Magnetic resonance image feature extraction method, system, device and feature application method

Abstract

The application provides a kind of magnetic resonance image feature extraction method, system, equipment and feature application method, it is related to image feature extraction technical field, the application includes obtaining the magnetic resonance imaging data of sample to be measured;According to magnetic resonance T1 weighted image, magnetic resonance imaging liquid attenuation inversion recovery sequence image and diffusion magnetic resonance image, construct the structure broken link network of sample to be measured;According to resting state functional magnetic resonance imaging, construct the functional broken link network of sample to be measured;According to structure broken link network and functional broken link network, construct individual injury degree network;The one-dimensional vector obtained by converting the upper triangular value on the matrix of individual injury degree network is used as the feature extraction result of the magnetic resonance image of sample to be measured.The application can accurately and reasonably extract the characteristics of magnetic resonance image by constructing structure broken link network and functional broken link network, to improve the precision of model classification.

Description

Technical Field

[0001] This invention relates to the field of image feature extraction technology, and in particular to a method, system, device, and feature application method for magnetic resonance imaging feature extraction. Background Technology

[0002] Functional magnetic resonance imaging (fMRI) plays a crucial role in monitoring brain activity and exploring connectivity between different brain regions. It is currently widely used in research on human cognitive function, such as in the auxiliary diagnosis of Alzheimer's disease, where it is frequently used for classification machine learning applications. Feature extraction is a critical step in machine learning; selecting appropriate features can significantly improve the model's classification performance (e.g., cognitive state prediction) while reducing model complexity. However, current feature extraction methods for fMRI images are insufficient to meet the accuracy requirements of current model classification. Therefore, there is an urgent need for a fMRI feature extraction technique to satisfy the classification accuracy requirements of model classification. Summary of the Invention

[0003] The purpose of this invention is to provide a method, system, device, and feature application method for magnetic resonance imaging feature extraction, which can accurately and reasonably extract features from magnetic resonance images to improve the accuracy of model classification.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] A method for feature extraction from magnetic resonance imaging includes:

[0006] Acquire magnetic resonance imaging data of the sample to be tested; the magnetic resonance imaging data includes magnetic resonance T1-weighted images, magnetic resonance imaging liquid attenuation inversion recovery sequence images, diffusion magnetic resonance images, and resting-state functional magnetic resonance imaging;

[0007] Based on the T1-weighted magnetic resonance image, the liquid attenuation inversion recovery sequence image, and the diffusion magnetic resonance image, a structural disconnection network of the sample to be tested is constructed.

[0008] Based on the resting-state functional magnetic resonance imaging, a functional disconnection network of the sample to be tested is constructed.

[0009] Based on the structural disconnection network and the functional disconnection network, an individual damage degree network is constructed.

[0010] The one-dimensional vector obtained by transforming the upper triangular value of the matrix of the individual damage degree network is used as the damage degree vector of the sample to be tested; the damage degree vector is the feature extraction result of the magnetic resonance image of the sample to be tested.

[0011] Optionally, based on the T1-weighted magnetic resonance image, the liquid attenuation inversion recovery sequence image, and the diffusion magnetic resonance image, a structural disconnection network of the sample under test is constructed, including:

[0012] Based on the T1-weighted magnetic resonance image and the liquid attenuation inversion recovery sequence image of magnetic resonance imaging, the high signal region map of white matter in the sample to be tested is determined;

[0013] The white matter high signal region map is transformed to the MNI standard space by linear transformation to obtain the white matter high signal region map in the MNI standard space.

[0014] The diffusion magnetic resonance image is preprocessed to obtain a preprocessed diffusion magnetic resonance image;

[0015] The preprocessed diffusion magnetic resonance images were registered to the MNI standard space and fiber tracing was performed to obtain the first deterministic fiber tracing structural connectivity image of the whole brain region.

[0016] The high-signal region map of white matter in the MNI standard space is superimposed on the diffuse magnetic resonance image in the MNI standard space to obtain the superimposed image.

[0017] The superimposed image is subjected to deterministic fiber tracing processing to obtain a second deterministic fiber tracing structure connection image of the high signal region of white matter;

[0018] The first structural connection matrix is ​​determined based on the first deterministic fiber tracking structural connection image;

[0019] The second structural connection matrix is ​​determined based on the second deterministic fiber tracking structural connection image;

[0020] Using the elements of the first structural connection matrix as the dividend and the elements of the second structural connection matrix as the divisor, the corresponding elements of the first structural connection matrix and the second structural connection matrix are divided to obtain the structural disconnection matrix of the whole brain region.

[0021] Based on the structural disconnection matrix, the structural disconnection network of the sample to be tested is determined.

[0022] Optionally, based on the resting-state functional magnetic resonance imaging, a functional disconnection network of the sample under test is constructed, including:

[0023] Extract the average signal from the high-signal region map of white matter in the MNI standard space;

[0024] The resting-state functional magnetic resonance imaging of the sample to be tested was preprocessed to obtain brain functional signal images in the MNI standard space;

[0025] The region outside the high signal area of ​​white matter in the brain functional signal image is divided to obtain multiple functional regions;

[0026] Based on the brain functional signal images, the average signal is processed with each functional region using Pearson correlation to obtain the functional connectivity values ​​between the high signal regions of white matter and each functional region.

[0027] Based on the multiple functional connectivity values, determine the functional disconnection matrix of the sample to be tested;

[0028] Based on the functional disconnection matrix, the functional disconnection network of the sample to be tested is determined.

[0029] Optionally, based on the structural disconnection network and the functional disconnection network, an individual impairment degree network is constructed, including:

[0030] Using the elements of the functional disconnection matrix as the dividend and the elements of the structural disconnection matrix as the divisor, the corresponding elements of the functional disconnection matrix and the structural disconnection matrix are used to perform quotient processing to obtain the individual damage degree matrix.

[0031] Based on the individual damage degree matrix, the individual damage degree network of the sample to be tested is determined.

[0032] A feature extraction system for magnetic resonance imaging, comprising:

[0033] The image acquisition module is used to acquire magnetic resonance imaging data of the sample to be tested; the magnetic resonance imaging data includes magnetic resonance T1-weighted images, magnetic resonance imaging liquid attenuation inversion recovery sequence images, diffusion magnetic resonance images, and resting-state functional magnetic resonance imaging.

[0034] The structural disconnection network construction module is used to construct a structural disconnection network of the sample under test based on the magnetic resonance T1-weighted image, the magnetic resonance imaging liquid attenuation inversion recovery sequence image and the diffusion magnetic resonance image.

[0035] A functional disconnection network construction module is used to construct a functional disconnection network for the sample under test based on the resting-state functional magnetic resonance imaging.

[0036] An individual damage severity network construction module is used to construct an individual damage severity network based on the structural disconnection network and the functional disconnection network.

[0037] The feature extraction module is used to convert the upper triangular values ​​of the matrix of the individual damage degree network into a one-dimensional vector, which serves as the feature extraction result of the magnetic resonance image of the sample to be tested.

[0038] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor running the computer program to enable the electronic device to perform the aforementioned feature extraction method for magnetic resonance imaging.

[0039] Optionally, the memory is a readable storage medium.

[0040] A feature application method, comprising:

[0041] The magnetic resonance imaging data of multiple individuals in the population to be tested are called population magnetic resonance imaging data.

[0042] Based on the group magnetic resonance imaging data, the damage degree vector of each individual in the group under test is determined using the aforementioned magnetic resonance image feature extraction method.

[0043] The damage degree vectors corresponding to the test group are superimposed to obtain the damage degree matrix of the test group.

[0044] The damage level matrix of the target group is input into the group cognitive level classification model to obtain the group cognitive level category of the target group; the group cognitive level classification model is a linear regression model constructed with the group damage level matrix of multiple groups as independent variables and the group cognitive level category of the corresponding group as dependent variable.

[0045] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0046] The present invention provides a method, system, device, and feature application method for magnetic resonance imaging feature extraction. By constructing a structural disconnection network and a functional disconnection network, and then constructing an individual damage degree network, the present invention can accurately and reasonably extract features of magnetic resonance imaging, which can be used to improve the accuracy of model classification (such as cognitive level classification model). Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of the magnetic resonance imaging feature extraction method according to Embodiment 1 of the present invention;

[0049] Figure 2 This is a flowchart of the feature application method in Embodiment 1 of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] The purpose of this invention is to provide a method, system, device, and feature application method for magnetic resonance imaging feature extraction, which can accurately and reasonably extract features from magnetic resonance images to improve the accuracy of model classification.

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] Example 1

[0054] like Figure 1 As shown, this embodiment provides a feature extraction method for magnetic resonance images, including:

[0055] Step 101: Acquire magnetic resonance imaging data of the sample to be tested; the magnetic resonance imaging data includes magnetic resonance T1-weighted images, magnetic resonance imaging liquid attenuation inversion recovery sequence images, diffusion magnetic resonance images, and resting-state functional magnetic resonance imaging.

[0056] Step 102: Construct the structural disconnection network of the sample under test based on the T1-weighted magnetic resonance imaging, the liquid attenuation inversion recovery sequence of magnetic resonance imaging, and the diffusion magnetic resonance imaging.

[0057] For example, step 102 includes:

[0058] Step 1021: Determine the high signal region of white matter in the sample under test based on the T1-weighted magnetic resonance imaging and the liquid attenuation inversion recovery sequence of magnetic resonance imaging.

[0059] Step 1022: Transform the white matter high signal region map to the MNI standard space through linear transformation to obtain the white matter high signal region map in the MNI standard space.

[0060] Step 1023: Preprocess the diffusion magnetic resonance image to obtain the preprocessed diffusion magnetic resonance image.

[0061] Step 1024: Register the preprocessed diffusion magnetic resonance images to the MNI standard space and perform fiber tracing processing to obtain the first deterministic fiber tracing structural connectivity images of the whole brain region.

[0062] Step 1025: Overlay the white matter high signal region map in the MNI standard space onto the diffuse magnetic resonance image in the MNI standard space to obtain the overlay image.

[0063] Step 1026: Perform deterministic fiber tracing processing on the superimposed image to obtain a second deterministic fiber tracing structure connection image of the high signal region of white matter.

[0064] Step 1027: Determine the first structure connection matrix based on the first deterministic fiber tracing structure connection image.

[0065] Step 1028: Determine the second structure connection matrix based on the second deterministic fiber tracking structure connection image.

[0066] Step 1029: Using the elements of the first structural connection matrix as the dividend and the elements of the second structural connection matrix as the divisor, perform quotient processing on the corresponding elements of the first and second structural connection matrices to obtain the structural disconnection matrix of the whole brain region.

[0067] Step 10210: Determine the structural disconnection network of the sample to be tested based on the structural disconnection matrix.

[0068] Specifically, constructing a structural disconnected network includes:

[0069] S1.1 Based on magnetic resonance T1-weighted images and FLAIR images (i.e., magnetic resonance imaging liquid attenuation inversion recovery sequence images), extract individual white matter high signal region maps of individual elderly samples.

[0070] S1.2 The white matter high signal region map is registered to the Montreal Neuroscience Institute’s brain standard space (i.e., MNI standard space) through linear transformation.

[0071] S1.3 preprocesses and performs diffusion-weighted imaging modeling on the diffusion-weighted magnetic resonance images of a single sample, and then performs fiber tracing. The specific process includes: DICOM to NII image format conversion, image resampling, skull stripping from the b0 image, eddy current correction, smoothing, registration to the MNI standard space, deterministic fiber tracing, etc. Finally, a deterministic fiber-traced structural connectivity image of the individual in the MNI standard space is constructed, and the structural connectivity network matrix (i.e., the first structural connectivity matrix) S of the individual is obtained through the Scheafer300 template.

[0072] S1.4 After processing the diffusion magnetic resonance image of a single sample in S1.3, the white matter high-signal region map obtained in S1.2 is superimposed onto the MNI standard space, and deterministic fiber tracing is performed. Finally, a deterministic fiber tracing structural connectivity image of the individual in the white matter high-signal region is obtained, and the structural connectivity matrix of the individual's white matter high-signal region (i.e., the second structural connectivity matrix) is obtained. wmh.

[0073] S1.5 is obtained by taking the quotient of the corresponding values ​​of the structural connectivity matrix obtained from S1.3 and S1.4 for each individual, thus obtaining the structural disconnection matrix S. D The value S of each node in the disconnection matrix of this structure D (m,n)=S wmh (m,n) / S(m,n) is used to obtain the structural disconnection network in the individual standard space; S D (m,n) represents the element in the m-th row and n-th column of the structural disconnection matrix; S wmh (m,n) represents the element in the m-th row and n-th column of the first structure connection matrix; S(m,n) represents the element in the m-th row and n-th column of the second structure connection matrix.

[0074] Step 103: Construct the functional disconnection network of the sample under test based on resting-state functional magnetic resonance imaging.

[0075] For example, step 103 includes:

[0076] Step 1031: Extract the average signal of the high-signal region map of white matter in the MNI standard space.

[0077] Step 1032: Preprocess the resting-state functional magnetic resonance imaging of the sample to be tested to obtain brain functional signal images in the MNI standard space.

[0078] Step 1033: Divide the region outside the high signal area of ​​white matter in the brain functional signal image to obtain multiple functional regions.

[0079] Step 1034: Based on the brain functional signal image, perform Pearson correlation processing on the average signal with each functional region to obtain the functional connectivity value between the high signal region of white matter and each functional region.

[0080] Step 1035: Determine the functional disconnection matrix of the sample to be tested based on multiple functional connectivity values.

[0081] Step 1036: Determine the functional disconnection network of the sample to be tested based on the functional disconnection matrix.

[0082] Specifically, building a functional disconnected network includes:

[0083] S2.1 preprocesses a single sample's resting-state functional magnetic resonance imaging (fMRI). The specific process includes: DICOM to NII image format conversion, removal of unstable time points, temporal correction, head motion correction, spatial normalization, delinear drift removal, spatial smoothing, noise variable removal, and filtering, ultimately obtaining a brain functional signal image in standard space.

[0084] S2.2 In the standard space, based on the high signal region of white matter obtained in S1.2, the average signal of the high signal region of white matter is extracted.

[0085] S2.3 The average signal extracted in the above steps is used to perform Pearson correlation with other regions of the whole brain. The functional connectivity values ​​of the whole brain and the high signal regions of white matter are calculated using the Schaefer300 template, which is the functional disconnection network under the individual standard space.

[0086] Step 104: Construct an individual damage degree network based on the structural disconnection network and the functional disconnection network.

[0087] For example, step 104 includes:

[0088] Step 1041: Using the elements of the functional disconnection matrix as the dividend and the elements of the structural disconnection matrix as the divisor, perform corresponding element-wise quotient processing on the functional disconnection matrix and the structural disconnection matrix to obtain the individual damage degree matrix.

[0089] Step 1042: Determine the individual damage degree network of the sample to be tested based on the individual damage degree matrix.

[0090] This invention yields a structural disconnection network for individual i. and functional disconnection network Network for calculating individual damage levels d i (m,n) represents the element in the m-th row and n-th column of the individual injury severity matrix; This represents the element in the m-th row and n-th column of the functional disconnection matrix; This represents the element in the m-th row and n-th column of the structural disconnection matrix; the upper triangular value of the damage degree network matrix is ​​taken and transformed into a one-dimensional vector.

[0091] Step 105: Transform the upper triangular values ​​of the individual damage degree network matrix into a one-dimensional vector, and use it as the damage degree vector of the sample to be tested; the damage degree vector is the feature extraction result of the magnetic resonance image of the sample to be tested.

[0092] Example 2

[0093] In order to perform the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a feature extraction system for magnetic resonance imaging is provided below, including:

[0094] The image acquisition module is used to acquire magnetic resonance imaging data of the sample to be tested; the magnetic resonance imaging data includes magnetic resonance T1-weighted images, magnetic resonance imaging liquid attenuation inversion recovery sequence images, diffusion magnetic resonance images, and resting-state functional magnetic resonance imaging.

[0095] The structural disconnection network construction module is used to construct the structural disconnection network of the sample under test based on magnetic resonance T1-weighted images, magnetic resonance imaging liquid attenuation inversion recovery sequence images, and diffusion magnetic resonance images.

[0096] The functional disconnection network construction module is used to construct a functional disconnection network for the sample under test based on resting-state functional magnetic resonance imaging.

[0097] The Individual Damage Degree Network Construction Module is used to construct an individual damage degree network based on the structural disconnection network and the functional disconnection network.

[0098] The feature extraction module is used to transform the upper triangular values ​​of the matrix of the individual damage degree network into a one-dimensional vector, which serves as the feature extraction result of the magnetic resonance image of the sample to be tested.

[0099] Example 3

[0100] This embodiment provides an electronic device, including a memory and a processor. The memory is used to store computer programs, and the processor runs the computer programs to enable the electronic device to perform a feature extraction method for magnetic resonance images according to Embodiment 1.

[0101] The memory is a readable storage medium.

[0102] Example 4

[0103] Changes in cognitive abilities are an inevitable consequence of aging, becoming one of the main factors affecting the quality of life in old age and closely related to neurodegenerative diseases such as Alzheimer's disease. Currently, commonly used scales for assessing cognitive function in clinical practice include the Montreal Cognitive Assessment (MoCA), the Activities of Daily Living (ADL) scale, the Mini-Mental State Examination (MMSE), and the Clinical Dementia Rating Scale (CDR). Assessment of the cognitive domain often requires combining multiple scales; for example, assessment of the memory domain typically involves combining the Logical Memory Story (A delayed recall), the Rey Auditory Verbal Learning Test, and the Benton Visual Retention Test recognition scale. Cognitive assessment is a very difficult process, and the results are easily influenced by the individual's education level and subjective factors.

[0104] White matter hyperintensity (WMH), also known as leukoaraiosis (LA), is characterized by high-signal areas around the ventricles in T2-weighted fluid-attenuated inversion recovery (FLAIR) images. WMH appears in some individuals after age 40, and community-based studies on aging have found WMH in 50%–98% of brain imaging, with the probability increasing with age. In non-dementia individuals, WMH load is negatively correlated with executive function, motor speed, and memory. Simultaneously, impaired executive function, memory delay, and decreased attention in individuals with mild cognitive impairment are significantly associated with white matter hyperintensity. Structural and functional magnetic resonance imaging (fMRI) of the brain can detect changes in the integrity and function of white matter, and studies have used MRI to construct brain networks to predict cognitive states. These results confirm the relationship between white matter hyperintensity and cognitive level, and imaging techniques provide a powerful tool for this analysis. Therefore, this embodiment uses group cognitive levels as an example to specifically illustrate the invention:

[0105] like Figure 2 This embodiment provides a feature application method, including:

[0106] Step 401: Obtain magnetic resonance imaging data of multiple individuals in the test group to form group magnetic resonance imaging data.

[0107] Step 402: Based on the group magnetic resonance imaging data, determine the damage degree vector of each individual in the group under test using a feature extraction method for magnetic resonance images as described in any one of claims 1 to 4.

[0108] Step 403: Superimpose multiple damage degree vectors corresponding to the test group to obtain the damage degree matrix of the test group.

[0109] Step 404: Input the damage level matrix of the test group into the group cognitive level classification model to obtain the group cognitive level category of the test group; the group cognitive level classification model is a linear regression model constructed with the group damage level matrix of multiple groups as independent variables and the group cognitive level category of the corresponding group as dependent variable.

[0110] Specifically, for each subject's image, the process of steps S1.1-S1.4 in Example 1 is repeated to obtain a total of N structural disconnection networks in standard spaces. For each subject's image, the process of steps S2.1-S2.3 in Example 1 is repeated to obtain a total of N functionally significant disconnection networks in standard spaces. The upper triangular values ​​of the impairment network matrix are transformed into one-dimensional vectors; the one-dimensional vectors obtained for each sample are superimposed to obtain the group impairment network D; using the group impairment network D as the independent variable and the group cognitive score, such as the MMSE value, as the dependent variable, a linear regression model is established using partial least squares regression. Ten regression models are obtained through the ten-fold cross-validation method. The model coefficients b0, b1, ... are averaged to obtain the normalized model coefficients b0, b1, ..., which can then be used to predict the group's cognitive score.

[0111] This invention uses a broad range of cognitively normal community residents as the model building sample. Their brain connectivity information is representative of the population and can accurately reflect the normal cognitive aging process associated with high white matter signal intensity. By using a brain damage network based on high white matter signal intensity to predict cognitive changes in the population, considering age-related cognitive changes from a damage perspective, it can more accurately reflect the damage to the brain system caused by aging. Based on MRI imaging data, the model building process is objective, effectively avoiding biases in cognitive assessments caused by subjective factors in medical staff and patients, and significantly saving assessment time.

[0112] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0113] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method of feature extraction from a magnetic resonance image, characterized by, The method comprises the following steps: acquiring magnetic resonance imaging data of a sample to be tested; the magnetic resonance imaging data comprises a magnetic resonance T1 weighted image, a magnetic resonance imaging liquid attenuation inversion recovery sequence image, a diffusion magnetic resonance image and a resting state functional magnetic resonance imaging; constructing a structural disconnection network of the sample to be tested according to the magnetic resonance T1 weighted image, the magnetic resonance imaging liquid attenuation inversion recovery sequence image and the diffusion magnetic resonance image; constructing a functional disconnection network of the sample to be tested according to the resting state functional magnetic resonance imaging; constructing an individual damage degree network according to the structural disconnection network and the functional disconnection network; converting a one-dimensional vector obtained by converting the upper triangular values of the matrix of the individual damage degree network into the one-dimensional vector as a damage degree vector of the sample to be tested; the damage degree vector is a feature extraction result of the magnetic resonance image of the sample to be tested; acquiring magnetic resonance imaging data of multiple individuals in a test group as group magnetic resonance imaging data; determining the damage degree vector of each individual in the test group according to the group magnetic resonance imaging data; superimposing the multiple damage degree vectors corresponding to the test group to obtain a damage degree matrix of the test group; inputting the damage degree matrix of the test group into a group cognitive level classification model to obtain the group cognitive level category of the test group; the group cognitive level classification model is a linear regression model constructed by taking the group damage degree matrices of multiple groups as independent variables and taking the group cognitive level categories of the corresponding groups as dependent variables; constructing a structural disconnection network of the sample to be tested according to the magnetic resonance T1 weighted image, the magnetic resonance imaging liquid attenuation inversion recovery sequence image and the diffusion magnetic resonance image, comprising: determining a white matter high signal area map of the sample to be tested according to the magnetic resonance T1 weighted image and the magnetic resonance imaging liquid attenuation inversion recovery sequence image; transforming the white matter high signal area map to the MNI standard space through linear transformation to obtain a white matter high signal area map in the MNI standard space; preprocessing the diffusion magnetic resonance image to obtain a preprocessed diffusion magnetic resonance image; registering the preprocessed diffusion magnetic resonance image to the MNI standard space and performing fiber tracking processing to obtain a first deterministic fiber tracking structural connection image of the whole brain region; superimposing the white matter high signal area map in the MNI standard space on the diffusion magnetic resonance image in the MNI standard space to obtain a superimposed image; performing deterministic fiber tracking processing on the superimposed image to obtain a second deterministic fiber tracking structural connection image of the white matter high signal area; determining a first structural connection matrix according to the first deterministic fiber tracking structural connection image; determining a second structural connection matrix according to the second deterministic fiber tracking structural connection image; performing corresponding element quotient processing on the first structural connection matrix and the second structural connection matrix by taking the elements of the first structural connection matrix as the dividend and the elements of the second structural connection matrix as the divisor to obtain a structural disconnection matrix of the whole brain region; determining a structural disconnection network of the sample to be tested according to the structural disconnection matrix; According to the structural disconnection network and the functional disconnection network, an individual injury degree network is constructed, comprising: Taking the elements of the functional disconnection matrix as the dividend and the elements of the structural disconnection matrix as the divisor, corresponding element quotient processing is performed on the functional disconnection matrix and the structural disconnection matrix to obtain an individual injury degree matrix; According to the individual injury degree matrix, the individual injury degree network of the sample to be tested is determined.

2. The method of claim 1, wherein, According to the resting state functional magnetic resonance imaging, a functional disconnection network of the sample to be tested is constructed, comprising: An average signal of a white matter hyperintensity region map in the MNI standard space is extracted; The resting state functional magnetic resonance imaging of the sample to be tested is preprocessed to obtain a brain functional signal image in the MNI standard space; Regions other than the white matter hyperintensity region map in the brain functional signal image are divided to obtain a plurality of functional regions; Based on the brain functional signal image, the average signal is subjected to Pearson correlation processing with each functional region to obtain a functional connection value of the white matter hyperintensity region and each functional region; According to a plurality of the functional connection values, a functional disconnection matrix of the sample to be tested is determined; According to the functional disconnection matrix, a functional disconnection network of the sample to be tested is determined.

3. A feature extraction system for magnetic resonance images, characterized in that The magnetic resonance image feature extraction system applies the magnetic resonance image feature extraction method according to any one of claims 1-2, and comprises: An image acquisition module is configured to acquire magnetic resonance imaging data of a sample to be tested, wherein the magnetic resonance imaging data comprises a magnetic resonance T1 weighted image, a magnetic resonance imaging liquid decay inversion recovery sequence image, a diffusion magnetic resonance image and a resting state functional magnetic resonance imaging; A structural disconnection network construction module is configured to construct a structural disconnection network of the sample to be tested according to the magnetic resonance T1 weighted image, the magnetic resonance imaging liquid decay inversion recovery sequence image and the diffusion magnetic resonance image; A functional disconnection network construction module is configured to construct a functional disconnection network of the sample to be tested according to the resting state functional magnetic resonance imaging; An individual injury degree network construction module is configured to construct an individual injury degree network according to the structural disconnection network and the functional disconnection network; A feature extraction module is configured to convert the upper triangular values of the matrix of the individual injury degree network into a one-dimensional vector as a feature extraction result of the magnetic resonance image of the sample to be tested.

4. An electronic device, comprising: The electronic device comprises a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the magnetic resonance image feature extraction method according to any one of claims 1-2.

5. The electronic device of claim 4, wherein, The memory is a readable storage medium.

Images