An intelligent respiratory sound analysis and diagnosis system

Through the intelligent breath sound analysis and diagnosis system, using high-sensitivity microphone arrays and deep learning technology, the time-frequency characteristics and slight variations of breath sound are extracted and analyzed, and the problem that traditional methods cannot identify subtle differences is solved, improving the accuracy of disease diagnosis and efficiency.

CN118553272BActive Publication Date: 2024-11-08JIANGSU RUIBO INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411010477.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-08
Estimated Expiration
2044-07-26

AI Technical Summary

Technical Problem

Traditional breath sound analysis methods have limitations in accurately identifying and processing breath sound data, and cannot fully explore subtle differences and disease characteristics in breath sound, resulting in a reduced possibility of early disease diagnosis, increasing the risk of misdiagnosis and misdiagnosis, affecting the effectiveness of medical diagnosis and efficiency.

Method used

The intelligent breath sound analysis and diagnosis system is adopted, including sound acquisition module, signal enhancement module, feature extraction module, pulmonary heart sound separation module, microvariance analysis module and disease identification module, and uses high-sensitivity microphone array, wavelet transformation, convolutional neural network and Long-term short-term memory network, combined with dynamic time-bending algorithms, extract and analyze the time-frequency characteristics and slight variations of respiratory sounds, and identify potential respiratory dysfunction and disease characteristics.

Benefits of technology

It realizes fine capture and analysis of respiratory sounds, improves the accuracy of automatic distinction between pulmonary heart sounds and the reliability of diagnostic data, reduces the error of human judgment, improves the accuracy and efficiency of disease diagnosis, and supports early identification of respiratory dysfunction .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118553272B_ABST
    Figure CN118553272B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of acoustic data analysis technology, specifically to an intelligent respiratory sound analysis and diagnosis system, which includes a sound collection module, a signal enhancement module, a feature extraction module, a lung and heart sound separation module, a micro-variation analysis module, a disease identification module, and a data integration module. In the present invention, advanced technologies such as high-sensitivity microphone arrays, signal enhancement, and wavelet transform are used to accurately capture and analyze the characteristics of the patient's respiratory sounds, especially through lung and heart sound separation and micro-variation analysis technology, to accurately identify small variations and abnormal patterns in respiratory sounds, to provide strong technical support for early identification of respiratory dysfunction, and to combine a hybrid model of a convolutional neural network and a long short-term memory network to optimize the automatic distinction between lung and heart sounds, enhance the accuracy and reliability of diagnostic data, reduce errors in human judgment, and improve the accuracy of disease diagnosis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of acoustic data analysis, and in particular to an intelligent respiratory sound analysis and diagnosis system. Background Art

[0002] The field of acoustic data analysis technology focuses on using sound data for information extraction, feature analysis and pattern recognition, combining signal processing, machine learning, data mining and acoustic principles to parse and understand the complex information contained in sound files. It is widely used in multiple industries and research fields, including but not limited to healthcare, environmental monitoring, smart home, security protection and audio recognition. In the healthcare field, acoustic data analysis technology can provide a non-invasive, low-cost diagnostic method to assist in the diagnosis and monitoring of diseases by analyzing the patient's voice characteristics, such as breathing sounds, heart sounds, etc., thereby achieving early detection, early diagnosis and monitoring of disease progression.

[0003] The intelligent respiratory sound analysis and diagnosis system is a medical diagnostic tool that integrates acoustic data analysis technology. Its main purpose is to use advanced signal processing and machine learning algorithms to automatically extract features from the patient's respiratory sounds, analyze and identify the acoustic patterns of respiratory diseases. It aims to provide doctors with an efficient and accurate auxiliary diagnostic method, improve the diagnostic process of respiratory-related diseases, shorten the diagnosis time, reduce the misdiagnosis rate, and improve the accuracy and efficiency of disease diagnosis, while providing patients with a more convenient and safe diagnosis and treatment experience.

[0004] Traditional breath sound analysis methods have significant limitations in accurately identifying and processing breath sound data. They rely on relatively primitive sound analysis tools and lack the ability to deeply analyze complex acoustic features, especially in detecting small variations and accurately separating lung and heart sounds. This limits the possibility of early disease diagnosis, increases the risk of misdiagnosis and missed diagnosis, and affects patients' chances of receiving timely and effective treatment. Fundamentally, these shortcomings stem from the limitations of traditional methods in acoustic data processing and analysis technology, which cannot fully tap into the subtle differences and disease characteristics in breath sounds, thereby reducing the overall effectiveness and efficiency of medical diagnosis. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent respiratory sound analysis and diagnosis system.

[0006] In order to achieve the above-mentioned object, the present invention adopts the following technical scheme: an intelligent respiratory sound analysis and diagnosis system includes a sound collection module, a signal enhancement module, a feature extraction module, a lung and heart sound separation module, a microvariation analysis module, a disease marker recognition module, and a data integration module;

[0007] The sound collection module uses a high-sensitivity microphone array to collect the patient's breathing sounds and heart sounds in a specific acoustic environment to obtain original breathing sound data;

[0008] The signal enhancement module reduces background noise and enhances the respiratory sound signal based on the original respiratory sound data, optimizes the clarity of the signal, and obtains a clear respiratory sound signal;

[0009] The feature extraction module applies wavelet transform to analyze the time-frequency characteristics of the breath sound according to the clear breath sound signal, extracts key features, including frequency distribution and energy peak, and generates key features of the breath sound;

[0010] The lung and heart sound separation module uses the key features of respiratory sounds and combines a hybrid model of a convolutional neural network and a long short-term memory network to automatically distinguish lung sounds from heart sounds and obtain respiratory sound separation data;

[0011] The microvariation analysis module performs dynamic time warping algorithm analysis on the respiratory sound separation data, identifies small periodic variations and abnormal patterns, reveals potential respiratory dysfunction, and generates a microvariation indicator set;

[0012] The disease identification module matches the microvariation index set with known disease characteristics, analyzes the patient's potential respiratory disease, and obtains disease prediction results;

[0013] The data integration module aggregates and formats disease prediction results, including potential respiratory and cardiac disease markers, to generate a comprehensive health data overview.

[0014] As a further solution of the present invention, the original respiratory sound data includes sound waveforms and sound intensity changes in different respiratory stages, the clear respiratory sound signal includes sampling frequency and number of quantization bits, the key features of respiratory sounds include frequency distribution, sound duration, and sound energy distribution, the respiratory sound separation data includes the duration of lung sounds, the rhythm and frequency of heart sounds, the micro-variation indicator set includes variation frequency, variation intensity, and variation time interval, the disease prediction results specifically include potential disease types and associated disease types, and the comprehensive health data overview includes disease prediction overview, key variation features, and recommended inspection items.

[0015] As a further solution of the present invention, the sound collection module includes a sound positioning submodule, a signal recording submodule, and a data preliminary processing submodule;

[0016] The sound localization submodule is based on a high-sensitivity microphone array to capture and locate the optimal collection points of the patient's breathing sounds and heart sounds in a specific acoustic environment and generate a localization result;

[0017] The signal recording submodule collects the patient's breathing sounds and heart sounds based on the positioning results using a high-sensitivity microphone array to obtain initial breathing sound data;

[0018] The data preliminary processing submodule performs preliminary signal sorting based on the initial respiratory sound data, removes invalid data and obvious errors, and obtains original respiratory sound data.

[0019] As a further solution of the present invention, the signal enhancement module includes a noise identification submodule, a signal adjustment submodule, and a clarity adjustment submodule;

[0020] The noise identification submodule analyzes the spectrum of the original breathing sound data, calibrates the frequency range of environmental and non-environmental noise, adjusts the filter parameters to match the frequency, and generates noise calibration data;

[0021] The signal adjustment submodule adjusts the signal processing parameters, including gain and frequency response, according to the noise calibration data, strengthens the breathing sound signal and suppresses the noise, and obtains the adjusted signal;

[0022] The clarity adjustment submodule uses the adjusted signal to perform dynamic range compression and equalization adjustment to optimize the overall clarity of the audio signal and generate a clear breathing sound signal.

[0023] As a further solution of the present invention, the feature extraction module includes a time-frequency analysis submodule, a feature recognition submodule, and a feature integration submodule;

[0024] The time-frequency analysis submodule performs time-frequency transformation based on the clear breathing sound signal, analyzes the performance of the signal in non-differential time and frequency, reveals the dynamic characteristics of the signal, and generates a time-frequency analysis result;

[0025] The feature recognition submodule recognizes key feature signals from the time-frequency analysis results, including frequency distribution and energy peaks, extracts the most valuable information from the results, and obtains a feature data set;

[0026] The feature integration submodule integrates and optimizes the extracted features based on the feature data set to ensure the integrity and availability of the data, and constructs the key features of respiratory sounds in combination with the importance and interrelationship of the features.

[0027] As a further solution of the present invention, the lung and heart sound separation module includes a feature screening submodule, a model training submodule, and an audio classification submodule;

[0028] The feature screening submodule is based on the key features of respiratory sounds. If the features meet the preset conditions, including the frequency range and energy threshold, the key features for distinguishing lung sounds from heart sounds are extracted to generate a screened feature set;

[0029] The model training submodule cyclically trains the hybrid model of the convolutional neural network and the long short-term memory network according to the screened feature set until the predetermined performance index is reached, the accuracy rate and the loss rate meet the threshold, and the training and perfecting model is obtained;

[0030] The audio classification submodule classifies the input respiratory sounds based on the trained and improved model. If the model determines that the audio features match the lung sound features, it is classified as lung sound, otherwise it is classified as heart sound, and respiratory sound separation data is constructed.

[0031] As a further solution of the present invention, the micro-variation analysis module includes a time alignment submodule, a difference positioning submodule, and an index evaluation submodule;

[0032] The time alignment submodule performs time alignment based on the breath sound separation data, adjusts the data points to match the signals of the difference time periods, ensures that the data has a consistent time reference before analysis, and generates time-aligned data;

[0033] The difference positioning submodule uses a dynamic time warping algorithm to identify periodic variations and abnormal patterns based on time-aligned data, and screens out significant differences by setting thresholds to generate difference positioning results;

[0034] The execution formula of the difference positioning submodule is as follows:

[0035]

[0036] in, Indicates the arrival point The minimum cumulative distance, Indicate point and Point The distance measure between Represents the time points in two time series respectively. Represents the sequence length adjustment factor, which adjusts the comparison weight of the difference length time series. represents the time elasticity coefficient, which adjusts the time bending flexibility in time series alignment. Indicates the weight of abnormal patterns and improves the sensitivity of identifying abnormal data points;

[0037] The index evaluation submodule analyzes the association between the identified differences and respiratory dysfunction based on the difference localization results, reveals potential respiratory dysfunction according to the type and degree of the differences, and constructs a microvariation index set.

[0038] As a further solution of the present invention, the disease identification module includes a feature matching submodule, a disease analysis submodule, and a diagnosis result generation submodule;

[0039] The feature matching submodule performs a matching process with a known disease feature library based on the microvariation index set, and generates a feature matching result by comparing the microvariation index set with the features of the disease in the feature library;

[0040] The disease analysis submodule analyzes the correlation between the matched disease features and the patient's respiratory sound data according to the feature matching results, and screens out potential respiratory diseases by calculating similarity scores to obtain potential disease analysis results;

[0041] The prediction result generation submodule constructs a disease prediction result based on the potential disease analysis results, the patient's microvariation index and the disease feature matching degree, including the potential disease name.

[0042] As a further solution of the present invention, the data integration module includes a result summary submodule, a formatting processing submodule, and an overview construction submodule;

[0043] The result summary submodule collects and summarizes all relevant health data, including potential respiratory and cardiac disease markers, based on the disease prediction results to generate a summary data set;

[0044] The formatting processing submodule formats the data according to the summary data set to ensure the clarity and readability of the information, and obtains formatted data by standardizing the display format of the data;

[0045] The overview building submodule comprehensively analyzes and organizes information based on formatted data, obtains key health indicators and assesses disease risks, and generates a comprehensive health data overview.

[0046] Compared with the prior art, the advantages and positive effects of the present invention are:

[0047] In the present invention, advanced technologies such as high-sensitivity microphone array, signal enhancement, wavelet transform, etc. are used to accurately capture and analyze the characteristics of the patient's respiratory sounds. In particular, through the separation of lung and heart sounds and micro-variation analysis technology, small variations and abnormal patterns in respiratory sounds can be accurately identified, providing strong technical support for the early identification of respiratory dysfunction. Combined with the hybrid model of convolutional neural network and long short-term memory network, the automatic distinction of lung and heart sounds is optimized, and the accuracy and reliability of diagnostic data are enhanced, ensuring that even in a complex acoustic environment, key information can be efficiently extracted from respiratory sounds, reducing errors in human judgment and improving the accuracy of disease diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a system flow chart of the present invention;

[0049] Figure 2 It is a schematic diagram of the system framework of the present invention;

[0050] Figure 3 It is a flow chart of the sound collection module of the present invention;

[0051] Figure 4 This is a flow chart of the signal enhancement module of the present invention;

[0052] Figure 5 It is a flow chart of the feature extraction module of the present invention;

[0053] Figure 6 This is a flow chart of the lung and heart sound separation module of the present invention;

[0054] Figure 7 This is a flow chart of the microvariation analysis module of the present invention;

[0055] Figure 8 This is a flow chart of the disease identification module of the present invention;

[0056] Fig. 9 It is a flow chart of the data integration module of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. 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.

[0058] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined. Embodiment 1

[0059] See also Figure 1 and Figure 2 ,An intelligent respiratory sound analysis and diagnosis system includes a sound acquisition module, a signal enhancement module, a feature extraction module, a lung and heart sound separation module, a micro-variation analysis module, a disease identification module, and a data integration module;

[0060] The sound collection module uses a high-sensitivity microphone array to collect the patient's breathing sounds and heart sounds in a specific acoustic environment to obtain the original breathing sound data;

[0061] The signal enhancement module reduces background noise and enhances the respiratory sound signal based on the original respiratory sound data, optimizes the signal clarity, and obtains a clear respiratory sound signal;

[0062] The feature extraction module uses wavelet transform to analyze the time-frequency characteristics of respiratory sounds based on clear respiratory sound signals, extracts key features, including frequency distribution and energy peak, and generates key features of respiratory sounds;

[0063] The lung and heart sound separation module uses the key features of respiratory sounds and combines a hybrid model of convolutional neural networks and long short-term memory networks to automatically distinguish lung sounds from heart sounds and obtain respiratory sound separation data;

[0064] The microvariation analysis module performs dynamic time warping algorithm analysis on the respiratory sound separation data to identify small periodic variations and abnormal patterns, reveal potential respiratory dysfunction, and generate a microvariation indicator set;

[0065] The disease identification module matches the microvariation indicator set with known disease characteristics, analyzes the patient's potential respiratory disease, and obtains disease prediction results;

[0066] The data integration module aggregates and formats disease prediction results, including potential respiratory and cardiac disease markers, to produce a comprehensive health data overview.

[0067] The original respiratory sound data include the sound waveform and sound intensity changes in different respiratory stages. The clear respiratory sound signal includes the sampling frequency and the number of quantization bits. The key features of respiratory sounds include frequency distribution, sound duration, and sound energy distribution. The respiratory sound separation data includes the duration of lung sounds, the rhythm and frequency of heart sounds. The micro-variation indicator set includes variation frequency, variation intensity, and variation time interval. The disease prediction results are specifically potential disease types and associated disease types. The comprehensive health data overview includes disease prediction overview, key variation characteristics, and recommended inspection items.

[0068] In the sound collection module, a high-sensitivity microphone array is used to collect the patient's breathing sounds and heart sounds in a specific acoustic environment to obtain raw breathing sound data. At this stage, the collected data format is mainly an audio file, including the sound waveform and sound intensity changes in the differential breathing phase. The high-sensitivity microphone ensures that sound signals can be effectively collected even at low volume or long distance, capturing subtle sound changes and laying the foundation for subsequent analysis.

[0069] In the signal enhancement module, based on the original breath sound data, noise reduction techniques and signal enhancement algorithms such as spectral subtraction and Wiener filtering are used to reduce background noise and enhance breath sound signals. This step generates a clear breath sound signal by optimizing the clarity of the signal, in the form of an adjusted audio file with sampling frequency and quantization bits. The application of the signal enhancement algorithm significantly improves the signal-to-noise ratio, ensuring the accuracy and effectiveness of subsequent feature extraction and analysis.

[0070] In the feature extraction module, based on the clear respiratory sound signal, wavelet transform is applied to analyze the time-frequency characteristics of respiratory sounds and extract key features, including frequency distribution, energy peak, sound duration and sound energy distribution. The application of wavelet transform can accurately analyze the frequency and time information of the signal at different scales. The key features are saved in the form of data sets, which provides a basis for the accurate separation of lung and heart sounds and the identification of disease characteristics.

[0071] In the lung and heart sound separation module, the key features of respiratory sounds are used, combined with a hybrid model of convolutional neural networks and long short-term memory networks, to automatically distinguish lung sounds from heart sounds. This process generates respiratory sound separation data, including the duration of lung sounds, the rhythm and frequency of heart sounds, by training a deep learning model to identify different sound features. The application of the hybrid model enhances the learning ability of complex sound features, improves separation accuracy, and provides an accurate sound basis for identifying specific diseases.

[0072] In the microvariation analysis module, the dynamic time warping algorithm is used to analyze the respiratory sound separation data to identify small periodic variations and abnormal patterns. This process generates a set of microvariation indicators, including variation frequency, variation intensity, and variation time interval. The application of the dynamic time warping algorithm enables the system to accurately match and identify pattern changes in time series, revealing potential respiratory dysfunction and providing important information for early diagnosis and treatment of diseases.

[0073] In the disease identification module, the microvariation indicator set is matched with known disease features, and machine learning algorithms such as support vector machines and random forests are used to predict diseases, obtaining disease prediction results, including potential disease types and associated disease types. This process improves the accuracy and efficiency of diagnosis by training algorithms to identify the relationship between specific sound patterns and diseases.

[0074] In the data integration module, the disease prediction results are summarized and formatted to generate a comprehensive health data overview, including disease prediction overview, key variant characteristics and recommended examination items. By integrating and analyzing all relevant data, a comprehensive and easy-to-understand health report is provided to doctors, helping them make accurate diagnoses and treatment decisions, and improving the quality and efficiency of medical services.

[0075] See also Figure 2 and Figure 3 ,The sound collection module includes a sound positioning submodule, a signal recording submodule, and a data preliminary processing submodule;

[0076] The sound localization submodule is based on a high-sensitivity microphone array, which captures and locates the optimal collection points of the patient's breathing and heart sounds in a specific acoustic environment and generates localization results.

[0077] The sound localization submodule is based on a high-sensitivity microphone array. It first performs spectrum analysis on the ambient noise to ensure that the recording is performed under low noise conditions. Next, the system quantitatively analyzes the direction of the sound source by transmitting and receiving sound waves of a specific frequency. Using the echolocation method, multiple sound wave detection points are set, and the sound source position is calculated by the reflection time difference of the sound wave. Assume that the sound source position is , the microphone receiving point is ( ), the speed of sound waves in air is , the propagation time of the sound wave from the sound source to the microphone is , then:

[0078]

[0079] Indicates the sound source To each receiving point The distance Represents the theoretical distance of sound wave propagation. The best sound source position is estimated by minimizing the sum of squared errors. , to accurately locate the optimal collection point of the patient's respiratory sounds and heart sounds, and generate positioning results.

[0080] Based on the positioning results, the signal recording submodule uses a high-sensitivity microphone array to collect the patient's breathing sounds and heart sounds to obtain initial breathing sound data;

[0081] Based on the positioning results, the signal recording submodule first adjusts the directivity of the microphone array to align with the optimal collection point. Then, the module adjusts the gain and frequency response to adapt to the breathing and heart sound characteristics of different patients. The dynamic threshold adjustment technology is used to perform real-time peak detection on the signal to ensure that the recorded sound data maintains the best quality within the dynamic range. In specific operations, each heartbeat and breathing cycle is collected independently, and the timestamp and signal strength are recorded separately. The calculation method is as follows:

[0082] Recording Data

[0083] in, It's in time The sound intensity, is a relative threshold, set as a percentage of the sound intensity to ensure that only sounds above this threshold are recorded. By adjusting the microphone parameters and threshold, the patient's breathing sounds and heart sounds are accurately captured to obtain initial breathing sound data.

[0084] The data preliminary processing submodule performs preliminary signal sorting based on the initial respiratory sound data, removes invalid data and obvious errors, and obtains the original respiratory sound data.

[0085] The data preliminary processing submodule first performs time-frequency analysis on the recorded sound data based on the initial breathing sound data to identify the periodic patterns of breathing and heartbeat. It uses frequency selective filtering technology to remove frequency components unrelated to breathing and heartbeat to improve the signal-to-noise ratio of the data. Subsequently, the system normalizes the frequency components to standardize the amplitude of the signal.

[0086] Processed data = (signal data - mean) / standard deviation

[0087] In this way, outliers and background noise are eliminated, and standardized operations make the data more consistent, which is convenient for subsequent analysis, and the original respiratory sound data is obtained, which retains the patient's physiological information and improves the data quality, which is convenient for subsequent detailed analysis and feature extraction.

[0088] See also Figure 2 and Figure 4 ,The signal enhancement module includes a noise identification submodule, a signal adjustment submodule, and a clarity adjustment submodule;

[0089] The noise identification submodule analyzes the spectrum of the original respiratory sound data, calibrates the frequency range of environmental and non-environmental noise, adjusts the filter parameters to match the frequency, and generates noise calibration data;

[0090] The noise identification submodule first performs a deep spectrum analysis on the original breathing sound data, and uses Fourier transform to convert the time series data into a frequency domain representation. During the analysis, the system identifies the frequency band that matches the human breathing and heartbeat frequency, and identifies the frequencies outside of it as environmental noise or non-environmental noise. The specific frequency calibration method is as follows:

[0091] Frequency calibration Average power density Ambient noise frequency range

[0092] in, Indicates the frequency The power density at It is a preset threshold coefficient used to distinguish significant noise from background noise. In this way, useful signals are effectively distinguished from noise, and the frequency range of each noise type is accurately calibrated for subsequent processing. Based on these analysis results, the system automatically adjusts the filter parameters to match the main noise frequency, weaken its influence, and generate noise calibration data.

[0093] The signal adjustment submodule adjusts the signal processing parameters, including gain and frequency response, according to the noise calibration data, strengthens the breathing sound signal and suppresses the noise, and obtains the adjusted signal;

[0094] The signal adjustment submodule accurately adjusts the key parameters of signal processing according to the noise calibration data, including gain adjustment and frequency response optimization. The module first applies adaptive filtering technology to apply noise reduction processing to the calibrated noise frequency range to enhance the recognizability of human breathing and heartbeat signals. The gain adjustment process is as follows:

[0095]

[0096] in, is the raw signal strength, Is the gain adjustment coefficient. The signal-to-noise ratio is used to measure the ratio of signal to noise. By increasing the gain related to the signal-to-noise ratio, the main part of the signal is strengthened while suppressing the noise. This precise parameter adjustment process ensures the effectiveness and clarity of the signal, thereby obtaining the adjusted signal.

[0097] The clarity adjustment submodule uses the adjusted signal to perform dynamic range compression and equalization adjustment to optimize the overall clarity of the audio signal and generate a clear breathing sound signal.

[0098] The clarity adjustment submodule uses the adjusted signal to perform dynamic range compression and frequency equalization adjustment to optimize the overall clarity of the audio signal. Dynamic range compression reduces the dynamic range difference of the signal, so that the low volume components in the audio are enhanced and the high volume components are controlled.

[0099]

[0100] here, is the adjusted signal, is the compressed output signal, is the compression factor, which controls the degree of compression. Frequency equalization adjusts the gain of different frequency bands to make the audio output more balanced and enhance the recognizability of speech and breathing sounds. These processing processes work together to improve the clarity of the breathing sound signal, generate the final clear breathing sound signal, and ensure the signal quality and accuracy of analysis.

[0101] See also Figure 2 and Figure 5 ,The feature extraction module includes a time-frequency analysis submodule, a feature recognition submodule, and a feature integration submodule;

[0102] The time-frequency analysis submodule performs time-frequency transformation based on the clear breathing sound signal, analyzes the performance of the signal at the same time and frequency, reveals the dynamic characteristics of the signal, and generates the time-frequency analysis results;

[0103] The time-frequency analysis submodule is based on the clear breathing sound signal. It first implements time-frequency transformation, specifically using short-time Fourier transform (STFT) to accurately analyze the performance of the signal at different times and frequencies. Through this transformation, the module can dynamically track the changes in signal frequency over time and reveal the complex dynamic characteristics of the breathing sound signal.

[0104]

[0105] in, is the signal at time The value of So The window function with the center as the center makes the time-frequency analysis localized. By adjusting the width of the window function, the resolution of time and frequency can be controlled, so as to finely capture the local characteristics of the signal, and the result includes the energy distribution of each frequency component over time.

[0106] The feature recognition submodule identifies key feature signals from the time-frequency analysis results, including frequency distribution and energy peaks, extracts the most valuable information from the results, and obtains feature data sets;

[0107] The feature recognition submodule conducts in-depth analysis of the time-frequency analysis results to identify key feature signals, including frequency distribution and energy peaks. The module sets energy thresholds to detect significant energy peaks, which usually represent key frequency features of breathing and heartbeat. During the recognition process, the module calculates the average energy of each frequency component and determines the peak by comparing its energy difference with the surrounding frequencies.

[0108]

[0109] in, is the frequency At all times After identifying these features, the module organizes them into a feature data set, which contains all the identified frequencies and corresponding energy values, providing an accurate data basis for subsequent analysis.

[0110] The feature integration submodule integrates and optimizes the extracted features based on the feature data set to ensure the integrity and availability of the data, and constructs the key features of respiratory sounds based on the importance and interrelationship of the features.

[0111] The feature integration submodule performs further feature integration and optimization based on the extracted feature data set. First, the feature data is statistically analyzed to evaluate the statistical significance of each feature and sort them according to their importance. Using principal component analysis (PCA) and correlation coefficient matrix, the module analyzes the interrelationships between features and selects the most critical feature combinations for diagnosis.

[0112]

[0113] in, Indicates each frequency In this way, the module not only ensures the integrity and availability of the data, but also builds a key feature model of respiratory sounds based on the correlation between features, providing strong data support for the final diagnosis and analysis.

[0114] See also Figure 2 and Figure 6 ,The pulmonary and heart sound separation module includes a feature screening submodule, a model training submodule, and an audio classification submodule;

[0115] The feature screening submodule is based on the key features of respiratory sounds. If the features meet the preset conditions, including frequency range and energy threshold, the key features used to distinguish lung sounds from heart sounds are extracted to generate a screened feature set.

[0116] The feature screening submodule operates based on the key features of respiratory sounds. First, specific screening criteria are set, including frequency range and energy threshold. The feature screening process checks each feature one by one to ensure that the frequency of each feature is within the set valid frequency range. For example, the human heartbeat and breathing frequency usually fall within a specific frequency band. At the same time, the module also evaluates whether the energy of each feature exceeds the preset energy threshold to determine its significance and relevance.

[0117]

[0118] in, and Represents the preset minimum and maximum frequency ranges, For frequency The energy value, is the energy threshold. The features that meet these conditions are extracted as key features for distinguishing lung sounds from heart sounds, generating a filtered feature set.

[0119] The model training submodule cyclically trains the hybrid model of the convolutional neural network and the long short-term memory network according to the filtered feature set until the predetermined performance indicators are reached, ensuring that the accuracy and loss rate meet the thresholds, and obtaining the trained and perfected model;

[0120] The model training submodule uses a loop training method to optimize a hybrid model of a convolutional neural network (CNN) and a long short-term memory network (LSTM). The model combines the spatial feature extraction capability of CNN and the time series prediction capability of LSTM to process time-dependent respiratory sound signals. During the training process, the module continuously adjusts network parameters, such as weights and biases, to gradually reduce errors and improve accuracy based on feedback from the training dataset. Training continues until the model's performance indicators, such as the loss rate, reach a predetermined threshold.

[0121]

[0122] in, is the loss rate, is the prediction result of the model, is the label value of the actual data. This training process ensures that the model has efficient learning and generalization capabilities, and ultimately obtains a trained and perfect model.

[0123] The audio classification submodule classifies the input respiratory sounds based on the trained and improved model. If the model determines that the audio features match the lung sound features, it is classified as lung sound, otherwise it is classified as heart sound, and the respiratory sound separation data is constructed.

[0124] The audio classification submodule effectively classifies the input respiratory sounds based on the trained model. The module uses the trained model to identify and distinguish lung sounds and heart sounds in the input signal. In specific operations, the module analyzes the input audio features in real time, and the model matches and classifies according to the feature patterns accumulated during training. If the model determines that the audio feature matches the feature template of lung sounds, it is classified as a lung sound, otherwise it is classified as a heart sound. This process is achieved through probability evaluation within the model to ensure the accuracy and efficiency of classification, while constructing respiratory sound separation data to support medical diagnosis and treatment decisions.

[0125] See also Figure 2 and Figure 7 ,The micro-variation analysis module includes the time alignment submodule, the difference ,localization submodule, and the index evaluation submodule;

[0126] The time alignment submodule performs time alignment based on the breath sound separation data, adjusts the data points to match the signals of the difference time periods, ensures that the data has a consistent time reference before analysis, and generates time-aligned data;

[0127] The time alignment submodule starts to perform time alignment based on the breath sound separation data. The purpose is to adjust the data points to match the signals of different time periods to ensure that all data have a consistent time reference before analysis. This process uses linear interpolation to time-standardize asynchronous data points to ensure data continuity and consistency. Specifically, for each data point in the time series, the time difference between adjacent points is calculated, and time interpolation is performed as needed to fill in missing data.

[0128]

[0129] in, and is the time stamp of the known data point, and are the data values ​​corresponding to these time points. In this way, all data points are aligned on the time axis before analysis, generating time-aligned data.

[0130] The difference localization submodule uses a dynamic time warping algorithm to identify periodic variations and abnormal patterns based on time-aligned data, and selects significant differences by setting thresholds to generate difference localization results.

[0131] In the difference positioning submodule, based on the time alignment data, the dynamic time warping algorithm is used, and the parameters including window size and distance measurement method are set to identify periodic variations and abnormal patterns. By comparing the variation amplitude with the preset threshold, significant differences are identified and the difference positioning results are generated.

[0132] The execution formula of the difference positioning submodule is as follows:

[0133]

[0134] : Arrival point The minimum cumulative distance of

[0135] :point and Point The distance measure between

[0136] : Represent the time points in two time series respectively;

[0137] : Sequence length adjustment factor, which adjusts the comparison weight of the time series with different lengths;

[0138] : Time elasticity coefficient, adjusts the flexibility of time bending in time series alignment;

[0139] : Abnormal pattern weight, improving the sensitivity of identifying abnormal data points;

[0140] The execution process is as follows:

[0141] Sequence length adjustment ( ): It is used to balance the effect of length on the matching process. It can be set according to the ratio of sequence lengths, such as the ratio of the length of the shorter sequence to the length of the longer sequence, to ensure that the matching results will not be adversely affected when the length difference is significant;

[0142] Time flexibility adjustment ( ): reflects the degree of time bending allowed when aligning time series. Higher Values ​​allow for larger time shifts and are suitable for series with loose alignment of time points, which can be determined by analyzing the need for time warping in the time series, for example, through exploratory data analysis during the preprocessing phase;

[0143] Abnormal mode weighting ( ): Increase the sensitivity to abnormal points in the time series by assigning higher weights to these points, which can be set based on the severity of previously identified abnormal patterns, for example, based on the impact of the outliers on the overall series or the frequency of anomalies;

[0144] Calculate DTW distance: After determining the above parameters, use the formula to calculate the DTW distance between two time series, traverse all point pairs in the time series, and use the recursive relationship to calculate the minimum cumulative distance, while taking into account the effects of sequence length, time elasticity, and abnormal patterns.

[0145] Threshold setting for significant differences: After identifying the variation between time series through the DTW method, a threshold is set to determine which variation is significant. The threshold can be adjusted dynamically according to the characteristics of the data set. For example, the threshold can be selected based on the distribution of variation in the data set to ensure that the identification of significant differences is neither too loose nor too strict.

[0146] By introducing innovative parameters and improving formulas, not only the algorithm's adaptability to different time series characteristics is enhanced, but also the flexibility and accuracy of significant difference identification are improved, providing a more accurate and efficient tool for time series data analysis, fully ensuring the accuracy of difference location results, and preparing for the subsequent construction of micro-variation indicator sets.

[0147] The indicator evaluation submodule analyzes the association between the identified differences and respiratory dysfunction based on the difference location results, reveals potential respiratory dysfunction according to the type and degree of the differences, and constructs a microvariation indicator set;

[0148] Based on the difference location results, the indicator evaluation submodule conducts in-depth analysis of the identified differences to reveal their potential association with respiratory dysfunction. The module uses statistical analysis methods, such as correlation analysis and regression analysis, to evaluate the relationship between the type and degree of differences and respiratory dysfunction. Through these analyses, a microvariation indicator set is constructed, which reflects the possible small changes in respiratory function and provides important quantitative data for medical diagnosis.

[0149] See also Figure 2 and Figure 8 ,The disease identification module includes a feature matching submodule, a disease analysis submodule, and a diagnosis result generation submodule;

[0150] The feature matching submodule performs the matching process with the known disease feature library based on the microvariation indicator set, and generates feature matching results by comparing the microvariation indicator set with the characteristics of the disease in the feature library;

[0151] The feature matching submodule performs a matching process with a known disease feature library based on the microvariation indicator set. First, the module compares the microvariation indicator set with the known features of each disease in the disease feature library. This comparison process uses a vector similarity algorithm, such as cosine similarity, to evaluate the similarity between the indicator set and each disease feature.

[0152]

[0153] in, is the vector representation of the microvariation index set, It is a vector representation of a disease feature in the disease feature library. Through this method, the module can identify the disease feature that is closest to the patient's microvariation indicator set and generate feature matching results, which helps to determine which disease features are most consistent with the patient's condition.

[0154] The disease analysis submodule analyzes the correlation between the matched disease features and the patient's respiratory sound data based on the feature matching results, and screens out potential respiratory diseases by calculating the similarity score to obtain potential disease analysis results;

[0155] The disease analysis submodule analyzes the correlation between the matched disease features and the patient's breath sound data based on the feature matching results. The module uses statistical analysis techniques, such as the Pearson correlation coefficient, to quantify the linear correlation between disease features and breath sound data. In addition, the module calculates a comprehensive similarity score, taking into account the weight and importance of each matching feature, and prioritizes those diseases that are highly correlated with the patient's data. In this way, the module can screen out potential respiratory diseases that are most likely to be related to the patient's current symptoms and obtain potential disease analysis results.

[0156] The prediction result generation submodule is based on the potential disease analysis results, comprehensively considers the patient's microvariation indicators and disease feature matching, including the potential disease name, to construct the disease prediction results.

[0157] The prediction result generation submodule is based on the potential disease analysis results, integrating the patient's microvariation indicators and disease feature matching. The module uses the obtained analysis results to build a disease prediction model, which will include the name of the potential disease, the relevance score, and the matching degree. In order to optimize the accuracy of the prediction, the module also considers the patient's clinical data and historical health records to provide more personalized disease predictions. In this way, the module can not only provide predictions for specific diseases, but also indicate which diseases are most relevant to the patient's specific situation, helping medical professionals to make more effective diagnosis and treatment planning.

[0158] See also Figure 2 and Fig. 9 ,The data integration module includes the result summary submodule, the formatting processing submodule, and the ,overview building submodule;

[0159] The result summary submodule collects and summarizes all relevant health data, including potential respiratory and cardiac disease markers, based on the disease prediction results to generate a summary data set;

[0160] The result summary submodule performs data collection and aggregation based on the disease prediction results to ensure that all relevant health data are included, which mainly involves potential respiratory and cardiac disease markers. The module first extracts data from various modules and databases, such as breath sound analysis results, feature matching results, and historical medical records. Then, this data is integrated to build a comprehensive data set, including the source, type, and contribution of each data to the patient's diagnosis. Through this comprehensive data processing, the module is able to generate a comprehensive summary data set that will serve as the basis for further analysis and reporting.

[0161] The formatting submodule formats the data according to the summary data set to ensure the clarity and readability of the information and obtains the formatted data by standardizing the display format of the data;

[0162] The formatting submodule formats the data based on the summary data set to ensure the clarity and readability of the information. The module uses standardized data formats and clear labels to organize information, such as using JSON or XML formats to standardize data structures. In addition, the module also applies data cleaning techniques such as deleting duplicate records, correcting data errors, and filling missing values ​​to further improve the quality and usability of the data. The formatting of the data not only improves the overall quality of the data, but also ensures the ease of use and consistency of the data in subsequent processing. Finally, the module outputs the formatted data to provide users with a clear and standardized data display.

[0163] The overview building submodule comprehensively analyzes and organizes information based on formatted data, obtains key health indicators and assesses disease risks, and generates a comprehensive health data overview;

[0164] The overview construction submodule is based on formatted data, comprehensive analysis and organization of information, focusing on obtaining key health indicators and conducting disease risk assessments. This module uses data visualization tools and techniques, such as dashboards and charts, to display changing trends in key health indicators such as heart rate, blood pressure, and respiratory rate. In addition, the module also applies risk assessment algorithms, such as risk score models based on patient data, to assess the probability of an individual developing a specific disease. These analyses provide a comprehensive overview of health data by taking into account an individual's health history, lifestyle, and genetic factors, which not only helps medical professionals quickly understand the patient's health status, but also facilitates patients to self-manage and monitor their health status.

[0165] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent respiratory sound analysis and diagnosis system, characterized in that: The intelligent respiratory sound analysis and diagnosis system includes a sound collection module, a signal enhancement module, a feature extraction module, a lung and heart sound separation module, a microvariation analysis module, a disease marker recognition module, and a data integration module; The sound collection module uses a high-sensitivity microphone array to collect the patient's breathing sounds and heart sounds in a specific acoustic environment to obtain original breathing sound data; The signal enhancement module reduces background noise and enhances the respiratory sound signal based on the original respiratory sound data, optimizes the clarity of the signal, and obtains a clear respiratory sound signal; The feature extraction module applies wavelet transform to analyze the time-frequency characteristics of the breath sound according to the clear breath sound signal, extracts key features, including frequency distribution and energy peak, and generates key features of the breath sound; The lung and heart sound separation module uses the key features of respiratory sounds and combines a hybrid model of a convolutional neural network and a long short-term memory network to automatically distinguish lung sounds from heart sounds and obtain respiratory sound separation data; The microvariation analysis module performs dynamic time warping algorithm analysis on the respiratory sound separation data, identifies small periodic variations and abnormal patterns, reveals potential respiratory dysfunction, and generates a microvariation indicator set; The micro-variation analysis module includes a time alignment submodule, a difference positioning submodule, and an index evaluation submodule; The time alignment submodule performs time alignment based on the breath sound separation data, adjusts the data points to match the signals of the difference time periods, ensures that the data has a consistent time reference before analysis, and generates time-aligned data; The difference positioning submodule uses a dynamic time warping algorithm to identify periodic variations and abnormal patterns based on time-aligned data, and screens out significant differences by setting thresholds to generate difference positioning results; The execution formula of the difference positioning submodule is as follows: ; in, Indicates the arrival point and The minimum cumulative distance, Indicate point and Point The distance measure between and Represents the time points in two time series respectively. represents the sequence length adjustment factor, represents the time elasticity coefficient, represents the abnormal pattern weight; The index evaluation submodule analyzes the association between the identified differences and respiratory dysfunction based on the difference location results, reveals potential respiratory dysfunction according to the type and degree of the differences, and constructs a microvariation index set; The disease identification module matches the microvariation index set with known disease characteristics, analyzes the patient's potential respiratory disease, and obtains disease prediction results; The data integration module aggregates and formats disease prediction results, including potential respiratory and cardiac disease markers, to generate a comprehensive health data overview.

2. The intelligent respiratory sound analysis and diagnosis system according to claim 1, characterized in that: The original respiratory sound data includes the sound waveform and sound intensity changes of different respiratory stages, the clear respiratory sound signal includes the sampling frequency and the number of quantization bits, the key features of respiratory sounds include frequency distribution, sound duration, and sound energy distribution, the respiratory sound separation data includes the duration of lung sounds, the rhythm and frequency of heart sounds, the micro-variation indicator set includes variation frequency, variation intensity, and variation time interval, the disease prediction results specifically include potential disease types and associated disease types, and the comprehensive health data overview includes disease prediction overview, key variation features, and recommended inspection items.

3. The intelligent respiratory sound analysis and diagnosis system according to claim 1, characterized in that: The sound collection module includes a sound positioning submodule, a signal recording submodule, and a data preliminary processing submodule; The sound localization submodule is based on a high-sensitivity microphone array to capture and locate the optimal collection points of the patient's breathing sounds and heart sounds in a specific acoustic environment and generate a localization result; The signal recording submodule collects the patient's breathing sounds and heart sounds based on the positioning results using a high-sensitivity microphone array to obtain initial breathing sound data; The data preliminary processing submodule performs preliminary signal sorting based on the initial respiratory sound data, removes invalid data and obvious errors, and obtains original respiratory sound data.

4. The intelligent respiratory sound analysis and diagnosis system according to claim 1, characterized in that: The signal enhancement module includes a noise identification submodule, a signal adjustment submodule, and a clarity adjustment submodule; The noise identification submodule analyzes the spectrum of the original breathing sound data, calibrates the frequency range of environmental and non-environmental noise, adjusts the filter parameters to match the frequency, and generates noise calibration data; The signal adjustment submodule adjusts the signal processing parameters, including gain and frequency response, according to the noise calibration data, strengthens the breathing sound signal and suppresses the noise, and obtains the adjusted signal; The clarity adjustment submodule uses the adjusted signal to perform dynamic range compression and equalization adjustment to optimize the overall clarity of the audio signal and generate a clear breathing sound signal.

5. The intelligent respiratory sound analysis and diagnosis system according to claim 1, characterized in that: The feature extraction module includes a time-frequency analysis submodule, a feature recognition submodule, and a feature integration submodule; The time-frequency analysis submodule performs time-frequency transformation based on the clear breathing sound signal, analyzes the performance of the signal in non-differential time and frequency, reveals the dynamic characteristics of the signal, and generates a time-frequency analysis result; The feature recognition submodule recognizes key feature signals from the time-frequency analysis results, including frequency distribution and energy peaks, extracts the most valuable information from the results, and obtains a feature data set; The feature integration submodule integrates and optimizes the extracted features based on the feature data set to ensure the integrity and availability of the data, and constructs the key features of respiratory sounds in combination with the importance and interrelationship of the features.

6. The intelligent respiratory sound analysis and diagnosis system according to claim 1, characterized in that: The lung and heart sound separation module includes a feature screening submodule, a model training submodule, and an audio classification submodule; The feature screening submodule is based on the key features of respiratory sounds. If the features meet the preset conditions, including the frequency range and energy threshold, the key features for distinguishing lung sounds from heart sounds are extracted to generate a screened feature set; The model training submodule cyclically trains the hybrid model of the convolutional neural network and the long short-term memory network according to the screened feature set until the predetermined performance index is reached, the accuracy rate and the loss rate meet the threshold, and the training and perfecting model is obtained; The audio classification submodule classifies the input respiratory sounds based on the trained and improved model. If the model determines that the audio features match the lung sound features, it is classified as lung sound, otherwise it is classified as heart sound, and respiratory sound separation data is constructed.

7. The intelligent respiratory sound analysis and diagnosis system according to claim 1, characterized in that: The disease identification module includes a feature matching submodule, a disease analysis submodule, and a diagnosis result generation submodule; The feature matching submodule performs a matching process with a known disease feature library based on the microvariation index set, and generates a feature matching result by comparing the microvariation index set with the features of the disease in the feature library; The disease analysis submodule analyzes the correlation between the matched disease features and the patient's respiratory sound data according to the feature matching results, and screens out potential respiratory diseases by calculating similarity scores to obtain potential disease analysis results; The prediction result generation submodule constructs a disease prediction result based on the potential disease analysis results, the patient's microvariation index and the disease feature matching degree, including the potential disease name.

8. The intelligent respiratory sound analysis and diagnosis system according to claim 1, characterized in that: The data integration module includes a result summary submodule, a formatting processing submodule, and an overview construction submodule; The result summary submodule collects and summarizes all relevant health data, including potential respiratory and cardiac disease markers, based on the disease prediction results to generate a summary data set; The formatting processing submodule formats the data according to the summary data set to ensure the clarity and readability of the information, and obtains formatted data by standardizing the display format of the data; The overview building submodule comprehensively analyzes and organizes information based on formatted data, obtains key health indicators and assesses disease risks, and generates a comprehensive health data overview.