A method and system for intelligent risk early warning of acute exacerbation of bronchiectasis

Abstract

This invention provides an intelligent risk warning method and system for acute exacerbation of bronchiectasis, relating to the field of intelligent medical technology. The method includes: acquiring audio signals and segmenting them into frames; identifying and retaining cough audio frames using an edge-side privacy filtering model; extracting four-dimensional features: fundamental frequency drift rate, peak power spectral slope, temporal waveform rise / fall ratio, and Mel energy distribution difference; statistically aggregating the four-dimensional features of all cough frames within the same file and inputting them into a low-discrepancy forest classification model; weighting and fusing the class probabilities output by each decision tree and comparing them with a threshold to output a binary classification warning result. This invention achieves low-latency, high-privacy, and high-stability intelligent warning for acute exacerbation of bronchiectasis by completing privacy filtering, feature extraction, and weighted classification at the edge.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical technology, and in particular to an intelligent risk warning method and system for acute exacerbation of bronchiectasis. Background Technology

[0002] Bronchiectasis is a chronic respiratory disease. During acute exacerbations, patients often experience increased cough frequency, changes in sputum characteristics, and worsening dyspnea. If the signs of acute exacerbations are not identified and interventions are not taken in time, the patient's respiratory system often suffers irreversible damage, requiring hospitalization and severely impacting quality of life and prognosis.

[0003] Current technologies for monitoring acute exacerbations of bronchiectasis primarily rely on patient subjective reporting, manual questionnaires, and regular follow-ups, which suffer from issues such as insufficient timeliness, poor continuity, and limited patient compliance. While some existing protocols can collect and identify respiratory-related signals, they often depend on cloud servers for data analysis, which carries risks such as long transmission links, significant end-to-end latency, and leakage of patient audio privacy. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an intelligent risk warning method and system for acute exacerbation of bronchiectasis.

[0005] To achieve the above objectives, a first aspect of the present invention provides an intelligent risk warning method for acute exacerbation of bronchiectasis, comprising: The patient's real-time audio signal is acquired, and the audio signal is divided into frames to obtain an audio frame sequence; The audio frame sequence is input into the end-side privacy filtering model to identify whether each audio frame is a cough audio frame and filter out non-cough audio frames to obtain a set of cough audio frames. For each cough audio frame in the set of cough audio frames, extract four-dimensional acute exacerbation discrimination features, which include: fundamental frequency drift rate, peak power spectral slope, time-domain waveform rise / fall ratio, and Mel energy distribution difference; The four-dimensional features of all cough audio frames within the same audio file are statistically aggregated to obtain a file-level feature vector; The file-level feature vector is input into a low-discrepancy forest classification model, which consists of multiple decision trees, each of which outputs the class probability of the corresponding input sample. The class probabilities output by each decision tree are weighted and fused to obtain a fusion probability. The fusion probability is compared with a preset threshold to output a binary classification warning result.

[0006] Furthermore, the weighted fusion includes: assigning fusion weights to each decision tree, the fusion weights being determined based on the tree's prior decay weights and the tree's classification accuracy on the training set, thereby differentiating the contributions of each decision tree.

[0007] Furthermore, the allocation method of the fusion weights is as follows: assuming a total of The first decision tree, the The prior decay weights of the decision trees are , No. The classification accuracy of the decision trees on the training set is Then the first Fusion weights of decision trees satisfy: ; For input samples , No. Class fusion probability satisfy: ; in For the first Decision tree output samples Belongs to the The probability of a class; when the fusion probability of the target class is greater than or equal to a preset threshold, the target class is output; otherwise, another class is output.

[0008] Furthermore, the edge privacy filtering model is a binary classification convolutional neural network used to identify whether the input audio frame is coughing or not; when the identification result is not coughing, the audio frame is cleared or discarded, terminating subsequent feature extraction.

[0009] Furthermore, the four-dimensional acute exacerbation discriminant features are defined as follows: Fundamental frequency drift rate: used to characterize the magnitude of the fundamental frequency change in cough audio within adjacent time windows; Peak power spectral slope: used to characterize the trend of peak power in the main frequency band as a function of frequency; Time-domain waveform rise / fall ratio: used to characterize the proportional relationship between the duration of the rising edge and the duration of the falling edge of the cough waveform; Mel energy distribution difference: used to characterize the degree of deviation of the Mel band energy distribution from the preset reference distribution.

[0010] Furthermore, the statistical aggregation of the four-dimensional features of all cough audio frames within the same audio file specifically includes: calculating the mean, standard deviation, and maximum value of each feature across all cough audio frames within the audio file, and concatenating the mean, standard deviation, and maximum value of each feature in order to form a file-level feature vector.

[0011] A second aspect of the present invention provides an intelligent risk warning system for acute exacerbations of bronchiectasis, comprising: The audio acquisition and framing module is used to acquire the patient's real-time audio signal, and to segment the audio signal into frames to obtain an audio frame sequence. The privacy filtering module is used to input the audio frame sequence into the end-side privacy filtering model, identify whether each audio frame is a cough audio frame, and filter out non-cough audio frames to obtain a set of cough audio frames. The feature extraction module is used to extract four-dimensional acute aggravation discrimination features for each cough audio frame in the cough audio frame set. The four-dimensional features include: fundamental frequency drift rate, peak power spectral slope, time-domain waveform rise / fall ratio, and Mel energy distribution difference. The statistical aggregation module is used to statistically aggregate the four-dimensional features of all cough audio frames within the same audio file to obtain a file-level feature vector. The low-discrepancy forest classification module is used to input the file-level feature vector into the low-discrepancy forest classification model, which consists of multiple decision trees, each of which outputs the class probability of the corresponding input sample; the class probabilities output by each decision tree are weighted and fused to obtain the fusion probability; the fusion probability is compared with a preset threshold to output a binary classification warning result.

[0012] A third aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the intelligent risk warning method for acute exacerbation of bronchiectasis as described in the first aspect of the present invention.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the intelligent risk warning method for acute exacerbation of bronchiectasis as described in the first aspect of the present invention.

[0014] A fifth aspect of the present invention provides a computer program product comprising software code, wherein the program in the software code performs the steps of the intelligent risk warning method for acute exacerbation of bronchiectasis as described in the first aspect of the present invention.

[0015] Compared with existing technologies, the present invention provides an intelligent risk warning method and system for acute exacerbation of bronchiectasis, which has the following beneficial effects: by performing privacy filtering and feature calculation on cough audio frames directly at the edge, only non-cough frames are discarded locally without being uploaded, and by combining a weighted fusion mechanism to highlight the contribution of high-quality decision trees, thereby protecting patient privacy while reducing end-to-end latency and improving the accuracy and stability of acute exacerbation risk warning. Attached Figure Description

[0016] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0017] Figure 1 A flowchart of the intelligent risk warning method for acute exacerbation of bronchiectasis provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the end-side privacy filtering process provided in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of a traditional cascaded forest structure provided in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the improved low-discrepancy forest structure provided in Embodiment 1 of the present invention. The diagram shows multiple decision trees outputting probabilities for four-dimensional features and then weighted and fused to obtain the final classification result. Figure 5 This is a schematic diagram of the data flow provided in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the interference judgment confusion matrix provided in Embodiment 1 of the present invention; Figure 7 This is a schematic diagram of the pneumonia judgment confusion matrix provided in Embodiment 1 of the present invention; Figure 8 This is a schematic diagram of the ROC curve provided in Embodiment 1 of the present invention; Figure 9 This is a schematic diagram comparing the end-side inference performance provided in Embodiment 1 of the present invention; Figure 10 This is an architecture diagram of the intelligent risk early warning system for acute exacerbation of bronchiectasis provided in Embodiment 2 of the present invention. Detailed Implementation

[0018] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0021] All data acquisition in this embodiment is carried out in accordance with laws and regulations and with user consent, and the data is used legally.

[0022] Example 1 like Figure 1 This embodiment provides an intelligent risk early warning method for acute exacerbation of bronchiectasis, including: S1. Acquire the patient's real-time audio signal, divide the audio signal into frames, and obtain an audio frame sequence.

[0023] Specifically, audio acquisition utilizes an INMP441 digital microphone with a sampling rate of 16kHz, and is configured with a fourth-order Butterworth bandpass filter, with a passband range of 60 Hz to 15kHz. Signal processing is implemented using an ESP32-Xiao-Sensor dual-core 240MHz processor, utilizing on-chip memory resources to complete audio buffering, amplitude normalization, framing, Mel spectrum generation, and feature vector preparation.

[0024] S2. Input the audio frame sequence into the end-side privacy filtering model, identify whether each audio frame is a cough audio frame, and filter out non-cough audio frames to obtain a set of cough audio frames.

[0025] like Figure 2As shown, the edge privacy filtering model is a binary classification convolutional neural network used to identify coughs and non-coughs in input audio frames. This CNN model employs a lightweight design: the input is MFCC 64×64 floating-point format; the Conv1 layer is a 3×3 convolution with a 16-filter, ReLU activation, and a stride of 2; the Conv2 layer is a 3×3 convolution with a 32-filter, ReLU activation, and a stride of 2; global average pooling is used; and the Softmax output is two logits (cough / non-cough). The total model parameters are 28KB, using INT8 quantization, and the edge inference time is less than 10ms. When the identification result is non-cough, the audio frame is zeroed out or discarded (e.g., using a memset(buffer, 0, frameSize) operation), terminating subsequent feature extraction. Only audio frames confirmed as coughs are analyzed, achieving edge privacy protection.

[0026] S3. For each cough audio frame in the set of cough audio frames, extract four-dimensional acute aggravation discrimination features, the four-dimensional features including: fundamental frequency drift rate, peak power spectrum slope, time-domain waveform rise / fall ratio and Mel energy distribution difference.

[0027] The calculation method for the four-dimensional acute exacerbation discriminant features is as follows: (a) Fundamental frequency drift rate: The fundamental frequency sequence of cough audio frames is extracted using the short-time autocorrelation method or the cepstral method, and the statistical value of the fundamental frequency change amplitude within adjacent time windows is calculated. This feature is used to characterize the amplitude of fundamental frequency change in cough audio within adjacent time windows. The fundamental frequency change rate during acute exacerbations is usually greater than 8% / s.

[0028] (b) Peak Power Spectral Slope: A Fast Fourier Transform (FFT) is performed on the cough audio frames to extract the linearly fitted slope of the peak power as a function of frequency within the main frequency band. This feature characterizes the trend of peak power variation with frequency in the main frequency band. The spectral slope during acute exacerbations is typically greater than 0.15 dB / Hz.

[0029] (c) Time-domain waveform rise / fall ratio: The rise time and fall time of the cough waveform are measured, and the ratio of the two is calculated. This feature characterizes the proportional relationship between the rise time and fall time of the cough waveform. The ratio is usually greater than 1.3 during acute exacerbations.

[0030] (d) Mel energy distribution difference: The cough audio frames are mapped to the Mel band, and the relative entropy (KL divergence) between the Mel energy distribution and the preset reference distribution is calculated. This feature is used to characterize the degree of deviation of the Mel band energy distribution from the preset reference distribution. The KL divergence during acute exacerbations is usually greater than 0.25.

[0031] These four-dimensional features were obtained through statistical analysis of clinical samples and are specifically used to distinguish between the acute exacerbation and stable phases of bronchiectasis. This solves the problem of the insensitivity of general acoustic features to disease specificity and achieves a high degree of discrimination in acute exacerbation.

[0032] like Figure 5 As shown, the data flow in this embodiment is: audio input (16kHz×1s=16000 samples) → preprocessing (MFCC 64×64) → privacy CNN (28KB model) → feature extraction (64-dimensional feature vector) → low-discrepancy forest classification (binary output).

[0033] S4. Statistically aggregate the four-dimensional features of all cough audio frames within the same audio file to obtain a file-level feature vector.

[0034] Coughing events are distinctly fragmented in time. Using only a single instantaneous four-dimensional feature for judgment can easily overlook the statistical differences between multiple coughing segments within the same audio file. Therefore, this step statistically aggregates the four-dimensional features of all coughing audio frames within the same audio file. Specifically, the mean, standard deviation, and maximum value of each feature across all coughing audio frames in the audio file are calculated. These values ​​are then concatenated sequentially to form a file-level feature vector. For example, for four-dimensional features, statistical aggregation yields a 4×3=12-dimensional file-level feature vector. By calculating the mean, standard deviation, and maximum value of each feature across all coughing frames in the file, the central tendency, dispersion, and extreme values ​​of cough features can be comprehensively reflected, enhancing the ability to represent temporal fluctuations and individual differences, and improving the model's robustness to complex environmental noise and sample variations. This step solves the technical problem that single-frame features are susceptible to noise interference and cannot represent the overall coughing pattern, thus enhancing feature robustness and model generalization ability.

[0035] S5. Input the file-level feature vector into the low-discrepancy forest classification model, which consists of multiple decision trees, each of which outputs the class probability of the corresponding input sample; perform weighted fusion of the class probabilities output by each decision tree to obtain the fusion probability; compare the fusion probability with a preset threshold and output the binary classification warning result.

[0036] For respiratory disease cough sound samples with high similarity, if unweighted traditional random forests are directly used for judgment, the contributions of different decision trees to the final result are usually considered equivalent, making it difficult to reflect the differences in the discrimination ability of individual trees, thus affecting the accuracy and stability of acute exacerbation risk warning. As a key improvement of this invention, the low-discrepancy forest classification model in this embodiment does not use equal-weighted voting for each decision tree, but assigns learnable fusion weights to each decision tree. This allows the model to reflect the differences in the contributions of high-quality decision trees without increasing the number of model layers, reducing the invalid interference caused by equal-weighted voting, and making it more suitable for resource-constrained edge deployment scenarios.

[0037] In addition, such as Figure 3 As shown, the traditional cascaded forest structure commonly used in existing technologies typically involves repeatedly cascading the original input to the next level of the forest input at each layer. This easily leads to a continuous increase in feature dimensions, increasing the burden on storage, transmission, and inference. For highly similar cough sound samples, it may also introduce highly repetitive information, affecting model training efficiency and generalization performance. In contrast, this embodiment uses a single-layer forest structure, avoiding a multi-layer cascaded input concatenation architecture. The low-discrepancy forest classification submodule uses a single-layer forest structure containing 50 decision trees. These 50 decision trees are trained by a random forest classifier and output class probabilities for each input sample. Figure 4 As shown, each decision tree outputs a classification probability for the input features, and these outputs are weighted and fused to obtain a comprehensive probability for determining the target category. The weights of each decision tree are determined by a preset decaying prior weight and the classification accuracy of each decision tree on the training samples. Specifically, let there be a total of The first decision tree, the The prior decay weights of the decision trees are , No. The classification accuracy of the decision trees on the training set is Then the first Fusion weights of decision trees satisfy: ; For input samples , No. The fusion probability of classes satisfies: ; in For the first Decision tree output samples Belongs to the The probability of a class is calculated; when the fusion probability of the target class is greater than or equal to a preset threshold τ, the target class (acute aggravation period) is output; otherwise, another class (stable period) is output. Through the above weight allocation and threshold decision mechanism, the performance difference of a single tree can be directly transformed into the weighted contribution of the final binary classification result, solving the problem that the traditional voting mechanism cannot quantify the difference in tree quality. This achieves a refined and quantifiable weighted fusion effect without increasing model complexity, thus improving classification stability.

[0038] In a preferred embodiment, the low-discrepancy forest classification model can further perform adaptive pruning on each decision tree to further reduce the risk of overfitting. For nodes... The loss function is defined as: The overall loss function for the entire decision tree T is defined as: ;in, Represents a node The training error, Indicates the node The number of leaf nodes in the child tree of the root node. This represents the training error of the decision tree. This represents the total number of leaf nodes in the decision tree. This is a non-negative hyperparameter used to balance the error term and model complexity. Pruning simplifies the tree structure while maintaining classification accuracy, further adapting to resource constraints at the edge.

[0039] The binary classification early warning result is either the acute exacerbation phase or the stable phase of bronchiectasis; when the acute exacerbation phase is output, an early warning message is generated and pushed to the mobile terminal.

[0040] Experimental verification results This invention uses a publicly available database for testing, with a 64% training, 16% validation, and 20% testing component. The experimental results are as follows: (a) Binary confusion matrix (interference judgment): such as Figure 6 As shown, the accuracy rate for interference judgment was 72.2%, the accuracy rate for the aggravation period was 82.4%, and the overall accuracy rate was 82.4%.

[0041] (b) Binary confusion matrix (pneumonia judgment): such as Figure 7 As shown, the accuracy rate for diagnosing pneumonia was 76.5%, the accuracy rate for bronchiectasis was 66.7%, and the overall accuracy rate was 76.5%.

[0042] (c) ROC curve: such as Figure 8 As shown, the AUC for the aggravation phase was 0.7908.

[0043] (d) End-side inference performance: such as Figure 9As shown, this invention achieves 182ms based on ESP32, which is superior to STM32F7 (156ms but without AI capability), RPi Pico (287ms) and cloud API (850ms).

[0044] Example 2 like Figure 10 As shown, this embodiment provides an intelligent risk early warning system for acute exacerbations of bronchiectasis, including: The audio acquisition and framing module is used to acquire the patient's real-time audio signal, and to segment the audio signal into frames to obtain an audio frame sequence. The privacy filtering module is used to input the audio frame sequence into the end-side privacy filtering model, identify whether each audio frame is a cough audio frame, and filter out non-cough audio frames to obtain a set of cough audio frames. The feature extraction module is used to extract four-dimensional acute aggravation discrimination features for each cough audio frame in the cough audio frame set. The four-dimensional features include: fundamental frequency drift rate, peak power spectral slope, time-domain waveform rise / fall ratio, and Mel energy distribution difference. The statistical aggregation module is used to statistically aggregate the four-dimensional features of all cough audio frames within the same audio file to obtain a file-level feature vector. The low-discrepancy forest classification module is used to input the file-level feature vector into the low-discrepancy forest classification model, which consists of multiple decision trees, each of which outputs the class probability of the corresponding input sample; the class probabilities output by each decision tree are weighted and fused to obtain the fusion probability; the fusion probability is compared with a preset threshold to output a binary classification warning result.

[0045] The binary early warning determination includes: acute exacerbation phase and stable phase of bronchiectasis. When an acute exacerbation phase is detected, an early warning message is generated and pushed to the mobile application module.

[0046] The mobile application module communicates with the low-discrepancy forest classification module to display status and push early warning information.

[0047] Example 3 Embodiment 3 of the present invention provides an electronic device.

[0048] An electronic device includes a memory, a processor, and a program stored in the memory and running on the processor. The processor includes, but is not limited to, at least one of a central processing unit (CPU), a graphics processing unit (GPU), a neural network processor (NPU), a tensor processor (TPU), or an artificial intelligence acceleration chip. The program is used to implement the steps in the intelligent risk warning method for acute exacerbation of bronchiectasis as described in Embodiment 1 of the present invention when executing the program.

[0049] The detailed steps are the same as those of the intelligent risk warning method for acute exacerbation of bronchiectasis provided in Example 1, and will not be repeated here.

[0050] Example 4 Embodiment 4 of the present invention provides a computer-readable storage medium.

[0051] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the intelligent risk warning method for acute exacerbation of bronchiectasis as described in Embodiment 1 of the present invention.

[0052] The detailed steps are the same as those of the intelligent risk warning method for acute exacerbation of bronchiectasis provided in Example 1, and will not be repeated here.

[0053] Example 5 Embodiment 5 of the present invention provides a computer program product.

[0054] A computer program product includes software code, wherein the program in the software code performs the steps of the intelligent risk warning method for acute exacerbation of bronchiectasis as described in Embodiment 1 of the present invention.

[0055] The detailed steps are the same as those of the intelligent risk warning method for acute exacerbation of bronchiectasis provided in Example 1, and will not be repeated here.

[0056] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product 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. The solutions in the embodiments of the present invention can be implemented using various computer languages. For example, in one implementation, the methods and systems can be developed based on deep learning frameworks (such as TensorFlow, PyTorch, etc.) and using the Python language. Those skilled in the art will understand that other suitable programming languages ​​or tools can also be used for implementation without departing from the core ideas of the present invention.

[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for intelligent risk early warning of acute exacerbation of bronchiectasis, characterized in that, include: The patient's real-time audio signal is acquired, and the audio signal is divided into frames to obtain an audio frame sequence; The audio frame sequence is input into the end-side privacy filtering model to identify whether each audio frame is a cough audio frame and filter out non-cough audio frames to obtain a set of cough audio frames. For each cough audio frame in the set of cough audio frames, extract four-dimensional acute exacerbation discrimination features, which include: fundamental frequency drift rate, peak power spectral slope, time-domain waveform rise / fall ratio, and Mel energy distribution difference; The four-dimensional features of all cough audio frames within the same audio file are statistically aggregated to obtain a file-level feature vector; The file-level feature vector is input into a low-discrepancy forest classification model, which consists of multiple decision trees, each of which outputs the class probability of the corresponding input sample. The class probabilities output by each decision tree are weighted and fused to obtain a fusion probability. The fusion probability is compared with a preset threshold to output a binary classification warning result.

2. The method according to claim 1, characterized in that, The weighted fusion includes assigning fusion weights to each decision tree, the fusion weights being determined based on the tree's prior decay weights and the tree's classification accuracy on the training set, thereby differentiating the contributions of each decision tree.

3. The method according to claim 2, characterized in that, The allocation method for the fusion weights is as follows: assuming a total of The first decision tree, the The prior decay weights of the decision trees are , No. The classification accuracy of the decision trees on the training set is Then the first Fusion weights of decision trees satisfy: ； For input samples , No. Class fusion probability satisfy: ； in For the first Decision tree output samples Belongs to the The probability of a class; when the fusion probability of the target class is greater than or equal to a preset threshold, the target class is output; otherwise, another class is output.

4. The method according to claim 1, characterized in that, The edge privacy filtering model is a binary classification convolutional neural network used to identify whether an input audio frame is coughing or not. When the identification result is not coughing, the audio frame is cleared or discarded, terminating subsequent feature extraction.

5. The method according to claim 1, characterized in that, The four-dimensional acute exacerbation discriminant features are defined as follows: Fundamental frequency drift rate: used to characterize the magnitude of the fundamental frequency change in cough audio within adjacent time windows; Peak power spectral slope: used to characterize the trend of peak power in the main frequency band as a function of frequency; Time-domain waveform rise / fall ratio: used to characterize the proportional relationship between the duration of the rising edge and the duration of the falling edge of the cough waveform; Mel energy distribution difference: used to characterize the degree of deviation of the Mel band energy distribution from the preset reference distribution.

6. The method according to claim 1, characterized in that, The statistical aggregation of the four-dimensional features of all cough audio frames in the same audio file specifically includes: calculating the mean, standard deviation, and maximum value of each feature across all cough audio frames in the audio file, and concatenating the mean, standard deviation, and maximum value of each feature in order to form a file-level feature vector.

7. An intelligent risk early warning system for acute exacerbation of bronchiectasis, characterized in that, include: The audio acquisition and framing module is used to acquire the patient's real-time audio signal, and to segment the audio signal into frames to obtain an audio frame sequence. The privacy filtering module is used to input the audio frame sequence into the end-side privacy filtering model, identify whether each audio frame is a cough audio frame, and filter out non-cough audio frames to obtain a set of cough audio frames. The feature extraction module is used to extract four-dimensional acute aggravation discrimination features for each cough audio frame in the cough audio frame set. The four-dimensional features include: fundamental frequency drift rate, peak power spectral slope, time-domain waveform rise / fall ratio, and Mel energy distribution difference. The statistical aggregation module is used to statistically aggregate the four-dimensional features of all cough audio frames within the same audio file to obtain a file-level feature vector. The low-discrepancy forest classification module is used to input the file-level feature vector into the low-discrepancy forest classification model, which consists of multiple decision trees, each of which outputs the class probability of the corresponding input sample; the class probabilities output by each decision tree are weighted and fused to obtain the fusion probability; the fusion probability is compared with a preset threshold to output a binary classification warning result.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the steps of the intelligent risk warning method for acute exacerbation of bronchiectasis according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the intelligent risk warning method for acute exacerbation of bronchiectasis according to any one of claims 1 to 6.

10. A computer program product, comprising software code, characterized in that, The program in the software code executes the steps of the intelligent risk warning method for acute exacerbation of bronchiectasis according to any one of claims 1 to 6.

Images