阅读说明：本技术 脑图像处理方法、计算机设备和可读存储介质 (Brain image processing method, computer device, and readable storage medium ) 是由 邢潇丹 石峰 于 2019-09-25 设计创作，主要内容包括：本发明涉及一种脑图像处理方法、计算机设备和可读存储介质,该方法包括：获取脑功能图像；从脑功能图像中获取各脑区的时域特征信息并进行傅里叶变换,得到各脑区的节点特征；节点特征包括频域实部特征和频域虚部特征；根据各脑区的节点特征,获取各脑区间的连接信息,将连接信息作为节点间的连接；将节点特征和节点间的连接构建成图特性矩阵；将图特性矩阵输入训练模型,得到分析结果,其中,训练模型为样本脑功能图像构建的样本图特性矩阵,输入图网络中训练得到的模型。该方法中是将各脑区的时域特征信息进行傅里叶变换得到的频域特征作为各脑区的节点特征,能够更好地对脑功能图像中的噪声进行区分,提高了得到的分析结果的准确度。(The invention relates to a brain image processing method, a computer device and a readable storage medium, wherein the method comprises the following steps: acquiring a brain function image; acquiring time domain characteristic information of each brain area from the brain function image and performing Fourier transform to obtain node characteristics of each brain area; the node features comprise frequency domain real part features and frequency domain imaginary part features; acquiring connection information of each brain area according to the node characteristics of each brain area, and using the connection information as connection between nodes; constructing the node characteristics and the connection among the nodes into a graph characteristic matrix; and inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by the sample brain function image, and the model obtained by training in a graph network is input. According to the method, the frequency domain characteristics obtained by performing Fourier transform on the time domain characteristic information of each brain area are used as the node characteristics of each brain area, so that the noise in the brain function image can be better distinguished, and the accuracy of the obtained analysis result is improved.)

1. A method of brain image processing, the method comprising:

acquiring a brain function image;

acquiring time domain feature information of each brain area from the brain function image and performing Fourier transform to obtain node features of each brain area; the node features comprise frequency domain real part features and frequency domain imaginary part features;

acquiring connection information of each brain area according to the node characteristics of each brain area, and using the connection information as the connection between the nodes;

constructing a graph characteristic matrix by using the node characteristics and the connection between the nodes;

and inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by a sample brain function image and is input into a model obtained by training in a graph network.

2. The method of claim 1, wherein the graph network comprises a graph convolution network.

3. The method according to claim 2, wherein the obtaining connection information between the brain regions according to the node characteristics of the brain regions, and using the connection information as the connection between the nodes comprises:

and calculating the Pearson correlation coefficient between the node characteristics of any two brain areas to obtain the connection information of each brain area, and taking the connection information as the connection between the nodes.

4. The method of claim 2, wherein the graph convolution network comprises at least one graph convolution layer, a fully-connected layer, a modulo computation layer, and a classification function layer; the full-connection layer is used for performing full-connection processing on the frequency domain real part characteristic and the frequency domain imaginary part characteristic output by the graph volume layer respectively to obtain a full-connection processing result; the module calculation layer is used for performing module calculation processing on a real part and an imaginary part in the full-connection processing result to obtain a characteristic value of each brain area; and the classification function layer is used for calculating the classification result of each brain region according to the characteristic value of each brain region.

5. The method of claim 2 or 4, wherein the convolution operation of the graph convolution layer in the graph convolution network comprises:

in the formula (I), the compound is shown in the specification,

6. The method according to claim 1, wherein the obtaining time-domain feature information of each brain region from the brain function image and performing fourier transform to obtain the node feature of each brain region comprises:

segmenting the brain function image according to a preset brain partition template to obtain an average value of blood oxygen concentration dependent contrast signals of all voxels in each brain area;

and carrying out Fourier transform on each average value to obtain the node characteristics of each brain area.

7. The method of claim 1, wherein the training process of the training model comprises:

acquiring a sample brain function image;

acquiring time domain characteristic information of each brain area from the sample brain function image and performing Fourier transform to obtain sample node characteristics of each brain area; the sample node features comprise real frequency-domain sample features and imaginary frequency-domain sample features;

acquiring sample connection information of each brain region according to the sample node characteristics of each brain region, and taking the sample connection information as the connection between the sample nodes;

constructing the sample node characteristics and the connection between the sample nodes into a sample graph characteristic matrix;

and inputting the sample graph characteristic matrix into the graph network for training to obtain the training model.

8. The method according to claim 7, wherein obtaining sample connection information between the brain regions according to the sample node characteristics of the brain regions, and using the sample connection information as the connection between the sample nodes, comprises:

and calculating a Pearson correlation coefficient between sample node characteristics of any two brain areas to obtain sample connection information of each brain area, and taking the sample connection information as the connection between the sample nodes.

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method according to any of claims 1-8.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 8.

Technical Field

The present invention relates to the field of images, and in particular, to a brain image processing method, a computer device, and a readable storage medium.

Background

Functional Magnetic Resonance (FMRI) detects brain activity by measuring Blood oxygen-level dependent contrast (BOLD) in Blood. There are two different ways to analyze the FMRI signal clinically, one is to analyze the connection characteristics of the brain from the perspective of the brain network by calculating the correlation of the time sequence signal between two brain areas; the other method is to detect the brain activity by extracting the low frequency Amplitude (ALFF), local consistency (ReHo) and other indicators of the FMRI signal, and classifying the FMRI signal by using a simple classifier, such as a support vector machine, a decision tree, and the like. In recent years, with the development of neural network technology, researchers have begun to classify FMRI signals using Graph Convolutional neural Networks (GCNs).

In the conventional technology, the FMRI signals are classified by using the GCN, and generally, each brain region is used as a node in a graph structure, and a connection between each brain region is used as a connection in the graph structure, so that the FMRI signals are classified. However, this method only considers the time-domain features contained in the FMRI signal, which are relatively non-quantitative and cannot classify the FMRI signal more accurately.

Therefore, the conventional technology has a problem that FMRI signals cannot be classified relatively accurately.

Disclosure of Invention

In view of the above, it is necessary to provide a brain image processing method, a computer device and a readable storage medium for solving the problem that the conventional technology cannot classify FMRI signals more accurately.

In a first aspect, an embodiment of the present invention provides a method for processing a brain image, where the method includes:

acquiring a brain function image;

acquiring time domain feature information of each brain area from the brain function image and performing Fourier transform to obtain node features of each brain area; the node features comprise frequency domain real part features and frequency domain imaginary part features;

acquiring connection information of each brain area according to the node characteristics of each brain area, and using the connection information as the connection between the nodes;

constructing a graph characteristic matrix by using the node characteristics and the connection between the nodes;

and inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by a sample brain function image and is input into a model obtained by training in a graph network.

In one embodiment, the graph network comprises a graph convolution network.

In one embodiment, the obtaining connection information between the brain regions according to the node characteristics of the brain regions, and using the connection information as the connection between the nodes, includes:

and calculating the Pearson correlation coefficient between the node characteristics of any two brain areas to obtain the connection information of each brain area, and taking the connection information as the connection between the nodes.

In one embodiment, the graph convolution network comprises at least one graph convolution layer, a full-link layer, a modulo computation layer, and a classification function layer; the full-connection layer is used for performing full-connection processing on the frequency domain real part characteristic and the frequency domain imaginary part characteristic output by the graph volume layer respectively to obtain a full-connection processing result; the module calculation layer is used for performing module calculation processing on a real part and an imaginary part in the full-connection processing result to obtain a characteristic value of each brain area; and the classification function layer is used for calculating the classification result of each brain region according to the characteristic value of each brain region.

In one embodiment, the convolution operation of the graph convolution layer in the graph convolution network includes:

in the formula (I), the compound is shown in the specification,

the frequency domain real part characteristic input for the graph convolution layer l,

for the frequency domain imaginary part characteristic of the input of the graph convolution layer l,

for the real part parameters of the volume layer l,

for the imaginary parameters of the map convolution layer/,

for the frequency domain real part feature output by the graph convolution layer l,

frequency domain imaginary part characteristic, A, output for said graph convolution layer l_rIs the real part of the connection between the nodes, A_cIs the imaginary part of the connection between the nodes.

In one embodiment, the obtaining time-domain feature information of each brain region from the brain function image and performing fourier transform to obtain node features of each brain region includes:

segmenting the brain function image according to a preset brain partition template to obtain an average value of blood oxygen concentration dependent contrast signals of all voxels in each brain area;

and carrying out Fourier transform on each average value to obtain the node characteristics of each brain area.

In one embodiment, the training process of the training model includes:

acquiring a sample brain function image;

acquiring time domain characteristic information of each brain area from the sample brain function image and performing Fourier transform to obtain sample node characteristics of each brain area; the sample node features comprise real frequency-domain sample features and imaginary frequency-domain sample features;

acquiring sample connection information of each brain region according to the sample node characteristics of each brain region, and taking the sample connection information as the connection between the sample nodes;

constructing the sample node characteristics and the connection between the sample nodes into a sample graph characteristic matrix;

and inputting the sample graph characteristic matrix into the graph network for training to obtain the training model.

In one embodiment, obtaining sample connection information of each brain region according to the sample node characteristics of each brain region, and using the sample connection information as a connection between the sample nodes includes:

and calculating a Pearson correlation coefficient between sample node characteristics of any two brain areas to obtain sample connection information of each brain area, and taking the sample connection information as the connection between the sample nodes.

In a second aspect, an embodiment of the present invention provides a brain image processing apparatus, including:

the first acquisition module is used for acquiring a brain function image;

the second acquisition module is used for acquiring time domain characteristic information of each brain area from the brain function image and performing Fourier transform to obtain node characteristics of each brain area; the node features comprise frequency domain real part features and frequency domain imaginary part features;

a third obtaining module, configured to obtain connection information between the brain areas according to node features of the brain areas, where the connection information is used as a connection between the nodes;

a first constructing module, configured to construct a graph property matrix from the node features and the connections between the nodes;

and the analysis module is used for inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by a sample brain function image and is input into a model obtained by training in a graph network.

In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and the processor implements the following steps when executing the computer program:

acquiring a brain function image;

acquiring time domain feature information of each brain area from the brain function image and performing Fourier transform to obtain node features of each brain area; the node features comprise frequency domain real part features and frequency domain imaginary part features;

acquiring connection information of each brain area according to the node characteristics of each brain area, and using the connection information as the connection between the nodes;

constructing a graph characteristic matrix by using the node characteristics and the connection between the nodes;

and inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by a sample brain function image and is input into a model obtained by training in a graph network.

In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the following steps:

acquiring a brain function image;

acquiring time domain feature information of each brain area from the brain function image and performing Fourier transform to obtain node features of each brain area; the node features comprise frequency domain real part features and frequency domain imaginary part features;

acquiring connection information of each brain area according to the node characteristics of each brain area, and using the connection information as the connection between the nodes;

constructing a graph characteristic matrix by using the node characteristics and the connection between the nodes;

and inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by a sample brain function image and is input into a model obtained by training in a graph network.

In the brain image processing method, the apparatus, the computer device, and the readable storage medium provided in the above embodiments, the computer device obtains a brain function image, obtains time domain feature information of each brain region from the brain function image, and performs fourier transform to obtain a node feature of each brain region; the node characteristics comprise frequency domain real part characteristics and frequency domain imaginary part characteristics, connection information of all brain regions is obtained according to the node characteristics of all brain regions, the connection information is used as connection between the nodes, the node characteristics and the connection between the nodes are constructed into a graph characteristic matrix, and the graph characteristic matrix is input into a training model to obtain an analysis result. In the method, the graph characteristic matrix input into the training model is constructed by the node characteristics of each brain area and the connection between nodes, the node characteristics of each brain area are obtained by acquiring the time domain characteristic information of each brain area from a brain function image and performing Fourier transform, the node characteristics of each brain area comprise frequency domain real part characteristics and frequency domain imaginary part characteristics, and the frequency domain characteristics can better distinguish noise in the brain function image, so that the graph characteristic matrix can be analyzed more comprehensively and accurately, and the accuracy of the obtained analysis result is improved; in addition, the graph characteristic matrix can be analyzed rapidly by using the training model, and the efficiency of obtaining an analysis result is improved.

Drawings

FIG. 1 is a schematic diagram of an internal structure of a computer device according to an embodiment;

fig. 2 is a flowchart illustrating a method for processing a brain image according to an embodiment;

FIG. 3 is a process diagram for constructing a graph property matrix, according to an embodiment;

FIG. 4 is a diagram illustrating a graph convolution network architecture, according to an exemplary embodiment;

FIG. 5 is a diagram illustrating a diagram of a convolutional layer structure, according to an embodiment;

FIG. 6 is a schematic diagram of a method for processing a brain image according to another embodiment;

fig. 7 is a flowchart illustrating a method for processing a brain image according to another embodiment;

fig. 8 is a schematic structural diagram of a brain image processing apparatus according to an embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The brain image processing method provided by the embodiment of the application can be applied to the computer equipment shown in fig. 1. The computer device comprises a processor and a memory connected by a system bus, wherein a computer program is stored in the memory, and the steps of the method embodiments described below can be executed when the processor executes the computer program. Optionally, the computer device may further comprise a network interface, a display screen and an input device. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a nonvolatile storage medium storing an operating system and a computer program, and an internal memory. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. Optionally, the computer device may be a server, a personal computer, a personal digital assistant, other terminal devices such as a tablet computer, a mobile phone, and the like, or a cloud or a remote server, and the specific form of the computer device is not limited in the embodiment of the present application.

In the brain image processing method provided in the embodiment of the present application, the execution subject may be a brain image processing apparatus, and the brain image processing apparatus may be implemented as part or all of a computer device by software, hardware, or a combination of software and hardware. In the following method embodiments, the execution subject is a computer device as an example.

The brain image processing method provided in the embodiment of the present application may be used for auxiliary diagnosis of early Mild Cognitive Impairment (MCI), and may also be used for auxiliary diagnosis of senile dementia, obsessive-compulsive disorder, autism, and the like. The following describes the technical solution of the present invention and how to solve the above technical problems with specific examples. The following several specific embodiments may be combined with each other, and details of the same or similar concepts or processes may not be repeated in some embodiments.

Fig. 2 is a flowchart illustrating a brain image processing method according to an embodiment. FIG. 3 is a process diagram for constructing a graph property matrix, according to an embodiment. The embodiment relates to a specific implementation process of constructing a graph characteristic matrix according to a brain function image and inputting the graph characteristic matrix into a training model to obtain an analysis result by computer equipment. As shown in fig. 2, the method may include:

s201, acquiring a brain function image.

The brain function image is a Functional Magnetic resonance image of the brain of the subject obtained by Functional Magnetic Resonance (FMRI). Alternatively, the computer device may acquire the brain function image from a PACS (Picture Archiving and Communication Systems) server, or may acquire the brain function image from an FMRI imaging device. Optionally, the computer device may further perform at least one of a temporal registration process, a head movement correction process, a normalization process, and a real space filtering process on the acquired brain function image.

S202, acquiring time domain characteristic information of each brain area from the brain function image and performing Fourier transform to obtain node characteristics of each brain area; the node features include frequency domain real part features and frequency domain imaginary part features.

It will be appreciated that in FMRI scanning of a subject's brain, a stack of functional images can be scanned in a short time, and therefore the resulting functional brain images have a high temporal resolution and a low spatial resolution. The time domain features contained in the brain function image are relative and non-quantitative, and the time domain features of the brain function image can be converted into frequency domain features through Fourier transform. Specifically, the computer device acquires time domain feature information of each brain region from the brain function image, and performs fourier transform on the acquired time domain feature information to obtain node features of each brain region, including frequency domain real part features and frequency domain imaginary part features.

S203, acquiring connection information of each brain area according to the node characteristics of each brain area, and using the connection information as connection between nodes;

specifically, the computer equipment is based onAnd calculating the connection strength between every two brain areas according to the node characteristics of the brain areas, acquiring the connection information between the brain areas, and taking the acquired connection information as the connection between the nodes. It is understood that the node feature of each brain region includes a frequency-domain real part feature of each brain region and a frequency-domain imaginary part feature of each brain region, and if the node feature of each brain region is represented by a matrix X, X ═ X_r+iX_cIn the formula, X_rRepresenting the real part of the frequency domain, X, of each brain region_cRepresenting the frequency domain imaginary part characteristics of each brain region, the connection information of each brain region can be represented as a ═ a_r+iA_cIn the formula, A_rReal part, A, representing connection information of each brain region_cRepresenting the imaginary part of the connection information between the brain regions.

And S204, constructing the node characteristics and the connection between the nodes into a graph characteristic matrix.

Specifically, after obtaining the node characteristics and the connections between the nodes, the computer device constructs the node characteristics and the connections between the nodes into a graph characteristic matrix. That is, the computer device may use, as the node feature, a feature obtained by performing fourier transform on the time domain feature information of each brain region, and use the connection information of each brain region as the connection between nodes, to construct the graph characteristic matrix. Illustratively, taking an example of dividing a brain function image into 4 brain areas as an example, acquiring time domain feature information of each 4 brain areas from the brain function image, performing fourier transform on the acquired time domain feature information to obtain node features of each 4 brain areas, acquiring connection information of each brain area according to the node features of each 4 brain areas, using the acquired connection information as connection between the nodes, constructing the connection between the node features and the nodes into a graph property matrix X,

wherein v is₁₁,v₂₂,v₃₃,v₄₄Respectively, feature information of each of 4 brain regions, and v₁₁,v₂₂,v₃₃,v₄₄Are all plural, v₁₂,v₂₁Is a functional connection of a first brain region and a second brain regionReceiving information, v₁₃,v₃₁Is the functional connection information of the first brain region and the third brain region, v₁₄,v₄₁Is the functional connection information of one brain region and four brain regions, v₂₃,v₃₂Is the functional connection information of the two brain region and the three brain region, v₂₄,v₄₂Is the functional connection information of the two brain region and the four brain region, v₃₄,v₄₃Is the functional connection information of the three brain region and the four brain region, and v₁₂,v₂₁,v₁₃,v₃₁,v₁₄,v₄₁,v₂₃,v₃₂,v₂₄,v₄₂,v₃₄,v₄₃Are all plural.

And S205, inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by the sample brain function image and is input into a model obtained by training in a graph network.

Specifically, the computer device inputs the constructed graph characteristic matrix into a training model to obtain an analysis result. The training model is a sample image characteristic matrix constructed by sample brain function images and is input into a model obtained by training in an image network. Illustratively, when analyzing whether the brain function image of the subject is a brain function image corresponding to an early MCI patient, optionally, the analysis result may be normal or early MCI.

In this embodiment, the graph characteristic matrix input into the training model is constructed by the node features of each brain region and the connection between nodes, the node features of each brain region are obtained by acquiring the time domain feature information of each brain region from the brain function image and performing fourier transform, the node features of each brain region include the frequency domain real part feature and the frequency domain imaginary part feature, and the frequency domain feature can better distinguish noise in the brain function image, so that the graph characteristic matrix can be analyzed more comprehensively and accurately, and the accuracy of the obtained analysis result is improved; in addition, the graph characteristic matrix can be analyzed rapidly by using the training model, and the efficiency of obtaining an analysis result is improved.

On the basis of the above embodiment, as an alternative implementation, the graph network includes a graph volume network.

Specifically, the graph network that inputs the constructed sample graph characteristic matrix into the graph network for training includes a graph convolution network. It can be understood that the graph convolution network performs information interaction on nodes in the graph by means of matrix operation, and therefore, the graph convolution network model is also referred to as an information interaction model. For a graph structure, defining a node feature matrix as X ∈ R^M×NWherein M is the number of nodes, and N is the characteristic quantity of each node; defining its connection matrix A ∈ R^M×MEach element in the matrix is the connection between corresponding nodes, and at the l-th layer of the neural network, the graph volume model based on information interaction is defined as follows:

wherein D is a metric matrix of the connection matrix A,

the method is used for normalizing a connection matrix, ReLU is an activation function and is used for introducing a nonlinear part in a graph volume network, and theta is a parameter needing graph volume network learning. Optionally, after the analysis result is obtained through the graph network, the obtained analysis result may be classified by using any one of a Support Vector Machine (SVM), an over-limit learning Machine (ELM), a decision tree, a random forest method, logistic regression, and ridge regression. Optionally, the graph network may further include a graph recursion network, a graph attention network, or a graph generation network.

In this embodiment, the graph network that inputs the constructed sample graph characteristic matrix into the graph network for training includes a graph convolution network, the graph convolution network can process connection between a plurality of node features and nodes, and can better analyze and process the sample graph characteristic matrix, and the training model is a model obtained by training the graph network, so that the training model can better analyze and process the graph characteristic matrix.

On the basis of the foregoing embodiment, as an optional implementation manner, the foregoing S203 includes: and calculating the Pearson correlation coefficient between the node characteristics of any two brain areas to obtain the connection information between the brain areas, and taking the connection information as the connection between the nodes.

Specifically, the computer device obtains connection information between the brain regions by calculating a pearson correlation coefficient between node features of any two brain regions, and arranges the obtained connection information according to the ordinal numbers of the brain regions to be used as the connection between the nodes. The pearson correlation coefficient is also called a pearson product moment correlation coefficient and is a linear correlation coefficient, the pearson correlation coefficient is a statistic for reflecting the linear correlation degree between the node characteristics of any two brain regions, the degree of the linear correlation strength between the node characteristics of any two brain regions is described, and the larger the absolute value of the pearson correlation coefficient is, the stronger the correlation between the node characteristics of the two brain regions is.

In this embodiment, the computer device obtains connection information between the brain regions by calculating a pearson correlation coefficient between node features of any two brain regions, and uses the connection information as the connection between the nodes, so that the characteristics of the brain function image can be better reflected, the constructed map characteristic matrix is more accurate, and the map characteristic matrix can be more accurately analyzed.

Fig. 4 is a schematic diagram of a graph convolution network structure according to an embodiment. FIG. 5 is a diagram illustrating a structure of a graph convolution layer according to an embodiment. On the basis of the above embodiment, as an optional implementation manner, the graph convolution network includes at least one graph convolution layer, a full connection layer, a modulo computation layer, and a classification function layer; the full-connection layer is used for performing full-connection processing on the frequency domain real part characteristic and the frequency domain imaginary part characteristic output by the graph volume layer respectively to obtain a full-connection processing result; the module calculation layer is used for performing module calculation processing on a real part and an imaginary part in the full-connection processing result to obtain a characteristic value of each brain area; and the classification function layer is used for calculating to obtain a classification result of each brain region according to the characteristic value of each brain region.

Specifically, as shown in fig. 4, the graph convolution network includes at least one graph convolution layer, a full connection layer, a modulo layer, and a classification function layer. The full-connection layer is used for performing full-connection processing on the frequency domain real part characteristic and the frequency domain imaginary part characteristic output by the graph convolution layer respectively to obtain a full-connection processing result; the module calculation layer is used for performing module calculation processing on a real part and an imaginary part in the full-connection processing result to obtain a characteristic value of each brain area; and the classification function layer is used for calculating to obtain a classification result of each brain region according to the characteristic value of each brain region.

Alternatively, as shown in fig. 5, the graph convolution layer may include a plurality of graph convolution layers, a plurality of pooling layers, a batch normalization layer, and an activation function layer. It will be appreciated that since batch normalization of complex numbers involves the computation of exponentials and trigonometric functions, the real and imaginary parts of the complex numbers can be normalized separately before batch normalization of the complex numbers is performed.

Specifically, as can be seen from the above description, the graph structure is at the l-th layer of the neural network, and the graph convolution network model is defined as:

considering the existence of complex numbers, the convolution operation for obtaining the complex number convolution layer in the graph convolution network according to the definition comprises the following steps:

in the formula (I), the compound is shown in the specification,

the frequency domain real part characteristic of the input of the graph convolution layer l,

for the frequency domain imaginary feature of the graph convolution layer input,

the real part parameters of the volume layer l,

as the imaginary parameters of the map convolution layer l,

is the frequency domain real part characteristic output by the graph volume layer l,

frequency domain imaginary part characteristic, A, output for graph convolution layer_rIs the real part of the connection between nodes, A_cIs the imaginary part of the connection between the nodes.

At layer l of the neural network, the pooling operation of the graph is defined as follows:

wherein, S is a pooling matrix to be learned and used for pooling the connection information of each brain interval, and Z is an embedded matrix and used for calculating the pooled node characteristics. S and Z are calculated as follows:

it will be appreciated that for a matrix V of size P Q, soft max is calculated by the formula:

this can result in V for a complex matrix of size P × Q_r+iV_cThe calculation formula of softmax is:

it will be appreciated that by multiplication of the matrix, the following batch normalization can be defined:

from this, it is possible to obtain a matrix having covariance for complex numbers

Since V is a semi-positive definite matrix, it can be guaranteed

Meaningful, convolutional neural networks are being calculated

Then, the scale transformation gamma and the offset beta are needed to be learned, and the output of the batch normalization layer is

In the output of our definition batch normalization layer

For the activation function layer, a modReLU function is employed in this embodiment,

where b is a parameter to be learned. Optionally, the activation function may also adopt a CReLU function, zReLU function, etc., where CReLU (x) equals ReLU (x)_r)+iReLU(x_c)，

In this embodiment, the convolutional network includes at least one convolutional layer, a full connection layer, a modulo computation layer, and a classification function layer, where the full connection layer can be used to perform full connection processing on frequency domain real part features and frequency domain features output by the convolutional layer, the modulo computation layer can be used to perform modulo computation on a real part and an imaginary part in a full connection processing result to obtain feature values of each brain region, the classification function layer can be used to calculate classification results of each brain region according to the feature values of each brain region, and a complex graph characteristic matrix can be processed and analyzed through the network structure to obtain an analysis result, thereby expanding a use scene of the convolutional network and also being capable of processing the complex graph characteristic matrix.

Fig. 6 is a schematic diagram of a brain image processing method according to another embodiment. The embodiment relates to a specific implementation process for obtaining node characteristics of each brain area by computer equipment. As shown in fig. 6, on the basis of the foregoing embodiment, as an optional implementation manner, the foregoing S202 includes:

s601, segmenting the brain function image according to a preset brain partition template, and obtaining the average value of the blood oxygen concentration dependent contrast signals of all voxels in each brain area.

Specifically, the computer device segments the brain function image according to a preset brain partition template, and obtains an average value of Blood oxygen-level dependent contrast (BOLD) signals of all voxels in each brain region. Optionally, the preset brain partition template may be an Anatomical Automatic Labeling (AAL) template, or may be another brain partition template, such as an SRI24 template. Alternatively, the computer device may divide the brain structure image into 116 brain regions according to a preset brain region template.

And S602, performing Fourier transform on each average value to obtain node characteristics of each brain area.

Specifically, the computer device performs fourier transform on the obtained average value of the BOLD signals of all voxels in each brain region to obtain the node characteristics of each brain region. It is understood that, after performing fourier transform, a complex number is introduced into the node features of each brain region, and the obtained node features of each brain region are node features including a real part and an imaginary part.

In this embodiment, the computer device segments the acquired brain function image according to a preset brain partition template, acquires an average value of blood oxygen concentration dependent contrast signals of all voxels in each brain region, and performs fourier transform on each acquired average value to obtain node features of each brain region.

Fig. 7 is a flowchart illustrating a method for processing a brain image according to another embodiment. The embodiment relates to a specific implementation process for training a training model by computer equipment. As shown in fig. 7, the training process of training the model may include:

and S701, acquiring a sample brain function image.

Alternatively, the computer device may obtain the sample brain function image from a PACS (Picture Archiving and communications systems) server, or may obtain the sample brain function image from an FMRI imaging device. Optionally, the computer device may further perform at least one of a temporal registration process, a cranial movement correction process, a normalization process, and a real space filtering process on the acquired sample brain function images.

S702, acquiring time domain characteristic information of each brain area from the sample brain function image and performing Fourier transform to obtain sample node characteristics of each brain area; the sample node features include frequency domain sample real features and frequency domain sample imaginary features.

Specifically, the computer device divides the sample brain function image according to a preset brain partition template, obtains time domain feature information of each brain area from the sample brain function image, and performs fourier transform to obtain sample node features of each brain area. Alternatively, the preset brain partition template may be an AAL template, or may be another brain partition template, for example, an SRI24 template. Optionally, the computer device may divide the sample brain function image into 116 brain regions according to a preset brain partition template, obtain time domain features of the 116 brain regions, and perform fourier transform to obtain sample node features of the brain regions.

And S703, acquiring sample connection information of each brain region according to the sample node characteristics of each brain region, and taking the sample connection information as the connection between the sample nodes.

Specifically, the computer device calculates the connection strength between every two brain areas according to the sample node characteristics of each brain area, obtains the connection information between each brain area, and uses the obtained connection information as the connection between the sample nodes. It is understood that the sample node characteristics of each brain region include frequency-domain real-part characteristics of each brain region and frequency-domain imaginary-part characteristics of each brain region, and the obtained connections between sample nodes include connections between sample nodes of real parts and connections between sample nodes of imaginary parts.

And S704, constructing the sample node characteristics and the connection between the sample nodes into a sample graph characteristic matrix.

Specifically, after obtaining the sample node features of each brain region and the connections between the sample nodes, the computer device constructs the connections between the sample node features and the sample nodes into a sample graph characteristic matrix. That is, the computer device may use, as the node feature, a feature obtained by performing fourier transform on the time domain feature information of each brain region, and use the connection information of each brain region as the connection between nodes, to construct the graph characteristic matrix.

S705, inputting the sample graph characteristic matrix into a graph network for training to obtain a training model.

Specifically, the computer device inputs the sample graph characteristic matrix into the graph network, and trains the graph network to obtain a training model. The computer device takes the sample map characteristic matrix as input, takes the analysis result of the sample map characteristic matrix as output, and trains the map network according to the analysis result of the sample map characteristic matrix and the labeled sample image to obtain a training model. Optionally, the graph network may be a graph convolution network, or may be a graph recursion network, a graph attention network, or a graph generation network.

In this embodiment, the computer device inputs a sample map characteristic matrix into a map network for training, so as to obtain a training model, where the sample map characteristic matrix is obtained by obtaining time-domain characteristic information of each brain region from a sample brain function image and performing fourier transform as a sample node characteristic, the sample node characteristic is a complex-form characteristic, obtaining sample connection information of each brain region according to the sample node characteristic of each brain region, and using the sample connection information as connection between sample nodes, and the connection between the sample nodes is also a complex-form characteristic, so that the sample map characteristic matrix can reflect frequency-domain characteristic information of the sample brain function image, and thus, inputting the sample map characteristic matrix into the map network for training can improve the accuracy of the obtained training model, so that the obtained training model is more accurate; the sample graph characteristic matrix of the input graph network is a matrix of complex numbers, and the obtained training model can process the complex numbers by training the graph network.

On the basis of the foregoing embodiment, as an optional implementation manner, the foregoing S703 includes: and calculating the Pearson correlation coefficient between the sample node characteristics of any two brain areas to obtain the sample connection information of each brain area, and taking the sample connection information as the connection between the sample nodes.

Specifically, the computer device obtains sample connection information of each brain region by calculating a pearson correlation coefficient between sample node features of any two brain regions, and arranges the obtained sample connection information according to the ordinal numbers of the brain regions to be used as the connection between the sample nodes.

In this embodiment, the computer device obtains sample connection information of each brain region by calculating a pearson correlation coefficient between sample node features of any two brain regions, and uses the sample connection information as connection between sample nodes, so that characteristics of a sample brain function image can be better reflected, a constructed sample map characteristic matrix is more accurate, and the sample map characteristic matrix can be more accurately analyzed.

It should be understood that although the various steps in the flow charts of fig. 2-7 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2-7 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternating with other steps or at least some of the sub-steps or stages of other steps.

Fig. 8 is a schematic structural diagram of a brain image processing apparatus according to an embodiment. As shown in fig. 8, the apparatus may include: a first acquisition module 10, a second acquisition module 11, a third acquisition module 12, a first construction module 13 and an analysis module 14.

Specifically, the first acquiring module 10 is configured to acquire a brain function image;

the second obtaining module 11 is configured to obtain time-domain feature information of each brain region from the brain function image and perform fourier transform on the time-domain feature information to obtain a node feature of each brain region; the node features comprise frequency domain real part features and frequency domain imaginary part features;

a third obtaining module 12, configured to obtain connection information between brain regions according to node features of the brain regions, and use the connection information as connection between nodes;

a first constructing module 13, configured to construct a graph characteristic matrix from the node features and the connections between nodes;

and the analysis module 14 is configured to input the graph characteristic matrix into a training model to obtain an analysis result, where the training model is a sample graph characteristic matrix constructed for the sample brain function image, and inputs the model obtained through training in a graph network.

Optionally, the graph network comprises a graph convolution network.

Optionally, the graph convolution network includes at least one graph convolution layer, a full connection layer, a modular computation layer, and a classification function layer; the full-connection layer is used for performing full-connection processing on the frequency domain real part characteristic and the frequency domain imaginary part characteristic output by the graph volume layer respectively to obtain a full-connection processing result; the module calculation layer is used for performing module calculation processing on a real part and an imaginary part in the full-connection processing result to obtain a characteristic value of each brain area; and the classification function layer is used for calculating to obtain a classification result of each brain region according to the characteristic value of each brain region.

Optionally, the convolution operation of the graph convolution layer in the graph convolution network includes:

in the formula (I), the compound is shown in the specification,

the frequency domain real part characteristic of the input of the graph convolution layer l,

for the frequency domain imaginary feature of the graph convolution layer input,

the real part parameters of the volume layer l,

as the imaginary parameters of the map convolution layer l,

is the frequency domain real part characteristic output by the graph volume layer l,

frequency domain imaginary part characteristic, A, output for graph convolution layer_rIs the real part of the connection between nodes, A_cIs the imaginary part of the connection between the nodes.

The brain image processing apparatus provided in this embodiment can execute the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.

On the basis of the foregoing embodiment, optionally, the third obtaining module 12 includes: a first acquisition unit.

Specifically, the first obtaining unit is configured to calculate a pearson correlation coefficient between node features of any two brain regions, obtain connection information between the brain regions, and use the connection information as connection between the nodes.

The brain image processing apparatus provided in this embodiment can execute the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.

On the basis of the foregoing embodiment, optionally, the second obtaining module 11 includes: a second acquisition unit and a third acquisition unit.

Specifically, the second obtaining unit is configured to segment the brain function image according to a preset brain partition template, and obtain an average value of blood oxygen concentration dependent contrast signals of all voxels in each brain region;

and the third acquisition unit is used for carrying out Fourier transform on each average value to obtain the node characteristics of each brain area.

The brain image processing apparatus provided in this embodiment can execute the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.

On the basis of the foregoing embodiment, optionally, the apparatus further includes: the device comprises a fourth acquisition module, a fifth acquisition module, a sixth acquisition module, a second construction module and a training module.

Specifically, the fourth acquisition module is used for acquiring a sample brain function image;

the fifth acquisition module is used for acquiring time domain characteristic information of each brain area from the sample brain function image and performing Fourier transform to obtain sample node characteristics of each brain area; the sample node characteristics comprise real part characteristics of the frequency domain samples and imaginary part characteristics of the frequency domain samples;

a sixth obtaining module, configured to obtain sample connection information between the brain regions according to sample node characteristics of each brain region, and use the sample connection information as connection between the sample nodes;

the second construction module is used for constructing the sample node characteristics and the connection between the sample nodes into a sample graph characteristic matrix;

and the training module is used for inputting the sample graph characteristic matrix into the graph network for training to obtain a training model.

The brain image processing apparatus provided in this embodiment can execute the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.

On the basis of the foregoing embodiment, optionally, the sixth obtaining module includes: and a fourth acquisition unit.

Specifically, the fourth obtaining unit is configured to calculate a pearson correlation coefficient between sample node features of any two brain regions, obtain sample connection information of each brain region, and use the sample connection information as connection between sample nodes.

The brain image processing apparatus provided in this embodiment can execute the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.

For specific limitations of the brain image processing apparatus, reference may be made to the above limitations of the brain image processing method, which are not described herein again. The modules in the brain image processing apparatus may be wholly or partially implemented by software, hardware, or a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:

acquiring a brain function image;

acquiring time domain characteristic information of each brain area from the brain function image and performing Fourier transform to obtain node characteristics of each brain area; the node features comprise frequency domain real part features and frequency domain imaginary part features;

acquiring connection information of each brain area according to the node characteristics of each brain area, and using the connection information as connection between nodes;

constructing the node characteristics and the connection among the nodes into a graph characteristic matrix;

and inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by the sample brain function image, and the model obtained by training in a graph network is input.

The implementation principle and technical effect of the computer device provided by the above embodiment are similar to those of the above method embodiment, and are not described herein again.

In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:

acquiring a brain function image;

acquiring time domain characteristic information of each brain area from the brain function image and performing Fourier transform to obtain node characteristics of each brain area; the node features comprise frequency domain real part features and frequency domain imaginary part features;

acquiring connection information of each brain area according to the node characteristics of each brain area, and using the connection information as connection between nodes;

constructing the node characteristics and the connection among the nodes into a graph characteristic matrix;

and inputting the graph characteristic matrix into a training model to obtain an analysis result, wherein the training model is a sample graph characteristic matrix constructed by the sample brain function image, and the model obtained by training in a graph network is input.

The implementation principle and technical effect of the computer-readable storage medium provided by the above embodiments are similar to those of the above method embodiments, and are not described herein again.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. 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. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.