Acute respiratory failure detection device based on oronasal airflow signals

Abstract

The application provides an acute respiratory failure detection device based on oral-nasal airflow signals. It is applied to the medical and health field. The operation of the device includes: obtaining the airflow signals of the oral-nasal of the measured person; extracting the time-frequency domain features, time domain features and frequency domain features of the airflow signals respectively; inputting the extracted time-frequency domain features, time domain features and frequency domain features into the pre-trained feature extraction neural network respectively to obtain the deep time-frequency domain features, deep time domain features and deep frequency domain features; performing feature fusion on the deep time-frequency domain features, deep time domain features and deep frequency domain features, and inputting them into the pre-trained classification neural network to obtain the dense classification probability; selecting the disease stage category c of the acute respiratory failure with the maximum probability at each time t in the dense classification probability. The application realizes convenient, safe and accurate detection of acute respiratory failure.

Description

Technical Field

[0001] This invention relates to the field of medical devices, and more specifically to an acute respiratory failure detection device based on airflow signals from the mouth and nose. Background Technology

[0002] Acute respiratory failure (ARF) is a highly fatal clinical syndrome characterized by acute impairment of gas exchange between the lungs and blood, leading to hypoxemia and / or hypercapnia. It often requires mechanical ventilation and can cause severe physiological disturbances and multiple organ failure. In the United States, approximately 360,000 cases of ARF occur annually due to various causes, with 36% of these patients dying during hospitalization. Patients with ARF frequently experience complications such as infections and cardiovascular events. These complications can further worsen the patient's condition and increase mortality. Based on the oxygenation index, ARF can be classified as mild, moderate, or severe. Therefore, timely screening and diagnosis of ARF are crucial for the diagnosis and prevention of physiological health, respiratory diseases, heart diseases, and cardiovascular diseases.

[0003] Currently, the main methods for detecting acute respiratory failure (ARF) include chest X-ray / CT scans and arterial blood gas analysis. However, radiological diagnostics such as plain chest X-rays, chest CT scans, pulmonary angiography, and ultrasound, or bronchoscopy and radionuclide ventilation with the assistance of a physician, are not only expensive and carry radiation risks, but are also difficult to perform for critically ill patients with limited mobility, which limits their application to some extent. Arterial blood gas analysis, on the other hand, requires collecting and measuring arterial blood for ARF diagnosis. However, blood gas analysis is affected by various factors such as age, altitude, and oxygen therapy, and the analysis needs to be combined with the specific clinical situation. It also requires highly skilled operators and is difficult to automate. Summary of the Invention

[0004] In view of this, the present invention provides an acute respiratory failure detection device based on nasal and oral airflow signals, comprising: a processor and a memory connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to cause the processor to perform operations including the following:

[0005] Acquire airflow signals from the mouth and nose of the test subject ;

[0006] For the airflow signal Time-frequency domain features, time-domain features, and frequency-domain features are extracted separately to obtain time-frequency domain features, time-domain features, and frequency-domain features;

[0007] The extracted time-frequency domain features, time-domain features, and frequency-domain features are respectively input into a pre-trained feature extraction neural network to obtain deep time-frequency domain features. Deep temporal features and depth frequency domain features Among them, the depth time-frequency domain features Deep temporal features and depth frequency domain features The sampling frequencies are the same;

[0008] The depth time-frequency domain features The depth temporal features and the depth frequency domain features Feature fusion is performed and the results are input into a pre-trained classification neural network to obtain dense classification probabilities. ;

[0009] Select the dense classification probability The disease stage c with the highest probability at each time t is the acute respiratory failure stage.

[0010] Optionally, the time t is determined by the total monitoring duration T of the nasal and oral airflow data and the update frequency of the acute respiratory failure detection results. Calculated.

[0011] Optionally, for the airflow signal Time-frequency domain feature extraction is performed to obtain time-frequency domain features, including:

[0012] For the airflow signal Short-time motion modal analysis was performed to obtain the respiratory time-frequency signal. ;

[0013] Regarding the respiratory time-frequency signal Smoothing along the time dimension yields respiratory motion signals. ;

[0014] The respiratory time-frequency signals respectively and the respiratory motion signal Perform non-overlapping frame segmentation along time t to separate the respiratory motion signal. The segmented non-overlapping frames are summed within each frame to obtain the first time-frequency domain features, and the respiratory time-frequency signal is calculated. The statistics of the segmented non-overlapping frames are used to obtain the second time-frequency domain features.

[0015] Optionally, for the airflow signal Temporal feature extraction is performed to obtain temporal features, including:

[0016] For the airflow signal Standardization processing is performed to obtain airflow signals. ;

[0017] Calculate the airflow signal The envelope, peak value changes, and valley value changes are used to obtain the first time-domain features;

[0018] The airflow signal Non-overlapping frames are segmented along time t, and the statistics of the segmented non-overlapping frames are calculated to obtain the second temporal feature.

[0019] The segmented non-overlapping frames are presented as scatter plots, and the long axis distance, short axis distance, and effective range of each scatter plot are extracted to obtain the third time-domain feature.

[0020] Optionally, the airflow signal can be processed in the following manner. Standardization processing is performed to obtain airflow signals. :

[0021] ;

[0022] in, This is the mean calculation function. This is the function for calculating standard deviation.

[0023] Optionally, for the airflow signal Frequency domain feature extraction is performed to obtain frequency domain features, including:

[0024] For the airflow signal Perform a Fourier transform to obtain the frequency domain distribution signal. ;

[0025] Calculate the frequency domain distribution signal The total energy is obtained as a characteristic. Calculate the frequency domain distribution signal Frequency less than preset frequency threshold The energy value is used to obtain the characteristic. Calculate the frequency domain distribution signal Frequency less than preset frequency threshold The energy value is used to obtain the characteristic. Calculate the frequency domain distribution signal Frequency greater than preset frequency threshold The energy value is used to obtain the characteristic. , the features The features The features The features As the first frequency domain feature;

[0026] Calculate the airflow signal Frequency domain distribution signal of the segmented non-overlapping frames And statistically analyze the frequency domain distribution signal within each frame. The characteristics, including the maximum value, minimum value, total energy value, and distributed energy value, are used to obtain the second frequency domain characteristics.

[0027] Optionally, airflow signals from the subject's mouth and nose can be acquired. ,include:

[0028] Acquire airflow signals from the nasal and oral airflow acquisition device and its sampling frequency ;

[0029] Determine the sampling frequency With expected sampling frequency Are they consistent?

[0030] When the sampling frequency With expected sampling frequency When consistent, the airflow signal Bandpass filtering is performed to obtain the airflow signal. ;

[0031] For the airflow signal Sparse reconstruction was performed to obtain the principal component airflow signal. ;

[0032] For the principal component airflow signal Wavelet denoising is performed to obtain the airflow signal. .

[0033] Optionally, when the sampling frequency With expected sampling frequency When inconsistent, the sampling frequency will be adjusted. With expected sampling frequency Compare;

[0034] like Then for the airflow signal Interpolation is performed to make the sampling frequency With expected sampling frequency To maintain consistency, and for the interpolated airflow signal Bandpass filtering is performed to obtain the airflow signal. ;

[0035] like Then for the airflow signal Perform downsampling to increase the sampling frequency. With expected sampling frequency To maintain consistency, and to process the downsampled airflow signal. Bandpass filtering is performed to obtain the airflow signal. .

[0036] Optionally, for the principal component airflow signal Wavelet denoising is performed to obtain the airflow signal. ,include:

[0037] The principal component signal is decomposed into w-level wavelet coefficients swa and swd using the wavelet kernel function wname.

[0038] The detailed data in the wavelet coefficients swd are denoised using soft thresholding to obtain the denoised wavelet detail coefficients swdn;

[0039] The airflow signal is obtained by reconstructing the signal using the wavelet coefficients sw and the denoised wavelet detail coefficients swdn. ;

[0040] Specifically, soft-threshold denoising is performed on the detail data in the wavelet coefficients SWD using the following method:

[0041] ;

[0042] in, For denoising wavelet detail coefficients SWDN, For wavelet coefficients swd, For the threshold, , Standard deviation This represents the data length.

[0043] This invention acquires airflow signals Comprehensive extraction of time-frequency domain features, time-domain features, and frequency-domain features is performed separately. This multi-dimensional feature extraction approach not only covers the basic attributes of the signal but also increases the diversity and information content of features through feature operations, providing rich information for subsequent high-level feature learning. Then, a pre-trained feature extraction neural network is used to perform deep learning and feature transformation on the extracted features, resulting in deep time-frequency domain features, deep time-domain features, and deep frequency-domain features. These deep features not only retain the key information of the original signal but also extract more discriminative and representative high-level feature representations through the nonlinear transformation of the neural network. Next, the three types of deep features are fused and input into a pre-trained classification neural network for classification. This fully utilizes the complementary information of multiple features, increasing the dimensionality and information content of the features, thereby improving the accuracy and robustness of the classification results. Finally, by selecting the class with the highest probability at each time point in the dense classification probability, real-time identification of the acute respiratory failure symptom stage is achieved. This step not only simplifies the processing flow of classification results but also improves the efficiency and practicality of symptom identification.

[0044] This invention detects acute respiratory failure based on airflow from the mouth and nose of the test subject. Compared to traditional CT scans and ultrasound, the data acquisition equipment of this invention is inexpensive, significantly reducing the testing cost for the test subject. By automatically analyzing the collected airflow data from the mouth and nose, this invention can directly obtain the test results without the need for additional manpower for data analysis, thereby further simplifying the operation process, lowering the barrier to entry for the equipment, and making it easier to widely promote. In addition, this invention has no negative impact on the health of the test subject, avoids potential hazards such as radiation, and ensures the comfort of the test subject, especially providing a more convenient and safer testing method for critically ill patients with limited mobility. Attached Figure Description

[0045] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0046] Figure 1 This is an application scenario diagram of the acute respiratory failure detection device based on oral and nasal airflow signals in an embodiment of the present invention;

[0047] Figure 2 This is a flowchart illustrating the acute respiratory failure detection method based on nasal and oral airflow signals in an embodiment of the present invention. Detailed Implementation

[0048] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] like Figure 1 As shown in the figure, an application scenario of an acute respiratory failure detection device based on oral and nasal airflow signals provided by an embodiment of the present invention is as follows:

[0051] This invention consists of a device and a cloud platform. The device is a nasal and oral airflow acquisition device, which is worn in a suitable position on the person being tested. The device acquires the nasal and oral airflow data of the person being tested and transmits the data to a processor in the cloud platform for processing via a transmission module. Figure 1 In this process, the nasal and oral airflow collection device 11 is worn over the mouth and nose of the person being tested. The nasal and oral airflow data is transmitted to the processor in the cloud 13 via the data transmission device 12 for data processing. After accumulating nasal and oral airflow data for a certain period of time, the algorithm analyzes all the received nasal and oral airflow data to obtain the detection result of acute respiratory failure and the disease severity classification result. The data transmission methods of the data transmission device 12 include, but are not limited to, WiFi network transmission, 4G wireless transmission, Bluetooth wireless transmission, and wired transmission.

[0052] like Figure 2 As shown, this embodiment of the invention provides a method for detecting acute respiratory failure based on airflow signals from the mouth and nose. This method can be executed by electronic devices such as computers or servers, specifically the aforementioned cloud platform 13. The operations performed include:

[0053] S1, Acquire airflow signals from the subject's mouth and nose. .

[0054] S2, for airflow signal Time-frequency domain features, time-domain features, and frequency-domain features are extracted separately to obtain time-frequency domain features, time-domain features, and frequency-domain features. The extracted features are not limited to these three types; more dimensional features can be obtained by performing operations on the above features.

[0055] S3, the extracted time-frequency domain features, time domain features, and frequency domain features are input into a pre-trained feature extraction neural network to obtain deep time-frequency domain features. Deep temporal features and depth frequency domain features Among them, the depth time-frequency domain features Deep temporal features and depth frequency domain features The sampling frequencies are the same. There are no requirements for the structure of the feature extraction neural network, as long as the extracted depth-time-frequency features are of the same quality. Deep temporal features and depth frequency domain features The sampling frequency should be the same.

[0056] By using frequency domain features, time domain features, and time-frequency domain features as input to a feature extraction neural network, advanced feature representations can be learned. These features can better reflect the structure and dynamic characteristics of airflow signals, providing more discriminative feature representations for subsequent classification tasks.

[0057] S4, depth time-frequency domain features Deep temporal features and depth frequency domain features Feature fusion is performed and the results are input into a pre-trained classification neural network to obtain dense classification probabilities. .

[0058] By fusing the three types of features and inputting them into a pre-trained classification neural network, we can comprehensively utilize the information from multiple features to extract higher-level feature representations, increase the dimensionality and information content of the features, and improve the accuracy of classification results for acute respiratory failure stages.

[0059] S5, Selecting Dense Classification Probabilities The disease stage c with the highest probability at each time t is the acute respiratory failure stage.

[0060] The time t is composed of the total monitoring duration T of the nasal and oral airflow data and the update frequency of the acute respiratory failure detection results. Calculated.

[0061] For example, a doctor decides to monitor the airflow data from the nose and mouth of a subject for 48 consecutive hours, with a total monitoring time T=48. Within this 48-hour monitoring period, the doctor wants to obtain the subject's acute respiratory failure test results hourly to adjust the treatment plan promptly. What is the update frequency? That is, once per hour. Therefore, time t is the exact moment of each hour, for example, at 8:00, the dense classification probability is obtained based on data processing. Dense classification probability If t=8:00, c: 30% asymptomatic, 20% mild, 80% moderate, and 9% severe, then the final test result is that at 8 o'clock, the person being tested was in the moderate stage of acute respiratory failure.

[0062] This invention embodiment acquires airflow signals Comprehensive extraction of time-frequency domain features, time-domain features, and frequency-domain features is performed separately. This multi-dimensional feature extraction approach not only covers the basic attributes of the signal but also increases the diversity and information content of features through feature operations, providing rich information for subsequent high-level feature learning. Then, a pre-trained feature extraction neural network is used to perform deep learning and feature transformation on the extracted features, resulting in deep time-frequency domain features, deep time-domain features, and deep frequency-domain features. These deep features not only retain the key information of the original signal but also extract more discriminative and representative high-level feature representations through the nonlinear transformation of the neural network. Next, the three types of deep features are fused and input into a pre-trained classification neural network for classification. This fully utilizes the complementary information of multiple features, increasing the dimensionality and information content of the features, thereby improving the accuracy and robustness of the classification results. Finally, by selecting the class with the highest probability at each time point in the dense classification probability, real-time identification of the acute respiratory failure symptom stage is achieved. This step not only simplifies the processing flow of classification results but also improves the efficiency and practicality of symptom identification.

[0063] This invention detects acute respiratory failure based on airflow from the mouth and nose of the test subject. Compared to traditional CT scans and ultrasound, the data acquisition equipment of this invention is inexpensive, significantly reducing the testing cost for the test subject. By automatically analyzing the collected airflow data from the mouth and nose, this invention can directly obtain the test results without the need for additional manpower for data analysis, thereby further simplifying the operation process, lowering the barrier to entry for the equipment, and making it easier to widely promote. In addition, this invention has no negative impact on the health of the test subject, avoids potential hazards such as radiation, and ensures the comfort of the test subject, especially providing a more convenient and safer testing method for critically ill patients with limited mobility.

[0064] In one embodiment, the airflow signal in step S2 Time-frequency domain feature extraction is performed to obtain time-frequency domain features, including:

[0065] S211, for airflow signal Short-time motion modal analysis was performed to obtain the respiratory time-frequency signal. Short-time motion mode analysis methods include, but are not limited to, Fast Fourier Transform and Wavelet Transform. f is the signal frequency.

[0066] S212, for respiratory time-frequency signals Smoothing along the time dimension yields respiratory motion signals. .

[0067] S213, respectively, the respiratory time-frequency signals and respiratory movement signals Perform non-overlapping frame segmentation along time t to separate the respiratory motion signal. The segmented non-overlapping frames are summed within each frame to obtain the first time-frequency domain features, and the respiratory time-frequency signal is calculated. The statistics of the segmented non-overlapping frames are used to obtain the second time-frequency domain features. The statistics include, but are not limited to, the maximum value, minimum value, median, and standard deviation.

[0068] Specifically, the signal and signal When performing frame segmentation, for example, two signals are divided into several non-overlapping time segments along time t. Each time segment is called a frame, and the length is ∆t. This applies to respiratory motion signals. The sum of signal values ​​corresponding to all frequencies f and time t within each frame is calculated to obtain the first time-frequency domain feature; for respiratory time-frequency signals... The maximum, minimum, median, and standard deviation of the signal values ​​corresponding to all frequencies f and time t within each frame are calculated to obtain the second time-frequency domain features.

[0069] This embodiment extracts airflow signals. The time-frequency domain characteristics enable comprehensive and in-depth analysis of airflow signals in both time and frequency dimensions, greatly enriching the analytical content of airflow signals. Through automated feature extraction algorithms, a large number of time-frequency domain features can be obtained quickly and accurately, focusing on both the dynamic changes of the signal in the time dimension and the spectral characteristics of the signal in the frequency dimension, achieving comprehensive coverage of signal characteristics. This balanced analysis approach helps improve the accuracy and reliability of subsequent acute respiratory failure classification, providing stronger support for clinical diagnosis and treatment.

[0070] In one embodiment, the airflow signal in step S2 Temporal feature extraction is performed to obtain temporal features, including:

[0071] S221, for airflow signal Standardization processing is performed to obtain airflow signals. .

[0072] S222, Calculate airflow signal The envelope, peak value variation, and trough value variation are used to obtain the first time-domain feature. The calculation of the first time-domain feature is not limited to the envelope, peak value variation, and trough value variation.

[0073] S223, airflow signal Non-overlapping frames are segmented along time t, and statistics of the segmented non-overlapping frames are calculated to obtain the second time-domain feature. The statistics include, but are not limited to, maximum, minimum, median, and standard deviation.

[0074] S224: The segmented non-overlapping frames are presented as scatter plots, and the long axis distance, short axis distance, and effective range of each scatter plot are extracted to obtain the third temporal feature. The extraction of the third temporal feature is not limited to the long axis distance, short axis distance, and effective range.

[0075] Specifically, the airflow signal When performing frame segmentation, for example, when dividing airflow signals... The time interval t is divided into several non-overlapping time segments, each called a frame, with a length of ∆t. The maximum, minimum, median, and standard deviation of the signal values ​​at all times t within each frame are calculated to obtain the second time-domain feature.

[0076] The peak points, zero-crossing points, or estimated points of the signal within each segmented frame are extracted. For example, for the k-th frame signal, the interbreathing interval signal is extracted as R(n), n=1,2,…,N. A scatter plot is constructed using N-1 coordinate values ​​of [R(k), R(k+1)], k=1,2,…,N-1. Then, the major axis distance, minor axis distance, and effective range of the scatter plot are calculated until the axis distance features of the scatter plots of all frames are extracted, resulting in the third time-domain feature. The major axis distance and minor axis distance are the longest and shortest distances measured at approximately 45 degrees and -45 degrees in the scatter plot, respectively. The effective range is the range formed by all points in the scatter plot.

[0077] This embodiment extracts airflow signals. The time-domain characteristics can directly describe the changes of the signal over time, and can intuitively reflect the temporal relationship of the signal. The short-term and long-term dynamic changes of the signal can be captured through the time-domain scatter plot, highlighting the detailed signal and accurately locating abnormal moments, which helps to improve the accuracy and reliability of subsequent acute respiratory failure classification.

[0078] In step S221, the airflow signal is processed in the following manner. Standardization processing is performed to obtain airflow signals. :

[0079] ;

[0080] in, This is the mean calculation function. This is the function for calculating standard deviation.

[0081] In one embodiment, the airflow signal in step S2 Frequency domain feature extraction is performed to obtain frequency domain features, including:

[0082] S231, for airflow signal Perform a Fourier transform to obtain the frequency domain distribution signal. .

[0083] S232, Calculate the frequency domain distribution signal The total energy is obtained as a characteristic. Calculate the frequency domain distribution signal Frequency less than preset frequency threshold The energy value is used to obtain the characteristic. Calculate the frequency domain distribution signal Frequency less than preset frequency threshold The energy value is used to obtain the characteristic. Calculate the frequency domain distribution signal Frequency greater than preset frequency threshold The energy value is used to obtain the characteristic. , will feature ,feature ,feature ,feature As the first frequency domain feature.

[0084] Specifically, preset frequency threshold For ultra-low frequencies, a preset frequency threshold is used. For low frequencies, a preset frequency threshold is used. For high frequencies, the three frequencies can be set according to the airflow signal characteristics of acute respiratory failure. Features It is a frequency domain distributed signal The sum of energy at all frequencies, such as a frequency domain distributed signal. If the value exists within the frequency range of 0-100Hz, and the energy distribution at these frequencies is known, then the characteristic... It is the total energy across the 0-100Hz frequency range. Characteristics It is a frequency domain distributed signal When the frequency is less than the preset frequency threshold The sum of energy on a portion of the surface, assuming =3Hz, then the characteristic It is the frequency domain distributed signal The total energy within the frequency range of 0-3Hz. Characteristics It is a frequency domain distributed signal When the frequency is less than the preset frequency threshold The sum of energy on a portion of the surface, assuming =15Hz, then the characteristic It is the frequency domain distributed signal The sum of energy within the frequency range of 3-15Hz. Characteristics It is a frequency domain distributed signal When the frequency is greater than the preset frequency threshold The total energy, assuming =50Hz, then the characteristic It is the frequency domain distributed signal The total energy within the frequency range of 50-100Hz.

[0085] S233, Calculate airflow signal Frequency domain distribution signal of the segmented non-overlapping frames And statistically analyze the frequency domain distribution signal within each frame. The characteristics, including the maximum value, minimum value, total energy value, and distributed energy value, are used to obtain the second frequency domain characteristics.

[0086] Specifically, the airflow signal When performing frame segmentation, for example, when dividing airflow signals... Divide the signal along time t into several non-overlapping time segments, each time segment is called a frame, and the length is ∆t. Calculate the frequency domain distribution of the signal within each frame. Then, the frequency domain distribution signal of each frame is statistically analyzed. The maximum value, minimum value, total energy value, and distributed energy value.

[0087] This embodiment extracts airflow signals. By examining the frequency domain characteristics, we can reveal the main frequency components of a signal, thereby gaining a deeper understanding of its periodicity and frequency characteristics. This process enhances our macroscopic grasp of the overall characteristics of a signal.

[0088] In one embodiment, step S1 involves acquiring the airflow signal from the subject's mouth and nose. ,include:

[0089] S11, Acquire the airflow signal collected by the oral and nasal airflow acquisition device. and its sampling frequency .

[0090] S12, Determine the sampling frequency With expected sampling frequency Whether they are consistent depends on the sampling frequency. With expected sampling frequency If they are consistent, proceed to step S13, when the sampling frequency... With expected sampling frequency If they match, proceed to step S14.

[0091] S13, for airflow signal Bandpass filtering is performed to obtain the airflow signal. Bandpass filtering can be performed using IIR or FIR filters.

[0092] S14, sampling frequency With expected sampling frequency If a comparison is made, If so, then execute step S15 first, then execute step S13. If so, then execute step S16 first, and then execute step S13.

[0093] S15, for airflow signal Interpolation is performed to make the sampling frequency With expected sampling frequency Maintain consistency. Interpolation methods include, but are not limited to, linear interpolation, nearest neighbor interpolation, bilinear interpolation, nearest point interpolation, spline interpolation, and conformal piecewise cubic interpolation.

[0094] S16, for airflow signal Perform downsampling to increase the sampling frequency. With expected sampling frequency Maintain consistency. Downsampling methods include, but are not limited to, linear sampling and sampling based on various interpolation methods.

[0095] This embodiment uses airflow signals Performing resampling operations to keep the sampling frequency consistent with the expected sampling frequency can achieve data alignment, maintain feature integrity, and improve the versatility of subsequent signal processing algorithms.

[0096] Furthermore, after performing step S13, the process also includes:

[0097] S17, for airflow signal Sparse reconstruction was performed to obtain the principal component airflow signal. Specifically, this includes:

[0098] S171, the airflow signal is transmitted in the following manner Rearranged as Data matrix :

[0099] ;

[0100] in, Indicates airflow signal The length.

[0101] S172, regarding the data matrix Principal component extraction was performed, and the specific implementation method is as follows:

[0102] First, the data matrix is ​​processed in the following way. Perform singular value decomposition:

[0103]

[0104] in, The singular value decomposition function is a well-known and publicly disclosed method in this field. It is a left singular value matrix. It is a right singular value matrix. and All are unitary matrices. Let the singular value matrix have eigenvalues ​​as its diagonal elements. The set of eigenvalues ​​is defined as follows: Then, the singular value matrix is ​​extracted using the following method. Eigenvalues ​​in:

[0105] ;

[0106] Then, select the k largest eigenvalues ​​using the following method. Constructing principal component feature sets This ensures that the sum of its eigenvalues ​​accounts for a proportion greater than the eigenvalue threshold. ,:

[0107] ;

[0108] Next, extract the principal component feature set. The corresponding k sets of feature vectors are then sparsely reconstructed using the following method to obtain the principal component data matrix. :

[0109] ;

[0110] Finally, the principal component data matrix was converted into principal component airflow signals using the diagonal averaging method. :

[0111] ;

[0112] The diagonal averaging method recovers the signal using the following formula:

[0113] ;

[0114] in, This represents the independent variable in the summation symbol.

[0115] This embodiment uses airflow signals Sparse reconstruction, by extracting eigenvalues ​​from the singular value matrix and selecting the k largest eigenvalues, can effectively reduce the redundancy of data information. By selecting appropriate eigenvalue thresholds and the corresponding number of eigenvalues, the most important features can be retained while discarding features with less impact on the data. By extracting eigenvectors corresponding to the selected eigenvalues, principal component data matrices can be obtained. These eigenvectors represent important patterns, structures, or features in the original data. By averaging each diagonal data set of the principal component data matrix using the diagonal averaging method, a set of synthesized signals can be obtained. These signals represent the most significant components in the original data and can be used for further analysis, identification, or visualization.

[0116] Furthermore, after performing step S17, the process also includes:

[0117] S18, for the principal component airflow signal Wavelet denoising is performed to obtain the airflow signal. Specifically, this includes:

[0118] S181, the principal component data signal is decomposed into w-level wavelet using the wavelet kernel function wname, as shown in the following formula. The wavelet kernel function includes, but is not limited to, the db-system wavelet kernel, the sym-system wavelet kernel, the bior wavelet kernel, and the rbio-system wavelet kernel.

[0119] ;

[0120] S182, perform soft-threshold denoising on the detail data in the wavelet coefficients swd to obtain the denoised wavelet detail coefficients swdn. The threshold function used is shown in the following formula:

[0121] ;

[0122] in, For denoising wavelet detail coefficients SWDN, For wavelet coefficients swd, For the threshold, , Standard deviation This represents the data length.

[0123] S183, using wavelet coefficients swa and denoised wavelet detail coefficients swdn, is used to recover the airflow signal. The specific formula is as follows:

[0124]

[0125] This embodiment uses wavelet decomposition to analyze the principal component airflow signal. By performing w-level wavelet decomposition, the components of a signal within different frequency ranges can be obtained to reflect local changes and characteristics of the signal, and corresponding coefficients can be acquired. The energy and statistical information of these coefficients are then used to denoise the signal, removing noise interference and improving signal quality. By fusing the denoised signals, information from different frequency ranges can be integrated, which helps extract important features from the signal, reduces noise interference, improves the signal-to-noise ratio, and makes the signal more identifiable. Using soft thresholding for denoising can remove noise without blurring signal edges, preserving more useful signals, which is crucial for retaining important signal features. The soft thresholding method proposed in this invention is more suitable for lower frequency signals, further removing difficult-to-identify noise while preserving respiratory signal components as much as possible.

[0126] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0128] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0130] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. An acute respiratory failure detection device based on airflow signals from the mouth and nose, characterized in that, include: A processor and a memory connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to cause the processor to perform operations including: Acquire airflow signals from the mouth and nose of the test subject ; For the airflow signal Time-frequency domain features, time-domain features, and frequency-domain features are extracted separately to obtain time-frequency domain features, time-domain features, and frequency-domain features; For the airflow signal Time-frequency domain feature extraction includes: extracting the airflow signal Short-time motion modal analysis was performed to obtain the respiratory time-frequency signal. ; Regarding the respiratory time-frequency signal Smoothing along the time dimension yields respiratory motion signals. ; respectively the respiratory time-frequency signals and the respiratory motion signal Perform non-overlapping frame segmentation along time t to separate the respiratory motion signal. The segmented non-overlapping frames are summed within each frame to obtain the first time-frequency domain features, and the respiratory time-frequency signal is calculated. The statistics of the segmented non-overlapping frames are used to obtain the second time-frequency domain features, while the first time-frequency domain features include respiratory motion signals. The sum of signal values ​​corresponding to all frequencies f and times t within each non-overlapping frame, and the second time-frequency domain feature including the respiratory time-frequency signal. The statistics of the signal values ​​corresponding to all frequencies f and time t within each non-overlapping frame; For the airflow signal Perform time-domain feature extraction, including: extracting the airflow signal Standardization processing is performed to obtain airflow signals. ; Calculate the airflow signal The envelope, peak value changes, and valley value changes of the airflow signal are used to obtain the first time-domain feature; the airflow signal is then... Non-overlapping frames are segmented along time t, and the statistics of the segmented non-overlapping frames are calculated to obtain the second time-domain feature; the segmented non-overlapping frames are presented as scatter plots, and the long axis distance, short axis distance, and effective range of each scatter plot are extracted to obtain the third time-domain feature. The extracted time-frequency domain features, time-domain features, and frequency-domain features are respectively input into a pre-trained feature extraction neural network to obtain deep time-frequency domain features. Deep temporal features and depth frequency domain features Among them, the depth time-frequency domain features Deep temporal features and depth frequency domain features The sampling frequencies are the same; The depth time-frequency domain features The depth temporal features and the depth frequency domain features Feature fusion is performed and the results are input into a pre-trained classification neural network to obtain dense classification probabilities. ; Select the dense classification probability The disease stage c with the highest probability at each time t is the acute respiratory failure stage.

2. The device according to claim 1, characterized in that, The time t is determined by the total monitoring duration T of the nasal and oral airflow data and the update frequency of the acute respiratory failure detection results. Calculated.

3. The device according to claim 1, characterized in that, The airflow signal is processed in the following manner Standardization processing is performed to obtain airflow signals. : ， in, This is the mean calculation function. This is the function for calculating standard deviation.

4. The device according to claim 1, characterized in that, For the airflow signal Frequency domain feature extraction is performed to obtain frequency domain features, including: For the airflow signal Perform a Fourier transform to obtain the frequency domain distribution signal. ; Calculate the frequency domain distribution signal The total energy is obtained as a characteristic. Calculate the frequency domain distribution signal Frequency less than preset frequency threshold The energy value is used to obtain the characteristic. Calculate the frequency domain distribution signal Frequency less than preset frequency threshold The energy value is used to obtain the characteristic. Calculate the frequency domain distribution signal Frequency greater than preset frequency threshold The energy value is used to obtain the characteristic. Among them, the preset frequency threshold <Preset frequency threshold <Preset frequency threshold ,feature It is a frequency domain distributed signal The sum of energy at all frequencies, characteristics It is a frequency domain distributed signal When the frequency is less than the preset frequency threshold The total energy and characteristics of the parts It is a frequency domain distributed signal When the frequency is less than the preset frequency threshold The total energy and characteristics of the parts It is a frequency domain distributed signal When the frequency is greater than the preset frequency threshold The total energy; The feature The features The features The features As the first frequency domain feature; Calculate the airflow signal Frequency domain distribution signal of the segmented non-overlapping frames And statistically analyze the frequency domain distribution signal within each frame. The characteristics, including the maximum value, minimum value, total energy value, and distributed energy value, are used to obtain the second frequency domain characteristics.

5. The device according to claim 1, characterized in that, Acquire airflow signals from the mouth and nose of the test subject ,include: Acquire airflow signals from the nasal and oral airflow acquisition device and its sampling frequency ; Determine the sampling frequency With expected sampling frequency Are they consistent? When the sampling frequency With expected sampling frequency When consistent, the airflow signal Bandpass filtering is performed to obtain the airflow signal. ; For the airflow signal Sparse reconstruction was performed to obtain the principal component airflow signal. ; For the principal component airflow signal Wavelet denoising is performed to obtain the airflow signal. .

6. The device according to claim 5, characterized in that, When the sampling frequency With expected sampling frequency When inconsistent, the sampling frequency will be adjusted. With expected sampling frequency Compare; like Then for the airflow signal Interpolation is performed to make the sampling frequency With expected sampling frequency To maintain consistency, and for the interpolated airflow signal Bandpass filtering is performed to obtain the airflow signal. ; like Then for the airflow signal Perform downsampling to increase the sampling frequency. With expected sampling frequency To maintain consistency, and to process the downsampled airflow signal. Bandpass filtering is performed to obtain the airflow signal. .

7. The device according to claim 5, characterized in that, For the principal component airflow signal Wavelet denoising is performed to obtain the airflow signal. ,include: The principal component signal is decomposed into w-level wavelet coefficients swa and swd using the wavelet kernel function wname. The detailed data in the wavelet coefficients swd are denoised using soft thresholding to obtain the denoised wavelet detail coefficients swdn; The airflow signal is obtained by reconstructing the signal using the wavelet coefficients sw and the denoised wavelet detail coefficients swdn. ; Specifically, soft-threshold denoising is performed on the detail data in the wavelet coefficients SWD using the following method: ， in, For denoising wavelet detail coefficients SWDN, For wavelet coefficients swd, For the threshold, , Standard deviation This represents the data length.

Images