Acute respiratory failure detection equipment based on mouth-nose airflow signal

Abstract

The invention provides acute respiratory failure detection equipment based on a mouth-nose airflow signal. Application in the field of medical health. The operation executed by the equipment comprises the following steps: acquiring an airflow signal # imgabs0 # of the mouth and nose of a tested person; performing time-frequency domain feature extraction, time-domain feature extraction and frequency-domain feature extraction on the airflow signal # imgabs1 # to obtain time-frequency domain features, time-domain features and frequency-domain features; respectively inputting the extracted time-frequency domain feature, the time domain feature and the frequency domain feature into a pre-trained feature extraction neural network to obtain a depth time-frequency domain feature # imgabs2 #, a depth time domain feature # imgabs3 # and a depth frequency domain feature # imgabs4 #; performing feature fusion on the depth time-frequency domain feature # imgabs5 #, the depth time-domain feature # imgabs6 # and the depth frequency-domain feature # imgabs7 #, and inputting a pre-trained classification neural network to obtain a dense classification probability # imgabs8 #; and selecting the disease stage category c of the acute respiratory failure with the maximum probability at each time t in the dense classification probability # imgabs9 #. According to the invention, acute respiratory failure detection can be carried out conveniently, safely and accurately.

Description

Technical Field

[0001] The present invention relates to the field of medical equipment, and in particular to an acute respiratory failure detection device based on oral and nasal airflow signals. Background Art

[0002] Acute respiratory failure (ARF) is a clinical syndrome with a high mortality rate. Its basic feature is the acute impairment of gas exchange between the lungs and the blood, leading to hypoxemia and / or hypercapnia. Mechanical ventilation support is often required and may lead to severe physiological disorders and multiple organ failure. In the United States, there are approximately 360,000 cases of acute respiratory failure due to various reasons each year, of which 36% of patients die during hospitalization. At the same time, patients with acute respiratory failure are often accompanied by complications such as infection and cardiovascular events. These complications may further aggravate the patient's condition and increase the mortality rate. According to the oxygenation index, ARF can be divided into mild, moderate and severe cases. Therefore, timely screening and diagnosis of ARF is of vital importance for the diagnosis and prevention of diseases such as human physiological health, respiratory diseases, heart diseases, cardiovascular and cerebrovascular diseases, etc.

[0003] At present, the detection methods of acute respiratory failure mainly include ARF detection based on lung X-ray / CT examination and arterial blood gas analysis. However, such as ordinary chest X-ray, chest CT and pulmonary angiography and ultrasound examination, or fiber bronchoscopy and radionuclide pulmonary ventilation perfusion with the help of a doctor. These radiological diagnoses are not only expensive but also accompanied by radiation risks. For critically ill patients with limited mobility, it is difficult to shoot, which limits its application to a certain extent. Based on arterial blood gas analysis, it is necessary to collect and measure the patient's arterial blood to diagnose ARF. However, blood gas will be affected by many factors such as age, altitude, oxygen therapy, etc. The analysis needs to be combined with the specific clinical situation, and the operator needs to be proficient in it and have a high level of professional literacy, and it is difficult to achieve automated monitoring. Summary of the invention

[0004] In view of this, the present invention provides, on one hand, an acute respiratory failure detection device based on oral and nasal airflow signals, comprising: a processor and a memory connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to perform operations including the following contents: Obtain the airflow signal from the mouth and nose of the person being tested ; The airflow signal Extracting time-frequency domain features, time domain features, and frequency domain features respectively to obtain time-frequency domain features, time domain features, and frequency domain features; The extracted time-frequency domain features, time domain features, and frequency domain features are respectively input into the pre-trained feature extraction neural network to obtain the deep time-frequency domain features. , deep temporal features and deep frequency domain features , where the deep time-frequency domain features , deep temporal features and deep frequency domain features The sampling frequency is the same; The deep time-frequency domain features , the deep temporal features And the deep frequency domain features Perform feature fusion and input into the pre-trained classification neural network to obtain dense classification probability ; Select the dense classification probability The symptom stage category c of acute respiratory failure with the highest probability at each time t.

[0005] Optionally, the time t is determined by the total monitoring time T of the oral and nasal airflow data and the update frequency of the acute respiratory failure detection result. Calculated.

[0006] Optionally, the airflow signal Perform time-frequency domain feature extraction to obtain time-frequency domain features, including: The airflow signal Perform short-term motion modal analysis to obtain respiratory time-frequency signals ; The respiratory time-frequency signal Smoothing along the time dimension to obtain the respiratory motion signal ; The respiratory time-frequency signal and the respiratory motion signal Segment the respiratory motion signal by non-overlapping frames along time t The segmented non-overlapping frames are summed within each frame to obtain the first time-frequency domain feature and calculate the respiratory time-frequency signal The statistics of the segmented non-overlapping frames are used to obtain the second time-frequency domain features.

[0007] Optionally, the airflow signal Perform time domain feature extraction to obtain time domain features, including: The airflow signal Perform standardization to obtain airflow signal ; Calculate the airflow signal The envelope, peak value change and valley value change of the , and the first time domain feature is obtained; The airflow signal Perform non-overlapping frame segmentation along time t, and calculate the statistics of the segmented non-overlapping frames to obtain a second time domain feature; The segmented non-overlapping frames are presented as scatter plots respectively, and the major axis distance, minor axis distance and effective range of each of the scatter plots are extracted to obtain a third time domain feature.

[0008] Optionally, the airflow signal is analyzed in the following manner: Perform standardization to obtain airflow signal : ; in, is the mean calculation function, Calculates the standard deviation function.

[0009] Optionally, the airflow signal Perform frequency domain feature extraction to obtain frequency domain features, including: The airflow signal Perform Fourier transform to obtain frequency domain distribution signal ; Calculate the frequency domain distribution signal The total energy of , calculate the frequency domain distribution signal The frequency is less than the preset frequency threshold The energy value of , calculate the frequency domain distribution signal The frequency is less than the preset frequency threshold The energy value of , calculate the frequency domain distribution signal The frequency is greater than the preset frequency threshold The energy value of , the features , the characteristics , the characteristics , the characteristics As the first frequency domain feature; Calculate the airflow signal Frequency domain distribution signal of each non-overlapping frame segmented , and count the frequency domain distribution signal in each frame The features of , including maximum value, minimum value, overall energy value, and distribution energy value, are used to obtain the second frequency domain features.

[0010] Optionally, obtain the airflow signal of the mouth and nose of the person being tested ,include: Obtain airflow signals collected by oral and nasal airflow collection equipment and its sampling frequency ; Determine the sampling frequency The expected sampling frequency Is it consistent? When the sampling frequency The expected sampling frequency When the airflow signal Perform bandpass filtering to obtain the airflow signal ; The airflow signal Perform sparse reconstruction to obtain the main component airflow signal ; The main component airflow signal Perform wavelet denoising to obtain the airflow signal .

[0011] Optionally, when the sampling frequency The expected sampling frequency If they are inconsistent, the sampling frequency The expected sampling frequency Make comparisons; like , then the airflow signal Interpolate so that the sampling frequency The expected sampling frequency Keep consistent, and the airflow signal after interpolation processing Perform bandpass filtering to obtain the airflow signal ; like , then the airflow signal Downsample so that the sampling frequency The expected sampling frequency Keep consistent, and the airflow signal after downsampling Perform bandpass filtering to obtain the airflow signal .

[0012] Optionally, the main component airflow signal Perform wavelet denoising to obtain the airflow signal ,include: Use the wavelet kernel function wname to perform w-layer wavelet decomposition on the principal component signal to obtain the wavelet coefficients swa and swd; Performing soft threshold denoising on the detail data in the wavelet coefficient swd to obtain denoised wavelet detail coefficient swdn; The airflow signal is obtained by restoring the wavelet coefficient swa and the denoising wavelet detail coefficient swdn. ; The following method is used to perform soft threshold denoising on the detail data in the wavelet coefficients swd: ; in, is the denoising wavelet detail coefficient swdn, is the wavelet coefficient swd, is the threshold value, , is the standard deviation, is the data length.

[0013] The present invention obtains the airflow signal by The time-frequency domain features, time domain features and frequency domain features are fully extracted respectively. This multi-dimensional feature extraction method not only covers the basic properties of the signal, but also increases the diversity and information content of the features through feature operations, providing rich information for subsequent advanced feature learning. Then, the extracted features are deeply learned and transformed using the pre-trained feature extraction neural network to obtain 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 can extract more discriminative and representative advanced feature representations through the nonlinear transformation of the neural network. Then, the three types of deep features are fused and input into the pre-trained classification neural network for classification, making full use of the information complementarity of multiple features, increasing the dimension and information content of the features, thereby improving the accuracy and robustness of the classification results. Finally, by selecting the maximum probability category at each time point in the dense classification probability, the real-time recognition of the stage of acute respiratory failure is achieved. This step not only simplifies the processing flow of the classification results, but also improves the efficiency and practicality of symptom recognition.

[0014] The present invention performs acute respiratory failure detection based on the airflow through the mouth and nose of the person being tested. Compared with traditional CT irradiation and ultrasonic detection methods, the data acquisition equipment of the present invention is low-cost, which significantly reduces the detection cost and expenses of the person being tested. By automatically analyzing the collected oral and nasal airflow data, the present invention can directly obtain the test results without the need for additional manpower for data analysis, thereby further simplifying the operating process, lowering the threshold for using the equipment, and making it easier to promote it widely. In addition, the present invention has no negative impact on the health of the person being tested, avoids potential hazards such as radiation, ensures the comfort of the person being tested, and provides a more convenient and safe detection method, especially for critically ill patients with limited mobility. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 This is a diagram of an application scenario of an acute respiratory failure detection device based on oral and nasal airflow signals in an embodiment of the present invention; Figure 2 Schematic diagram of the process of detecting acute respiratory failure based on oral and nasal airflow signals in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] In addition, 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.

[0019] like Figure 1 As shown, 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: The present invention is composed of a device end and a cloud. The device end is an oral and nasal airflow collection device. The oral and nasal airflow collection device is worn at a suitable position of the person being tested. The oral and nasal airflow collection device acquires the oral and nasal airflow data of the person being tested and transmits the data to a cloud processor for processing through a transmission module. Figure 1 In the present invention, the oral and nasal airflow collection device 11 is worn on the mouth and nose of the person being tested, and the oral and nasal airflow data will be transmitted to the processor of the cloud 13 through the data transmission device 12 for data processing. After accumulating oral and nasal airflow data for a certain period of time, the algorithm analyzes all the received oral and nasal airflow data to obtain the detection results of acute respiratory failure and the classification results of the severity of the disease. The data transmission method of the data transmission device 12 includes but is not limited to WiFi network transmission, 4G wireless transmission, Bluetooth wireless transmission, wired transmission, etc.

[0020] like Figure 2 As shown, an embodiment of the present invention provides an acute respiratory failure detection method based on oral and nasal airflow signals. The method can be executed by an electronic device such as a computer or a server, specifically the cloud 13, and the operations performed include: S1, obtain the airflow signal of the mouth and nose of the person being tested .

[0021] S2, airflow signal The time-frequency domain features, time-domain features, and frequency-domain features are extracted respectively to obtain the time-frequency domain features, time-domain features, and frequency-domain features. The extracted features are not limited to these three types of features, and the above features can also be operated to obtain features of more dimensions.

[0022] S3, the extracted time-frequency domain features, time domain features and frequency domain features are respectively input into the pre-trained feature extraction neural network to obtain the deep time-frequency domain features , deep temporal features and deep frequency domain features , where the deep time-frequency domain features , deep temporal features and deep frequency domain features The sampling frequency is the same as that of the feature extraction neural network. There is no requirement for the structure of the feature extraction neural network, as long as the extracted deep time-frequency domain features , deep temporal features and deep frequency domain features The sampling frequency is the same.

[0023] By extracting frequency domain features, time domain features, and time-frequency domain features as input feature extraction neural networks, advanced feature representations can be learned, which can better reflect the structure and dynamic characteristics of airflow signals and provide more discriminative feature representations for subsequent classification tasks.

[0024] S4, deep time-frequency domain features , deep temporal features and deep frequency domain features Perform feature fusion and input into the pre-trained classification neural network to obtain dense classification probability .

[0025] By fusing the three types of features and inputting them into a pre-trained classification neural network, it is possible to comprehensively utilize the information of multiple features to extract more advanced feature representations, increase the dimension and amount of information of the features, and improve the accuracy of the classification results of the acute respiratory failure symptom stages.

[0026] S5, select dense classification probability The symptom stage category c of acute respiratory failure with the highest probability at each time t.

[0027] The time t is determined by the total monitoring time T of the oral and nasal airflow data and the update frequency of the acute respiratory failure test results. Calculated.

[0028] For example, the doctor decides to monitor the oral and nasal airflow data of the tested person for 48 hours continuously, and the total monitoring time is T=48. During the above 48 hours of monitoring time, the doctor hopes to obtain the test results of the tested person's acute respiratory failure every hour so as to adjust the treatment plan in time. Then the update frequency That is once every hour. Therefore, time t is the time point of each hour. For example, at 8:00, the dense classification probability is obtained according to data processing , dense classification probability That is, t=8:00, c: asymptomatic 30%, mild symptoms 20%, moderate symptoms 80%, severe symptoms 9%, so the final test result is that the acute respiratory failure stage of the tested person at 8 o'clock is moderate.

[0029] In the embodiment of the present invention, the airflow signal is obtained by The time-frequency domain features, time domain features and frequency domain features are fully extracted respectively. This multi-dimensional feature extraction method not only covers the basic properties of the signal, but also increases the diversity and information content of the features through feature operations, providing rich information for subsequent advanced feature learning. Then, the extracted features are deeply learned and transformed using the pre-trained feature extraction neural network to obtain 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 can extract more discriminative and representative advanced feature representations through the nonlinear transformation of the neural network. Then, the three types of deep features are fused and input into the pre-trained classification neural network for classification, making full use of the information complementarity of multiple features, increasing the dimension and information content of the features, thereby improving the accuracy and robustness of the classification results. Finally, by selecting the maximum probability category at each time point in the dense classification probability, the real-time recognition of the stage of acute respiratory failure is achieved. This step not only simplifies the processing flow of the classification results, but also improves the efficiency and practicality of symptom recognition.

[0030] The present invention performs acute respiratory failure detection based on the airflow through the mouth and nose of the person being tested. Compared with traditional CT irradiation and ultrasonic detection methods, the data acquisition equipment of the present invention is low-cost, which significantly reduces the detection cost and expenses of the person being tested. By automatically analyzing the collected oral and nasal airflow data, the present invention can directly obtain the test results without the need for additional manpower for data analysis, thereby further simplifying the operating process, lowering the threshold for using the equipment, and making it easier to promote it widely. In addition, the present invention has no negative impact on the health of the person being tested, avoids potential hazards such as radiation, ensures the comfort of the person being tested, and provides a more convenient and safe detection method, especially for critically ill patients with limited mobility.

[0031] In one embodiment, the airflow signal in step S2 Perform time-frequency domain feature extraction to obtain time-frequency domain features, including: S211, air flow signal Perform short-term motion modal analysis to obtain respiratory time-frequency signals The short-term motion modal analysis methods include but are not limited to fast Fourier transform, wavelet transform, etc. f is the signal frequency.

[0032] S212, respiratory time-frequency signal Smoothing along the time dimension to obtain the respiratory motion signal .

[0033] S213, respectively, the respiratory time-frequency signal and respiratory motion signals Segment the respiratory motion signal by non-overlapping frames along time t The segmented non-overlapping frames are summed within each frame to obtain the first time-frequency domain feature and calculate the respiratory time-frequency signal 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 maximum value, minimum value, median, standard deviation, etc.

[0034] Specifically, the signal and signal When performing frame segmentation, for example, two signals are segmented into several non-overlapping time periods along time t, each time period is called a frame, and the length is ∆t. , calculate the sum of the signal values ​​corresponding to all frequencies f and time t in each frame to obtain the first time-frequency domain feature; for the respiratory time-frequency signal , calculate the maximum value, minimum value, median value and standard deviation of the signal values ​​corresponding to all frequencies f and time t in each frame to obtain the second time-frequency domain features.

[0035] In this embodiment, the airflow signal is extracted The time-frequency domain features of the airflow signal can achieve a comprehensive and in-depth analysis of the airflow signal in the time and frequency dimensions, greatly enriching the analysis content of the airflow signal. Through the automated feature extraction algorithm, a large number of time-frequency domain features can be quickly and accurately obtained, 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 the signal characteristics. This balanced analysis method helps to improve the accuracy and reliability of subsequent acute respiratory failure classification and provide more powerful support for clinical diagnosis and treatment.

[0036] In one embodiment, the airflow signal in step S2 Perform time domain feature extraction to obtain time domain features, including: S221, air flow signal Perform standardization to obtain airflow signal .

[0037] S222, calculate airflow signal The envelope, peak value change and valley value change of the first time domain feature are calculated to obtain the first time domain feature. The calculation of the first time domain feature is not limited to the envelope, peak value change and valley value change.

[0038] S223, the airflow signal Non-overlapping frames are segmented along time t, and statistics of the segmented non-overlapping frames are calculated to obtain a second time domain feature. The statistics include but are not limited to maximum value, minimum value, median, standard deviation, etc.

[0039] S224, presenting the segmented non-overlapping frames as scatter plots respectively, and extracting the long wheelbase, short wheelbase and effective range of each scatter plot to obtain a third time domain feature. The third time domain feature extraction is not limited to the long wheelbase, short wheelbase and effective range.

[0040] Specifically, the airflow signal When performing frame segmentation, for example, the airflow signal The time t is divided into several non-overlapping time periods, each of which is called a frame with a length of ∆t. The maximum, minimum, median, and standard deviation of the signal values ​​corresponding to all time t in each frame are calculated to obtain the second time domain feature.

[0041] Extract the peak point, zero crossing point or valuation point of the signal in each segmented frame. For example, for the k-th frame signal, the respiratory interval signal is extracted as R(n), n=1,2,…,N, and a scatter plot is formed with [R(k), R(k+1)], k=1,2,…,N-1 as N-1 coordinate values. Then, the long wheelbase, short wheelbase and effective range of the scatter plot are calculated until the wheelbase features of the scatter plots of all frames are extracted to obtain the third time domain feature, wherein the long wheelbase and short wheelbase are the longest and shortest distances measured in the directions of approximately 45 degrees and -45 degrees of the scatter plot, respectively, and the effective range is the range formed by all points in the scatter plot.

[0042] In this embodiment, the airflow signal is extracted The time domain characteristics can directly describe the changes of signals in time and can intuitively reflect the timing relationship of signals. The time domain scatter plot can capture the short-term and long-term dynamic changes of signals, highlight the detailed signals, and accurately locate the abnormal moment, which helps to improve the accuracy and reliability of subsequent acute respiratory failure classification.

[0043] Among them, step S221 uses the following method to analyze the airflow signal Perform standardization to obtain airflow signal : ; in, is the mean calculation function, Calculates the standard deviation function.

[0044] In one embodiment, the airflow signal in step S2 Perform frequency domain feature extraction to obtain frequency domain features, including: S231, air flow signal Perform Fourier transform to obtain frequency domain distribution signal .

[0045] S232, calculate the frequency domain distribution signal The total energy of , calculate the frequency domain distribution signal The frequency is less than the preset frequency threshold The energy value of , calculate the frequency domain distribution signal The frequency is less than the preset frequency threshold The energy value of , calculate the frequency domain distribution signal The frequency is greater than the preset frequency threshold The energy value of , the feature ,feature ,feature ,feature As the first frequency domain feature.

[0046] Specifically, the preset frequency threshold For ultra-low frequency, preset frequency threshold For low frequency, preset frequency threshold The three frequencies can be set according to the characteristics of the airflow signal of acute respiratory failure. is a frequency domain distributed signal The sum of the energy at all frequencies, i.e. a frequency-domain distributed signal If there are values ​​in the frequency range 0-100 Hz and the energy distribution at these frequencies is known, then the characteristic It is the total energy from 0 to 100 Hz. is a frequency domain distributed signal When the frequency is less than the preset frequency threshold The sum of the energy on the parts of =3Hz, then the characteristic It is the frequency domain distribution signal The sum of the energy in the frequency range 0-3Hz. is a frequency domain distributed signal When the frequency is less than the preset frequency threshold The sum of the energy on the parts of =15Hz, then the characteristic It is the frequency domain distribution signal The sum of the energy in the frequency range 3-15Hz. is a frequency domain distributed signal When the frequency is greater than the preset frequency threshold The total energy of =50Hz, then the characteristic It is the frequency domain distribution signal The sum of the energy in the frequency range 50-100 Hz.

[0047] S233, calculate airflow signal Frequency domain distribution signal of each non-overlapping frame segmented , and count the frequency domain distribution signal in each frame The features of , including maximum value, minimum value, overall energy value, and distribution energy value, are used to obtain the second frequency domain features.

[0048] Specifically, the airflow signal When performing frame segmentation, for example, the airflow signal Divide the time t into several non-overlapping time periods, each of which is called a frame with a length of ∆t. Calculate the frequency domain distribution signal of the signal in each frame. , and then count the frequency domain distribution signals of each frame The maximum value, minimum value, overall energy value, and distributed energy value.

[0049] In this embodiment, the airflow signal is extracted The frequency domain characteristics of the signal can reveal the main frequency components in the signal, and then deeply understand the periodicity and frequency characteristics of the signal. This process strengthens the macroscopic grasp of the comprehensive characteristics of the signal.

[0050] In one embodiment, in step S1, the airflow signal of the mouth and nose of the person being tested is obtained. ,include: S11, obtain the airflow signal collected by the oral and nasal airflow collection device and its sampling frequency .

[0051] S12, determine the sampling frequency The expected sampling frequency Is it consistent? When the sampling frequency The expected sampling frequency If they are consistent, step S13 is executed. When the sampling frequency The expected sampling frequency If they are consistent, execute step S14.

[0052] S13, air flow signal Perform bandpass filtering to obtain the airflow signal Bandpass filtering can use either an IIR filter or a FIR filter.

[0053] S14, the sampling frequency The expected sampling frequency For comparison, if , then execute step S15 first, then execute step S13, if , then execute step S16 first, and then execute step S13.

[0054] S15, air flow signal Interpolate so that the sampling frequency The expected sampling frequency The interpolation methods include, but are not limited to, linear interpolation, nearest neighbor interpolation, bilinear interpolation, adjacent point interpolation, spline interpolation, shape-preserving piecewise cubic interpolation, etc.

[0055] S16, air flow signal Downsample so that the sampling frequency The expected sampling frequency The downsampling methods include but are not limited to linear sampling and sampling based on various interpolation methods.

[0056] In this embodiment, the airflow signal Resampling is performed to make the sampling frequency consistent with the expected sampling frequency, which can achieve data alignment, maintain feature integrity, and improve the versatility of subsequent signal processing algorithms.

[0057] Furthermore, after executing step S13, the method further includes: S17, air flow signal Perform sparse reconstruction to obtain the main component airflow signal . Specifically include: S171, the airflow signal is converted into Rearrange to The data matrix : ; in, Indicates air flow signal Length.

[0058] S172, for the data matrix Perform principal component extraction, the specific implementation method is: First, the data matrix is ​​transformed as follows Perform singular value decomposition:

[0059] in, is a singular value decomposition function, which is a well-known and publicly recognized method in the art. is the left singular value matrix, is the right singular value matrix, and are all unitary matrices, is a singular value matrix, whose diagonal elements are eigenvalues, and the eigenvalue set is defined as Then, the singular value matrix is ​​extracted as follows The eigenvalues ​​in : ; Then, the largest k eigenvalues ​​are selected as follows Constructing the principal component feature set , so that the sum of its eigenvalues ​​accounts for a greater proportion of the total eigenvalues ​​than the eigenvalue threshold , : ; Next, extract the principal component feature set The corresponding k groups of eigenvectors are sparsely reconstructed in the following way to obtain the principal component data matrix : ; Finally, the principal component data matrix is ​​converted into the principal component airflow signal using the diagonal average method : ; Among them, the diagonal averaging method restores the signal according to the following formula: ; in, Represents the independent variable in the summation symbol.

[0060] In this embodiment, the airflow signal Sparse reconstruction is performed by extracting the eigenvalues ​​of the singular value matrix and selecting the largest k eigenvalues, which can effectively reduce the redundancy of data information. By selecting an appropriate eigenvalue threshold and the corresponding number of eigenvalues, the most important features can be retained and features that have little impact on the data can be discarded. By extracting the eigenvectors corresponding to the selected eigenvalues, the principal component data matrix can be obtained. These eigenvectors represent important patterns, structures or features in the original data. By averaging each set of diagonal data in the principal component data matrix using the diagonal averaging method, a set of synthetic 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.

[0061] Furthermore, after executing step S17, the method further includes: S18, for the main component airflow signal Perform wavelet denoising to obtain the airflow signal . Specifically include: S181, use the wavelet kernel function wname to perform w-layer wavelet decomposition on the principal component data signal, as shown in the following formula, where the wavelet kernel function includes but is not limited to db series wavelet kernel, sym series wavelet kernel, bior wavelet kernel, rbio series wavelet kernel, etc.

[0062] ; S182, performing soft threshold denoising on the detail data in the wavelet coefficient swd to obtain denoised wavelet detail coefficient swdn, and the threshold function used is shown in the following formula: ; in, is the denoising wavelet detail coefficient swdn, is the wavelet coefficient swd, is the threshold value, , is the standard deviation, is the data length.

[0063] S183, using the wavelet coefficient swa and the denoising wavelet detail coefficient swdn to restore the airflow signal , the specific formula is as follows:

[0064] In this embodiment, the main component airflow signal is decomposed by wavelet By performing w-layer wavelet decomposition, the components of the signal in different frequency ranges can be obtained to reflect the local changes and characteristics of the signal, and the corresponding coefficients can be obtained. The energy and statistical information of the coefficients are used to reduce the noise of the signal, remove noise interference, and improve the signal quality; by fusing the denoised signal, the information in different frequency ranges can be integrated, which helps to extract the important features in the signal, reduce the interference of noise, improve the signal-to-noise ratio, and make the signal more recognizable. Soft thresholding is used for denoising, and the edge of the signal can be not blurred while removing noise, and more useful signals can be retained, which is very critical for retaining the important features of the signal. The soft thresholding method proposed by the present invention is more suitable for lower frequency signals, and further removes the unrecognizable noise while retaining the components belonging to the respiratory signal as much as possible.

[0065] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented 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.

[0066] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0067] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0069] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. An acute respiratory failure detection device based on oral and nasal airflow signals, characterized in that: include: A processor and a memory connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to perform operations including the following contents: Obtain the airflow signal from the mouth and nose of the person being tested ; The airflow signal Extracting time-frequency domain features, time domain features, and frequency domain features respectively to obtain time-frequency domain features, time domain features, and frequency domain features; The extracted time-frequency domain features, time domain features, and frequency domain features are respectively input into the pre-trained feature extraction neural network to obtain the deep time-frequency domain features. , deep temporal features and deep frequency domain features , where the deep time-frequency domain features , deep temporal features and deep frequency domain features The sampling frequency is the same; The deep time-frequency domain features , the deep temporal features And the deep frequency domain features Perform feature fusion and input into the pre-trained classification neural network to obtain dense classification probability ; Select the dense classification probability The symptom stage category c of acute respiratory failure with the highest probability at each time t.

2. The device according to claim 1, characterized in that The time t is determined by the total monitoring time T of the oral and nasal 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 Perform time-frequency domain feature extraction to obtain time-frequency domain features, including: The airflow signal Perform short-term motion modal analysis to obtain respiratory time-frequency signals ; The respiratory time-frequency signal Smoothing along the time dimension to obtain the respiratory motion signal ; The respiratory time-frequency signal and the respiratory motion signal Segment the respiratory motion signal by non-overlapping frames along time t The segmented non-overlapping frames are summed within each frame to obtain the first time-frequency domain feature and calculate the respiratory time-frequency signal The statistics of the segmented non-overlapping frames are used to obtain the second time-frequency domain features.

4. The device according to claim 1, characterized in that The airflow signal Perform time domain feature extraction to obtain time domain features, including: The airflow signal Perform standardization to obtain airflow signal ; Calculate the airflow signal The envelope, peak value change and valley value change of the , and the first time domain feature is obtained; The airflow signal Perform non-overlapping frame segmentation along time t, and calculate the statistics of the segmented non-overlapping frames to obtain a second time domain feature; The segmented non-overlapping frames are presented as scatter plots respectively, and the major axis distance, minor axis distance and effective range of each of the scatter plots are extracted to obtain a third time domain feature.

5. The device according to claim 4, characterized in that The airflow signal is analyzed by the following method: Perform standardization to obtain airflow signal : ； in, is the mean calculation function, Calculates the standard deviation function.

6. The device according to claim 4, characterized in that The airflow signal Perform frequency domain feature extraction to obtain frequency domain features, including: The airflow signal Perform Fourier transform to obtain frequency domain distribution signal ; Calculate the frequency domain distribution signal The total energy of , calculate the frequency domain distribution signal The frequency is less than the preset frequency threshold The energy value of , calculate the frequency domain distribution signal The frequency is less than the preset frequency threshold The energy value of , calculate the frequency domain distribution signal The frequency is greater than the preset frequency threshold The energy value of , the features , the characteristics , the characteristics , the characteristics As the first frequency domain feature; Calculate the airflow signal Frequency domain distribution signal of each non-overlapping frame segmented , and count the frequency domain distribution signal in each frame The features of , including maximum value, minimum value, overall energy value, and distribution energy value, are used to obtain the second frequency domain features.

7. The device according to claim 1, characterized in that Obtain the airflow signal from the mouth and nose of the person being tested ,include: Obtain airflow signals collected by oral and nasal airflow collection equipment and its sampling frequency ; Determine the sampling frequency The expected sampling frequency Is it consistent? When the sampling frequency The expected sampling frequency When the airflow signal Perform bandpass filtering to obtain the airflow signal ; The airflow signal Perform sparse reconstruction to obtain the main component airflow signal ; The main component airflow signal Perform wavelet denoising to obtain the airflow signal .

8. The device according to claim 7, characterized in that When the sampling frequency The expected sampling frequency If they are inconsistent, the sampling frequency The expected sampling frequency Make comparisons; like , then the airflow signal Interpolate so that the sampling frequency The expected sampling frequency Keep consistent, and the airflow signal after interpolation processing Perform bandpass filtering to obtain the airflow signal ; like , then the airflow signal Downsample so that the sampling frequency The expected sampling frequency Keep consistent, and the airflow signal after downsampling Perform bandpass filtering to obtain the airflow signal .

9. The device according to claim 7, characterized in that The main component airflow signal Perform wavelet denoising to obtain the airflow signal ,include: Use the wavelet kernel function wname to perform w-layer wavelet decomposition on the principal component signal to obtain the wavelet coefficients swa and swd; Performing soft threshold denoising on the detail data in the wavelet coefficient swd to obtain denoised wavelet detail coefficient swdn; The airflow signal is obtained by restoring the wavelet coefficient swa and the denoising wavelet detail coefficient swdn. ; The following method is used to perform soft threshold denoising on the detail data in the wavelet coefficients swd: ； in, is the denoising wavelet detail coefficient swdn, is the wavelet coefficient swd, is the threshold value, , is the standard deviation, is the data length.

Images