A classification method for resting-state functional magnetic resonance imaging images of the brain

Abstract

The present invention discloses a classification method for resting-state functional magnetic resonance imaging images of the brain. The present invention adopts a combination of traditional machine learning and deep learning methods to analyze the user's neuroimaging results, solving the problem that the analysis method of resting-state functional magnetic resonance imaging images in the prior art cannot have both high classification performance and low training data volume.

Description

Technical Field

[0001] The present invention relates to the field of image classification, and in particular to a classification method for functional magnetic resonance imaging images of a brain in a resting state. Background Art

[0002] Neuroimaging technology is a powerful tool for capturing brain structure and function, and offers great potential for understanding the mechanisms of brain diseases. Resting-state functional magnetic resonance imaging (rs-fMRI) is a powerful neuroimaging method that can capture blood oxygen level-dependent (BOLD) signals and reflect the activity state of the brain. Existing analysis methods for resting-state functional magnetic resonance imaging images are mainly divided into two strategies: the first is to use traditional machine learning methods to build diagnostic classification models. Since traditional machine learning usually uses human prior knowledge to preprocess raw data into various features and then classify the features, this learning method does not require a large amount of data, but has limited performance. The second method uses deep learning methods, which allow the neural network to learn how to capture the features of the data by itself. Therefore, this learning method can effectively improve the classification performance, but requires a large amount of data to train the model. In short, the analysis methods of resting-state functional magnetic resonance imaging images in the prior art have the problem of not being able to have both high classification performance and low training data volume.

[0003] Therefore, the existing technology still needs to be improved and developed. Summary of the invention

[0004] The technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, a classification method for resting-state functional magnetic resonance imaging images of the brain is provided, aiming to solve the problem that the analysis method of resting-state functional magnetic resonance imaging images in the prior art cannot have both high classification performance and low training data volume.

[0005] The technical solution adopted by the present invention to solve the problem is as follows:

[0006] In a first aspect, an embodiment of the present invention provides a method for classifying a resting-state functional magnetic resonance imaging image of the brain, wherein the method comprises:

[0007] Acquire a plurality of time series data, wherein the plurality of time series data are generated based on a resting state functional magnetic resonance imaging image of a target user's brain, and the plurality of time series data correspond to different brain regions of the target user's brain;

[0008] Inputting the time series data into a preset brain function connection network, performing feature extraction on the time series data through the brain function connection network, and obtaining a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions;

[0009] Inputting the connection relationship matrix into a fused sparse autoencoder, and performing dimension reduction on the connection relationship matrix by the fused sparse autoencoder to obtain a target feature vector;

[0010] The target feature vector is classified to obtain a classification category corresponding to the target user.

[0011] In one implementation, the acquiring of a number of time series data includes:

[0012] Acquiring the resting state functional magnetic resonance imaging image of the target user;

[0013] Acquire a preset segmentation template, and segment the resting state functional magnetic resonance imaging image according to the segmentation template to obtain a plurality of local images, wherein the plurality of local images respectively correspond to different brain regions;

[0014] Filtering is performed on the plurality of local images respectively to obtain the plurality of time series data.

[0015] In one embodiment, the brain functional connection network is constructed based on a group constrained sparse model, and the first objective function corresponding to the group constrained sparse model includes a group sparsity regularization parameter and a smoothing regularization parameter.

[0016] In one embodiment, the fused sparse autoencoder includes a first symmetric neural network and a second symmetric neural network, the first symmetric neural network and the second symmetric neural network respectively include different numbers of hidden layers, the connection relationship matrix is ​​input into the fused sparse autoencoder, and the connection relationship matrix is ​​reduced in dimension by the fused sparse autoencoder to obtain a target feature vector, including:

[0017] Inputting the connection relationship matrix into the first symmetric neural network, performing feature extraction on the connection relationship matrix through the first symmetric neural network to obtain a first feature vector;

[0018] Inputting the connection relationship matrix into the second symmetric neural network, performing feature extraction on the connection relationship matrix through the second symmetric neural network to obtain a second feature vector;

[0019] The first feature vector and the second feature vector are fused to obtain the target feature vector.

[0020] In one embodiment, the first symmetric neural network includes a plurality of cascaded hidden layers, and the second symmetric neural network includes one hidden layer.

[0021] In one implementation, the second objective function corresponding to each hidden layer includes a Euclidean norm.

[0022] In one implementation, classifying the target feature vector to obtain a user label corresponding to the target user includes:

[0023] Inputting the target feature vector into a preset classifier;

[0024] The target feature vector is classified by the classifier to obtain the classification category.

[0025] In a second aspect, an embodiment of the present invention further provides a device for classifying resting state functional magnetic resonance imaging images of the brain, wherein the device comprises:

[0026] A data acquisition module, used to acquire a plurality of time series data, wherein the plurality of time series data are generated based on a resting state functional magnetic resonance imaging image of a target user's brain, and the plurality of time series data correspond to different brain regions of the target user's brain;

[0027] A data extraction module, used for performing feature extraction on the time series data to obtain a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions;

[0028] The data classification module is used to extract features from the connection relationship matrix to obtain a target feature vector, and classify the target feature vector to obtain a classification category corresponding to the target user.

[0029] In a third aspect, an embodiment of the present invention further provides a terminal, wherein the terminal includes a memory and one or more processors; the memory stores one or more programs; the program includes instructions for executing a classification method for resting-state functional magnetic resonance imaging images of the brain as described in any one of the above; and the processor is used to execute the program.

[0030] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a plurality of instructions stored thereon, wherein the instructions are suitable for being loaded and executed by a processor to implement the steps of any of the above-mentioned methods for classifying resting-state functional magnetic resonance imaging images of the brain.

[0031] Beneficial effects of the present invention: The embodiment of the present invention obtains a number of time series data, wherein the time series data are generated based on the resting state functional magnetic resonance imaging of the target user's brain, and the time series data correspond to different brain regions of the target user's brain; the time series data are input into a preset brain function connection network, and feature extraction is performed on the time series data through the brain function connection network to obtain a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions; the connection relationship matrix is ​​input into a fused sparse autoencoder, and the connection relationship matrix is ​​reduced in dimension through the fused sparse autoencoder to obtain a target feature vector; the target feature vector is classified to obtain the classification category corresponding to the target user. The present invention adopts a combination of traditional machine learning and deep learning methods to analyze the user's neuroimaging results, which solves the problem that the analysis method of resting state functional magnetic resonance imaging in the prior art cannot have both high classification performance and low training data volume. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 It is a flow chart of a classification method for resting-state functional magnetic resonance imaging images of the brain provided by an embodiment of the present invention.

[0034] Figure 2 It is a structural schematic diagram of a brain functional connection network and a fused sparse autoencoder provided in an embodiment of the present invention.

[0035] Figure 3 It is an internal module diagram of a classification device for resting-state functional magnetic resonance imaging images of the brain provided by an embodiment of the present invention.

[0036] Figure 4 It is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention discloses a classification method for functional magnetic resonance imaging images of the brain in a resting state. In order to make the purpose, technical scheme and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0038] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0039] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0040] Neuroimaging technology is a powerful tool for capturing brain structure and function, and offers great potential for understanding the mechanisms of brain diseases. Resting-state functional magnetic resonance imaging (rs-fMRI) is a powerful neuroimaging method that can capture blood oxygen level-dependent (BOLD) signals and reflect the activity state of the brain. Existing analysis methods for resting-state functional magnetic resonance imaging images are mainly divided into two strategies: the first is to use traditional machine learning methods to build diagnostic classification models. Since traditional machine learning usually uses human prior knowledge to preprocess raw data into various features and then classify the features, this learning method does not require a large amount of data, but has limited performance. The second method uses deep learning methods, which allow the neural network to learn how to capture the features of the data by itself. Therefore, this learning method can effectively improve the classification performance, but requires a large amount of data to train the model. In short, the analysis methods of resting-state functional magnetic resonance imaging images in the prior art have the problem of not being able to have both high classification performance and low training data volume.

[0041] In view of the above-mentioned defects of the prior art, the present invention provides a classification method for resting state functional magnetic resonance imaging images of the brain, the method comprising: obtaining a number of time series data, wherein the time series data are all generated based on the resting state functional magnetic resonance imaging images of the target user's brain, and the time series data correspond to different brain regions of the target user's brain; inputting the time series data into a preset brain function connection network, extracting features of the time series data through the brain function connection network, and obtaining a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions; inputting the connection relationship matrix into a fused sparse autoencoder, reducing the dimension of the connection relationship matrix through the fused sparse autoencoder, and obtaining a target feature vector; classifying the target feature vector to obtain the classification category corresponding to the target user. The present invention adopts a combination of traditional machine learning and deep learning methods to analyze the user's neuroimaging results, solving the problem that the analysis method of resting state functional magnetic resonance imaging images in the prior art cannot have both high classification performance and low training data volume.

[0042] like Figure 1 As shown, the method comprises the following steps:

[0043] Step S100, obtaining a plurality of time series data, wherein the plurality of time series data are generated based on a resting-state functional magnetic resonance imaging image of a target user's brain, and the plurality of time series data correspond to different brain regions of the target user's brain.

[0044] Specifically, neuroimaging technology is a powerful tool for capturing brain structure and function. In order to analyze the resting state functional magnetic resonance imaging image of the target user's brain, this embodiment first needs to convert the resting state functional magnetic resonance imaging image of the target user's brain into a time series form. In addition, since different areas of the brain reflect different information, it is also necessary to divide the brain regions according to the resting state functional magnetic resonance imaging image of the brain, so as to obtain the time series data corresponding to each brain region.

[0045] In one implementation, step S100 specifically includes the following steps:

[0046] Step S101, obtaining the resting state functional magnetic resonance imaging image of the target user;

[0047] Step S102, obtaining a preset segmentation template, and segmenting the resting state functional magnetic resonance imaging image according to the segmentation template to obtain a plurality of local images, wherein the plurality of local images respectively correspond to different brain regions;

[0048] Step S103: performing filtering processing on the plurality of local images respectively to obtain a plurality of time series data.

[0049] Specifically, in order to segment the resting state functional magnetic resonance imaging image of the target user, this embodiment pre-sets a segmentation template, through which the brain area images that do not contribute to the analysis results can be deleted, and the local images of other brain areas of interest can be retained. Then, these local images are filtered using a preset filter to obtain the time series data corresponding to each local image.

[0050] As an example, a 3.0-Tesla MR system (Philips Medical Systems, USA) equipped with an eight-channel phased array head coil was used for data acquisition. During data collection, the subjects were asked to close their eyes, relax, stay awake, and not move. The resting-state functional magnetic resonance imaging data of the brain were acquired using a gradient echo planar imaging (EPI) sequence with the following parameters: echo time (TE) = 60 ms; temporal repetition (TR) = 2000 ms; 33 slices, flip angle = 90°, field of view [FOV] = 240 mm × 240 mm, matrix = 64 × 64; slice thickness = 4.0 mm. For spatial normalization and positioning, a magnetization-prepared gradient echo sequence was also used. The specific parameters were as follows: TE = 1.7 ms, TR = 8 ms, flip angle = 20°, FOV = 240 mm × 240 mm, matrix = 256 × 256, and slice thickness = 1.0 mm.

[0051] The target user's resting-state fMRI images were then preprocessed using the Resting-State fMRI Data Processing Assistant (DPARSFA) and the Statistical Parametric Mapping Toolbox (SPM8) through a standardized data preprocessing process. The first 10 volumes of rs-fMRI data were discarded before processing to maintain equal magnetization intensity. The remaining 170 volumes were corrected using an interleaved sequence of slice acquisition, and echo planar scanning was used to ensure that the data on each slice corresponded to a consistent time point. The acquired image data were preprocessed as follows: slice timing correction; head motion correction; realignment with the corresponding T1 volume; regression of nuisance covariates (six head motion parameters, white matter signal, and cerebrospinal fluid signal); spatial normalization to the Montreal Neurological Institute stereotactic space and resampling to 3 × 3 × 3 mm3; spatial smoothing using a 6 mm full-width half-maximum isotropic Gaussian kernel and bandpass filtering (0.01–0.08 Hz). The rs-fMRI data were segmented using the AAL template. Twenty-six cerebellar ROIs were removed, and 90 brain regions of interest (ROIs) were retained for study. In addition, the mean RS fMRI time series of each brain region were thinned using a high-pass filter. In addition, the mean BOLD time series of head motion parameters, cerebrospinal fluid, and white matter were regressed out. Finally, the mean value of the BOLD signal was extracted as the raw RS fMRI data.

[0052] like Figure 1 As shown, the method further comprises the following steps:

[0053] Step S200: inputting the time series data into a preset brain function connection network, performing feature extraction on the time series data through the brain function connection network, and obtaining a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions.

[0054] Specifically, each time series data is input into a preset brain function connection network. Since the brain function connection network in this embodiment has been pre-trained (the training uses the subject's rs-fMRI time series data as input data to learn the relationships between the brains), the brain function connection network can perform feature extraction on the input time series data, and then output a connection relationship matrix that reflects the connection relationship between the brain regions.

[0055] In one implementation, the brain functional connection network is constructed based on a group constrained sparse model, and the first objective function corresponding to the group constrained sparse model includes a group sparsity regularization parameter and a smoothing regularization parameter.

[0056] In short, since most biological networks are sparse in nature, sparse learning should remove noise features and retain important information, so this embodiment constructs a brain functional connection network through sparse learning. Specifically, this embodiment uses a group-constrained sparse model to construct a brain functional connection network, and in order to utilize the smoothness characteristics of different objects, this embodiment also adds a group sparse regularization parameter and a smoothing regularization parameter to the first objective function corresponding to the group-constrained sparse model.

[0057] For example, there are N subjects in total, and the brain is segmented into R regions of interest (ROIs) using a preset segmentation template. The bold region represents the average time series of the Rth ROI (length M), which can be estimated as a linear combination of the time series of the other R-1 ROIs. To include except The signal matrix of all ROI data except is a weighted regression coefficient vector used to measure the influence of R-1 ROI on the target ROI (the rth ROI). The group constrained sparse (GCS) model can be used to construct a sparse network representing the brain functional connectivity network (BFCN). The first objective function is defined as follows:

[0058]

[0059] Among them, λ1 represents the group regularization parameter, yes The sum of the l2-norms of W r Specifically, for different matrices The l2 norm constraint is imposed on the same elements of . By this strategy, the weights corresponding to the connections between different subjects are grouped together. As a sparse regression network method, GCS ensures that all models in unequal groups have similar connections. The l2 norm not only encourages common connection topology between subjects, but also promotes the variation of connection values ​​between subjects. r The coefficients in the matrix represent the importance of the relationship between the other R-1 ROIs and the target ROI. The model uses the remaining ROIs to represent the relationship between ROIs and can linearly reconstruct the target ROI. In order to refer to the smoothness characteristics of different objects in the model, this embodiment adds group sparse regularization and smooth regularization to the model to jointly learn the shared BFCN of each subject. The objective function is defined as:

[0060]

[0061] The first term in equation (2) is the data fitting term, the second term is the group regularization term, and the last term represents the smoothness regularization. λ1 and λ2 are the parameters of group regularization and smoothness regularization, respectively. The last term The difference between two consecutive weighted vectors from the same position is limited to be as small as possible. The l1 norm is used in the smooth regularization, so there will be many zero components in the weight vector, which makes the weight vector have sparse characteristics. The smooth term is introduced to smooth the connectivity coefficients of the subjects, and the non-zero weights in the matrix represent the information characteristics of the target ROI. The sparse learning model is defined as a smooth sparse network.

[0062] like Figure 1 As shown, the method further comprises the following steps:

[0063] Step S300: input the connection relationship matrix into a fused sparse autoencoder, and reduce the dimension of the connection relationship matrix by the fused sparse autoencoder to obtain a target feature vector.

[0064] Simply put, the application of brain functional connection networks usually has the problem of too high data dimension. Therefore, in order to reduce the dimension of the data, this embodiment pre-trains a fused sparse autoencoder, and performs feature learning and dimensionality reduction on the connection relationship matrix output by the brain functional connection network through the fused sparse autoencoder, thereby outputting the target feature vector.

[0065] In one implementation, the original edge weights of the connection relationship matrix are used as features. Specifically, the upper triangular weights of the connection relationship matrix are used as features.

[0066] For example, if the brain is divided into 90 ROIs, the extracted connectivity matrix will contain 8100 elements. To ensure symmetry, simply define Where W n is the BFCN matrix (connection relationship matrix) finally obtained by the nth subject, It is W n The transposed matrix of . Since the BFCN matrix is ​​symmetric, only the upper triangular weights of the BFCN matrix are extracted as features. Assuming there are n nodes in the undirected BFCN, this will generate n(n-1) / 2 edges, which is equivalent to obtaining a feature vector containing 4005 elements.

[0067] In one implementation, the fused sparse autoencoder includes a first symmetric neural network and a second symmetric neural network, the first symmetric neural network and the second symmetric neural network respectively include different numbers of hidden layers, and the step S300 specifically includes the following steps:

[0068] Step S301, inputting the connection relationship matrix into the first symmetric neural network, performing feature extraction on the connection relationship matrix through the first symmetric neural network, and obtaining a first feature vector;

[0069] Step S302: input the connection relationship matrix into the second symmetric neural network, and extract features from the connection relationship matrix through the second symmetric neural network to obtain a second feature vector;

[0070] Step S303: The first feature vector and the second feature vector are merged to obtain the target feature vector.

[0071] Specifically, Figure 2 As shown, the fused sparse autoencoder in this embodiment includes two networks, namely a first symmetric neural network and a second symmetric neural network, and the difference between the two networks is the number of hidden layers. Therefore, by inputting the connection relationship matrix into the two networks respectively, features of different levels can be extracted, and finally the first feature vector output by the first symmetric neural network and the second feature vector output by the second symmetric neural network are fused to obtain the target feature vector, which includes the multi-level features of the connection relationship matrix.

[0072] In one implementation, the first symmetric neural network includes a plurality of cascaded hidden layers, and the second symmetric neural network includes one hidden layer.

[0073] Specifically, the first symmetric neural network in the present embodiment is constructed by stacking multiple sparse autoencoders (SAE), for example, two sparse autoencoders are used. The first symmetric neural network can learn high-level information from input data, which is superimposed by multiple sparse autoencoders, improving the performance of traditional sparse autoencoders. The second symmetric neural network is constructed by traditional sparse autoencoders. The principle of sparse autoencoders is to punish hidden unit deviations to learn relatively sparse features. Since the traditional sparse autoencoder has only one hidden layer, the second symmetric neural network can retain a lot of useful input data information. For sparse autoencoders, in the encoder network, the input data in the high-dimensional space is converted into a low-dimensional feature space. Then the decoding network is used to reconstruct the input data from the feature space to the high-dimensional space. Therefore, the target feature vector obtained by fusing the first feature vector output by the first symmetric neural network and the second feature vector output by the second symmetric neural network not only includes the multi-level features of the connection relationship matrix, but also reduces the data dimension of the connection relationship matrix at the same time.

[0074] In one implementation, the second objective function corresponding to each hidden layer includes a Euclidean norm.

[0075] Specifically, the fusion sparse autoencoder in this embodiment also introduces the Euclidean norm (l2 norm) to improve the classification accuracy and prevent overfitting. Therefore, the second objective function corresponding to each hidden layer can be expressed as:

[0076]

[0077]

[0078]

[0079] It contains three terms: mean square error, l2 norm and Kullback-Leibler (KL) divergence. β1 and β2 are the coefficients of l2 regularization and KL divergence respectively. The input data is represented as The reconstructed data is represented as For N subjects, the dimensions of the input data and the reconstructed data are F and D respectively. The activation matrix of the hidden layer is expressed as It has K nodes. The input data X is encoded into the activation matrix Z using bias b1 and weight W1. The bias b2 and weight W2 decode the activation matrix Z and the typical nonlinear form is shown below:

[0080] Z=f(w1X+b1)(6)

[0081] Y=f(w2Z+b2)(7)

[0082] The average activation of the kth hidden node is expressed as:

[0083]

[0084] Where ρ represents a constant and the scaled conjugate gradient descent algorithm is used to optimize the weights W and bias b.

[0085] like Figure 1 Said method further comprises the following steps:

[0086] Step S400: classify the target feature vector to obtain a classification category corresponding to the target user.

[0087] Since the target feature vector can reflect the multi-level features of the connection relationship between different brain regions, a more accurate classification category can be obtained by classifying according to the target feature vector. This classification category can assist relevant staff in evaluating the group corresponding to the target user, such as a high-risk group or a normal group.

[0088] In one implementation, step S400 specifically includes the following steps:

[0089] Step S401, inputting the target feature vector into a preset classifier;

[0090] Step S402: classify the target feature vector by the classifier to obtain the classification category.

[0091] Specifically, Figure 2 As shown, in order to implement the classification process, this embodiment pre-sets a classifier, and the target feature vector is input into the classifier, and the classifier automatically performs classification based on the input target feature vector, thereby outputting the classification category with the highest probability.

[0092] Based on the above embodiments, the present invention also provides a classification device for resting state functional magnetic resonance imaging images of the brain, such as Figure 3 As shown, the device comprises:

[0093] The data acquisition module 01 is used to acquire a plurality of time series data, wherein the plurality of time series data are generated based on a resting state functional magnetic resonance imaging image of the target user's brain, and the plurality of time series data correspond to different brain regions of the target user's brain respectively;

[0094] A data extraction module 02 is used to extract features from the time series data to obtain a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions;

[0095] The data classification module 03 is used to extract features from the connection relationship matrix to obtain a target feature vector, and classify the target feature vector to obtain a classification category corresponding to the target user.

[0096] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 4 As shown. The terminal includes a processor, a memory, a network interface, and a display screen connected through a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a classification method for functional magnetic resonance imaging images of the brain in a resting state is implemented. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0097] Those skilled in the art will understand that Figure 4 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0098] In one implementation, one or more programs are stored in the memory of the terminal, and the terminal is configured to be executed by one or more processors. The one or more programs include instructions for performing a classification method for resting state functional magnetic resonance imaging images of the brain.

[0099] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0100] In summary, the present invention discloses a classification method for resting state functional magnetic resonance imaging images of the brain, the method is by obtaining a number of time series data, wherein the time series data are all generated based on the resting state functional magnetic resonance imaging images of the target user's brain, and the time series data correspond to different brain regions of the target user's brain; the time series data are input into a preset brain function connection network, and the features of the time series data are extracted through the brain function connection network to obtain a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions; the connection relationship matrix is ​​input into a fused sparse autoencoder, and the connection relationship matrix is ​​reduced in dimension through the fused sparse autoencoder to obtain a target feature vector; the target feature vector is classified to obtain the classification category corresponding to the target user. The present invention adopts a combination of traditional machine learning and deep learning methods to analyze the user's neuroimaging results, solving the problem that the analysis method of resting state functional magnetic resonance imaging images in the prior art cannot have both high classification performance and low training data volume.

[0101] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A classification method for resting state functional magnetic resonance imaging images of the brain, characterized in that: The method comprises: Acquire a plurality of time series data, wherein the plurality of time series data are generated based on a resting state functional magnetic resonance imaging image of a target user's brain, and the plurality of time series data correspond to different brain regions of the target user's brain; Inputting the time series data into a preset brain function connection network, performing feature extraction on the time series data through the brain function connection network, and obtaining a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions; Inputting the connection relationship matrix into a fused sparse autoencoder, and performing dimension reduction on the connection relationship matrix by the fused sparse autoencoder to obtain a target feature vector; Classifying the target feature vector to obtain a classification category corresponding to the target user; The fused sparse autoencoder comprises a first symmetric neural network and a second symmetric neural network, wherein the first symmetric neural network and the second symmetric neural network respectively comprise different numbers of hidden layers, and the connection relationship matrix is ​​input into the fused sparse autoencoder, and the connection relationship matrix is ​​reduced in dimension by the fused sparse autoencoder to obtain a target feature vector, comprising: Inputting the connection relationship matrix into the first symmetric neural network, performing feature extraction on the connection relationship matrix through the first symmetric neural network to obtain a first feature vector; Inputting the connection relationship matrix into the second symmetric neural network, performing feature extraction on the connection relationship matrix through the second symmetric neural network to obtain a second feature vector; The first feature vector and the second feature vector are fused to obtain the target feature vector.

2. The method for classifying resting state functional magnetic resonance imaging images of the brain according to claim 1, characterized in that: The obtaining of a number of time series data includes: Acquiring the resting state functional magnetic resonance imaging image of the target user; Acquire a preset segmentation template, and segment the resting state functional magnetic resonance imaging image according to the segmentation template to obtain a plurality of local images, wherein the plurality of local images respectively correspond to different brain regions; Filtering is performed on the plurality of local images respectively to obtain the plurality of time series data.

3. The classification method of resting state functional magnetic resonance imaging of the brain according to claim 1, characterized in that: The brain functional connection network is constructed based on a group constrained sparse model, and the first objective function corresponding to the group constrained sparse model includes a group sparse regularization parameter and a smoothing regularization parameter.

4. The method for classifying resting state functional magnetic resonance imaging images of the brain according to claim 1, characterized in that: The first symmetric neural network includes a plurality of cascaded hidden layers, and the second symmetric neural network includes one hidden layer.

5. The method for classifying resting state functional magnetic resonance imaging images of the brain according to claim 4, characterized in that: The second objective function corresponding to each hidden layer includes a Euclidean norm.

6. The method for classifying resting state functional magnetic resonance imaging images of the brain according to claim 1, characterized in that: The step of classifying the target feature vector to obtain a user label corresponding to the target user includes: Inputting the target feature vector into a preset classifier; The target feature vector is classified by the classifier to obtain the classification category.

7. A device for classifying resting state functional magnetic resonance imaging images of the brain, the device being used to implement the method for classifying resting state functional magnetic resonance imaging images of the brain as claimed in claim 1, characterized in that: The device comprises: A data acquisition module, used to acquire a plurality of time series data, wherein the plurality of time series data are generated based on a resting state functional magnetic resonance imaging image of a target user's brain, and the plurality of time series data correspond to different brain regions of the target user's brain; A data extraction module, used for performing feature extraction on the time series data to obtain a connection relationship matrix, wherein the connection relationship matrix is ​​used to reflect the connection relationship between the brain regions; The data classification module is used to extract features from the connection relationship matrix to obtain a target feature vector, and classify the target feature vector to obtain a classification category corresponding to the target user.

8. A terminal, characterized in that: The terminal includes a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the classification method of resting state functional magnetic resonance imaging images of the brain as described in any one of claims 1-6; and the processor is used to execute the program.

9. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are suitable for being loaded and executed by a processor to implement the steps of the method for classifying resting state functional magnetic resonance imaging images of the brain as described in any one of claims 1 to 6.

Images