Brain age prediction method, device, electronic device and storage medium

Abstract

The present invention is applicable to the field of image processing technology, and provides a brain age prediction method, device, equipment and medium. The brain age prediction device comprises a preprocessing unit for preprocessing three-dimensional brain magnetic resonance imaging data of a subject to be tested, and a brain age prediction unit for inputting the preprocessed imaging data into a brain age prediction network and outputting a brain age prediction result. The brain age prediction network comprises a first and a second basic convolution module, a first, a second and a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module which are sequentially connected. The brain age prediction network also comprises a first cross-layer auxiliary residual module whose input end and output end are respectively connected to the output end of the second basic convolution module and the input end of the second residual convolution module, and a second cross-layer auxiliary residual module whose input end and output end are respectively connected to the output end of the first residual convolution module and the input end of the third residual convolution module, thereby realizing brain age prediction.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and in particular relates to a brain age prediction method, device, electronic equipment and storage medium. Background Art

[0002] Magnetic resonance imaging (MRI) is now widely used. It can provide multimodal information such as structure, function, and metabolism, and is an indispensable tool in clinical diagnosis. Its advantages such as non-invasiveness, no radiation, high clinical penetration rate, and reasonable testing costs also make it an important choice for disease screening.

[0003] Currently, brain age assessment based on magnetic resonance imaging and early diagnosis and prediction technology of Alzheimer's disease are basically based on structural magnetic resonance imaging. Brain aging or disease conditions are evaluated by performing machine learning or deep learning analysis on morphological indicators of specific brain regions or the entire brain.

[0004] In terms of using artificial intelligence technology to predict brain age, current research mainly uses various methods to extract features from three-dimensional magnetic resonance data, and then inputs the extracted features into machine learning models for prediction, or directly uses convolutional neural networks in deep learning for training and prediction. There are two main difficulties in current research: one is how to effectively extract features from imaging data, and the other is how to choose a suitable machine learning model or neural network model to fit these features. Both of these will bring challenges to brain age prediction. Summary of the invention

[0005] The purpose of the present invention is to provide a brain age prediction method, device, electronic device and storage medium, aiming to solve the problem of how to select a suitable model to effectively extract features and predict brain age in the prior art.

[0006] In one aspect, the present invention provides a method for predicting brain age, the method comprising the following steps:

[0007] Preprocessing the three-dimensional brain magnetic resonance imaging data of the subject to be tested to obtain preprocessed imaging data;

[0008] The preprocessed image data is input into a pre-trained brain age prediction network to predict the brain age of the subject, wherein the brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module which are sequentially connected, and the brain age prediction network also includes a first and a second cross-layer auxiliary residual module, wherein the input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module, and the input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module.

[0009] Preferably, the step of preprocessing the three-dimensional brain magnetic resonance imaging data of the subject includes:

[0010] The three-dimensional brain magnetic resonance imaging data of the subject are registered to obtain registered imaging data.

[0011] Preferably, before preprocessing the three-dimensional brain magnetic resonance imaging data of the subject, the method includes:

[0012] Get the basic data set;

[0013] Preprocessing each three-dimensional brain magnetic resonance imaging data in the basic data set to obtain a training data set;

[0014] Based on the training data set, the brain age prediction network is trained using a stochastic gradient descent algorithm to obtain the trained brain age prediction network.

[0015] Preferably, the step of predicting the brain age of the subject comprises:

[0016] Extracting basic features of the preprocessed image data through the first and second basic convolution modules;

[0017] Based on the basic features, extracting deep features through the first, second, and third residual convolution modules and the first and second cross-layer auxiliary residual modules;

[0018] Unifying the data size of the deep features through the two-dimensional maximum pooling module;

[0019] Based on the deep features after the data size is unified, the brain age prediction result is output through the fully connected module and the sigmoid function module.

[0020] Preferably, the first basic convolution module and the second basic convolution module have the same structure, the first basic convolution module includes three basic convolution blocks connected in sequence and a two-dimensional maximum pooling layer, and each of the convolution blocks includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence.

[0021] Preferably, the first residual convolution module, the second residual convolution module and the third residual convolution module have the same structure, the first residual convolution module includes three residual blocks connected in sequence and a two-dimensional maximum pooling layer, the output value of the residual block is the sum of the input value of the top layer and the output value of the input value after passing through two basic convolution blocks, and the basic convolution block includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence.

[0022] Preferably, the first cross-layer auxiliary residual module and the second cross-layer auxiliary residual module have the same structure, and the first cross-layer auxiliary residual module includes a two-dimensional maximum pooling layer and an attention layer connected in sequence.

[0023] Preferably, the attention layer adopts SeNet.

[0024] In another aspect, the present invention provides a brain age prediction device, comprising:

[0025] A preprocessing unit, used for preprocessing the three-dimensional brain magnetic resonance imaging data of the subject to be tested to obtain preprocessed imaging data; and

[0026] A brain age prediction unit is used to input the preprocessed image data into a pre-trained brain age prediction network, and output the brain age prediction result of the subject, wherein the brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module connected in sequence, and the brain age prediction network also includes a first cross-layer auxiliary residual module and a second cross-layer auxiliary residual module, the input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module, the input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module.

[0027] Preferably, the preprocessing unit is used to register the three-dimensional brain magnetic resonance imaging data of the subject to be tested to obtain registered image data.

[0028] Preferably, the device further comprises a data set acquisition unit and a model training unit, wherein:

[0029] The data set acquisition unit is used to acquire a basic data set;

[0030] The preprocessing unit is further used to preprocess each three-dimensional brain magnetic resonance imaging data in the basic data set to obtain a training data set; and

[0031] The model training unit is used to train the brain age prediction network based on the training data set using a stochastic gradient descent algorithm to obtain the trained brain age prediction network.

[0032] Preferably, the brain age prediction unit further includes:

[0033] A basic feature extraction unit, used for extracting basic features of the preprocessed image data through the first and second basic convolution modules;

[0034] A deep feature extraction unit, configured to extract deep features based on the basic features through the first, second, and third residual convolution modules and the first and second cross-layer auxiliary residual modules;

[0035] A pooling unit, configured to unify the data size of the deep features through the two-dimensional maximum pooling module; and

[0036] The prediction unit is used to output the brain age prediction result through the fully connected module and the sigmoid function module based on the deep features after the data size is unified.

[0037] Preferably, the first basic convolution module and the second basic convolution module have the same structure, the first basic convolution module includes three basic convolution blocks connected in sequence and a two-dimensional maximum pooling layer, and each of the convolution blocks includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence.

[0038] Preferably, the first residual convolution module, the second residual convolution module and the third residual convolution module have the same structure, the first residual convolution module includes three residual blocks connected in sequence and a two-dimensional maximum pooling layer, the output value of the residual block is the sum of the input value of the top layer and the output value of the input value after passing through two basic convolution blocks, and the basic convolution block includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence.

[0039] Preferably, the first cross-layer auxiliary residual module and the second cross-layer auxiliary residual module have the same structure, and the first cross-layer auxiliary residual module includes a two-dimensional maximum pooling layer and an attention layer connected in sequence.

[0040] Preferably, the attention layer adopts SeNet.

[0041] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the functions of the above-mentioned device when executing the computer program.

[0042] On the other hand, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the functions of the above-mentioned device are realized.

[0043] In the present invention, the brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module which are connected in sequence. The brain age prediction network also includes a first and a second cross-layer auxiliary residual module. The input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module. The input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module. When predicting the brain age of a subject, the three-dimensional brain magnetic resonance imaging data of the subject is preprocessed to obtain preprocessed image data, and the preprocessed image data is input into a pre-trained brain age prediction network to predict the brain age of the subject, thereby realizing brain age prediction through the brain age prediction network. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1A is a flowchart of the implementation of the brain age prediction method provided in the first embodiment of the present invention;

[0045] Figure 1B is a schematic diagram of the structure of a brain age prediction network provided in Embodiment 1 of the present invention;

[0046] Figure 1C is a schematic diagram of the structure of the first basic convolution module provided in Embodiment 1 of the present invention;

[0047] Figure 1D is a schematic diagram of the structure of a first residual convolution module provided in Embodiment 1 of the present invention;

[0048] Figure 2 is a flowchart of the implementation of the training method of the brain age prediction network provided in the second embodiment of the present invention;

[0049] Figure 3 is a schematic diagram of the structure of a brain age prediction device provided in Embodiment 3 of the present invention; and

[0050] Figure 4 It is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0052] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments:

[0053] Embodiment 1:

[0054] Figure 1A The implementation process of the brain age prediction method provided in the first embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0055] In step S101, the three-dimensional brain magnetic resonance imaging data of the subject is preprocessed to obtain preprocessed image data.

[0056] The embodiments of the present invention are applicable to electronic devices, which may be terminal devices such as mobile phones, tablet computers, wearable devices, laptop computers, personal computers, servers, etc. The embodiments of the present application do not impose any restrictions on the specific types of electronic devices.

[0057] In an embodiment of the present invention, the three-dimensional brain magnetic resonance imaging data of the subject is obtained and preprocessed. Considering that the convolution kernel used in the brain age prediction network in this embodiment is two-dimensional, when the three-dimensional brain magnetic resonance imaging data of the subject is preprocessed, preferably, the three-dimensional brain magnetic resonance imaging data of the subject is aligned to eliminate the differences in size and structure between individual image data before the image data is input into the network. Specifically, the image data can be aligned to a preset standard template space, for example, in the MNI152 space. Taking the MNI152 space as an example, the size of the data after alignment is 91×109×91. When configuring the image data, the alignment tool used can be FSL, etc. Among them, FSL is a comprehensive FMRI, MRI and DTI brain imaging data analysis tool library, which can be used for the analysis and processing of FMRI, MRI and DTI brain imaging data.

[0058] In step S102, the preprocessed image data is input into a pre-trained brain age prediction network to predict the brain age of the subject.

[0059] In the embodiment of the present invention, Figure 1BAs shown, the brain age prediction network is a convolutional neural network constructed based on VGG-16 and residual ideas. The brain age prediction network includes a first basic convolution module 10, a second basic convolution module 11, a first residual convolution module 12, a second residual convolution module 13, a third residual convolution module 14, a two-dimensional maximum pooling module 15, a fully connected module 16 and a sigmoid function module 17 connected in sequence. The brain age prediction network also includes a first cross-layer auxiliary residual module 18 and a second cross-layer auxiliary residual module 19. The input end of the first cross-layer auxiliary residual module 18 is connected to the output end of the second basic convolution module 11, and the output end is connected to the input end of the second residual convolution module 13. The input end of the second cross-layer auxiliary residual module 19 is connected to the output end of the first residual convolution module 12, and the output end is connected to the input end of the third residual convolution module 14.

[0060] When predicting the brain age of the subject through the brain age prediction network, specifically, the basic features of the preprocessed image data are extracted through the first and second basic convolution modules, based on the basic features, the deep features are extracted through the first, second, and third residual convolution modules and the first and second cross-layer auxiliary residual modules, the data size of the deep features is unified through the two-dimensional maximum pooling module, based on the deep features after the unified data size, the brain age prediction result is output through the full connection module and the sigmoid function module, so as to predict the brain age of the subject through the constructed brain age prediction network. Among them, the first cross-layer auxiliary residual module and the second cross-layer auxiliary residual module are respectively used to obtain and transfer the potential features of the cross-layer, specifically, the first cross-layer auxiliary residual module takes the output features of the second basic convolution module as input, learns the potential features and passes them to the second residual convolution module, and the second cross-layer auxiliary residual module takes the output features of the first residual convolution module as input, learns the potential features and passes them to the third residual convolution module. When transferring the learned potential features, the output features of each cross-layer auxiliary residual module are multiplied by the preset weight coefficient and input into the corresponding residual convolution module. For example, the weight coefficient is set to 0.3. The specific training method of the brain age prediction network can refer to the relevant description of the second embodiment.

[0061] The first basic convolution module and the second basic convolution module usually have the same structure, such as Figure 1CAs shown, preferably, the first basic convolution module 10 includes three basic convolution blocks 101 connected in sequence and a two-dimensional maximum pooling layer 102, and each convolution block includes a two-dimensional convolution layer 1011, a normalization layer 1012 and a Relu activation layer 1013 connected in sequence to realize the extraction of basic features. Among them, the purpose of adding the normalization layer is to increase the learning speed of the model, simplify the process of parameter adjustment, and make the learning process of the brain age prediction network more stable. The normalization layer can be a BN (Batch Normalization) layer. The purpose of adding the Relu activation function is to increase nonlinearity and alleviate the gradient disappearance problem in the training process of the brain age prediction network. After the three basic convolution blocks, there is a two-dimensional maximum pooling layer. After pooling, the size of the feature map is halved, which is similar to a downsampling operation. In a specific implementation, the convolution kernel size of the two-dimensional convolution layer can be 3×3, the convolution kernel size of the two-dimensional pooling layer can be 2×2, and the step size can be set to 2.

[0062] like Figure 1D As shown, the first residual convolution module, the second residual convolution module and the third residual convolution module are generally of the same structure. Preferably, the first residual convolution module 12 includes three residual blocks 121 connected in sequence and a two-dimensional maximum pooling layer 122. The output value of the residual block is the sum of the input value of the top layer and the output value after the input value passes through two basic convolution blocks. The basic convolution block includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence to achieve the extraction of deep features. In a specific implementation, the parameters of the basic convolution block in the residual block are generally the same as the parameters of the basic convolution block in the basic convolution module.

[0063] The first cross-layer auxiliary residual module and the second cross-layer auxiliary residual module generally have the same structure. Preferably, each cross-layer auxiliary residual module includes a two-dimensional maximum pooling layer and an attention layer connected in sequence, so that each cross-layer auxiliary residual module can be specifically used to learn more efficient features through the attention mechanism and transfer them across layers, so that the brain age prediction network can autonomously learn the importance of the upper-layer features. Further preferably, the attention layer uses SeNet (Squeeze-and-Excitation Networks) to improve the performance of the network at the feature channel level through SeNet, allowing the network to learn the importance of each feature channel by itself.

[0064] After predicting the brain age of the subject, the brain health status of the subject can be evaluated based on the brain age prediction result. Considering that there is a certain error between the brain age prediction value of the brain age prediction network and the true value of the subject, the mean square error (MSE) of the brain age prediction network on the healthy group test data set can be evaluated, and an error range can be set according to the mean square error, for example, the error range is within 3 years. When evaluating the brain health status of the subject, the predicted brain age value is subtracted from the true age to determine whether the difference between the predicted brain age value and the true age is within the error range. If so, it indicates that the brain of the subject is in a healthy state. If not, it indicates that the brain of the subject is in an unhealthy state and there is a risk of illness.

[0065] Furthermore, in order to evaluate the unhealthy state, it is necessary to evaluate the mean square error of the trained brain age prediction network on the data sets of people with Alzheimer's disease (AD) and people with mild cognitive impairment (MCI), and determine the above error range based on the mean square error of the brain age prediction network on the diseased group data sets.

[0066] In an embodiment of the present invention, a brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module which are connected in sequence. The brain age prediction network also includes a first and a second cross-layer auxiliary residual module. The input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module. The input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module. When predicting the brain age of a subject, the three-dimensional brain magnetic resonance imaging data of the subject is preprocessed to obtain preprocessed image data, and the preprocessed image data is input into a pre-trained brain age prediction network to predict the brain age of the subject, thereby realizing brain age prediction through the brain age prediction network.

[0067] Embodiment 2:

[0068] Figure 2 The implementation process of the training method of the brain age prediction network provided by the second embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0069] In step S201, a basic data set is obtained.

[0070] In an embodiment of the present invention, a basic data set can be obtained based on an existing public data set. Since the amount of data in the normal control group in ADNI1 (Alzheimer's Disease Neuroimaging Initiative) is very small, and the age distribution of the subjects in the data set is mainly over 50 years old, and deep learning training requires a large number of data sets, and the more evenly the label distribution is distributed, the better, therefore, the data set of ICBM (International Consortium for Brain Mapping) can be selected as an expansion. For example, 1201 data are used in ICBM, and 406 data of the normal control group (healthy subjects) are used in ADNI-1, totaling 1607 data as the basic data set, and the modality is selected as structural image. Among them, ADNI is currently the authoritative data center for studying Alzheimer's disease. It was jointly funded and established by the National Institute of Health and the National Institute on Aging in 2004. ICBM is a public data set provided by the International Brain Imaging Alliance.

[0071] In step S202, each three-dimensional brain magnetic resonance imaging data in the basic data set is preprocessed to obtain a training data set.

[0072] In the embodiment of the present invention, considering that the magnetic resonance equipment used for ICBM and ADNI1 data acquisition is not completely the same, the data size in ADNI1 is 166*256*256, while the data size in ICBM is 124*256*256, that is, there is a difference in the original data size between the two. Therefore, it is necessary to unify the data size in the two sets of data sets, so that preferably, each three-dimensional brain magnetic resonance imaging data in the training data set is registered to obtain the registered image data, so as to eliminate the differences in size and structure between the data set and the individual, unify the coordinates, and facilitate the training of the brain age prediction network. Specifically, the image data can be registered to a preset standard template space, for example, MNI152 space. Taking MNI152 space as an example, the registered data size is 91×109×91.

[0073] In step S203, based on the training data set, the brain age prediction network is trained using a stochastic gradient descent algorithm to obtain a trained brain age prediction network.

[0074] In the embodiment of the present invention, the stochastic gradient descent algorithm SGD is used to train the brain age prediction network, and the error is calculated according to the label value of the brain age and the output value of the model. The parameters of the network training are adjusted and optimized according to the error. Among them, the loss function uses the mean square loss function:

[0075] loss = (x i -y i )2

[0076] Among them, loss represents the loss value, x i represents the true value corresponding to sample i, y i Represents the feature expression corresponding to sample i.

[0077] In an embodiment of the present invention, a basic data set is obtained, and each three-dimensional brain magnetic resonance imaging data in the basic data set is preprocessed to obtain a training data set. Based on the training data set, a stochastic gradient descent algorithm is used to train a brain age prediction network to obtain a trained brain age prediction network, thereby realizing the training of the brain age prediction network.

[0078] Embodiment three:

[0079] Figure 3 The structure of the brain age prediction device provided by the third embodiment of the present invention is shown. For the convenience of description, only the part related to the embodiment of the present invention is shown, including:

[0080] A preprocessing unit 31 is used to preprocess the three-dimensional brain magnetic resonance imaging data of the subject to be tested to obtain preprocessed imaging data; and

[0081] The brain age prediction unit 32 is used to input the preprocessed image data into a pre-trained brain age prediction network, and output the brain age prediction result of the subject to be tested, wherein the brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module connected in sequence, and the brain age prediction network also includes a first cross-layer auxiliary residual module and a second cross-layer auxiliary residual module, the input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module, the input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module.

[0082] Preferably, the preprocessing unit is used to register the three-dimensional brain magnetic resonance imaging data of the subject to obtain registered image data.

[0083] Preferably, the device further comprises a data set acquisition unit and a model training unit, wherein:

[0084] A data set acquisition unit, used to acquire a basic data set;

[0085] The preprocessing unit is also used to preprocess each three-dimensional brain magnetic resonance imaging data in the basic data set to obtain a training data set; and

[0086] The model training unit is used to train the brain age prediction network based on the training data set using a stochastic gradient descent algorithm to obtain a trained brain age prediction network.

[0087] Preferably, the brain age prediction unit further includes:

[0088] A basic feature extraction unit, used to extract basic features of the preprocessed image data through the first and second basic convolution modules;

[0089] A deep feature extraction unit, used for extracting deep features based on basic features through first, second, and third residual convolution modules and first and second cross-layer auxiliary residual modules;

[0090] A pooling unit for unifying the data size of deep features through a two-dimensional maximum pooling module; and

[0091] The prediction unit is used to output the brain age prediction results through the fully connected module and the sigmoid function module based on the deep features after the data size is unified.

[0092] Preferably, the first basic convolution module and the second basic convolution module have the same structure, the first basic convolution module includes three basic convolution blocks connected in sequence and a two-dimensional maximum pooling layer, and each convolution block includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence.

[0093] Preferably, the first residual convolution module, the second residual convolution module and the third residual convolution module have the same structure. The first residual convolution module includes three residual blocks connected in sequence and a two-dimensional maximum pooling layer. The output value of the residual block is the sum of the input value of the top layer and the output value after the input value passes through two basic convolution blocks. The basic convolution block includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence.

[0094] Preferably, the first cross-layer auxiliary residual module and the second cross-layer auxiliary residual module have the same structure, and the first cross-layer auxiliary residual module includes a two-dimensional maximum pooling layer and an attention layer connected in sequence.

[0095] Preferably, the attention layer adopts SeNet.

[0096] In the embodiment of the present invention, each unit of the brain age prediction device can be implemented by a corresponding hardware or software unit, and each unit can be an independent software or hardware unit, or can be integrated into a software or hardware unit, which is not intended to limit the present invention. The specific implementation of each unit of the brain age prediction device can refer to the description of the aforementioned method embodiment, which will not be repeated here.

[0097] Embodiment 4:

[0098] Figure 4The structure of an electronic device provided by a fourth embodiment of the present invention is shown. For the convenience of description, only the parts related to the embodiment of the present invention are shown.

[0099] The electronic device 4 of the embodiment of the present invention includes a processor 40, a memory 41, and a computer program 42 stored in the memory 41 and executable on the processor 40. When the processor 40 executes the computer program 42, the steps in the above-mentioned method embodiments are implemented, for example: Figure 1A Alternatively, when the processor 40 executes the computer program 42, the functions of each unit in the above-mentioned device embodiments are realized, for example Figure 2 The functions of the units 21 to 22 are shown.

[0100] In an embodiment of the present invention, a brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module which are connected in sequence. The brain age prediction network also includes a first and a second cross-layer auxiliary residual module. The input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module. The input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module. When predicting the brain age of a subject, the three-dimensional brain magnetic resonance imaging data of the subject is preprocessed to obtain preprocessed image data, and the preprocessed image data is input into a pre-trained brain age prediction network to predict the brain age of the subject, thereby realizing brain age prediction through the brain age prediction network.

[0101] Embodiment five:

[0102] In an embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above method embodiment are implemented, for example, Figure 1A Alternatively, when the computer program is executed by a processor, the functions of each unit in the above-mentioned device embodiments are realized, for example Figure 3 The functions of the units 31 to 32 are shown.

[0103] In an embodiment of the present invention, a brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module which are connected in sequence. The brain age prediction network also includes a first and a second cross-layer auxiliary residual module. The input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module. The input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module. When predicting the brain age of a subject, the three-dimensional brain magnetic resonance imaging data of the subject is preprocessed to obtain preprocessed image data, and the preprocessed image data is input into a pre-trained brain age prediction network to predict the brain age of the subject, thereby realizing brain age prediction through the brain age prediction network.

[0104] The computer-readable storage medium of the embodiment of the present invention may include any entity or device or recording medium capable of carrying computer program code, for example, ROM / RAM, magnetic disk, optical disk, flash memory and other memories.

[0105] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting brain age, characterized in that: The method comprises: Preprocessing the three-dimensional brain magnetic resonance imaging data of the subject to be tested to obtain preprocessed imaging data; The preprocessed image data is input into a pre-trained brain age prediction network to predict the brain age of the subject, wherein the brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module connected in sequence, and the brain age prediction network also includes a first and a second cross-layer auxiliary residual module with the same structure, the input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module, the input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module, and the first and second cross-layer auxiliary residual modules both include a two-dimensional maximum pooling layer and an attention layer connected in sequence, so as to learn more efficient features through the attention mechanism and transfer them across layers.

2. A brain age prediction device, characterized in that: The device comprises: A preprocessing unit, used for preprocessing the three-dimensional brain magnetic resonance imaging data of the subject to be tested to obtain preprocessed imaging data; and A brain age prediction unit is used to input the preprocessed image data into a pre-trained brain age prediction network, and output the brain age prediction result of the subject, wherein the brain age prediction network includes a first basic convolution module, a second basic convolution module, a first residual convolution module, a second residual convolution module, a third residual convolution module, a two-dimensional maximum pooling module, a fully connected module and a sigmoid function module connected in sequence, and the brain age prediction network also includes a first cross-layer auxiliary residual module and a second cross-layer auxiliary residual module with the same structure, the input end of the first cross-layer auxiliary residual module is connected to the output end of the second basic convolution module, and the output end is connected to the input end of the second residual convolution module, the input end of the second cross-layer auxiliary residual module is connected to the output end of the first residual convolution module, and the output end is connected to the input end of the third residual convolution module, and the first and second cross-layer auxiliary residual modules both include a two-dimensional maximum pooling layer and an attention layer connected in sequence, so as to learn more efficient features through the attention mechanism and transfer them across layers.

3. The device according to claim 2, characterized in that The preprocessing unit is used to register the three-dimensional brain magnetic resonance imaging data of the subject to be tested to obtain registered image data.

4. The device according to claim 2, characterized in that The device also includes a data set acquisition unit and a model training unit, wherein: The data set acquisition unit is used to acquire a basic data set; The preprocessing unit is further used to preprocess each three-dimensional brain magnetic resonance imaging data in the basic data set to obtain a training data set; and The model training unit is used to train the brain age prediction network based on the training data set using a stochastic gradient descent algorithm to obtain the trained brain age prediction network.

5. The device according to claim 2, characterized in that The brain age prediction unit also includes: A basic feature extraction unit, used for extracting basic features of the preprocessed image data through the first and second basic convolution modules; A deep feature extraction unit, configured to extract deep features based on the basic features through the first, second, and third residual convolution modules and the first and second cross-layer auxiliary residual modules; A pooling unit, configured to unify the data size of the deep features through the two-dimensional maximum pooling module; and The prediction unit is used to output the brain age prediction result through the fully connected module and the sigmoid function module based on the deep features after the data size is unified.

6. The device according to claim 2, characterized in that The first basic convolution module and the second basic convolution module have the same structure. The first basic convolution module includes three basic convolution blocks connected in sequence and a two-dimensional maximum pooling layer. Each of the convolution blocks includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence.

7. The device according to claim 2, characterized in that The first residual convolution module, the second residual convolution module and the third residual convolution module have the same structure. The first residual convolution module includes three residual blocks connected in sequence and a two-dimensional maximum pooling layer. The output value of the residual block is the sum of the input value of the top layer and the output value of the input value after passing through two basic convolution blocks. The basic convolution block includes a two-dimensional convolution layer, a normalization layer and a Relu activation layer connected in sequence.

8. The device according to claim 2, characterized in that The attention layer in the first and second cross-layer auxiliary residual modules adopts SeNet.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the function of the device according to any one of claims 2 to 8 is realized.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the function of the device according to any one of claims 2 to 8 is realized.

Images