Respiratory parameter estimation method and system fusing physical model and attention mechanism

Abstract

The application discloses a kind of fusion physical model and respiratory parameter estimation method and system of attention mechanism, it is related to respiratory parameter estimation technical field.User information, a plurality of left lung respiratory discrimination points and a plurality of right lung respiratory discrimination points are acquired.Through attention mechanism, first estimation parameter characteristic vector and second estimation parameter characteristic vector are detected.First estimation parameter vector and second estimation parameter vector are superimposed and then input into second discrimination neural network, to obtain respiratory parameter value.The attention mechanism and second discrimination neural network are included in the respiratory parameter estimation model.The network of physical model is fused in the training of respiratory parameter estimation model.According to the features of local and central chest and abdominal fluctuation extracted by convolutional neural network, a three-dimensional lung feature map is constructed.The use of attention mechanism and gating mechanism achieves the technical effect of more accurately using the state of chest and abdominal fluctuation to judge the respiratory parameter.

Description

Technical Field

[0001] This invention relates to the field of respiratory parameter estimation technology, and more specifically, to a respiratory parameter estimation method and system that integrates physical models and attention mechanisms. Background Technology

[0002] Estimation of respiratory parameters is typically based on readily available personal information such as age, height, and weight, using specific formulas or rules. It is not a direct measurement but rather a predicted value derived from population statistical patterns, which is very useful in both clinical and routine health assessments. Respiratory parameters are indicators used to quantitatively assess an individual's respiratory movement characteristics and lung function, encompassing multiple dimensions from basic respiratory rate and inhaled air volume to more complex breathing patterns and gas exchange efficiency.

[0003] Currently, there are methods for estimating respiratory parameters using deep learning networks. However, existing deep learning-based respiratory parameter estimation methods still have many shortcomings.

[0004] First, most models rely on a single temporal signal, such as an electrocardiogram (ECG) or piezoelectric signal, as input, neglecting the fact that respiratory patterns are influenced by the coupling of multiple physiological factors. This leads to a significant decrease in the model's generalization ability when there are large individual differences or during physical activity. Furthermore, network designs that estimate respiratory parameter changes based on chest and abdominal variations often employ standard convolutional or recurrent neural networks. While these can capture temporal dependencies, they struggle to effectively extract multi-scale, non-stationary local features from respiratory signals, especially when tidal volume and respiratory rate change drastically, resulting in significant fluctuations in estimation accuracy. Therefore, a network capable of accurately estimating respiratory parameter changes based on chest and abdominal variations is needed. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for estimating respiratory parameters that integrates physical models and attention mechanisms, in order to solve the above-mentioned problems existing in the prior art.

[0006] In a first aspect, embodiments of the present invention provide a method for estimating respiratory parameters that integrates a physical model and an attention mechanism, including:

[0007] Obtain user information; the user information includes the user's age, height, weight, and medical conditions; the medical conditions refer to respiratory symptoms that the user is suffering from at the current point in time;

[0008] Multiple left lung breathing discrimination points and multiple right lung breathing discrimination points are obtained; the left lung breathing discrimination point represents the location on the body surface of the left lung where a sensor is connected; the right lung breathing discrimination point represents the location on the body surface of the right lung where a sensor is connected.

[0009] Through an attention mechanism, based on multiple left lung breathing discrimination points and multiple right lung breathing discrimination points, a first estimation parameter feature vector and a second estimation parameter feature vector are obtained. The first estimation parameter feature vector represents the correlation between the fluctuation state of multiple points around the lung surface during breathing and the breathing parameters. The second estimation parameter feature vector represents the correlation between the fluctuation state of the lung surface during breathing and user information and the breathing parameters.

[0010] The first and second estimated parameter vectors are superimposed and then input into the second discriminative neural network to obtain respiratory parameter values; the respiratory parameter values ​​represent parameters about breathing predicted from user information and respiratory status;

[0011] The attention mechanism and the second discriminant neural network are included in the breathing parameter estimation model; the breathing parameter estimation model is a network that integrates a physical model during training.

[0012] Optionally, the step of obtaining the first estimated parameter feature vector based on the multiple left lung breathing discrimination points and multiple right lung breathing discrimination points through an attention mechanism includes:

[0013] Based on the multiple left lung breathing discrimination points and multiple right lung breathing discrimination points, multiple left lung breathing maps, multiple right lung breathing maps, left lung center point, right lung center point, left lung center breathing map, and right lung center breathing map are obtained;

[0014] The left lung central respiratory map, the right lung central respiratory map, multiple left lung respiratory maps, and multiple right lung respiratory maps are respectively input into the first feature extraction network to detect the features of the body surface due to the fluctuation state of lung respiration, and obtain the right lung central feature vector, the left lung central feature vector, multiple left lung feature vectors, and multiple right lung feature vectors.

[0015] The first estimation parameter feature vector is obtained by using the first discriminative neural network and attention mechanism based on the right lung center feature vector, the left lung center feature vector, multiple left lung feature vectors, and multiple right lung feature vectors;

[0016] The second feature extraction network is used to obtain the second estimation parameter feature vector based on the feature vectors of the left lung center, the right lung center, and multiple user information.

[0017] Optionally, the step of obtaining the first estimated parameter feature vector based on the right lung center feature vector, the left lung center feature vector, multiple left lung feature vectors, and multiple right lung feature vectors through the first discriminative neural network and attention mechanism includes:

[0018] Calculate the distances between the multiple left lung center points and the left lung breathing discrimination points to obtain multiple first distances;

[0019] Using an attention mechanism, based on multiple left lung breathing discrimination points and the left lung center point, and based on the multiple first distances, the left lung center feature vector, and multiple left lung feature vectors, a three-dimensional feature map of the left lung is obtained.

[0020] The right lung center feature vector and multiple right lung feature vectors correspond to obtain a three-dimensional feature map of the right lung;

[0021] The three-dimensional feature map of the left lung is input into a first three-dimensional convolutional network to obtain the feature vector of the left lung; the three-dimensional feature map of the right lung is input into a second three-dimensional convolutional network to obtain the feature vector of the right lung.

[0022] The feature vectors of the left lung and the three-dimensional feature vectors of the right lung are input into the first discriminative neural network to obtain the first estimated parameter feature vector.

[0023] Optionally, the step of obtaining a three-dimensional feature map of the left lung through an attention mechanism, based on multiple left lung breathing discrimination points and the left lung center point, and on the multiple first distances, the left lung center feature vector, and multiple left lung feature vectors, includes:

[0024] Based on multiple left lung feature vectors, multiple left lung distance feature vectors are obtained;

[0025] Multiple left lung distance feature vectors are input into the attention mechanism to obtain multiple left lung attention feature vectors; the left lung attention feature vectors represent the weighted left lung distance feature vectors after determining the association state with the breathing state;

[0026] The left lung attention feature vector is filled into the position of the left lung breathing discrimination point, and the left lung center feature vector is filled into the position of the left lung center point to obtain the three-dimensional feature map of the left lung.

[0027] Optionally, the process of obtaining multiple left lung distance feature vectors based on multiple left lung feature vectors includes:

[0028] Subtract the left lung center feature vector from the left lung feature vector to obtain the left lung difference vector;

[0029] Divide the left lung difference vector by the corresponding first distance to obtain the left lung distance feature vector; the left lung distance feature vector represents the correlation between the surface undulation features of the left lung breathing discrimination point and the distance to the center point of the left lung.

[0030] Multiple left lung feature vectors correspond to multiple left lung distance feature vectors.

[0031] Optionally, the step of obtaining the second estimation parameter feature vector through the second feature extraction network, based on the left lung center feature vector, the right lung center feature vector, and multiple user information, includes:

[0032] User information is input into the second feature extraction network to obtain the user feature vector;

[0033] The lung feature vector is obtained by superimposing the feature vectors of the left and right lung centers.

[0034] Based on the user feature vector and lung feature vector, a second estimation parameter feature vector is obtained through a gating mechanism;

[0035] The method for obtaining the feature vector of the second estimation parameter is shown in the following formula:

[0036]

[0037] in, This represents the eigenvector of the second estimated parameter. Represents the lung feature vector. This represents the user feature vector.

[0038] Optionally, the first discriminant neural network, the second discriminant neural network, the first three-dimensional convolutional network, the second three-dimensional convolutional network, the first feature extraction network, the second feature extraction network, and the attention mechanism constitute a breathing parameter estimation model.

[0039] Optionally, the training method for the respiratory parameter estimation model includes:

[0040] Acquire training data and labeled data; the training data consists of user information at historical time points, left lung central respiratory map, right lung central respiratory map, multiple left lung respiratory maps, and multiple right lung respiratory maps; the labeled data represents the measured respiratory parameters.

[0041] The training data is input into the respiratory parameter estimation model to obtain the first respiratory parameter value;

[0042] Based on the first respiratory parameter value, the loss value of the first respiratory parameter is obtained through a physical model;

[0043] The loss is calculated using the first respiratory parameter value and the labeled data to obtain the second respiratory parameter loss value;

[0044] The first respiratory parameter loss value and the second respiratory parameter loss value are added together, and the respiratory parameter estimation model is trained backward.

[0045] Optionally, the step of obtaining multiple left lung respiratory maps, multiple right lung respiratory maps, left lung center point, right lung center point, left lung center respiratory map, and right lung center respiratory map based on the multiple left lung respiratory discrimination points and multiple right lung respiratory discrimination points includes:

[0046] Multiple left lung breathing maps were obtained at multiple left lung breathing discrimination points; the left lung breathing map represents the degree of skin undulation during breathing.

[0047] Multiple right lung breathing maps were obtained at multiple right lung breathing discrimination points; the right lung breathing map represents the degree of skin undulation during breathing.

[0048] Based on multiple left lung breathing images, the location of the greatest skin undulation is detected using a Bayesian optimization algorithm to obtain the center point of the left lung; the center point of the right lung is obtained corresponding to multiple right lung breathing images.

[0049] At the center point of the left lung, obtain the respiratory image of the left lung; at the center point of the right lung, obtain the respiratory image of the right lung.

[0050] Secondly, embodiments of the present invention provide a respiratory parameter estimation system that integrates a physical model and an attention mechanism, comprising:

[0051] The acquisition module is used to acquire user information, including the user's age, height, weight, and symptoms. The symptoms refer to respiratory symptoms the user has at the current time. Multiple left lung respiratory discrimination points and multiple right lung respiratory discrimination points are acquired. The left lung respiratory discrimination point indicates the location of the sensor connected to the left lung on the body surface; the right lung respiratory discrimination point indicates the location of the sensor connected to the right lung on the body surface.

[0052] The feature detection module is used to obtain a first estimated parameter feature vector and a second estimated parameter feature vector based on multiple left lung breathing discrimination points and multiple right lung breathing discrimination points through an attention mechanism. The first estimated parameter feature vector represents the correlation between the fluctuation state of multiple points around the lung surface during breathing and the breathing parameters. The second estimated parameter feature vector represents the correlation between the fluctuation state of the lung surface during breathing and user information and the breathing parameters.

[0053] The discrimination module is used to superimpose the first estimated parameter vector and the second estimated parameter vector and input them into the second discrimination neural network to obtain respiratory parameter values; the respiratory parameter values ​​represent parameters about breathing predicted by user information and respiratory state;

[0054] The attention mechanism and the second discriminant neural network are included in the breathing parameter estimation model; the breathing parameter estimation model is a network that integrates a physical model during training.

[0055] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0056] This invention also provides a method and system for estimating respiratory parameters that integrates physical models and attention mechanisms.

[0057] In this invention, because the chest and abdomen fluctuate as a whole during respiration, changes at multiple points during respiration are considered, thus more accurately detecting the relationship between chest and abdominal fluctuations and respiratory parameters. A Bayesian optimization algorithm is used to define the regions of the left and right lungs, identifying the unsampled locations with the greatest change during a single breath as the center points of the left and right lungs. At multiple time points, the distances of chest and abdominal fluctuations to the center points of the left and right lungs are determined using surrounding left and right lung respiratory discrimination points, respectively, and features are extracted using a convolutional neural network. Convolution is performed on the three-dimensional feature maps of the left and right lungs constructed based on these features, enabling not only the detection of changes but also the determination of the relationship between the point of change and the center point of the left or right lung. Furthermore, an attention mechanism is employed to find the correlation between the chest and abdominal fluctuations of multiple surrounding left lung respiratory discrimination points and the center point of the left lung on respiratory parameter discrimination, and these correlations are weighted. A gating mechanism is used, employing user information as a constraint to determine whether the user's height, weight, or medical condition will lead to changes in respiratory parameters. A physical model is also integrated during network training. This technology achieves the effect of more accurately judging respiratory parameters by using the rise and fall of the chest and abdomen. Attached Figure Description

[0058] Figure 1 This is a flowchart of a breathing parameter estimation method that integrates a physical model and an attention mechanism, provided by an embodiment of the present invention. Detailed Implementation

[0059] The present invention will now be described in detail with reference to the accompanying drawings.

[0060] Example 1:

[0061] This invention relates to a respiratory parameter estimation method that integrates a physical model and an attention mechanism to estimate respiratory parameters by utilizing the rise and fall of the chest and abdomen.

[0062] like Figure 1 As shown, this embodiment of the invention provides a method for estimating respiratory parameters that integrates a physical model and an attention mechanism. The method includes:

[0063] S101: Obtain user information; the user information includes the user's age, height, weight, and illness; the illness refers to the respiratory illness the user has at the current time.

[0064] The conditions mentioned include bronchitis, pneumonia, and rhinitis.

[0065] S102: Obtain multiple left lung breathing discrimination points and multiple right lung breathing discrimination points; the left lung breathing discrimination point indicates the location on the body surface of the left lung where a sensor is connected; the right lung breathing discrimination point indicates the location on the body surface of the right lung where a sensor is connected.

[0066] In this embodiment, the sensor is a strain sensor.

[0067] S103: Through an attention mechanism, based on multiple left lung breathing discrimination points and multiple right lung breathing discrimination points, a first estimation parameter feature vector and a second estimation parameter feature vector are obtained; the first estimation parameter feature vector represents the correlation between the fluctuation state of multiple points around the lung surface during breathing and the breathing parameters; the second estimation parameter feature vector represents the correlation between the fluctuation state of the lung surface during breathing and user information and the breathing parameters.

[0068] S104: The first estimated parameter vector and the second estimated parameter vector are superimposed and then input into the second discriminative neural network to obtain the respiratory parameter value; the respiratory parameter value represents the parameters about breathing predicted by user information and respiratory state.

[0069] Specifically, the first estimated parameter vector is concatenated with the second estimated parameter vector to obtain the respiratory fusion vector. The number of elements in the respiratory fusion vector is equal to the sum of the number of elements in the second estimated parameter vector and the number of elements in the first estimated parameter vector.

[0070] The attention mechanism and the second discriminant neural network are included in the breathing parameter estimation model; the breathing parameter estimation model is a model that introduces a physical model to construct loss constraints for network training.

[0071] Among them, the breathing parameter estimation model incorporates the physical laws described by the physical model as a loss term during training, and imposes physical consistency constraints on the network output.

[0072] The second discriminant neural network is a fully connected neural network (FCN).

[0073] Optionally, the step of obtaining the first estimated parameter feature vector based on the multiple left lung breathing discrimination points and multiple right lung breathing discrimination points through an attention mechanism includes:

[0074] Based on the multiple left lung breathing discrimination points and multiple right lung breathing discrimination points, multiple left lung breathing maps, multiple right lung breathing maps, left lung center point, right lung center point, left lung center breathing map, and right lung center breathing map are obtained.

[0075] The left lung central respiratory map, the right lung central respiratory map, multiple left lung respiratory maps, and multiple right lung respiratory maps are respectively input into the first feature extraction network to detect the features of the body surface due to the fluctuation of lung breathing, and obtain the right lung central feature vector, the left lung central feature vector, multiple left lung feature vectors, and multiple right lung feature vectors.

[0076] The first feature extraction network is a shared-weight convolutional neural network (CNN). The flatten function is used to transform the feature map into a feature vector.

[0077] The first estimation parameter feature vector is obtained by using the first discriminative neural network and attention mechanism based on the right lung center feature vector, the left lung center feature vector, multiple left lung feature vectors, and multiple right lung feature vectors.

[0078] Optionally, the step of obtaining the first estimated parameter feature vector based on the right lung center feature vector, the left lung center feature vector, multiple left lung feature vectors, and multiple right lung feature vectors through the first discriminative neural network and attention mechanism includes:

[0079] Calculate the distances between the multiple left lung center points and the left lung breathing discrimination points to obtain multiple first distances.

[0080] In this embodiment, the calculation is performed using the Euclidean distance method.

[0081] The first estimation parameter feature vector is obtained by using the first discriminative neural network and attention mechanism based on the right lung center feature vector, the left lung center feature vector, multiple left lung feature vectors, and multiple right lung feature vectors.

[0082] The right lung central feature vector and multiple right lung feature vectors correspond to obtain the right lung three-dimensional feature map.

[0083] The three-dimensional feature map of the left lung is input into a first three-dimensional convolutional network to obtain the feature vector of the left lung; the three-dimensional feature map of the right lung is input into a second three-dimensional convolutional network to obtain the feature vector of the right lung.

[0084] The first and second three-dimensional convolutional networks are three-dimensional convolutional neural networks (3D-CNN) containing 2*2*2 three-dimensional convolutional kernels.

[0085] The feature vectors of the left lung and the three-dimensional feature vectors of the right lung are input into the first discriminative neural network to obtain the first estimated parameter feature vector.

[0086] In this embodiment, the feature vectors of the left lung and the three-dimensional feature vectors of the right lung are superimposed and then input into the first discriminative neural network. In this embodiment, the first discriminative neural network is a fully connected neural network (FCN).

[0087] Optionally, the step of obtaining a three-dimensional feature map of the left lung through an attention mechanism, based on multiple left lung breathing discrimination points and the left lung center point, and on the multiple first distances, the left lung center feature vector, and multiple left lung feature vectors, includes:

[0088] Based on multiple left lung feature vectors, multiple left lung distance feature vectors are obtained;

[0089] Multiple left lung distance feature vectors are input into the attention mechanism to obtain multiple left lung attention feature vectors; the left lung attention feature vectors represent the weighted left lung distance feature vectors after determining the association state with the breathing state.

[0090] Specifically, the left lung distance feature vector is weighted using a trained attention mechanism.

[0091] The left lung attention feature vector is filled into the position of the left lung breathing discrimination point, and the left lung center feature vector is filled into the position of the left lung center point to obtain the three-dimensional feature map of the left lung.

[0092] Specifically, a coordinate axis is established based on the surface region of the lungs. Based on the position of the left lung breathing discrimination point on the coordinate axis, the 1 representing the left lung breathing discrimination point is replaced with the left lung attention feature vector, the 2 representing the left lung center point is replaced with the left lung center feature vector, and other points are padded with zeros to obtain the three-dimensional feature map of the left lung.

[0093] The above method utilizes the principle that breathing involves the movement of the entire chest and abdomen. By identifying a central point on the left lung and multiple surrounding left lung respiratory reference points, the correlation between the left lung central respiratory map and the left lung respiratory map is determined by comparing the differences in surface undulations of the chest and abdomen between these reference points and the central point. Finally, using a three-dimensional feature map of the left lung, a correlation is established between the respiratory state of the central point and the respiratory states of the surrounding left lung respiratory reference points, thus enabling further judgment.

[0094] Optionally, the process of obtaining multiple left lung distance feature vectors based on multiple left lung feature vectors includes:

[0095] Subtracting the left lung center feature vector from the left lung feature vector yields the left lung difference vector.

[0096] In this context, the value of one subscript in the left lung difference vector represents the value of the corresponding subscript in the left lung feature vector minus the value of the corresponding subscript in the left lung center feature vector.

[0097] Divide the left lung difference vector by the corresponding first distance to obtain the left lung distance feature vector; the left lung distance feature vector represents the correlation between the surface undulation features of the left lung breathing discrimination point and the distance to the center point of the left lung.

[0098] Multiple left lung feature vectors correspond to multiple left lung distance feature vectors.

[0099] Optionally, the step of obtaining the second estimation parameter feature vector through the second feature extraction network, based on the left lung center feature vector, the right lung center feature vector, and multiple user information, includes:

[0100] User information is input into the second feature extraction network to obtain the user feature vector.

[0101] The lung feature vector is obtained by superimposing the feature vectors of the left and right lung centers.

[0102] The number of elements in the lung feature vector is equal to the sum of the number of elements in the left lung central feature vector and the number of elements in the right lung central feature vector.

[0103] Based on the user feature vector and lung feature vector, a second estimation parameter feature vector is obtained through a gating mechanism.

[0104] The method for obtaining the feature vector of the second estimation parameter is shown in the following formula:

[0105]

[0106] in, This represents the eigenvector of the second estimated parameter. Represents the lung feature vector. This represents the user feature vector.

[0107] Optionally, the first discriminant neural network, the second discriminant neural network, the first three-dimensional convolutional network, the second three-dimensional convolutional network, the first feature extraction network, the second feature extraction network, and the attention mechanism constitute a breathing parameter estimation model.

[0108] The inputs to the first feature extraction network are the left lung central respiratory map, the right lung central respiratory map, multiple left lung respiratory maps, and multiple right lung respiratory maps; the output of the first feature extraction network is the input of the attention mechanism; the outputs of the attention mechanism are the inputs of the first three-dimensional convolutional network and the second three-dimensional convolutional network, respectively; the outputs of the first three-dimensional convolutional network and the second three-dimensional convolutional network are the inputs of the first discriminative neural network; the input of the first feature extraction network is user information; the output of the first feature extraction network is the input of the second feature extraction network; the outputs of the first discriminative neural network and the second feature extraction network are the inputs of the second discriminative neural network; and the output of the second discriminative neural network is respiratory parameter values.

[0109] Optionally, the training method for the respiratory parameter estimation model includes:

[0110] Acquire training data and labeled data; the training data consists of user information at historical time points, left lung central respiratory map, right lung central respiratory map, multiple left lung respiratory maps, and multiple right lung respiratory maps; the labeled data represents the measured respiratory parameters.

[0111] The training data is input into the respiratory parameter estimation model to obtain the first respiratory parameter value.

[0112] In this embodiment, the first respiratory parameter value is tidal volume.

[0113] Based on the first respiratory parameter value, the loss value of the first respiratory parameter is obtained through a physical model.

[0114] In this embodiment, the physical model used is the respiratory motion equation.

[0115] Using the above method, the role of the physical model is to constrain the output to conform to the physical state of the human body.

[0116] The loss is calculated by combining the first respiratory parameter value and the labeled data to obtain the second respiratory parameter loss value.

[0117] In this embodiment, the mean squared error loss function is used to calculate the loss.

[0118] The first respiratory parameter loss value and the second respiratory parameter loss value are added together, and the respiratory parameter estimation model is trained backward.

[0119] Using the above method, the difference between the detected and actual measured tidal volume, together with whether the physical model satisfies the respiratory motion equation, is used as a loss to train the respiratory parameter estimation model.

[0120] Optionally, the step of obtaining multiple left lung respiratory maps, multiple right lung respiratory maps, left lung center point, right lung center point, left lung center respiratory map, and right lung center respiratory map based on the multiple left lung respiratory discrimination points and multiple right lung respiratory discrimination points includes:

[0121] Multiple left lung breathing images are obtained at multiple left lung breathing discrimination points; multiple right lung breathing images are obtained at multiple right lung breathing discrimination points; the left lung breathing image represents the degree of skin undulation during breathing in the left lung; the right lung breathing image represents the degree of skin undulation during breathing in the right lung.

[0122] In this embodiment, a strain sensor is used to obtain the surface undulation values ​​of the chest and abdomen corresponding to the left lung during breathing.

[0123] Based on multiple left lung breathing images, the location of the greatest skin undulation is detected using a Bayesian optimization algorithm to obtain the center point of the left lung; the center point of the right lung is obtained corresponding to multiple right lung breathing images.

[0124] Among them, the maximum distance difference of skin undulation during one breath is detected, and the average of multiple maximum distance differences during multiple breaths is used as the value corresponding to each left lung breathing discrimination point.

[0125] In this embodiment, the `gpminimize` function in the `skopt` library of Python is used to compute the Bayesian optimization algorithm. The region containing the left lung is used as the search space; in this embodiment, the center point of the left lung region is designated as [0,0]. 20 is used as the number of iterations. The values ​​corresponding to the left lung breathing discrimination points are used as the initial data. The acquisition function is set to EI (Expected Improvement). The acquisition function optimizer is set to random search.

[0126] At the center point of the left lung, obtain the respiratory image of the left lung; at the center point of the right lung, obtain the respiratory image of the right lung.

[0127] In this embodiment, strain sensors are used to acquire respiratory images of the left and right lung centers.

[0128] Example 2:

[0129] Based on the above-described method for estimating respiratory parameters by fusing physical models and attention mechanisms, this invention also provides a respiratory parameter estimation system that fuses physical models and attention mechanisms, the system comprising:

[0130] The acquisition module is used to acquire user information, including the user's age, height, weight, and symptoms. The symptoms refer to respiratory symptoms the user has at the current time. Multiple left lung respiratory discrimination points and multiple right lung respiratory discrimination points are acquired. The left lung respiratory discrimination point indicates the location of the sensor connected to the left lung on the body surface; the right lung respiratory discrimination point indicates the location of the sensor connected to the right lung on the body surface.

[0131] The feature detection module is used to obtain a first estimated parameter feature vector and a second estimated parameter feature vector based on multiple left lung breathing discrimination points and multiple right lung breathing discrimination points through an attention mechanism. The first estimated parameter feature vector represents the correlation between the fluctuation state of multiple points around the lung surface during breathing and the breathing parameters. The second estimated parameter feature vector represents the correlation between the fluctuation state of the lung surface during breathing and user information and the breathing parameters.

[0132] The discrimination module is used to superimpose the first estimated parameter vector and the second estimated parameter vector and input them into the second discrimination neural network to obtain respiratory parameter values; the respiratory parameter values ​​represent parameters about breathing predicted by user information and respiratory state.

[0133] The attention mechanism and the second discriminant neural network are included in the breathing parameter estimation model; the breathing parameter estimation model is a network that integrates a physical model during training.

[0134] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0135] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0136] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. A method for estimating respiratory parameters that integrates physical models and attention mechanisms, characterized in that, include: Obtain user information; the user information includes the user's age, height, weight, and medical conditions; The symptoms described refer to respiratory symptoms that the user is experiencing at the current point in time. Multiple left lung breathing discrimination points and multiple right lung breathing discrimination points are acquired; the left lung breathing discrimination points represent the locations on the body surface where sensors are connected; The right lung breathing discrimination point indicates the location on the body surface of the right lung where the sensor is connected; Through an attention mechanism, based on multiple left lung breathing discrimination points and multiple right lung breathing discrimination points, a first estimation parameter feature vector and a second estimation parameter feature vector are obtained. The first estimation parameter feature vector represents the correlation between the fluctuation state of multiple points around the lung surface during breathing and the breathing parameters. The second estimation parameter feature vector represents the correlation between the fluctuation state of the lung surface during breathing and user information and the breathing parameters. The first and second estimated parameter vectors are superimposed and then input into the second discriminative neural network to obtain respiratory parameter values; the respiratory parameter values ​​represent parameters about breathing predicted from user information and respiratory status; The attention mechanism and the second discriminant neural network are included in the breathing parameter estimation model; the breathing parameter estimation model is a model that introduces a physical model to construct loss constraints for network training.

2. The method for estimating respiratory parameters by integrating a physical model and an attention mechanism according to claim 1, characterized in that, The first estimated parameter feature vector is obtained through an attention mechanism based on the multiple left lung breathing discrimination points and multiple right lung breathing discrimination points, including: Based on the multiple left lung breathing discrimination points and multiple right lung breathing discrimination points, multiple left lung breathing maps, multiple right lung breathing maps, left lung center point, right lung center point, left lung center breathing map, and right lung center breathing map are obtained; The left lung central respiratory map, the right lung central respiratory map, multiple left lung respiratory maps, and multiple right lung respiratory maps are respectively input into the first feature extraction network to detect the features of the body surface due to the fluctuation state of lung respiration, and obtain the right lung central feature vector, the left lung central feature vector, multiple left lung feature vectors, and multiple right lung feature vectors. The first estimation parameter feature vector is obtained by using the first discriminative neural network and attention mechanism based on the right lung center feature vector, the left lung center feature vector, multiple left lung feature vectors, and multiple right lung feature vectors; The second feature extraction network is used to obtain the second estimation parameter feature vector based on the feature vectors of the left lung center, the right lung center, and multiple user information.

3. The method for estimating respiratory parameters by integrating a physical model and an attention mechanism according to claim 2, characterized in that, The first estimated parameter feature vector is obtained by using a first discriminative neural network and an attention mechanism, based on the right lung center feature vector, the left lung center feature vector, multiple left lung feature vectors, and multiple right lung feature vectors, including: Calculate the distances between the multiple left lung center points and the left lung breathing discrimination points to obtain multiple first distances; Using an attention mechanism, based on multiple left lung breathing discrimination points and the left lung center point, and based on the multiple first distances, the left lung center feature vector, and multiple left lung feature vectors, a three-dimensional feature map of the left lung is obtained. The right lung center feature vector and multiple right lung feature vectors correspond to obtain a three-dimensional feature map of the right lung; The three-dimensional feature map of the left lung is input into a first three-dimensional convolutional network to obtain the feature vector of the left lung; the three-dimensional feature map of the right lung is input into a second three-dimensional convolutional network to obtain the feature vector of the right lung. The feature vectors of the left lung and the three-dimensional feature vectors of the right lung are input into the first discriminative neural network to obtain the first estimated parameter feature vector.

4. The method for estimating respiratory parameters by integrating a physical model and an attention mechanism according to claim 3, characterized in that, The process involves using an attention mechanism to obtain a three-dimensional feature map of the left lung based on multiple left lung breathing discrimination points and the left lung center point, along with multiple first distances, the left lung center feature vector, and multiple left lung feature vectors. Based on multiple left lung feature vectors, multiple left lung distance feature vectors are obtained; Multiple left lung distance feature vectors are input into the attention mechanism to obtain multiple left lung attention feature vectors; the left lung attention feature vectors represent the weighted left lung distance feature vectors after determining the association state with the breathing state; The left lung attention feature vector is filled into the position of the left lung breathing discrimination point, and the left lung center feature vector is filled into the position of the left lung center point to obtain the three-dimensional feature map of the left lung.

5. The method for estimating respiratory parameters by integrating a physical model and an attention mechanism according to claim 4, characterized in that, The process involves obtaining multiple left lung distance feature vectors based on multiple left lung feature vectors, including: Subtract the left lung center feature vector from the left lung feature vector to obtain the left lung difference vector; Divide the left lung difference vector by the corresponding first distance to obtain the left lung distance feature vector; the left lung distance feature vector represents the correlation between the surface undulation features of the left lung breathing discrimination point and the distance to the center point of the left lung. Multiple left lung feature vectors correspond to multiple left lung distance feature vectors.

6. The method for estimating respiratory parameters by integrating a physical model and an attention mechanism according to claim 2, characterized in that, The second feature extraction network obtains the second estimation parameter feature vector based on the left lung center feature vector, the right lung center feature vector, and multiple user information, including: User information is input into the second feature extraction network to obtain the user feature vector; The lung feature vector is obtained by superimposing the feature vectors of the left and right lung centers. Based on the user feature vector and lung feature vector, a second estimation parameter feature vector is obtained through a gating mechanism; The method for obtaining the feature vector of the second estimation parameter is shown in the following formula: ； in, This represents the eigenvector of the second estimated parameter. Represents the lung feature vector. This represents the user feature vector.

7. The method for estimating respiratory parameters by integrating a physical model and an attention mechanism according to claim 5, characterized in that, The first discriminant neural network, the second discriminant neural network, the first three-dimensional convolutional network, the second three-dimensional convolutional network, the first feature extraction network, the second feature extraction network, and the attention mechanism constitute a breathing parameter estimation model.

8. The method for estimating respiratory parameters by integrating a physical model and an attention mechanism according to claim 6, characterized in that, The training method for the respiratory parameter estimation model includes: Acquire training data and labeled data; the training data consists of user information at historical time points, left lung central respiratory map, right lung central respiratory map, multiple left lung respiratory maps, and multiple right lung respiratory maps; the labeled data represents the measured respiratory parameters. The training data is input into the respiratory parameter estimation model to obtain the first respiratory parameter value; Based on the first respiratory parameter value, the loss value of the first respiratory parameter is obtained through a physical model; The loss is calculated using the first respiratory parameter value and the labeled data to obtain the second respiratory parameter loss value; The first respiratory parameter loss value and the second respiratory parameter loss value are added together, and the respiratory parameter estimation model is trained backward.

9. The method for estimating respiratory parameters by integrating a physical model and an attention mechanism according to claim 2, characterized in that, The process of obtaining multiple left lung respiratory discriminant points, multiple right lung respiratory maps, left lung center point, right lung center point, left lung center respiratory map, and right lung center respiratory map based on the multiple left lung respiratory discriminant points and multiple right lung respiratory discriminant points includes: Multiple left lung breathing maps were obtained at multiple left lung breathing discrimination points; the left lung breathing map represents the degree of skin undulation during breathing. Multiple right lung breathing maps were obtained at multiple right lung breathing discrimination points; the right lung breathing map represents the degree of skin undulation during breathing. Based on multiple left lung breathing images, the location of the greatest skin undulation is detected using a Bayesian optimization algorithm to obtain the center point of the left lung; the center point of the right lung is obtained corresponding to multiple right lung breathing images. At the center point of the left lung, obtain the respiratory image of the left lung; at the center point of the right lung, obtain the respiratory image of the right lung.

10. A respiratory parameter estimation system integrating physical models and attention mechanisms, characterized in that, include: The acquisition module is used to acquire user information, including the user's age, height, weight, and medical conditions. The symptoms refer to respiratory symptoms that the user has at the current point in time; multiple left lung respiratory discrimination points and multiple right lung respiratory discrimination points are obtained; the left lung respiratory discrimination points refer to the locations on the body surface where sensors are connected to the left lung; The right lung breathing discrimination point indicates the location on the body surface of the right lung where the sensor is connected; The feature detection module is used to obtain a first estimated parameter feature vector and a second estimated parameter feature vector based on multiple left lung breathing discrimination points and multiple right lung breathing discrimination points through an attention mechanism. The first estimated parameter feature vector represents the correlation between the fluctuation state of multiple points around the lung surface during breathing and the breathing parameters. The second estimated parameter feature vector represents the correlation between the fluctuation state of the lung surface during breathing and user information and the breathing parameters. The discrimination module is used to superimpose the first estimated parameter vector and the second estimated parameter vector and input them into the second discrimination neural network to obtain the breathing parameter value; the breathing parameter value represents the parameters about breathing predicted by user information and breathing state; the attention mechanism and the second discrimination neural network are included in the breathing parameter estimation model; the breathing parameter estimation model is a network that integrates the physical model during training.

Images