A wireless sensing method for quantitatively assessing the degree of upper airway obstruction

Abstract

The application discloses a wireless sensing method for quantitatively evaluating upper airway obstruction degree, and belongs to the technical field of biomedical signal processing. The method comprises the following steps: collecting chest and abdominal movement and respiratory sound signals, and obtaining physiological signals and labels to process the chest and abdominal movement and respiratory sound signals; performing feature extraction on the processed chest and abdominal movement and respiratory sound signals through a recurrent neural network; iteratively training a hypergraph neural network to obtain a trained respiratory parameter quantification model, fusing the extracted features and inputting the features into the respiratory parameter quantification model, outputting ventilation drive and tidal volume prediction results through operation of the respiratory parameter quantification model, and thus obtaining an upper airway obstruction degree result. Compared with the prior art, the method has the advantages that the upper airway obstruction degree quantification method based on hypergraph dual-modal signal fusion is combined with LSTM and self-attention mechanism to extract deep learning features, and the completeness and accuracy of the evaluation result are improved.

Description

Technical Field

[0001] This invention relates to the field of biomedical signal processing technology, and more specifically, to a wireless sensing method for quantitatively assessing the degree of upper airway obstruction. Background Technology

[0002] The quantitative assessment of upper airway obstruction aims to accurately capture and quantify the degree of upper airway collapse during sleep using scientific methods, providing objective evidence for the research and monitoring of obstructive sleep apnea (OSA). OSA is a common sleep disorder characterized by recurrent complete or partial collapse of the upper airway during sleep. Complete collapse leads to apnea, while partial collapse triggers hypoventilation events. The hourly total number of these events, known as the apnea-hypopnea index (AHI), is widely used to assess the severity of OSA.

[0003] However, existing measurement methods such as the apnea-hypopnea index (AHI) and esophageal manometry (EPM) have significant limitations. As research progresses, the limitations of AHI as an assessment standard have become increasingly apparent. It cannot comprehensively reflect the true state of OSA among different individuals. Even with similar AHI values, the severity of airway obstruction and health impacts can vary greatly from individual to individual. Therefore, AHI cannot comprehensively and accurately reflect the assessment results of OSA. To more accurately assess the severity of upper airway obstruction, esophageal manometry (EPM) was developed. It quantifies the degree of upper airway obstruction by measuring pressure changes within the esophagus. However, although EPM is considered an effective method for assessing the severity of obstruction, its invasive nature requires the subject to wear an esophageal tube throughout sleep during the detection and assessment process. This not only causes discomfort for the subject but also affects the accuracy of the assessment results due to the impact on the subject's sleep quality.

[0004] Currently, most non-invasive alternatives developed due to the limitations of EPM are complex and lack sufficient validation, and most detection and assessment methods still use contact measurements, raising questions about their accuracy and reliability. The literature “Mann DL, Terrill PI, Azarbarzin A, et al. Quantifying the magnitude of pharyngeal obstruction during sleep using airflow shape[J]. European Respiratory Journal, 2019, 54(1)” proposes an assessment method that non-invasively quantifies the severity of upper airway obstruction by analyzing the temporal shape characteristics of individual airflow in nocturnal sleep studies. It combines the assessment of airflow and ventilation drive (measured by calibrated intraesophageal diaphragmatic electromyography), and estimates the ratio of respiratory flow to ventilation drive through multivariate regression analysis and by using airflow shape characteristics (such as inspiratory / expiratory duration, flatness, and shovel shape) to assess the severity of upper airway obstruction. While this method theoretically enables automated and non-invasive quantification of upper airway obstruction, its practical application still relies on contact devices (such as oronasal masks and esophageal diaphragmatic electromyography electrodes). This limitation restricts its non-invasive advantage during actual testing. Therefore, despite various attempts and improvements in existing technologies, a truly effective, simple, and accurate non-contact method for assessing the degree of upper airway obstruction without disrupting normal sleep is still lacking. Summary of the Invention

[0005] 1. Technical problems to be solved

[0006] To address the problem of improving the accuracy of upper airway obstruction assessment results in the process of achieving non-contact assessment of upper airway obstruction, this invention provides a wireless sensing method for quantitative assessment of upper airway obstruction. This method can assess the degree of upper airway obstruction through wireless sensing. It is based on a dual-modal signal fusion method of hypergraph for upper airway obstruction quantification and combines LSTM and self-attention mechanism to extract deep learning features. This effectively integrates signal data from different modalities, thereby improving the completeness and accuracy of the assessment results.

[0007] 2. Technical Solution

[0008] The objective of this invention is achieved through the following technical solutions.

[0009] The summary section of this application is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0010] Some embodiments of this application propose a wireless sensing method for quantitatively assessing the degree of upper airway obstruction, in order to solve the technical problems mentioned in the background section above.

[0011] As a first aspect of this application, some embodiments of this application provide a wireless sensing method for quantitatively assessing the degree of upper airway obstruction, comprising the following steps: acquiring chest and abdominal movement and respiratory sound signals, and obtaining physiological signals and tags for processing the chest and abdominal movement and respiratory sound signals; extracting features from the processed chest and abdominal movement and respiratory sound signals using a recurrent neural network; iteratively training a hypergraph neural network to obtain a trained respiratory parameter quantification model; fusing the extracted features and inputting them into the respiratory parameter quantification model; and outputting ventilation drive and tidal volume prediction results by running the respiratory parameter quantification model, thereby obtaining the result of the degree of upper airway obstruction.

[0012] Furthermore, chest and abdominal movement and respiratory sound signals are collected simultaneously; physiological signals and tags are acquired, and the physiological signals are divided into data in time periods according to a fixed duration. Ventilation drive and tidal volume data are obtained based on the physiological signals and tags in each time period.

[0013] Furthermore, the specific process for obtaining ventilation drive and tidal volume data is as follows: physiological signals are divided into time periods of fixed duration in minutes; ventilation drive and tidal volume are calculated based on the airflow signal and label of each time period; ventilation drive data of each time period are fused into ventilation drive time series data, and tidal volume data of each time period are fused into tidal volume time series data, thus obtaining ventilation drive and tidal volume data.

[0014] Furthermore, the ventilation drive and tidal volume data are segmented into fixed respiratory cycles, with each fixed respiratory cycle as a segment. The start and end times of each segment are recorded to obtain segmented ventilation drive and tidal volume data. Based on the start and end times of each segment, the processed chest and abdominal movement and respiratory sound signals are segmented in time synchronization, ensuring that the length of the chest and abdominal movement and respiratory sound signals in each segment is consistent.

[0015] Furthermore, the acquired respiratory sound signals are subjected to noise reduction processing, including framing the acquired respiratory sound signals and performing short-time Fourier transform on each frame of the respiratory sound signal after framing.

[0016] Furthermore, the expression for framing the breathing sound signal is as follows:

[0017] x k (n)=x(n)·w(n-kM);

[0018] In the formula, x k x(n) represents the breathing sound signal in the k-th frame after framing, x(n) represents the breathing sound signal, k represents the current frame number, and k is a natural number; n is the n-th sampling point in the k-th frame; w() is the windowing function; M represents the frame shift.

[0019] Furthermore, the expression for the short-time Fourier transform of the framed breathing sound signal is as follows:

[0020] X k (f)=F{x k (n)};

[0021] In the formula, X k (f) represents the frequency domain of the k-th frame, where f represents the frequency, F{} represents the short-time Fourier transform process, and x k x(n) represents the breathing sound signal in the kth frame after framing, and x(n) represents the breathing sound signal.

[0022] Furthermore, regarding X k (f), the noise power spectrum estimate of the k-th frame is expressed as: The expression is as follows:

[0023]

[0024] In the formula, α is the smoothing factor, α = 0.9; d represents the type of noise;

[0025] Noise is suppressed and the clarity of the breathing sound signal is preserved by using the gain function G(k,f). The expression for the gain function G(k,f) is as follows:

[0026]

[0027] In the formula, p(u|γ(k,f)) is the probability density function of the posterior signal-to-noise ratio; u represents the distribution of noise power; ξ(k,f) is used to characterize the relationship between the noise estimate and the energy of the current signal; γ(k,f) represents the ratio of the current signal power to the noise power.

[0028] Furthermore, the relationship between the noise estimate and the energy of the current signal, ξ(k,f), is expressed as follows:

[0029]

[0030] The expression for the ratio of current signal power to noise power γ(k,f) is as follows:

[0031]

[0032] As a second aspect of this application, some embodiments of this application provide a processing system based on the wireless sensing method for quantitative assessment of upper airway obstruction, including a processing module: acquiring chest and abdominal movement and respiratory sound signals, and obtaining physiological signals and tags to process the chest and abdominal movement and respiratory sound signals; an extraction module: extracting features from the processed chest and abdominal movement and respiratory sound signals through a recurrent neural network; and an output module: iteratively training a hypergraph neural network to obtain a trained respiratory parameter quantification model, fusing the extracted features and inputting them into the respiratory parameter quantification model, and outputting ventilation drive and tidal volume prediction results by running the respiratory parameter quantification model, thereby obtaining the result of the upper airway obstruction degree.

[0033] 3. Beneficial effects

[0034] Compared with the prior art, the advantages of this invention are:

[0035] (1) This scheme integrates chest and abdominal movement and respiratory sound signals to form a dual-modal monitoring system. Its dual-modal characteristics enable the system to obtain physiological information of the assessment object from multiple aspects, realizing information complementarity and comprehensive analysis. Compared with single-modal monitoring methods, the dual-modal system can provide more comprehensive and accurate physiological state assessment and enhance the ability to analyze complex physiological phenomena. In addition, the integration of dual-modal sensors can also improve the robustness of data, effectively cope with interference under different environments and conditions, and improve the reliability and accuracy of monitoring.

[0036] (2) This scheme uses a hypergraph-based dual-modal signal fusion method for OSA upper airway obstruction quantification, and combines LSTM and self-attention mechanism to extract deep learning features. This effectively integrates signal data from different modalities, thereby improving the completeness and accuracy of information. The LSTM network can capture long-term dependencies in time-series signals, while the self-attention mechanism can dynamically adjust the importance of features, making the respiratory parameter quantification model pay more attention to features that are crucial to identifying the degree of upper airway obstruction. The structure of the hypergraph allows the respiratory parameter quantification model to learn features at a higher level, capturing more complex patterns and relationships, improving the prediction accuracy of ventilation drive and tidal volume, and thus improving the accuracy of the assessment results. In addition, compared with traditional contact-based upper airway obstruction monitoring technology, this scheme avoids direct contact with the subject through wireless sensing of radio frequency sensors and audio sensors, reducing the risk of infection for the assessment subjects during the detection process. Attached Figure Description

[0037] Figure 1This is a flowchart of a wireless sensing method for quantitatively assessing the degree of upper airway obstruction in one embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the structure of a respiratory parameter quantification model based on multimodal fusion in one embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the input signal for a respiratory parameter quantification model based on radio frequency sensing of chest and abdominal motion in one embodiment of the present invention.

[0040] Figure 4 This is a schematic diagram of the input signal for a respiratory parameter quantization model based on audio sensing respiratory sound signals in one embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram showing the output results of the respiratory parameter quantification model and the degree of upper airway obstruction in one embodiment of the present invention;

[0042] Figure 6 This is a schematic diagram illustrating the consistency of the degree of obstruction output by the respiratory parameter quantification model in one embodiment of the present invention.

[0043] Figure 7 This is a schematic diagram of the Bland-Altman output of the respiratory parameter quantification model in one embodiment of the present invention. Detailed Implementation

[0044] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0045] like Figure 1 As shown, the present invention provides a wireless sensing method for quantitatively assessing the degree of upper airway obstruction, comprising the following steps: acquiring chest and abdominal movement and respiratory sound signals, and obtaining physiological signals and tags for processing the chest and abdominal movement and respiratory sound signals; extracting features from the processed chest and abdominal movement and respiratory sound signals using a recurrent neural network; iteratively training the hypergraph neural network to obtain a trained respiratory parameter quantification model; fusing the extracted features and inputting them into the respiratory parameter quantification model; and outputting ventilation drive and tidal volume prediction results by running the respiratory parameter quantification model, thereby obtaining the result of the degree of upper airway obstruction.

[0046] Specifically, the wireless sensing method for quantitatively assessing the degree of upper airway obstruction follows this process:

[0047] S1. Collect and process data

[0048] Acquire physiological signals, tags, and chest and abdominal movement and respiratory sound signals, and process them.

[0049] Specifically, data was collected during the subject's entire sleep period. Radiofrequency and audio sensors were used to simultaneously acquire chest and abdominal movement and respiratory sound signals during sleep; polysomnography (PSG) was performed to obtain physiological signals and labels; and based on the physiological signals and labels obtained from PSG, the ventilation drive and tidal volume per breath were calculated.

[0050] Among them, the radio frequency sensor is used to capture the minute movements of the chest and abdomen of the evaluation subject, while the audio sensor is used to record breathing sound signals.

[0051] More specifically, physiological signals include airflow signals and electroencephalogram (EEG) signals. Based on the airflow signals and tags acquired by PSG, the beginning and end of each respiratory cycle are identified, and then the ventilation drive and tidal volume per breath are calculated. Ventilation drive refers to the airflow generation force required during breathing, and tidal volume refers to the amount of gas inhaled or exhaled with each breath. The calculation of ventilation drive involves fitting a ventilation control model (including the effects of chemical substances and arousal substances on ventilation drive) to the ventilation pattern of sleep apnea; the calculation of tidal volume involves calculating the average airflow rate, or tidal volume, for each breath based on the amplitude of the airflow signal.

[0052] In this embodiment, the acquired airflow signal is divided into segments of fixed duration, resulting in data segmented into time periods for analysis, ensuring the continuity and stability of the data analysis. Within each time period, ventilation drive and tidal volume are calculated based on the airflow signal and the tag.

[0053] In one specific embodiment, the fixed-duration segmentation process divides the time into 7-minute intervals. Ventilation drive and tidal volume are calculated based on the airflow signal for each interval. The ventilation drive data from each interval are then concatenated to form a complete overnight ventilation drive time series. Similarly, the tidal volume data from each interval are concatenated to form a complete overnight tidal volume time series. This yields breath-by-breath ventilation drive and tidal volume data for the entire night.

[0054] A label is a marker associated with the physiological signals of the subject being assessed, used to indicate a specific physiological state or event. Labels are used to identify sleep-disordered breathing, specifically respiratory events related to upper airway obstruction. In this embodiment, labels include sleep stage labels, respiratory event labels, and other physiological state labels. Sleep stage labels identify sleep stages, such as non-rapid eye movement (NREM) and rapid eye movement (REM) sleep. These labels help analyze the impact of different sleep stages on the degree of upper airway obstruction. Respiratory event labels identify respiratory events, such as apnea and hypoventilation. Other physiological state labels include other physiological state labels related to breathing or sleep, such as heart rate variability and blood oxygen saturation.

[0055] Tags were obtained through a polysomnography (PSG) monitoring system. PSG is a comprehensive sleep monitoring technology that can simultaneously record multiple physiological signals of the subject being assessed and automatically or semi-automatically generate corresponding tags based on these signals. Obtaining tags provides important reference information for subsequent data analysis and processing.

[0056] In one specific embodiment, the acquisition process of physiological signals and tags is as follows: During sleep, a PSG monitoring system is used to simultaneously collect various physiological signals, including airflow signals, electroencephalogram (EEG), eye movement (EOG), mandibular electromyography (EMG), and electrocardiogram (ECG). Based on the PSG monitoring results, sleep stages (such as non-rapid eye movement (NREM) and rapid eye movement (REM)) and respiratory events (such as apnea and hypoventilation) are labeled as tags for subsequent analysis. The acquisition process of chest and abdominal movement and respiratory sound signals is as follows: Radio frequency sensors (such as RFID) are used to monitor chest and abdominal movement; simultaneously, audio sensors (such as microphone arrays) are used to capture respiratory sound signals. During the acquisition process, it is ensured that the sensors and monitoring equipment work synchronously to obtain time-consistent signal data.

[0057] Specifically, such as Figure 2 As shown, noise reduction processing is performed on the collected respiratory sound signals to reduce background noise interference.

[0058] like Figure 3 The diagram shows the input signal of the respiratory parameter quantization model based on radio frequency sensing for chest and abdominal movement in this embodiment. The vertical axis represents the normalized amplitude of chest and abdominal movement. The input respiratory sound signal is subjected to a short-time Fourier transform (STFT) using the OMLSA method to obtain frequency domain information. The optimal gain function is obtained by calculating the noise power spectrum and the respiratory sound signal power spectrum, and their ratios. The spectrum is then scaled according to the gain coefficient to reduce noise components. After processing, the respiratory sound signal is restored to the time domain using an inverse short-time Fourier transform (ISTFT). Combined with... Figure 2It can be seen that this noise reduction process can effectively suppress background noise and improve the clarity of breathing sound signals.

[0059] In one specific embodiment, the acquired respiratory sound signal x(n) is first processed by framing, and a short-time Fourier transform (STFT) is performed on each frame of the respiratory sound signal.

[0060] First, the expression for framing the breathing sound signal x(n) is as follows:

[0061] x k (n)=x(n)·w(n-kM);

[0062] In the formula, x k (n) represents the breathing sound signal of the k-th frame after framing, where k represents the current frame number and is a natural number; n represents the n-th sampling point in the k-th frame; w() is the windowing function; M represents the frame shift.

[0063] Secondly, the expression for the short-time Fourier transform of the framed breathing sound signal is as follows:

[0064] X k (f)=F{x k (n)};

[0065] In the formula, X k (f) represents the frequency domain of the k-th frame, where f represents the frequency, and F{} represents the short-time Fourier transform process.

[0066] For X k (f) An adaptive noise estimation algorithm is used to achieve dynamic estimation of the noise power spectrum.

[0067] Specifically, the noise power spectrum estimate for the k-th frame is expressed as: Its expression is as follows:

[0068]

[0069] In the formula, α is a smoothing factor, and in this embodiment, α is 0.9; d represents the type of noise, and in this embodiment, d is background noise.

[0070] According to x k (n) and The resulting value is further processed using a gain function G(k,f) to suppress noise while preserving the clarity of the breathing sound signal. The expression for the gain function G(k,f) is as follows:

[0071]

[0072] In the formula, p(u|γ(k,f)) is the probability density function of the posterior signal-to-noise ratio; u represents the distribution of noise power;

[0073] ξ(k,f) is used to characterize the relationship between the noise estimate and the energy of the current signal, and its expression is as follows:

[0074]

[0075] γ(k,f) represents the ratio of the current signal power to the noise power, and its expression is as follows:

[0076]

[0077] The enhancement of the respiratory sound spectrum is performed by targeting the gain function. The expression for the enhancement process is as follows:

[0078] |Y k (f)|=G(k,f)·|X k (f)|;

[0079] In the formula, Y k (f) represents the enhanced frequency domain signal.

[0080] For the enhanced frequency domain signal Y k (f) Perform inverse short-time Fourier transform, overlap and add the obtained time-domain frame signals to reconstruct the time-domain denoised breathing sound signal. The expression is as follows:

[0081]

[0082] Through the above steps, processed chest and abdominal movement and respiratory sound signals are obtained.

[0083] Subsequently, the obtained overnight, breath-by-breath ventilatory drive and tidal volume data were segmented. Each segment consisted of 10 respiratory cycles, and the start and end times of each segment were recorded to obtain segmented ventilatory drive and tidal volume data. Based on the recorded start and end times of each segment, the processed chest and abdominal movements and respiratory sound signals were time-synchronized and segmented, ensuring that the length of the chest and abdominal movements and respiratory sound signals in each segment was consistent. Figure 5 As shown, the chest and abdominal movements are synchronized with the respiratory sound signals in terms of time and duration.

[0084] Based on the obtained segmented ventilation drive and tidal volume data, let P be... j It is the ventilation drive of segment j, V j This represents the tidal volume data for segment j, where j is a natural number. Each ventilation drive segment contains 10 ventilation drive values, and each tidal volume data segment contains 10 tidal volume values. For each segment P... j and Vj Start time j and End Time k They are represented as follows:

[0085] Start k =t 10j+1 ;

[0086] End j =t 10j+10 ;

[0087] Where t represents time. For the obtained start and end times, within the time range [Start...] j End j Within this section, the signals of chest and abdominal movement and respiratory sounds are extracted. The process expressions for extracting these signals are as follows:

[0088] R j (t)={R(t)t∈[Start j End j ]};

[0089] A j (t)={A(t)t∈[Start j End j ]};

[0090] Among them, R j (t) represents the chest and abdominal movements of the j-th segment, A j (t) is the breathing sound signal of segment j.

[0091] In a specific embodiment, after obtaining the segmented chest and abdominal motion signals and the noise-reduced respiratory sound signals through the above process, signal interpolation or clipping is performed on the segmented chest and abdominal motion signals and the respiratory sound signals to make the segmented chest and abdominal motion signals and the respiratory sound signals have the same length, thereby ensuring that the different signals have consistency in the time dimension.

[0092] Interpolation involves inserting known data points between two known points to estimate unknown data points. In signal processing, interpolation can construct a continuous or approximately continuous function from known discrete signal points to estimate values ​​between or outside of those points. Interpolation can refine a signal along the time axis, ensuring different signals have the same length at the same time resolution. Clipping, on the other hand, directly removes data points from a signal to achieve the desired length. Clipping can reduce a signal to the same length as other signals for subsequent analysis or comparison.

[0093] Interpolating or cropping segmented chest and abdominal motion signals to the same length as respiratory sound signals is to ensure consistency and comparability of these signals in subsequent analysis, comparison, or fusion.

[0094] Specifically, since an adult's breath lasts 3-5 seconds, this embodiment selects to interpolate or trim the segmented chest and abdominal movements and respiratory sound signals to a length of 40 seconds. Segmented chest and abdominal movements With respiratory sound signals The final expression is:

[0095]

[0096] in, This refers to the chest and abdominal movements after interpolation. `Interpolate()` represents the interpolation operation, and `length()` represents the length of the chest and abdominal movements or respiratory sound signal before interpolation or clipping. It is a chest and abdominal exercise that has been tailored. It is the breath sound signal after interpolation. It is a edited version of the breathing sound signal.

[0097] S2, Feature Extraction

[0098] Feature extraction of processed chest and abdominal movement and respiratory sound signals is performed using a recurrent neural network.

[0099] Specifically, the chest and abdominal movement and respiratory sound signals of consistent length obtained in step S1 are input into a Long Short-Term Memory (LSTM) network incorporating a self-attention mechanism to extract deep features f. thorax-abdomen with f sound ;

[0100] Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network (RNN) used to process and predict long-term dependencies in sequential data. LSTMs overcome the vanishing and exploding gradient problems common in traditional RNNs during long sequence training by effectively controlling the inflow, forgetting, and output of information through the introduction of memory units and gating mechanisms (including input gates, forget gates, and output gates). This makes LSTMs particularly suitable for processing signals with temporal order. In this embodiment, LSTM is used to process chest and abdominal movement and respiratory sound signals to extract key features while preserving contextual information.

[0101] Self-attention dynamically focuses on the relationships between different parts of a sequence when processing sequential data. By calculating the correlation between each element in the input sequence and other elements, self-attention can assign different weights to each element, thus focusing on the parts most important to the current task. Combining LSTM and self-attention can effectively extract deep features f from chest and abdominal movements and respiratory sound signals. thorax-abdomen with f sound This allows for a more comprehensive understanding of the information within complex signals.

[0102] In one specific embodiment, segmented chest and abdominal exercises are targeted. and breathing sound signals Normalization was performed to obtain normalized chest and abdominal movement and respiratory sound signals. Its expression is:

[0103]

[0104] In the formula, μ R Mean value of chest and abdominal movements, σ R It is the standard deviation of chest and abdominal movements, μ A It is the mean of the respiratory sound signal, σ A is the standard deviation of the respiratory sound signal, and [] is used to represent the normalization result of chest and abdominal movements and respiratory sound signals.

[0105] Will The input is fed into an LSTM for processing. Divide the data into multiple time series samples, define the time step as t1, and then the input for each time step is x. t1 The computation process in LSTM is as follows:

[0106] Input Gate:

[0107] Forgotten Gate:

[0108] Candidate memory units:

[0109] Update memory units:

[0110] Output gate:

[0111] Final output:

[0112] Where h represents the hidden state, This represents the hidden state at the current time step... This represents the hidden state of the previous time step, that is, the output at time step t1-1; This represents the cell state at the previous time step, i.e., the memory information at time step t1-1; tanh() is the activation function. Indicates the current cell state, combining information from the forget gate and the input gate; W i W f W c W o b is the weight matrix, used for linear transformation of the input data and hidden states; i b f b C b o Both represent paranoia vectors; i t f represents the activation value of the input gate; t Indicates the activation value of the forget gate; o t σ represents the activation value of the output gate; σ represents the Sigmoid activation function; tanh represents the hyperbolic tangent activation function; ⊙ represents the element-wise multiplication operation.

[0113] After processing all time steps, the hidden state outputs of chest and abdominal movement and respiratory sound signals can be obtained, with the hidden state output H of chest and abdominal movement being... thorax-abdomen And the hidden state output H of the breathing sound signal sound .

[0114] For the hidden state outputs of the obtained chest and abdominal movement and respiratory sound signals, a self-attention mechanism is applied respectively. First, a linear transformation is used, i.e., using the weight matrix W. Q W K W V The hidden state outputs of chest and abdominal movement and respiratory sound signals are mapped to queries (Q), keys (K), and values ​​(V), and the expressions for this process are as follows:

[0115] Q=(H thorax-abdomen / H sound W Q ;

[0116] K = (H thorax-abdomen / H sound W K ;

[0117] V=(H thorax-abdomen / H sound W V ;

[0118] Next, calculate the attention score Attention(Q,K,V):

[0119]

[0120] Where, d kIt represents the dimension of the key, and softmax() is the activation function.

[0121] Therefore, the deep features f in the chest and abdominal movements after attention calculation can be obtained. thorax-abdomen And deep features f in the respiratory sound signal after attention calculation siund , is represented as:

[0122] f thorax-abdomen =Attention(Q,K,V);

[0123] f sound =Attention(Q,K,V);

[0124] This allows us to obtain features extracted from chest and abdominal movements and respiratory sound signals.

[0125] S3. Calculate and output the results.

[0126] The hypergraph neural network is iteratively trained to obtain a trained respiratory parameter quantification model. The extracted features are fused and input into the respiratory parameter quantification model. By running the respiratory parameter quantification model, the ventilation drive and tidal volume prediction results are output, thereby obtaining the result of the degree of upper airway obstruction.

[0127] Specifically, f obtained in step S2 thorax-abdomen with f sound The data are spliced ​​together and input into a pre-trained respiratory parameter quantification model in a hypergraph neural network to obtain the ventilation driving force and tidal volume output by the respiratory parameter quantification model, thereby obtaining the result of the degree of upper airway obstruction.

[0128] Hypergraph Neural Network (HGNN) is a network architecture that extends the traditional Graph Neural Network (GNN) to more effectively process data with hyperedge structures. Unlike traditional graphs where each edge connects only two nodes, in a hypergraph neural network, the degree of each edge is an arbitrary non-negative integer, and the same hyperedge can connect different vertices to represent correlations between connected vertices. This allows for better modeling of complex relationships between data.

[0129] Specifically, the process of using a hypergraph neural network to predict ventilation driving force and tidal volume is as follows:

[0130] By using a hypergraph neural network and leveraging the correlation of bimodal data, corresponding hyperedge groups H1 and H2 are constructed, and the hyperedge groups are concatenated to obtain the correlation matrix H.

[0131] The correlation matrix H is a matrix representation of a graph, used to represent the correlation between vertices and hyperedges in a hypergraph neural network, expressed as follows:

[0132] H = [H1; H2];

[0133] The correlation matrix H consists of a set of nodes and a set of hyperedges, where each hyperedge e can connect multiple nodes v; for each node in the hypergraph, it is initialized with the concatenated feature h, expressed as:

[0134]

[0135] In the formula: v represents a vertex of the hypergraph; e represents a hyperedge. If a vertex is contained in a hyperedge, the corresponding position in the incidence matrix is ​​marked as 1; if a vertex is not contained in a hyperedge, the corresponding position in the incidence matrix is ​​marked as 0.

[0136] Then, the association matrix in the hypergraph and the node features contained in the dataset are fed into the hyperedge convolutional layer for convolution operations to learn the latent features of the high-dimensional data. The expression for the convolution operation is as follows:

[0137] X (0) =X∈R N×C ;

[0138]

[0139] In the formula: X (0) X represents the input after dual-modal branch feature fusion; (l+1) X represents the feature matrix at layer l+1; (l) D represents the feature matrix at layer l; σ is the activation function; D v and D e Indicates normalization; θ (l) H represents the filtering matrix, W represents the correlation matrix, N represents the diagonal matrix of the weights of each hyperedge, C represents the number of nodes, and C represents the feature dimension of each node.

[0140] The nodes of the last layer of the hypergraph neural network are used to generate predicted values ​​for each segment, and the prediction results are ventilation-driven. and tidal volume Prediction results The expression is as follows:

[0141]

[0142] The predicted results are constrained using a loss function, and then the hypergraph neural network is iteratively trained to obtain a trained respiratory parameter quantification model. The loss function is expressed as follows:

[0143]

[0144] In the formula: L MSE-pL represents the loss function for a ventilation-driven prediction task. MSE-v The loss function represents the tidal volume prediction task; N represents the number of samples of ventilation drive and tidal volume at input and output. In this embodiment, N = 10.

[0145] Finally, the chest and abdominal movements and respiratory sound signals to be tested are put into the trained respiratory parameter quantification model, and the ventilation drive and tidal volume prediction results are finally output to accurately assess the degree of upper airway obstruction of the subject.

[0146] The ratio of tidal volume to ventilation drive in each respiratory cycle is calculated using the following expression:

[0147]

[0148] Specifically, the ratio of tidal volume to ventilation drive in each respiratory cycle, obtained through a wireless sensing method that quantifies the degree of upper airway obstruction, can serve as a key indicator for assessing the degree of upper airway obstruction in obstructive sleep apnea (OSA). The degree of upper airway obstruction is assessed by judging the ratio of tidal volume to ventilation drive in each respiratory cycle; if the ratio is >90%, the degree of upper airway obstruction is normal; if the ratio is 70%–90%, the degree of upper airway obstruction is mild; if the ratio is 50%–70%, the degree of upper airway obstruction is moderate; if the ratio is 30%–50%, the degree of upper airway obstruction is severe; and if the ratio is <30%, the degree of upper airway obstruction is very severe.

[0149] This approach integrates chest and abdominal movement signals with respiratory sound signals, forming a dual-modal monitoring system. Its dual-modal characteristics allow for the acquisition of physiological information from multiple perspectives, enabling complementary and comprehensive analysis. Compared to single-modal monitoring methods, this dual-modal approach provides a more comprehensive and accurate assessment of physiological states, enhancing the ability to interpret complex physiological phenomena. Furthermore, the integration of dual-modal sensors improves data robustness, effectively addressing interference under different environmental conditions and enhancing the reliability and accuracy of the assessment results.

[0150] In a specific embodiment, when the ratio of tidal volume to ventilation drive is >90%, it indicates sufficient ventilation drive and no significant airway obstruction, so the assessment result is output as normal; when the ratio of tidal volume to ventilation drive is 70% to 90%, it indicates mild airway obstruction, so the assessment result is output as mild; when the ratio of tidal volume to ventilation drive is 50% to 70%, it indicates significant airway obstruction, so the assessment result is output as moderate; when the ratio of tidal volume to ventilation drive is 30% to 50%, it indicates severe airway obstruction and significantly affected breathing, so the assessment result is output as severe; when the ratio of tidal volume to ventilation drive is <30%, it indicates extremely severe airway obstruction and a very high risk of restricted breathing, so the assessment result is output as very severe.

[0151] In a specific embodiment, such as Figure 4 The diagram shows the input signal of the respiratory parameter quantification model based on audio sensing respiratory sound signals in this embodiment. The vertical axis represents the normalized amplitude of the respiratory sound signal. It illustrates the results of ventilation drive and tidal volume, reflecting the degree of upper airway obstruction of the assessed subject within the corresponding time period, i.e., the ratio of ventilation drive to tidal volume. Furthermore, it can be seen that the assessed subject experienced a process from open to obstructed upper airway, with a mild degree of obstruction. Additionally, during the non-obstructive period, the ventilation drive level was higher, indicating that the assessed subject was experiencing obstructive apnea or hypoventilation.

[0152] like Figures 3 to 5 As shown, the respiratory parameter quantification model designed by the wireless sensing method for quantifying the degree of upper airway obstruction according to the present invention inputs a signal into the respiratory parameter quantification model and runs the respiratory parameter quantification model to output the result. The output result of the respiratory parameter quantification model reflects the degree of upper airway obstruction.

[0153] like Figures 6 to 7 The image shows the consistency of obstruction severity obtained by running a respiratory parameter quantification model and the Bland-Altman results. Among them, Figure 7 This is a Bland-Altman diagram of the respiratory parameter quantification model output in one embodiment of the present invention. A Bland-Altman diagram, also known as a consistency analysis diagram, is a graphical method that visually reflects the consistency limits between two sets of measurement results. Its basic idea is to calculate the difference and average of the two sets of measurement results, and then visually display the relationship between these differences and the average, thereby assessing the consistency between the two sets of data.

[0154] In one specific embodiment, the processing system of the wireless sensing method for quantifying the degree of upper airway obstruction includes:

[0155] Processing module: Collects chest and abdominal movement and respiratory sound signals, and acquires physiological signals and tags to process the chest and abdominal movement and respiratory sound signals;

[0156] Extraction module: Features are extracted from the processed chest and abdominal movement and respiratory sound signals using a recurrent neural network;

[0157] Output module: Iteratively trains the hypergraph neural network to obtain a trained respiratory parameter quantification model, fuses the extracted features and inputs them into the respiratory parameter quantification model, and outputs the ventilation drive and tidal volume prediction results by running the respiratory parameter quantification model, thereby obtaining the result of the degree of upper airway obstruction.

[0158] This approach utilizes a hypergraph-based dual-modal signal fusion method for quantifying upper airway obstruction. By combining LSTM and self-attention mechanisms to extract deep learning features, it effectively integrates signal data from different modalities, thereby improving the completeness and accuracy of the information. Secondly, the LSTM network captures long-term dependencies in time-series signals, while the self-attention mechanism dynamically adjusts the importance of features, allowing the respiratory parameter quantification model to focus more on features crucial for identifying the degree of upper airway obstruction. The hypergraph structure allows the respiratory parameter quantification model to learn features at a higher level, capturing more complex patterns and relationships, improving the prediction accuracy of ventilation drive and tidal volume, and thus enhancing the accuracy of the assessment results.

[0159] The invention and its embodiments have been described above illustratively. This description is not restrictive, and the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. The accompanying drawings are only one embodiment of the invention, and the actual structure is not limited thereto. No reference numerals in the claims should limit the scope of the claims. Therefore, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the invention, such design should fall within the scope of protection of this patent. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Multiple elements stated in the product claims may also be implemented by a single element through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A wireless sensing method for quantitatively assessing the degree of upper airway obstruction, comprising the following steps: Acquire chest and abdominal movement and respiratory sound signals, and obtain physiological signals and tags for processing of chest and abdominal movement and respiratory sound signals; Feature extraction of processed chest and abdominal movement and respiratory sound signals was performed using a recurrent neural network; The hypergraph neural network is iteratively trained to obtain a trained respiratory parameter quantification model. The extracted features are fused and input into the respiratory parameter quantification model. By running the respiratory parameter quantification model, the ventilation drive and tidal volume prediction results are output, thereby obtaining the result of the degree of upper airway obstruction. The noise reduction process for the collected respiratory sound signals includes framing the collected respiratory sound signals and performing a short-time Fourier transform on each frame of the respiratory sound signal after framing. The expression for framing the breathing sound signal is as follows: ； In the formula, Indicates the first frame after frame division The breathing sound signal of the frame, Indicates a breathing sound signal. Indicates the current frame number. It is a natural number; It is the kth frame. One sampling point; It is a windowing function; Indicates frame shift; The expression for the short-time Fourier transform of the framed breathing sound signal is as follows: ； In the formula, Indicates the first The frequency domain of a frame Represents frequency, This represents the process of short-time Fourier transform. Indicates the first frame after frame division The breathing sound signal of the frame, Indicates a breathing sound signal; against , No. The noise power spectrum estimate of the frame is expressed as follows: The expression is as follows: ； In the formula, It is a smoothing factor. =0.9; Indicate the type of noise; Through the gain function To suppress noise and preserve the clarity of the breath sound signal, the gain function... The expression is as follows: ； In the formula, It is the probability density function of the posterior signal-to-noise ratio; Represents the distribution of noise power; The relationship between the noise estimate and the energy of the current signal is used to characterize the noise. This represents the ratio of the current signal power to the noise power. Relationship between noise estimation and the energy of the current signal The expression is as follows: ； The ratio of current signal power to noise power The expression is as follows: ； The enhancement of the respiratory sound spectrum is performed by targeting the gain function. The expression for the enhancement process is as follows: ； In the formula, This represents the enhanced frequency domain signal; For the enhanced frequency domain signal Perform an inverse short-time Fourier transform, then overlap and sum the resulting time-domain frame signals to reconstruct the denoised breathing sound signal in the time domain. The expression is as follows: 。 2. The wireless sensing method for quantitative assessment of upper airway obstruction as described in claim 1, characterized in that, The acquisition of chest and abdominal movement and respiratory sound signals was carried out simultaneously. Physiological signals and tags were acquired, and the physiological signals were divided into data in time periods according to a fixed duration. Ventilation drive and tidal volume data were acquired based on the physiological signals and tags in each time period.

3. The wireless sensing method for quantitative assessment of upper airway obstruction as described in claim 2, characterized in that, The specific process for obtaining ventilation drive and tidal volume data is as follows: physiological signals are divided into time periods of fixed duration in minutes, and ventilation drive and tidal volume are calculated based on the airflow signal and label of each time period. Ventilation drive data for each time period is fused into ventilation drive time series data, and tidal volume data for each time period is fused into tidal volume time series data, thus obtaining ventilation drive and tidal volume data.

4. The wireless sensing method for quantitative assessment of upper airway obstruction degree according to claim 3, characterized in that, The ventilation drive and tidal volume data were segmented into fixed respiratory cycles, with each fixed respiratory cycle as a segment. The start and end times of each segment were recorded to obtain the segmented ventilation drive and tidal volume data. Based on the start and end times of each segment, the processed chest and abdominal movement and respiratory sound signals are segmented in time synchronization, ensuring that the length of the chest and abdominal movement and respiratory sound signals in each segment is consistent.

5. A processing system based on the wireless sensing method for quantitative assessment of upper airway obstruction as described in any one of claims 1-4, characterized in that, Includes a processing module: acquiring chest and abdominal movement and respiratory sound signals, and obtaining physiological signals and tags for processing chest and abdominal movement and respiratory sound signals; Extraction module: Features are extracted from the processed chest and abdominal movement and respiratory sound signals using a recurrent neural network; Output module: Iteratively trains the hypergraph neural network to obtain a trained respiratory parameter quantification model, fuses the extracted features and inputs them into the respiratory parameter quantification model, and outputs the ventilation drive and tidal volume prediction results by running the respiratory parameter quantification model, thereby obtaining the result of the degree of upper airway obstruction.

Images