Semi-supervised in-vitro examination information cognition device and method based on random configuration network

Abstract

The invention discloses a semi-supervised in-vitro examination information cognition device and method based on a random configuration network, and the device comprises a data preprocessing module which is used for carrying out the data preprocessing of in-vitro examination data of each mode, and extracting corresponding interpretable biological features; and the semi-supervised in-vitro examination information analysis module is used for generating a pseudo label of unmarked in-vitro examination data by utilizing a random configuration network based on manifold learning according to the extracted interpretable biological characteristics, and classifying by utilizing a random configuration network based on direct fuzzy weighting to obtain a disease data classification result based on in-vitro examination information. According to the invention, the internal relation between the labeled data and the unlabeled data is learned through the semi-supervised in-vitro examination information analysis module, and the distribution information of the labeled sample data and the unlabeled data sample with the pseudo label is learned, so that the disease classification accuracy of the in-vitro examination information is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data analysis, and in particular to a semi-supervised in vitro diagnostic information recognition device and method based on a random configuration network. Background Art

[0002] In order to promote the objectivity and standardization of traditional Chinese medicine, intelligent sensing technology and signal processing technology are used to collect and process digital body surface information, and samples are marked through precise Western pathology. Finally, the diagnosis of different diseases is achieved through data-driven artificial intelligence algorithms. It not only provides a strict basis for the classification of case samples, but also meets the standards of current mainstream research at home and abroad, which is conducive to promoting the internationalization and standardization of traditional Chinese medicine. However, in the process of collecting body surface information, it is easy to cause problems such as missing sample labels due to human factors, equipment failures, etc., which affects the classification accuracy. If only labeled data is used for model learning, the intrinsic information of unlabeled data is easily ignored. The rational and effective use of labeled and unlabeled data is conducive to improving the accuracy of disease diagnosis.

[0003] Different from models such as deep belief networks, deep Boltzmann machines, and convolutional neural networks, random configuration networks have shown great development potential in machine learning due to their rapid and effective modeling. Random configuration networks adopt an incremental learning method and introduce a supervision mechanism. They use inequality constraints to assign values ​​to random parameters and adaptively select the range of values ​​for random parameters. They calculate the output weights of the hidden layer through the least squares method, which ensures the global approximation of the learning model and improves the learning accuracy and efficiency of the network with less human intervention.

[0004] Although there are currently methods for semi-supervised medical data, there has been no research on using semi-supervised random configuration networks to solve in vitro diagnostic data analysis tasks; in the absence of sample labels for surface information collection, existing semi-supervised medical data classification devices only use labeled data for model learning, which easily ignores the intrinsic information of unlabeled data, resulting in low disease classification accuracy of in vitro diagnostic information.

[0005] Therefore, the prior art needs to be improved. Summary of the invention

[0006] The technical problem to be solved by the present invention is that, in view of the defects of the prior art, the present invention provides a semi-supervised in vitro diagnostic information recognition device and method based on a randomly configured network to solve the problem of low disease classification accuracy of in vitro diagnostic information by the existing semi-supervised medical data classification device.

[0007] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0008] In a first aspect, the present invention provides a semi-supervised in vitro diagnostic information recognition device based on a random configuration network, comprising:

[0009] A data preprocessing module is used to preprocess the in vitro diagnostic data of each modality and extract corresponding interpretable biological features;

[0010] The semi-supervised in vitro diagnostic information analysis module is used to generate pseudo labels for unlabeled in vitro diagnostic data based on the extracted interpretable biological features using a random configuration network based on manifold learning, and to classify the data using a random configuration network based on direct fuzzy weighting to obtain disease data classification results based on in vitro diagnostic information.

[0011] In one implementation, the data preprocessing module includes:

[0012] A data acquisition unit, used to acquire in vitro diagnostic data of each modality; wherein the in vitro diagnostic data includes: labeled in vitro diagnostic data and unlabeled in vitro diagnostic data;

[0013] The feature extraction unit is used to extract features according to the signal feature extraction method corresponding to the in vitro diagnostic data of each modality, and obtain the interpretable biological features corresponding to the in vitro diagnostic data of each modality after normalization.

[0014] In one implementation, the semi-supervised in vitro diagnostic information analysis module includes:

[0015] A random configuration network unit based on manifold learning, used to generate a pseudo label for the unlabeled in vitro diagnostic data by learning the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data using the random configuration network unit based on manifold learning according to the extracted interpretable biological features;

[0016] A direct fuzzy weighted random configuration network unit is used to use the direct fuzzy weighted random configuration network to learn the distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label.

[0017] In one implementation, the random configuration network unit based on manifold learning includes:

[0018] An intrinsic connection learning subunit, used for constructing the random configuration network based on manifold learning, and learning the intrinsic connection between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data by a manifold learning method;

[0019] The pseudo-label generation subunit is used to generate a pseudo-label for the unlabeled in vitro diagnostic data according to the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data.

[0020] In one implementation, the direct fuzzy weighted random configuration network unit includes:

[0021] A construction subunit, used for constructing the random configuration network based on direct fuzzy weighting;

[0022] A weighted value assigning subunit, used for assigning a weighted value to the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label by using the intuitionistic fuzzy set theory through the random configuration network based on direct fuzzy weighting;

[0023] The distribution information learning subunit is used to learn the membership and non-membership of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label to the category based on the assigned weighted values ​​to obtain corresponding distribution information.

[0024] In a second aspect, the present invention provides a method for processing semi-supervised in vitro diagnostic information of a semi-supervised in vitro diagnostic information recognition device based on a random configuration network, comprising:

[0025] The in vitro diagnostic data of each modality is preprocessed through the data preprocessing module to extract the corresponding interpretable biological features;

[0026] The extracted interpretable biological features are analyzed through the semi-supervised in vitro diagnostic information analysis module, and pseudo labels of unlabeled in vitro diagnostic data are generated using a random configuration network based on manifold learning. The data are then classified using a random configuration network based on direct fuzzy weighting to obtain disease data classification results based on in vitro diagnostic information.

[0027] In one implementation, the data preprocessing module performs data preprocessing on the in vitro diagnostic data of each modality to extract corresponding interpretable biological features, including:

[0028] The in vitro diagnostic data of each modality is acquired through the data preprocessing module, and feature extraction is performed according to the signal feature extraction method corresponding to the in vitro diagnostic data of each modality, and interpretable biological features corresponding to the in vitro diagnostic data of each modality are obtained after normalization; wherein the in vitro diagnostic data includes: labeled in vitro diagnostic data and unlabeled in vitro diagnostic data.

[0029] In one implementation, the extracting interpretable biological features are analyzed by a semi-supervised in vitro diagnostic information analysis module, pseudo labels of unlabeled in vitro diagnostic data are generated by a random configuration network based on manifold learning, and classification is performed by a random configuration network based on direct fuzzy weighting, including:

[0030] According to the extracted interpretable biological features, the semi-supervised in vitro diagnostic information analysis module uses a random configuration network based on manifold learning to learn the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data, and generates a pseudo label for the unlabeled in vitro diagnostic data;

[0031] The distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo labels is learned by using the random configuration network based on direct fuzzy weighting of the semi-supervised in vitro diagnostic information analysis module.

[0032] In one implementation, the method of using the manifold learning-based random configuration network of the semi-supervised in vitro diagnostic information analysis module to learn the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data to generate pseudo labels for the unlabeled in vitro diagnostic data includes:

[0033] Constructing the manifold learning-based random configuration network through the semi-supervised in vitro diagnostic information analysis module;

[0034] The intrinsic connection between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data is learned by the manifold learning method through the random configuration network based on manifold learning, and based on the intrinsic connection between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data, a pseudo label for the unlabeled in vitro diagnostic data is generated.

[0035] In one implementation, the distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo labels learned by the random configuration network based on direct fuzzy weighting of the semi-supervised in vitro diagnostic information analysis module includes:

[0036] Constructing the random configuration network based on direct fuzzy weighting through the semi-supervised in vitro diagnostic information analysis module;

[0037] Through the random configuration network based on direct fuzzy weighting, intuitive fuzzy set theory is used to assign a weighted value to the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label, and based on the assigned weighted values, the membership and non-membership of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label to the category are learned respectively to obtain corresponding distribution information.

[0038] The present invention adopts the above technical solution to achieve the following effects:

[0039] The present invention performs data preprocessing on the in vitro diagnostic data of each modality through a data preprocessing module, and can extract corresponding interpretable biological features; and learns the intrinsic connection between labeled data and unlabeled data through a semi-supervised in vitro diagnostic information analysis module, as well as learns the distribution information of labeled sample data and unlabeled data samples with pseudo-labels, thereby improving the disease classification accuracy of in vitro diagnostic information. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0041] Figure 1 It is a functional principle diagram of the semi-supervised in vitro diagnostic information recognition device based on a randomly configured network in the present invention.

[0042] Figure 2 It is a flow chart of the semi-supervised in vitro diagnostic information recognition method based on a randomly configured network in the present invention.

[0043] Figure 3 It is a flow chart of a method for recognizing in vitro diagnostic information in an application scenario of the present invention.

[0044] Figure 4 It is a processing flow chart of a data preprocessing module in an application scenario of the present invention.

[0045] Figure 5 It is a processing flow chart of a semi-supervised in vitro diagnostic information analysis module in an application scenario of the present invention.

[0046] Figure 6 It is a functional principle diagram of a terminal in one implementation of the present invention.

[0047] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0049] Exemplary Devices

[0050] In the process of collecting body surface information, it is easy to cause problems such as missing sample labels due to human factors, equipment failure, etc., which affects the classification accuracy. The existing semi-supervised medical data classification device only uses labeled data for model learning, which easily ignores the intrinsic information of unlabeled data, resulting in low disease classification accuracy of in vitro diagnostic information.

[0051] In response to the above technical problems, an embodiment of the present invention provides a semi-supervised in vitro diagnostic information recognition device based on a randomly configured network. The data preprocessing module of the device performs data preprocessing on the in vitro diagnostic data of each modality to extract the corresponding interpretable biological features; and the semi-supervised in vitro diagnostic information analysis module of the device can learn the intrinsic connection between labeled data and unlabeled data, as well as the distribution information of labeled sample data and unlabeled data samples with pseudo-labels; the device of the embodiment of the present invention improves the disease classification accuracy of in vitro diagnostic information.

[0052] like Figure 1 As shown, an embodiment of the present invention provides a semi-supervised in vitro diagnostic information recognition device based on a random configuration network, comprising:

[0053] The data preprocessing module 100 is used to preprocess the in vitro diagnostic data of each modality and extract corresponding interpretable biological features;

[0054] The semi-supervised in vitro diagnostic information analysis module 200 is used to generate pseudo labels for unlabeled in vitro diagnostic data based on the extracted interpretable biological features using a random configuration network based on manifold learning, and to classify the data using a random configuration network based on direct fuzzy weighting to obtain disease data classification results based on in vitro diagnostic information.

[0055] Preferably, the data preprocessing module 100 includes:

[0056] A data acquisition unit, used to acquire in vitro diagnostic data of each modality; wherein the in vitro diagnostic data includes: labeled in vitro diagnostic data and unlabeled in vitro diagnostic data;

[0057] The feature extraction unit is used to extract features according to the signal feature extraction method corresponding to the in vitro diagnostic data of each modality, and obtain the interpretable biological features corresponding to the in vitro diagnostic data of each modality after normalization.

[0058] Preferably, the semi-supervised in vitro diagnostic information analysis module 200 comprises:

[0059] A random configuration network unit based on manifold learning, used to generate a pseudo label for the unlabeled in vitro diagnostic data by learning the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data using the random configuration network unit based on manifold learning according to the extracted interpretable biological features;

[0060] A direct fuzzy weighted random configuration network unit is used to use the direct fuzzy weighted random configuration network to learn the distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label.

[0061] Preferably, the random configuration network unit based on manifold learning includes:

[0062] An intrinsic connection learning subunit, used for constructing the random configuration network based on manifold learning, and learning the intrinsic connection between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data by a manifold learning method;

[0063] The pseudo-label generation subunit is used to generate a pseudo-label for the unlabeled in vitro diagnostic data according to the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data.

[0064] Preferably, the direct fuzzy weighted random configuration network unit comprises:

[0065] A construction subunit, used for constructing the random configuration network based on direct fuzzy weighting;

[0066] A weighted value assigning subunit, used for assigning a weighted value to the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label by using the intuitionistic fuzzy set theory through the random configuration network based on direct fuzzy weighting;

[0067] The distribution information learning subunit is used to learn the membership and non-membership of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label to the category based on the assigned weighted values ​​to obtain corresponding distribution information.

[0068] This embodiment achieves the following technical effects through the above technical solution:

[0069] This embodiment performs data preprocessing on the in vitro diagnostic data of each modality through a data preprocessing module, and can extract corresponding interpretable biological features; and uses a semi-supervised in vitro diagnostic information analysis module to learn the intrinsic connection between labeled data and unlabeled data, as well as the distribution information of labeled sample data and unlabeled data samples with pseudo-labels, thereby improving the disease classification accuracy of in vitro diagnostic information.

[0070] Exemplary Methods

[0071] Based on the above embodiments, the present invention further provides a semi-supervised in vitro diagnostic information processing method of a semi-supervised in vitro diagnostic information recognition device based on a randomly configured network.

[0072] like Figure 2 As shown, an embodiment of the present invention provides a semi-supervised in vitro diagnostic information processing method of a semi-supervised in vitro diagnostic information recognition device based on a random configuration network, comprising the following steps:

[0073] Step S100, performing data preprocessing on the in vitro diagnostic data of each modality through a data preprocessing module to extract corresponding interpretable biological features.

[0074] In this embodiment, for in vitro diagnostic information, a data preprocessing module is used to extract original signal features of different modalities to obtain interpretable feature information; the semi-supervised in vitro diagnostic information analysis module adopts a two-stage learning process. In the first stage, a random configuration network based on manifold learning is used to learn the intrinsic connection between labeled data and unlabeled data, and generate pseudo labels for labeled missing samples. In the second stage, a random configuration network based on direct fuzzy weighting is used to learn labeled sample data and unlabeled data samples with pseudo labels to improve the accuracy of disease diagnosis.

[0075] Specifically, in an implementation of this embodiment, step S100 includes the following steps:

[0076] Step S101, acquiring the in vitro diagnostic data of each modality through the data preprocessing module, and performing feature extraction according to the signal feature extraction method corresponding to the in vitro diagnostic data of each modality, and obtaining the interpretable biological features corresponding to the in vitro diagnostic data of each modality after normalization; wherein the in vitro diagnostic data includes: labeled in vitro diagnostic data and unlabeled in vitro diagnostic data.

[0077] In this embodiment, the in vitro diagnostic data collected by each body surface acquisition device is first obtained, and then the in vitro diagnostic data are preprocessed by using a data preprocessing module, and interpretable feature information is extracted by a corresponding signal feature extraction method; wherein the body surface acquisition device includes but is not limited to: a visual signal acquisition device, an auditory signal acquisition device, an olfactory signal acquisition device and a tactile signal acquisition device; correspondingly, the acquired in vitro diagnostic data include: visual signal data, auditory signal data, olfactory signal data and tactile signal data; among these in vitro diagnostic data, some are marked in vitro diagnostic data, and others are unmarked in vitro diagnostic data; taking tactile signal data (such as pulse wave data) as an example, the data includes marked in vitro diagnostic data and unmarked in vitro diagnostic data.

[0078] For the acquired in vitro diagnostic data of different modalities, feature extraction can be performed using the corresponding signal feature extraction method, and then, after normalization, the interpretable biological features corresponding to the in vitro diagnostic data of each modality can be obtained; taking pulse wave data as an example, for a small amount of labeled in vitro diagnostic data and a large amount of unlabeled in vitro diagnostic data, feature extraction can be performed using short-time Fourier transform. Assuming that the labeled sample is The number of labeled samples is l, and the number of unlabeled samples is The number of unlabeled samples is u.

[0079] In this embodiment, by extracting interpretable biological features corresponding to the in vitro diagnostic data of each modality, a random configuration network based on manifold learning can be used to analyze the relationship between unlabeled in vitro diagnostic data and labeled in vitro diagnostic data, thereby automatically labeling the unlabeled in vitro diagnostic data to form pseudo-labels for classification.

[0080] like Figure 2 As shown, in one implementation of the embodiment of the present invention, the semi-supervised in vitro diagnostic information processing method of the semi-supervised in vitro diagnostic information recognition device based on the random configuration network further includes the following steps:

[0081] Step S200, analyze the extracted interpretable biological features through the semi-supervised in vitro diagnostic information analysis module, generate pseudo labels for unlabeled in vitro diagnostic data using a random configuration network based on manifold learning, and classify them using a random configuration network based on direct fuzzy weighting to obtain disease data classification results based on in vitro diagnostic information.

[0082] In this embodiment, after the interpretable biological features are extracted, the extracted features are used as inputs of a semi-supervised in vitro diagnostic information analysis module for model training to obtain disease data classification results based on the in vitro diagnostic information.

[0083] Specifically, in an implementation of this embodiment, step S200 includes the following steps:

[0084] Step S201, based on the extracted interpretable biological features, the semi-supervised in vitro diagnostic information analysis module uses a random configuration network based on manifold learning to learn the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data, and generates pseudo labels for the unlabeled in vitro diagnostic data.

[0085] Specifically, the method of utilizing the manifold learning-based random configuration network of the semi-supervised in vitro diagnostic information analysis module to learn the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data to generate pseudo labels for the unlabeled in vitro diagnostic data includes: constructing the manifold learning-based random configuration network through the semi-supervised in vitro diagnostic information analysis module; learning the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data by the manifold learning method through the manifold learning-based random configuration network, and generating pseudo labels for the unlabeled in vitro diagnostic data based on the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data.

[0086] In this embodiment, the semi-supervised in vitro diagnostic information analysis module adopts a two-stage learning process. In the first stage, a random configuration network based on manifold learning is used to learn the intrinsic relationship between labeled in vitro diagnostic data and unlabeled in vitro diagnostic data, and to generate pseudo labels for labeled missing samples (i.e., unlabeled in vitro diagnostic data). The main learning process is as follows:

[0087] For the constructed random configuration network based on manifold learning, assuming that the L-1 hidden layer nodes of the random configuration network based on manifold learning have been configured, the current network output result can be calculated by formula (1):

[0088]

[0089] In the formula, X=[X l , X u ] represents the features corresponding to the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data; β j represents the output weight of hidden layer node j; g j (·) represents the activation function; w j and b j They represent the input weight and bias of the jth node in the hidden layer respectively.

[0090] The current network residual is defined as shown in formula (2):

[0091]

[0092] In the formula, it is assumed that the labeled in vitro diagnostic data has m dimensions and the error of m dimensions; f L-1 (X l ) represents the output of labeled in vitro diagnostic data, which is used to calculate the network residual; m represents the original output dimension; N represents the number of samples.

[0093] For the current network residual, such as ||e L-1 || 2 (error norm) does not reach the preset error ε or does not reach the maximum number of nodes Lmax , we need to add the Lth hidden layer node and then randomly generate T max candidate nodes, and determine their input weights and biases according to the supervision mechanism of formula (3). The candidate node parameter with the maximum value is used as the Lth node parameter:

[0094]

[0095] in:

[0096]

[0097] In the formula, w L and b L They represent the input weight and bias of the Lth node in the hidden layer respectively; g L (·) represents the activation function adopted by node L; q = 1, 2, ..., m; represents the output of the hidden layer node L of the labeled data; r∈(0,1); {μ L} represents a non-negative real number sequence, where μ L ≤1-r,lim L→+∞ μ L =0.

[0098] In this embodiment, in order to effectively learn the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data, a manifold regularized objective function is adopted for semi-supervised learning, and the objective function is defined as follows:

[0099]

[0100] Where Y∈R (l+u)×m is the target output, the first l rows of which are equal to the output values ​​of the labeled in vitro diagnostic data, and the remaining u rows are equal to 0; C∈R (l+u)(l+u) represents a diagonal matrix of dimension (l+u)×(l+u), whose first l rows are equal to The remaining u rows are equal to 0; H∈R (l+u)×L represents the output matrix of the hidden layer; Lp represents the graph Laplacian matrix, which is calculated using the manifold learning method; C and λ represent regularization parameters.

[0101] By setting the derivative of formula (5) to zero, we can get the hidden layer output matrix of formula (6). According to formula (7), we can calculate the current network model output. According to the model output of formula (7), we can obtain the pseudo label Y of the unlabeled in vitro diagnostic data: u .

[0102] β=(I L +H T CH+λH T L p H)-1 H T CY (6);

[0103] Y = Hβ (7);

[0104] In this embodiment, through the first stage of learning of the semi-supervised in vitro diagnostic information analysis module, the unlabeled in vitro diagnostic data can be automatically labeled, thereby generating pseudo labels for the unlabeled in vitro diagnostic data that are easy to classify.

[0105] Specifically, in an implementation of this embodiment, step S200 further includes the following steps:

[0106] Step S202 , using the random configuration network based on direct fuzzy weighting of the semi-supervised in vitro diagnostic information analysis module to learn the distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo labels.

[0107] In this embodiment, the second stage of the semi-supervised in vitro diagnostic information analysis module utilizes the distribution information of labeled in vitro diagnostic data and unlabeled in vitro diagnostic data with the pseudo-labels learned by a random configuration network based on direct fuzzy weighting to improve the accuracy of disease diagnosis.

[0108] Specifically, the distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label is learned by using the random configuration network based on direct fuzzy weighting of the semi-supervised in vitro diagnostic information analysis module, including: constructing the random configuration network based on direct fuzzy weighting through the semi-supervised in vitro diagnostic information analysis module; assigning a weighted value to the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label respectively through the random configuration network based on direct fuzzy weighting using intuitive fuzzy set theory, and learning the membership and non-membership of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label to the category based on the assigned weighted values ​​to obtain corresponding distribution information.

[0109] In this embodiment, the complete training data {X u , Y u} and {X l , Y l}, in order to further learn the distribution information of labeled in vitro diagnostic data and unlabeled in vitro diagnostic data with the pseudo-labels, and reduce the impact of potential noise in the generated pseudo-labels on the classification results, the second stage uses a random configuration network based on direct fuzzy weighting to learn the distribution information of labeled in vitro diagnostic data and unlabeled in vitro diagnostic data with the pseudo-labels, thereby improving the accuracy of disease diagnosis.

[0110] Different from the first-stage network configuration process, the random configuration network based on direct fuzzy weighting adopts the objective function of formula (8) and uses the intuitive fuzzy set theory to assign a weighted value s to each in vitro diagnostic data. i By learning the membership and non-membership of each in vitro diagnostic data to the category, more information is provided for classification and the influence of outliers on the classification results is reduced.

[0111] The intuitionistic fuzzy weighted random configuration network can assign a score value to each in vitro diagnostic data to deal with outliers in the data set. The objective function is defined as follows:

[0112]

[0113] By setting the derivative of equation (8) to zero, we can get the hidden layer output matrix of equation (9), and the model output can be calculated according to equation (10).

[0114]

[0115] Y = HB (10);

[0116] In the formula, S=diag(Θ1, Θ2,..., Θ N ), s i =Θ i ; I represents the identity matrix, and C2 represents the regularization parameter.

[0117] In this embodiment, the distribution information of labeled in vitro diagnostic data and unlabeled in vitro diagnostic data with the pseudo-labels is learned by using a random configuration network based on direct fuzzy weighting to determine the membership and non-membership of each in vitro diagnostic data to a category, thereby providing more information for classification and reducing the impact of outliers on the classification results.

[0118] The following is an explanation of the technical solution in this embodiment through specific application examples:

[0119] like Figure 3 As shown, the overall process of the semi-supervised in vitro diagnostic information recognition method based on a random configuration network in this embodiment includes the following steps:

[0120] S10, obtaining visual signals, auditory signals, olfactory signals and tactile signals, and marking missing in vitro diagnostic information;

[0121] S20, data preprocessing, extracting interpretable biometric features;

[0122] S30, random configuration network based on manifold learning learns the intrinsic correlation between labeled samples and unlabeled samples;

[0123] S40, generating pseudo label information of unlabeled data;

[0124] S50, classification based on intuitionistic fuzzy weighted random configuration network;

[0125] S60, obtaining disease classification results.

[0126] like Figure 4 As shown, the above steps S10 to S20 are processed by a data preprocessing module, specifically including:

[0127] S21, inputting in vitro diagnostic information of each modality;

[0128] S22, determining a feature extraction method for different signals;

[0129] S23, feature normalization;

[0130] S24, output the preprocessing result.

[0131] like Figure 5 As shown, the above steps S30 to S60 are processed by a semi-supervised in vitro diagnostic information analysis module, specifically including:

[0132] S31, input labeled and unlabeled data samples;

[0133] S32, learning the intrinsic relationship between labeled and unlabeled data samples through manifold learning method;

[0134] S33, constructing random configuration networks based on manifold learning;

[0135] S34, generating pseudo label information of unlabeled data;

[0136] S35, using direct fuzzy theory to learn the membership and non-membership functions of sample points;

[0137] S36, constructing random configuration networks based on intuitionistic fuzzy weighting;

[0138] S37, obtain disease classification results.

[0139] The above steps S32 to S34 are the first stage of the learning process of the semi-supervised in vitro diagnostic information analysis module; and the above steps S35 to S37 are the second stage of the learning process of the semi-supervised in vitro diagnostic information analysis module.

[0140] In this embodiment, based on some clinical data collected by Guangdong Provincial Hospital of Traditional Chinese Medicine and based on pulse signals, 200 samples of diabetic patients and 200 samples of healthy people are selected for diabetes diagnosis as an example, among which 300 samples are randomly selected as training data and 100 samples are randomly selected as test data. Only 10% of the training data is used for labeling, and feature extraction is performed using short-time Fourier transform, and the features are normalized. The above process is used for verification, and a semi-supervised in vitro diagnosis information analysis module (random configuration network based on manifold learning + random configuration network based on direct fuzzy weighting) can obtain a classification accuracy of 93.47%. The classification accuracy of only using the random configuration network based on manifold learning is 90.71%, and the classification accuracy of only using the random configuration network to learn labeled data is 89.83%. The experimental results verify the effectiveness of the scheme in this embodiment.

[0141] This embodiment achieves the following technical effects through the above technical solution:

[0142] This embodiment uses a random configuration network based on manifold learning to learn the intrinsic relationship between labeled data and unlabeled data, and uses a random configuration network based on direct fuzzy weighting to learn the distribution information of labeled sample data and unlabeled data samples with pseudo labels, thereby improving the disease classification accuracy of in vitro diagnostic information.

[0143] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be as follows: Figure 6 shown.

[0144] The terminal includes: a processor, a memory, an interface, a display screen and a communication module connected through a system bus; wherein the processor of the terminal is used to provide computing and control capabilities; the memory of the terminal includes a storage medium and an internal memory; the storage medium stores an operating system and a computer program; the internal memory provides an environment for the operation of the operating system and the computer program in the storage medium; the interface is used to connect to external devices; the display screen is used to display corresponding information; and the communication module is used to communicate with a cloud server or other devices.

[0145] When the computer program is executed by a processor, it is used to implement the operation of a semi-supervised in vitro diagnostic information processing method of a semi-supervised in vitro diagnostic information recognition device based on a random configuration network.

[0146] It can be understood by those skilled in the art that Figure 6 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0147] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and volatile memory.

[0148] In summary, the present invention provides a semi-supervised in vitro diagnostic information recognition device and method based on a random configuration network, including: a data preprocessing module, which is used to perform data preprocessing on the in vitro diagnostic data of each modality and extract the corresponding interpretable biological features; a semi-supervised in vitro diagnostic information analysis module, which is used to generate pseudo labels for unlabeled in vitro diagnostic data based on the extracted interpretable biological features using a random configuration network based on manifold learning, and classify using a random configuration network based on direct fuzzy weighting to obtain disease data classification results based on in vitro diagnostic information. The present invention learns the intrinsic connection between labeled data and unlabeled data through the semi-supervised in vitro diagnostic information analysis module, and learns the distribution information of labeled sample data and unlabeled data samples with pseudo labels, thereby improving the disease classification accuracy of in vitro diagnostic information.

[0149] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A semi-supervised in vitro diagnostic information recognition device based on a random configuration network, characterized in that: include: A data preprocessing module is used to preprocess the in vitro diagnostic data of each modality and extract corresponding interpretable biological features; The semi-supervised in vitro diagnostic information analysis module is used to generate pseudo labels for unlabeled in vitro diagnostic data based on the extracted interpretable biological features using a random configuration network based on manifold learning, and to classify the data using a random configuration network based on direct fuzzy weighting to obtain disease data classification results based on in vitro diagnostic information.

2. The semi-supervised in vitro diagnostic information recognition device based on a random configuration network according to claim 1, characterized in that: The data preprocessing module comprises: A data acquisition unit, used to acquire in vitro diagnostic data of each modality; wherein the in vitro diagnostic data includes: labeled in vitro diagnostic data and unlabeled in vitro diagnostic data; The feature extraction unit is used to extract features according to the signal feature extraction method corresponding to the in vitro diagnostic data of each modality, and obtain the interpretable biological features corresponding to the in vitro diagnostic data of each modality after normalization.

3. The semi-supervised in vitro diagnostic information recognition device based on a random configuration network according to claim 1, characterized in that: The semi-supervised in vitro diagnostic information analysis module includes: A random configuration network unit based on manifold learning, used to generate a pseudo label for the unlabeled in vitro diagnostic data by learning the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data using the random configuration network unit based on manifold learning according to the extracted interpretable biological features; A direct fuzzy weighted random configuration network unit is used to use the direct fuzzy weighted random configuration network to learn the distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label.

4. The semi-supervised in vitro diagnostic information recognition device based on a random configuration network according to claim 3, characterized in that: The random configuration network unit based on manifold learning includes: An intrinsic connection learning subunit, used for constructing the random configuration network based on manifold learning, and learning the intrinsic connection between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data by a manifold learning method; The pseudo-label generation subunit is used to generate a pseudo-label for the unlabeled in vitro diagnostic data according to the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data.

5. The semi-supervised in vitro diagnostic information recognition device based on random configuration network according to claim 3 is characterized in that: The direct fuzzy weighted random configuration network unit includes: A construction subunit, used for constructing the random configuration network based on direct fuzzy weighting; A weighted value assigning subunit, used for assigning a weighted value to the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label by using the intuitionistic fuzzy set theory through the random configuration network based on direct fuzzy weighting; The distribution information learning subunit is used to learn the membership and non-membership of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label to the category based on the assigned weighted values ​​to obtain corresponding distribution information.

6. A method for processing semi-supervised in vitro diagnostic information based on a semi-supervised in vitro diagnostic information recognition device of a random configuration network, characterized in that: include: The in vitro diagnostic data of each modality is preprocessed through the data preprocessing module to extract the corresponding interpretable biological features; The extracted interpretable biological features are analyzed through the semi-supervised in vitro diagnostic information analysis module, and pseudo labels of unlabeled in vitro diagnostic data are generated using a random configuration network based on manifold learning. The data are then classified using a random configuration network based on direct fuzzy weighting to obtain disease data classification results based on in vitro diagnostic information.

7. The semi-supervised in vitro diagnostic information recognition method based on random configuration network according to claim 6 is characterized in that: The data preprocessing module performs data preprocessing on the in vitro diagnostic data of each modality to extract corresponding interpretable biological features, including: The in vitro diagnostic data of each modality is acquired through the data preprocessing module, and feature extraction is performed according to the signal feature extraction method corresponding to the in vitro diagnostic data of each modality, and interpretable biological features corresponding to the in vitro diagnostic data of each modality are obtained after normalization; wherein the in vitro diagnostic data includes: labeled in vitro diagnostic data and unlabeled in vitro diagnostic data.

8. The semi-supervised in vitro diagnostic information recognition method based on random configuration network according to claim 6, characterized in that: The method analyzes the extracted interpretable biological features through a semi-supervised in vitro diagnostic information analysis module, generates pseudo labels for unlabeled in vitro diagnostic data using a random configuration network based on manifold learning, and classifies using a random configuration network based on direct fuzzy weighting, including: According to the extracted interpretable biological features, the semi-supervised in vitro diagnostic information analysis module uses a random configuration network based on manifold learning to learn the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data, and generates a pseudo label for the unlabeled in vitro diagnostic data; The distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo labels is learned by using the random configuration network based on direct fuzzy weighting of the semi-supervised in vitro diagnostic information analysis module.

9. The semi-supervised in vitro diagnostic information recognition method based on random configuration network according to claim 8, characterized in that: The method of using the random configuration network based on manifold learning of the semi-supervised in vitro diagnostic information analysis module to learn the intrinsic relationship between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data to generate pseudo labels for the unlabeled in vitro diagnostic data includes: Constructing the manifold learning-based random configuration network through the semi-supervised in vitro diagnostic information analysis module; The intrinsic connection between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data is learned by the manifold learning method through the random configuration network based on manifold learning, and based on the intrinsic connection between the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data, a pseudo label for the unlabeled in vitro diagnostic data is generated.

10. The semi-supervised in vitro diagnostic information recognition method based on random configuration network according to claim 8, characterized in that: The distribution information of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo labels learned by the random configuration network based on direct fuzzy weighting using the semi-supervised in vitro diagnostic information analysis module includes: Constructing the random configuration network based on direct fuzzy weighting through the semi-supervised in vitro diagnostic information analysis module; Through the random configuration network based on direct fuzzy weighting, intuitive fuzzy set theory is used to assign a weighted value to the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label, and based on the assigned weighted values, the membership and non-membership of the labeled in vitro diagnostic data and the unlabeled in vitro diagnostic data with the pseudo-label to the category are learned respectively to obtain corresponding distribution information.

Images