Lung function examination and diagnosis method and device based on artificial intelligence mixed feature recognition

Abstract

The invention provides a pulmonary function examination and diagnosis method and device based on artificial intelligence mixed feature recognition, and the method comprises the steps: obtaining personal information representing the health features of a subject, and carrying out the extraction to obtain information features; feature extraction is carried out on the multiple lung function examination images of the subject, and the time sequence feature of each lung function examination image is obtained; and constructing a multi-channel lung function diagram based on the plurality of lung function examination images of the subject, and performing feature extraction on the multi-channel lung function diagram to obtain multi-channel image features. The invention provides a pulmonary function examination and diagnosis method based on artificial intelligence mixed feature recognition. Fusion examination and diagnosis of various types of pulmonary function examination images are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a lung function examination diagnosis method and device based on artificial intelligence mixed feature recognition. BACKGROUND

[0002] Lung function examination refers to the detection and evaluation of the respiratory function of the examinee by using specific means and instruments, which is an important method for describing respiratory function, involving respiratory mechanics, fluid mechanics and thermodynamics, etc. It is theoretically complex. After a series of tests and complex computer calculations, a large number of professional data and real-time monitoring images are listed, and then artificial interpretation is performed on these data and images to provide guiding diagnostic opinions for the diagnosis and treatment of clinicians. Lung function examination has important significance in the diagnosis of chest and lung diseases, state detection and evaluation of lung function damage.

[0003] Lung function examination requires a professional operator to guide the examinee to cooperate with the instructions to perform a series of inhalation and exhalation actions, and then the lung function instrument detects and calculates a series of professional data indicators, and displays the images obtained by the examination. Finally, artificial interpretation is required to issue a diagnostic report.

[0004] However, artificial interpretation often has problems such as image interpretation difficulty, data interpretation error, and different interpretation standards, which is time-consuming and labor-intensive, causing lung function examination reports issued by different hospitals to be incompatible. The above problems of report interpretation, which are heavily dependent on the experience and professional quality of the operator, combined with the lack of emphasis on lung function examination in primary hospitals and the low popularity of related knowledge, result in a serious shortage of professional lung function examination operators, so that lung function examination has not been widely popularized in China so far. SUMMARY

[0005] The present application provides a lung function examination diagnosis method and device based on artificial intelligence mixed feature recognition to solve the defects of difficult artificial interpretation of lung function examination reports and the lack of professional lung function examination operators in the prior art, and to realize a lung function examination diagnosis method and device based on artificial intelligence mixed feature recognition.

[0006] The present application provides a lung function examination diagnosis method based on artificial intelligence mixed feature recognition, comprising: Obtaining personal information representing the health characteristics of the examinee and extracting information features; Respectively extracting features from multiple lung function examination images of the examinee to obtain time sequence features of each lung function examination image; Constructing a multi-channel lung function graph based on the multiple lung function examination images of the examinee, extracting features from the multi-channel lung function graph, and obtaining multi-channel image features; The temporal features, information features, and multi-channel image features of each lung function test image are fused and then input into a pre-trained classifier to obtain the lung function diagnosis result of the subject output by the classifier as normal, obstructed, restricted, or mixed.

[0007] According to the present invention, a lung function test diagnostic method based on artificial intelligence hybrid feature recognition is provided, wherein the lung function test images include slow vital capacity map, forced vital capacity map, and maximum voluntary ventilation map.

[0008] According to the present invention, a method for diagnosing lung function tests using artificial intelligence hybrid feature recognition includes the step of extracting features from multiple lung function test images of the subject to obtain the temporal features of each lung function test image, specifically comprising: Each lung function test image of the subject is input into a long short-term memory network to obtain the output timing of each lung function test image output by the long short-term memory network; The output time sequence is weighted using an attention mechanism, and the weighted output time sequence is used as the time sequence feature of the lung function test image.

[0009] According to the present invention, a method for pulmonary function testing and diagnosis based on artificial intelligence hybrid feature recognition, the step of constructing a multi-channel pulmonary function map based on multiple pulmonary function test images of the subject specifically includes: Multiple lung function test images of the subject were cropped to the same size; Multiple cropped lung function test images are combined to obtain a multi-channel lung function map.

[0010] According to the present invention, a method for pulmonary function testing and diagnosis using artificial intelligence hybrid feature recognition, the step of extracting features from the multi-channel pulmonary function map to obtain multi-channel image features specifically includes: The convolutional neural network is used to extract features from the multi-channel pulmonary function images to obtain multi-channel image features. The convolutional neural network includes a set of convolutional kernels, and the number of the multiple convolutional kernels is the same as the number of the multiple pulmonary function examination images.

[0011] The lung function test diagnostic method based on artificial intelligence hybrid feature recognition provided by the present invention further includes: A cross-entropy loss function is designed. If the value of the cross-entropy loss function is less than a preset threshold, the lung function diagnostic model composed of the long short-term memory network, the convolutional neural network, and the classifier is considered to have completed training.

[0012] The present invention also provides a lung function testing and diagnostic device based on artificial intelligence hybrid feature recognition, comprising: The personal information acquisition module is used to acquire personal information that characterizes the health features of the examinee and extract information features; The temporal feature acquisition module is used to extract features from multiple lung function test images of the subject to obtain the temporal features of each lung function test image. The spatial feature acquisition module is used to construct a multi-channel pulmonary function map based on multiple pulmonary function test images of the subject, and to extract features from the multi-channel pulmonary function map to obtain multi-channel image features; The diagnostic module is used to fuse the temporal features, information features, and multi-channel image features of each lung function test image and input them into a pre-trained classifier to obtain the lung function diagnosis result of the subject output by the classifier as normal, obstructed, restricted, or mixed.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the lung function test diagnosis method of artificial intelligence hybrid feature recognition as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the lung function test diagnosis method of artificial intelligence hybrid feature recognition as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a lung function test and diagnosis method based on artificial intelligence hybrid feature recognition as described above.

[0016] The present invention provides a method and apparatus for pulmonary function examination diagnosis based on artificial intelligence hybrid feature recognition. By combining information features that characterize the health status of the examinee, temporal features that characterize the temporal characteristics of the examinee's pulmonary function examination images, and multi-channel image features that characterize the spatial characteristics of the examinee's multi-channel images, the diagnostic results of the examinee's pulmonary function status are obtained. This improves the accuracy and consistency of the diagnostic report, reduces the error rate of human judgment, saves judgment time, and is conducive to promotion. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1This is a flowchart illustrating the lung function test and diagnosis method based on artificial intelligence hybrid feature recognition provided by the present invention. Figure 2 This is a schematic diagram of the neural network used to extract personal information in the lung function test and diagnosis method of artificial intelligence hybrid feature recognition provided by the present invention; Figure 3 This is a schematic diagram of a multi-channel pulmonary function map in the pulmonary function test and diagnosis method based on artificial intelligence hybrid feature recognition provided by the present invention; Figure 4 This is a schematic diagram of the chronic vital capacity map in the lung function test and diagnosis method based on artificial intelligence hybrid feature recognition provided by the present invention; Figure 5 This is a schematic diagram of the forced vital capacity graph in the lung function test and diagnosis method of artificial intelligence hybrid feature recognition provided by the present invention; Figure 6 This is a schematic diagram of the maximum spontaneous ventilation map in the lung function test and diagnosis method of artificial intelligence hybrid feature recognition provided by the present invention; Figure 7 This is a schematic diagram of the convolutional neural network structure in the lung function examination and diagnosis method of artificial intelligence hybrid feature recognition provided by the present invention; Figure 8 This is a schematic diagram of the feature splicing and straightening process in the lung function test diagnosis method of artificial intelligence hybrid feature recognition provided by the present invention; Figure 9 This is a schematic diagram of the diagnostic model in the lung function test diagnostic method based on artificial intelligence hybrid feature recognition provided by the present invention; Figure 10 This is a schematic diagram of the structure of the artificial intelligence hybrid feature recognition lung function testing and diagnostic device provided by the present invention; Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] The following is combined Figures 1 to 9 This application introduces a lung function test diagnostic method based on artificial intelligence hybrid feature recognition, such as... Figure 1 As shown, it includes: Step 101: Obtain personal information characterizing the health characteristics of the examinee and extract the information features; The examinee is a patient who needs to be diagnosed with lung function. The examinee's personal information is used to characterize the examinee's health characteristics, mainly including the examinee's gender (G), height (H), weight (W) and age (A).

[0021] Optionally, other data that can directly characterize the subject's physical condition, such as waist circumference, abdominal circumference, and body fat, may also be included.

[0022] Optionally, data that indirectly affects the physical condition of the examinee, such as season, region, temperature, and humidity, may also be included.

[0023] Based on the acquired personal information, construct a personal information vector. P In one feasible implementation:

[0024] Personal information vectors are processed through a fully connected layer. P Perform feature extraction, such as Figure 2 As shown, the subject's gender, height, weight, and age are used as inputs to the four neural networks in the input layer, respectively. After passing through subsequent hidden layers, the output feature vector is obtained. P 1. As an information feature: ; Understandably, the number of different categories of personal information corresponds to the number of neural networks designed for the input layer.

[0025] Step 102: Extract features from multiple lung function test images of the subject to obtain temporal features of each lung function test image; Optionally, the lung function test images represent medical images of the lung function under different instructions obtained after the subject performs a series of operations according to the operator's instructions. The operator's instructions are determined according to the actual needs of medical diagnosis.

[0026] Understandably, the type and number of multiple lung function test images are determined based on the actual needs of medical diagnosis.

[0027] For example, the actual needs of medical diagnosis require obtaining slow vital capacity maps, forced vital capacity maps, and maximum spontaneous ventilation maps. In this case, corresponding instructions are determined for each type of medical image, and the test images of the examinee under the corresponding instructions are recorded as medical images of the corresponding category. Finally, three lung function test images are obtained for feature extraction.

[0028] Optionally, a Long Short-Term Memory (LSTM) network can be used to extract features from each lung function test image to obtain the temporal features of each lung function image.

[0029] It is understandable that, regardless of the category of lung function test images, they all represent a set of data sequences. Therefore, by pre-training an LSTM model, it is possible to extract features from lung function test images and obtain the temporal features corresponding to each lung function test image.

[0030] Step 103: Construct a multi-channel pulmonary function map based on multiple pulmonary function test images of the subject, and extract features from the multi-channel pulmonary function map to obtain multi-channel image features; Integrating multiple lung function test images of a subject into a single multichannel lung function map, it is understood that the number of channels is the same as the number of categories of lung function test images.

[0031] Optionally, a convolutional neural network can be pre-trained to extract features from the multi-channel pulmonary function map, thereby obtaining multi-channel image features that characterize the spatial features of the multi-channel pulmonary function map.

[0032] Compared to temporal features extracted from pulmonary function test images, multi-channel image features focus more on the two-dimensional image spatial features of pulmonary function information represented by pulmonary function images.

[0033] In one feasible implementation, when multiple pulmonary function test images are respectively a slow vital capacity map, a forced vital capacity map, and a maximum spontaneous ventilation map, the generated three-channel pulmonary function map is as follows: Figure 3 As shown.

[0034] It should be noted that, for example Figure 3 As shown, the curves in the lung function test images can all characterize the changes of independent variables over time to a certain extent. Therefore, the multi-channel image features extracted based on the multi-channel lung function map to characterize the spatial features can also characterize the temporal features of the lung function test images to a certain extent.

[0035] Step 104: The temporal features, information features, and multi-channel image features of each lung function test image are fused and then input into a pre-trained classifier to obtain the lung function diagnosis result of the subject output by the classifier as normal, obstructed, restricted, or mixed.

[0036] After the extracted information features, multi-channel image features, and multiple temporal features are spliced ​​and straightened to obtain combined features, the combined features are input into a pre-trained classifier to obtain the lung function diagnosis results of the examinee output by the classifier.

[0037] The classifier is configured as a four-class classifier, representing the four common states of a patient's lung function: normal, obstructive, restrictive, and mixed.

[0038] Optionally, the classifier includes a fully connected layer (FC), with the last layer of the FC connecting four neurons and passing through a softmax function. The softmax function transforms a vector containing multiple values ​​into a vector representing the probability distributions. In this embodiment, it transforms the vector into the probability of each lung function state, and the sum of the probabilities of the four lung function states is 1.

[0039] Finally, the lung function state with the highest output probability is used as the lung function diagnosis result for the examinee, namely, normal output, obstructed, restricted, or mixed output.

[0040] The above approach addresses the issues of incomplete lung function information extraction caused by traditional methods that rely on human experience or make judgments based solely on local information, thus solving the problems in diagnosis, status monitoring, and evaluation of lung function impairment in the areas of chest and lung disease. It provides an artificial intelligence-based hybrid feature recognition method for lung function examination and diagnosis.

[0041] Furthermore, while overcoming the limitations of manual interpretation of lung function test images, artificial intelligence is used to achieve fusion diagnosis of multiple types of lung function test images. Based on information features, temporal features, and multi-channel image features, the final lung function diagnosis result is obtained. It integrates the spatial and temporal features of lung function data and combines the individual characteristics of the examinee to achieve accurate judgment of the examinee's lung function status.

[0042] This invention combines information features characterizing the health status of the examinee, temporal features characterizing the temporal characteristics of the examinee's pulmonary function test images, and multi-channel image features characterizing the spatial characteristics of the examinee's multi-channel images to obtain diagnostic results of the examinee's pulmonary function status. This improves the accuracy and consistency of diagnostic reports, reduces the error rate of human judgment, saves judgment time, and is conducive to promotion.

[0043] In the lung function test and diagnosis method of artificial intelligence hybrid feature recognition of the present invention, the lung function test images include slow vital capacity map, forced vital capacity map and maximum voluntary ventilation map.

[0044] In this embodiment, the multiple lung function test images include a slow vital capacity (V) image, a forced vital capacity (F) image, and a maximum spontaneous ventilation (M) image.

[0045] Instruct the subject to perform inhalation and exhalation according to the operator's commands. Use a spirometer to measure the change in respiratory volume over time during slow breathing, obtaining a slow vital capacity (CVC) map. Figure 4 As shown.

[0046] Optionally, the slow vital capacity chart V includes at least the following information: measured maximum vital capacity (VC MAX, VCm), measured inspiratory volume (IC), measured tidal volume (VT), measured respiratory rate (BF), measured minute ventilation (MV), and measured expiratory reserve volume (ERV).

[0047] The examinee is instructed to inhale and exhale according to the operator's commands. A spirometer is used to measure the change in respiratory flow rate versus respiratory volume during forced breathing. Flow rate is the volume of fluid passing through a cross-section per unit time. In pulmonary function tests, the time integral of respiratory flow rate equals the respiratory volume, thus obtaining the forced vital capacity (F) graph. Figure 5 As shown.

[0048] Optionally, the forced vital capacity chart F includes at least the following information: measured forced vital capacity (FVC), measured volume in one second (FEV1), measured rate in one second (FEV1%), measured peak expiratory flow (PEF), measured instantaneous expiratory flow at 25% of vital capacity (MEF 75), measured instantaneous expiratory flow at 50% of vital capacity (MEF 50), measured instantaneous expiratory flow at 75% of vital capacity (MEF 25), measured instantaneous expiratory flow at 25%-75% of vital capacity (MMEF 75 / 25), forced expiratory time (FET), and extrapolated volume (Ve).

[0049] Instruct the subject to perform inhalation and exhalation according to the operator's commands. Use a spirometer to measure the change in respiratory volume over time during the subject's maximum spontaneous ventilation, obtaining the maximum spontaneous ventilation map M. Figure 6 As shown.

[0050] Optionally, the maximum spontaneous ventilation map M includes at least the following information: the measured maximum spontaneous minute ventilation (MVV).

[0051] In the lung function test diagnosis method based on artificial intelligence hybrid feature recognition of the present invention, the step of extracting features from multiple lung function test images of the subject to obtain the temporal features of each lung function test image specifically includes: Each lung function test image of the subject is input into a long short-term memory network to obtain the output timing of each lung function test image output by the long short-term memory network; The output time sequence is weighted using an attention mechanism, and the weighted output time sequence is used as the time sequence feature of the lung function test image.

[0052] Specifically, in this embodiment, features are extracted from multiple lung function test images by fusing long short-term memory networks and self-attention mechanisms.

[0053] Taking multiple lung function test images as examples, namely slow vital capacity (SVC), forced vital capacity (FVC), and maximum spontaneous ventilation (VPV), the following explanation is given: First, the SVC, FVC, and VPV images are input to the Long Short-Term Memory (LSTM) network, respectively, to obtain the output time sequence of each image output by the LSM network.

[0054] The two outputs of the Long Short-Term Memory network are the output sequences of all time steps. and the last time step Hidden state .

[0055] because Representing data characteristics, Representing temporal characteristics, therefore, it is possible to establish and The target attention relationship, that is, establishing the output at each time step. right The weights. Since LSTM itself takes position information into account, no additional position encoding is needed.

[0056] Based on this, the output of each time step After linear transformation, it serves as the key and value, representing the last time step. Multiply by matrix As a query matrix (Query).

[0057] For time step t, , And the score corresponding to each time step in the query. and weight It is calculated as follows: ; In the formula, Query does not change with time step. , and All of these are hyperparameters, obtained through training.

[0058] Alternatively, LSTM can be pre-trained as a standalone model for feature extraction; or LSTM can be combined with other modules (such as classifiers) to form a diagnostic model and trained synchronously.

[0059] Based on this, the weights of each time step are... and Weighted summation yields the attention-based temporal feature vector of lung function: This serves as a temporal feature for each lung function test image.

[0060] By introducing attention mechanisms into the temporal characteristics of the lung function data obtained through the above methods, we can help the long short-term memory neural network to better focus on the important parts of the lung function temporal data. For example, the slow vital capacity curve (V) needs to focus on the difference between the volume of the respiratory peak and the trough in the curve, the forced vital capacity curve (F) needs to focus on whether there is an expiratory peak in the curve and whether there is a depression in the expiratory image, and the maximum voluntary ventilation curve (M) needs to focus on whether the respiratory amplitude of the curve is consistent.

[0061] Self-attention mechanisms endow models with the ability to capture interactions between different parts of the input data. In Long Short-Term Memory (LSTM) networks, by introducing a self-attention layer at each time step, the similarity between the current time step and all previous time steps can be calculated. This method generates a weighted context vector, which is further merged with the input vector at the current time step to form an enhanced input representation. This fusion allows the model to utilize historical information more flexibly and effectively, thereby generating a more accurate current output.

[0062] When multiple pulmonary function test images are slow vital capacity (SVC), forced vital capacity (FVC), and maximum spontaneous ventilation (VVO) images, their corresponding time-series characteristics are represented as follows: , and .

[0063] In the lung function test diagnosis method based on artificial intelligence hybrid feature recognition of the present invention, the step of constructing a multi-channel lung function map based on multiple lung function test images of the subject specifically includes: Multiple lung function test images of the subject were cropped to the same size; Multiple cropped lung function test images are combined to obtain a multi-channel lung function map.

[0064] To construct a multi-channel pulmonary function map, multiple pulmonary function test images first need to be preprocessed.

[0065] Optionally, preprocessing first includes image cropping. Taking multiple lung function test images—CLP, forced vital capacity (FVC), and maximum spontaneous ventilation (MVC)—as an example, the CLP image (V), forced vital capacity image (F), and maximum spontaneous ventilation image (M) are cropped to the same size. .

[0066] The images of the cropped slow vital capacity chart (V), forced vital capacity chart (F), and maximum spontaneous ventilation chart (M) are used. Combined into VFM three-channel pulmonary function chart ,like Figure 3 As shown.

[0067] Understandably, multiple lung function test images are cropped and combined to ultimately obtain a multi-channel lung function map.

[0068] In the lung function test diagnosis method based on artificial intelligence hybrid feature recognition of the present invention, the step of extracting features from the multi-channel lung function map to obtain multi-channel image features specifically includes: The convolutional neural network is used to extract features from the multi-channel pulmonary function images to obtain multi-channel image features. The convolutional neural network includes a set of convolutional kernels, and the number of the multiple convolutional kernels is the same as the number of the multiple pulmonary function examination images.

[0069] In this embodiment, a convolutional neural network is constructed to extract features from the multi-channel lung function map to obtain multi-channel image features.

[0070] Specifically, first, a convolutional layer is constructed, convolution calculations are performed, and then the results are summed: the size of the convolutional kernel is designed to be... The convolution kernel parameters are Design fill size p Fill refers to filling around edge pixels, typically with "0"; design step size. s .

[0071] Under this padding mechanism, the resolution of the image after convolution will be the same as the resolution of the image before convolution, and there is no downsampling.

[0072] Furthermore, the convolution kernel is designed, and its calculation process can be represented by the following mathematical formula: ; ; In the formula, a Indicates the input image. o This represents the output feature map. w These are the convolution kernel parameters; taking VFM three-channel as an example, a set of convolution kernels consists of three kernels, each a two-dimensional array. This indicates that the parameters of the convolution kernel are iterated and summed. It is the floor function, used to round down the result when it is not an integer.

[0073] It is understandable that the number of convolutional kernels varies depending on the number of channels.

[0074] Based on this, an incentive layer is designed, and incentive functions are used in the incentive layer. Relu : .

[0075] Then, the pooling layer is designed. The pooling layer selects (or calculates) a value from a certain range (i.e., the size of the pooling layer) as the representative of this region. For example, if the pooling function is MAX, the maximum value of the image region is selected as the pooled value for that region, as shown below: .

[0076] like Figure 7 As shown, a fully connected layer is designed last. The flattening step mainly connects the convolutional network layer and the fully connected layer, primarily because the FC layer requires a one-dimensional input. Other common methods include global average pooling. The fully connected layer (FC) mainly performs the final feature extraction, ultimately obtaining the final result. Multi-channel image features.

[0077] Optionally, a convolutional neural network can be trained separately for feature extraction; alternatively, the convolutional neural network can be used as a module to form a diagnostic model together with other modules (such as LSTM modules and classifiers) and trained synchronously.

[0078] The lung function test diagnostic method based on artificial intelligence hybrid feature recognition of the present invention further includes: A cross-entropy loss function is designed. If the value of the cross-entropy loss function is less than a preset threshold, the lung function diagnostic model composed of the long short-term memory network, the convolutional neural network, and the classifier is considered to have completed training.

[0079] like Figure 8 As shown, taking multiple lung function test images—chronic vital capacity (CVC), forced vital capacity (FVC), and maximum voluntary ventilation (VVO)—as an example, the information features were finally extracted. Temporal characteristics , and and multi-channel image features The combined feature vector is obtained by splicing and straightening the above features. As input to the classifier.

[0080] In this embodiment, a neural network for extracting information features, an LSTM network for extracting temporal features, a convolutional neural network for extracting multi-channel image features, and a classifier are constructed into a lung function diagnostic model and trained together. Figure 9 As shown.

[0081] During training, the cross-entropy loss function is used, treating the output values ​​of each item as probabilities. The goal is to narrow the gap between the predicted probabilities and the actual classification values. The training objective is for the cross-entropy loss function value to be less than a preset threshold. Typically, the preset threshold is set to [value missing]. .

[0082] Once the diagnostic model has been trained to the training objective, it is believed that it can perform tests on other subjects. That is, by inputting the personal information of other subjects and multiple lung function test images, the diagnostic model can output lung function diagnostic results as normal, obstructive, restrictive, or mixed.

[0083] The following describes the lung function testing and diagnostic device based on artificial intelligence hybrid feature recognition provided by the present invention. The lung function testing and diagnostic device based on artificial intelligence hybrid feature recognition described below can be referred to in correspondence with the lung function testing and diagnostic method based on artificial intelligence hybrid feature recognition described above.

[0084] like Figure 10 As shown, the artificial intelligence-based hybrid feature recognition lung function testing and diagnostic device includes a personal information acquisition module 1001, a temporal feature acquisition module 1002, a spatial feature acquisition module 1003, and a diagnostic module 1004. The personal information acquisition module 1001 is used to acquire personal information that characterizes the health characteristics of the examinee and extract information features; The examinee is a patient who needs to be diagnosed with lung function. The examinee's personal information is used to characterize the examinee's health characteristics, mainly including the examinee's gender (G), height (H), weight (W) and age (A).

[0085] Optionally, other data that can directly characterize the subject's physical condition, such as waist circumference, abdominal circumference, and body fat, may also be included.

[0086] Optionally, data that indirectly affects the physical condition of the examinee, such as season, region, temperature, and humidity, may also be included.

[0087] Based on the acquired personal information, construct a personal information vector. P In one feasible implementation:

[0088] Personal information vectors are processed through a fully connected layer. P Perform feature extraction, such as Figure 2 As shown, the subject's gender, height, weight, and age are used as inputs to the four neural networks in the input layer, respectively. After passing through subsequent hidden layers, the output feature vector is obtained. P 1. As an information feature: ; Understandably, the number of different categories of personal information corresponds to the number of neural networks designed for the input layer.

[0089] The temporal feature acquisition module 1002 is used to extract features from multiple lung function test images of the subject respectively to obtain the temporal features of each lung function test image; Optionally, the lung function test images represent medical images of the lung function under different instructions obtained after the subject performs a series of operations according to the operator's instructions. The operator's instructions are determined according to the actual needs of medical diagnosis.

[0090] Understandably, the type and number of multiple lung function test images are determined based on the actual needs of medical diagnosis.

[0091] For example, the actual needs of medical diagnosis require obtaining slow vital capacity maps, forced vital capacity maps, and maximum spontaneous ventilation maps. In this case, corresponding instructions are determined for each type of medical image, and the test images of the examinee under the corresponding instructions are recorded as medical images of the corresponding category. Finally, three lung function test images are obtained for feature extraction.

[0092] Optionally, a Long Short-Term Memory (LSTM) network can be used to extract features from each lung function test image to obtain the temporal features of each lung function image.

[0093] It is understandable that, regardless of the category of lung function test images, they all represent a set of data sequences. Therefore, by pre-training an LSTM model, it is possible to extract features from lung function test images and obtain the temporal features corresponding to each lung function test image.

[0094] The spatial feature acquisition module 1003 is used to construct a multi-channel pulmonary function map based on multiple pulmonary function test images of the subject, and to extract features from the multi-channel pulmonary function map to obtain multi-channel image features. Integrating multiple lung function test images of a subject into a single multichannel lung function map, it is understood that the number of channels is the same as the number of categories of lung function test images.

[0095] Optionally, a convolutional neural network can be pre-trained to extract features from the multi-channel pulmonary function map, thereby obtaining multi-channel image features that characterize the spatial features of the multi-channel pulmonary function map.

[0096] Compared to temporal features extracted from pulmonary function test images, multi-channel image features focus more on the two-dimensional image spatial features of pulmonary function information represented by pulmonary function images.

[0097] In one feasible implementation, when multiple pulmonary function test images are respectively a slow vital capacity map, a forced vital capacity map, and a maximum spontaneous ventilation map, the generated three-channel pulmonary function map is as follows: Figure 3 As shown.

[0098] It should be noted that, for exampleFigure 3 As shown, the curves in the lung function test images can all characterize the changes of independent variables over time to a certain extent. Therefore, the multi-channel image features extracted based on the multi-channel lung function map to characterize the spatial features can also characterize the temporal features of the lung function test images to a certain extent.

[0099] The diagnostic module 1004 is used to fuse the temporal features, information features, and multi-channel image features of each lung function test image and input them into a pre-trained classifier to obtain the lung function diagnosis result of the subject output by the classifier as normal, obstructed, restricted, or mixed.

[0100] After the extracted information features, multi-channel image features, and multiple temporal features are spliced ​​and straightened to obtain combined features, the combined features are input into a pre-trained classifier to obtain the lung function diagnosis results of the examinee output by the classifier.

[0101] The classifier is configured as a four-class classifier, representing the four common states of a patient's lung function: normal, obstructive, restrictive, and mixed.

[0102] Optionally, the classifier includes a fully connected layer (FC), with the last layer of the FC connecting four neurons and passing through a softmax function. The softmax function transforms a vector containing multiple values ​​into a vector representing the probability distributions. In this embodiment, it transforms the vector into the probability of each lung function state, and the sum of the probabilities of the four lung function states is 1.

[0103] Finally, the lung function state with the highest output probability is used as the lung function diagnosis result for the examinee, namely, normal output, obstructed, restricted, or mixed output.

[0104] The above approach addresses the issues of incomplete lung function information extraction caused by traditional methods that rely on human experience or make judgments based solely on local information, thus solving the problems in diagnosis, status monitoring, and evaluation of lung function impairment in the areas of chest and lung disease. It provides an artificial intelligence-based hybrid feature recognition method for lung function examination and diagnosis.

[0105] Furthermore, while overcoming the limitations of manual interpretation of lung function test images, artificial intelligence is used to achieve fusion diagnosis of multiple types of lung function test images. Based on information features, temporal features, and multi-channel image features, the final lung function diagnosis result is obtained. It integrates the spatial and temporal features of lung function data and combines the individual characteristics of the examinee to achieve accurate judgment of the examinee's lung function status.

[0106] This invention combines information features characterizing the health status of the examinee, temporal features characterizing the temporal characteristics of the examinee's pulmonary function test images, and multi-channel image features characterizing the spatial characteristics of the examinee's multi-channel images to obtain diagnostic results of the examinee's pulmonary function status. This improves the accuracy and consistency of diagnostic reports, reduces the error rate of human judgment, saves judgment time, and is conducive to promotion.

[0107] Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11 As shown, the electronic device may include: a processor 1110, a communications interface 1120, a memory 1130, and a communications bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other through the communications bus 1140. The processor 1110 can call logical instructions in the memory 830 to execute a lung function test diagnosis method based on artificial intelligence hybrid feature recognition. The method includes: acquiring personal information characterizing the health characteristics of the examinee and extracting information features; extracting features from multiple lung function test images of the examinee to obtain temporal features of each lung function test image; constructing a multi-channel lung function map based on the multiple lung function test images of the examinee, extracting features from the multi-channel lung function map to obtain multi-channel image features; and fusing the temporal features, information features, and multi-channel image features of each lung function test image and inputting them into a pre-trained classifier to obtain the lung function diagnosis result of the examinee output by the classifier as normal, obstructive, restrictive, or mixed.

[0108] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0109] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the artificial intelligence hybrid feature recognition lung function test diagnosis method provided by the above methods. The method includes: acquiring personal information characterizing the health characteristics of the examinee and extracting information features; extracting features from multiple lung function test images of the examinee to obtain temporal features of each lung function test image; constructing a multi-channel lung function map based on the multiple lung function test images of the examinee, extracting features from the multi-channel lung function map to obtain multi-channel image features; and fusing the temporal features, information features, and multi-channel image features of each lung function test image and inputting them into a pre-trained classifier to obtain the lung function diagnosis result of the examinee output by the classifier as normal, obstructive, restrictive, or mixed.

[0110] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a lung function test diagnosis method based on artificial intelligence hybrid feature recognition provided by the above methods. The method includes: acquiring personal information characterizing the health characteristics of a subject and extracting information features; extracting features from multiple lung function test images of the subject to obtain temporal features for each lung function test image; constructing a multi-channel lung function map based on the multiple lung function test images of the subject, extracting features from the multi-channel lung function map to obtain multi-channel image features; and fusing the temporal features, information features, and multi-channel image features of each lung function test image and inputting them into a pre-trained classifier to obtain the lung function diagnosis result of the subject output by the classifier as normal, obstructive, restrictive, or mixed.

[0111] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0112] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for pulmonary function testing and diagnosis using artificial intelligence-based hybrid feature recognition, characterized in that, The method comprises the following steps: acquiring personal information representing health characteristics of a subject and extracting information features; extracting features from multiple lung function test images of the subject respectively to obtain time sequence features of each of the lung function test images; constructing a multi-channel lung function graph based on the multiple lung function test images of the subject, and extracting features from the multi-channel lung function graph to obtain multi-channel image features; inputting the time sequence features of each of the lung function test images, the information features and the multi-channel image features into a pre-trained classifier after feature fusion to obtain a lung function diagnosis result of the subject output by the classifier, which is normal, obstruction, restriction or mixed. 2.The lung function test diagnosis method using artificial intelligence hybrid feature recognition according to claim 1, characterized in that, The lung function test images include slow vital capacity images, forced vital capacity images and maximum voluntary ventilation images. 3.The lung function test diagnosis method using artificial intelligence hybrid feature recognition according to claim 2, characterized in that, The step of extracting features from the multiple lung function test images of the subject respectively to obtain time sequence features of each of the lung function test images specifically comprises the following steps: inputting each of the lung function test images of the subject into a long short-term memory network to obtain an output time sequence of each of the lung function test images output by the long short-term memory network; using an attention mechanism to weight the output time sequence, and taking the weighted output time sequence as the time sequence features of the lung function test image. 4.The lung function test diagnosis method using artificial intelligence hybrid feature recognition according to claim 1, characterized in that, The step of constructing a multi-channel lung function graph based on the multiple lung function test images of the subject specifically comprises the following steps: cropping the multiple lung function test images of the subject to the same size; combining the cropped multiple lung function test images to obtain a multi-channel lung function graph. 5.The lung function test diagnosis method using artificial intelligence hybrid feature recognition according to claim 3, characterized in that, The step of extracting features from the multi-channel lung function graph to obtain multi-channel image features specifically comprises the following step: extracting features from the multi-channel lung function graph using a convolutional neural network to obtain multi-channel image features, wherein a set of convolution kernels in the convolutional neural network comprises a plurality of convolution kernels, and the number of the plurality of convolution kernels is the same as the number of the multiple lung function test images. 6.The lung function test diagnosis method using artificial intelligence hybrid feature recognition according to claim 5, characterized in that, Further comprising: designing a cross-entropy loss function, and considering that a lung function diagnosis model composed of the long short-term memory network, the convolutional neural network and the classifier is trained when the value of the cross-entropy loss function is less than a preset threshold.

7. An artificial intelligence hybrid feature recognition lung function test diagnosis device, characterized by, The method comprises the following steps: a personal information acquisition module is configured to acquire personal information representing health characteristics of a subject and extract information features; a time sequence feature acquisition module is configured to extract features from multiple lung function test images of the subject respectively to obtain time sequence features of each of the lung function test images; a spatial feature acquisition module is configured to construct a multi-channel lung function graph based on the multiple lung function test images of the subject, and extract features from the multi-channel lung function graph to obtain multi-channel image features; a diagnosis module is configured to input the time sequence features of each of the lung function test images, the information features and the multi-channel image features into a pre-trained classifier after feature fusion to obtain a lung function diagnosis result of the subject output by the classifier, which is normal, obstruction, restriction or mixed.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the artificial intelligence mixed feature recognition lung function test diagnosis method according to any one of claims 1 to 6 when executing the program. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the artificial intelligence mixed feature recognition lung function test diagnosis method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by a processor to implement the artificial intelligence mixed feature recognition lung function test diagnosis method according to any one of claims 1 to 6.

Images