Deep learning-based central airway stenosis degree evaluation method and system

Abstract

The invention relates to the field of medical diagnosis, in particular to a central airway stenosis degree evaluation method and system based on deep learning, and the method comprises the steps: obtaining breath sound simulation data of the body surface of a patient; denoising and windowing the breath sound simulation data, and extracting features to obtain breath sound pathological features; the patient is examined to obtain the airway stenosis degree and airway stenosis partitions, a preset basic network is adjusted according to the breath sound simulation data, the breath sound pathological features, the airway stenosis degree and the airway stenosis partitions, and an adjusted network structure is obtained; constructing an initial breath sound analysis model according to the adjusted network structure, and training by using breath sound simulation data to obtain a standard breath sound analysis model; acquiring to-be-diagnosed breath sound simulation data of a to-be-diagnosed patient, and diagnosing the to-be-diagnosed breath sound simulation data by using the standard breath sound analysis model to obtain an illness state diagnosis result of the to-be-diagnosed patient. The reliability of the diagnosis result is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical diagnosis, and in particular to a method and system for assessing the degree of central airway stenosis based on deep learning. Background Art

[0002] Chronic respiratory diseases are a major public health issue and a leading cause of illness and death worldwide. Traditionally, doctors diagnose chronic respiratory diseases by listening to the lungs and identifying breath sounds based on experience. However, this method can be highly inaccurate and affected by environmental factors and the doctor's subjective judgment.

[0003] Different lung diseases require the collection of different types of breath sounds for analysis. Currently, there is no simple and effective screening method for central airway stenosis. In clinical practice, auscultation of breath sounds is often used for preliminary screening, but most of the judgments are based on clinical experience. The accuracy is related to the doctor's familiarity with the disease, so the misdiagnosis rate is relatively high. Summary of the Invention

[0004] The present invention provides a method for assessing the degree of central airway stenosis based on deep learning, the main purpose of which is to improve the reliability of medical diagnosis results.

[0005] To achieve the above objectives, the present invention provides a method for assessing the degree of central airway stenosis based on deep learning, comprising:

[0006] Acquiring body surface sounds of the patient based on a preset sound collection device to obtain comprehensive body surface sounds, and extracting respiratory sounds from the comprehensive body surface sounds to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sounds are collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus;

[0007] Obtaining a frequency band range of the respiratory sound analog signal data, selecting an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, performing windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data, and constructing a respiratory sound feature matrix based on the windowed respiratory sound analog signal data;

[0008] Acquiring respiratory sound pathological characteristics based on the respiratory sound feature matrix, performing an airway examination on the patient using a preset examination device to obtain the degree of airway stenosis and airway stenosis zones, and adjusting a preset basic network based on the respiratory sound simulation signal data, the respiratory sound pathological characteristics, the degree of airway stenosis and the airway stenosis zones to obtain an adjusted network structure;

[0009] constructing an initial respiratory sound analysis model according to the adjusted network structure, and training the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model;

[0010] Acquire the respiratory sound simulation signal data of the patient to be diagnosed, obtain the respiratory sound simulation signal data to be diagnosed, and use the standard respiratory sound analysis model to diagnose the respiratory sound simulation signal data to obtain the diagnosis result of the patient to be diagnosed.

[0011] Optionally, extracting the respiratory sounds from the comprehensive body surface sounds to obtain respiratory sound simulation signal data includes:

[0012] Segmenting the comprehensive body surface sound based on a preset time period to obtain comprehensive sound samples;

[0013] Extracting sound features of the respiratory sound from the comprehensive sound sample to obtain a respiratory sound feature, and converting the respiratory sound feature into a vector to obtain a respiratory sound feature vector;

[0014] Constructing a projection matrix based on the respiratory sound feature vector to obtain a respiratory sound projection matrix;

[0015] The respiratory sound projection matrix is ​​used to separate the heartbeat sound and the respiratory sound in the comprehensive body surface sound to obtain a respiratory sound vector, and signal restoration processing is performed based on the respiratory sound vector to obtain respiratory sound simulation signal data, wherein the signal restoration processing includes inverse discrete cosine transform, feature matching and denoising optimization.

[0016] Optionally, extracting the sound features of the respiratory sound in the comprehensive sound sample to obtain a respiratory sound feature, and converting the respiratory sound feature into a vector to obtain a respiratory sound feature vector includes:

[0017] Sorting the integrated sound samples according to time sequence to obtain an integrated sound sample sequence;

[0018] Obtaining a time-frequency signal of the integrated sound sample sequence, calculating a covariance matrix based on the time-frequency signal of the integrated sound sample sequence to obtain a respiratory sound covariance matrix, and extracting off-diagonal element energy of the respiratory sound covariance matrix using a pre-built joint approximate diagonalization algorithm to obtain independent signal components;

[0019] The main diagonal elements in the respiratory sound covariance matrix are extracted based on the independent signal components, and the main diagonal elements are used as feature vectors to construct the respiratory sound feature vector.

[0020] Optionally, the acquiring a frequency band range of the respiratory sound analog signal data, and selecting an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, includes:

[0021] Acquiring a frequency band range of the respiratory sound analog signal data, and selecting a filter corresponding to the filtering range according to the frequency band range to obtain a screening filter;

[0022] The cutoff frequency of the screening filter is set according to the frequency band range to obtain a target filter, and the target filter is used to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data.

[0023] Optionally, performing windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data includes:

[0024] Dividing the denoised respiratory sound analog signal data based on a pre-built sliding window method to obtain a plurality of short-time window respiratory sound analog signal data, wherein the sliding window corresponding to the sliding window method has a window length of 25 milliseconds and an overlap rate of 50%±5%;

[0025] Obtaining a window function, performing windowing processing on the multiple short-time windowed respiratory sound analog signal data using the window function to obtain multiple short-time windowed respiratory sounds, and splicing the multiple short-time windowed respiratory sounds in chronological order to obtain windowed respiratory sound analog signal data;

[0026] The short-term windowed breath sound is expressed as:

[0027] S windowed (n) = S filtered (n)·w(n)

[0028] Among them, S filtered (n) is the sampling value of the n-th short-time window respiratory sound analog signal data, w(n) is the value of the window function corresponding to the n-th short-time window respiratory sound analog signal data, S windowed (n) is the short-time windowed respiratory sound corresponding to the n-th short-time window respiratory sound analog signal data.

[0029] Optionally, constructing a respiratory sound feature matrix according to the windowed respiratory sound simulation signal data includes:

[0030] The windowed respiratory sound analog signal data is converted into cepstral coefficients using a pre-constructed discrete cosine transform method, and a matrix is ​​constructed according to the cepstral coefficients to obtain a respiratory sound feature matrix.

[0031] Optionally, obtaining respiratory sound pathological features based on the respiratory sound feature matrix includes:

[0032] Using a pre-built correlation analysis method, a feature having a correlation with a preset pathological feature greater than a preset threshold is obtained from the respiratory sound feature matrix to obtain a pathology-related feature, and a pathology-related feature matrix is ​​constructed based on the pathology-related feature;

[0033] The pathology-related feature matrix is ​​subjected to dimensionality reduction processing using a pre-constructed principal component analysis method to obtain a reduced-dimensional respiratory sound feature matrix, and features in the reduced-dimensional respiratory sound matrix are extracted to obtain respiratory sound pathology features.

[0034] Optionally, adjusting the preset basic network according to the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree and airway stenosis partition to obtain an adjusted network structure includes:

[0035] Performing convolution on the basic network using a convolution layer with a preset accuracy based on the respiratory sound simulation signal data and the respiratory sound pathological characteristics, thereby obtaining parameters of one or two convolution layers in the basic network;

[0036] performing convolution in the basic network using a 4×4 convolution kernel according to the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree, and airway stenosis partition, and comparing the convolution accuracy of the convolution kernel with a preset network diagnostic target to obtain a convolution kernel comparison difference;

[0037] The basic network is adjusted according to the parameters of the first and second convolution layers in the basic network and the difference between the convolution kernels to obtain an adjusted network structure.

[0038] Optionally, the using the respiratory sound simulation signal data to train the initial respiratory sound analysis model to obtain a standard respiratory sound analysis model includes:

[0039] Dividing the respiratory sound simulation signal data into a training set, a test set, and a validation set to obtain a respiratory sound training set, a respiratory sound test set, and a respiratory sound validation set, wherein the ratio of the respiratory sound training set, the respiratory sound test set, and the respiratory sound validation set is set to 8:1:1;

[0040] Optimizing the parameters of the initial respiratory sound analysis model using a preset Adam optimizer to obtain an optimized respiratory sound analysis model, wherein the initial learning rate of the Adam optimizer is 0.001, the batch size is 64, and the total number of iterations is 100;

[0041] Using the respiratory sound training set to train the optimized respiratory sound analysis model to obtain a preliminary trained respiratory sound analysis model;

[0042] The preliminary trained respiratory sound analysis model is tested using the respiratory sound test set to obtain a test respiratory sound analysis model, and the test respiratory sound analysis model is verified using the respiratory sound verification set. When the verification passes, a standard respiratory sound analysis model is obtained.

[0043] To achieve the above objectives, the present invention further provides a central airway stenosis assessment system based on deep learning, comprising:

[0044] a data acquisition module, configured to acquire body surface sounds of the patient using a preset sound acquisition device to obtain a comprehensive body surface sound, and extract respiratory sounds from the comprehensive body surface sound to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sound is collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus;

[0045] a data processing module, configured to obtain a frequency band range of the respiratory sound analog signal data, select an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, perform windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data, and construct a respiratory sound feature matrix based on the windowed respiratory sound analog signal data;

[0046] a network adjustment module, configured to obtain respiratory sound pathological characteristics based on the respiratory sound feature matrix, perform an airway examination on the patient using a preset examination device to obtain the degree of airway stenosis and airway stenosis zones, and adjust a preset basic network based on the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree, and airway stenosis zones to obtain an adjusted network structure;

[0047] a model application module, configured to construct an initial respiratory sound analysis model according to the adjusted network structure, and train the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model;

[0048] Acquire the respiratory sound simulation signal data of the patient to be diagnosed, obtain the respiratory sound simulation signal data to be diagnosed, and use the standard respiratory sound analysis model to diagnose the respiratory sound simulation signal data to obtain the diagnosis result of the patient to be diagnosed.

[0049] In order to solve the above problem, the present invention further provides an electronic device, comprising:

[0050] A memory stores at least one instruction.

[0051] In order to solve the above problems, the present invention also provides a computer-readable storage medium, which stores at least one instruction to implement the above-mentioned method for assessing the degree of central airway stenosis based on deep learning.

[0052] The present invention is to solve the problem described in the background technology. The present invention obtains the sound of the patient's body surface based on a preset sound collection device to obtain a comprehensive body surface sound, and extracts the respiratory sound in the comprehensive body surface sound to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sound is collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus; obtains the frequency band range of the respiratory sound simulation signal data, selects a filter based on the frequency band range to denoise the respiratory sound simulation signal data, obtains denoised respiratory sound simulation signal data, and performs windowing processing on the denoised respiratory sound simulation signal data to obtain windowed respiratory sound simulation signal data, and constructs a respiratory sound feature matrix according to the windowed respiratory sound simulation signal data; based on the respiratory sound feature matrix Acquire pathological characteristics of respiratory sounds, perform an airway examination on the patient using a preset examination device, obtain the degree of airway stenosis and airway stenosis zones, and adjust a preset basic network based on the simulated respiratory sound signal data, respiratory sound pathological characteristics, degree of airway stenosis, and airway stenosis zones to obtain an adjusted network structure; construct an initial respiratory sound analysis model based on the adjusted network structure, and train the initial respiratory sound analysis model using the simulated respiratory sound signal data to obtain a standard respiratory sound analysis model; acquire simulated respiratory sound signal data of the patient to be diagnosed to obtain simulated respiratory sound signal data to be diagnosed, and diagnose the simulated respiratory sound signal data to be diagnosed using the standard respiratory sound analysis model to obtain a diagnosis result for the patient's condition. The present invention utilizes multi-site sound acquisition, denoising processing, feature extraction and analysis, and multi-source data fusion, combined with a deep learning model, to achieve accurate and efficient diagnosis of central airway stenosis, reduce the misdiagnosis rate, and provide patients with more reliable diagnostic results and personalized treatment support. Therefore, the present invention can improve the reliability of diagnostic results. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A flowchart of a method for assessing the degree of central airway stenosis based on deep learning provided by one embodiment of the present invention;

[0054] Figure 2 This is a functional module diagram of a central airway stenosis assessment system based on deep learning provided by one embodiment of the present invention;

[0055] Figure 3 A schematic structural diagram of an electronic device for implementing the deep learning-based central airway stenosis assessment method provided in one embodiment of the present invention.

[0056] Description of reference numerals:

[0057] 1. Electronic device; 10. Processor; 11. Memory; 12. Bus.

[0058] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0059] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0060] The present invention provides a method for assessing the degree of central airway stenosis based on deep learning. The method can be performed by, but is not limited to, at least one of electronic devices such as a server and a terminal that can be configured to execute the method provided by the present invention.

[0061] Reference Figure 1 FIG2 is a flow chart of a method for assessing the degree of central airway stenosis based on deep learning according to an embodiment of the present invention. In this embodiment, the method for assessing the degree of central airway stenosis based on deep learning includes:

[0062] S1. Acquire body surface sounds of the patient based on a preset sound collection device to obtain a comprehensive body surface sound, and extract respiratory sounds from the comprehensive body surface sound to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sound is collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus.

[0063] It is understandable that based on the preset sound collection equipment, the respiratory sounds of the patient's body surface are obtained from key parts of the patient's body such as the suprasternal fossa, left main bronchus, right main bronchus, etc., thereby obtaining respiratory sound simulation signal data, which can provide comprehensive and accurate original data for subsequent respiratory sound analysis, ensuring the integrity and representativeness of the respiratory sound signal.

[0064] The left main bronchus includes the front side of the left main bronchus and the back side of the left main bronchus, and the right main bronchus includes the front side of the right main bronchus and the back side of the right main bronchus.

[0065] In an embodiment of the present invention, the sound collection device is a digital stethoscope, which converts and processes respiratory sound signals into analog sound signals that can continuously and wirelessly monitor the patient's body through various sensors (such as capacitive microphones and voltage sensors). It can also transmit high-precision audio signals through the Bluetooth protocol, and the detectable sound frequency range is 20 to 2000 Hz.

[0066] Comprehensive body surface sounds refer to a collection of sound signals acquired from multiple specific locations on the patient's body (such as the suprasternal fossa, left mainstem bronchus, and right mainstem bronchus) using a pre-set sound acquisition device. These sound signals contain raw sound information from various parts of the patient's body, and may include breathing sounds, heartbeats, and other body surface sounds. By collecting and analyzing these comprehensive body surface sounds, useful breathing sound simulation signal data can be extracted for subsequent diagnosis and analysis.

[0067] Breath sounds are the sounds produced by air inhaled through the mouth and nose as it vibrates or strikes the airway walls as it passes through the lungs and alveoli. These sounds are transmitted to the body surface through lung tissue and the chest wall, effectively reflecting the condition of the airways and lungs. Breath sounds can be categorized as normal and abnormal. Normal breath sounds include alveolar, bronchial, and bronchoalveolar sounds, and their acoustic characteristics are related to the location and transmission method of the breath sounds.

[0068] In an embodiment of the present invention, auscultation at the suprasternal fossa can clearly capture the breath sounds of the trachea and main bronchi. If there is an abnormality, such as tracheal inflammation or obstruction, abnormal breath sounds, such as wheezing or moist rales, may be heard in this area; auscultation at the anterior left main bronchus can evaluate the breath sounds of the left upper and lower lobes of the lung. If there are lesions in the left upper and lower lobes of the lung, such as pneumonia or a tumor, abnormal breath sounds, such as moist rales or bronchial breath sounds, may be heard in this area; auscultation at the anterior right main bronchus can evaluate the breath sounds of the right upper and lower lobes of the lung. If there are lesions in the right upper and lower lobes of the lung, such as pneumonia or a tumor, abnormal breath sounds may be heard in this area.

[0069] Furthermore, by auscultating at the three key locations mentioned above, doctors can conduct a comprehensive assessment of the patient's respiratory system, including different types of breath sounds (such as bronchial breath sounds, alveolar breath sounds, and moist rales) and their abnormal changes. For example, abnormal bronchial breath sounds may indicate airway inflammation or obstruction, while moist rales may indicate lung infection or fluid accumulation. By comprehensively analyzing this information, doctors can more accurately diagnose and treat respiratory diseases.

[0070] In an embodiment of the present invention, extracting the respiratory sound from the comprehensive body surface sound to obtain the respiratory sound analog signal data includes:

[0071] Segmenting the comprehensive body surface sound based on a preset time period to obtain comprehensive sound samples;

[0072] Extracting sound features of the respiratory sound from the comprehensive sound sample to obtain a respiratory sound feature, and converting the respiratory sound feature into a vector to obtain a respiratory sound feature vector;

[0073] Constructing a projection matrix based on the respiratory sound feature vector to obtain a respiratory sound projection matrix;

[0074] The respiratory sound projection matrix is ​​used to separate the heartbeat sound and the respiratory sound in the comprehensive body surface sound to obtain a respiratory sound vector, and signal restoration processing is performed based on the respiratory sound vector to obtain respiratory sound simulation signal data, wherein the signal restoration processing includes inverse discrete cosine transform, feature matching and denoising optimization.

[0075] In an embodiment of the present invention, the time period may be set to 15 seconds, thereby obtaining a plurality of respiratory sound samples with a duration of 15 seconds, so as to ensure consistency in subsequent analysis.

[0076] Among them, the respiratory sound vector refers to the vector representation of sound features related to respiratory sounds separated from the comprehensive body surface sounds. The respiratory sound vector retains the key features of respiratory sounds and can be used for subsequent signal restoration processing, including inverse discrete cosine transform, feature matching and denoising optimization, to obtain purer and clearer respiratory sound simulation signal data.

[0077] Furthermore, extracting the sound features of the respiratory sound in the comprehensive sound sample to obtain a respiratory sound feature, and converting the respiratory sound feature into a vector to obtain a respiratory sound feature vector includes:

[0078] Sorting the integrated sound samples according to time sequence to obtain an integrated sound sample sequence;

[0079] Obtaining a time-frequency signal of the integrated sound sample sequence, calculating a covariance matrix based on the time-frequency signal of the integrated sound sample sequence to obtain a respiratory sound covariance matrix, and extracting off-diagonal element energy of the respiratory sound covariance matrix using a pre-built joint approximate diagonalization algorithm to obtain independent signal components;

[0080] The main diagonal elements in the respiratory sound covariance matrix are extracted based on the independent signal components, and the main diagonal elements are used as feature vectors to construct the respiratory sound feature vector.

[0081] In the embodiment of the present invention, when the integrated sound samples are sorted according to the time sequence to obtain the integrated sound sample sequence, the integrated sound samples can be obtained by sorting them in the time sequence.

[0082] Among them, the time-frequency signal of the integrated sound sample sequence refers to the signal representation obtained after converting the separated respiratory sound samples from the time domain to the frequency domain, which usually includes the frequency components of the signal and its changes over time. The time-frequency signal can more comprehensively describe the characteristics of respiratory sounds, including frequency, energy distribution, duration, etc., which is helpful for subsequent feature extraction and analysis; the off-diagonal element energy refers to the value of the off-diagonal elements in the covariance matrix. These elements reflect the covariance or correlation between different features. By extracting and minimizing the off-diagonal element energy, the independence of features can be achieved, the redundancy and correlation between features can be reduced, the main features can be highlighted, and the feature discrimination and model performance can be improved.

[0083] It is understandable that among the comprehensive body surface sounds, heart sounds and lung sounds often overlap with each other, especially in the signals collected in the heart valve auscultation area, where heart sounds and respiratory sounds seriously overlap in the time domain and frequency domain. If they are not separated, it will interfere with the accurate extraction of respiratory sounds, increase the complexity of signal analysis, and affect the respiratory rate, respiratory intensity and accurate judgment, which is not conducive to the accurate assessment and diagnosis of respiratory function. After separation, the analysis accuracy can be improved to better meet the needs of medical diagnosis and monitoring.

[0084] The joint approximate diagonalization algorithm is a matrix processing technique whose goal is to find a transformation matrix that simultaneously approximates the diagonal form of multiple covariance matrices under this transformation matrix. Specifically, the algorithm seeks a projection matrix such that, in the projected space, the off-diagonal elements (i.e., the correlations between different sampling points) of the breath sound covariance matrix are minimized while the diagonal elements (i.e., the variance of each sampling point) are preserved. This allows the extraction of stable features of the breath sound signal across different periods, as these features are similar across periods and are retained and highlighted during the joint diagonalization process.

[0085] S2. Obtain a frequency band range of the respiratory sound analog signal data, select a filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, perform windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data, and construct a respiratory sound feature matrix based on the windowed respiratory sound analog signal data.

[0086] It is understandable that by performing a series of processing on the respiratory sound simulation signal data, including denoising, windowing, and feature extraction, a valuable respiratory sound feature matrix for subsequent analysis can be obtained, laying the foundation for subsequent accurate diagnosis of the patient's condition.

[0087] In the embodiment of the present invention, the frequency band range refers to the distribution interval of the respiratory sound analog signal data in the frequency domain, covering different frequency components contained in the respiratory sound signal, and is used to determine the existence range of each frequency component in the respiratory sound signal.

[0088] Furthermore, windowed respiratory sound analog signal data refers to data obtained by applying a window function to the denoised respiratory sound analog signal data. Specifically, windowing involves truncating or weighting the signal in the time domain to reduce spectral leakage and improve the accuracy of frequency domain analysis. By applying a window function at the start and end points of the signal, the signal smoothly transitions to zero, thereby reducing artifacts and errors in spectral analysis. The resulting windowed respiratory sound analog signal data is then used for subsequent feature extraction and analysis, such as constructing a respiratory sound feature matrix to better capture the signal's frequency and temporal characteristics.

[0089] In an embodiment of the present invention, obtaining a frequency band range of the respiratory sound analog signal data, and selecting an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, includes:

[0090] Acquiring a frequency band range of the respiratory sound analog signal data, and selecting a filter corresponding to the filtering range according to the frequency band range to obtain a screening filter;

[0091] The cutoff frequency of the screening filter is set according to the frequency band range to obtain a target filter, and the target filter is used to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data.

[0092] In embodiments of the present invention, by setting a filter with an appropriate cutoff frequency, it is possible to retain the target signal frequency and remove noise frequencies. For example, since breathing sounds primarily occur between 100 and 2000 Hz, the cutoff frequency can be set outside this range, and a bandpass filter can be used to retain the target frequency signal and achieve denoising. An appropriate cutoff frequency can effectively remove noise while retaining the useful signal. A cutoff frequency that is too high or too low may result in loss of useful signal or inadequate noise removal.

[0093] In another embodiment of the present invention, other parameters of the screening filter may be set to obtain a target filter, such as passband ripple, transition bandwidth, filter order, etc.

[0094] In an embodiment of the present invention, a filter is selected based on the frequency band range to denoise the respiratory sound analog signal data. After obtaining the denoised respiratory sound analog signal data, the denoising effect can be evaluated by comparing the signals before and after filtering. Specifically, indicators such as signal-to-noise ratio (SNR) and root mean square error (RMSE) can be used to quantify the denoising effect. If the denoising effect is not ideal, the parameters of the filter can be adjusted or other types of filters can be tried, and the filtering and evaluation process can be repeated until a satisfactory denoising effect is achieved. Other types of filters include low-pass filters, band-pass filters, and band-stop filters.

[0095] Furthermore, performing windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data includes:

[0096] Dividing the denoised respiratory sound analog signal data based on a pre-built sliding window method to obtain a plurality of short-time window respiratory sound analog signal data, wherein the sliding window corresponding to the sliding window method has a window length of 25 milliseconds and an overlap rate of 50%±5%;

[0097] Obtaining a window function, performing windowing processing on the multiple short-time windowed respiratory sound analog signal data using the window function to obtain multiple short-time windowed respiratory sounds, and splicing the multiple short-time windowed respiratory sounds in chronological order to obtain windowed respiratory sound analog signal data;

[0098] The short-term windowed breath sound is expressed as:

[0099] S windowed (n) = S filtered (n)·w(n)

[0100] Among them, S filtered (n) is the sampling value of the n-th short-time window respiratory sound analog signal data, w(n) is the value of the window function corresponding to the n-th short-time window respiratory sound analog signal data, S windowed (n) is the short-time windowed respiratory sound corresponding to the n-th short-time window respiratory sound analog signal data.

[0101] The sliding window method is a signal processing technique used to segment long signals into multiple shorter segments for short-term analysis. The sliding window method uses a fixed-length window that slides across the signal, each step size being a fixed amount, to segment the signal into multiple short-term windows. The length and step size of the sliding window can be adjusted based on analysis needs.

[0102] Among them, the window function is a mathematical function used to intercept or weight the signal in signal processing to reduce spectral leakage and improve the accuracy of frequency domain analysis.

[0103] Furthermore, constructing a respiratory sound feature matrix based on the windowed respiratory sound simulation signal data includes:

[0104] The windowed respiratory sound analog signal data is converted into cepstral coefficients using a pre-constructed discrete cosine transform method, and a matrix is ​​constructed according to the cepstral coefficients to obtain a respiratory sound feature matrix.

[0105] The discrete cosine transform (DCT) is a mathematical transformation method used to convert signals from the time domain to the frequency domain. It is particularly suitable for signal feature extraction and data compression. In breath sound feature extraction, the DCT is used to help extract the main features of the signal and improve the robustness of the features.

[0106] In an embodiment of the present invention, a preset set of window functions is used to intercept the windowed respiratory audio spectrum information to obtain multiple spectrum signal energies. Each filter corresponds to a window function, and the signal energy corresponds to the signal energy (amplitude spectrum square) within the frequency range covered by the window function.

[0107] S3. Acquire respiratory sound pathological characteristics based on the respiratory sound feature matrix, perform airway examination on the patient using a preset examination device to obtain the degree of airway stenosis and airway stenosis zones, and adjust the preset basic network according to the respiratory sound simulation signal data, respiratory sound pathological characteristics, degree of airway stenosis and airway stenosis zones to obtain an adjusted network structure.

[0108] It's understandable that by combining the pathological characteristics of respiratory sounds in the respiratory sound feature matrix with the degree and zoning of airway stenosis derived from the examination equipment, targeted adjustments are made to the basic network, thereby constructing a more accurate network structure that better meets the needs of actual disease diagnosis. This enables subsequent models based on this adjusted network structure to more accurately assess the degree of airway stenosis in the patient's respiratory system.

[0109] Pathological respiratory sound features refer to features extracted from the respiratory sound feature matrix that reflect the pathological state of the respiratory system. These features may include abnormal frequency components, intensity changes, duration, and characteristic differences between different respiratory phases. They can be used by physicians to determine the presence, type, and severity of respiratory diseases.

[0110] Among them, airway stenosis zoning refers to dividing the patient's airway structure into multiple specific areas to accurately locate the stenosis. This zoning method covers the upper, middle, and lower segments of the trachea, as well as the origin and distal ends of the left and right main bronchi. It helps to quickly determine the location of the lesion during diagnosis and treatment, and provides a basis for formulating personalized treatment plans. For example, stenosis of the upper main airway means that the patient may experience symptoms such as hoarseness and inspiratory dyspnea; stenosis of the origin of the left main bronchus means that the patient may experience weakened or absent breath sounds in the left lung, accompanied by a persistent dry cough.

[0111] Inspection equipment refers to medical devices used to examine a patient's airway. Common inspection equipment includes bronchoscopes, CT scanners, and X-ray machines. These devices can help doctors examine the internal structure and condition of the airway and determine the extent and location of airway stenosis.

[0112] Among them, the basic network is a basic network based on VGGNet, specifically based on the VGG-16 model, and the network structure can be adjusted according to factors such as respiratory sounds and the degree of airway stenosis, airway stenosis division, and pathological characteristics of respiratory sounds.

[0113] In an embodiment of the present invention, obtaining respiratory sound pathological features based on the respiratory sound feature matrix includes:

[0114] Using a pre-built correlation analysis method, a feature having a correlation with a preset pathological feature greater than a preset threshold is obtained from the respiratory sound feature matrix to obtain a pathology-related feature, and a pathology-related feature matrix is ​​constructed based on the pathology-related feature;

[0115] The pathology-related feature matrix is ​​subjected to dimensionality reduction processing using a pre-constructed principal component analysis method to obtain a reduced-dimensional respiratory sound feature matrix, and features in the reduced-dimensional respiratory sound matrix are extracted to obtain respiratory sound pathology features.

[0116] Correlation analysis is a statistical method used to evaluate the strength and direction of the linear relationship between two variables. In feature selection, correlation analysis can be used to identify features that are highly correlated with target pathological features (such as the degree and location of airway stenosis), thereby screening the most valuable feature subsets (pathology-related features) for diagnosis and improving the accuracy and efficiency of adjusting the network structure.

[0117] Principal component analysis (PCA) is used to reduce the dimensionality of the normalized respiratory sound feature matrix. Through linear transformation, pathology-related features are combined into new, mutually orthogonal principal components, removing redundant information and noise while highlighting key features. This improves the model's computational efficiency and its ability to identify respiratory sound pathology features.

[0118] Among them, the pathological characteristics of respiratory sounds include energy characteristics, time domain characteristics, spectrum characteristics and spatial characteristics. The spatial characteristics include the characteristic differences of different auscultation areas.

[0119] In an embodiment of the present invention, the preset basic network is adjusted according to the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree and airway stenosis partition to obtain an adjusted network structure, including:

[0120] Performing convolution on the basic network using a convolution layer with a preset accuracy based on the respiratory sound simulation signal data and the respiratory sound pathological characteristics, thereby obtaining parameters of one or two convolution layers in the basic network;

[0121] performing convolution in the basic network using a 4×4 convolution kernel according to the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree, and airway stenosis partition, and comparing the convolution accuracy of the convolution kernel with a preset network diagnostic target to obtain a convolution kernel comparison difference;

[0122] The basic network is adjusted according to the parameters of the first and second convolution layers in the basic network and the difference between the convolution kernels to obtain an adjusted network structure.

[0123] Among them, the convolution accuracy of the convolution kernel refers to the degree of fitting of the basic network after using a convolution kernel of a specific size for convolution operation, which is usually expressed by the loss function value or accuracy, reflecting the model's processing effect on the data under the current convolution kernel size; the convolution kernel contrast difference refers to the difference between the convolution kernel convolution accuracy and the network diagnosis target, such as the gap between the current accuracy and the target accuracy. The size of the convolution kernel is adjusted according to this difference.

[0124] S4. Constructing an initial respiratory sound analysis model according to the adjusted network structure, and training the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model.

[0125] It's clear that by using the adjusted network structure for model building and training it with simulated respiratory sound data, an accurate and reliable trained model can be obtained for subsequent diagnosis. This process ensures that the model learns effective features and patterns from the data and demonstrates good performance on unseen data, ultimately improving diagnostic accuracy and reliability.

[0126] The initial breath sound analysis model is a machine learning or deep learning model built based on an adjusted network structure, used to diagnose and analyze simulated breath sound signal data. By learning the relationship between information such as the breath sound feature matrix and the degree of airway stenosis and the condition, the model can analyze new simulated breath sound signal data to be diagnosed, outputting an airway stenosis assessment result, and assisting doctors in making more accurate medical diagnoses.

[0127] Furthermore, the using the respiratory sound simulation signal data to train the initial respiratory sound analysis model to obtain a standard respiratory sound analysis model includes:

[0128] Dividing the respiratory sound simulation signal data into a training set, a test set, and a validation set to obtain a respiratory sound training set, a respiratory sound test set, and a respiratory sound validation set, wherein the ratio of the respiratory sound training set, the respiratory sound test set, and the respiratory sound validation set is set to 8:1:1;

[0129] Optimizing the parameters of the initial respiratory sound analysis model using a preset Adam optimizer to obtain an optimized respiratory sound analysis model, wherein the initial learning rate of the Adam optimizer is 0.001, the batch size is 64, and the total number of iterations is 100;

[0130] Using the respiratory sound training set to train the optimized respiratory sound analysis model to obtain a preliminary trained respiratory sound analysis model;

[0131] The preliminary trained respiratory sound analysis model is tested using the respiratory sound test set to obtain a test respiratory sound analysis model, and the test respiratory sound analysis model is verified using the respiratory sound verification set. When the verification passes, a standard respiratory sound analysis model is obtained.

[0132] Among them, using the preset Adam optimizer to optimize the parameters of the initial respiratory sound analysis model means that during the training process of the initial respiratory sound analysis model, the Adam optimization algorithm is used to automatically adjust the model parameters to minimize the difference between the predicted value and the true value, thereby improving the performance of the optimized respiratory sound analysis model.

[0133] Among them, when the test respiratory sound analysis model passes the verification, it means that when the performance of the model on the verification set meets the preset performance standards (for example, the loss value is lower than a certain threshold or the accuracy is higher than a certain threshold), the test respiratory sound model is considered to have successfully passed the verification and can be used as a standard model for actual respiratory sound analysis tasks to ensure that it has good generalization ability and stability.

[0134] In another embodiment of the present invention, when the optimized respiratory sound analysis model is trained using the respiratory sound training set, the training loss of each iteration is recorded. If the training loss does not improve during several consecutive iterations, the training is stopped to avoid overfitting.

[0135] S5. Acquire the respiratory sound simulation signal data of the patient to be diagnosed, obtain the respiratory sound simulation signal data to be diagnosed, and use the standard respiratory sound analysis model to diagnose the respiratory sound simulation signal data to be diagnosed, to obtain the diagnosis result of the patient's condition to be diagnosed.

[0136] It is understandable that the analysis of newly collected respiratory sound simulation signal data of patients to be diagnosed has achieved accurate assessment of central airway stenosis, improved diagnostic efficiency and accuracy, reduced misdiagnosis rate and medical costs, and at the same time reduced the risk of complications caused by frequent examinations for patients, providing an objective basis for early diagnosis and intervention, improving patient prognosis, and promoting the application and development of artificial intelligence technology in the medical field.

[0137] Among them, the respiratory sound simulation signal data to be diagnosed refers to the respiratory sound simulation signal data collected from the patient to be diagnosed. This data contains the patient's respiratory sound information and is analyzed and evaluated through the trained model to determine the degree of airway stenosis of the patient.

[0138] Among them, the standard respiratory sound analysis model refers to a machine learning model or deep learning model that has been trained and optimized. During the training process, the model learns the relationship between a large amount of respiratory sound simulation signal data and the disease condition, and can perform diagnostic analysis on new respiratory sound simulation signal data.

[0139] Furthermore, the standard respiratory sound analysis model has an adjusted VGGNet, which is more suitable for analyzing the simulated respiratory sound signal data to be diagnosed, extracting key features and assessing the degree of airway stenosis.

[0140] Furthermore, the disease diagnosis results include normal airway, mild airway stenosis, moderate airway stenosis, severe airway stenosis, airway stenosis location and cause of airway stenosis, etc.

[0141] The present invention is to solve the problem described in the background technology. The present invention obtains the sound of the patient's body surface based on a preset sound collection device to obtain a comprehensive body surface sound, and extracts the respiratory sound in the comprehensive body surface sound to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sound is collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus; obtains the frequency band range of the respiratory sound simulation signal data, selects a filter based on the frequency band range to denoise the respiratory sound simulation signal data, obtains denoised respiratory sound simulation signal data, and performs windowing processing on the denoised respiratory sound simulation signal data to obtain windowed respiratory sound simulation signal data, and constructs a respiratory sound feature matrix according to the windowed respiratory sound simulation signal data; based on the respiratory sound feature matrix Acquire pathological characteristics of respiratory sounds, perform an airway examination on the patient using a preset examination device, obtain the degree of airway stenosis and airway stenosis zones, and adjust a preset basic network based on the simulated respiratory sound signal data, respiratory sound pathological characteristics, degree of airway stenosis, and airway stenosis zones to obtain an adjusted network structure; construct an initial respiratory sound analysis model based on the adjusted network structure, and train the initial respiratory sound analysis model using the simulated respiratory sound signal data to obtain a standard respiratory sound analysis model; acquire simulated respiratory sound signal data of the patient to be diagnosed to obtain simulated respiratory sound signal data to be diagnosed, and diagnose the simulated respiratory sound signal data to be diagnosed using the standard respiratory sound analysis model to obtain a diagnosis result for the patient's condition. The present invention utilizes multi-site sound acquisition, denoising processing, feature extraction and analysis, and multi-source data fusion, combined with a deep learning model, to achieve accurate and efficient diagnosis of central airway stenosis, reduce the misdiagnosis rate, and provide patients with more reliable diagnostic results and personalized treatment support. Therefore, the present invention can improve the reliability of diagnostic results.

[0142] like Figure 2 , which is a functional module diagram of a central airway stenosis assessment system based on deep learning provided by one embodiment of the present invention.

[0143] The deep learning-based central airway stenosis assessment system 100 described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the deep learning-based central airway stenosis assessment system 100 may include a data acquisition module 101, a data processing module 102, a network adjustment module 103, and a model application module 104. A module, also referred to as a unit, is a series of computer program segments that can be executed by an electronic device processor and perform a fixed function. These modules are stored in the electronic device's memory.

[0144] The data acquisition module 101 is configured to acquire body surface sounds of the patient using a preset sound acquisition device to obtain comprehensive body surface sounds, and extract respiratory sounds from the comprehensive body surface sounds to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sounds are collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus;

[0145] The data processing module 102 is configured to obtain a frequency band range of the respiratory sound analog signal data, select a filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, perform windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data, and construct a respiratory sound feature matrix based on the windowed respiratory sound analog signal data;

[0146] The network adjustment module 103 is configured to obtain respiratory sound pathological characteristics based on the respiratory sound feature matrix, perform an airway examination on the patient using a preset examination device to obtain the degree of airway stenosis and airway stenosis zones, and adjust the preset basic network based on the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree, and airway stenosis zones to obtain an adjusted network structure;

[0147] The model application module 104 is configured to construct an initial respiratory sound analysis model according to the adjusted network structure, and train the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model;

[0148] Acquire the respiratory sound simulation signal data of the patient to be diagnosed, obtain the respiratory sound simulation signal data to be diagnosed, and use the standard respiratory sound analysis model to diagnose the respiratory sound simulation signal data to obtain the diagnosis result of the patient to be diagnosed.

[0149] In detail, each module in the central airway stenosis assessment system 100 based on deep learning in the embodiment of the present invention adopts the same method as above when used. Figure 1 The technical means are the same as the deep learning-based central airway stenosis assessment method described in , and can produce the same technical effects, so they will not be repeated here.

[0150] like Figure 3 , which is a structural diagram of an electronic device for implementing a method for assessing the degree of central airway stenosis based on deep learning provided by one embodiment of the present invention.

[0151] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a method program for assessing the degree of central airway stenosis based on deep learning.

[0152] The memory 11 includes at least one type of readable storage medium, including flash memory, mobile hard disk, multimedia card, card-type memory (e.g., SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 1. Furthermore, the memory 11 also includes an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed on the electronic device 1, such as the code of the central airway stenosis assessment method program based on deep learning, but can also be used to temporarily store data that has been output or is to be output.

[0153] In some embodiments, the processor 10 may be comprised of an integrated circuit, such as a single packaged integrated circuit or multiple packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting the various components of the electronic device using various interfaces and circuits. It executes programs or modules stored in the memory 11 (e.g., a deep learning-based central airway stenosis assessment method program) and accesses data stored in the memory 11 to perform various functions and process data.

[0154] The bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus 12 may be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to enable communication between the memory 11 and at least one processor 10.

[0155] Figure 3 Only the electronic device with components is shown, and it can be understood by those skilled in the art that Figure 3The structure shown does not constitute a limitation on the electronic device 1 , and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0156] Furthermore, the electronic device 1 may further include a network interface. Optionally, the network interface may include a wired interface, which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0157] Optionally, the electronic device 1 may further include a user interface, which may be a display or an input unit (such as a keyboard). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light Emitting Diode) touchscreen. The display may also be appropriately referred to as a display screen or a display unit, and is used to display information processed in the electronic device 1 and to display a visual user interface.

[0158] The central airway stenosis assessment method program based on deep learning stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve the following:

[0159] Acquiring body surface sounds of the patient based on a preset sound collection device to obtain comprehensive body surface sounds, and extracting respiratory sounds from the comprehensive body surface sounds to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sounds are collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus;

[0160] Obtaining a frequency band range of the respiratory sound analog signal data, selecting an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, performing windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data, and constructing a respiratory sound feature matrix based on the windowed respiratory sound analog signal data;

[0161] Acquiring respiratory sound pathological characteristics based on the respiratory sound feature matrix, performing an airway examination on the patient using a preset examination device to obtain the degree of airway stenosis and airway stenosis zones, and adjusting a preset basic network based on the respiratory sound simulation signal data, the respiratory sound pathological characteristics, the degree of airway stenosis and the airway stenosis zones to obtain an adjusted network structure;

[0162] constructing an initial respiratory sound analysis model according to the adjusted network structure, and training the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model;

[0163] Acquire the respiratory sound simulation signal data of the patient to be diagnosed, obtain the respiratory sound simulation signal data to be diagnosed, and use the standard respiratory sound analysis model to diagnose the respiratory sound simulation signal data to obtain the diagnosis result of the patient to be diagnosed.

[0164] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0165] Furthermore, if the modules / units integrated into the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0166] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. When the computer program is executed by a processor of an electronic device, the computer program can implement:

[0167] Acquiring body surface sounds of the patient based on a preset sound collection device to obtain comprehensive body surface sounds, and extracting respiratory sounds from the comprehensive body surface sounds to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sounds are collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus;

[0168] Obtaining a frequency band range of the respiratory sound analog signal data, selecting an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, performing windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data, and constructing a respiratory sound feature matrix based on the windowed respiratory sound analog signal data;

[0169] Acquiring respiratory sound pathological characteristics based on the respiratory sound feature matrix, performing an airway examination on the patient using a preset examination device to obtain the degree of airway stenosis and airway stenosis zones, and adjusting a preset basic network based on the respiratory sound simulation signal data, the respiratory sound pathological characteristics, the degree of airway stenosis and the airway stenosis zones to obtain an adjusted network structure;

[0170] constructing an initial respiratory sound analysis model according to the adjusted network structure, and training the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model;

[0171] Acquire the respiratory sound simulation signal data of the patient to be diagnosed, obtain the respiratory sound simulation signal data to be diagnosed, and use the standard respiratory sound analysis model to diagnose the respiratory sound simulation signal data to obtain the diagnosis result of the patient to be diagnosed.

[0172] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only exemplary, and actual implementations may have other division methods.

[0173] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0174] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0175] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for assessing the degree of central airway stenosis based on deep learning, characterized in that: The method comprises: Acquiring body surface sounds of the patient based on a preset sound collection device to obtain comprehensive body surface sounds, and extracting respiratory sounds from the comprehensive body surface sounds to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sounds are collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus; Obtaining a frequency band range of the respiratory sound analog signal data, selecting an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, performing windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data, and constructing a respiratory sound feature matrix based on the windowed respiratory sound analog signal data; Acquiring respiratory sound pathological characteristics based on the respiratory sound feature matrix, performing an airway examination on the patient using a preset examination device to obtain the degree of airway stenosis and airway stenosis zones, and adjusting a preset basic network based on the respiratory sound simulation signal data, the respiratory sound pathological characteristics, the degree of airway stenosis and the airway stenosis zones to obtain an adjusted network structure; constructing an initial respiratory sound analysis model according to the adjusted network structure, and training the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model; Acquire the respiratory sound simulation signal data of the patient to be diagnosed, obtain the respiratory sound simulation signal data to be diagnosed, and use the standard respiratory sound analysis model to diagnose the respiratory sound simulation signal data to obtain the diagnosis result of the patient to be diagnosed.

2. The method for assessing the degree of central airway stenosis based on deep learning according to claim 1, wherein: The step of extracting the respiratory sound from the comprehensive body surface sound to obtain respiratory sound simulation signal data includes: Segmenting the comprehensive body surface sound based on a preset time period to obtain comprehensive sound samples; Extracting sound features of the respiratory sound from the comprehensive sound sample to obtain a respiratory sound feature, and converting the respiratory sound feature into a vector to obtain a respiratory sound feature vector; Constructing a projection matrix based on the respiratory sound feature vector to obtain a respiratory sound projection matrix; The respiratory sound projection matrix is ​​used to separate the heartbeat sound and the respiratory sound in the comprehensive body surface sound to obtain a respiratory sound vector, and signal restoration processing is performed based on the respiratory sound vector to obtain respiratory sound simulation signal data, wherein the signal restoration processing includes inverse discrete cosine transform, feature matching and denoising optimization.

3. The method for assessing the degree of central airway stenosis based on deep learning according to claim 2, wherein: The extracting the sound features of the respiratory sound from the comprehensive sound sample to obtain a respiratory sound feature, and converting the respiratory sound feature into a vector to obtain a respiratory sound feature vector, includes: Sorting the integrated sound samples according to time sequence to obtain an integrated sound sample sequence; Obtaining a time-frequency signal of the integrated sound sample sequence, calculating a covariance matrix based on the time-frequency signal of the integrated sound sample sequence to obtain a respiratory sound covariance matrix, and extracting off-diagonal element energy of the respiratory sound covariance matrix using a pre-built joint approximate diagonalization algorithm to obtain independent signal components; The main diagonal elements in the respiratory sound covariance matrix are extracted based on the independent signal components, and the main diagonal elements are used as feature vectors to construct the respiratory sound feature vector.

4. The method for assessing the degree of central airway stenosis based on deep learning according to claim 3, wherein: The acquiring of the frequency band range of the respiratory sound analog signal data, and selecting an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, includes: Acquiring a frequency band range of the respiratory sound analog signal data, and selecting a filter corresponding to the filtering range according to the frequency band range to obtain a screening filter; The cutoff frequency of the screening filter is set according to the frequency band range to obtain a target filter, and the target filter is used to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data.

5. The method for assessing the degree of central airway stenosis based on deep learning according to claim 4, wherein: The performing windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data includes: Dividing the denoised respiratory sound analog signal data based on a pre-built sliding window method to obtain a plurality of short-time window respiratory sound analog signal data, wherein the sliding window corresponding to the sliding window method has a window length of 25 milliseconds and an overlap rate of 50%±5%; Obtaining a window function, performing windowing processing on the multiple short-time windowed respiratory sound analog signal data using the window function to obtain multiple short-time windowed respiratory sounds, and splicing the multiple short-time windowed respiratory sounds in chronological order to obtain windowed respiratory sound analog signal data; The short-term windowed breath sound is expressed as: S windowed (n)=S filtered (n)·w(n) Among them, S filtered (n) is the sampling value of the n-th short-time window respiratory sound analog signal data, w(n) is the value of the window function corresponding to the n-th short-time window respiratory sound analog signal data, S windowed (n) is the short-time windowed respiratory sound corresponding to the n-th short-time window respiratory sound analog signal data.

6. The method for assessing the degree of central airway stenosis based on deep learning according to claim 5, wherein: The step of constructing a respiratory sound feature matrix according to the windowed respiratory sound simulation signal data comprises: The windowed respiratory sound analog signal data is converted into cepstral coefficients using a pre-constructed discrete cosine transform method, and a matrix is ​​constructed according to the cepstral coefficients to obtain a respiratory sound feature matrix.

7. The method for assessing the degree of central airway stenosis based on deep learning according to claim 6, wherein: The acquiring of respiratory sound pathological features based on the respiratory sound feature matrix includes: Using a pre-built correlation analysis method, a feature having a correlation with a preset pathological feature greater than a preset threshold is obtained from the respiratory sound feature matrix to obtain a pathology-related feature, and a pathology-related feature matrix is ​​constructed based on the pathology-related feature; The pathology-related feature matrix is ​​subjected to dimensionality reduction processing using a pre-constructed principal component analysis method to obtain a reduced-dimensional respiratory sound feature matrix, and features in the reduced-dimensional respiratory sound matrix are extracted to obtain respiratory sound pathology features.

8. The method for assessing the degree of central airway stenosis based on deep learning according to claim 7, wherein: The preset basic network is adjusted according to the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree and airway stenosis partition to obtain an adjusted network structure, including: Performing convolution on the basic network using a convolution layer with a preset accuracy based on the respiratory sound simulation signal data and the respiratory sound pathological characteristics, thereby obtaining parameters of one or two convolution layers in the basic network; performing convolution in the basic network using a 4×4 convolution kernel according to the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree, and airway stenosis partition, and comparing the convolution accuracy of the convolution kernel with a preset network diagnostic target to obtain a convolution kernel comparison difference; The basic network is adjusted according to the parameters of the first and second convolution layers in the basic network and the difference between the convolution kernels to obtain an adjusted network structure.

9. The method for assessing the degree of central airway stenosis based on deep learning according to claim 8, wherein: The method of training the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model includes: Dividing the respiratory sound simulation signal data into a training set, a test set, and a validation set to obtain a respiratory sound training set, a respiratory sound test set, and a respiratory sound validation set, wherein the ratio of the respiratory sound training set, the respiratory sound test set, and the respiratory sound validation set is set to 8:1:1; Optimizing the parameters of the initial respiratory sound analysis model using a preset Adam optimizer to obtain an optimized respiratory sound analysis model, wherein the initial learning rate of the Adam optimizer is 0.001, the batch size is 64, and the total number of iterations is 100; Using the respiratory sound training set to train the optimized respiratory sound analysis model to obtain a preliminary trained respiratory sound analysis model; The preliminary trained respiratory sound analysis model is tested using the respiratory sound test set to obtain a test respiratory sound analysis model, and the test respiratory sound analysis model is verified using the respiratory sound verification set. When the verification passes, a standard respiratory sound analysis model is obtained.

10. A central airway stenosis assessment system based on deep learning, characterized in that: The system comprises: a data acquisition module, configured to acquire body surface sounds of the patient using a preset sound acquisition device to obtain a comprehensive body surface sound, and extract respiratory sounds from the comprehensive body surface sound to obtain respiratory sound simulation signal data, wherein the comprehensive body surface sound is collected from the patient's suprasternal fossa, left main bronchus, and right main bronchus; a data processing module, configured to obtain a frequency band range of the respiratory sound analog signal data, select an analog signal filter based on the frequency band range to denoise the respiratory sound analog signal data to obtain denoised respiratory sound analog signal data, perform windowing processing on the denoised respiratory sound analog signal data to obtain windowed respiratory sound analog signal data, and construct a respiratory sound feature matrix based on the windowed respiratory sound analog signal data; a network adjustment module, configured to obtain respiratory sound pathological characteristics based on the respiratory sound feature matrix, perform an airway examination on the patient using a preset examination device to obtain the degree of airway stenosis and airway stenosis zones, and adjust a preset basic network based on the respiratory sound simulation signal data, respiratory sound pathological characteristics, airway stenosis degree, and airway stenosis zones to obtain an adjusted network structure; a model application module, configured to construct an initial respiratory sound analysis model according to the adjusted network structure, and train the initial respiratory sound analysis model using the respiratory sound simulation signal data to obtain a standard respiratory sound analysis model; Acquire the respiratory sound simulation signal data of the patient to be diagnosed, obtain the respiratory sound simulation signal data to be diagnosed, and use the standard respiratory sound analysis model to diagnose the respiratory sound simulation signal data to obtain the diagnosis result of the patient to be diagnosed.

Images