CN107193993A - The medical data sorting technique and device selected based on local learning characteristic weight - Google Patents

Abstract

The invention discloses a kind of medical data sorting technique selected based on local learning characteristic weight, the property value of sample is obtained according to training sample set first, the corresponding weight vectors of weight update mode computation attribute that gradient declines are utilized according to property value, therefore convergence can be ensured, the stopping criterion of algorithm can be reached quickly, the calculating time is reduced, computation complexity is reduced；Feature selecting, which is carried out, according to the weight vectors calculated obtains optimal characteristics collection, feature selecting is carried out in optimal feature subset again after data sample to be assessed is standardized, the data sample to be assessed after feature selecting, which is classified, again can just make data sample realize dimensionality reduction, therefore method provided in an embodiment of the present invention realizes the complexity for reducing calculating while dimensionality reduction again, reduces the calculating time.Present invention also offers a kind of medical data sorter selected based on local learning characteristic weight, above-mentioned technique effect can be equally realized.

Description

Medical data classification method and device based on local learning feature weight selection

technical field

The present invention relates to the field of medical diagnosis, more specifically, to a medical data classification method and device based on local learning feature weight selection.

Background technique

With the development of artificial intelligence, computer technology has also played an important role in the medical field, realizing artificial intelligence in the medical field. Combining computer technology with a large amount of authoritative knowledge and experience of human medical experts in many fields, a medical diagnosis system has been developed, which can effectively solve various clinical problems and play a role in assisting doctors in diagnosis.

In the medical diagnosis system, DNA microarray technology, that is, gene chip, is introduced. The application of gene chip can quantitatively analyze the level of a large amount of gene expression data at the same time, and through these data, the essence of biology can be studied. However, due to the development of DNA microarray technology, the explosive growth of gene expression data has led to the selection of important genes in these large amounts of gene expression data, which poses new challenges to the existing technology.

The Local Hyperlane (LH-Relief) algorithm can reduce the dimensionality of a large amount of gene expression data, that is, filter out useless gene expression data, select important genes, and reduce redundancy. However, in the application of this algorithm to noisy data and high-level data, the convergence cannot be guaranteed, resulting in high computational complexity of the algorithm.

Therefore, how to reduce the dimensionality of a large amount of genetic data while reducing the computational complexity of the algorithm is a problem to be solved by those skilled in the art.

Contents of the invention

The purpose of the present invention is to provide a medical data classification method based on local learning feature weight selection, so as to reduce the dimensionality of a large amount of genetic data and reduce the computational complexity of the algorithm.

In order to achieve the above object, the embodiment of the present invention provides the following technical solutions:

A medical data classification method based on local learning feature weight selection, including:

S101: Obtain a first sample set of medical data, and obtain attributes of the first sample;

S102: Set an initial weight vector of the first sample attribute, and use the initial weight vector as the current weight vector;

S103: Update the current weight vector by gradient descent to obtain the next weight vector after one iteration;

S104: Determine whether the determination rule is established, if yes, use the weight vector as the final weight vector, and execute S105; if not, use the next weight vector as the current weight vector, and return to S103; where ||w ^t+1 - w ^t ||≤θ is the determination rule, w ^t is the current weight vector, w ^t+1 is the next weight vector, and θ is the stop criterion;

S105: Perform feature selection according to the final weight vector to obtain a feature index subset;

S106: Perform feature selection on the first sample set according to the feature index subset to obtain a second sample set after feature selection;

S107: Obtain the first data to be evaluated, and perform feature selection according to the feature index subset to obtain second data to be evaluated;

S108: Classify the second data to be evaluated on the second sample set to obtain a classification result.

Preferably, said obtaining the first sample set of medical data to obtain the first sample attributes includes:

Acquiring a first sample set of medical data, obtaining attributes of the first sample, and performing dispersion standardization processing on the first sample set;

Preferably, the update method of gradient descent is used to update the current weight vector to obtain the next weight vector after one iteration, including:

pass the rules Update the current weight vector to obtain the next weight vector w ^t+1 after one iteration, J(w) is the optimization objective function, by maximizing J(w)=(z _i ^t+1 ) ^T w ^{t+ 1} is calculated.

Preferably, the acquiring the first data to be evaluated, and performing feature selection according to the feature index subset to obtain the second data to be evaluated includes:

Acquire the first data to be evaluated, perform dispersion standardization processing, and perform feature selection according to the feature index subset to obtain the second data to be evaluated.

Preferably, the second data to be evaluated is classified on the second sample set to obtain classification results, including:

Classify the second data to be evaluated on the second sample set using a K-nearest neighbor classifier to obtain a classification result.

A medical data classification device based on local learning feature weight selection, comprising:

The first sample set obtaining module is used to obtain the first sample set of medical data and obtain the first sample attributes;

The initial weight limit setting module is used to set the initial weight vector of the first sample attribute, and use the initial weight vector as the current weight vector;

The next weight vector acquisition module is used to update the current weight vector through gradient descent to obtain the next weight vector after one iteration;

A judging module, used to judge whether the determination rule is established, if so, use the next weight vector as the final weight vector, and call the feature index subset acquisition module; if not, use the next weight vector as the current weight vector, call The next weight vector acquisition module; wherein the determination rule is ||w ^t+1 -w ^t ||≤θ, w ^t is the current weight vector, w ^t+1 is the next weight vector, and θ is the stop criterion;

The feature index subset acquisition module is configured to perform feature selection according to the final weight vector to obtain a feature index subset;

A second sample set acquisition module, configured to perform feature selection on the first sample set according to the feature index subset, to obtain a second sample set after feature selection;

The second data to be evaluated acquisition module is used to obtain the first data to be evaluated, and perform feature selection according to the feature index subset to obtain the second data to be evaluated;

A classification module, configured to classify the second data to be evaluated on the second sample set to obtain a classification result.

Preferably, the first sample set acquisition module is specifically used for:

Acquire a first sample set of medical data, obtain attributes of the first sample, and perform dispersion standardization processing on the first sample set.

Preferably, the next weight vector acquisition module is specifically used for:

pass the rules Update the current weight vector to obtain the next weight vector w ^t+1 after one iteration, J(w) is the optimization objective function, by maximizing J(w)=(z _i ^t+1 ) ^T w ^{t+ 1} is calculated.

Preferably, the second data acquisition module to be evaluated is specifically used for:

Acquire the first data to be evaluated, perform dispersion standardization processing, and perform feature selection according to the feature index subset to obtain the second data to be evaluated.

Preferably, the classification module is specifically used for:

Classify the second data to be evaluated on the second sample set using a K-nearest neighbor classifier to obtain a classification result.

From the above schemes, it can be seen that in the embodiment of the present invention, a medical data classification method based on local learning feature weight selection, first obtains the attribute value of the sample according to the training sample set, and uses the weight update method of gradient descent to calculate the corresponding attribute value according to the attribute value. Weight vector, so the convergence can be guaranteed, the stopping criterion of the algorithm can be reached quickly, the calculation time is reduced, and the calculation complexity is reduced; according to the calculated weight vector, the feature selection is performed to obtain the optimal feature set, and the data samples to be evaluated are standardized. Then perform feature selection in the optimal feature subset, and then classify the data samples to be evaluated after feature selection, so that the data samples can achieve dimensionality reduction. Therefore, the method provided by the embodiment of the present invention realizes dimensionality reduction while reducing calculation complexity, reducing computation time. The present invention also provides a medical data classification device based on local learning feature weight selection, which can also achieve the above technical effects.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is a flow chart of a medical data classification method disclosed in an embodiment of the present invention;

Fig. 2 is a schematic structural diagram of a medical data classification device disclosed in an embodiment of the present invention;

Fig. 3 is a graph comparing the convergence results of a medical data classification method disclosed in an embodiment of the present invention and LH-RELIEF.

Fig. 4 is a comparison chart of the average performance between a medical data classification method disclosed in the embodiment of the present invention and LH-RELIEF.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Referring to FIG. 1 , an embodiment of the present invention discloses a medical data classification method based on local learning feature weight selection. specifically:

S101: Obtain a first sample set of medical data, and obtain attributes of the first sample.

Specifically, to obtain the first sample set of medical data The sample attribute of the first sample set is obtained as the first sample attribute. Where _xi ∈ R ^I , y _i ∈ {1,2,...,C} is the label of _xi , indicating the category of _xi , N is the number of training samples, I is the dimension of the sample, C is total number of categories.

S102: Set an initial weight vector of the first sample attribute, and use the initial weight vector as a current weight vector.

Specifically, set the initial weight vector w ⁰ =[1/I,1/I,...,1/I] ^t , where t is the number of iterations, and the current t=0, that is, the iteration is not started, the initial weight vector w ⁰ as this weight vector w ^t .

S103: Update the current weight vector by gradient descent to obtain a next weight vector after one iteration.

Specifically, one iteration is performed, that is, the current weight vector w ^t is updated once using a gradient descent update method to obtain the next weight vector w ^t+1 .

S104: Determine whether the determination rule is established, if yes, use the weight vector as the final weight vector, and execute S105; if not, use the next weight vector as the current weight vector, and return to S103; where the determination rule is ||w ^{t +1} -w ^t ||≤θ, w ^t is the current weight vector, w ^t+1 is the next weight vector, and θ is the stopping criterion.

Specifically, set a stopping criterion θ, and judge whether ||w ^t+1 -w ^t ||≤θ holds true, and if so, take the next weight vector w ^t+1 as the final weight vector w, w=[w ₁ ,w ₂ ,...,w _I ] ^t ∈ R, proceed to S105; if not established, use the next weight vector w ^t+1 as the current weight vector w ^t , and return to S103 for a new iteration.

S105: Perform feature selection according to the final weight vector to obtain a feature index subset.

Specifically, feature selection is performed by classification accuracy according to the final weight vector w, and the corresponding feature index subset is obtained Realize feature dimensionality reduction for the first sample, thereby reducing calculation amount and calculation time.

S106: Perform feature selection on the first sample set according to the feature index subset, to obtain a second sample set after feature selection.

Specifically, the first sample set Index subset by feature Perform feature selection to obtain the second sample set Each sample x _i ∈R ^|F| , |F|<I.

S107: Acquire the first data to be evaluated, and perform feature selection according to the feature index subset to obtain second data to be evaluated.

Specifically, the first data sample x to be evaluated is obtained, x∈R ^I , the current sample x is not subjected to dimensionality reduction processing, and the sample dimension is I. Subset data samples according to feature index Perform feature selection to obtain the second data x' to be evaluated.

S108: Classify the second data to be evaluated on the second sample set to obtain a classification result.

Specifically, in the second sample set Classify the second data x′ to be evaluated to obtain a classification result and obtain a classification result. The classification result can be used to diagnose the first data sample x to be evaluated.

Therefore, a medical data classification method based on local learning feature weight selection provided by the embodiment of the present invention first obtains the attribute value of the sample according to the training sample set, and calculates the weight vector corresponding to the attribute by using the weight update method of gradient descent according to the attribute value, Therefore, the convergence can be guaranteed, the stopping criterion of the algorithm can be reached quickly, the calculation time is reduced, and the calculation complexity is reduced; according to the calculated weight vector, the feature selection is performed to obtain the optimal feature set, and the data samples to be evaluated are standardized before the final Performing feature selection in the optimal feature subset, and then classifying the data samples to be evaluated after feature selection can reduce the dimensionality of the data samples. Therefore, the method provided by the embodiment of the present invention reduces the complexity of calculation while achieving dimensionality reduction. Reduced computation time.

The embodiment of the present invention discloses a specific medical data classification method based on local learning feature weight selection. Different from the previous embodiment, the present invention specifically limits S101, and other steps are roughly the same as the previous embodiment. For details, refer to the previous embodiment, which will not be repeated here. Specifically, S101 includes:

Acquiring a first sample set of medical data, obtaining attributes of the first sample, and performing dispersion standardization processing on the first sample set;

Specifically, to obtain the first sample set of medical data The sample attribute of the first sample set is obtained as the first sample attribute. Where _xi ∈ R ^I , y _i ∈ {1,2,...,C} is the label of _xi , indicating the category of _xi , N is the number of training samples, I is the dimension of the sample, C is total number of categories.

It should be noted that different feature attributes often have different dimensions and dimensional units, which will affect the results of data analysis. In order to eliminate the impact of different dimensions and dimensional units, it is necessary to analyze the first sample set Dispersion standardization is performed to solve the comparability of feature attribute data. The conversion function of dispersion standardization is Among them, x _ij is the jth attribute of the i-th sample, In order to take the maximum value of attribute j in all training sample data, is the minimum value of attribute j in all data. After the standardization process, all the indicators of the feature data are of the same order of magnitude, which is more conducive to comprehensive comparison and evaluation of these data. The feature data used in the embodiments of the present invention are all data after the standardization process of dispersion.

The embodiment of the present invention discloses a specific medical data classification method based on local learning feature weight selection. Different from the previous embodiment, the present invention specifically limits S103, and other steps are roughly the same as the previous embodiment. For details, refer to the previous embodiment, which will not be repeated here. Specifically, S103 includes:

pass the rules Update the current weight vector to obtain the next weight vector w ^t+1 after one iteration, J(w) is the optimization objective function, by maximizing J(w)=(z _i ^t+1 ) ^T w ^{t+ 1} is calculated.

Specifically, maximizing Solve J(w), and update the next weight vector w ^t+1 .

in with They are the neighboring sample matrix of the sample x _i in heterogeneous samples and similar samples, and k is the number of neighbors set a priori. α _i and β _i are the coefficient vectors of heterogeneous samples and similar samples _xi , respectively. solve The optimization problem of can obtain α _i ; solve The optimization problem of can obtain β _i ,

Therefore, by optimizing the objective function J(w), the formula Update the current weight vector w ^t to obtain the next weight vector w ^t+1 after one iteration.

The weight update method using gradient descent can guarantee the convergence. When the convergence can be guaranteed, the stopping criterion of the algorithm can be reached quickly, so the complexity of the calculation can be reduced, and the calculation time can be reduced.

The embodiment of the present invention discloses a specific medical data classification method based on local learning feature weight selection. Different from the previous embodiment, the present invention specifically limits S107, and other steps are roughly the same as the previous embodiment. For details, refer to the previous embodiment, which will not be repeated here. Specifically, S107 includes:

Acquire the first data to be evaluated, perform dispersion standardization processing, and perform feature selection according to the feature index subset to obtain the second data to be evaluated.

Specifically, the credit data sample x to be evaluated is obtained as the first data to be evaluated, where x∈R ^I , and the first data to be evaluated is standardized using the deviation standardization method introduced in the above embodiment, namely

It should be noted that the first data to be evaluated used in the present invention are all data after standardization of spread, and standardization of dispersion is carried out on the first data to be evaluated, which also avoids the dimension and dimension unit between characteristic data. The different impact data analysis results, the data are standardized, so that the indicators of the data to be evaluated are in the same order of magnitude, which is suitable for comprehensive comparison and evaluation.

The embodiment of the present invention discloses a specific medical data classification method based on local learning feature weight selection. Different from the previous embodiment, the present invention specifically limits S108, and the other steps are roughly the same as the previous embodiment. For details, refer to the previous embodiment, which will not be repeated here. Specifically, S108 includes:

Classify the second data to be evaluated on the second sample set using a K-nearest neighbor classifier to obtain a classification result.

Specifically, in the second sample set On the basis of , use the K-nearest neighbor classifier to classify the second data x′ to be evaluated, and obtain the classification result. The classification result can be used to diagnose the first data sample x to be evaluated.

A medical data classification device based on local learning feature weight selection provided by an embodiment of the present invention is introduced below. The medical data classification device described below and the medical data classification method described above can be referred to each other.

Referring to Fig. 2, an embodiment of the present invention provides a medical data classification device based on local learning feature weight selection, which specifically includes:

The first sample set acquiring module 201 is configured to acquire a first sample set of medical data to obtain a first sample attribute.

Specifically, the first sample set acquisition module 201 acquires the first sample set of medical data The sample attribute of the first sample set is obtained as the first sample attribute. Where _xi ∈ R ^I , y _i ∈ {1,2,...,C} is the label of _xi , indicating the category of _xi , N is the number of training samples, I is the dimension of the sample, C is total number of categories.

The initial weight limit setting module 202 is configured to set the initial weight vector of the first sample attribute, and use the initial weight vector as the current weight vector.

Specifically, the initial weight limit setting module 202 sets the initial weight vector, that is, the initial weight vector is w ⁰ =[1/I,1/I,...,1/I] ^t , where t is the number of iterations, and the current t =0, that is, the iteration is not started, and the initial weight vector w ⁰ is used as the current weight vector w ^t .

The next weight vector acquisition module 203 is configured to update the current weight vector by gradient descent to obtain the next weight vector after one iteration.

Specifically, the current weight vector is iterated through the next weight vector acquisition module 203 , that is, the current weight vector w ^t is updated once using gradient descent to obtain the next weight vector w ^t+1 .

Judging module 204, used to judge whether the rule of determination is established, if so, then use the next weight vector as the final weight vector, and call the feature index subset acquisition module; if not, use the next weight vector as the current weight vector , call the next weight vector acquisition module; wherein the determination rule is ||w ^t+1 -w ^t ||≤θ, w ^t is the current weight vector, w ^t+1 is the next weight vector, and θ is the stop guidelines.

Specifically, a stop criterion θ is set in the judging module 204, and it is judged whether ||w ^t+1 -w ^t ||≤θ is established, and if yes, the next weight vector w ^t+1 is taken as the final weight vector w, w=[w ₁ ,w ₂ ,...,w _I ] ^t ∈ R, call the feature index subset acquisition module 205; if not, use the next weight vector w ^t+1 as the current weight vector w ^t , call the next weight vector acquisition module 203 again to perform a new iteration.

The feature index subset obtaining module 205 is configured to perform feature selection according to the final weight vector to obtain a feature index subset.

Specifically, the feature index subset acquisition module 205 performs feature selection according to the final weight vector w through the classification accuracy, and obtains the corresponding feature index subset Realize feature dimensionality reduction for the first sample, thereby reducing calculation amount and calculation time.

The second sample set acquisition module 206 is configured to perform feature selection on the first sample set according to the feature index subset, to obtain a second sample set after feature selection.

Specifically, in the second sample set acquisition module 206, the first sample set Index subset by feature Perform feature selection to obtain the second sample set Each sample x _i ∈R ^|F| , |F|<I.

The second data to be evaluated acquisition module 207 is configured to acquire the first data to be evaluated, and perform feature selection according to the feature index subset to obtain second data to be evaluated.

Specifically, the second data to be evaluated acquisition module 207 acquires the first data to be evaluated sample x, x∈R ^I , the current sample x has not undergone dimensionality reduction processing, and the sample dimension is I. Subset data samples according to feature index Perform feature selection to obtain the second data x' to be evaluated.

A classification module 208, configured to classify the second data to be evaluated on the second sample set to obtain a classification result.

Specifically, the classification module 208 puts the second data to be evaluated x' in the second sample set Classify, get the classification result, get the classification result. The classification result can be used to diagnose the first data sample x to be evaluated.

Therefore, in the medical data classification method based on local learning feature weight selection provided by the embodiment of the present invention, the attribute value of the sample is first obtained through the first sample set acquisition module 201, and in the next weight vector acquisition module 203 according to the attribute value, Use the weight update method of gradient descent to calculate the weight vector corresponding to the attribute, so the convergence can be guaranteed, the stop criterion of the algorithm can be reached quickly, the calculation time is reduced, and the calculation complexity is reduced; the second sample set acquisition module 206 according to the calculated The weight vector performs feature selection to obtain the optimal feature set, the second data sample to be evaluated 207 standardizes the data samples to be evaluated, and then performs feature selection in the optimal feature subset, and then classifies the data samples to be evaluated after feature selection. Dimensionality reduction can be achieved for data samples, so the method provided by the embodiment of the present invention not only realizes dimensionality reduction, but also reduces computation complexity and computation time.

The embodiment of the present invention discloses a specific medical data classification device based on local learning feature weight selection. Different from the previous embodiment, the present invention makes specific limitations on the first sample set acquisition module 201, and the contents of other steps are the same as The previous embodiment is roughly the same, and details can be referred to the previous embodiment, which will not be repeated here. The above-mentioned first sample set acquisition module 201 is specifically used for:

Acquire a first sample set of medical data, obtain attributes of the first sample, and perform dispersion standardization processing on the first sample set.

Specifically, the first sample set acquisition module 201 acquires the first sample set of medical data The sample attribute of the first sample set is obtained as the first sample attribute. Where _xi ∈ R ^I , y _i ∈ {1,2,...,C} is the label of _xi , indicating the category of _xi , N is the number of training samples, I is the dimension of the sample, C is total number of categories.

It should be noted that different feature attributes often have different dimensions and dimensional units, which will affect the results of data analysis. In order to eliminate the impact of different dimensions and dimensional units, it is necessary to analyze the first sample set Dispersion standardization is performed to solve the comparability of feature attribute data. The conversion function of dispersion standardization is Among them, x _ij is the jth attribute of the i-th sample, In order to take the maximum value of attribute j in all training sample data, is the minimum value of attribute j in all data. After the standardization process, all the indicators of the feature data are of the same order of magnitude, which is more conducive to comprehensive comparison and evaluation of these data. The feature data used in the embodiments of the present invention are all data after the standardization process of dispersion.

The embodiment of the present invention discloses a specific medical data classification device based on local learning feature weight selection. Different from the previous embodiment, the present invention makes specific limitations on the next weight vector acquisition module 203, and the content of other steps is the same as the above The first embodiment is roughly the same, and for details, refer to the previous embodiment, and details are not repeated here. The above-mentioned next weight vector acquisition module 203 is specifically used for:

pass the rules Update the current weight vector to obtain the next weight vector w ^t+1 after one iteration, and J(w) is calculated by maximizing the optimization objective function J(w)=(z _i ^t+1 ) ^T w ^t+1 get.

Specifically, in the next weight vector acquisition module 203, first maximize Solve J(w), and update the next weight vector w ^t+1 .

in with They are the neighboring sample matrix of the sample x _i in heterogeneous samples and similar samples, and k is the number of neighbors set a priori. α _i and β _i are the coefficient vectors of heterogeneous samples and similar samples _xi , respectively. solve The optimization problem of can obtain α _i ; solve The optimization problem of can obtain β _i ,

Therefore, J(w) can be used to use the formula Update the current weight vector w ^t to obtain the next weight vector w ^t+1 after one iteration. Wherein, the optimization objective function J(w) is calculated by maximizing J(w)=(z _i ^t+1 ) ^T w ^t+1 .

The weight update method using gradient descent can guarantee the convergence. When the convergence can be guaranteed, the stopping criterion of the algorithm can be reached quickly, so the complexity of the calculation can be reduced, and the calculation time can be reduced.

The embodiment of the present invention discloses a specific medical data classification device based on local learning feature weight selection. Different from the previous embodiment, the present invention makes specific limitations on the second data acquisition module 207 to be evaluated, and the content of other steps is the same as The previous embodiment is roughly the same, and details can be referred to the previous embodiment, which will not be repeated here. The above-mentioned second data acquisition module 207 to be evaluated is specifically used for:

Acquire the first data to be evaluated, perform dispersion standardization processing, and perform feature selection according to the feature index subset to obtain the second data to be evaluated.

Specifically, the second data to be evaluated acquisition module 207 obtains the credit data sample x to be evaluated as the first data to be evaluated, where x∈R ^I uses the deviation standardization method introduced in the above-mentioned embodiments to perform standardized treatment, that is

It should be noted that the first data to be evaluated used in the present invention are all data after standardization of spread, and standardization of dispersion is carried out on the first data to be evaluated, which also avoids the dimension and dimension unit between characteristic data. The different impact data analysis results, the data are standardized, so that the indicators of the data to be evaluated are in the same order of magnitude, which is suitable for comprehensive comparison and evaluation.

The embodiment of the present invention discloses a specific medical data classification device based on local learning feature weight selection. Different from the previous embodiment, the present invention specifically limits the classification module 208, and other steps are roughly the same as the previous embodiment. Same, for details, refer to the previous embodiment, which will not be repeated here. The above classification module 208 is specifically used for:

Classify the second data to be evaluated on the second sample set using a K-nearest neighbor classifier to obtain a classification result.

Specifically, the classification module 208 in the second sample set On the basis of , use the K-nearest neighbor classifier to classify the second data x′ to be evaluated, and obtain the classification result. The classification result can be used to diagnose the first data sample x to be evaluated.

The embodiment of the present invention discloses a medical data classification method based on local learning feature weight, which specifically includes:

The embodiment of the present invention is tested in the embryo data set (CNS) data set, which contains 34 patient samples, and each sample has 7129 genes. These 34 samples included 25 classical medulloblastomas (C) and 9 desmoplastic medulloblastomas (D), so there were 2 categories. The CNS dataset is divided into two subsets: 23 training samples (6 C, 17 D), used to select genes and adjust the weight of the classifier, and 11 test samples (3 C, 8 D), used to Evaluate the performance of the results obtained by the system. Each sample has 7129 features. We consider C as the first class and D as the second class. The specific implementation steps are divided into two modules, as follows:

Model training module:

S301, input medical data sample set As the first sample set, where _xi ∈ R ^I , y _i ∈ {1,2,...,C} is the label of _xi , indicating the category of _xi , N is the number of training samples, I is The dimension of the sample, C is the total number of categories. Here N=23, I=7129, C=2.

S302. Perform dispersion standardization processing on the first sample set, and the conversion function is Among them, x _ij is the jth attribute of the i-th sample, In order to take the maximum value of attribute j in all training sample data, is the minimum value of attribute j in all data.

S303. Set an initial weight vector w ⁰ =[1/I,1/I,...,1/I] ^t of the first sample attribute, and use the initial weight vector as the current weight vector. Where t is the number of iterations, currently t=0, that is, the iteration is not started, and the initial weight vector w ⁰ is used as the weight vector w ^t of this time, and the number of iterations is 30 in total, that is, a total of 30 iterations are performed.

S304. Update the current weight vector by gradient descent to obtain a next weight vector after one iteration.

Specifically, maximizing Solve the optimization objective function J(w), and update the next weight vector w ^t+1 .

in with They are the neighboring sample matrix of the sample x _i in heterogeneous samples and similar samples, and k is the number of neighbors set a priori. α _i and β _i are the coefficient vectors of heterogeneous samples and similar samples _xi , respectively. solve The optimization problem of can obtain α _i ; solve The optimization problem of can obtain β _i ,

Therefore, J(w) can be used to use the formula Update the current weight vector w ^t to obtain the next weight vector w ^t+1 after one iteration.

S305, judge whether the determination rule is established, if so, use the weight vector as the final weight vector, and execute S306; if not, use the next weight vector as the current weight vector, and return to S304; wherein the determination rule is ||w ^{t +1} -w ^t ||≤θ, w ^t is the current weight vector, w ^t+1 is the next weight vector, and θ is the stopping criterion.

Specifically, set a stopping criterion θ=0.001, and judge whether ||w ^t+1 -w ^t ||≤θ holds true, and if so, take the next weight vector w ^t+1 as the final weight vector w, w= [w ₁ ,w ₂ ,...,w _I ] ^t ∈ R ⁷¹²⁹ , proceed to S306; if not established, take the next weight vector w ^t+1 as the current weight vector w ^t , and return to S304 to perform a new one iteration.

S306. Perform feature selection according to the final weight vector to obtain a feature index subset.

Specifically, feature selection is performed by classification accuracy according to the final weight vector w, and the corresponding feature index subset is obtained Realize feature dimensionality reduction for the first sample, thereby reducing calculation amount and calculation time.

S307. Perform feature selection on the first sample set according to the feature index subset to obtain a second sample set after feature selection.

Specifically, the first sample set Index subset by feature Perform feature selection to obtain the second sample set Wherein each sample x _i ∈R ^|F| , |F|<7129.

Evaluation modules:

S308. Acquire first data to be evaluated.

Specifically, input the credit data sample x to be evaluated as the first data sample to be evaluated, x∈R ^I .

S309. Perform dispersion standardization processing on the first data to be evaluated.

Specifically, the credit data sample x to be evaluated is obtained as the first data to be evaluated, where x∈R ^I , and the first data to be evaluated is standardized using the deviation standardization method introduced in the above embodiment, namely

S310, index the subset according to the feature Feature selection is performed on the first data to be evaluated to obtain the second data to be evaluated x′.

S311. Classify the second data to be evaluated on the second sample set using a K-nearest neighbor classifier to obtain a classification result.

Specifically, in the second sample set On the basis of , use the K-nearest neighbor classifier to classify the second data x′ to be evaluated, and obtain the classification result. The classification result can be used to diagnose the first data sample x to be evaluated.

Through the present invention, a medical data classification method based on local learning feature weight is proposed, the feature selection method of LH-RELIEF is improved, and the combination F of features in 23 7129-dimensional training samples is extracted, 1≤length(F) ≤7129, classify 11 test samples of 7129 dimensions. The method proposed in this experiment is compared with the LH-RELIEF algorithm on the same data set, and 78 training samples are randomly selected 10 times. The average convergence results are shown in Figure 3, and the average performance results are shown in Figure 4. It can be seen that the present invention converges faster than the MSVM-RFE algorithm, and has better classification performance under the same selection of the same number of genes.

Table 1 gives a comparison of the best average classification performance obtained by each of the two methods. The present invention has an improvement of about 2 percentage points over the LH-RELIEF method.

Table 1 Comparison of LH-RELIEF and the best classification performance of the present invention

method Recognition rate(%) this invention 70.91(10) LH-RELIEF 69.09(10)

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A medical data classification method based on local learning feature weight selection, characterized in that, comprising: S101: Obtain a first sample set of medical data, and obtain attributes of the first sample; S102: Set an initial weight vector of the first sample attribute, and use the initial weight vector as the current weight vector; S103: Update the current weight vector by gradient descent to obtain the next weight vector after one iteration; S104: Determine whether the determination rule is established, if yes, use the weight vector as the final weight vector, and execute S105; if not, use the next weight vector as the current weight vector, and return to S103; where ||w ^t+1 - w ^t ||≤θ is the determination rule, w ^t is the current weight vector, w ^t+1 is the next weight vector, and θ is the stop criterion; S105: Perform feature selection according to the final weight vector to obtain a feature index subset; S106: Perform feature selection on the first sample set according to the feature index subset to obtain a second sample set after feature selection; S107: Obtain the first data to be evaluated, and perform feature selection according to the feature index subset to obtain second data to be evaluated; S108: Classify the second data to be evaluated on the second sample set to obtain a classification result. 2. The medical data classification method according to claim 1, wherein the first sample set of the obtained medical data is obtained to obtain the first sample attributes, including: Acquire a first sample set of medical data, obtain attributes of the first sample, and perform dispersion standardization processing on the first sample set. 3. The medical data classification method according to claim 1, characterized in that, the weight vector is updated this time by the update mode of gradient descent, and the next weight vector after one iteration is obtained, including: pass the rules Update the current weight vector to obtain the next weight vector w ^t+1 after one iteration, J(w) is the optimization objective function, by maximizing J(w)=(z _i ^t+1 ) ^T w ^{t+ 1} is calculated. 4. The medical data classification method according to claim 1, wherein said obtaining the first data to be evaluated, and performing feature selection according to the feature index subset to obtain the second data to be evaluated comprises: Acquire the first data to be evaluated, perform dispersion standardization processing, and perform feature selection according to the feature index subset to obtain the second data to be evaluated. 5. The medical data classification method according to any one of claims 1 to 4, wherein the second data to be evaluated is classified on the second sample set to obtain a classification result, including: Classify the second data to be evaluated on the second sample set using a K-nearest neighbor classifier to obtain a classification result. 6. A medical data classification device based on local learning feature weight selection, characterized in that it comprises: The first sample set obtaining module is used to obtain the first sample set of medical data and obtain the first sample attributes; The initial weight limit setting module is used to set the initial weight vector of the first sample attribute, and use the initial weight vector as the current weight vector; The next weight vector acquisition module is used to update the current weight vector through gradient descent to obtain the next weight vector after one iteration; A judging module, used to judge whether the determination rule is established, if so, use the next weight vector as the final weight vector, and call the feature index subset acquisition module; if not, use the next weight vector as the current weight vector, call The next weight vector acquisition module; wherein the determination rule is ||w ^t+1 -w ^t ||≤θ, w ^t is the current weight vector, w ^t+1 is the next weight vector, and θ is the stop criterion; The feature index subset acquisition module is configured to perform feature selection according to the final weight vector to obtain a feature index subset; A second sample set acquisition module, configured to perform feature selection on the first sample set according to the feature index subset, to obtain a second sample set after feature selection; The second data to be evaluated acquisition module is used to obtain the first data to be evaluated, and perform feature selection according to the feature index subset to obtain the second data to be evaluated; A classification module, configured to classify the second data to be evaluated on the second sample set to obtain a classification result. 7. The medical data classification device according to claim 6, wherein the first sample set acquisition module is specifically used for: Acquire a first sample set of medical data, obtain attributes of the first sample, and perform dispersion standardization processing on the first sample set. 8. The medical data classification device according to claim 6, wherein the next weight vector acquisition module is specifically used for: pass the rules Update the current weight vector to obtain the next weight vector w ^t+1 after one iteration, J(w) is the optimization objective function, by maximizing J(w)=(z _i ^t+1 ) ^T w ^{t+ 1} is calculated. 9. The medical data classification device according to claim 6, wherein the second data acquisition module to be evaluated is specifically used for: Acquire the first data to be evaluated, perform dispersion standardization processing, and perform feature selection according to the feature index subset to obtain the second data to be evaluated. 10. The medical data classification device according to any one of claims 6 to 9, wherein the classification module is specifically used for: Classify the second data to be evaluated on the second sample set using a K-nearest neighbor classifier to obtain a classification result.