Cough sound-based screening aid system and electronic device for copd

Abstract

The present application relates to the technical field of medical data assistance, and particularly relates to a screening auxiliary system and electronic equipment for chronic obstructive pulmonary disease based on cough sound, which comprises a data acquisition module for acquiring audio data to be processed; a cough sound extraction module for acquiring a mel-frequency spectrum diagram of the audio data, and extracting a cough sound segment from the mel-frequency spectrum diagram by using a pre-trained Conformer-BiLSTM network; and a chronic obstructive pulmonary disease screening module for preprocessing the cough sound segment and converting it into a frequency domain signal, obtaining FBank feature parameters corresponding to the frequency domain signal by using a mel filter, inputting the FBank feature parameters into a pre-trained classification model with dynamic convolution and a multi-scale attention mechanism, and outputting classification results of chronic obstructive pulmonary disease cough sound and non-chronic obstructive pulmonary disease cough sound. The present application reduces the possibility of classification errors and improves classification efficiency.

Description

Technical Field

[0001] This invention relates to the field of medical data assistance technology, and in particular to a screening assistance system and electronic device for COPD based on cough sounds. Background Technology

[0002] COPD is a respiratory disease affecting hundreds of millions of people worldwide. Currently, the standard diagnostic method for COPD is pulmonary function testing, but this method is costly and time-consuming, making it difficult to meet the needs of large-scale population screening. In contrast, cough sounds, as a biosignal reflecting abnormal lung function, have the advantages of being non-invasive and easy to collect.

[0003] However, current COPD screening using cough sounds still relies on manual annotation and simple models, which are easily affected by environmental noise, resulting in high rates of missed and false detections. Consequently, the extracted cough sounds may not be accurately graded or may be graded incorrectly in subsequent processing, affecting the system's grading accuracy. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a screening auxiliary system and electronic device for COPD based on cough sounds, which solves the technical problem that the prior art cannot accurately extract and classify cough sounds.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0008] In a first aspect, the COPD screening assistance system based on cough sounds proposed in the embodiments of the present invention includes a data acquisition module for acquiring audio data to be processed;

[0009] A cough sound extraction module is used to acquire the Mel spectrogram of the audio data and extract cough sound segments from the Mel spectrogram using a pre-trained Conformer-BiLSTM network;

[0010] The COPD screening module is used to preprocess the cough sound fragments and convert them into frequency domain signals. The FBank feature parameters corresponding to the frequency domain signals are obtained through a Mel filter. The FBank feature parameters are input into a pre-trained classification model with dynamic convolution and multi-scale attention mechanisms to output the classification results of COPD cough sounds and non-COPD cough sounds.

[0011] Optionally, the system further includes a COPD grading module, which, when the classification result is COPD cough sound, inputs the FBank feature parameters into a pre-trained screening model to obtain the level to which the COPD cough sound belongs; the pre-trained screening model outputs at least five levels.

[0012] Optionally, the screening model uses the softmax function to obtain the level of the COPD cough sound; wherein, the softmax function is a five-level classification activation function responsible for outputting probabilities;

[0013] The structure of the screening model, except for the cross-entropy loss function, is the same as the structure of the non-cross-entropy loss function in the classification model.

[0014] Optionally, the cough sound extraction module includes: a conversion unit for performing frequency domain conversion on audio data with a sampling rate of 16000Hz and a frame length greater than 32ms, and obtaining a Mel spectrogram with time-frequency features; a detection unit for extracting local time-frequency features with spatial dimension compression performance from the Mel spectrogram through the CNN layer of the Conformer-BiLSTM network; and obtaining the temporal dependencies of the Mel spectrogram through the BiLSTM layer of the Conformer-BiLSTM network; the Conformer layer of the Conformer-BiLSTM network performs residual connections based on the local time-frequency features and temporal dependencies to obtain the position of the cough frame corresponding to the audio data; and an extraction unit for extracting cough sound segments from the audio data based on the position of the cough frame.

[0015] Optionally, the Conformer-BiLSTM network includes: three CNN layers, two BiLSTM layers, and three stacked Conformer structures; each Conformer structure includes: a multi-head attention branch for capturing global context relationships and a convolutional branch for local feature enhancement; the Conformer structure is also used to fuse the residual connections output by the multi-head attention branch and the convolutional branch respectively; during the training phase, the Adam optimizer is used, with an initial learning rate of 0.001, a batch size of 16, and an epoch of 100. Training stops after reaching the maximum number of epochs, and the loss function is Focal Loss.

[0016] Optionally, the Conformer-BiLSTM network further includes a BatchNorm layer, which is used to normalize the output of the Conformer structure to obtain and output a frame-level cough probability sequence.

[0017] Optionally, the COPD screening module includes: a preprocessing unit for pre-emphasizing cough sound segments, dividing the pre-emphasized signal into short frames of a specified duration, windowing each frame, and converting the windowed frames into frequency domain signals using a fast Fourier transform; an FBank feature parameter acquisition unit for inputting the frequency domain signal into a Mel filter to obtain FBank feature parameters; and a classification unit for inputting the FBank feature parameters into a pre-trained classification model with dynamic convolution and multi-scale attention mechanisms to output classification results for COPD cough sounds and non-COPD cough sounds.

[0018] Optionally, the classification model includes: a full-dimensional dynamic convolutional module with a four-branch parallel architecture, a residual network, and a statistical pooling layer;

[0019] The full-dimensional dynamic convolution module is used to perform cross-dimensional dynamic weighting processing on the number of convolution kernels, input channels, output channel feature information, and spatial dimension of FBank feature parameters to generate an adaptive weight matrix to generate preliminary features.

[0020] The residual network includes multiple Res2Net residual modules. Each residual module achieves cross-group feature interaction through grouped convolution and channel recombination techniques, and uses a multi-scale attention mechanism in the residual path to achieve weighted fusion of multi-scale features in multiple residual modules, and strengthens pathology-related features to output the first feature.

[0021] The statistical pooling layer is used to normalize the first feature and output the classification results of COPD cough sound and non-COPD cough sound through the sigmoid function; wherein, the sigmoid function is a binary classification activation function.

[0022] Optionally, the statistical pooling layer is a one-dimensional pooling layer.

[0023] Secondly, an electronic device proposed in the embodiments of the present invention includes any of the above-described screening assistance systems for COPD based on cough sounds.

[0024] (III) Beneficial Effects

[0025] The beneficial effects of this invention are as follows: This invention provides a screening assistance system and electronic device for COPD based on cough sounds. The invention utilizes a data acquisition module to acquire audio data to be processed, and a cough sound extraction module to acquire the Mel spectrogram of the audio data. A pre-trained Conformer-BiLSTM network is used to extract cough sound segments from the Mel spectrogram. Then, a COPD screening module preprocesses and performs frequency domain transformation on the cough sound segments to obtain feature parameters. These FBank feature parameters are input into a pre-trained classification model with dynamic convolution and multi-scale attention mechanisms, outputting classification results for COPD cough sounds and non-COPD cough sounds. Thus, COPD classification can be achieved.

[0026] Furthermore, this invention utilizes Mel spectrograms to locate cough sounds in both the time and frequency domains, improving the accuracy of cough sound localization. Moreover, the classification and cough sound extraction in this invention are performed using modules, rather than relying on the experience of human operators, thus reducing the possibility of classification errors and improving classification efficiency to a certain extent. Attached Figure Description

[0027] Figure 1 A schematic diagram of a screening assistance system for COPD based on cough sounds provided in an embodiment of the present invention;

[0028] Figure 2 This is a schematic diagram of the structure of a cough sound extraction module provided in an embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram of another cough sound extraction module provided in an embodiment of the present invention;

[0030] Figure 4 This is a schematic diagram of the structure of a COPD grading module provided in an embodiment of the present invention;

[0031] Figure 5 This is a flowchart illustrating another screening assistance system for COPD based on cough sounds, provided as an embodiment of the present invention. Detailed Implementation

[0032] Chronic obstructive pulmonary disease (COPD) is a common chronic disease characterized by airflow obstruction, involving chronic bronchitis and / or emphysema, which can further develop into pulmonary heart disease and respiratory failure. Currently, COPD is a respiratory disease affecting the health of hundreds of millions of people worldwide. The current standard diagnostic method for COPD is pulmonary function testing, but this method is costly and time-consuming, making it difficult to meet the needs of large-scale population screening. In contrast, cough sounds, as a biosignal reflecting abnormal lung function, have the advantages of being non-invasive and easy to collect.

[0033] In recent years, research on cough sounds as an aid in the diagnosis of COPD has gradually attracted attention. However, traditional methods rely on manual annotation or simple models, which are easily affected by environmental noise, resulting in high rates of missed and false detections. Existing technologies are mostly limited to COPD screening and are insufficient in capturing pathological features such as abnormal low-frequency energy distribution and spectral discretization in cough sounds, thus failing to meet the clinical need for accurate grading of COPD severity.

[0034] To address the aforementioned problems, this invention provides a cough sound-based COPD screening assistance system. The system utilizes a data acquisition module to acquire audio data to be processed, and a cough sound extraction module to obtain the Mel spectrogram of the audio data. A pre-trained Conformer-BiLSTM network is then used to extract cough sound segments from the Mel spectrogram. The COPD screening module preprocesses and performs frequency domain transformation on the cough sound segments to obtain feature parameters. These FBank feature parameters are input into a pre-trained classification model with dynamic convolution and multi-scale attention mechanisms, outputting classification results for COPD and non-COPD cough sounds. This enables COPD classification. Furthermore, this invention utilizes the Mel spectrogram to locate cough sounds in both the time and frequency domains, improving the accuracy of cough sound localization. Moreover, both classification and cough sound extraction are performed by modules, rather than relying on the experience of operators, thus reducing the possibility of classification errors and improving classification efficiency to a certain extent.

[0035] For a better explanation of the invention, please refer to [link / reference]. Figure 1 ,like Figure 1 As shown, this system can be applied to any electronic device in the medical field, or within a medical computer system, specifically including:

[0036] The data acquisition module is used to acquire the audio data to be processed. In this embodiment, the audio data of the patient to be tested can be acquired and stored in advance through a mobile phone or other audio acquisition means, and then the data acquisition module can acquire and process this audio data.

[0037] A cough sound extraction module is used to acquire the Mel spectrogram of the audio data and extract cough sound segments from the Mel spectrogram using a pre-trained Conformer-BiLSTM network.

[0038] The COPD screening module is used to preprocess the cough sound fragments and convert them into frequency domain signals. The FBank feature parameters corresponding to the frequency domain signals are obtained through a Mel filter. The FBank feature parameters are input into a pre-trained classification model with dynamic convolution and multi-scale attention mechanisms to output the classification results of COPD cough sounds and non-COPD cough sounds.

[0039] Specifically, the system includes at least one of the following: a data acquisition module, a cough sound extraction module, and a COPD screening module. The data acquisition module is primarily used to acquire audio data collected by the acquisition device. This acquisition device includes, but is not limited to, at least one of the following: a smartphone, a microphone, a voice recorder, and a lung acoustic detector. The audio data acquired by the data acquisition module has temporal domain characteristics.

[0040] During the data collection process, to maintain the purity of the audio signal and the clarity of the speech, participants were required to remove their masks during recording. The microphones on the recording equipment were carefully disinfected after each use to ensure hygiene and safety. During recording, the microphone was placed 30-40 centimeters from the participant's mouth at a 45-degree angle upwards from the horizontal. This distance and angle were chosen to capture the most detail in the cough sound. Participants coughed in a standardized manner: first taking a deep breath, then forcefully coughing it out.

[0041] Because the environment varies during data collection, the audio data acquired by the acquisition device may include not only the participants' cough sounds but also background noise, equipment noise, and other sounds. Therefore, a cough sound extraction module can be used to extract the participants' cough sounds from the audio data, thus avoiding the influence of other noises on the subsequent classification results. This embodiment does not specifically limit the method by which the cough sound extraction module extracts cough sounds.

[0042] In one illustrative embodiment, the cough sound extraction module specifically obtains the Mel spectrogram of audio data and extracts cough sound segments from the Mel spectrogram using a pre-trained Conformer-BiLSTM network. Specifically, since audio data inherently possesses time-domain characteristics, only a frequency-domain transformation is needed to obtain the frequency-domain features of the audio data. Then, combining these with the inherent time-domain features of the audio data, the Mel spectrogram can be obtained. Because cough sound recognition depends not only on frequency distribution but also on dynamic changes over time, the Mel spectrogram can display the sound distribution pattern in the audio data from both time and frequency perspectives, thereby enabling rapid and accurate localization of cough sounds.

[0043] The cough sound extraction module in this embodiment specifically includes a conversion unit, a detection unit, and an extraction unit. The conversion unit performs frequency domain conversion on the audio data. This embodiment does not specifically limit the format of the audio data; for example, it can be audio data with a sampling rate of 16000Hz and a frame length greater than 32ms. A Mel spectrogram with time-frequency characteristics is obtained based on the frequency and time domain characteristics of the audio data. The detection unit can extract local time-frequency features with spatial dimension compression performance from the Mel spectrogram using the CNN layer of the Conformer-BiLSTM network. Specifically, since the Mel spectrogram has both time (i.e., time domain) and frequency (i.e., frequency domain) dimensions, these two dimensions together construct the spatial structure of the Mel spectrogram. The CNN layer can retain the most critical features of the Mel spectrogram in both dimensions through the stride and pooling operations of the convolutional layers. For example, cough sounds often have sudden and brief pulses, which are characteristics of the time domain. In the frequency domain, the frequency of cough sounds is often relatively concentrated. Combining these characteristics for analysis is equivalent to comprehensively considering both spatial and temporal dimensions, which can improve efficiency in the subsequent cough sound determination stage. For example, some cough sounds have abnormal low-frequency energy distribution and discretized spectrum. By analyzing both the frequency energy distribution and time of the cough sound, we can determine whether its ability is abnormal and whether the spectrum is discretized.

[0044] The detection unit can obtain the temporal dependencies of the Mel spectrogram through the BiLSTM layer of the Conformer-BiLSTM network. In other words, the BiLSTM layer can capture the contextual information of the Mel spectrogram in chronological order. This contextual information refers to analyzing the relationships between consecutive frames, on a frame-by-frame basis. It's important to note that using frames as the unit here is merely an example for ease of understanding; the specific approach can be adjusted according to actual needs, and the relationships between consecutive seconds can also be analyzed. Based on the potential temporal relationships between multiple frames in the audio data (e.g., cough sounds have continuous pulses, and multiple pulses occur within a certain time frame; frames with pulses and close timing can be considered part of a single cough sound), and using the Conformer layer to perform residual connections based on local time-frequency features and temporal dependencies, the location of the corresponding cough frame in the audio data is obtained. Here, the Conformer layer mainly plays a role in feature fusion, that is, capturing the correlation between frames while preserving intra-frame time-frequency features, thereby improving the accuracy of cough sound localization.

[0045] For easier understanding, please refer to Figure 2 ,like Figure 2As shown, the Mel spectrogram can be passed through the CNN layers in the Conformer-BiLSTM network to obtain local time-frequency features. Then, the Mel spectrogram is passed through two layers of BiLSTM to obtain the temporal dependencies of the Mel spectrogram. Finally, the temporal dependencies and local time-frequency features are passed through three stacked Conformers to output the position of the cough frame.

[0046] The extraction unit can extract the audio from the audio data by detecting the position of the cough frame obtained by the detection unit, and obtain the cough sound segment.

[0047] It is important to note that this embodiment does not specifically limit the specific structure of the Conformer-BiLSTM network. In practice, the network structure may include: three CNN layers, two BiLSTM layers, and three stacked Conformer structures. This configuration allows for sufficient feature extraction from the Mel spectrogram. Furthermore, due to the limited spatial complexity of the Mel spectrogram, the three CNN layers can avoid overfitting during feature extraction. The two-layer BiLSTM can learn not only the temporal relationships between individual frames but also capture the temporal relationships between frames throughout the entire audio data. For example, the first BiLSTM layer can capture the temporal dependencies between each frame and its adjacent frames in the Mel spectrogram, while the second BiLSTM layer can capture the temporal dependencies between non-adjacent frames based on the output of the first BiLSTM layer. This allows for more precise location of cough frames, improving the accuracy of cough frame identification.

[0048] Optionally, any Conformer structure may include: a multi-head attention branch for capturing global contextual relationships, and a convolutional branch for local feature enhancement. Please refer to [link / reference] for details. Figure 3 ,like Figure 3As shown, the Attention Module is the multi-head attention branch, the Conv Module is the convolution branch, the Feed forward Module is the feedforward module, which is used to perform nonlinear transformation on temporal dependencies and local time-frequency features to enhance feature representation ability, LayerNorm is the layer normalization, which is used for normalization processing, and the Attention Score is the attention score, which is used to learn the correlation between frames. The module input is the temporal features of the Mel spectrogram time-series dependencies extracted by the BiLSTM layer. It first enters the first residual structure, where the features initially processed by the feedforward module are added to the original input. The output of this feedforward module is simultaneously fed into two branches, each containing a residual structure. The first branch on the left outputs the input features, which are first processed by a convolutional branch and then by a multi-head attention branch before being added to the original features. The second branch on the right inputs the features, which are first processed by a multi-head attention branch and then by a convolutional branch before being added to the original features. Simultaneously, the key attention scores of the two branches are fused, achieving information fusion. After processing by these two branches, the outputs of the left and right branches are finally connected by an adder to complete the third residual connection. The resulting comprehensive features are then fed into the encoder's feedforward module for further transformation and integration. The output of this module is finally processed by layer normalization to obtain the final output of the entire encoder module. This output effectively combines the advantages of convolution and attention, while preserving original information and promoting feature reuse through multi-level residual connections, thus enhancing the model's multi-scale feature extraction capability and long-range dependency modeling capability.

[0049] Optionally, the Conformer-BiLSTM network may further include a BatchNorm layer, which normalizes the output of the Conformer structure to obtain a frame-level cough probability sequence. Simultaneously, a dynamic threshold detection algorithm is used to analyze temporal distribution characteristics, enabling precise localization of the start and end points of cough sounds. For example, after normalizing the output of the Conformer structure in the BatchNorm layer, the probability of each frame being a cough sound can be obtained. When this probability is greater than a certain threshold, the frame is considered a cough frame. This embodiment does not impose a specific limitation on this threshold; for example, the threshold could be 80%.

[0050] During the training of the Conformer-BiLSTM network, the Adam optimizer can be used to optimize the network parameters. The optimizer's initial learning rate is 0.001, the batch size is 16, and the epochs are set to 100. Training stops after reaching the maximum number of epochs, and the loss function is Focal Loss. Thus, in the training samples used during training, the number of cough frames and non-cough frames is often imbalanced, with non-cough frames potentially far exceeding cough frames. Therefore, this embodiment utilizes Focal Loss by introducing a modulation factor to make the network pay more attention to the difficult-to-classify cough frame samples, thereby effectively alleviating the problem caused by class imbalance and improving the network's ability to detect cough frames.

[0051] The specific relationship can be satisfied as follows:

[0052] ;

[0053] Wherein, FL is the frame-level cough probability sequence; It's a real label. To predict probabilities for the model, It is a parameter used to balance the weights of positive and negative samples. This is the focusing parameter that adjusts the weights of easily classified samples. In this embodiment, it is set to 2, and the base of the log function is e, which is the natural logarithm.

[0054] After obtaining the cough sound fragments output by the cough sound extraction module, the COPD screening module can use the preprocessing unit, the FBank feature parameter acquisition unit, and the classification unit to obtain the classification results.

[0055] In one illustrative embodiment, a preprocessing unit pre-emphasizes the cough sound segment, enhancing the high-frequency components to better capture audio details. The pre-emphasized signal is then divided into short frames of a specified duration, and each frame is windowed using a window function to reduce signal discontinuities between frames. Finally, the windowed frames are converted to frequency domain signals using a Fast Fourier Transform. This yields a clearer, more continuous frequency domain signal of the cough sound segment.

[0056] Next, the frequency domain signal can be input into a Mel filter to filter the frequency domain signal of each frame, obtaining the FBank feature parameters of the cough sound segment. The Mel filter bank is a set of triangular filters constructed based on the Mel scale. When the frequency domain signal passes through the Mel filter, the frequency components can be integrated according to the way the human ear perceives sound, making it easier for subsequent units to obtain the essential characteristics of the cough sound. Furthermore, the FBank feature parameters output by the Mel filter are obtained by filtering the spectrum and then combining them, meaning the FBank feature parameters have both time-domain and frequency-domain characteristics.

[0057] Since the FBank feature parameters retain both frequency domain and time domain features, the FBank feature parameters can be further input into a pre-trained classification model with dynamic convolution and multi-scale attention mechanisms to fully explore the correlation between the two dimensions (e.g., pulse changes at a certain frequency at a specific time point), thereby outputting classification results for COPD cough sounds and non-COPD cough sounds.

[0058] It should be noted that this embodiment does not specifically limit the structure of the classification model. For example, the classification model may include: a full-dimensional dynamic convolutional module with a four-branch parallel architecture, a residual network, and a statistical pooling layer.

[0059] In other words, the full-dimensional dynamic convolution module is used to perform cross-dimensional dynamic weighting processing on the number of convolution kernels, input channels, output channels, and spatial dimensions of FBank feature parameters. This breaks the isolation between dimensions, captures the time-frequency intersection features in the FBank feature parameters, generates an adaptive weight matrix, and generates preliminary features. Specifically, the full-dimensional dynamic convolution module can determine which dimension is more effective in distinguishing between COPD and non-COPD cough sounds. The input and output channels can be considered as frequency domain dimensions, the spatial dimension as the time domain dimension, and the number of convolution kernels as the importance of each dimension. For example, COPD cough is often accompanied by low-frequency resonance caused by airway narrowing, which has strong discriminative power. Therefore, in the frequency domain dimension, the number of convolution kernels will be given higher weight.

[0060] The specific relationship can be satisfied as follows:

[0061] ;

[0062] in, Indicates the convolution kernel The weight, This represents the weights in the convolution kernel space; This represents the weights for the input channels. This represents the weights for the output channels, where i ranges from 1 to n, x is the initial convolution, and y is the final convolution. It is the tensor product.

[0063] Residual networks can learn more complex pathological features layer by layer from initial features, providing a foundation for subsequent judgments. This embodiment does not specifically limit the structure of the residual network; for example, the residual network may include multiple Res2Net residual modules. Each residual module achieves cross-group feature interaction through grouped convolution and channel recombination techniques. That is, each group uses an independent convolutional kernel to extract different types of features. This embodiment does not specifically limit the number of residual modules; for example, it can be four. For instance, the first group focuses on low-frequency features, the second group focuses on high-frequency features, and so on. Furthermore, a multi-scale attention mechanism is used in the residual path to achieve weighted fusion of multi-scale features from multiple residual modules, strengthening pathology-related features and outputting the first feature.

[0064] The statistical pooling layer normalizes the first feature and outputs the classification results of COPD cough sounds and non-COPD cough sounds through the sigmoid function. In other words, it uses the sigmoid function to output the probability that a cough sound segment is a COPD cough sound or not, and based on this probability, it can determine whether the cough sound segment is a COPD cough sound. The sigmoid function is a binary classification activation function. In this embodiment, the statistical pooling layer is a one-dimensional pooling layer, which can retain key frequency band temporal information while compressing features along the time axis. It strengthens the channel response of pathology-related features through a frequency gating mechanism, making it more suitable for the characteristics of time-series data such as audio. In COPD cough sound analysis, it can accurately capture the changing trend of frequency channels over time, allowing channel attention to focus more on pathology-related features such as duration and time-frequency linkage, ultimately improving the model's accuracy in distinguishing COPD cough sounds.

[0065] Optionally, the system may further include a COPD grading module, used to input the FBank feature parameters into a pre-trained screening model when the classification result is COPD cough sound, to obtain the grade to which the COPD cough sound belongs; the pre-trained screening model outputs at least five grades. This grade screening model uses a softmax function to obtain the grade to which the COPD cough sound belongs; wherein, the softmax function is a five-level classification activation function, and the softmax function can be used to output the probability that the cough sound segment belongs to a certain grade. The structure of the screening model, except for the five-level classification activation function, is the same as the structure in the classification model. The grades include, but are not limited to, at least one of the following: normal, mild, moderate, severe, and very severe.

[0066] It should be noted that the COPD grading module may also include other processing structures, and this embodiment does not impose specific limitations here.

[0067] In one illustrative example, please refer to [reference needed]. Figure 4 ,like Figure 4As shown, the COPD grading module takes the FBank features of COPD cough sounds as input. First, it undergoes dynamic multi-dimensional feature extraction through a full-dimensional dynamic convolution module, followed by batch normalization layers to standardize the feature distribution. Then, it compresses the feature dimensions through max pooling layers, and subsequently enters the residual module sequence of the backbone network: the first 2-layer residual module processes basic features, the second 3-layer residual module deepens feature abstraction, the third 5-layer residual module strengthens feature representation, and finally, a 3-layer residual module completes high-order feature compression. Afterward, the model selects a pooling type based on requirements: a uniform size can be selected from adaptive spatial pooling (ASP), spatial adaptive pooling (SAP), temporal adaptive pooling (TAP), or spatiotemporal pooling (TSP) to eliminate input scale differences. The pooling output is fed into a fully connected layer for class space mapping, and finally, a softmax activation function is used to generate the probability distribution for COPD diagnosis, completing the intelligent decision-making classification result.

[0068] The above describes a screening assistance system for COPD based on cough sounds provided by the present invention. The above embodiments can be combined according to actual conditions, and the present invention does not impose specific limitations here.

[0069] For a better understanding of the solution provided by this invention, please refer to [the relevant documentation / reference]. Figure 5 ,like Figure 5 As shown, after the participants complete the data collection, the audio data will be uploaded to the audio database. Next, the cough sound extraction module can be used to detect and extract the cough sounds. Furthermore, the extracted cough sounds are classified. If classified as COPD, then it can be graded.

[0070] In addition, embodiments of the present invention also provide an electronic device, which may include a memory and a processor, wherein the memory stores a computer program and the processor executes the program in the memory, such as executing the screening assistance system for COPD based on cough sounds of any of the above embodiments.

[0071] The advantages of this electronic device correspond to those of any of the aforementioned screening assistance systems. It can quickly and accurately distinguish whether the audio signal belongs to COPD, and further, if it does belong to COPD, it can further subdivide its level, ensuring the accuracy of the judgment.

[0072] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0073] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0074] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "over," or "on top" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," or "beneath" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0075] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0076] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A screening assistance system for COPD based on cough sounds, characterized in that, include: The data acquisition module is used to acquire the audio data to be processed; A cough sound extraction module is used to acquire the Mel spectrogram of the audio data and extract cough sound segments from the Mel spectrogram using a pre-trained Conformer-BiLSTM network; The Conformer-BiLSTM network comprises: three CNN layers, two BiLSTM layers, and three stacked Conformer structures; Each Conformer structure includes: a multi-head attention branch for capturing global contextual relationships and a convolutional branch for local feature enhancement; the Conformer structure is also used to fuse the residual connections output by the multi-head attention branch and the convolutional branch respectively; The COPD screening module is used to preprocess the cough sound fragments and convert them into frequency domain signals. The FBank feature parameters corresponding to the frequency domain signals are obtained through a Mel filter. The FBank feature parameters are input into a pre-trained classification model with dynamic convolution and multi-scale attention mechanisms to output the classification results of COPD cough sounds and non-COPD cough sounds.

2. The system according to claim 1, characterized in that, The system also includes a COPD grading module, which, when the classification result is COPD cough sound, inputs the FBank feature parameters into a pre-trained screening model to obtain the level to which the COPD cough sound belongs; the levels output by the pre-trained screening model include at least five levels.

3. The system according to claim 2, characterized in that, The screening model uses the softmax function to determine the level of the cough sound in COPD; wherein, the softmax function is a five-class activation function responsible for outputting probabilities. The structure of the screening model, except for the cross-entropy loss function, is the same as the structure of the non-cross-entropy loss function in the classification model.

4. The system according to claim 1, characterized in that, The cough sound extraction module includes: The conversion unit is used to perform frequency domain conversion on audio data with a sampling rate of 16000Hz and a frame length greater than 32ms, and to obtain a Mel spectrogram with time-frequency characteristics. The detection unit extracts local time-frequency features with spatial dimension compression performance from the Mel spectrogram through the CNN layer of the Conformer-BiLSTM network; Furthermore, the temporal dependencies of the Mel spectrogram are obtained through the BiLSTM layer of the Conformer-BiLSTM network; The Conformer layer of the Conformer-BiLSTM network performs residual connections based on local time-frequency features and temporal dependencies to obtain the position of the cough frame in the corresponding audio data. An extraction unit is used to extract cough sound segments from the audio data based on the position of the cough frame.

5. The system according to claim 4, characterized in that, During the training phase, the Adam optimizer was used with an initial learning rate of 0.001, a batch size of 16, and an epoch of 100. Training stopped after reaching the maximum number of epochs, and the loss function was Focal Loss.

6. The system according to claim 5, characterized in that, The Conformer-BiLSTM network also includes a BatchNorm layer, which is used to normalize the output of the Conformer structure to obtain and output a frame-level cough probability sequence.

7. The system according to claim 1, characterized in that, COPD screening module, including: The preprocessing unit is used to pre-emphasize the cough sound segment, divide the pre-emphasized signal into short frames of a specified duration, perform windowing processing on each frame, and convert the windowed frame into a frequency domain signal through fast Fourier transform. The FBank feature parameter acquisition unit is used to input the frequency domain signal into the Mel filter to obtain the FBank feature parameters. The classification unit is used to input the FBank feature parameters into a pre-trained classification model with dynamic convolution and multi-scale attention mechanisms, and output the classification results of COPD cough sounds and non-COPD cough sounds.

8. The system according to claim 7, characterized in that, The classification model includes: a full-dimensional dynamic convolutional module with a four-branch parallel architecture, a residual network, and a statistical pooling layer; The full-dimensional dynamic convolution module is used to perform cross-dimensional dynamic weighting processing on the number of convolution kernels, input channels, output channel feature information, and spatial dimension of FBank feature parameters to generate an adaptive weight matrix to generate preliminary features. The residual network includes multiple Res2Net residual modules. Each residual module achieves cross-group feature interaction through grouped convolution and channel recombination techniques, and uses a multi-scale attention mechanism in the residual path to achieve weighted fusion of multi-scale features in multiple residual modules, and strengthens pathology-related features to output the first feature. The statistical pooling layer is used to normalize the first feature and output the classification results of COPD cough sound and non-COPD cough sound through the sigmoid function; wherein, the sigmoid function is a binary classification activation function.

9. The system according to claim 8, characterized in that, The statistical pooling layer is a one-dimensional pooling layer.

10. An electronic device, characterized in that, The screening assistance system for COPD based on cough sounds, as described in any one of claims 1 to 9 above.

Images