A health condition assessment method based on wavelet packet decomposition

Abstract

The present application relates to the technical field of signal processing, in particular to a health condition evaluation method based on wavelet packet decomposition, comprising: constructing a classification and recognition network model; recording patient breathing sound audio, and preprocessing the breathing sound audio to obtain an audio sequence; extracting features of the audio sequence through a wavelet packet decomposition algorithm to obtain a feature vector; inputting the feature vector into the classification and recognition network model for training to obtain a state change; and assigning scores after comparing the state change with a health state to obtain an evaluation result, wherein the present application evaluates the health condition through the constructed classification and recognition network model based on sound signals, improves the accuracy of the examination result, and solves the problem of low accuracy of the examination result of the existing evaluation method.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a health status assessment method based on wavelet packet decomposition. Background Technology

[0002] When a person breathes, airflow passes through the respiratory tract and alveoli, creating turbulence that causes vibration and produces sound. This sound, transmitted through the lung tissue and chest wall to the body surface, is called breath sounds. Patients with lung diseases may experience abnormal breath sounds due to abnormal lung tissue structure.

[0003] Traditional assessment of abnormal breath sounds relies heavily on auscultation by doctors. However, due to the limited frequency range of human hearing, some important sound signals can be inadvertently lost, leading to biased test results and reducing the accuracy of the findings. Summary of the Invention

[0004] The purpose of this invention is to provide a method for assessing lung health status based on wavelet packet decomposition, which aims to solve the problem of low accuracy of examination results in existing assessment methods.

[0005] To achieve the above objectives, this invention provides a method for assessing lung health status based on wavelet packet decomposition, comprising the following steps:

[0006] S1. Construct a classification and recognition network model;

[0007] S2. Record the patient's respiratory sounds audio and preprocess the respiratory sounds audio to obtain an audio sequence;

[0008] S3. The audio sequence is subjected to feature extraction using the wavelet packet decomposition algorithm to obtain the feature vector;

[0009] S4. Input the feature vector into the classification and recognition network model for training to obtain the state changes;

[0010] S5. After comparing the state changes with the health status, a score is assigned to obtain the evaluation result.

[0011] The specific method for constructing the classification and recognition network model is as follows:

[0012] Constructing neural networks;

[0013] Obtain the audio dataset;

[0014] The neural network is trained and tested using the audio dataset to obtain a classification and recognition network model.

[0015] The neural network in question is an RNN neural network.

[0016] The specific method for obtaining the audio dataset is as follows:

[0017] Access to publicly available audio;

[0018] The open-source audio from the publicly available audio is extracted to obtain an audio dataset.

[0019] The specific method for training and testing the neural network using the audio dataset to obtain the classification and recognition network model is as follows:

[0020] The audio dataset is divided into a training set and a test set;

[0021] The neural network is trained using the training set to obtain a pre-trained model;

[0022] The pre-trained model is tested using the test set. If the test is passed, a classification and recognition network model is obtained.

[0023] The specific method for extracting feature vectors from the audio sequence using the wavelet packet decomposition algorithm is as follows:

[0024] The audio sequence is decomposed into multiple wavelet packet bases.

[0025] Feature vectors are constructed based on multiple wavelet packet basis parameters.

[0026] The specific method for performing wavelet packet decomposition on the audio sequence to obtain multiple wavelet packet bases is as follows:

[0027] After setting the number of wavelet basis decomposition layers of the audio sequence to 3, wavelet packet decomposition is performed on the audio sequence to obtain multiple wavelet packet bases.

[0028] The specific method for constructing the feature vector based on multiple wavelet packet basis parameters is as follows:

[0029] Calculate energy coefficients under multiple wavelet packet bases;

[0030] A feature vector is constructed based on the energy coefficient.

[0031] The specific method for inputting the feature vector into the classification and recognition network model for training to obtain the state change is as follows:

[0032] The feature vector is input into the classification and recognition network model for training to obtain the first data result;

[0033] The first data result is recorded in the buffer, and the feature vector is further trained through the classification and recognition network model to obtain the second data result;

[0034] The state change is obtained by comparing the first data result with the second data result.

[0035] The specific method for comparing the state changes with the health status and assigning scores to obtain the evaluation results is as follows:

[0036] Set the score for breath sounds to 100 out of 100, with a minimum of 0, to obtain the score range.

[0037] Based on the set score range, the cosine similarity between the vector corresponding to the state change and the vector corresponding to the healthy state is calculated to obtain the evaluation result.

[0038] The beneficial effects of this invention are as follows: This invention provides a health status assessment method based on wavelet packet decomposition. The method extracts features from the audio sequence using a wavelet packet decomposition algorithm to obtain feature vectors. For breath sounds, it exhibits better characteristics in the time-frequency domain than Mel-spectrum cepstral coefficients and other higher-order statistical data. The feature vectors are then input into a classification and recognition network model for training to obtain state changes. These state changes are compared with health status and assigned scores to obtain the assessment result. This invention improves the accuracy of examination results by constructing a classification and recognition network model to assess health status through sound signals, thus solving the problem of low accuracy in existing assessment methods. Attached Figure Description

[0039] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of eight energy coefficients.

[0041] Figure 2 This is a flowchart of a health status assessment method based on wavelet packet decomposition provided by the present invention.

[0042] Figure 3 This is a flowchart of constructing a classification and recognition network model.

[0043] Figure 4 This is a flowchart illustrating the process of extracting features from the audio sequence using a wavelet packet decomposition algorithm to obtain feature vectors.

[0044] Figure 5 The flowchart shows how the feature vector is input into the classification and recognition network model for training to obtain the state change process.

[0045] Figure 6 It is a flowchart that compares the changes in the state with the health state and assigns scores to obtain the evaluation results. Detailed Implementation

[0046] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0047] Please see Figures 1 to 6 This invention provides a health status assessment method based on wavelet packet decomposition, comprising the following steps:

[0048] S1 constructs a classification and recognition network model;

[0049] The specific method is as follows:

[0050] S11 constructs a neural network;

[0051] Specifically, the neural network is a recurrent neural network. A recurrent neural network (RNN) is a type of recurrent neural network that takes sequential data as input, recursively processes data in the direction of sequence evolution, and connects all nodes in a chain-like manner. RNNs have applications in natural language processing, such as speech recognition, language modeling, and machine translation, and are also used for various time series forecasting applications.

[0052] S12 retrieves the audio dataset;

[0053] Specifically, obtain publicly available audio; extract open-source audio from the publicly available audio to obtain an audio dataset.

[0054] S13 uses the audio dataset to train and test the neural network to obtain a classification and recognition network model.

[0055] The specific method is as follows:

[0056] S131 divides the audio dataset into a training set and a test set;

[0057] Specifically, the ratio of the training set to the test set is 8:2.

[0058] S132 uses the training set to train the neural network to obtain a pre-trained model;

[0059] S133 uses the test set to test the pre-trained model. If the test is passed, a classification and recognition network model is obtained.

[0060] S2 records the patient's respiratory sounds and preprocesses the respiratory sounds to obtain an audio sequence;

[0061] Specifically, the purpose of preprocessing is to remove noise and, to some extent, acquire the characteristics of the signal, transforming the audio signal into a discrete audio sequence.

[0062] S3 extracts features from the audio sequence using a wavelet packet decomposition algorithm to obtain a feature vector;

[0063] Specifically, after signal processing, different signals can exhibit certain differences in multiple dimensions, and these dimensions that can reflect these differences are selected to form feature vectors. In this invention, wavelet packet decomposition algorithm is chosen for feature extraction.

[0064] The specific method is as follows:

[0065] S31 performs wavelet packet decomposition on the audio sequence to obtain multiple wavelet packet bases;

[0066] Specifically, the audio sequence is decomposed into multiple wavelet packet bases by setting the number of wavelet basis decomposition layers to three. Figure 1 This is a schematic diagram of decomposing the signal into three layers to obtain eight energy coefficients. S represents the original signal, A represents the low-frequency part of the signal, and D represents the high-frequency part of the signal. Decomposing the signal into three layers yields eight different frequency components.

[0067] S32 constructs eigenvectors based on multiple wavelet packet bases.

[0068] Specifically, energy coefficients are calculated under multiple wavelet packet bases; feature vectors are constructed based on the energy coefficients.

[0069] The construction of feature vectors based on the energy coefficients includes: normalizing the energy coefficients (energy coefficients of the optimal basis) Ei (i = 1, 2, 3, ...) obtained after wavelet packet decomposition by comparing them with the total energy E0; constructing a vector [E1 / E0, E2 / E0, ..., E7 / E0] in sequence from the eight normalized values; and feeding the vector into a neural network for classification and recognition.

[0070] S4 inputs the feature vector into the classification and recognition network model for training to obtain the state changes;

[0071] The specific method is as follows:

[0072] S41 The feature vector is input into the classification and recognition network model for training to obtain the first data result;

[0073] Specifically, the feature vector is a multidimensional feature vector.

[0074] S42 records the first data result into the buffer, and continues to train the feature vector through the classification and recognition network model to obtain the second data result;

[0075] S43 compares the first data result with the second data result to obtain the state change.

[0076] S5 compares the state changes with the health status and assigns scores to obtain the evaluation results.

[0077] The specific method is as follows:

[0078] S51 sets the breath sound score to 100 points (maximum) and 0 points (minimum) to obtain the score range.

[0079] S52 calculates the cosine similarity between the vector corresponding to the state change and the vector corresponding to the healthy state based on the set range of the score, and obtains the evaluation result.

[0080] Specifically, the score for breath sounds is set to 100 points, with a minimum of 0 points. The comparison method is to calculate the cosine similarity between the two vectors. Since the range of cosine similarity is [-1, 1], let the cosine similarity be 'a', and assign it a score of 100*(1+a) / 2. The formula for calculating cosine similarity is as follows:

[0081]

[0082] Here, A and B are two vectors, representing the health feature vector and the patient's current breath sound feature vector, respectively. n represents the dimension of the vector, which here corresponds to the number of classes in the classifier. Ai and Bi represent each term of the vector. For example, for the health feature vector A, during model training, the first term of the health data result value can be set to 1, and the remaining terms to 0, i.e., A is [1,0,0,0,0,0]; the prediction result for a specific breath sound is B. The cosine value of the angle θ between the two vectors, cos(θ), is obtained by comparing the inner product of A and B with the product of their magnitudes, i.e., the cosine similarity.

[0083] The beneficial effects of this invention are as follows: The method of this invention can provide health assessments for patients without the need for manual intervention by doctors, and can serve as a reference for doctors' treatment. It makes full use of the patient's respiratory sound signals, taking into account both high-frequency and low-frequency components of the signal, resulting in a more complete and comprehensive signal analysis. Furthermore, by utilizing the self-learning and error-correcting capabilities of neural networks, the accuracy of the analysis results can be continuously improved.

[0084] The above-disclosed embodiments are merely preferred embodiments of the health status assessment method based on wavelet packet decomposition of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art can understand that implementing all or part of the above embodiments and making equivalent changes in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. An evaluation method based on wavelet packet decomposition, characterized in that, Includes the following steps: Construct a classification and recognition network model; Record the patient's respiratory sounds audio and preprocess the respiratory sounds audio to obtain an audio sequence; The feature vector is obtained by extracting features from the audio sequence using the wavelet packet decomposition algorithm; The feature vector is input into the classification and recognition network model for training to obtain the state changes; The changes in the state are compared with the health status and then scored to obtain the evaluation result; The specific method for comparing the state changes with the health status and assigning scores to obtain the evaluation results is as follows: Set the score for breath sounds to 100 out of 100, with a minimum of 0, to obtain the score range; calculate the cosine similarity 'a' between vectors A and B, and assign it a score of 100*(1+a) / 2; Based on the set score range, the cosine similarity between the vector corresponding to the state change and the vector corresponding to the healthy state is calculated to obtain the evaluation result; The formula for calculating cosine similarity is as follows: Where A is the health feature vector, B is the feature vector of the patient's current breath sound, and n is the vector dimension. i and B i This represents each term of the vector.

2. The evaluation method based on wavelet packet decomposition as described in claim 1, characterized in that, The specific method for constructing the classification and recognition network model is as follows: Constructing neural networks; Obtain the audio dataset; The neural network is trained and tested using the audio dataset to obtain a classification and recognition network model.

3. The evaluation method based on wavelet packet decomposition as described in claim 2, characterized in that, The neural network is a BP neural network.

4. The evaluation method based on wavelet packet decomposition as described in claim 3, characterized in that, The specific method for obtaining the audio dataset is as follows: Access to publicly available audio; The open-source audio from the publicly available audio is extracted to obtain an audio dataset.

5. The evaluation method based on wavelet packet decomposition as described in claim 4, characterized in that, The specific method for training and testing the neural network using the audio dataset to obtain the classification and recognition network model is as follows: The audio dataset is divided into a training set and a test set; The neural network is trained using the training set to obtain a pre-trained model; The pre-trained model is tested using the test set. If the test is passed, a classification and recognition network model is obtained.

6. The evaluation method based on wavelet packet decomposition as described in claim 5, characterized in that, The specific method for extracting features from the audio sequence using the wavelet packet decomposition algorithm to obtain the feature vector is as follows: The audio sequence is decomposed into multiple wavelet packet bases. The optimal basis is obtained by screening multiple wavelet packet bases. Construct eigenvectors based on the optimal basis.

7. The evaluation method based on wavelet packet decomposition as described in claim 6, characterized in that, The specific method for performing wavelet packet decomposition on the audio sequence to obtain multiple wavelet packet bases is as follows: After setting the number of wavelet basis decomposition layers of the audio sequence to three, wavelet packet decomposition is performed on the audio sequence to obtain multiple wavelet packet bases.

8. The evaluation method based on wavelet packet decomposition as described in claim 7, characterized in that, The specific method for constructing the feature vector based on the optimal basis is as follows: The energy coefficients are calculated under multiple wavelet packet bases; feature vectors are constructed based on the energy coefficients.

9. The evaluation method based on wavelet packet decomposition as described in claim 8, characterized in that, The specific method for inputting the feature vector into the classification and recognition network model for training to obtain the state change is as follows: The feature vector is input into the classification and recognition network model for training to obtain the first data result; The first data result is recorded in the buffer, and the feature vector is further trained through the classification and recognition network model to obtain the second data result; The state change is obtained by comparing the first data result with the second data result.

Images