A brain tumor image feature extraction method and system

Abstract

The application discloses a brain tumor image feature extraction method and system, electronic equipment and computer readable storage medium, and belongs to the technical field of brain tumor image feature extraction; the deep belief network composed of three consecutive Boltzmann machines effectively extracts the depth features of the image; then the dimensionality reduction is performed through the local linear embedding, and under the premise of reducing the redundant features, the sufficient expression of the image is effectively ensured, and the space is saved. The problems that the brain tumor image features are difficult to extract, and the image features extracted by the traditional convolutional neural network (CNN) are accompanied by redundant information or noise are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of brain tumor image feature extraction, in particular to a brain tumor image feature extraction method and system. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] Unlike other medical images, brain tumor medical diagnosis mainly relies on the accumulation of personal knowledge and the judgment of practical experience of doctors, and this dependence has great potential problems, such as too large time cost, or doctors are prone to over fatigue under the condition of high-intensity repetitive work, which may cause misdiagnosis and missed diagnosis. Therefore, we choose to assist doctors in analyzing medical images by computer.

[0004] In addition, due to the existence of multiple regions of brain tumors, complex tumor morphology, blurred edges of each region, insufficient image data, low MRI image resolution and other problems, it brings many challenges to image feature extraction work.

[0005] In recent years, the performance of convolutional neural network (CNN) in brain tumor image analysis task is very competitive, and CNN has great advantages in extracting local features of images. However, on the one hand, the limitation of convolution operation, it cannot well handle long-distance relationship; on the other hand, after convolutional neural network, the dimension of the extracted image features is still large, which may have redundant information, even with noise interference classification. SUMMARY

[0006] In order to solve the problems of the prior art, the present application provides a brain tumor image feature extraction method, system, electronic device and computer readable storage medium, which effectively extracts the deep features of the image by the deep belief network composed of three consecutive Boltzmann machines; then it is reduced by local linear embedding, which effectively ensures the sufficient expression of the image under the premise of reducing redundant features, and saves the space.

[0007] In the first aspect, the present application provides a brain tumor image feature extraction method;

[0008] A brain tumor image feature extraction method comprises:

[0009] Obtaining a brain tumor image;

[0010] Inputting the brain tumor image into a deep belief network for deep feature extraction to obtain primary deep features;

[0011] Reducing the dimension of the primary deep features by local linear embedding algorithm.

[0012] By adopting the technical scheme, the deep belief network is used to extract the deep image features of the brain tumor image, the final data features can be obtained by gradually learning the network features through the hidden layer, the local linear embedding is used to perform dimension reduction operation on the previously extracted deep features of the brain tumor image for the first time, the key features of the image are maximally reserved, the redundant features are removed, the space cost is saved, and the operation efficiency of the downstream task is improved.

[0013] In a second aspect, the present application provides a brain tumor image feature extraction system.

[0014] The brain tumor image feature extraction system comprises:

[0015] A brain tumor image acquisition module is configured to acquire a brain tumor image.

[0016] A primary deep feature acquisition module is configured to input the brain tumor image into a deep belief network to extract deep features and acquire primary deep features.

[0017] A dimension reduction module is configured to perform dimension reduction on the primary deep features by using a local linear embedding algorithm.

[0018] In a third aspect, the present application provides an electronic device.

[0019] The electronic device comprises a memory and a processor, and computer instructions stored in the memory and running on the processor, when the computer instructions are run by the processor, the steps of the brain tumor image feature extraction method are completed.

[0020] In a fourth aspect, the present application provides a computer readable storage medium.

[0021] The computer readable storage medium is configured to store computer instructions, when the computer instructions are executed by a processor, the steps of the brain tumor image feature extraction method are completed.

[0022] Compared with the prior art, the present application has the following beneficial effects:

[0023] 1. The present application innovatively uses a deep belief network (DBN) to extract deep image features of a brain tumor, the method has strong unsupervised learning ability, and the final data features can be obtained by gradually learning the network features through the hidden layer;

[0024] 2. The present application uses local linear embedding to perform dimension reduction operation on the previously extracted deep features of the brain tumor image for the first time, eliminates redundant features, saves space, improves the operation efficiency of the downstream task, and realizes more effective classification. BRIEF DESCRIPTION OF DRAWINGS

[0025] The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the application, and are incorporated in and constitute a part of this specification, the embodiments of the application, and their description, are to explain the application without imposing any undue limitations.

[0026] Figure 1 A flowchart provided for the embodiments of the application;

[0027] Figure 2 A structure diagram provided for the application of the brain tumor image feature extraction method to the brain tumor image classification task. DETAILED DESCRIPTION

[0028] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0029] It should be noted that the terms used herein are only intended to describe specific embodiments, and are not intended to limit the exemplary embodiments according to the application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should be understood that the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0030] The embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0031] First, some technical terms related to the specific implementation of the technical solutions of the application are explained:

[0032] Brain tumor image: medical image used by doctors to diagnose brain tumors.

[0033] In the prior art, on the one hand, brain tumors have multiple regions, complex tumor shapes, blurred edges of each region, insufficient image data, low MRI image resolution, etc., which bring many challenges to brain tumor image feature extraction; on the other hand, there are problems of redundant information and noise interference in the process of brain tumor image feature extraction; therefore, the present application provides a brain tumor image feature extraction method capable of extracting more effective brain tumor image features.

[0034] A brain tumor image feature extraction method, comprising:

[0035] Obtaining a brain tumor image;

[0036] inputting the brain tumor image into a deep belief network to perform deep feature extraction and obtain primary deep features;

[0037] performing dimension reduction on the primary deep features by using a local linear embedding algorithm.

[0038] Further, the deep belief network is formed by three Boltzmann machines RBM stacked in series.

[0039] Further, a local reconstruction weight matrix is obtained according to the primary deep feature vector.

[0040] According to the local reconstruction weight matrix and the primary deep feature vector, a dimension-reduced primary deep feature vector is obtained.

[0041] Further, a local reconstruction weight matrix is obtained according to the primary deep feature vector.

[0042] According to the local reconstruction weight matrix and the primary deep feature vector, a dimension-reduced primary deep feature vector is obtained.

[0043] Further, the nearest neighbor of each primary deep feature vector is obtained according to the primary deep feature vector.

[0044] According to the primary deep feature vector and the nearest neighbor, a reconstruction error is obtained.

[0045] According to the nearest neighbor and the reconstruction error, a local reconstruction weight matrix is obtained.

[0046] Further, the method further comprises: inputting the dimension-reduced primary deep features into a brain tumor image classifier to obtain a brain tumor image recognition result.

[0047] Further, according to the dimension-reduced primary deep feature vector and a parameter set of the brain tumor image classifier, a probability that the brain tumor image contains a tumor is obtained.

[0048] Further, the method further comprises:

[0049] According to the number of brain tumor images and the brain tumor image recognition result, a classification loss of the brain tumor classifier is obtained.

[0050] The parameter set of the brain tumor classifier is optimized by minimizing the classification loss.

[0051] Next, the brain tumor image feature extraction method disclosed in the embodiment will be described in detail. Figures 1-2 The brain tumor image feature extraction method disclosed in the embodiment will be described in detail.

[0052] Embodiment One

[0053] The embodiment provides a brain tumor image feature extraction method.

[0054] A brain tumor image feature extraction method, comprising:

[0055] S1, acquiring a brain tumor image.

[0056] S2, inputting the brain tumor image into a deep belief network for deep feature extraction to obtain primary deep features;

[0057] Since the DBN has strong unsupervised learning ability, it is selected to extract the primary deep features of the brain tumor image. The DBN is a deep neural network realized by stacking multiple Boltzmann machines (RBM).

[0058] First, the basic principle of RBM is introduced. The RBM constructs the DBN through a greedy hierarchical training model. The RBM is a two-layer network including a visible layer = {0,1} D and a hidden layer = {0,1} K , as shown in Figure 2 . The visible layer and the hidden layer both contain multiple units, and the input data is generally input to the visible layer v, which is defined as:

[0059]

[0060] where w ij is the weight associated with the visible unit i and the hidden unit j, c i is the bias term of the visible unit i, b j is the bias term of the hidden unit j, the hidden layer is represented as h, and the visible layer is represented as v. In the DBN, the weight is an important structural parameter.

[0061] The joint distribution of the layer is:

[0062]

[0063] where Z is a standardization constant.

[0064] The conditional distribution of the hidden layer h and the visible layer v is generated as follows:

[0065]

[0066]

[0067]

[0068] By setting each v i in to 1, the visible layer can be reconstructed. The contrast divergence method is used to update the weight. The change of a weight is defined as:

[0069] Δw ij = ε(vi h jdata -v i h jrec )

[0070] where ε is the learning rate. Each RBM has a learning rate, and proper weights can be generated by proper learning rate.

[0071] Generally, only one RBM cannot learn the characteristics well. Therefore, it is necessary to stack multiple RBMs to extract features. The features extracted by the first RBM are input to the next RBM. When inputting data, we can realize a more abstract representation of data or features, and extract effective deep features through DBN.

[0072] In this embodiment, the deep belief network is stacked by three RBMs. The primary deep feature vector X = {X1, X2,..., XN} is extracted from the low-level features of the brain tumor image by the three RBMs. N}, X i ∈R D , i = 1, 2,..., N.

[0073] S3, dimension reduction is performed on the primary deep features by a local linear embedding algorithm. Specifically, the following steps are included.

[0074] S301, find K adjacent points of each sample point X i in the high-dimensional space, and calculate the Euclidean distance:

[0075] d ij = ||X i -X j ||

[0076] where the value of K can be calculated through multiple experiments, or can be calculated by using the calc_k() function of R language.

[0077] S302, calculate the local reconstruction weight matrix of the sample point from the adjacent points; specifically, a function for measuring reconstruction error is defined:

[0078]

[0079]

[0080] where N is the number of sample points, X ij is the i-th sample point represented by the j-th adjacent point, and W ij is the coefficient (i.e., weight) of the i-th sample point represented by the j-th adjacent point.

[0081] When a data point does not belong to the adjacent points of the reconstructed data point (i.e., it is not the nearest sample point to the reconstructed data point), Wij = 0. is the result of reconstructing the i-th sample point, and the difference between the i-th sample point and the reconstructed result is the reconstruction error, and all the coefficients W ij Finally, the local reconstruction weight matrix W is formed.

[0082] S303, mapping all sample points X i from the high-dimensional space to the low-dimensional space Y i vector, so that X ij is mapped to Y ij ; wherein Y = {Y1, Y2,..., Y N}, Y j ∈ R d , j = 1, 2,..., N, d << D, Y = {Y1, Y2,..., Y N} is the mapping output of the d-th dimensional space.

[0083] Specifically, define the cost function:

[0084]

[0085] s.t. Y Y T = N * I

[0086] M = (I - W) (I - W) T ,

[0087] wherein I is the unit matrix.

[0088] Find the value Y that minimizes the cost function, thereby generating the low-dimensional output vector we need. Finding the value of the cost function is equivalent to finding the smallest d eigenvalues of the M matrix. The first eigenvalue should be rounded off because it is zero. Finally, the eigenvectors corresponding to the 2nd ~ d+1 eigenvalues of M are the output results after dimensionality reduction, that is, the features of the brain tumor image after dimensionality reduction.

[0089] S4, input the primary deep features after dimensionality reduction into the brain tumor image classifier to obtain the brain tumor image recognition result.

[0090] Specifically, input the features Y after dimensionality reduction into the brain tumor image classifier, with the purpose of preparing to identify the brain tumor image. An image p j is output by the brain tumor image classifier to represent the probability that the image is a tumor, which is defined as follows:

[0091]

[0092] wherein θ d represents the parameter set of the brain tumor image classifier, and y j represents the image pj where D denotes a mapping function of the classifier.

[0093] As shown in Table 1, the inventors conducted experiments on the Kaggle brain tumor dataset, and used the brain tumor image feature extraction method of the present application for the brain tumor image classification task. The experimental results are better than other existing methods, which shows that the method can accurately capture more effective brain tumor image features.

[0094] Table 1 Experimental results of different methods on the Kaggle brain tumor dataset

[0095]

[0096] In addition, in order to optimize the parameter set of the brain tumor image classifier, the following operations are performed:

[0097] According to the number of brain tumor images and the brain tumor image recognition result, the classification loss of the brain tumor classifier is obtained. Specifically, a cross-entropy loss function is used, and the formula is as follows:

[0098]

[0099] Where N represents the number of images, and y e {0, 1} represents whether it is a brain tumor.

[0100] Minimizing the classification loss optimizes the parameter set of the brain tumor classifier. Specifically, the following definition is made:

[0101]

[0102] The present application provides a brain tumor image feature extraction method, which effectively extracts the deep features of the image through a deep belief network composed of three consecutive Boltzmann machines; then it is reduced through local linear embedding. Under the premise of reducing redundant features, it effectively ensures sufficient expression of the image, saves space, and improves the operation efficiency of downstream tasks.

[0103] Embodiment Two

[0104] The present embodiment discloses a brain tumor image feature extraction system, comprising:

[0105] A brain tumor image acquisition module is configured to acquire brain tumor images.

[0106] A primary deep feature acquisition module is configured to input the brain tumor images into a deep belief network to extract deep features and acquire primary deep features.

[0107] A dimension reduction module is configured to reduce the dimension of the primary deep features through a local linear embedding algorithm.

[0108] It should be noted that the brain tumor image acquisition module, the primary deep feature acquisition module and the dimension reduction module correspond to the steps in Embodiment 1, and the modules and the examples and application scenarios realized by the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the modules as part of the system can be executed in a computer system such as a set of computer executable instructions.

[0109] Embodiment 3

[0110] Embodiment 3 of the present application provides an electronic device, comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, when the computer instructions are executed by the processor, the steps of the brain tumor image feature extraction method are completed.

[0111] Embodiment 4

[0112] Embodiment 4 of the present application provides a computer readable storage medium for storing computer instructions, which are executed by a processor to complete the steps of the brain tumor image feature extraction method.

[0113] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows or blocks.

[0114] These computer program instructions can also be stored in a computer readable memory capable of guiding the computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable memory produce a product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows or blocks.

[0115] These computer program instructions can also be loaded into a computer or other programmable data processing device to perform a series of operation steps to produce a computer implemented process, so that the instructions executed on the computer or other programmable device provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1steps of the functions specified in the one or more blocks.

[0116] The above description of the various embodiments can be focused on one embodiment or another, and the description of a certain embodiment can be not detailed in another embodiment. Please refer to the relevant description of other embodiments.

[0117] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Those skilled in the art can make various modifications and changes to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A brain tumor image feature extraction method, characterized by, The method comprises the following steps: obtaining a brain tumor image; inputting the brain tumor image into a deep belief network to extract deep features and obtain primary deep features; the deep belief network is formed by three stacked RBMs; dimensionality reduction is performed on the primary deep features by using a locally linear embedding algorithm; a local reconstruction weight matrix is obtained according to the primary deep feature vectors; a dimension-reduced primary deep feature vector is obtained according to the local reconstruction weight matrix and the primary deep feature vector, specifically as follows: find K adjacent points of each sample point Xi in a high-dimensional space and calculate the Euclidean distance; a local reconstruction weight matrix of the sample point is calculated from the adjacent points, and a reconstruction error function is measured as follows: where N is the number of sample points, X ij is the i-th sample point represented using the j-th neighboring point, W ij is the coefficient of the i-th sample point represented using the j-th neighboring point. W ij = 0, is the result of reconstructing the i-th sample point, the difference between the i-th sample point and the reconstructed i-th sample point is the reconstruction error, and all the coefficients W ij Finally, the local reconstruction weight matrix W is formed. all sample points X from a high dimensional space i are mapped to a low dimensional space Y i vector, such that X ij is mapped to Y ij ; where Y = {Y1, Y2,..., Y N}, Y j ∈R d , j = 1, 2,... N, d << D, Y = {Y1, Y2,..., Y N} is the mapping output of the dthdimensional space; a cost function is defined as follows: M = (I - W)(I - W) T where I is an identity matrix; find the value Y that minimizes the cost function to obtain a low-dimensional output vector; input the dimension-reduced primary deep features into a brain tumor image classifier to obtain a brain tumor image recognition result; obtain a classification loss of the brain tumor classifier according to the number of brain tumor images and the brain tumor image recognition result; specifically, a cross-entropy loss function is used: wherein N represents the number of images, indicates whether it is a brain tumor or not; minimize the classification loss to optimize the parameter set of the brain tumor classifier, specifically defined as follows: wherein, denotes a parameter set of a brain tumor image classifier.

2. The brain tumor image feature extraction method according to claim 1, wherein obtain the nearest neighbor of each primary deep feature vector according to the primary deep feature vector; obtain a reconstruction error according to the primary deep feature vector and the nearest neighbor; obtain a local reconstruction weight matrix according to the nearest neighbor and the reconstruction error.

3. The brain tumor image feature extraction method according to claim 1, wherein obtain the probability of containing a tumor in the brain tumor image according to the dimension-reduced primary deep feature vector and the parameter set of the brain tumor image classifier.

4. A brain tumor image feature extraction system characterized by comprising: The method comprises the following steps: a brain tumor image acquisition module is configured to obtain a brain tumor image; a primary deep feature acquisition module is configured to input the brain tumor image into a deep belief network to extract deep features and obtain primary deep features; the deep belief network is formed by three stacked RBMs; a dimensionality reduction module is configured to perform dimensionality reduction on the primary deep features by using a locally linear embedding algorithm; a local reconstruction weight matrix is obtained according to the primary deep feature vectors; a dimension-reduced primary deep feature vector is obtained according to the local reconstruction weight matrix and the primary deep feature vector, specifically as follows: find K adjacent points of each sample point Xi in a high-dimensional space and calculate the Euclidean distance; a local reconstruction weight matrix of the sample point is calculated from the adjacent points, and a reconstruction error function is measured as follows: where N is the number of sample points, X ij is the i-th sample point represented using the j-th neighboring point, W ij is the coefficient of the i-th sample point represented using the j-th neighboring point. W ij = 0, is the result of reconstructing the i-th sample point, the difference between the i-th sample point and the reconstructed i-th sample point is the reconstruction error, and all the coefficients W ij Finally, the local reconstruction weight matrix W is formed. all sample points X from a high dimensional space i are mapped to a low dimensional space Y i vector, such that X ij is mapped to Y ij ; where Y = {Y1, Y2,..., Y N}, Y j ∈R d , j = 1, 2,... N, d << D, Y = {Y1, Y2,..., Y N} is the mapping output of the dthdimensional space; a cost function is defined as follows: M = (I - W)(I - W) T where I is an identity matrix; find the value Y that minimizes the cost function to obtain a low-dimensional output vector; input the dimension-reduced primary deep features into a brain tumor image classifier to obtain a brain tumor image recognition result; obtain a classification loss of the brain tumor classifier according to the number of brain tumor images and the brain tumor image recognition result; specifically, a cross-entropy loss function is used: wherein N represents the number of images, indicates whether it is a brain tumor or not; minimize the classification loss to optimize the parameter set of the brain tumor classifier, specifically defined as follows: wherein, denotes a parameter set of a brain tumor image classifier.

5. An electronic device, comprising: The computer program product comprises a memory and a processor, and computer instructions stored in the memory and run on the processor, and when the computer instructions are run by the processor, the method of any one of claims 1-3 is completed.

6. A computer-readable storage medium, characterized in that, The computer program product is configured to store computer instructions, and when the computer instructions are executed by the processor, the method of any one of claims 1-3 is completed.

Images