Asthma detection method and portable asthma detection device

Abstract

The application discloses a kind of asthma detection method and portable asthma detection equipment, the method includes obtaining user breathing sound;Segmentation after de-noising, and into frequency spectrum by short-time Fourier transform and logarithmic transformation conversion;And frequency spectrum is input to neural network model to obtain wheezing and wheeze sound.Detection equipment includes shell, main control unit is built in shell, bottom is embedded with contact microphone connected with main control unit;Wherein, main control unit is used to control contact microphone to collect breathing sound and utilize the asthma detection method described above to identify wheezing and wheeze sound.The detection method of the application is fast in operation, good in real-time, can realize the timely detection to the asthma condition of patient, and has good distinguishing effect;While detection equipment is portable, simple to operate, easy to carry.

Description

Technical Field

[0001] This invention relates to the field of asthma detection technology, and more specifically to an asthma detection method and a portable asthma detection device. Background Technology

[0002] Asthma is a high-pitched, musical breathing sound caused by narrowing or obstruction of the airways. It is the sound produced by the local vibration caused by the eddies created when airflow passes through narrowed bronchi or bronchioles.

[0003] As a common chronic inflammatory airway disease in children, asthma is showing an increasing incidence rate. However, because children, especially young children, cannot accurately describe their symptoms and triggering factors, and caregivers have limited ability to identify the condition, it is difficult to apply early and effective interventions to prevent the progression of acute exacerbations. Therefore, it is necessary to develop asthma monitoring methods for children to achieve early identification and timely intervention.

[0004] However, current methods for monitoring asthma in children mainly rely on implanted devices to collect patients' expiratory information, thermal imaging of the patient's environment, or wrist breathing tests to diagnose asthma. These methods are often inaccurate and inconvenient to carry. For example:

[0005] 1) Implantable detection methods, such as fixing an implantable respiratory sensor into the airway through minimally invasive surgery, not only have large device size, which does not meet the needs of daily carrying and use, but also, although they can obtain accurate data, there is a risk of foreign body rejection and poor compliance.

[0006] 2) Non-contact monitoring based on environmental thermal imaging relies on high-resolution infrared cameras, which are usually only suitable for medical institutions; at the same time, they are easily affected by environmental temperature fluctuations and have insufficient signal-to-noise ratio of respiratory signals.

[0007] 3) The core algorithm of wrist-worn respiratory monitoring devices relies on multimodal sensor fusion (such as accelerometers and gyroscopes), resulting in high hardware circuit complexity. Furthermore, it indirectly infers respiratory parameters through chest and abdominal movements, making it difficult to capture high-frequency wheezing characteristics.

[0008] In addition, existing algorithms generally employ complex network structures (such as deep residual networks or 3D convolutional networks), resulting in a large number of model parameters and making it difficult to achieve real-time data processing.

[0009] Therefore, how to solve this problem is a problem that urgently needs to be addressed by those skilled in the art. Summary of the Invention

[0010] To at least partially solve the above-mentioned technical problems, the present invention proposes an asthma detection method and a portable asthma detection device for detecting asthma sounds in children.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] Firstly, this application provides a method for asthma detection, the steps of which include:

[0013] Obtain user's breathing sounds;

[0014] After denoising, the segments are divided and transformed into a spectrum using short-time Fourier transform and logarithmic transform;

[0015] The spectrograms are input into a neural network model to obtain wheezing and dyspnea sounds.

[0016] In one optional embodiment, wavelet thresholding is used to filter and denoise the user's breathing sounds.

[0017] In an optional embodiment, the short-time Fourier transform is calculated as follows:

[0018]

[0019] In the formula, t represents the center time point of the current window, f represents the frequency, x(τ) represents the value of the denoised and segmented respiratory signal at τ, h(τ-t) represents the time-shifted window function for truncating the signal segment, centered at t, τ represents the integration time variable, and e -j2πfτ This represents the complex exponential basis functions of the Fourier transform.

[0020] In an optional embodiment, the logarithmic transformation calculation formula is as follows:

[0021]

[0022] In the formula, imf(i,j) represents the spectrum obtained through STFT, i represents the time frame index, and j represents the frequency index.

[0023] In one optional embodiment, the neural network model includes: multiple feature extraction modules and fully connected layers; each feature extraction module includes,

[0024] Basic unit: one or more consecutive convolutional layers;

[0025] Normalized unit: The batch normalization layer and ReLU activation layer immediately following the convolutional layer;

[0026] Downsampling unit: An optional max-pooling layer used to reduce the resolution of the feature map.

[0027] Secondly, this application discloses a portable asthma detection device, including a housing, a main control unit built into the housing, and a contact microphone connected to the main control unit embedded at the bottom; wherein, the main control unit is used to control the contact microphone to collect respiratory sounds and perform asthma detection methods as described above to identify wheezing and whimpering sounds.

[0028] In an optional embodiment, buttons and a display panel are arranged on the top of the housing. The buttons are used to manually control the detection, and the display panel is used to display the recognition results.

[0029] In an optional embodiment, the main control unit is integrated on a circuit board that is connected to a rechargeable lithium battery, a TYPE-C interface, an SD card interface, and a speaker.

[0030] In an optional embodiment, the circuit board also integrates an analog data processing module for amplifying, filtering, and converting the acquired breathing sounds into digital data.

[0031] Thirdly, this application discloses a portable asthma detection device, including a housing, a main control unit and a wireless module built into the housing, and a contact microphone connected to the unit embedded at the bottom; wherein, the main control unit is used to control the contact microphone to collect respiratory sounds and transmit them to a terminal through the wireless module, and the terminal executes the asthma detection method described above to identify wheezing and whimpering sounds.

[0032] The asthma detection method and portable asthma detection device disclosed in this invention can solve the problems of current detection devices being cumbersome to implant and unable to be used in daily life. Compared with the prior art, the advantages of this application include:

[0033] (1) By performing noise reduction, segmentation, short-time Fourier transform and logarithmic transform on the user's breath sounds, the time domain signal is transformed into a spectrum, which can more clearly show the frequency characteristics of the breath sounds; furthermore, the lightweight neural network model specially designed in this application has fewer model parameters, reduces computational complexity, realizes real-time processing of breath sound data, and ensures high detection accuracy, which can meet the needs of real-time asthma monitoring.

[0034] (2) The testing equipment provided in this application is lightweight and compact, easy to operate and use, and is convenient for users to conduct asthma testing anytime and anywhere. It is especially suitable for children with asthma who need long-term monitoring and can effectively assist caregivers in judging the condition. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0036] Figure 1 This is a flowchart of the asthma detection method provided by the present invention;

[0037] Figure 2This is a schematic diagram of the neural network model structure provided by the present invention;

[0038] Figure 3 This is a schematic diagram of the structure of the portable asthma detection device provided by the present invention;

[0039] Figure 4 A three-dimensional view of the portable asthma detection device provided by the present invention;

[0040] Figure 5 A top view of the portable asthma detection device provided by the present invention;

[0041] Figure 6 Top view of the portable asthma detection device provided by the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0044] Example 1:

[0045] This invention first discloses an asthma detection method, aiming to quickly and accurately detect and assess childhood asthma. The steps include:

[0046] Obtain user's breathing sounds;

[0047] After denoising, the segments are divided and transformed into a spectrum using short-time Fourier transform and logarithmic transform;

[0048] The spectrograms are input into a neural network model to obtain wheezing and dyspnea sounds.

[0049] In this embodiment, the asthma detection method mainly includes three parts: signal processing (denoising), feature extraction, and asthma identification; the specific process is as follows. Figure 1 ;

[0050] In one optional embodiment, wavelet thresholding is used for denoising. By setting an appropriate threshold, the effective information is separated from the energy of the noise, thereby removing high-frequency noise from the acquired audio signal. Preferably, the sym8 wavelet is selected, with a wavelet decomposition level of 3, and the decomposition coefficients of each level of the signal are calculated. A threshold is set for each level of coefficients; signals above the threshold are retained, while wavelet coefficients below the threshold are set to 0. Finally, wavelet reconstruction is performed to recover the signal based on the low-frequency coefficients and the high-frequency coefficients of each level, resulting in the denoised audio.

[0051] In one optional embodiment, feature extraction includes: dividing the filtered breath sound into 5-second segments, windowing the signal into frames using a short-time Fourier transform (STFT), and then obtaining its DFT to obtain a spectrogram; the formula is expressed as:

[0052]

[0053] In the formula, t represents the center time point of the current window, f represents the frequency, x(τ) represents the value of the denoised and segmented respiratory signal at τ, h(τ-t) represents the time-shifted window function for truncating the signal segment, centered at t, τ represents the integration time variable, and e -j2πfτ This represents the complex exponential basis functions of the Fourier transform.

[0054] Since the power spectrum generated by directly performing a short-time Fourier transform on the signal results in low power values ​​for most pixels, making them difficult to distinguish, this application performs additional processing on the power spectrum to enhance its features. Specifically, a logarithmic transform is used to make the transformed logarithmic power spectrum contain more information than the original power spectrum. The formula for calculating the logarithmic transform is:

[0055]

[0056] In the formula, imf(i,j) represents the spectrum obtained through STFT, i represents the time frame index, and j represents the frequency index.

[0057] In one alternative embodiment, a neural network model is constructed for asthma identification; including:

[0058] First, a simplified convolutional neural network is constructed. All convolutional layers in the initial part are replaced with pooling layers or removed (this will not affect the model results). In the middle part, some convolutional layers are replaced with pooling layers, using consecutive 2*1 convolutions + 1*2 pooling. In the final part, fully connected layers are retained, the number of convolutional layers and convolutional filters is reduced to lower the parameters, and at least one 3*3 convolutional layer is retained to ensure that the model's test accuracy is not affected.

[0059] The final simplified neural network model includes: multiple feature extraction modules and fully connected layers; each feature extraction module consists of the following layer combination:

[0060] a. Basic unit: one or more consecutive convolutional layers;

[0061] b. Normalized Unit: The batch normalization layer and ReLU activation layer immediately following the convolutional layer;

[0062] c. Downsampling unit: An optional max-pooling layer used to reduce the resolution of the feature map;

[0063] Then, the Softmax function is applied to the output of the fully connected layer to obtain the classification result.

[0064] In one embodiment, the neural network model structure is as follows: Figure 2 Specifically, it includes:

[0065] The input layer is used to receive grayscale images with a size of 1024×256×1.

[0066] The normalization layer is used to normalize the input image.

[0067] The first convolutional layer uses a 2×1 convolutional kernel, has 3 output channels, and a stride of [2,1].

[0068] The second convolutional layer uses a 5×3 convolutional kernel and has 4 output channels;

[0069] First batch normalization layer and first ReLU activation layer;

[0070] The third convolutional layer uses a 2×1 convolutional kernel, has 8 output channels, and a stride of [2,1].

[0071] The first max pooling layer has a pooling window size of 2×2 and a step size of 2.

[0072] The fourth convolutional layer uses a 5×5 convolutional kernel, has 10 output channels, and 1 padding.

[0073] The fifth convolutional layer uses a 5×5 convolutional kernel, has 12 output channels, and 1 padding.

[0074] The sixth convolutional layer uses a 5×5 convolutional kernel, has 14 output channels, and 1 padding.

[0075] The second batch normalization layer and the second ReLU activation layer;

[0076] The second max pooling layer has a pooling window size of 2×2 and a step size of 2.

[0077] The seventh convolutional layer uses a 3×3 convolutional kernel, has 18 output channels, and 1 padding.

[0078] The third batch normalization layer and the third ReLU activation layer;

[0079] The third max pooling layer has a pooling window size of 2×2 and a stride of 2.

[0080] The eighth convolutional layer uses a 3×3 convolutional kernel and has 30 output channels;

[0081] The fourth batch normalization layer and the fourth ReLU activation layer;

[0082] The fourth max pooling layer has a pooling window size of 2×2 and a stride of 2.

[0083] The ninth convolutional layer uses a 5×5 convolutional kernel, has 56 output channels, and padding of 1.

[0084] The tenth convolutional layer uses a 5×5 convolutional kernel, has 72 output channels, and padding of 1.

[0085] The fifth batch normalization layer and the fifth ReLU activation layer;

[0086] The fifth max pooling layer has a pooling window size of 2×2 and a step size of 2.

[0087] The eleventh convolutional layer uses a 5×5 convolutional kernel, has 120 output channels, and padding of 1.

[0088] The twelfth convolutional layer uses a 3×3 convolutional kernel and has 300 output channels;

[0089] The sixth batch normalization layer and the sixth ReLU activation layer;

[0090] A random deactivation layer with a deactivation probability of 0.5;

[0091] Fully connected layer with an output dimension of 4;

[0092] The Softmax layer is used to generate classification results.

[0093] Preferably, all max pooling layers employ either global max pooling or local max pooling, with local max pooling being preferred.

[0094] The network was trained using a pre-constructed dataset of respiratory sounds containing wheezing and stridor. After achieving a high accuracy, the model was used as the final recognition model to distinguish between wheezing and stridor in respiratory sounds.

[0095] In this embodiment, the neural network model has 14.3K parameters, which not only has fast computing speed and good real-time performance, enabling timely detection of patients' asthma conditions; but also has good discrimination effect and robustness in recognizing asthma sounds through a simplified neural network.

[0096] Example 2:

[0097] This embodiment discloses a portable asthma detection device, including a housing, a main control unit built into the housing, and a contact microphone connected to the main control unit embedded at the bottom; wherein, the main control unit is used to control the contact microphone to collect breath sounds and perform asthma detection methods as described above to identify wheezing and whimpering sounds.

[0098] This application uses a contact microphone to collect the user's breathing sounds, avoiding the risk of foreign body rejection and compliance issues associated with implanted sensors, thus improving the comfort and acceptance of the test, and is especially suitable for special groups such as children.

[0099] In an optional embodiment, buttons and a display panel are arranged on the top of the housing. The buttons are used for manual control of the detection, and the display panel is used to display the recognition results. See details... Figures 3-6 ;

[0100] Furthermore, in this embodiment, the main control unit is integrated into a circuit board, which is connected to a rechargeable lithium battery, a TYPE-C interface, an SD card interface, and a speaker. The TYPE-C interface, SD card interface, and speaker are externally mounted on the housing. The TYPE-C interface is used to charge the rechargeable lithium battery, the SD card interface is used to store respiratory sound signals collected by the contact microphone, and the speaker is used to trigger an alarm when asthma is detected.

[0101] When conducting asthma testing, the collected breath sound signals are first stored in the tablet's built-in memory card. Then, a built-in machine learning model is used to identify asthma symptoms and generate a result. Finally, the breath signal waveform and the identification result are displayed on the display panel.

[0102] In a preferred embodiment, the circuit board also integrates an analog data processing module for amplifying, filtering, and converting the collected breathing sounds into digital data.

[0103] Example 3:

[0104] In this embodiment, the portable asthma detection device housing both a main control unit and a wireless module. The main control unit controls the contact microphone to collect respiratory sounds and transmits them to the terminal via the wireless module. The terminal executes any of the asthma detection methods described above to identify wheezing and whimpering sounds.

[0105] In this embodiment, the terminal can be a tablet device or a mobile phone device, and the wireless module is a Bluetooth module;

[0106] The asthma detection method is presented as an app and includes functions related to the operation of the data collection device, such as controlling the start and end of the data collection and sound playback via a wireless module.

[0107] In the above embodiments,

[0108] The main control unit uses an STM32 series chip;

[0109] The contact microphone model is CM-01B. This microphone relies on the mechanical vibration of the human body surface to collect signals. It can significantly suppress background noise and can still clearly collect target body sounds in noisy environments. It has a good response to the low to mid frequency range and can accurately capture physiological sounds such as breathing sounds and asthma sounds.

[0110] The Bluetooth module is an HC-04 Bluetooth serial communication module, operating in the 2.4GHz ISM band, using GFSK modulation, and capable of Bluetooth transmission within 10m.

[0111] The rechargeable lithium battery is a 1200mAh rechargeable lithium battery with a rated power of 4.4Wh and an output voltage of 3.7V.

[0112] The terminal uses a Huawei HONOR Pad X8 tablet computer with a built-in asthma detection app. It can transmit data with the data collection device via Bluetooth, process and analyze the data, and display the collected data and recognition results. In this embodiment, the app can remotely control the device, including controlling the start of data collection and setting the collection mode (normal mode or calibration mode).

[0113] Preferably, a buzzer and LED lights are also installed on the casing to indicate power on / off status, battery level, and Bluetooth connection status.

[0114] This detection device collects breath sounds, calculates the sound waveforms, and uses machine learning methods to identify and diagnose asthma over a period of time. It uses a Bluetooth acquisition device and a receiving tablet to convert the incoming sound signals into electrical signals. The entire device is battery-powered and interacts via Bluetooth. The Bluetooth acquisition device uses button operations to collect, store, upload, and recognize sound. The receiving tablet can also control various operations of the acquisition device and display the collected signals, recognition results, and other information.

[0115] This asthma detection device utilizes non-invasive, high-precision breath sound analysis, combined with a lightweight neural network model and portable design, to achieve real-time and accurate asthma monitoring. It overcomes the shortcomings of existing technologies in terms of invasiveness, accuracy, real-time performance, and portability. It can be applied to assist in the assessment of asthma in children, providing an effective technical means for the early identification and timely intervention of childhood asthma. Specifically, the device is placed at a specific location on the chest for a period of time to collect data. The start and end of the data collection process are controlled in the host computer software to assist in the assessment of the child's asthma condition and help caregivers confirm the child's status in real time.

[0116] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0117] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A portable asthma detection device, characterized in that, The device includes a housing with a built-in main control unit and a contact microphone embedded at the bottom connected to the main control unit. The main control unit is used to acquire the user's breathing sounds, denoise and segment them, and then convert them into a spectrum using short-time Fourier transform and logarithmic transform. The spectrum is then input into a neural network model to obtain wheezing and dyspnea sounds. In the neural network model described above, all convolutional layers in the initial part are replaced or removed by pooling layers; in the middle part, some convolutional layers are replaced by pooling layers, using continuous 2*1 convolutions + 1*2 pooling; and in the final part, fully connected layers are retained, the number of convolutional layers and convolutional filters is reduced to lower the parameters, and at least one 3*3 convolutional layer is retained to ensure that the model's test accuracy is not affected. The final simplified neural network model includes: multiple feature extraction modules and fully connected layers; each feature extraction module consists of the following layer combination: a. Basic unit: one or more consecutive convolutional layers; b. Normalized Unit: The batch normalization layer and ReLU activation layer immediately following the convolutional layer; c. Downsampling unit: Max pooling layer, used to reduce the resolution of feature maps; Then, the Softmax function is applied to the output of the fully connected layer to obtain the classification result.

2. The asthma detection device according to claim 1, characterized in that, Wavelet thresholding is used to filter and denoise user breathing sounds.

3. The asthma detection device according to claim 1, characterized in that, The formula for calculating the short-time Fourier transform is: In the formula, t represents the center time point of the current window, f represents the frequency, x(τ) represents the value of the denoised and segmented respiratory signal at τ, h(τ-t) represents the time-shifted window function for truncating the signal segment, centered at t, τ represents the integration time variable, and e -j2πfτ This represents the complex exponential basis functions of the Fourier transform.

4. The asthma detection device according to claim 1, characterized in that, The formula for calculating the logarithmic transformation is: In the formula, img(i,j) represents the spectrum obtained through STFT, i represents the time frame index, and j represents the frequency index.

5. The portable asthma detection device according to claim 1, characterized in that, The top of the housing has buttons and a display panel. The buttons are used for manual control of the detection, and the display panel is used to display the recognition results.

6. The portable asthma detection device according to claim 1, characterized in that, The main control unit is integrated on a circuit board, which is connected to a rechargeable lithium battery, a TYPE-C interface, an SD card interface, and a speaker.

7. The portable asthma detection device according to claim 1, characterized in that, The circuit board also integrates an analog data processing module, which is used to amplify, filter, and convert the collected breathing sounds into digital data.

8. The portable asthma detection device according to claim 1, characterized in that, It includes a wireless module and a contact microphone embedded at the bottom that is connected to the unit; the main control unit is used to control the contact microphone to collect breathing sounds and transmit them to the terminal through the wireless module.

Images