Attribute reduction method of pneumonia CT images based on multi-view neighborhood evidence entropy
Through the multi-view neighborhood evidence entropy method, the problem of insufficient data information fusion and redundant attributes in pneumonia CT image data processing is solved, and more efficient and accurate simplification of pneumonia CT image attributes is achieved, and intelligent auxiliary diagnosis of pneumonia is improved. value.
Patent Information
- Application Number
- CN202310226858.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-07
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2043-03-07
AI Technical Summary
The prior art fails to effectively integrate data information in pneumonia CT image data processing, resulting in low detection efficiency and easy misdiagnosis and misdiagnosis, and redundant pathological attribute data, affecting diagnostic efficiency.
Using a method based on multi-view neighborhood evidence entropy, the neighborhood set of pneumonia CT images is portrayed from two views sparse representation and distance measurement, and redundant attributes are removed by constructing neighborhood evidence entropy to improve detection efficiency and accuracy.
Through the multi-view neighborhood evidence entropy method, the neighborhood of pneumonia CT images is accurately portrayed, errors are reduced, and the efficiency and accuracy of intelligent auxiliary diagnosis of pneumonia are improved, thereby enhancing the ability to analyze pneumonia lesions.
Smart Images

Figure CN116596836B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent medical technology, and in particular to a method for reducing pneumonia CT image attributes based on multi-view neighborhood evidence entropy. Background Art
[0002] Pneumonia is caused by pathogens invading the lung parenchyma and overgrowth in the lung parenchyma beyond the host's defense capacity, resulting in the appearance of exudates in the alveolar cavity and pneumonia. The occurrence and severity of pneumonia are mainly determined by the balance between pathogen factors and host factors. Common symptoms include coughing and expectoration, sometimes with thick or bloody sputum, accompanied by shortness of breath and dyspnea. Symptoms are often atypical, the disease progresses rapidly, and missed and misdiagnoses are prone to occur.
[0003] With the continuous development of computer technology, artificial intelligence is used to realize automatic detection of pneumonia CT images to ensure the correctness of the diagnosis results. At present, scholars at home and abroad have also conducted in-depth research on this and proposed some better detection models. When faced with a wide range of data information, computer technology can effectively process, delete redundant information, and extract effective information, thereby continuously improving detection efficiency. However, the currently proposed method does not effectively fuse the data information, but only considers the connection between the data from a single perspective. Therefore, how to effectively fuse the data information is one of the issues that the present invention needs to consider.
[0004] In addition, the current effective way to judge the condition of pneumonia lesions is to conduct pathological attribute experiments on the etiology and pathogenesis of pneumonia. However, during the experiment, there is too much pathological attribute data. There is an urgent need for a new method that can effectively reduce the redundant attributes in the classification information of pneumonia CT image data, reduce the detection time of pneumonia CT image data and improve the detection efficiency, so as to help doctors effectively analyze the lesions of pneumonia.
[0005] How to solve the above technical problems is the subject faced by the present invention. Summary of the invention
[0006] In order to solve the above problems, the present invention provides a pneumonia CT image attribute simplification method based on multi-view neighborhood evidence entropy, which characterizes the neighborhood set of pneumonia CT images from two views: distance measurement and sparse representation, and introduces neighborhood entropy and evidence theory to construct neighborhood evidence entropy, characterize the importance of each attribute, and remove redundant attributes, which effectively improves the efficiency and accuracy of pneumonia CT image attribute simplification and has strong application value for intelligent auxiliary diagnosis of pneumonia.
[0007] In order to achieve the above objectives, the present invention adopts a technical solution: a pneumonia CT image attribute simplification method based on multi-view neighborhood evidence entropy, comprising the following steps:
[0008] S10, reading the pneumonia CT image dataset and converting the dataset into a four-tuple decision information system;
[0009] S20, using sparse representation and distance measurement to represent the correlation between pneumonia CT image data samples, and characterizing the neighborhood range through two stages of fusion and deletion;
[0010] S30, integrating evidence theory and neighborhood entropy to construct neighborhood evidence entropy, which is used to characterize the attribute importance of pneumonia CT image data;
[0011] S40. Calculate a reduced attribute set in the pneumonia CT image dataset.
[0012] Furthermore, the step S10 includes the following steps:
[0013] S11, read the pneumonia CT image data set, determine its attribute set and decision classification, the decision information system S=<U,C∪D,V,f> , where U = {x1,…,x i ,…,x N} represents the sample set of pneumonia CT image data, N represents the number of pneumonia CT images, x i represents the i-th sample, x N represents the Nth sample; C={a1,…,a i ,…,a n} represents a non-empty finite set of pneumonia CT image data attributes, n represents the number of attributes in pneumonia CT image data, a i represents the i-th attribute, a n represents the nth attribute; D = {d1,…,d i ,…,d m} represents a non-empty finite set of decision attributes of pneumonia CT image data, m represents the number of decision categories in pneumonia CT image data, d i represents the i-th decision category, d m represents the mth decision category; V = ∪ a∈C∪D V a , V a is a possible case of attribute a of pneumonia CT image data; f:U×(C∪D)→V is an information function that assigns an information value to each CT image, that is, x∈U,f(x,a)∈V a ;
[0014] S12, according to the number of different information values of the decision attribute D in the data set, the pneumonia CT image data set S is divided into t data subsets, and the Where S i represents the i-th data subset, S jRepresents the j-th data subset, i≠j, i=1,2,…,t, j=1,2,…,t.
[0015] Furthermore, the step S20 includes the following steps:
[0016] S21. Set the pneumonia CT image dataset to X, the reconstruction weight matrix of X to W, and calculate the sample x to be examined through the sparse constraint function. i and other samples {x1,x2,…,x i-1 ,x i+1 ,…,x N}, the sparse correlation between the samples to be examined is x i The objective function is as follows:
[0017]
[0018] Among them, x i-1 represents the i-1th sample, x i+1 represents the i+1th sample, is the square of the Frobenius norm, F is the norm selection method, ||·||1 is the 1 norm, T is the transpose transformation of the matrix, η and μ are parameters with values between 0 and 1, R(W)=Tr(W T X T LXW) is a locality preserving projection regularization term, where L is the Laplace matrix, which represents the relationship information between attributes, and Tr is the trace of the matrix, i.e., the sum of the diagonal elements of the matrix.
[0019] S22. The weight matrix W reconstructed from the pneumonia CT image dataset shows the sparse correlation between samples. If the sparse correlation is greater than 0, it is considered that the two samples are positively correlated and are placed in the sparse neighborhood set. The number of samples with positive correlation is recorded as k. Sample x i The sparse neighborhood set SN obtained by sparse representation A (x i ) is defined as follows:
[0020] SN A (x i )={x j |x j ∈U,W(x i ,x j )>0}
[0021] Where A represents a subset of attributes in a non-empty finite set C of data attributes, x j represents the jth sample, W(x i ,x j ) represents xi and x j The sparse correlation weights between .
[0022] S23. Calculate the pneumonia CT image sample x i and other samples x j The Euclidean distance dis(x i ,x j ), which is defined as follows:
[0023]
[0024] Among them, f(x,a t ) indicates that sample x has the tth attribute a t The value in .
[0025] S24. Based on the Euclidean distance and k value, the k nearest neighbor concept is adopted, that is, the first k neighbors are selected according to the distance to solve the distance neighborhood set. Sample x i The distance neighborhood set KN obtained by k nearest neighbors A (x i ) is defined as follows:
[0026]
[0027] in, Indicates the distance x i The nth sample of Indicates distance from x i The distance to the nearest nth sample.
[0028] S25, the sparse neighborhood set and the distance neighborhood set of the pneumonia CT image data are fused to obtain a multi-view neighborhood set MN of the pneumonia CT image data set. A (x i ),Right now:
[0029] MN A (x i )={x j |x j ∈SN A (x i )∪x j ∈KN A (x i )}
[0030] S26. Perform a first-order deletion operation on the fused neighborhood set to calculate the sample x in the pneumonia CT image dataset i and the sample x in its neighborhood set j The average distance is denoted as ave. Compare dis(x i ,xj ) and ave, if dis(x i ,x j )>ave, then it is considered that sample x j With sample x i The first-order neighborhood relationship is not satisfied, so the sample x j From x i is deleted from the multi-view neighborhood set. Sample x i The first-order neighborhood set obtained after the first deletion is The calculation formula is as follows:
[0031]
[0032] S27. According to the perpendicular median theorem, a second-order operation is performed on the first-order neighborhood set of samples in the pneumonia CT image dataset, and samples that do not meet the second-order requirements are deleted. The perpendicular bisector is used to determine the sample x in the neighborhood set. j Whether there is a perpendicular bisector depends on the sample x to be examined i Count all samples in the neighborhood set. If the sample x j In the sample x to be examined i If the value is on one side, the value remains unchanged, otherwise the value increases by 1. When the counting is finished, the sample with the largest value is considered to be the same as the sample to be examined x i The two samples do not satisfy the second-order neighborhood relationship and are considered redundant samples. After deleting the redundant samples, the second-order neighborhood set of pneumonia CT image data samples is obtained. According to the perpendicular median theorem, Sample x i The second-order neighborhood set obtained after the second deletion is Deletion is defined as follows:
[0033]
[0034] Among them, count means calculating the sample x according to the perpendicular bisector j Is it consistent with the sample x to be examined? i The value on the same side, n' represents the sample x to be examined i The number of first-order neighborhood sets of Indicates that x is calculated based on the perpendicular bisector i The samples in the first-order neighborhood are not in the sample to be examined x i Maximum value on the same side.
[0035] Furthermore, the step S30 includes the following steps:
[0036] S31. Calculate the pth sample x to be examined in the pneumonia CT image data set pand other qth samples x q The distance dis(x p ,x q ) and sparse representation W(x p ,x q ), respectively represented by dis p,q and spa p,q , according to dis p,q and spa p,q Calculate the pneumonia CT image sample x p The second-order neighborhood of
[0037] S32. Calculate other samples x in the pneumonia CT image dataset q The sample x to be examined p The evidence information provided, Ω=U / D={D1,D2,...,D j} is the decision-making classification set of the pneumonia CT image dataset, where j is the number of decision classifications, D j For the j-th decision class division, if the pneumonia CT image sample x q The category label is l t and decision classes are divided into D t , then (x q ,D t ) can be independently the sample x to be examined p Provide evidence support, including the evidence information The definition is as follows:
[0038]
[0039]
[0040] Among them, D t is the t-th decision classification, w1 and w2 represent the importance weights of distance metric and sparse representation, w1+w2=1. μ is the relevant parameter of distance metric, η is the relevant parameter of sparse representation, and μ>0 and η>0 are satisfied.
[0041] S33. Calculate the pneumonia CT image sample x p The second-order neighborhood of The category label is l t The sample set is Sample Set x p The evidence supporting this can be expressed as Right now:
[0042]
[0043]
[0044] S34. Calculate the pneumonia CT image sample x p Evidence information under different category labels And the evidence information is merged, which can be specifically expressed as:
[0045]
[0046]
[0047] in, j is the number of decision partitions.
[0048] S35. Calculate pneumonia CT image sample x p Divide the decision classes into Ω=U / D={D1,D2,...,D j D in t Trust function of the class The specific definitions are as follows:
[0049]
[0050] S36. Calculate the pneumonia CT image sample x p Neighborhood evidence entropy H(D|A), x p The second-order neighborhood set of [x p ] D is x p is an equivalent class of , then the neighborhood evidence entropy is defined as follows:
[0051]
[0052] Among them, |U| is the number of domains.
[0053] Furthermore, the step S40 includes the following steps:
[0054] S41. Calculate each pathological attribute in the pneumonia CT image dataset The external importance measure sig relative to the reduced set red out (red,a,D)=H(D|red)-H(D|red∪{a}), select the attribute a with the maximum attribute importance k ;
[0055] S42, judge sig out (red,a k ,D) is greater than 0, if sig out (red,a k ,D)>0, then attribute a k Add it to the reduced set red and delete it from the candidate attribute subset C;
[0056] S43. Calculate and compare H(D|red) and H(D|C). If H(D|red)≥H(D|C) and sig out (red,a k ,D)≤0, then attribute a k Remove from the reduced set red until
[0057] S44. For each attribute in the reduced set red Recalculate the internal importance measure of its attributes again, sig in (red,a w ,D)=H(D|red-{a w})-H(D|red), delete attribute a in turn w , compare sig in (red,a w ,D) is greater than 0, if sig in (red,a w ,D)≤0, then attribute a w Delete from the reduced set red. After looping through all the attributes in red, output the reduced set red of the pneumonia CT image dataset.
[0058] Compared with the prior art, the present invention has the following beneficial effects:
[0059] (1) The present invention provides a pneumonia CT image attribute simplification method based on multi-view neighborhood evidence entropy. First, the neighborhood of the pneumonia CT image is characterized from two views, sparse representation and distance measurement, and the neighborhood is characterized more accurately in a multi-view manner. Among them, an adaptive k value is used, k is the number of neighbors of a certain pneumonia CT image data, and there is no need to manually define parameters, which reduces errors to a certain extent.
[0060] (2) The evidence theory is introduced into the neighborhood entropy, and the trust function is put into the neighborhood entropy as the supporting information between samples to judge the uncertainty of the attribute. Finally, a heuristic reduction algorithm is designed based on the neighborhood evidence entropy to select the simplified attributes. This method improves the efficiency and accuracy of the attribute simplification of pneumonia CT image data and improves the application value of intelligent auxiliary diagnosis of pneumonia. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.
[0062] Figure 1It is an overall flow chart of the pneumonia CT image attribute simplification method based on multi-view neighborhood evidence entropy of the present invention.
[0063] Figure 2 This is an overall data processing framework diagram of the pneumonia CT image attribute simplification method based on multi-view neighborhood evidence entropy of the present invention.
[0064] Figure 3 The present invention is an attribute reduction flow chart of the pneumonia CT image attribute reduction method based on multi-view neighborhood evidence entropy. DETAILED DESCRIPTION
[0065] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.
[0066] Example 1
[0067] See also Figures 1 to 3 This embodiment provides a method for reducing pneumonia CT image attributes based on multi-view neighborhood evidence entropy, including the following steps:
[0068] S10 reads the pneumonia CT image dataset and converts the dataset into a four-tuple decision information system;
[0069] S20 represents the correlation between pneumonia CT image data samples through two views of sparse representation and distance measurement, and characterizes the neighborhood range through two stages of fusion and deletion;
[0070] S30 integrates evidence theory and neighborhood entropy to construct neighborhood evidence entropy, which is used to characterize the attribute importance of pneumonia CT image data;
[0071] S40 computes a reduced set of attributes in a pneumonia CT image dataset.
[0072] The step S10 comprises the following steps:
[0073] S11 reads the pneumonia CT image data set, determines its attribute set and decision classification, and the decision information system S=<U,C∪D,V,f> , where U = {x1,…,x i ,…,x N} represents the sample set of pneumonia CT image data, N represents the number of pneumonia CT images, x i represents the i-th sample, x N represents the Nth sample; C={a1,…,a i ,…,a n} represents a non-empty finite set of pneumonia CT image data attributes, n represents the number of attributes in pneumonia CT image data, ai represents the i-th attribute, a n represents the nth attribute; D = {d1,…,d i ,…,d m} represents a non-empty finite set of decision attributes of pneumonia CT image data, m represents the number of decision categories in pneumonia CT image data, d i represents the i-th decision category, d m represents the mth decision category; V = ∪ a∈C∪D V a , V a is a possible case of attribute a of pneumonia CT image data; f:U×(C∪D)→V is an information function that assigns an information value to each CT image, that is, x∈U,f(x,a)∈V a ;
[0074] S12 divides the pneumonia CT image dataset S into t data subsets according to the number of different information values of the decision attribute D in the dataset, and satisfies Where S i represents the i-th data subset, S j Represents the j-th data subset, i≠j, i=1,2,…,t, j=1,2,…,t.
[0075] The original data set is converted into a four-tuple decision information system as follows:
[0076]
[0077] Where U = {x1, x2, …, x 10} indicates that there are 10 samples in total in this embodiment, and C = {a1, a2, ..., a5} represents the data attributes of pneumonia CT images, which are respectively represented as respiratory rate, white blood cells, platelets, blood pressure and age, and normalization operations are performed on them.
[0078] The step S20 comprises the following steps:
[0079] S21 sets the pneumonia CT image dataset to X, the reconstruction weight matrix of X is W, and the sample to be examined x is calculated by the sparse constraint function. i and other samples {x1,x2,…,x i-1 ,x i+1 ,…,x N}, the sparse correlation between the samples to be examined is x i The objective function is as follows:
[0080]
[0081] Among them, x i-1represents the i-1th sample, x i+1 represents the i+1th sample, is the square of the Frobenius norm, F is the norm selection method, ||·||1 is the 1 norm, T is the transpose transformation of the matrix, η and μ are parameters with values between 0 and 1, R(W)=Tr(W T X T LXW) is a locality preserving projection regularization term, where L is the Laplace matrix, which represents the relationship information between attributes, and Tr is the trace of the matrix, i.e., the sum of the diagonal elements of the matrix.
[0082] S22 reconstructs the weight matrix W based on the pneumonia CT image dataset, and it can be seen that the sparse correlation between samples. If the sparse correlation is greater than 0, it is considered that the two samples are positively correlated, and they are placed in the sparse neighborhood set, and the number of samples with positive correlation is recorded as k. Sample x i The sparse neighborhood set SN obtained by sparse representation A (x i ) is defined as follows:
[0083] SN A (x i )={x j |x j ∈U,W(x i ,x j )>0}
[0084] Where A represents a subset of attributes in a non-empty finite set C of data attributes, x j represents the jth sample, W(x i ,x j ) represents x i and x j According to the sparse representation, the reconstruction weight matrix W can be obtained, which is expressed as the correlation between pneumonia CT image samples. For the convenience of understanding, the sample x1 is described here, and the sparse correlation matrix W of the sample x1 is expressed as follows:
[0085] W=[-0.3063 0 0.5823 0 0.1555 -0.0630 -0.6656 0.5211 0.8318]
[0086] It can be seen that sample x1 is different from samples x4, x6, x9 and x 10 There is a positive correlation between them, so the sparse neighborhood set of x1 obtained by the correlation is {x1,x4,x6,x9,x 10}.
[0087] S23 calculates pneumonia CT image sample x i and other samples x j The Euclidean distance dis(x i ,x j ), which is defined as follows:
[0088]
[0089] Among them, f(x,a t ) indicates that sample x has the tth attribute a t S24 adopts the k-nearest neighbor concept based on the Euclidean distance and k value, that is, the first k neighbors are selected according to the distance to solve the distance neighborhood set. Sample x i The distance neighborhood set KN obtained by k nearest neighbors A (x i ) is defined as follows:
[0090]
[0091] in, Indicates the distance x i The nth sample of Indicates distance from x i The distance to the nearest nth sample. From S22, it can be concluded that there is a positive correlation between sample x1 and the other 4 samples, so k = 4. The distance between sample x1 and the other 9 samples is expressed as follows:
[0092] dis=[0.6765 0.3453 0.2593 1.3251 1.0378 1.1568 2.0710 2.07142.0648]
[0093] Therefore, the distance neighborhood set of x1 obtained by k nearest neighbors is {x1,x2,x3,x4,x6}.
[0094] S25 fuses the sparse neighborhood set and the distance neighborhood set of the pneumonia CT image data to obtain a multi-view neighborhood set MN of the pneumonia CT image data set. A (x i ),Right now:
[0095] MN A (x i )={x j |x j ∈SN A (x i )∪x j ∈KN A (x i )}
[0096] After fusion, the multi-view neighborhood set of x1 is {x1,x2,x3,x4,x6,x9,x 10}.
[0097] S26 performs a first-order deletion operation on the fused neighborhood set to calculate the sample x in the pneumonia CT image dataset i and the sample x in its neighborhood set j The average distance is denoted as ave. Compare dis(x i ,x j ) and ave, if dis(x i ,x j )>ave, then it is considered that sample x j With sample x i The first-order neighborhood relationship is not satisfied, so the sample x j From x i is deleted from the multi-view neighborhood set. Sample x i The first-order neighborhood set obtained after the first deletion is The calculation formula is as follows:
[0098]
[0099] Calculate the distance between x1 and the samples in the fused neighborhood set respectively, and find the average distance ave = 1.0668. 10 The distance of exceeds ave, so it is deleted from the neighborhood set. The first-order neighborhood set after the first deletion is {x1,x2,x3,x4,x6}.
[0100] S27 performs a second-order operation on the first-order neighborhood set of samples in the pneumonia CT image dataset according to the perpendicular median theorem, and deletes samples that do not meet the second-order requirements. Use the perpendicular bisector to determine the sample x in the neighborhood set. j Whether there is a perpendicular bisector depends on the sample x to be examined i Count all samples in the neighborhood set. If the sample x j In the sample x to be examined i If the value is on one side, the value remains unchanged, otherwise the value increases by 1. When the counting is finished, the sample with the largest value is considered to be the same as the sample to be examined x i The two samples do not satisfy the second-order neighborhood relationship and are considered redundant samples. After deleting the redundant samples, the second-order neighborhood set of pneumonia CT image data samples is obtained. According to the perpendicular median theorem, Sample x i The second-order neighborhood set obtained after the second deletion is Deletion is defined as follows:
[0101]
[0102] Among them, count means calculating the sample x according to the perpendicular bisector j Is it consistent with the sample x to be examined? i The value on the same side, n' represents the sample x to be examined i The number of first-order neighborhood sets of Indicates that x is calculated based on the perpendicular bisector i The samples in the first-order neighborhood are not in the sample to be examined x i The maximum value on the same side. According to the perpendicular median theorem, after analysis, for sample x1, the count values of its neighbors x2, x3, x4, and x6 are 3, 2, 2, and 3 respectively, so the maximum count value is 3, and the corresponding samples are x2 and x6, which are deleted from the neighborhood set. Finally, after the second deletion, the second-order neighborhood set is {x1, x3, x4}.
[0103] The step S30 includes the following steps: S31 calculates the pth sample x to be examined in the pneumonia CT image data set: p and other qth samples x q The distance dis(x p ,x q ) and sparse representation W(x p ,x q ), respectively represented by dis p,q and spa p,q , according to dis p,q and spa p,q Calculate the pneumonia CT image sample x p The second-order neighborhood of
[0104] S32 calculates other samples x in the pneumonia CT image dataset q The sample x to be examined p The evidence information provided, Ω=U / D={D1,D2,...,D j} is the decision-making classification set of the pneumonia CT image dataset, where j is the number of decision classifications, D j For the j-th decision class division, if the pneumonia CT image sample x q The category label is l t and decision classes are divided into D t , then (x q ,D t ) can be independently the sample x to be examined p Provide evidence support, including the evidence information The definition is as follows:
[0105]
[0106]
[0107] Among them, D t is the t-th decision classification, w1 and w2 represent the importance weights of distance metric and sparse representation, w1+w2=1. μ is the relevant parameter of distance metric, η is the relevant parameter of sparse representation, and μ>0 and η>0 are satisfied.
[0108] In this embodiment, w1=0.4, w2=0.6, and the evidence information between the pneumonia CT image sample x1 to be investigated and its neighboring sample x3 can be calculated.
[0109] S33 calculates pneumonia CT image sample x p The second-order neighborhood of The category label is l t The sample set is Sample Set x p The evidence supporting this can be expressed as Right now:
[0110]
[0111]
[0112] Calculate the second-order neighborhood of the pneumonia CT image sample x1 to be examined The sample set with category label l1 The sample set Evidence for x1
[0113] S34 calculates pneumonia CT image sample x p Evidence information under different category labels And the evidence information is merged, which can be specifically expressed as:
[0114]
[0115]
[0116] in, j is the number of decision divisions. Calculate the evidence information of the pneumonia CT image sample x1 under different category labels
[0117] S35 calculates pneumonia CT image sample x p Divide the decision classes into Ω=U / D={D1,D2,...,D j D in t Trust function of the class The specific definitions are as follows:
[0118]
[0119] Calculate the trust function of the pneumonia CT image sample x1 to be investigated for D1
[0120] S36 Calculate pneumonia CT image sample x p Neighborhood evidence entropy H(D|A), x p The second-order neighborhood set of [x p ]D is x p is an equivalent class of , then the neighborhood evidence entropy is defined as follows:
[0121]
[0122] Among them, |U| is the number of domains.
[0123] The step S40 includes the following steps: S41 calculates each pathological attribute in the pneumonia CT image data set: The external importance measure sig relative to the reduced set red out (red,a,D)=H(D|red)-H(D|red∪{a}), select the attribute a with the maximum attribute importance k ;
[0124] S42 judge sig out (red,a k ,D) is greater than 0, if sig out (red,a k ,D)>0, then attribute a k Add it to the reduced set red and delete it from the candidate attribute subset C;
[0125] S43 calculates and compares H(D|red) and H(D|C), if H(D|red)≥H(D|C) and sig out (red,a k ,D)≤0, then attribute a k Remove from the reduced set red until
[0126] S44 For each attribute in the reduced set red Recalculate the internal importance measure of its attributes again, sig in (red,a w ,D)=H(D|red-{a w})-H(D|red), delete attribute a in turnw , compare sig in (red,a w ,D) is greater than 0, if sig in (red,a w ,D)≤0, then attribute a w Delete from the reduced set red. After looping through all the attributes in red, output the reduced set red of the pneumonia CT image dataset.
[0127] The above description is only an exemplary embodiment of the present invention, and does not limit the scope of patent protection of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the scope of patent protection of the present invention.
Claims
1. A method for reducing the attributes of pneumonia CT images based on multi-view neighborhood evidence entropy, characterized in that: The steps include: S10, reading the pneumonia CT image dataset and converting the dataset into a four-tuple decision information system; S20, using sparse representation and distance measurement to represent the correlation between pneumonia CT image data samples, and characterizing the neighborhood range through two stages of fusion and deletion; S30, integrating evidence theory and neighborhood entropy to construct neighborhood evidence entropy, which is used to characterize the attribute importance of pneumonia CT image data; The step S30 includes the following steps: S31. Calculate the pth sample x to be examined in the pneumonia CT image data set p and other qth samples x q The distance between dis(x p ,x q ) and sparse representation W(x p ,x q ), respectively represented by dis p,q and spa p,q , according to dis p,q and spa p,q Calculate the pneumonia CT image sample x p The second-order neighborhood of S32. Calculate other samples x in the pneumonia CT image dataset q The sample x to be examined p The evidence information provided, Ω=U / D={D1,D2,...,D j } is the decision-making classification set of the pneumonia CT image dataset, where j is the number of decision classifications, D j For the j-th decision class division, if the pneumonia CT image sample x q The category label is l t and decision classes are divided into D t , then (x q ,D t ) independently for the sample x to be examined p Provide evidence support, including the evidence information The definition is as follows: Among them, D t is the t-th decision classification, w1 and w2 represent the importance weights of distance metric and sparse representation, w1+w2=1, α is the relevant parameter of distance metric, β is the relevant parameter of sparse representation, and α>0 and β>0 are satisfied; S33. Calculate the pneumonia CT image sample x p The second-order neighborhood of The category label is l t The sample set is Sample Set x p The evidence supports that Right now: S34. Calculate the pneumonia CT image sample x p Evidence information under different category labels And the evidence information is merged, which is specifically expressed as follows: in, j is the number of decision partitions; S35. Calculate pneumonia CT image sample x p Divide the decision classes into Ω=U / D={D1,D2,...,D j D in t Trust function of the class The specific definitions are as follows: S36. Calculate the pneumonia CT image sample x p Neighborhood evidence entropy H(D|A), x p The second-order neighborhood set of [x p ] D is x p is an equivalent class of , then the neighborhood evidence entropy is defined as follows: Among them, |U| is the number of domains; S40, calculating a reduced attribute set in a pneumonia CT image dataset; The step S40 comprises the following steps: S41. Calculate each pathological attribute in the pneumonia CT image dataset The external importance measure sig relative to the reduced set red out (red,a,D)=H(D|red)-H(D|red∪{a}), select the attribute a with the maximum attribute importance k ; S42, judge sig out (red,a k ,D) is greater than 0, if sig out (red,a k ,D)>0, then attribute a k Add it to the reduced set red and delete it from the candidate attribute subset C; S43. Calculate and compare H(D|red) and H(D|C). If H(D|red)≥H(D|C) and sig out (red,a k ,D)≤0, then attribute a k Remove from the reduced set red until S44. For each attribute in the reduced set red Recalculate the internal importance measure of its attributes again, sig in (red,a w ,D)=H(D|red-{a w })-H(D|red), delete attribute a in turn w , compare sig in (red,a w ,D) is greater than 0, if sig in (red,a w ,D)≤0, then attribute a w Delete it from the reduced set red, loop through all the attributes in red one by one, and output the reduced set red of the pneumonia CT image dataset.
2. The pneumonia CT image attribute reduction method based on multi-view neighborhood evidence entropy according to claim 1 is characterized in that: The step S10 comprises the following steps: S11, read the pneumonia CT image data set, determine its attribute set and decision classification, the decision information system S=<U,C∪D,V,f> , where U = {x1,…,x i ,…,x N } represents the sample set of pneumonia CT image data, N represents the number of pneumonia CT images, x i represents the i-th sample, x N represents the Nth sample; C={a1,…,a i ,…,a n } represents a non-empty finite set of pneumonia CT image data attributes, n represents the number of attributes in pneumonia CT image data, a i represents the i-th attribute, a n represents the nth attribute; D = {d1,…,d i ,…,d m } represents a non-empty finite set of decision attributes of pneumonia CT image data, m represents the number of decision categories in pneumonia CT image data, d i represents the i-th decision category, d m represents the mth decision category; V = ∪ a∈C∪D V a , V a is a possible case of attribute a of pneumonia CT image data; f:U×(C∪D)→V is an information function that assigns an information value to each CT image, that is, x∈U,f(x,a)∈V a ; S12, according to the number of different information values of the decision attribute D in the data set, the pneumonia CT image data set S is divided into t data subsets, and the Where S i represents the i-th data subset, S j Represents the j-th data subset, i≠j, i=1,2,…,t, j=1,2,…,t.
3. The pneumonia CT image attribute reduction method based on multi-view neighborhood evidence entropy according to claim 1 is characterized in that: The step S20 comprises the following steps: S21. Set the pneumonia CT image dataset to X, the reconstruction weight matrix of X to W, and calculate the sample x to be examined through the sparse constraint function. i and other samples {x1,x2,…,x i-1 ,x i+1 ,…,x N }, the sparse correlation between the samples to be examined is x i The objective function is as follows: Among them, x i-1 represents the i-1th sample, x i+1 represents the i+1th sample, is the square of the Frobenius norm, F is the norm selection method, ||·||1 is the 1 norm, T is the transpose transformation of the matrix, η and μ are parameters with values between 0 and 1, R(W)=Tr(W T X T LXW) is the locality preserving projection regularization term, where L is the Laplace matrix, which represents the relationship information between attributes, and Tr is the trace of the matrix, i.e., the sum of the diagonal elements of the matrix; S22. The weight matrix W reconstructed from the pneumonia CT image dataset is used to find the sparse correlation between samples. If the sparse correlation is greater than 0, it is considered that the two samples are positively correlated. They are placed in the sparse neighborhood set and the number of samples with positive correlation is recorded as k. Sample x i The sparse neighborhood set SN obtained by sparse representation A (x i ) is defined as follows: SN A (x i )={x j |x j ∈U,W(x i ,x j )>0} Where A represents a subset of attributes in a non-empty finite set C of data attributes, x j represents the jth sample, W(x i ,x j ) represents x i and x j The sparse correlation weight value between ; S23. Calculate the pneumonia CT image sample x i and other samples x j The Euclidean distance dis(x i ,x j ), which is defined as follows: Among them, f(x,a t ) indicates that sample x has the tth attribute a t The value in ; S24, according to the Euclidean distance and k value, the k nearest neighbor idea is adopted, that is, the first k neighbors are selected according to the distance, and the distance neighborhood set is solved. Sample x i The distance neighborhood set KN obtained by k nearest neighbors A (x i ) is defined as follows: in, Indicates the distance x i The nth sample of Indicates distance from x i The distance to the nearest nth sample; S25, fusing the sparse neighborhood set and the distance neighborhood set of the pneumonia CT image data to obtain a multi-view neighborhood set MN of the pneumonia CT image data set A (x i ),Right now: MN A (x i )={x j |x j ∈SN A (x i )∪x j ∈KN A (x i )} S26. Perform a first-order deletion operation on the fused neighborhood set to calculate the sample x in the pneumonia CT image dataset i and the sample x in its neighborhood set j The average distance, denoted as ave, is compared with dis(x i ,x j ) and ave, if dis(x i ,x j )>ave, then it is considered that sample x j With sample x i The first-order neighborhood relationship is not satisfied, so the sample x j From x i In the multi-view neighborhood set, Sample x i The first-order neighborhood set obtained after the first deletion is The calculation formula is as follows: S27. According to the perpendicular median theorem, a second-order operation is performed on the first-order neighborhood set of samples in the pneumonia CT image dataset, samples that do not meet the second-order requirements are deleted, and the perpendicular bisector is used to determine the sample x in the neighborhood set. j Whether there is a perpendicular bisector depends on the sample x to be examined i On one side, count all samples in the neighborhood set. If the sample x j In the sample x to be examined i If the value is on one side, the value remains unchanged, otherwise the value increases by 1, and the counting ends. It is considered that the sample with the largest value is the same as the sample to be examined x i The second-order neighborhood relationship between them does not meet the requirements and is regarded as a redundant sample. After deleting the redundant samples, the second-order neighborhood set of the pneumonia CT image data samples is obtained. According to the perpendicular median theorem, Sample x i The second-order neighborhood set obtained after the second deletion is Deletion is defined as follows: Among them, count j Indicates that the sample x is calculated according to the perpendicular bisector j Is it consistent with the sample x to be examined? i The value on the same side, n' represents the sample x to be examined i The number of first-order neighborhood sets of Indicates that x is calculated based on the perpendicular bisector i The samples in the first-order neighborhood are not in the sample to be examined x i Maximum value on the same side.
