Artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology

Abstract

The present application discloses an artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology, which uses artificial intelligence technology and meta-learning technology to analyze the gene sequence data of pathogen samples of different respiratory infectious diseases, and uses the semantic features of these gene sequences as prior information meta-features, and then measures the semantic interaction and matching between the sample gene sequence semantics of the suspected pathogen of the detected object and the meta-features, thereby achieving effective detection and identification of pathogens, thereby identifying the probability that the detected object suffers from respiratory infectious diseases. This method, with the guidance of meta-knowledge, improves the generalization ability of the model and its applicability in different regions and different populations, while improving the intelligence level of respiratory infectious disease screening to better assist in the screening of respiratory infectious diseases.

Description

Technical Field

[0001] The present application relates to the field of intelligent screening, and more specifically, to an artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology. Background Art

[0002] Respiratory infectious diseases are common diseases worldwide. They are usually caused by pathogens such as viruses and bacteria, posing a serious threat to human health. With the acceleration of globalization and the increase in population mobility, the speed and scope of the spread of infectious diseases are also expanding, which poses a huge challenge to the public health system. Therefore, rapid and accurate screening and identification of respiratory infectious disease pathogens are of great significance for controlling epidemics and reducing disease transmission.

[0003] However, traditional screening methods usually require professional medical staff to collect samples and conduct laboratory tests, which requires a lot of human resources and is inefficient. When an epidemic breaks out, there may be a shortage of human resources, resulting in insufficient screening capabilities. At the same time, the process of medical staff conducting laboratory analysis to screen and identify pathogens for patients' respiratory infectious diseases is less intelligent and is easily affected by human factors and operational experience. Not only that, the scope of application of traditional screening methods is limited by the distribution of professional laboratory facilities and professionals. It does not have sufficient flexibility and universality, and it is difficult to popularize in areas with limited resources. Due to the diversity and complexity of pathogens, traditional methods may find it difficult to accurately identify all types of pathogens, especially when faced with new or mutated pathogens.

[0004] Therefore, it is expected that artificial intelligence and meta-learning techniques can be used to assist in the screening of respiratory infectious diseases. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology, which uses artificial intelligence technology and meta-learning technology to analyze the gene sequence data of pathogenic samples of different respiratory infectious diseases, and uses the semantic features of these gene sequences as prior information meta-features, and then measures the semantic interaction and matching between the sample gene sequence semantics of the suspected pathogen of the detected object and the meta-feature, thereby achieving effective detection and identification of pathogens, thereby identifying the probability that the detected object suffers from respiratory infectious diseases. This method, with the guidance of meta-knowledge, improves the generalization ability of the model and its applicability in different regions and different populations, while improving the intelligence level of screening for respiratory infectious diseases, so as to better assist in the screening of respiratory infectious diseases.

[0006] According to one aspect of the present application, a respiratory infectious disease artificial intelligence-assisted screening system based on meta-learning technology is provided, which includes:

[0007] A pathogen sample gene sequence collection module is used to obtain the gene sequences of pathogen samples of various respiratory infectious diseases to obtain a collection of gene sequences of pathogen samples of respiratory infectious diseases;

[0008] A meta-feature semantic learning module between gene sequences of pathogenic samples, used for performing meta-feature semantic learning between gene sequences on each gene sequence of pathogenic samples of respiratory infectious diseases in the set of gene sequences of pathogenic samples of respiratory infectious diseases to obtain a set of meta-feature vectors of gene sequences of pathogenic samples of respiratory infectious diseases;

[0009] A suspected pathogen sample gene sequence collection module is used to obtain the sample gene sequence of the suspected pathogen of the detected object;

[0010] A suspected pathogen sample gene sequence semantic coding module is used to perform semantic coding processing on the sample gene sequence of the suspected pathogen of the detected object to obtain a semantic coding feature vector of the suspected pathogen sample gene sequence;

[0011] A gene sequence semantic one-way matching query module is used to use the semantic encoding feature vector of the gene sequence of the suspected pathogen sample as a query feature vector, and input the query feature vector and the set of the gene sequence meta-feature vector of the respiratory infectious disease pathogen sample into a one-way attention granular scanning interactive matching network to obtain a one-way matching result representation vector;

[0012] The respiratory infectious disease detection module is used to determine the probability value of the detected object suffering from a respiratory infectious disease based on the one-way matching result representation vector.

[0013] Compared with the prior art, the present application provides an artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology. It uses artificial intelligence technology and meta-learning technology to analyze the gene sequence data of pathogen samples of different respiratory infectious diseases, and uses the semantic features of these gene sequences as prior information meta-features, and then measures the semantic interaction and matching between the sample gene sequence semantics of the suspected pathogen of the detected object and the meta-features, thereby achieving effective detection and identification of pathogens, thereby identifying the probability that the detected object suffers from respiratory infectious diseases. This method, with the guidance of meta-knowledge, improves the generalization ability of the model and its applicability in different regions and different populations, while improving the intelligence level of respiratory infectious disease screening to better assist in the screening of respiratory infectious diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0015] Figure 1 A block diagram of an artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to an embodiment of the present application;

[0016] Figure 2 This is a system architecture diagram of a respiratory infectious disease artificial intelligence-assisted screening system based on meta-learning technology according to an embodiment of the present application;

[0017] Figure 3 A block diagram of the training phase of the respiratory infectious disease artificial intelligence-assisted screening system based on meta-learning technology according to an embodiment of the present application;

[0018] Figure 4 This is a block diagram of a gene sequence semantic one-way matching query module in an artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0020] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0021] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0022] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] Meta-learning is a machine learning paradigm that aims to enable models to quickly adapt to new tasks, even if there is only a small amount of labeled data for these tasks. The core idea of ​​meta-learning is to enable the model to learn how to quickly learn from a small amount of data by training it on a series of related tasks. Meta-learning technology is a machine learning method that uses prior knowledge to quickly learn new tasks. In the screening system for respiratory infectious diseases, by combining meta-learning technology with artificial intelligence technology, the genetic sequence semantics of pathogen samples can be quickly and effectively learned to assist in the screening of pathogens and realize intelligent detection of respiratory infectious diseases.

[0025] In the technical solution of the present application, an artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology is proposed. Figure 1 This is a block diagram of an artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to an embodiment of the present application. Figure 2 : is a system architecture diagram of a respiratory infectious disease artificial intelligence-assisted screening system based on meta-learning technology according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to an embodiment of the present application, an artificial intelligence-assisted screening system 300 for respiratory infectious diseases based on meta-learning technology includes: a pathogen sample gene sequence acquisition module 310, which is used to obtain pathogen sample gene sequences of multiple respiratory infectious diseases to obtain a set of respiratory infectious disease pathogen sample gene sequences; a pathogen sample gene sequence meta-feature semantic learning module 320, which is used to perform meta-feature semantic learning between gene sequences on each respiratory infectious disease pathogen sample gene sequence in the set of respiratory infectious disease pathogen sample gene sequences to obtain a set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors; a suspected pathogen sample gene sequence acquisition module 330, which is used to obtain sample meta-feature vectors of suspected pathogens of the detected object; due to sequence; a suspected pathogen sample gene sequence semantic encoding module 340, which is used to perform semantic encoding processing on the sample gene sequence of the suspected pathogen of the detected object to obtain a semantic encoding feature vector of the suspected pathogen sample gene sequence; a gene sequence semantic one-way matching query module 350, which is used to use the semantic encoding feature vector of the suspected pathogen sample gene sequence as a query feature vector, and input the query feature vector and the set of the respiratory infectious disease pathogen sample gene sequence meta-feature vector into a one-way attention granular scanning interactive matching network to obtain a one-way matching result representation vector; a respiratory infectious disease detection module 360, which is used to determine the probability value of the detected object suffering from a respiratory infectious disease based on the one-way matching result representation vector.

[0026] In particular, the pathogen sample gene sequence acquisition module 310 and the pathogen sample gene sequence meta-feature semantic learning module 320 are used to obtain pathogen sample gene sequences of multiple respiratory infectious diseases to obtain a set of respiratory infectious disease pathogen sample gene sequences; and perform meta-feature semantic learning based on gene sequence meta-feature semantics on each respiratory infectious disease pathogen sample gene sequence in the set of respiratory infectious disease pathogen sample gene sequences to obtain a set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors. In particular, in a specific example of the present application, each respiratory infectious disease pathogen sample gene sequence in the set of respiratory infectious disease pathogen sample gene sequences is respectively input into a gene sequence meta-feature semantic learner containing a gene embedding encoder to obtain a set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors. It should be understood that, since each respiratory infectious disease pathogen sample gene sequence in the set of respiratory infectious disease pathogen sample gene sequences contains a specific respiratory infectious disease pathogen sample gene sequence semantic feature, this gene sequence semantic feature can reflect the essential attributes and biological significance of the gene sequence, thereby helping to effectively identify the gene similarity and difference characteristics between pathogen samples of different respiratory infectious diseases, thereby assisting in the subsequent detection of pathogen samples and respiratory infectious diseases. Based on this, in the technical solution of the present application, each respiratory infectious disease pathogen sample gene sequence in the set of respiratory infectious disease pathogen sample gene sequences is respectively input into a gene sequence inter-feature semantic learner containing a gene embedding encoder to obtain a set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors. In particular, here, the gene embedding encoder is similar to the word embedding technology in natural language processing, which can convert the gene sequence into a vector representation in a high-dimensional space, which helps the subsequent analysis and learning process. Then, through the processing of the meta-feature semantic learner between gene sequences, the contextual encoding module between gene sequences can be used to extract and learn the contextual semantic association feature information between the embedded coding features of each gene in the gene sequence of each respiratory infectious disease pathogen sample, and use this information as a priori information meta-feature to guide and optimize the subsequent screening and identification process, providing strong support for subsequent gene sequence semantic matching and pathogen identification.

[0027] In particular, the suspected pathogen sample gene sequence acquisition module 330 and the suspected pathogen sample gene sequence semantic encoding module 340 are used to obtain the sample gene sequence of the suspected pathogen of the detected object; and the sample gene sequence of the suspected pathogen of the detected object is semantically encoded to obtain the semantic encoding feature vector of the suspected pathogen sample gene sequence. In particular, in a specific example of the present application, the sample gene sequence of the suspected pathogen of the detected object is input into a gene sequence semantic encoder containing a gene embedding encoder to obtain the semantic encoding feature vector of the suspected pathogen sample gene sequence. It should be understood that the original sample gene sequence data of the suspected pathogen can be embedded and encoded into a vector form through the gene embedding encoder, and this conversion makes the data more suitable for subsequent machine learning model processing. In addition, through the processing of the gene sequence semantic encoder, the implicit association relationship between the semantic embedding encoding features of each gene in the sample gene sequence of the suspected pathogen can be learned and captured, thereby reflecting the semantic features of the sample gene sequence of the suspected pathogen, which includes not only the direct biological information of the gene sequence, but also may include the semantic information of the sequence pattern, variation, functional region, etc., which is helpful for the subsequent matching and recognition process.

[0028] In particular, the gene sequence semantic one-way matching query module 350 is used to use the suspected pathogen sample gene sequence semantic encoding feature vector as the query feature vector, and input the query feature vector and the set of the respiratory infectious disease pathogen sample gene sequence meta-feature vector into the one-way attention granular scanning interactive matching network to obtain a one-way matching result representation vector. Since the suspected pathogen sample gene sequence semantic encoding feature vector contains the sample gene sequence semantic features of the suspected pathogen of the detected object, each respiratory infectious disease pathogen sample gene sequence meta-feature vector in the set of respiratory infectious disease pathogen sample gene sequence meta-feature vector contains gene sequence meta-feature semantic information about various pathogen samples. Therefore, in order to be able to detect whether the detected object is suffering from a respiratory infectious disease, the sample gene sequence semantic features of the suspected pathogen of the detected object need to be used as a query feature to scan and query the gene sequence semantic features in the gene sequence meta-feature semantic information of each respiratory infectious disease pathogen sample, so as to interactively match the one-way matching results between the suspected pathogen sample gene sequence semantics and the various pathogen sample gene sequence semantics, so as to calculate the probability that the detected object suffers from a respiratory infectious disease. Based on this, in the technical solution of the present application, the semantic coding feature vector of the gene sequence of the suspected pathogen sample is further used as the query feature vector, and the set of the query feature vector and the gene sequence meta-feature vector of the respiratory infectious disease pathogen sample is input into the one-way attention granular scanning interactive matching network to obtain the one-way matching result representation vector. Through the processing of the one-way attention granular scanning interactive matching network, the correlation relationship and one-way interactive matching information between the semantic coding query feature of the gene sequence of the suspected pathogen sample and the meta-features of the gene sequence of various respiratory infectious disease pathogen samples can be automatically learned and captured, and the one-way interactive matching information is used as the weight to optimize the expression of the meta-feature vector of each respiratory infectious disease pathogen sample gene sequence. In this way, it is possible to identify which meta-features of the gene sequence of the respiratory infectious disease pathogen sample are related to the subsequent matching and recognition tasks, so as to give higher attention to the gene sequence meta-features with higher correlation and contribution in the process of fusion of these meta-features, and give lower weights to the meta-features with less contribution to the subsequent tasks. Then, the difference between the query feature vector and the semantic fusion information of these gene sequence meta-features is used to effectively determine whether the gene sequence semantics of the suspected pathogen sample is a pathogen, thereby determining whether the patient suffers from a respiratory infectious disease. In particular, in a specific example of the present application, Figure 4As shown, the gene sequence semantic one-way matching query module 350 includes: a semantic similarity measurement unit 351, which is used to calculate the semantic similarity between the query feature vector and each respiratory infectious disease pathogen sample gene sequence meta-feature vector in the set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors to obtain a sequence of gene sequence meta-feature one-way matching semantic measurement coefficients; a normalization unit 352, which is used to normalize the sequence of gene sequence meta-feature one-way matching semantic measurement coefficients to obtain a sequence of gene sequence meta-feature one-way matching semantic measurement weight values; a gene sequence meta-feature one-way matching interactive reshaping unit 353, which is used to calculate the weighted sum of the set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors using the sequence of gene sequence meta-feature one-way matching semantic measurement weight values ​​as a weight to obtain a gene sequence meta-feature one-way matching interactive reshaping feature vector; a differential calculation unit 354, which is used to calculate the differential feature vector between the gene sequence meta-feature one-way matching interactive reshaping feature vector and the query feature vector as the one-way matching result representation vector.

[0029] Specifically, the semantic similarity measurement unit 351 is used to calculate the semantic similarity between the query feature vector and each respiratory infectious disease pathogen sample gene sequence meta-feature vector in the set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors to obtain a sequence of gene sequence meta-feature one-way matching semantic measurement coefficients. In a specific example of the present application, the query feature vector and each respiratory infectious disease pathogen sample gene sequence meta-feature vector in the set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors are first cascaded to obtain a set of query gene sequence meta-feature semantic encoding joint feature vectors; then, after calculating the matrix product of each query gene sequence meta-feature semantic encoding joint feature vector in the set of query gene sequence meta-feature semantic encoding joint feature vectors and the weight matrix, the obtained feature vector is added to the bias vector by position to obtain a sequence of query gene sequence meta-feature semantic encoding joint bias feature vectors; and then the sequence of query gene sequence meta-feature semantic encoding joint bias feature vectors is input into the sigmoid activation function to obtain a sequence of gene sequence meta-feature one-way matching semantic measurement coefficients.

[0030] Specifically, the normalization unit 352 is used to normalize the sequence of the one-way matching semantic metric coefficients of the gene sequence meta-features to obtain a sequence of one-way matching semantic metric weight values ​​of the gene sequence meta-features. In a specific example of the present application, each one-way matching semantic metric coefficient of the gene sequence meta-features in the sequence of the one-way matching semantic metric coefficients of the gene sequence meta-features is first used as an exponent of a natural constant to calculate an exponential function value with the natural constant as the base to obtain a sequence of logarithmic values ​​of the one-way matching semantic metric coefficients of the gene sequence meta-features; then, the sum of the logarithmic values ​​of the one-way matching semantic metric coefficients of each gene sequence meta-feature in the sequence of the logarithmic values ​​of the one-way matching semantic metric coefficients of the gene sequence meta-features is calculated to obtain the total sum of the one-way matching semantic metric of the gene sequence meta-features; and then, the logarithmic values ​​of each one-way matching semantic metric coefficient of the gene sequence meta-features in the sequence of the logarithmic values ​​of the one-way matching semantic metric coefficients of the gene sequence meta-features are divided by the total sum of the one-way matching semantic metric of the gene sequence meta-features to obtain a sequence of one-way matching semantic metric weight values ​​of the gene sequence meta-features.

[0031] Specifically, the gene sequence meta-feature one-way matching interactive reshaping unit 353 is used to calculate the weighted sum of the set of the respiratory infectious disease pathogen sample gene sequence meta-feature vectors using the sequence of the gene sequence meta-feature one-way matching semantic metric weight values ​​as weights to obtain the gene sequence meta-feature one-way matching interactive reshaping feature vector. In a specific example of the present application, each gene sequence meta-feature one-way matching semantic metric weight value in the sequence of the gene sequence meta-feature one-way matching semantic metric weight values ​​is first multiplied by position with the corresponding respiratory infectious disease pathogen sample gene sequence meta-feature vector in the set of the respiratory infectious disease pathogen sample gene sequence meta-feature vectors to obtain a set of weight-optimized respiratory infectious disease pathogen sample gene sequence meta-feature vectors; then the position-wise sum of the set of weight-optimized respiratory infectious disease pathogen sample gene sequence meta-feature vectors is calculated to obtain the gene sequence meta-feature one-way matching interactive reshaping feature vector.

[0032] Specifically, the difference calculation unit 354 is used to calculate the difference feature vector between the gene sequence meta-feature one-way matching interactive reshaping feature vector and the query feature vector as the one-way matching result representation vector. It should be understood that by calculating the difference feature vector between the gene sequence meta-feature one-way matching interactive reshaping feature vector and the query feature vector, the similarity or difference between the gene sequence meta-feature and the query feature can be obtained. This helps to determine the matching degree of the gene sequence element in the query sequence.

[0033] In summary, in the above embodiment, the semantic encoding feature vector of the gene sequence of the suspected pathogen sample is used as the query feature vector, and the query feature vector and the set of the meta-feature vector of the gene sequence of the respiratory infectious disease pathogen sample are input into the one-way attention granular scanning interactive matching network to obtain a one-way matching result representation vector, including: using the semantic encoding feature vector of the gene sequence of the suspected pathogen sample as the query feature vector, and inputting the query feature vector and the set of the meta-feature vector of the gene sequence of the respiratory infectious disease pathogen sample into the one-way attention granular scanning interactive matching network to process the query feature vector and the set of the meta-feature vector of the gene sequence of the respiratory infectious disease pathogen sample into the one-way attention granular scanning interactive matching network to obtain a one-way matching result representation vector; wherein, the granular scanning interactive matching formula is:

[0034]

[0035]

[0036]

[0037]

[0038] in, and The query feature vector and the respiratory infectious disease pathogen sample gene sequence element feature vector are respectively The gene sequence element feature vector of respiratory infectious disease pathogen samples, It is cascade processing. is the first in the sequence of one-way matching semantic metric coefficients of gene sequence meta-features Gene sequence meta-feature one-way matching semantic metric coefficient, is the weight matrix, is the bias vector, is the sigmoid function, Indicated by natural constant The exponential function with base , is the number of feature vectors in the set of feature vectors of gene sequence elements of respiratory infectious disease pathogen samples, is the first in the sequence of the semantic metric weight value of the one-way matching of the gene sequence meta-feature The weight value of the semantic metric for the one-way matching of gene sequence meta-features, It is the gene sequence meta-feature unidirectional matching interactive reshaping feature vector, It means subtracting by position. is the one-way matching result representation vector.

[0039] It is worth mentioning that in other specific examples of the present application, the semantic encoding feature vector of the gene sequence of the suspected pathogen sample can also be used as the query feature vector in other ways, and the query feature vector and the set of the gene sequence meta-feature vector of the respiratory infectious disease pathogen sample are input into the one-way attention granular scanning interactive matching network to obtain a one-way matching result representation vector, for example: input the semantic encoding feature vector of the gene sequence of the suspected pathogen sample and the set of the gene sequence meta-feature vector of the respiratory infectious disease pathogen sample; input the query feature vector and the set of the gene sequence meta-feature vector of the respiratory infectious disease pathogen sample into the one-way attention granular scanning interactive matching network; the one-way attention granular scanning interactive matching network will process each respiratory infectious disease pathogen sample gene sequence meta-feature vector one by one, and interactively match it with the query feature vector to evaluate the similarity or correlation between them; scan each respiratory infectious disease pathogen sample gene sequence meta-feature vector on a granular basis and match it with the query feature vector; in the process of granular scanning, generate the one-way matching result representation vector, which is used to represent the matching result between the query feature vector and the gene sequence meta-feature vector of the respiratory infectious disease pathogen sample.

[0040] In particular, the respiratory infectious disease detection module 360 ​​is used to determine the probability value of the detected object suffering from a respiratory infectious disease based on the one-way matching result representation vector. In particular, in a specific example of the present application, the one-way matching result representation vector is passed through a classifier-based respiratory infectious disease identifier to obtain a recognition result, and the recognition result is used to represent the probability value of the detected object suffering from a respiratory infectious disease. That is, the one-way matching features between the gene sequence semantics of the suspected pathogen sample and the meta-feature aggregation semantics of the pathogen sample gene sequence of various respiratory infectious diseases are used to identify respiratory infectious diseases. In this way, effective detection and identification of pathogens are achieved, thereby identifying the probability of the detected object suffering from a respiratory infectious disease. This method, with the guidance of meta-knowledge, improves the generalization ability of the model and its applicability in different regions and different populations, while improving the intelligence of respiratory infectious disease screening to better assist in the screening of respiratory infectious diseases.

[0041] Here, considering that the set of the semantic coding feature vectors of the suspected pathogen sample gene sequence and the set of the respiratory infectious disease pathogen sample gene sequence meta-feature vectors respectively represent the gene embedding coding semantic features and gene embedding coding sequence meta-features of the set of sample gene sequences of the suspected pathogen and the respiratory infectious disease pathogen sample gene sequences of the detected object, in this way, after the set of the semantic coding feature vectors of the suspected pathogen sample gene sequence and the set of the respiratory infectious disease pathogen sample gene sequence meta-feature vectors are input into the one-way attention granular scanning interactive matching network, the one-way matching result representation vector obtained will also have complex semantic feature interactions based on the one-way attention semantic scanning interactive matching under different gene semantic coding dimensions, thereby affecting the overall expression effect.

[0042] Based on this, in a preferred embodiment of the present application, the one-way matching result representation vector is passed through a classifier-based respiratory infectious disease identifier to obtain a recognition result, including: optimizing the one-way matching result representation vector to obtain an optimized one-way matching result representation vector; passing the optimized one-way matching result representation vector through a classifier-based respiratory infectious disease identifier to obtain a recognition result.

[0043] Specifically, optimizing the one-way matching result representation vector to obtain an optimized one-way matching result representation vector includes the following steps:

[0044] Calculate the mean and variance of all eigenvalues ​​of the one-way matching result representation vector, perform dot addition of the one-way matching result representation vector with the mean and the variance respectively, take the inverse of each eigenvalue and further perform dot multiplication with the one-way matching result representation vector to obtain a first intermediate eigenvector and a second intermediate eigenvector, then transpose the one-way matching result representation vector into a column vector and perform matrix multiplication with the first intermediate eigenvector as a row vector to obtain a first intermediate feature matrix, and transpose the second intermediate eigenvector into a column vector and perform matrix multiplication with the one-way matching result representation vector to obtain a second intermediate feature matrix, and finally perform matrix multiplication of the first intermediate feature matrix, the second intermediate feature matrix and the transpose of the one-way matching result representation vector to obtain an optimized one-way matching result representation vector.

[0045] The one-way matching result representation vector is represented as , then the optimization process is expressed as:

[0046]

[0047]

[0048]

[0049] in, represents the one-way matching result representation vector, and represent the mean and variance of all eigenvalues ​​of the one-way matching result representation vector respectively, It means adding by position. It means point multiplication by position. It means to calculate the inverse of each eigenvalue of the vector. and are the first intermediate eigenvector and the second intermediate eigenvector respectively, represents matrix multiplication, represents the transpose of a vector, The optimized one-way matching result representation vector.

[0050] That is, in order to address the collapse of the tail feature distribution caused by the loss of tail feature information in the interactive feature distribution ecology of the one-way matching result representation vector, the descriptive ecological discovery of the feature set of the one-way matching result representation vector is achieved through the stable diffusion of the distribution probability of interaction based on the global statistical characteristics of the one-way matching result representation vector, thereby forming a hybrid intuitive feature ecology through variational autoencoding based on the stable diffusion probability distribution, so as to enhance the universality of the acquisition of the real interactive value of the one-way matching result representation vector as an interactive synthetic feature, improve the overall expression effect of the one-way matching result representation vector, and enhance the accuracy of the classification results obtained by the one-way matching result representation vector.

[0051] It should be understood that before using the above-mentioned neural network model for inference, it is necessary to train the gene sequence meta-feature semantic learner including the gene embedding encoder, the gene sequence semantic encoder including the gene embedding encoder, the unidirectional attention granular scanning interactive matching network and the classifier-based respiratory infectious disease identifier. That is to say, according to the respiratory infectious disease artificial intelligence-assisted screening system 300 based on meta-learning technology of the present application, it also includes a training phase for training the gene sequence meta-feature semantic learner including the gene embedding encoder, the gene sequence semantic encoder including the gene embedding encoder, the unidirectional attention granular scanning interactive matching network and the classifier-based respiratory infectious disease identifier.

[0052] Figure 3 FIG. 1 is a block diagram of the training phase of the respiratory infectious disease artificial intelligence-assisted screening system based on meta-learning technology according to an embodiment of the present application. Figure 3As shown, according to the embodiment of the present application, the artificial intelligence-assisted screening system 300 for respiratory infectious diseases based on meta-learning technology includes: a training module 400, including: a training data acquisition unit 410, used to acquire training data, the training data including gene sequences of pathogen samples of multiple respiratory infectious diseases to obtain a set of training respiratory infectious disease pathogen sample gene sequences; and training sample gene sequences of suspected pathogens of the detected object; an inter-sequence meta-feature semantic learning unit 420, used to input each training respiratory infectious disease pathogen sample gene sequence in the set of training respiratory infectious disease pathogen sample gene sequences into a gene sequence inter-feature semantic learner containing a gene embedding encoder to obtain a set of training respiratory infectious disease pathogen sample gene sequence meta-feature vectors; a sequence semantic encoding unit 430, used to perform semantic encoding processing on the training sample gene sequence of the suspected pathogen of the detected object to obtain a semantic encoding feature vector of the training suspected pathogen sample gene sequence; a sequence semantic one-way matching query unit 440, used to obtain a set of training respiratory infectious disease pathogen sample gene sequence meta-feature vectors; The semantic encoding feature vector of the training suspected pathogen sample gene sequence is used as a query feature vector, and the query feature vector and the set of the training respiratory infectious disease pathogen sample gene sequence meta-feature vector are input into a one-way attention granular scanning interactive matching network to obtain a training one-way matching result representation vector; a classification loss unit 450 is used to pass the training one-way matching result representation vector through a classifier-based respiratory infectious disease identifier to obtain a classification loss function value; a one-way matching result loss unit 460 is used to calculate the one-way matching result loss function value; a loss calculation unit 470 is used to calculate the weighted sum of the classification loss function value and the one-way matching result loss function value to obtain a loss function value; a training unit 480 is used to train the gene sequence meta-feature semantic learner containing a gene embedding encoder, the gene sequence semantic encoder containing a gene embedding encoder, the one-way attention granular scanning interactive matching network and the classifier-based respiratory infectious disease identifier based on the loss function value through gradient back propagation.

[0053] In particular, here, considering that the set of semantic coding feature vectors of the training suspected pathogen sample gene sequence and the set of meta-feature vectors of the training respiratory infectious disease pathogen sample gene sequence respectively represent the gene embedding coding semantic features and gene embedding coding sequence meta-features of the set of training sample gene sequences of suspected pathogens of the detected object and training respiratory infectious disease pathogen sample gene sequences, in this way, after the set of semantic coding feature vectors of the training suspected pathogen sample gene sequence and the set of meta-feature vectors of the training respiratory infectious disease pathogen sample gene sequence are input into the one-way attention granular scanning interactive matching network, the obtained training one-way matching result representation vector will also have class regression recognition difficulties based on complex semantic feature interaction based on the one-way attention semantic scanning interactive matching under different gene semantic coding dimensions, thereby affecting the classification training efficiency.

[0054] Therefore, the applicant of the present application further introduces a predetermined loss function other than the classification loss function during the model training process, that is, training the model through gradient back propagation based on the loss function includes the following steps:

[0055] Calculate a training first one-way matching result weight matrix and a training second one-way matching result weight matrix based on the training one-way matching result representation vector, wherein the first training one-way matching result weight matrix The eigenvalues ​​of the positions are the first Eigenvalues ​​and The mean of the eigenvalues, and the training second one-way matching result weight matrix The eigenvalues ​​of the positions are the first Eigenvalues ​​and The difference of the eigenvalues ​​is taken as the absolute value and divided by two;

[0056] Performing query matrix multiplication of the training one-way matching result representation vector with the training first one-way matching result weight matrix and the training second one-way matching result weight matrix to obtain a training first one-way matching result intermediate vector and a training second one-way matching result intermediate vector;

[0057] Calculate the vector inner product of the intermediate vector of the first one-way matching result of the training and the intermediate vector of the second one-way matching result of the training to obtain a first one-way matching result loss term;

[0058] The first one-way matching result weight matrix and the training second one-way matching result weight matrix are matrix-multiplied, and the result matrix is ​​calculated. norm to obtain the second one-way matching result loss term;

[0059] Subtracting the product of a predetermined weight hyperparameter and the second one-way matching result loss term from the first one-way matching result loss term to obtain a one-way matching result loss function value; and

[0060] Based on the weighted sum of the one-way matching result loss function value and the classification loss function value, the model parameters are optimized through gradient back propagation.

[0061] The one-way matching result loss function is specifically expressed as:

[0062]

[0063]

[0064] in, is the training one-way matching result representation vector, and The first one-way matching result weight matrix and the second one-way matching result weight matrix are trained respectively. and are respectively the first training one-way matching result weight matrix and the second training one-way matching result weight matrix The eigenvalues ​​of the positions, and are the first Eigenvalues ​​and Eigenvalues, is matrix multiplication, To calculate the matrix norm, is the predetermined weight hyperparameter, is the loss function value of the one-way matching result.

[0065] That is, the one-way matching result loss function uses the structured feature representation of the short-range-long-range cross-scale detail link of the one-way matching result representation vector to perform the query composition of the detail inner product space in the training one-way matching result representation vector, so as to approximate the low-rank independent observable composition of the link detail composition provided by the structured detail interaction of the training one-way matching result representation vector. In this way, by training with the one-way matching result loss function, the distributed detail group of the training one-way matching result representation vector can be used to decompose the detail group based on the complexity of the details, so as to promote the class regression decomposition recognition of the complex feature structure of the training one-way matching result representation vector and improve the classification training efficiency. In this way, pathogens can be detected and identified more effectively, thereby assisting the screening of respiratory infectious diseases and improving the intelligence level of respiratory infectious disease screening.

[0066] As described above, the respiratory infectious disease artificial intelligence-assisted screening system 300 based on meta-learning technology according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with an artificial intelligence-assisted screening algorithm for respiratory infectious diseases based on meta-learning technology. In one possible implementation, the respiratory infectious disease artificial intelligence-assisted screening system 300 based on meta-learning technology according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the respiratory infectious disease artificial intelligence-assisted screening system 300 based on meta-learning technology can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the respiratory infectious disease artificial intelligence-assisted screening system 300 based on meta-learning technology can also be one of the many hardware modules of the wireless terminal.

[0067] Alternatively, in another example, the meta-learning technology-based artificial intelligence-assisted screening system 300 for respiratory infectious diseases and the wireless terminal may also be separate devices, and the meta-learning technology-based artificial intelligence-assisted screening system 300 for respiratory infectious diseases may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0068] This application breaks the construction concept and model of large-sample training of traditional medical AI. By calculating the loss function between the standard labeled samples and the target sample features, the classification probability of the target sample is determined. This method is suitable for the AI ​​construction of statutory diseases (statutory infectious diseases, statutory occupational diseases) and rare diseases with standard examples. It is no longer necessary to use expert consultation as an alternative endpoint to label a large number of samples. The Transformer algorithm in natural language process research is innovatively applied to clinical multimodal fusion. On the basis of enriching clinical features to improve robustness (fuzzy recognition ability), it solves the problem of low tolerance of neural networks, including twin neural networks, for missing input sample modalities. For the learning of different sudden unexplained respiratory infectious diseases, the model is given the ability to converge through multimodal small samples to quickly respond to the identification of unknown pathogenic viruses and avoid large-scale medical runs. At the same time, the model can also calculate the loss function between the clinical features caused by different viruses to assist in the preliminary classification of unknown pathogenic virus types.

[0069] In a specific example, transfer learning is a scenario where an architecture optimized on a similar domain dataset can be used to learn low-data system goals. It uses a trained neural network for generalization and then uses the current dataset to improve these hyperparameters. In this study, we intend to use the constructed COVID-19 multimodal twin network for transfer learning to achieve the expansion of the algorithm for other sudden unexplained respiratory infectious disease screening / diagnosis scenarios. Specifically, it is planned to be implemented in two steps: ① Focusing on the identification of different coronaviruses, using the multimodal twin network constructed with COVID-19 as the basis, learning and optimizing the model-related hyperparameters for coronaviruses such as SARS and MARS as the validation set to obtain a high-precision but narrowly applied sudden new coronavirus infectious disease screening / diagnosis model; ② On this basis, the data of patients infected with non-coronaviruses such as H1N1, H9N7, and influenza A virus are successively used for retraining and verification, and the hyperparameters are improved to obtain a generally applicable but relatively low-precision sudden unexplained respiratory infectious disease screening / diagnosis model. In addition, we summarize the hyperparameter settings of AI models for various types of viruses, use the clinical characteristic patterns of various sudden unexplained respiratory infectious diseases as "data beacons", compare the relationship and differences between hyperparameters, compare the loss functions and different / same characteristics between the same / different types of viruses, draw the virus cluster classification network structure based on clinical characteristics, and locate the clinical characteristics in the network to determine the virus type and species. This will provide reliable data evidence and effective auxiliary technical means for the rapid judgment of sudden unexplained respiratory infectious diseases in the future.

[0070] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology, characterized in that: include: A pathogen sample gene sequence collection module is used to obtain the gene sequences of pathogen samples of various respiratory infectious diseases to obtain a collection of gene sequences of pathogen samples of respiratory infectious diseases; A meta-feature semantic learning module between gene sequences of pathogenic samples, used for performing meta-feature semantic learning between gene sequences on each gene sequence of pathogenic samples of respiratory infectious diseases in the set of gene sequences of pathogenic samples of respiratory infectious diseases to obtain a set of meta-feature vectors of gene sequences of pathogenic samples of respiratory infectious diseases; A suspected pathogen sample gene sequence collection module is used to obtain the sample gene sequence of the suspected pathogen of the detected object; A suspected pathogen sample gene sequence semantic coding module is used to perform semantic coding processing on the sample gene sequence of the suspected pathogen of the detected object to obtain a semantic coding feature vector of the suspected pathogen sample gene sequence; A gene sequence semantic one-way matching query module is used to use the semantic encoding feature vector of the gene sequence of the suspected pathogen sample as a query feature vector, and input the query feature vector and the set of the gene sequence meta-feature vector of the respiratory infectious disease pathogen sample into a one-way attention granular scanning interactive matching network to obtain a one-way matching result representation vector; A respiratory infectious disease detection module, used to determine the probability value of the detected object suffering from a respiratory infectious disease based on the one-way matching result representation vector; The respiratory infectious disease detection module is used to: pass the one-way matching result representation vector through a classifier-based respiratory infectious disease identifier to obtain a recognition result, wherein the recognition result is used to represent the probability value of the detected object suffering from a respiratory infectious disease; The one-way matching result representation vector is passed through a classifier-based respiratory infectious disease identifier to obtain a recognition result, comprising: optimizing the one-way matching result representation vector to obtain an optimized one-way matching result representation vector; passing the optimized one-way matching result representation vector through a classifier-based respiratory infectious disease identifier to obtain a recognition result; The step of optimizing the one-way matching result representation vector to obtain an optimized one-way matching result representation vector comprises the following steps: Calculate the mean and variance of all eigenvalues ​​of the one-way matching result representation vector, perform dot addition of the one-way matching result representation vector with the mean and the variance respectively, take the inverse of each eigenvalue and further perform dot multiplication with the one-way matching result representation vector to obtain a first intermediate eigenvector and a second intermediate eigenvector, then transpose the one-way matching result representation vector into a column vector and perform matrix multiplication with the first intermediate eigenvector as a row vector to obtain a first intermediate feature matrix, and transpose the second intermediate eigenvector into a column vector and perform matrix multiplication with the one-way matching result representation vector to obtain a second intermediate feature matrix, and finally perform matrix multiplication of the first intermediate feature matrix, the second intermediate feature matrix and the transpose of the one-way matching result representation vector to obtain an optimized one-way matching result representation vector; The one-way matching result representation vector is represented as , then the optimization process is expressed as: in, represents the one-way matching result representation vector, and represent the mean and variance of all eigenvalues ​​of the one-way matching result representation vector respectively, It means adding by position. It means point multiplication by position. It means to calculate the inverse of each eigenvalue of the vector. and are the first intermediate eigenvector and the second intermediate eigenvector respectively, represents matrix multiplication, represents the transpose of a vector, The optimized one-way matching result representation vector.

2. The artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to claim 1 is characterized in that: The pathogen sample gene sequence meta-feature semantic learning module is used to: input each respiratory infectious disease pathogen sample gene sequence in the set of respiratory infectious disease pathogen sample gene sequences into a gene sequence meta-feature semantic learner including a gene embedding encoder to obtain a set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors.

3. The artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to claim 2 is characterized in that: The suspected pathogen sample gene sequence semantic encoding module is used to: input the sample gene sequence of the suspected pathogen of the detected object into a gene sequence semantic encoder including a gene embedding encoder to obtain the suspected pathogen sample gene sequence semantic encoding feature vector.

4. The artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to claim 3 is characterized in that: The gene sequence semantic one-way matching query module includes: A semantic similarity measurement unit, used for calculating the semantic similarity between the query feature vector and each respiratory infectious disease pathogen sample gene sequence meta-feature vector in the set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors to obtain a sequence of gene sequence meta-feature unidirectional matching semantic measurement coefficients; A normalization unit, used for normalizing the sequence of gene sequence meta-feature one-way matching semantic measurement coefficients to obtain a sequence of gene sequence meta-feature one-way matching semantic measurement weight values; A gene sequence meta-feature one-way matching interactive reshaping unit is used to calculate the weighted sum of the set of gene sequence meta-feature vectors of the respiratory infectious disease pathogen sample using the sequence of the gene sequence meta-feature one-way matching semantic measurement weight values ​​as weights to obtain a gene sequence meta-feature one-way matching interactive reshaping feature vector; The differential calculation unit is used to calculate the differential feature vector between the gene sequence meta-feature one-way matching interactive reshaping feature vector and the query feature vector as the one-way matching result representation vector.

5. The artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to claim 4 is characterized in that: The semantic similarity measurement unit is used to: Cascading the query feature vector and each respiratory infectious disease pathogen sample gene sequence meta-feature vector in the set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors to obtain a set of query gene sequence meta-feature semantically encoded joint feature vectors; After respectively calculating the matrix product of each query gene sequence meta-feature semantic encoding joint feature vector in the set of the query gene sequence meta-feature semantic encoding joint feature vector and the weight matrix, the obtained feature vector and the bias vector are added by position to obtain a sequence of query gene sequence meta-feature semantic encoding joint bias feature vectors; The sequence of the query gene sequence meta-feature semantic encoding combined with the bias feature vector is input into a sigmoid activation function to obtain a sequence of the gene sequence meta-feature unidirectional matching semantic metric coefficients.

6. The artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to claim 5 is characterized in that: The normalization unit is used for: Using each gene sequence meta-feature one-way matching semantic measurement coefficient in the sequence of gene sequence meta-feature one-way matching semantic measurement coefficients as an exponent of a natural constant to calculate an exponential function value with the natural constant as a base to obtain a sequence of logarithmic values ​​of gene sequence meta-feature one-way matching semantic measurement coefficients; Calculating the sum of the logarithmic values ​​of the one-way matching semantic metric coefficients of each gene sequence meta-feature in the sequence of the logarithmic values ​​of the one-way matching semantic metric coefficients of the gene sequence meta-feature to obtain the total sum of the one-way matching semantic metric values ​​of the gene sequence meta-feature; Each logarithm of the one-way matching semantic metric coefficient of the gene sequence meta-feature in the sequence of logarithm of the one-way matching semantic metric coefficient of the gene sequence meta-feature is divided by the total sum of the one-way matching semantic metric of the gene sequence meta-feature to obtain a sequence of weight values ​​of the one-way matching semantic metric of the gene sequence meta-feature.

7. The artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to claim 6 is characterized in that: The gene sequence meta-feature one-way matching interactive reshaping unit is used to: Multiply each gene sequence meta-feature one-way matching semantic metric weight value in the sequence of the gene sequence meta-feature one-way matching semantic metric weight values ​​by the corresponding respiratory infectious disease pathogen sample gene sequence meta-feature vector in the set of respiratory infectious disease pathogen sample gene sequence meta-feature vectors by position to obtain a set of weight-optimized respiratory infectious disease pathogen sample gene sequence meta-feature vectors; The positional sum of the set of weight-optimized respiratory infectious disease pathogen sample gene sequence meta-feature vectors is calculated to obtain the gene sequence meta-feature one-way matching interactive reshaping feature vector.

8. The artificial intelligence-assisted screening system for respiratory infectious diseases based on meta-learning technology according to claim 7 is characterized in that: A training module for training the gene sequence inter-meta feature semantic learner including the gene embedding encoder, the gene sequence semantic encoder including the gene embedding encoder, the unidirectional attention granular scanning interactive matching network and the classifier-based respiratory infectious disease identifier; Wherein, the training module includes: A training data acquisition unit, used to acquire training data, wherein the training data includes gene sequences of pathogenic samples of multiple respiratory infectious diseases to obtain a set of training gene sequences of pathogenic samples of respiratory infectious diseases; and gene sequences of training samples of suspected pathogens of the detected object; An inter-sequence meta-feature semantic learning unit is used to input each training respiratory infectious disease pathogen sample gene sequence in the set of training respiratory infectious disease pathogen sample gene sequences into a gene sequence inter-meta-feature semantic learner including a gene embedding encoder to obtain a set of training respiratory infectious disease pathogen sample gene sequence meta-feature vectors; A sequence semantic encoding unit, used for performing semantic encoding processing on the training sample gene sequence of the suspected pathogen of the detected object to obtain a semantic encoding feature vector of the training suspected pathogen sample gene sequence; A sequence semantic one-way matching query unit is used to use the semantic encoding feature vector of the training suspected pathogen sample gene sequence as a query feature vector, and input the query feature vector and the set of the training respiratory infectious disease pathogen sample gene sequence meta-feature vector into a one-way attention granular scanning interactive matching network to obtain a training one-way matching result representation vector; A classification loss unit, used for passing the training one-way matching result representation vector through a classifier-based respiratory infectious disease identifier to obtain a classification loss function value; A one-way matching result loss unit, used to calculate the one-way matching result loss function value; A loss calculation unit, used for calculating a weighted sum of the classification loss function value and the one-way matching result loss function value to obtain a loss function value; A training unit is used to train the gene sequence inter-meta feature semantic learner including the gene embedding encoder, the gene sequence semantic encoder including the gene embedding encoder, the unidirectional attention granular scanning interaction matching network and the classifier-based respiratory infectious disease identifier through gradient back propagation based on the loss function value.

Images