in memory unit 15, and can reduce memory capacity.

[0101] 1 to the lung sound analysis device 21 via the network 23. For example, the server device 22 executes at least a part of the processes of steps S1 to S2 in Fig. 7, steps S11 to S14 in Fig. 9, and steps S21 to S33 in Fig. 10, or the learning data generation process, on behalf of the lung sound analysis device 21. Therefore, the lung sound analysis device 21 can simplify the configuration of the calculation processing unit 16 compared to the lung sound analysis device 10 in Fig. 1.

[0102] Furthermore, terminal device 24 has a function of playing back lung sound data for each auscultation position recorded in analysis target lung sound information received by email or the like from multiple lung sound analyzers 21, a function of inputting auscultation findings from a heart failure specialist, and a function of transmitting analysis target lung sound information with auscultation findings by email or the like to lung sound analyzer 21. Although only one terminal device 24 is shown in Fig. 14, there may be a number of terminal devices 24 corresponding to the number of heart failure specialists.

[0103] [Third embodiment] Fig. 15 is a block diagram of a lung sound analysis system 30 according to a third embodiment of the present invention. Referring to Fig. 14, the lung sound analysis system 30 includes an acquisition unit 31, a detection unit 32, a finding acquisition unit 33, and a generation unit 34.

[0104] The acquiring means 31 is configured to acquire a time-series acoustic signal including a lung sound of a patient with heart failure. The acquiring means 31 can be configured similarly to, for example, step S24 in FIG. 10, but is not limited thereto.

[0105] The detection means 32 is configured to detect abnormal lung sounds from the time-series acoustic signal acquired by the acquisition means 31. The detection means 32 can be configured, for example, in the same manner as in step S30 of Fig. 10, but is not limited thereto.

[0106] The finding acquiring means 33 is configured to transmit analysis target lung sound information in which the time-series acoustic signal acquired by the acquiring means 31 corresponds to the detection result of the detecting means 32 to a terminal device of a specialist, and to receive analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device. The finding acquiring means 33 can be configured in the same manner as, for example, the analysis result output means 164 in Fig. 1, but is not limited thereto.

[0107] The generating means 34 is configured to generate learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added, received by the findings acquiring means 33. The generating means 34 can be configured similarly to, for example, the learning data generating means 165 in Fig. 1, but is not limited thereto.

[0108] The lung sound analysis system 30 configured as above functions as follows. That is, first, the acquisition means 31 acquires a time-series acoustic signal including the lung sounds of a patient with heart failure. Next, the detection means 32 detects abnormal lung sounds from the acquired time-series acoustic signal. Next, the findings acquisition means 33 transmits analysis target lung sound information in which the time-series acoustic signal acquired by the acquisition means 31 is associated with the detection result of the detection means 32 to a terminal device of a specialist, and receives analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device. Next, the generation means 34 generates learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added, received by the findings acquisition means 33.

[0109] The lung sound analysis system 30 configured and operated as described above can efficiently collect learning data for detecting abnormal lung sounds in clinics in towns without specialists, patients' homes, etc. The reason for this is that the lung sound analysis system 30 acquires a time-series acoustic signal including the lung sounds of a patient with heart failure, detects abnormal lung sounds from this acquired time-series acoustic signal, transmits analysis target lung sound information in which the acquired time-series acoustic signal is associated with the detection result to a terminal device of a specialist, and upon receiving analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added, generates learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added.

[0110] Although the present invention has been described with reference to the above-mentioned embodiments, the present invention is not limited to the above-mentioned embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention. For example, the following configurations are also included in the present invention.

[0111] For example, in the lung sound analysis device 10 shown in Fig. 1, the lung sound abnormality detection means 163 detects abnormalities in the patient's lung sound data acquired by the electronic stethoscope 11 using an abnormality detection method based on supervised learning. However, the lung sound abnormality detection means 163 may detect abnormal lung sounds using an abnormality detection method other than the above, for example, an abnormality detection method based on unsupervised learning. An example of an abnormality detection method based on unsupervised learning is a method in which abnormal sounds are not used in learning, normal lung sound data is learned, and lung sounds that deviate from the learned normal sounds are detected as abnormal lung sounds.

[0112] Furthermore, for example, the analysis target lung sound acquisition means may instruct the subject to take a deeper breath when it is determined that the lung sound has not been recorded correctly. The analysis target lung sound acquisition means may instruct the operator on the position of auscultation by an AR (Augmented Reality) display. The analysis target lung sound acquisition means may change the position of auscultation based on pre-registered information such as the gender of the subject. The analysis target lung sound acquisition means may start instructing the operator to breathe when it detects that the stethoscope has been placed on the subject, i.e., that the chest piece has come into contact with the subject's body. The analysis target lung sound acquisition means may instruct the operator to breathe by displaying or voice of an avatar designated by the subject. The analysis target lung sound acquisition means may urge the operator to acquire lung sound by displaying or voice of an avatar designated by the subject when lung sound acquisition has not been performed within a predetermined period of time.

[0113] Furthermore, the analysis result output means may display the analysis results, the lung sound records used in the analysis, and the stored information including the analyzed lung sound information in chronological order on a screen display unit, etc. Furthermore, the analysis result output means may transmit information to the server even if no abnormality is detected. [Industrial Applicability]

[0114] INDUSTRIAL APPLICABILITY The present invention can be used in a system for analyzing human lung sounds, and in particular in a system for creating learning data for detecting lung sound abnormalities.

[0115] A part or all of the above-described embodiments may be described as, but is not limited to, the following supplementary notes. [Appendix 1] An acquisition means for acquiring a time-series acoustic signal including lung sounds of a patient with heart failure; a detection means for detecting abnormal lung sounds from the acquired time-series acoustic signals; a finding acquisition means for transmitting analysis target lung sound information in which the acquired time-series acoustic signal is associated with a result of the detection to a terminal device of a specialist, and receiving analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device; a generating means for generating learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added; A lung sound analysis system comprising: [Appendix 2] The finding acquisition means, before performing the transmission, determines whether or not an abnormal lung sound has been detected as a result of the detection, and performs the transmission only when an abnormal lung sound has been detected. 2. A lung sound analysis system as described in appended claim 1. [Appendix 3] The finding acquisition means, prior to the transmission, determines whether or not consent has been obtained from the patient with heart failure for the time-series acoustic signal to be used as the learning data, and performs the transmission only if consent has been obtained. 3. A lung sound analysis system according to claim 1 or 2. [Appendix 4] The finding acquisition means, before performing the transmission, determines whether or not the personal information of the heart failure patient matches the personal information of a target for collecting lung sound data, and performs the transmission only if the personal information matches. 4. A lung sound analysis system according to any one of claims 1 to 3. [Appendix 5] The acquisition means determines a resting phase of the breathing of the heart failure patient, and divides the time-series acoustic signal into a period of the resting phase of the heart failure patient and a period other than the resting phase according to a result of the determination. 5. A lung sound analysis system according to any one of claims 1 to 4. [Appendix 6] The detection means detects lung sound abnormality from the time-series acoustic signal in a period other than the rest phase after the division. 6. The lung sound analysis system of claim 5. [Appendix 7] Acquire a time-series acoustic signal including lung sounds of a patient with heart failure; Detecting abnormal lung sounds from the acquired time-series acoustic signals; transmitting analysis target lung sound information in which the acquired time-series acoustic signal and the result of the detection are associated with each other to a terminal device of a specialist, and receiving analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device; Based on the analysis target lung sound information to which the specialist's findings have been added, learning data is generated for detecting abnormal lung sounds. Lung sound analysis method. [Appendix 8] In the transmission, before the transmission is performed, it is determined whether or not an abnormal lung sound is detected as a result of the detection, and the transmission is performed only when an abnormal lung sound is detected. 8. A method for lung sound analysis as described in appendix 7. [Appendix 9] In the transmission, before the transmission, it is determined whether or not consent has been obtained from the patient of heart failure regarding the time-series acoustic signal being used as the learning data, and the transmission is performed only if consent has been obtained. 9. The lung sound analysis method according to claim 7 or 8. [Appendix 10] In the transmission, before the transmission, it is determined whether or not the personal information of the heart failure patient matches the personal information of a target for collecting lung sound data, and the transmission is performed only if the personal information matches. 10. A lung sound analysis method according to any one of claims 7 to 9. [Appendix 11] On the computer, acquiring a time-series acoustic signal including lung sounds of a patient with heart failure; detecting abnormal lung sounds from the acquired time-series acoustic signals; transmitting analysis target lung sound information in which the acquired time-series acoustic signal and the result of the detection are associated with each other to a terminal device of a specialist, and receiving the analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device; A process of generating learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added; A computer-readable recording medium having a program recorded thereon for causing a computer to carry out the above. [Explanation of symbols]

[0116] 10 Lung sound analyzer 11 Electronic stethoscope 12 Communication I / F section 13 Operation input section 14 Screen display section 15 Storage section 16 Processing unit 151 Programs 152 Lung Sound Recording 153 Lung sound information to be analyzed 154 Learning Data DB 161 Lung sound recording acquisition method 162 Means of acquiring lung sounds to be analyzed 163 Lung sound abnormality detection method 164 Analysis result output means 165 Training Data Generation Means 166 Learning Tools

Claims

1. An acquisition means for acquiring a time-series acoustic signal including lung sounds of a patient with heart failure, and the gender, age, BMI, and medical history of the patient with heart failure; a detection means for detecting abnormal lung sounds from the acquired time-series acoustic signals; a finding acquisition means for transmitting analysis target lung sound information in which the acquired time-series acoustic signal is associated with a result of the detection to a terminal device of a specialist, and receiving analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device; a generating means for generating learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added; Equipped with The finding acquisition means, before transmitting, determines whether the acquired gender, age, BMI, and medical history of the heart failure patient match a gender, age group, BMI range, and medical history set in advance, and transmits the findings only if they match. Lung sound analysis system.

2. The finding acquisition means, before performing the transmission, determines whether or not an abnormal lung sound has been detected as a result of the detection, and performs the transmission only when an abnormal lung sound has been detected.

2. The lung sound analysis system of claim 1.

3. The finding acquisition means, prior to the transmission, determines whether or not consent has been obtained from the patient with heart failure for the time-series acoustic signal to be used as the learning data, and performs the transmission only if consent has been obtained.

3. A lung sound analysis system according to claim 1 or 2.

4. The acquisition means determines a resting phase of the breathing of the heart failure patient, and divides the time-series acoustic signal into a period of the resting phase of the heart failure patient and a period other than the resting phase according to a result of the determination.

4. A lung sound analysis system according to any one of claims 1 to 3.

5. The detection means detects lung sound abnormality from the time-series acoustic signal in a period other than the rest phase after the division.

5. A lung sound analysis system according to claim 4.

6. 1. A method for computerized lung sound analysis, comprising: The computer acquires a time-series acoustic signal including lung sounds of a heart failure patient, and the gender, age, BMI, and medical history of the heart failure patient; The computer detects abnormal lung sounds from the acquired time series acoustic signals; The computer determines whether the acquired gender, age, BMI, and medical history of the heart failure patient match a gender, age group, BMI range, and medical history set in advance; the computer transmits to a terminal device of a specialist lung sound information to be analyzed, the lung sound information being associated with the acquired time-series acoustic signal and the detection result, only when the acquired gender, age, BMI, and medical history of the heart failure patient match the pre-set gender, age group, BMI range, and medical history; the computer receives from the terminal device the lung sound information to be analyzed, to which the specialist's findings on the time-series acoustic signal have been added; The computer generates learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added. Lung sound analysis method.

7. The computer, In the transmission, before the transmission is performed, it is determined whether or not an abnormal lung sound is detected as a result of the detection, and the transmission is performed only when an abnormal lung sound is detected.

7. A method for analyzing lung sounds according to claim 6.

8. The computer, In the transmission, before the transmission, it is determined whether or not consent has been obtained from the patient of heart failure regarding the time-series acoustic signal being used as the learning data, and the transmission is performed only if consent has been obtained.

8. A method for analyzing lung sounds according to claim 6 or 7.

9. On the computer, Obtaining a time-series acoustic signal including lung sounds of a patient with heart failure, and the gender, age, BMI, and medical history of the patient with heart failure; detecting abnormal lung sounds from the acquired time-series acoustic signals; A process of determining whether the acquired gender, age, BMI, and medical history of the heart failure patient match a gender, age group, BMI range, and medical history set in advance; transmitting analysis target lung sound information in which the acquired time-series acoustic signal corresponds to a result of the detection to a terminal device of a specialist only when the acquired gender, age, BMI, and medical history of the heart failure patient match the pre-set gender, age range, BMI range, and medical history, and receiving the analysis target lung sound information to which the specialist's findings on the time-series acoustic signal have been added from the terminal device; A process of generating learning data for detecting abnormal lung sounds based on the analysis target lung sound information to which the specialist's findings have been added; A program to carry out the above.

Images