Breathing monitoring and early warning method for breathing critical patient

Through multimodal data acquisition and analysis, the problem of insufficient single modal data acquisition is solved, comprehensive and accurate monitoring and abnormal detection of respiratory status are achieved, and clinical intervention efficiency and patient safety are improved.

CN119055225APending Publication Date: 2024-12-03CHUANGRUN LIANKE (QINGDAO) MEDICAL EQUIPMENT CO LTD

Patent Information

Application Number
CN202411306862.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-19
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Due to insufficient single-modal data acquisition and data fusion analysis, existing respiratory monitoring methods and equipment have reduced monitoring accuracy when the patient's position changes, breathing mode is unstable or environmental interference, and cannot accurately reflect the patient's actual respiratory status, affecting the monitoring. Accuracy and continuity.

Method used

Using multimodal data acquisition and preprocessing, including imaging equipment, audio sensors and bioimpedance sensors to synchronize data acquisition, and real-time monitoring through automatic attitude recognition and correction, multimodal feature extraction and fusion, abnormal detection and early warning generation. and analyze the patient's respiratory status, and generate comprehensive respiratory characteristic parameters to detect abnormalities.

Benefits of technology

Improve the accuracy and continuity of respiratory monitoring, reduce data deviations caused by posture changes, enhance the ability to adapt to patients' dynamic status, ensure timely identification and early warning of respiratory abnormalities, improve the efficiency of clinical intervention and patient safety sex.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119055225A_ABST
    Figure CN119055225A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of respiration monitoring, in particular to a respiration monitoring and early warning method for a critical respiration patient, which comprises the following steps: S1, multi-modal data acquisition and preprocessing: acquiring respiration image data, respiration audio data and biological impedance data of the patient; s2, automatically identifying and correcting the posture of the patient: correcting the posture of the multi-mode breathing signal data; s3, multi-modal feature extraction and fusion: extracting breathing feature parameters of each modal to obtain comprehensive breathing feature parameters; s4, anomaly detection: analyzing through a breathing anomaly detection model, and detecting whether breathing of the patient is abnormal or not; and S5, early warning generation: reminding a user to take corresponding intervention measures when abnormal breathing is detected. According to the invention, the adaptive capacity of the system to the dynamic state of the patient is improved, the monitoring error is reduced under the condition of frequent posture change, and continuous breathing data monitoring is ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of respiratory monitoring, and in particular to a respiratory monitoring and early warning method for patients with severe respiratory illness. Background Art

[0002] With the development of medical technology, the monitoring and management of patients with critical respiratory diseases has become an important part of clinical treatment. Respiratory monitoring helps medical staff detect respiratory abnormalities in a timely manner by collecting data such as the patient's respiratory rate, respiratory depth, and respiratory rhythm, so as to take appropriate intervention measures. Current respiratory monitoring equipment usually uses single-modality data, such as ventilator pressure sensors, blood oxygen monitors, or pulmonary function testers, to assess the patient's respiratory condition. These devices are mainly used in intensive care units or emergency environments to provide support for clinical treatment by continuously monitoring the patient's respiratory function.

[0003] However, existing respiratory monitoring methods and equipment have many limitations. A single-modality data source often cannot fully and accurately reflect the patient's actual respiratory status. In particular, when the patient's body position changes, the breathing pattern is unstable, or there is environmental interference, the monitoring accuracy will be significantly reduced. In addition, the existing technology has limited fusion analysis of multimodal data and cannot effectively integrate data from different sensors (such as images, audio, bioimpedance, etc.), which affects the reliability of abnormality detection. Especially in the dynamic monitoring of critically ill respiratory patients, factors such as posture changes often lead to data deviations, thereby affecting the accuracy and continuity of monitoring.

[0004] In order to address the deficiencies of the prior art, the present invention proposes a respiratory monitoring and early warning method for patients with severe respiratory illness, which can improve the efficiency of clinical intervention and the safety of patients, and overcome the deficiencies of the prior art in monitoring accuracy and data fusion. Summary of the invention

[0005] The present invention provides a respiratory monitoring and early warning method for patients with severe respiratory diseases.

[0006] A respiratory monitoring and early warning method for patients with severe respiratory illness, comprising the following steps:

[0007] S1, multimodal data acquisition and preprocessing: using a camera, an audio sensor and a bioimpedance sensor to synchronously acquire the patient's respiratory image data, respiratory audio data and bioimpedance data to form multimodal respiratory signal data, and preprocessing the multimodal respiratory signal data, including denoising and signal enhancement;

[0008] S2, automatic identification and correction of patient posture: Analyze the pre-processed multi-modal respiratory signal data, identify the patient's posture information in real time, including lying flat, lying on the side, and sitting up, and perform posture correction on the multi-modal respiratory signal data based on the identified posture results to compensate for the data deviation caused by posture changes;

[0009] S3, multimodal feature extraction and fusion: based on the corrected multimodal respiratory signal data, extract the respiratory feature parameters of each modality, including image features, audio features and bioimpedance features, and fuse the respiratory feature parameters of each modality to obtain comprehensive respiratory feature parameters;

[0010] S4, abnormality detection: Based on the integrated respiratory feature parameters obtained by fusion, the respiratory abnormality detection model is used to analyze whether the patient's breathing is abnormal, including abnormal respiratory rate, disordered respiratory rhythm, and insufficient respiratory depth;

[0011] S5, warning generation: When abnormal breathing is detected, the monitoring equipment will immediately inform the medical staff by means of sound alarms, visual prompts or remote notifications, reminding them to take appropriate intervention measures.

[0012] Optionally, the multimodal data collection and preprocessing in S1 includes:

[0013] S11, data acquisition: using a camera to collect respiratory image data of the patient's chest and abdomen, using an audio sensor to collect the patient's respiratory audio data, and using a bioimpedance sensor to monitor the patient's lung volume changes and respiratory impedance;

[0014] S12, denoising processing: performing denoising processing on the collected multi-modal respiratory signal data, including:

[0015] Image data denoising: denoising is performed using a denoising algorithm based on a Gaussian filter;

[0016] Audio data denoising: A bandpass filter is used to remove environmental noise from the breathing audio signal, and the frequency range of the bandpass filter is set to 0.1Hz to 1kHz;

[0017] Bioimpedance data denoising: The Kalman filter algorithm is used to denoise the bioimpedance data;

[0018] S13, signal enhancement: performing signal enhancement on the denoised multimodal respiratory signal data, including:

[0019] Image data enhancement: The histogram equalization algorithm is used to enhance the contrast of the respiratory image by adjusting the grayscale distribution of the image;

[0020] Audio data enhancement: Spectral enhancement of breathing audio signals is performed, and high-frequency components are amplified using a pre-emphasis filter;

[0021] Bioimpedance data enhancement: A sliding window mean filter is used to smooth and enhance the bioimpedance data.

[0022] Optionally, the automatic recognition and correction of the patient posture in S2 includes:

[0023] S21, posture recognition: Analyze the preprocessed multimodal respiratory signal data and use convolutional neural network (CNN) to perform posture recognition on the patient's image data, including lying flat, lying on the side, and sitting up;

[0024] S22, posture change detection and tracking: Based on the patient's posture recognition, the patient's posture changes are continuously monitored, and the trend of posture changes is tracked in real time by using image difference detection and time series analysis between consecutive frames;

[0025] S23, multimodal data correction: based on the real-time recognized posture information, the influence of posture changes on image data, audio data and bio-impedance data is calculated, and the posture compensation model is applied to correct the data deviation.

[0026] Optionally, the gesture recognition in S21 includes:

[0027] S211, posture recognition: using convolutional neural network (CNN) to perform posture recognition on the patient's image data, and extracting features from the input image data through the convolution layer;

[0028] S212, Posture classification and output: The convolutional neural network (CNN) maps the extracted features to the posture classification space through the fully connected layer, and classifies them through the Softmax function to identify the patient's posture status, including lying flat, lying on the side, and sitting up.

[0029] Optionally, the posture change detection and tracking in S22 includes:

[0030] S221, posture change detection: based on the identification of the patient's current posture, using an image frame difference algorithm to analyze pixel differences between consecutive image frames to detect changes in the patient's posture;

[0031] S222, posture change trend tracking: based on the current posture state, calculate the speed and direction of the posture change, and use the Kalman filter to predict the trend of the detected posture change;

[0032] S223, posture tracking result output: according to the results of posture change detection and tracking, the patient's posture state is updated in real time.

[0033] Optionally, the multimodal data correction in S23 includes:

[0034] S231, analysis of the impact of posture changes on multi-modal signals: based on the real-time recognized posture information, calculate the impact of posture changes on different modal data, including:

[0035] Analysis of the impact of posture changes on image data: Using a perspective projection transformation algorithm, calculate the impact of posture changes on the position and angle of respiratory features in the image;

[0036] Analysis of the impact of posture changes on audio data: Use the sound wave propagation algorithm to estimate the impact of posture changes on audio data propagation;

[0037] Analysis of the influence of posture changes on bioimpedance data: quantify the influence of posture changes on bioimpedance data through the relationship between posture angle and impedance value;

[0038] S232, applying a posture compensation model: based on the calculated influence of the posture change on the multimodal signal, applying the posture compensation model to correct the signal, including:

[0039] Image data correction: Use a geometric correction model based on perspective projection to correct the image data;

[0040] Audio data correction: Use adaptive filters to adjust the amplitude of the audio signal and correct the audio data;

[0041] Bioimpedance data correction: adopts posture-based impedance compensation algorithm to correct bioimpedance data;

[0042] S233, output of multimodal data correction results: after real-time multimodal data correction, output corrected image data, audio data and bioimpedance data.

[0043] Optionally, the multimodal feature extraction and fusion in S3 includes:

[0044] S31, image feature extraction: based on the corrected image data, extract image features related to the patient's breathing, including breathing amplitude and breathing frequency, specifically including:

[0045] Respiratory amplitude extraction: Extract the respiratory amplitude by calculating the movement amplitude of the chest and abdomen in consecutive frames;

[0046] Respiratory frequency extraction: Extract respiratory frequency through Fourier transform;

[0047] S32, audio feature extraction: extracting audio features, including breathing sound intensity and audio breathing frequency, by performing spectrum analysis on the corrected audio data and using fast Fourier transform (FFT);

[0048] S33, bioimpedance feature extraction: extract lung volume changes and respiratory impedance based on the corrected bioimpedance data;

[0049] S34, multimodal feature fusion: normalization and weighted fusion are used to fuse and analyze the extracted image features, audio features, and bioimpedance features to obtain comprehensive respiratory feature parameters.

[0050] Optionally, the anomaly detection in S4 includes:

[0051] S41, data analysis: Analyze the integrated respiratory characteristic parameters (including image features, audio features and bioimpedance features) obtained by the respiratory anomaly detection model to obtain the predicted value of respiratory frequency Respiratory rhythm prediction And the predicted value of breathing depth

[0052] S42, abnormal respiratory rate detection: based on the obtained respiratory rate prediction value By comparing with the respiratory rate threshold in the normal range, if the predicted respiratory rate Below the respiratory rate lower threshold f min Or exceeds the upper respiratory rate threshold f max , it is judged as abnormal respiratory rate;

[0053] S43, respiratory rhythm disorder detection: based on the obtained respiratory rhythm prediction value By comparing with the respiratory rhythm threshold T max Compared with the predicted respiratory rhythm Above the respiratory rhythm threshold T max , it is judged as respiratory rhythm disorder;

[0054] S44, insufficient breathing depth detection: based on the obtained breathing depth prediction value By comparing with the breathing depth threshold D min Compared with the predicted breathing depth Below the breathing depth threshold D min , it is judged as insufficient breathing depth.

[0055] Optionally, the breathing abnormality detection model in S41 adopts a graph attention network (GAT) model, and the graph attention network (GAT) model includes:

[0056] S411, input data construction: organize the multi-modal respiratory feature parameters (including image features, audio features and bio-impedance features) into a graph structure, set the respiratory feature parameters of each modality as nodes in the graph, each node corresponds to a different feature vector, the image feature node is the respiratory image feature, the audio feature node is the respiratory audio feature, the bio-impedance feature node is the bio-impedance feature, and the node feature is represented by h i, where i represents different modal features;

[0057] S412, graph structure definition: Establish the connection relationship between nodes, representing the mutual influence between different modal data, and construct the graph structure of the inter-modal dependency relationship by defining edges, where the edge between node i and node j is defined as (i, j), indicating the correlation between the two modal features;

[0058] S413, attention coefficient calculation: Use the graph attention mechanism to assign attention weights to the connections between each node (the size of the weight reflects the importance between the nodes), and calculate the attention coefficient α by learning the contribution of neighboring nodes to the current node. ij ;

[0059] S414, node feature update: according to the attention coefficient α ij Perform weighted updates on node features and aggregate information from neighboring nodes to update the features of each node;

[0060] S415, global feature fusion and classification: After the multi-layer attention mechanism, the features of all nodes are updated to feature representations containing multimodal information. The updated feature vectors are used as input for global feature fusion, followed by respiratory abnormality detection and classification.

[0061] S416, model training: Use the annotated breathing data to train the graph attention network (GAT) model, and use the cross entropy loss function to optimize the model weights.

[0062] Optionally, the warning generation in S5 includes:

[0063] S51, abnormal detection trigger: when abnormal breathing is detected, an early warning signal is automatically generated;

[0064] S52, early warning notification: according to the generated early warning signal, an alarm is immediately issued to medical staff through the monitoring equipment, including an audible alarm, a visual prompt or a remote notification;

[0065] S53, intervention reminder: through early warning notifications, medical staff are reminded to take appropriate intervention measures in a timely manner, including oxygen therapy, ventilator adjustment, body position adjustment, drug intervention, and suction or airway clearing.

[0066] Beneficial effects of the present invention:

[0067] The present invention solves the problem of insufficient accuracy in traditional single data source monitoring through steps such as multimodal data acquisition and preprocessing, patient posture recognition and correction, etc., and ensures that consistent and accurate respiratory monitoring data can be obtained under different postures by integrating image, audio and bioimpedance signals and correcting the monitoring data deviation caused by posture changes in real time. This process improves the comprehensiveness and accuracy of the patient's respiratory status, effectively enhances the system's adaptability to the patient's dynamic state, and reduces monitoring errors, especially in the case of frequent posture changes, to ensure continuous respiratory data monitoring.

[0068] The present invention uses a graph attention network model to perform fusion analysis on multimodal respiratory features, dynamically learns the dependencies between different modal features, and automatically assigns feature weights to achieve accurate fusion analysis of multimodal respiratory features. The GAT model can comprehensively and flexibly capture the interrelationships between different features such as images, audio, and bioimpedance, thereby improving the robustness and accuracy of respiratory abnormality detection, especially in complex or dynamic clinical environments, ensuring that respiratory abnormalities can be identified and warned in a timely and accurate manner.

[0069] The present invention uses a variety of warning generation methods, including sound alarms, visual prompts and remote notifications, to ensure that medical staff can be informed of abnormal breathing conditions in the first place, and quickly take intervention measures according to the warning prompts, such as oxygen therapy, ventilator adjustment, body position adjustment, etc. This warning mechanism improves the emergency response speed of abnormal breathing and reduces the risk of delayed treatment, thereby improving the efficiency of overall clinical monitoring and intervention, ensuring that patients can receive timely and effective treatment, and significantly improving the safety of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 A schematic diagram of a warning method flow chart of an embodiment of the present invention;

[0072] Figure 2 Schematic diagram of automatic patient posture recognition and correction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0073] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0074] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0075] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0076] like Figure 1-Figure 2 As shown, a respiratory monitoring and early warning method for patients with severe respiratory diseases comprises the following steps:

[0077] S1, multimodal data acquisition and preprocessing: using a camera, an audio sensor and a bioimpedance sensor to synchronously acquire the patient's respiratory image data, respiratory audio data and bioimpedance data to form multimodal respiratory signal data, and preprocessing the multimodal respiratory signal data, including denoising and signal enhancement;

[0078] S2, automatic identification and correction of patient posture: Analyze the pre-processed multi-modal respiratory signal data, identify the patient's posture information in real time, including lying flat, lying on the side, and sitting up, and perform posture correction on the multi-modal respiratory signal data based on the identified posture results to compensate for the data deviation caused by posture changes and ensure data consistency under different postures;

[0079] S3, multimodal feature extraction and fusion: Based on the corrected multimodal respiratory signal data, the respiratory feature parameters of each modality are extracted, including image features, audio features, and bioimpedance features, and the respiratory feature parameters of each modality are fused to obtain comprehensive respiratory feature parameters to improve the comprehensiveness and accuracy of the respiratory status;

[0080] S4, abnormality detection: Based on the integrated respiratory feature parameters obtained by fusion, the respiratory abnormality detection model is used to analyze whether the patient's breathing is abnormal, including abnormal respiratory rate, disordered respiratory rhythm, and insufficient respiratory depth;

[0081] S5, warning generation: when abnormal breathing is detected, the monitoring device immediately informs the medical staff by means of sound alarm, visual prompt or remote notification, reminding them to take appropriate intervention measures;

[0082] Through the above steps, combined with automatic recognition and correction of patient posture, comprehensive and accurate monitoring of critically ill respiratory patients is achieved, and data consistency is maintained under different postures, overcoming the problem of insufficient monitoring accuracy caused by traditional single data source and posture changes. By comprehensively analyzing multimodal feature parameters, respiratory abnormalities are detected in time and early warning signals are generated, which helps medical staff to quickly take intervention measures and improve the efficiency of clinical monitoring and patient safety.

[0083] Multimodal data acquisition and preprocessing in S1 include:

[0084] S11, data acquisition: using a camera to collect respiratory image data of the patient's chest and abdomen, using an audio sensor to collect the patient's respiratory audio data, and using a bioimpedance sensor to monitor the patient's lung volume changes and respiratory impedance;

[0085] S12, denoising processing: performing denoising processing on the collected multi-modal respiratory signal data, including:

[0086] Image data denoising: The denoising algorithm based on Gaussian filter is used for denoising, which is expressed as:

[0087]

[0088] Where I'(x,y) is the pixel value of the denoised image, I(x+i,y+j) is the pixel value at the pixel coordinate (x+i,y+j) in the image, σ is the standard deviation of the Gaussian filter, and k is the size of the filter kernel;

[0089] Audio data denoising: A bandpass filter is used to remove environmental noise from the breathing audio signal. The frequency range of the bandpass filter is set to 0.1Hz to 1kHz, expressed as:

[0090]

[0091] Where f is the frequency of the input audio, f0 is the center frequency, and f b is the bandwidth, Q is the quality factor;

[0092] Bioimpedance data denoising: The Kalman filter algorithm is used to denoise the bioimpedance data, which is expressed as:

[0093]

[0094] in, is the predicted state, z k is the observed value, K k is the Kalman gain, H is the observation matrix;

[0095] S13, signal enhancement: Perform signal enhancement on the denoised multimodal respiratory signal data to improve the quality and identifiability of the signal, including:

[0096] Image data enhancement: The histogram equalization algorithm is used to enhance the contrast of the respiratory image by adjusting the grayscale distribution of the image, which is expressed as:

[0097]

[0098] Among them, I eq (x, y) is the pixel value of the enhanced image, L is the grayscale level of the image, MN is the total number of pixels in the image, h(i) is the number of pixels with grayscale level i, and I(x, y) is the pixel value of the original image;

[0099] Audio data enhancement: The spectrum of the breathing audio signal is enhanced, and the high-frequency component is amplified using a pre-emphasis filter, which is expressed as:

[0100] S′(f)=S(f)·αf;

[0101] Where S(f) is the spectrum of the original audio signal, S'(f) is the enhanced audio spectrum, and α is the amplification factor;

[0102] Bioimpedance data enhancement: A sliding window mean filter is used to smooth and enhance the bioimpedance data, which is expressed as:

[0103]

[0104] Where Z(t) is the original impedance value, Z'(t) is the smoothed impedance value, and N is the window size;

[0105] Through the above contents, the quality and availability of data are improved, and the impact of environmental noise, equipment interference and posture changes on monitoring accuracy are reduced.

[0106] Automatic patient posture recognition and correction in S2 includes:

[0107] S21, posture recognition: Analyze the preprocessed multimodal respiratory signal data and use convolutional neural network (CNN) to perform posture recognition on the patient's image data, including lying flat, lying on the side, and sitting up;

[0108] S22, posture change detection and tracking: Based on the patient's posture recognition, the patient's posture changes are continuously monitored, and the image difference detection and time series analysis between consecutive frames are used to track the posture change trend in real time to ensure a rapid response to posture changes;

[0109] S23, multimodal data correction: according to the real-time recognized posture information, by calculating the influence of posture change on image data, audio data and bioimpedance data, the posture compensation model is applied to correct the data deviation to ensure the consistency and accuracy of the respiratory signal under different postures, thereby compensating for the signal distortion caused by the patient's posture change;

[0110] Through the above content, the problem of data deviation caused by posture changes in traditional monitoring is solved, the monitoring system's adaptability to the patient's dynamic state is improved, the monitoring error caused by patient movement or posture changes is reduced, the continuity and reliability of respiratory monitoring is ensured, and it helps to provide medical staff with more accurate monitoring results.

[0111] Gesture recognition in S21 includes:

[0112] S211, posture recognition: Use convolutional neural network (CNN) to perform posture recognition on the patient's image data, and extract features from the input image data through the convolution layer, which is expressed as:

[0113]

[0114] Among them, Z l (x, y) represents the value of the pixel position (x, y) in the output feature map of convolution layer l, W l (i, j) is the weight parameter of the convolution kernel, the size of the convolution kernel is (2k+1)×(2k+1), I l-1 (x+i, y+j) is the pixel value of the previous layer input image or feature map, b l is the bias term of the convolution layer, and k is half the size of the convolution kernel;

[0115] S212, posture classification and output: The convolutional neural network (CNN) maps the extracted features to the posture classification space through the fully connected layer, and classifies them through the Softmax function to identify the patient's posture state, including lying flat, lying on the side, and sitting up, which is expressed as:

[0116]

[0117] Among them, P(c|x) represents the probability that the image data x belongs to category c (lying flat, lying on the side, sitting up), z c is the unnormalized score of the corresponding category c in the output layer, and C is the total number of classification categories (lying flat, lying on the side, sitting up);

[0118] Through the above content, it is possible to quickly adapt to changes in patient posture and ensure that the respiratory monitoring data remains consistent and accurate under various body postures. At the same time, posture recognition reduces monitoring errors caused by changes in patient posture, improves the robustness and intelligence of the system, and provides a reliable basis for subsequent respiratory signal correction, ultimately improving the accuracy and effectiveness of overall monitoring.

[0119] Posture change detection and tracking in S22 includes:

[0120] S221, posture change detection: Based on the identification of the patient's current posture, the image frame difference algorithm is used to analyze the pixel differences between consecutive image frames to detect changes in the patient's posture, which is expressed as:

[0121] D t (x,y)=|I t (x,y)-I t-1 (x,y)|;

[0122] Among them, D t (x, y) represents the difference value at the pixel position (x, y) at time t, I t (x,y) and I t-1 (x, y) are the pixel values ​​of the image frame at time t and t-1 respectively;

[0123] S222, posture change trend tracking: Based on the current posture state, the speed and direction of the posture change are calculated, and the Kalman filter is used to predict the trend of the detected posture change, which is expressed as:

[0124]

[0125] in, is the estimated value of the posture at time k, z k is the actual observed posture data, K k is the Kalman gain, H is the observation matrix, P k|k-1 is the estimated covariance, R is the measurement noise covariance;

[0126] S223, posture tracking result output: according to the results of posture change detection and tracking, the posture state of the patient is updated in real time;

[0127] Through the above content, the patient's posture change trend can be captured in real time, and the posture transition can be accurately predicted. It can not only quickly identify the changes in the patient's posture, but also smoothly track the continuous changes in posture through the Kalman filter, avoiding monitoring errors caused by instantaneous posture changes, ensuring the robustness of the system in dynamic conditions, and providing reliable support for respiratory signal correction and consistency of monitoring data, thereby improving the accuracy and real-time performance of monitoring.

[0128] The multimodal data correction in S23 includes:

[0129] S231, analysis of the impact of posture changes on multi-modal signals: based on the real-time recognized posture information, calculate the impact of posture changes on different modal data, including:

[0130] Analysis of the influence of posture changes on image data: Through the perspective projection transformation algorithm, the influence of posture changes on the position and angle of the breathing features in the image is calculated, which is expressed as:

[0131]

[0132] Where x', y' are the coordinates of the patient's chest and abdomen in the image after the posture change, x, y are the coordinates in the original posture, f is the focal length of the camera, x0, y0 are the offsets of the image center, z is the distance between the camera and the patient, and d z is the depth offset caused by posture change;

[0133] Analysis of the influence of posture changes on audio data: The influence of posture changes on the propagation of audio data is estimated through the sound wave propagation algorithm, which is expressed as:

[0134]

[0135] Where d0 is the sound wave propagation distance in the reference posture, and d(t) is the propagation distance after the posture change, which depends on the change of the three-dimensional space distance between the patient and the sensor;

[0136] Analysis of the influence of posture changes on bioimpedance data: The influence of posture changes on bioimpedance data is quantified through the relationship between posture angle and impedance value, expressed as:

[0137]

[0138] Where Z(t) is the impedance during posture change, ρ is the resistivity of body tissue (depending on the characteristics of body tissue), L(t) is the path length between electrodes, which may change due to posture changes (such as bending, stretching, etc.), and A(t) is the area of ​​the body cross-section, which may also change due to posture changes;

[0139] S232, applying a posture compensation model: based on the calculated influence of the posture change on the multimodal signal, applying the posture compensation model to correct the signal, including:

[0140] Image data correction: The image data is corrected using a geometric correction model based on perspective projection, which is expressed as:

[0141]

[0142] Among them, x', y' are the pixel coordinates of the corrected image;

[0143] Audio data correction: An adaptive filter is used to adjust the amplitude of the audio signal and correct the audio data, which is expressed as:

[0144]

[0145] Wherein, S'(t) is the corrected audio data, and S(t) is the original audio data;

[0146] Bioimpedance data correction: The bioimpedance data is corrected using a posture-based impedance compensation algorithm, which is expressed as:

[0147] Z′(t)=Z(t)·(1+α·Δθ);

[0148] Where Z'(t) is the corrected bioimpedance value after posture change, α is the impedance sensitivity coefficient related to the patient's body structure, and Δθ is the angle of posture change (e.g., the angle change from lying flat to lying on the side);

[0149] S233, output of multimodal data correction result: after real-time multimodal data correction, output corrected image data, audio data and bioimpedance data to ensure consistency of multimodal data under different postures, avoid data deviation caused by posture changes, and provide accurate respiratory monitoring information;

[0150] Through the above content, the consistency of various types of data under different postures is ensured, especially the changes in viewing angle, sound wave propagation path and bioimpedance caused by posture changes can be compensated, thereby greatly improving the accuracy and reliability of the data. Multimodal data correction provides more stable input data for subsequent respiratory monitoring, ensuring the accuracy of monitoring in dynamic scenes, making the monitoring results more in line with the actual situation, and significantly improving the adaptability and robustness of the system.

[0151] Multimodal feature extraction and fusion in S3 include:

[0152] S31, image feature extraction: based on the corrected image data, extract image features related to the patient's breathing, including breathing amplitude and breathing frequency, specifically including:

[0153] Respiratory amplitude extraction: The respiratory amplitude is extracted by calculating the movement amplitude of the chest and abdomen in consecutive frames, which is expressed as:

[0154] M(t)=∑ x,y |I t (x,y)-I t-1 (x,y)|;

[0155] Where M(t) is the chest and abdomen movement amplitude at time t, I t (x,y) and I t-1 (x, y) are the pixel values ​​of the image frame at time t and t-1 respectively;

[0156] Respiratory frequency extraction: Respiratory frequency is extracted by Fourier transform, expressed as:

[0157]

[0158] Among them, f breath represents the respiratory frequency, FFT(M(t)) represents the Fourier transform of the chest and abdomen motion amplitude signal, and T is the time window of signal acquisition;

[0159] S32, audio feature extraction: by performing spectrum analysis on the corrected audio data, the audio features are extracted using fast Fourier transform (FFT), including the intensity of breathing sound and the audio breathing frequency, which are expressed as:

[0160]

[0161] Where, S(f) is the respiratory spectrum value at frequency f, and s(t) is the respiratory audio signal;

[0162]

[0163] Where I(t) is the sound intensity value at time t, and s(n) is the amplitude value of the sampled signal;

[0164] S33, bioimpedance feature extraction: Based on the corrected bioimpedance data, extract lung volume changes and respiratory impedance, expressed as:

[0165]

[0166] Where Z(t) is the bioimpedance value, ρ is the resistivity, L(t) is the path length between electrodes, and A(t) is the cross-sectional area of ​​the chest;

[0167]

[0168] Where V(t) represents the change in the patient's lung volume, Z(t) is the integral of the impedance value, and T represents the time length of a complete respiratory cycle;

[0169] S34, multimodal feature fusion: normalization and weighted fusion are used to fuse and analyze the extracted image features, audio features, and bioimpedance features to obtain comprehensive respiratory feature parameters, which are expressed as:

[0170]

[0171] Among them, x' is the normalized eigenvalue, x is the original eigenvalue, μ and σ are the mean and standard deviation of the feature respectively;

[0172] F combined =w1·f image +w2·f audio +w3·f imedance ;

[0173] Among them, F combined is the integrated respiratory characteristic parameter after fusion, f image 、f audio 、f impedance are the feature values ​​of image, audio and bioimpedance respectively, w1, w2 and w3 are the weights of each modality feature;

[0174] The weights w1, w2, and w3 are set based on the signal quality, stability, and relevance of each modality data to the respiratory state. The modality with higher signal quality and lower noise is usually assigned a higher weight, satisfying w1+w2+w3=1, w1 is 0.2, w2 is 0.3, and w3 is 0.5;

[0175] Through the above content, the patient's respiratory status can be reflected more comprehensively and accurately. It not only overcomes the possible limitations of single modality data, but also ensures that high-quality and highly relevant modalities occupy a larger proportion in the final result through weight allocation. Multimodal fusion improves the robustness and accuracy of respiratory monitoring, especially in complex or dynamic clinical environments, providing a more reliable basis for respiratory abnormality detection.

[0176] Anomaly detection in S4 includes:

[0177] S41, data analysis: Analyze the integrated respiratory characteristic parameters (including image features, audio features and bioimpedance features) obtained by the respiratory anomaly detection model to obtain the predicted value of respiratory frequency Respiratory rhythm prediction And the predicted value of breathing depth

[0178] S42, abnormal respiratory rate detection: based on the obtained respiratory rate prediction value By comparing with the respiratory rate threshold in the normal range, if the predicted respiratory rate Below the respiratory rate lower threshold f min Or exceeds the upper respiratory rate threshold f max , it is judged as abnormal respiratory rate;

[0179] Respiratory rate threshold f min 、f max By collecting a large number of historical respiratory rate data of healthy individuals and patients with respiratory abnormalities, the frequency distribution within the normal range is calculated and the upper and lower thresholds are set, including:

[0180] Collect historical respiratory rate data of healthy people and abnormal patients to form a data set {f1,f2,...,f n}, where f i is the respiratory rate of the ith person;

[0181] Fit the historical data to a normal distribution, assuming that the distribution of respiratory frequency conforms to the normal distribution N(μ f ,σ f ), where μ f is the mean, σ f is the standard deviation;

[0182] According to the confidence interval of the data, set the upper and lower thresholds f min and f max , the threshold can usually be set by a confidence interval that statistically covers 95% of the population, expressed as:

[0183] f min =μ f -1.96·σ f ;

[0184] f max =μ f +1.96·σ f ;

[0185] Among them, 1.96 is the Z value corresponding to the 95% confidence interval;

[0186] S43, respiratory rhythm disorder detection: based on the obtained respiratory rhythm prediction value By comparing with the respiratory rhythm threshold T max Compared with the predicted respiratory rhythm Above the respiratory rhythm threshold T max , it is judged as respiratory rhythm disorder;

[0187] Respiratory rhythm threshold T max It is set based on the rate of change of the respiratory cycle in the time series data, mainly by calculating the fluctuation range of each cycle, including:

[0188] Collect respiratory cycle time series data {T1,T2,...,T n}, where T i represents the cycle of the i-th breathing;

[0189] Calculate the fluctuation ΔT of each respiratory cycle relative to the previous cycle i =T i -T i-1 ;

[0190] The changes in respiratory cycle fluctuations are statistically analyzed, assuming that the fluctuations follow a normal distribution N(μ T ,σ T ), and the mean μ is obtained by fitting historical data T and standard deviation σ T ;

[0191] By setting a confidence interval covering 95% of the population, the maximum allowable period fluctuation range is obtained, which is expressed as:

[0192] T max =μ T +1.96·σ T ;

[0193] S44, insufficient breathing depth detection: based on the obtained breathing depth prediction value By comparing with the breathing depth threshold D min Compared with the predicted breathing depth Below the breathing depth threshold D min , it is judged as insufficient breathing depth;

[0194] Respiratory depth threshold D min Through historical data, the breathing depth is set based on the bioimpedance signal. The breathing depth is mainly calculated by the integral of bioimpedance, including:

[0195] Collect historical data of breathing depth of healthy individuals and abnormal patients, and record the changes in breathing depth {D1,D2,...,D n}, where D i is the depth of the ith breath;

[0196] The breathing depth data is statistically analyzed, assuming that the depth variation follows a normal distribution N(μ D ,σ D ), where μ D is the mean, σ D is the standard deviation;

[0197] Set the lower depth limit D min The standard deviation is 1.96 times the mean, which ensures that 95% of the normal population is within this range, expressed as:

[0198] D min =μ D -1.96·σ D ;

[0199] Through the above content, the accuracy of abnormality detection is improved, and the misjudgment that may be caused by single signal monitoring is avoided. At the same time, setting a reasonable threshold range can quickly identify and warn of abnormal respiratory rate, rhythm disorders and insufficient breathing depth, ensuring efficient and accurate assessment of the patient's respiratory condition, and improving the timeliness and reliability of clinical intervention.

[0200] The breathing anomaly detection model in S41 adopts the graph attention network (GAT) model, which includes:

[0201] S411, input data construction: organize the multi-modal respiratory feature parameters (including image features, audio features and bio-impedance features) into a graph structure, set the respiratory feature parameters of each modality as nodes in the graph, each node corresponds to a different feature vector, the image feature node is the respiratory image feature, the audio feature node is the respiratory audio feature, the bio-impedance feature node is the bio-impedance feature, and the node feature is represented by h i , where i represents different modal features;

[0202] S412, graph structure definition: Establish the connection relationship between nodes, representing the mutual influence between different modal data, and construct the graph structure of the inter-modal dependency relationship by defining edges, where the edge between node i and node j is defined as (i, j), indicating the correlation between the two modal features;

[0203] S413, attention coefficient calculation: Use the graph attention mechanism to assign attention weights to the connections between each node (the size of the weight reflects the importance between the nodes), and calculate the attention coefficient α by learning the contribution of neighboring nodes to the current node. ij , expressed as:

[0204]

[0205] Among them, α ij is the attention weight between node i and node j, a is the attention weight vector, W is the learnable weight matrix, h i and h j is the feature representation of node i and node j, is the set of neighbor nodes of node i;

[0206] S414, node feature update: according to the attention coefficient α ij The features of the nodes are weighted updated, and the information of neighboring nodes is aggregated to update the features of each node, which can be expressed as:

[0207]

[0208] Among them, h i ' is the updated feature representation of node i, σ is the ReLU activation function, α ij is the attention coefficient between nodes i and j, Wh j is the representation of the features of neighbor node j after being transformed by a learnable weight matrix;

[0209] S415, global feature fusion and classification: After the multi-layer attention mechanism, the features of all nodes are updated to feature representations containing multimodal information. The updated feature vector is used as input for global feature fusion, followed by respiratory abnormality detection and classification, which is expressed as:

[0210] h out =∑ i∈Nodes h i ′;

[0211] Among them, h out is the feature representation after global fusion, which contains the information of all modal nodes, h i ' is the updated node feature, Nodes is all the nodes in the graph, that is, a set of nodes containing different modal features (such as image, audio, bioimpedance, etc.);

[0212] S416, model training: Use the labeled breathing data to train the graph attention network (GAT) model, and use the cross entropy loss function to optimize the model weights, expressed as:

[0213]

[0214] Among them, L is the cross-picking loss, y i is the true label (0 means normal, 1 means abnormal), is the probability value predicted by the model, and N is the total number of samples;

[0215] Through the above content, it is possible to effectively process multimodal data, such as images, audio, and bioimpedance features, use the attention mechanism to dynamically learn the interdependence between these modalities, and automatically assign feature weights. Secondly, the graph structure of GAT can flexibly model the complex relationship between modal features, thereby providing a more comprehensive respiratory feature analysis. By aggregating key information from different modalities, GAT improves the accuracy of respiratory abnormality detection and can achieve more reliable monitoring and early warning in complex clinical environments.

[0216] Alert generation in S5 includes:

[0217] S51, abnormal detection trigger: when abnormal breathing is detected, an early warning signal is automatically generated;

[0218] S52, early warning notification: Based on the generated early warning signal, the monitoring device immediately sends an alarm to the medical staff, including an audible alarm, a visual prompt or a remote notification, specifically including:

[0219] Sound alarm: The monitoring device sends out a sound alarm to remind medical staff to pay attention to the patient's abnormal breathing;

[0220] Visual prompts: Display abnormal information on the device display screen, providing visual prompts in the form of colors or flashing icons;

[0221] Remote notification: Send remote notifications to medical staff via the Internet or mobile devices to ensure that medical staff can be informed of abnormal breathing conditions in a timely manner even when they are not on site;

[0222] S53, intervention reminder: through early warning notifications, medical staff are reminded to take appropriate intervention measures in a timely manner, including oxygen therapy, ventilator adjustment, body position adjustment, drug intervention, and suction or airway clearing to ensure that the patient's breathing abnormalities are effectively treated;

[0223] Specific intervention measures include:

[0224] Oxygen therapy: If the patient has insufficient breathing depth or abnormal breathing rate, medical staff can immediately adjust the patient's oxygen supply concentration to ensure that the patient can get enough oxygen support and improve breathing conditions;

[0225] Ventilator adjustment: For patients using mechanical ventilation, medical staff can adjust the ventilator parameters, such as respiratory rate, tidal volume or respiratory pressure, according to early warning signals to meet the patient's actual respiratory needs and prevent respiratory failure;

[0226] Body position adjustment: If the patient is detected to have abnormal breathing due to body position, such as respiratory rhythm disorder when lying flat, medical staff can adjust the patient's body position (such as lying on the side or raising the upper body) in time to help improve breathing;

[0227] Drug intervention: For specific types of respiratory abnormalities (such as acute respiratory distress), medical staff may use drugs such as bronchodilators, sedatives, or antibiotics to help relieve the patient's symptoms and stabilize the patient's breathing status;

[0228] Suctioning or clearing the airway: When the patient has respiratory obstruction or accumulation of airway secretions, medical staff can suction or clear the airway to ensure that the airway is unobstructed and improve the patient's ventilation.

[0229] Through the above content, medical staff can be informed instantly in a variety of ways to ensure that abnormal situations receive timely attention. At the same time, combined with different types of intervention measures, medical staff can quickly take effective response actions, which not only improves the accuracy and response speed of respiratory monitoring, but also effectively reduces the risk of delayed treatment, helping patients receive timely treatment and care, and significantly improving the efficiency and effectiveness of clinical intervention.

[0230] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0231] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A respiratory monitoring and early warning method for patients with severe respiratory diseases, characterized in that: The following steps are involved: S1, multimodal data acquisition and preprocessing: using a camera, an audio sensor and a bioimpedance sensor to synchronously acquire the patient's respiratory image data, respiratory audio data and bioimpedance data to form multimodal respiratory signal data, and preprocessing the multimodal respiratory signal data, including denoising and signal enhancement; S2, automatic identification and correction of patient posture: Analyze the pre-processed multi-modal respiratory signal data, identify the patient's posture information in real time, including lying flat, lying on the side, and sitting up, and perform posture correction on the multi-modal respiratory signal data based on the identified posture results to compensate for the data deviation caused by posture changes; S3, multimodal feature extraction and fusion: based on the corrected multimodal respiratory signal data, extract the respiratory feature parameters of each modality, including image features, audio features and bioimpedance features, and fuse the respiratory feature parameters of each modality to obtain comprehensive respiratory feature parameters; S4, abnormality detection: Based on the integrated respiratory feature parameters obtained by fusion, the respiratory abnormality detection model is used to analyze whether the patient's breathing is abnormal, including abnormal respiratory rate, disordered respiratory rhythm, and insufficient respiratory depth; S5, warning generation: When abnormal breathing is detected, the monitoring equipment will immediately inform the medical staff by means of sound alarms, visual prompts or remote notifications, reminding them to take appropriate intervention measures.

2. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 1, characterized in that: The multimodal data acquisition and preprocessing in S1 includes: S11, data acquisition: using a camera to collect respiratory image data of the patient's chest and abdomen, using an audio sensor to collect the patient's respiratory audio data, and using a bioimpedance sensor to monitor the patient's lung volume changes and respiratory impedance; S12, denoising processing: performing denoising processing on the collected multi-modal respiratory signal data, including: Image data denoising: denoising is performed using a denoising algorithm based on a Gaussian filter; Audio data denoising: A bandpass filter is used to remove environmental noise from the breathing audio signal, and the frequency range of the bandpass filter is set to 0.1Hz to 1kHz; Bioimpedance data denoising: The Kalman filter algorithm is used to denoise the bioimpedance data; S13, signal enhancement: performing signal enhancement on the denoised multimodal respiratory signal data, including: Image data enhancement: The histogram equalization algorithm is used to enhance the contrast of the respiratory image by adjusting the grayscale distribution of the image; Audio data enhancement: Spectral enhancement of breathing audio signals is performed, and high-frequency components are amplified using a pre-emphasis filter; Bioimpedance data enhancement: A sliding window mean filter is used to smooth and enhance the bioimpedance data.

3. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 1, characterized in that: The automatic recognition and correction of the patient's posture in S2 includes: S21, posture recognition: Analyze the pre-processed multimodal respiratory signal data and use convolutional neural network to perform posture recognition on the patient's image data, including lying flat, lying on the side, and sitting up; S22, posture change detection and tracking: Based on the patient's posture recognition, the patient's posture changes are continuously monitored, and the trend of posture changes is tracked in real time by using image difference detection and time series analysis between consecutive frames; S23, multimodal data correction: based on the real-time recognized posture information, the influence of posture changes on image data, audio data and bioimpedance data is calculated, and the posture compensation model is applied to correct the data deviation.

4. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 3, characterized in that: The gesture recognition in S21 includes: S211, posture recognition: using a convolutional neural network to perform posture recognition on the patient's image data, and extracting features from the input image data through a convolutional layer; S212, posture classification and output: The convolutional neural network maps the extracted features to the posture classification space through the fully connected layer, and classifies them through the Softmax function to identify the patient's posture status, including lying flat, lying on the side, and sitting up.

5. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 4, characterized in that: The posture change detection and tracking in S22 includes: S221, posture change detection: based on the identification of the patient's current posture, using an image frame difference algorithm to analyze pixel differences between consecutive image frames to detect changes in the patient's posture; S222, posture change trend tracking: based on the current posture state, calculate the speed and direction of the posture change, and use the Kalman filter to predict the trend of the detected posture change; S223, posture tracking result output: according to the results of posture change detection and tracking, the patient's posture state is updated in real time.

6. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 5, characterized in that: The multimodal data correction in S23 includes: S231, analysis of the impact of posture changes on multi-modal signals: based on the real-time recognized posture information, calculate the impact of posture changes on different modal data, including: Analysis of the impact of posture changes on image data: Using a perspective projection transformation algorithm, calculate the impact of posture changes on the position and angle of respiratory features in the image; Analysis of the impact of posture changes on audio data: Use the sound wave propagation algorithm to estimate the impact of posture changes on audio data propagation; Analysis of the impact of posture changes on bioimpedance data: quantify the impact of posture changes on bioimpedance data through the relationship between posture angle and impedance value; S232, applying a posture compensation model: based on the calculated influence of the posture change on the multimodal signal, applying the posture compensation model to correct the signal, including: Image data correction: Use a geometric correction model based on perspective projection to correct the image data; Audio data correction: Use adaptive filters to adjust the amplitude of the audio signal and correct the audio data; Bioimpedance data correction: adopts posture-based impedance compensation algorithm to correct bioimpedance data; S233, output of multimodal data correction results: after real-time multimodal data correction, output corrected image data, audio data and bioimpedance data.

7. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 1, characterized in that: The multimodal feature extraction and fusion in S3 includes: S31, image feature extraction: based on the corrected image data, extract image features related to the patient's breathing, including breathing amplitude and breathing frequency, specifically including: Respiratory amplitude extraction: Extract the respiratory amplitude by calculating the movement amplitude of the chest and abdomen in consecutive frames; Respiratory frequency extraction: Extract respiratory frequency through Fourier transform; S32, audio feature extraction: by performing spectrum analysis on the corrected audio data, using fast Fourier transform to extract audio features, including breathing sound intensity and audio breathing frequency; S33, bioimpedance feature extraction: extract lung volume changes and respiratory impedance based on the corrected bioimpedance data; S34, multimodal feature fusion: normalization and weighted fusion are used to fuse and analyze the extracted image features, audio features, and bioimpedance features to obtain comprehensive respiratory feature parameters.

8. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 1, characterized in that: The anomaly detection in S4 includes: S41, data analysis: Analyze the integrated respiratory characteristic parameters obtained by fusion through the respiratory abnormality detection model to obtain the predicted value of respiratory frequency Predicted respiratory rhythm And the predicted value of breathing depth S42, abnormal respiratory rate detection: based on the obtained respiratory rate prediction value By comparing with the respiratory rate threshold in the normal range, if the predicted respiratory rate Below the respiratory rate lower threshold f min Or exceeds the upper respiratory rate threshold f max , it is judged as abnormal respiratory rate; S43, respiratory rhythm disorder detection: based on the obtained respiratory rhythm prediction value By comparing the respiratory rhythm threshold T max Compared with the predicted respiratory rhythm Above the respiratory rhythm threshold T max , it is judged as respiratory rhythm disorder; S44, insufficient breathing depth detection: based on the obtained breathing depth prediction value By comparing with the breathing depth threshold D min Compared with the predicted breathing depth Below the breathing depth threshold D min , it is judged as insufficient breathing depth.

9. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 8, characterized in that: The breathing abnormality detection model in S41 adopts a graph attention network model, and the graph attention network model includes: S411, input data construction: organize the multi-modal respiratory feature parameters into a graph structure, set the respiratory feature parameters of each modality as nodes in the graph, each node corresponds to a different feature vector, the image feature node is the respiratory image feature, the audio feature node is the respiratory audio feature, the bio-impedance feature node is the bio-impedance feature, and the node feature is represented by h i , where i represents different modal features; S412, graph structure definition: Establish the connection relationship between nodes, representing the mutual influence between different modal data, and construct the graph structure of the inter-modal dependency relationship by defining edges, where the edge between node i and node j is defined as (i, j), indicating the correlation between the two modal features; S413, attention coefficient calculation: Use the graph attention mechanism to assign attention weights to the connections between each node, and calculate the attention coefficient α by learning the contribution of neighboring nodes to the current node. ij ; S414, node feature update: according to the attention coefficient α ij Perform weighted updates on node features and aggregate information from neighboring nodes to update the features of each node; S415, global feature fusion and classification: After the multi-layer attention mechanism, the features of all nodes are updated to feature representations containing multimodal information. The updated feature vectors are used as input for global feature fusion, followed by respiratory abnormality detection and classification. S416, model training: Use annotated breathing data to train the graph attention network model and use the cross entropy loss function to optimize the model weights.

10. A respiratory monitoring and early warning method for patients with severe respiratory diseases according to claim 9, characterized in that: The warning generation in S5 includes: S51, abnormal detection trigger: when abnormal breathing is detected, an early warning signal is automatically generated; S52, early warning notification: according to the generated early warning signal, an alarm is immediately issued to medical staff through the monitoring equipment, including an audible alarm, a visual prompt or a remote notification; S53, intervention reminder: through early warning notifications, medical staff are reminded to take appropriate intervention measures in a timely manner, including oxygen therapy, ventilator adjustment, body position adjustment, drug intervention, and suction or airway clearing.

Citation Information

Patent Citations

  • Respiratory early warning scoring systems and methods

    CN109996488A

  • Device for measuring sleep apnea and method therefor

    CN110035691A

  • Breathing monitoring and early warning method for breathing critical patient

    CN117593308A

  • Intelligent sleep breathing state monitoring system and method

    CN118648868A

  • Multi-modal system and method for tracking respiratory health

    US20230346253A1

Cited By

  • Respiration data reminding system for patient with critical respiratory disease

    CN119655714A

  • A respiratory data alerting system for a critically ill patient

    CN119655714B

  • Biofeedback-based respiratory therapy system and method

    CN119656472A

  • Disease condition analysis and early warning method for pneumology department nursing based on big data

    CN119724629A

  • Multi-modal data fusion method and system based on identification analysis

    CN119851081A