An image classification method based on magnetic resonance imaging data

Abstract

The present invention discloses an image classification method based on magnetic resonance imaging data, the method comprising: acquiring three-dimensional data of brain neural images and preprocessing the three-dimensional data to obtain magnetic resonance imaging data; inputting the magnetic resonance imaging data into a trained image classification model, and outputting image categories through the image classification model; wherein the image classification model includes extracted spatial dimension features and channel dimension features of the brain neural images. The embodiments of the present invention can obtain high-level features of the magnetic resonance imaging data through deep learning of three-dimensional magnetic resonance imaging data. The features of the magnetic resonance imaging data extracted by the image classification model are extracted by introducing an attention mechanism and self-supervised learning to obtain high-level features of spatial relationships with greater discriminability, thereby obtaining more accurate image categories.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an image classification method based on magnetic resonance imaging data. Background Art

[0002] With the development of imaging technology, deep learning has been widely used for neuroimaging analysis. Currently, the most widely used brain imaging data is magnetic resonance imaging (MRI), which can reveal abnormalities in the human brain structure. For example, Parkinson's patients show significant atrophy of key structures such as the substantia nigra, putamen, and striatum on MRI, which provides a basis for image classification of diseases such as Parkinson's. Existing Parkinson's disease image classification based on MRI data mostly relies on machine learning, focusing on feature selection. Through feature selection, discriminative features are selected to reduce feature dimensionality. However, traditional machine learning image classification methods ignore the spatial relationships of MRI data and the high-level features between slices in three-dimensional data, resulting in inaccurate image classification results.

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

[0004] The technical problem to be solved by the present invention is to provide an image classification method based on magnetic resonance imaging data in response to the above-mentioned defects of the prior art, aiming to solve the problem that the image classification method in the prior art ignores the spatial relationship of MRI data and the high-level features between slices in three-dimensional data, resulting in inaccurate image classification results.

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

[0006] In a first aspect, an embodiment of the present invention provides an image classification method based on magnetic resonance imaging data, wherein the method comprises:

[0007] Acquiring three-dimensional data of brain nerve images and preprocessing the three-dimensional data to obtain magnetic resonance imaging data;

[0008] The magnetic resonance imaging data is input into a trained image classification model, and the image category is output by the image classification model; wherein the image classification model includes the extracted spatial dimension features and channel dimension features of the brain neural image.

[0009] In one implementation, the image classification model includes several stage modules, three-dimensional deconvolution blocks, average pooling layers and fully connected layers, and each of the stage modules includes two three-dimensional residual modules and one three-dimensional attention shuffling module.

[0010] In one implementation, the three-dimensional attention shuffling module includes a global average pooling layer, a gating mechanism, an activation function, a group normalization layer, and a splicing layer.

[0011] In one implementation, inputting the magnetic resonance imaging data into a trained image classification model, and outputting an image category through the image classification model includes:

[0012] Inputting the magnetic resonance imaging data into a plurality of stage modules to obtain a first feature;

[0013] Input the first feature into the average pooling layer to obtain the second feature;

[0014] The second feature is input into a fully connected layer to obtain an image category.

[0015] In one implementation, inputting the magnetic resonance imaging data into a trained image classification model, and outputting the image category through the image classification model further comprises:

[0016] The second feature is input into the three-dimensional deconvolution block to obtain predicted voxel information.

[0017] In one implementation, inputting the magnetic resonance imaging data into a plurality of stage modules to obtain the first feature includes:

[0018] inputting the magnetic resonance imaging data into a plurality of stage modules in sequence;

[0019] For each stage module, the two 3D residual modules in each stage module output initial features, and the initial features are input into the 3D attention shuffling module to obtain the stage module output features;

[0020] The stage module output feature output by the last stage module is used as the first feature.

[0021] In one implementation, inputting the initial features into the three-dimensional attention shuffling module to obtain the stage module output features includes:

[0022] Dividing the initial features into several groups to obtain several sub-features;

[0023] For each sub-feature, each sub-feature is divided into two parts to obtain a first sub-feature and a second sub-feature, the first sub-feature is sequentially input to a global average pooling layer, a gating mechanism, and an activation function to obtain a third sub-feature, the first sub-feature is sequentially input to a group normalization layer, a gating mechanism, and an activation function to obtain a fourth sub-feature, the third sub-feature and the fourth sub-feature are input to a splicing layer to obtain a merged sub-feature corresponding to each sub-feature; wherein the third sub-feature includes a channel feature; and the fourth sub-feature includes a spatial feature;

[0024] Merge all merged sub-features corresponding to all sub-features to obtain the stage module output feature.

[0025] In one implementation, the training method of the image classification model includes:

[0026] Get training data and labels;

[0027] Inputting the training data into an initial image classification model to obtain model output data;

[0028] Obtaining a loss function based on the label and the model output data; wherein the loss function includes a self-supervised learning loss function and a classification loss function;

[0029] Based on the loss function, the initial image classification model is trained to obtain a trained image classification model.

[0030] In a second aspect, an embodiment of the present invention further provides an image classification device based on magnetic resonance imaging data, wherein the device comprises:

[0031] A preprocessing module, configured to obtain three-dimensional data of brain nerve images and preprocess the three-dimensional data to obtain magnetic resonance imaging data;

[0032] A classification module is used to input the magnetic resonance imaging data into a trained image classification model and output image categories through the image classification model; wherein the image classification model includes the extracted spatial dimension features and channel dimension features of the brain neural image.

[0033] In a third aspect, an embodiment of the present invention further provides an intelligent terminal comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, wherein the one or more programs include a method for executing an image classification method based on magnetic resonance imaging data as described in any one of the above.

[0034] In a fourth aspect, an embodiment of the present invention further provides a non-temporary computer-readable storage medium, which, when the instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute an image classification method based on magnetic resonance imaging data as described in any one of the above.

[0035] Beneficial effects of the present invention: The embodiment of the present invention first obtains three-dimensional data of brain neural images, and preprocesses the three-dimensional data to obtain magnetic resonance imaging data; then the magnetic resonance imaging data is input into a trained image classification model, and the image category is output through the image classification model; wherein, the image classification model includes the extracted spatial dimension features and channel dimension features of the brain neural images; it can be seen that the embodiment of the present invention can obtain high-level features of magnetic resonance imaging data through deep learning of three-dimensional magnetic resonance imaging data, and extracts features of magnetic resonance imaging data through the image classification model by introducing attention mechanism and self-supervised learning, so as to obtain more discriminative high-level features of spatial relationships, so that more accurate image categories can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 A schematic flow chart of an image classification method based on magnetic resonance imaging data provided by an embodiment of the present invention.

[0038] Figure 2 This is a structural diagram of an image classification model for an implementation method provided by an embodiment of the present invention.

[0039] Figure 3 A diagram of the shuffling attention structure of an implementation method provided in an embodiment of the present invention.

[0040] Figure 4 This is a principle block diagram of an image classification device based on magnetic resonance imaging data provided by an embodiment of the present invention.

[0041] Figure 5 This is a block diagram of the internal structure of the smart terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention discloses an image classification method based on magnetic resonance imaging data. To make the objectives, technical solutions, and effects of the present invention more clear and explicit, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are intended only to illustrate the present invention and are not intended to limit the present invention.

[0043] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the description of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when the present invention refers to an element being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any units and all combinations of one or more associated listed items.

[0044] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0045] In the existing technology, traditional machine learning image classification methods ignore the spatial relationship of MRI data and the high-level features between slices in three-dimensional data, resulting in inaccurate image classification results.

[0046] To address the problems of the prior art, this embodiment provides an image classification method based on magnetic resonance imaging data. By deep learning three-dimensional magnetic resonance imaging data, high-level features of the magnetic resonance imaging data can be obtained. The features of the magnetic resonance imaging data are extracted through an image classification model. By introducing an attention mechanism and self-supervised learning, high-level features of spatial relationships with greater discriminative power are obtained, thereby enabling more accurate image classification. In a specific implementation, three-dimensional brain neuroimaging data is first acquired and preprocessed to obtain magnetic resonance imaging data. The magnetic resonance imaging data is then input into a trained image classification model, which outputs image categories. The image classification model includes the extracted spatial and channel dimensional features of the brain neuroimaging.

[0047] Exemplary Methods

[0048] This embodiment provides an image classification method based on magnetic resonance imaging data, which can be applied to intelligent terminals for image processing. Figure 1 As shown, the method includes:

[0049] Step S100: Acquire three-dimensional data of brain nerve images, and pre-process the three-dimensional data to obtain magnetic resonance imaging data;

[0050] Specifically, the present invention uses the statistical parametric mapping (SPM) toolbox to preprocess brain neuroimages through three steps: bias field correction, skull stripping, and image registration. After preprocessing, magnetic resonance imaging (MRI) data is obtained. Magnetic resonance imaging (MRI) data is three-dimensional, and can more fully learn the spatial relationship within and between slices, and mine more discriminative high-level information, thereby achieving more accurate neurodegenerative image classification, and preparing for the subsequent acquisition of high-level features of magnetic resonance imaging data.

[0051] After obtaining the three-dimensional magnetic resonance imaging data, you can perform the following Figure 1 The following steps are shown: S200, inputting the magnetic resonance imaging data into a trained image classification model, and outputting the image category through the image classification model; wherein the image classification model includes the extracted spatial dimension features and channel dimension features of the brain neural image.

[0052] Specifically, the image classification model is obtained through training in advance and can extract the spatial and channel features of brain neural images and the high-level features of three-dimensional magnetic resonance imaging data, and finally output the image category from the image classification model.

[0053] In one implementation, Figure 2 As shown, the image classification model includes several stage modules, three-dimensional deconvolution blocks, average pooling layers and fully connected layers. Each of the stage modules includes two three-dimensional residual modules and a three-dimensional attention shuffle module. The three-dimensional attention shuffle module includes a global average pooling layer, a gating mechanism, an activation function, a group normalization layer, and a splicing layer.

[0054] In this embodiment, the image classification model includes four stage modules, a three-dimensional deconvolution block (3D deconvolution block), an average pooling layer and a fully connected layer (FC), each of which includes two three-dimensional residual modules B1 and B2 (forming Stage (F) = B1 (F) + B2 (F)) and a three-dimensional attention shuffle module (3D shuffleattention). The three-dimensional attention shuffle module includes a global average pooling layer (GAP), a simple gating mechanism (Fc (.)), an activation function (σ (·)), a group normalization layer (GN), and a concatenation layer (CONCAT). Among them, Fc (.) is a linear function w (x) plus a bias, and σ (·) is an activation function.

[0055] Step S200 includes the following steps:

[0056] S201, inputting the magnetic resonance imaging data into several stage modules to obtain a first feature;

[0057] S202: Input the first feature into an average pooling layer to obtain a second feature;

[0058] S203: Input the second feature into a fully connected layer to obtain an image category.

[0059] In this embodiment, after the magnetic resonance imaging data is input into the four-stage module, the first feature is output, and then the first feature is input into the average pooling layer to obtain a second feature containing two dimensions of space and channel of the brain neural image. Finally, the second feature is input into the fully connected layer (FC) to obtain the image category.

[0060] In one implementation, inputting the magnetic resonance imaging data into a trained image classification model, and outputting the image category through the image classification model further includes the following steps: inputting the second feature into a three-dimensional deconvolution block to obtain predicted voxel information.

[0061] Specifically, the present application sets up a forward task for predicting voxel information. By executing the forward task, some potential high-level feature representations can be better learned from the MRI data, so as to facilitate adjustments in the classification task and improve the classification performance. At the same time, the second feature is input into the three-dimensional deconvolution block to obtain the predicted voxel information. The loss function can be obtained through the predicted voxel information, and finally the model parameters of the image classification model are adjusted through the loss function.

[0062] Step S201 includes the following steps:

[0063] S2011, inputting the magnetic resonance imaging data into several stage modules in sequence;

[0064] S2012: For each stage module, output initial features through the two three-dimensional residual modules in each stage module, and input the initial features into the three-dimensional attention shuffling module to obtain stage module output features;

[0065] S2013. Use the stage module output feature output by the last stage module as the first feature.

[0066] Specifically, inputting the magnetic resonance imaging data into the first stage module will output a feature, which is then input into the next stage module, and the output feature of the last stage module is used as the first feature. For each stage module, the initial features are output by the two three-dimensional residual modules in each stage module, and then the initial features are input into the three-dimensional attention shuffle module to obtain the stage module output features. A three-dimensional attention shuffle module (3D shuffle attention) is added after every two three-dimensional residual modules to eliminate the influence of redundant features from the output features and extract more important feature information. The present invention uses a "channel shuffle" operation to transform different information groups along the channel dimension, which effectively combines the spatial and channel attention mechanisms while reducing the calculation time of the module.

[0067] In step S2012, the initial features are input into the three-dimensional attention shuffling module to obtain the stage module output features, including the following steps: dividing the initial features into several groups to obtain several sub-features; for each sub-feature, dividing each sub-feature into two parts to obtain a first sub-feature and a second sub-feature, inputting the first sub-feature into the global average pooling layer, the gating mechanism, and the activation function in sequence to obtain a third sub-feature, inputting the first sub-feature into the group normalization layer, the gating mechanism, and the activation function in sequence to obtain a fourth sub-feature, inputting the third sub-feature and the fourth sub-feature into the splicing layer to obtain a merged sub-feature corresponding to each sub-feature; wherein the third sub-feature includes a channel feature; the fourth sub-feature includes a spatial feature; merging all merged sub-features corresponding to all sub-features to obtain the stage module output features.

[0068] Specifically, if Figure 3 The detailed structure of the 3D shuffle attention module is shown, which converts the initial features They are divided into eight groups along the channel dimension, namely Sub-feature F k Divide into two parts again along the channel dimension: the first sub-feature and the second sub-feature And put them into channel attention and spatial attention channels respectively. The channel attention module first takes the first sub-feature F k1 Generate global information of each channel through global feature pooling, and then use F c (·) and activation function (σ(·)) to guide the selection of key features. Channel attention can be expressed as:

[0069] M c (F k1 )=σ(F c (GAP(F k1 )))·F k1=σ(w c GN(F k1 )+b)·F k1 ,

[0070] The spatial attention module first takes the second sub-feature F k2 The spatial information is obtained by group normalization, and then F c (·) and activation function (σ(·)) to guide the selection of key features. Spatial attention can be expressed as:

[0071] M s (F k2 )=σ(F c (GN(F k2 )))·F k2 =σ(w c GN(F k2 )+b)·F k2 ,

[0072] The feature information obtained by the spatial attention and channel attention modules is fused by splicing to obtain the internal feature information of a group (that is, the merged sub-features). Finally, the eight groups of merged sub-features are resized and arranged through “channel shuffling” and then merged, and the stage module outputs features to fully utilize the information learned within each group of features.

[0073] In one implementation, the training method of the image classification model includes the following steps: obtaining training data and labels; inputting the training data into the initial image classification model to obtain model output data; obtaining a loss function based on the labels and the model output data; wherein the loss function includes a self-supervised learning loss function and a classification loss function; based on the loss function, training the initial image classification model to obtain a trained image classification model.

[0074] Specifically, training data and labels corresponding to the training data are first obtained, and then the training data is input into an initial image classification model to obtain model output data; a loss function is obtained based on the labels and the model output data; wherein the loss function includes a self-supervised learning loss function and a classification loss function; Figure 2 (b) The self-supervised learning module can better learn some potential high-level feature information from MRI data by predicting voxel information, thereby improving the classification performance. It is calculated by five layers of deconvolution blocks, and the kernel size of the deconvolution block is the same as the kernel size of the forward convolution block. Given the input data Loss Function for Self-Supervised Learning It can be expressed as:

[0075]

[0076] Among them, m represents the number of samples, and i represents the i-th sample.

[0077] Figure 2 The classification module in (c) inputs the first feature and obtains the model output data. The category label of the training data is y' i (i=1,2,3,…,m). The present invention adopts a simple fully connected layer as the classification module, given the true label y i (i=1,2,3,…,m), classification loss It can be expressed as:

[0078]

[0079] Among them, m represents the number of samples, and i represents the i-th sample.

[0080] Therefore, the loss function of the entire model of the present invention is As shown below:

[0081]

[0082] Based on the loss function The initial image classification model is trained to obtain a trained image classification model.

[0083] In summary, the method of the present invention uses a feature extraction network to learn potential, inherent feature information from MRI data. This feature information is then input into a classification module to train a multi-classification model, thereby achieving early multi-stage image classification results for neurodegenerative diseases.

[0084] Key innovation points:

[0085] (1) Based on the three-dimensional data of brain neuroimaging, the rich brain structure information contained in the original data is utilized through the three-dimensional deep learning method to explore the relationship between brain neuroimaging slices and the intrinsic relationship within the slice space, and select the features that are most relevant to the disease image and have discriminative ability.

[0086] (2) Based on the idea of ​​the attention mechanism, by assigning feature weight coefficients from the channel and spatial dimensions to emphasize important information and suppress irrelevant redundant information, the bottom-up feedforward process is guided, so that the classification can focus on the lesion area from the spatial and channel dimensions and reduce the influence of irrelevant features.

[0087] (3) Predicting voxel information from learned features can fully understand the characteristics of three-dimensional spatial information, which is an important and transferable knowledge. Therefore, a self-supervised learning forward task is designed to predict voxel information to better learn some potential high-level feature representations from MRI data, so as to facilitate adjustment in the classification task and improve classification performance.

[0088] Exemplary devices

[0089] like Figure 4 As shown in , an embodiment of the present invention provides an image classification device based on magnetic resonance imaging data, the device comprising a preprocessing module 301 and a classification module 302, wherein:

[0090] A preprocessing module 301 is used to obtain three-dimensional data of brain nerve images and preprocess the three-dimensional data to obtain magnetic resonance imaging data;

[0091] The classification module 302 is configured to input the magnetic resonance imaging data into a trained image classification model, and output an image category through the image classification model.

[0092] Based on the above embodiment, the present invention also provides an intelligent terminal, whose principle block diagram can be shown as follows: Figure 5 As shown. The intelligent terminal includes a processor, a memory, a network interface, a display screen, and a temperature sensor connected via a system bus. The processor of the intelligent terminal is used to provide computing and control capabilities. The memory of the intelligent terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the intelligent terminal is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for image classification of diseases based on magnetic resonance imaging data is implemented. The display screen of the intelligent terminal can be a liquid crystal display screen or an electronic ink display screen, and the temperature sensor of the intelligent terminal is pre-set inside the intelligent terminal to detect the operating temperature of the internal device.

[0093] Those skilled in the art will understand that Figure 5 The schematic diagram is only a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the smart terminal to which the solution of the present invention is applied. The specific smart terminal may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0094] In one embodiment, a smart terminal is provided, comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations:

[0095] Acquiring three-dimensional data of brain nerve images and preprocessing the three-dimensional data to obtain magnetic resonance imaging data;

[0096] The magnetic resonance imaging data is input into a trained image classification model, and the image category is output by the image classification model; wherein the image classification model includes the extracted spatial dimension features and channel dimension features of the brain neural image.

[0097] Those skilled in the art will appreciate that all or part of the processes in the above-described embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described embodiments. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various 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 RAM bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0098] In summary, the present invention discloses an image classification method based on magnetic resonance imaging data, the method comprising: acquiring three-dimensional data of brain neural images, and preprocessing the three-dimensional data to obtain magnetic resonance imaging data; inputting the magnetic resonance imaging data into a trained image classification model, and outputting image categories through the image classification model; wherein the image classification model includes extracted spatial dimension features and channel dimension features of brain neural images. The embodiment of the present invention can obtain high-level features of magnetic resonance imaging data through deep learning of three-dimensional magnetic resonance imaging data, and extracting features of magnetic resonance imaging data through the image classification model by introducing an attention mechanism and self-supervised learning to obtain high-level features of spatial relationships with higher discriminability, thereby obtaining more accurate image categories.

[0099] Based on the above embodiments, the present invention discloses an image classification method based on magnetic resonance imaging data. It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, they can make improvements or changes based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. An image classification method based on magnetic resonance imaging data, characterized in that: The method comprises: Acquiring three-dimensional data of brain nerve images and preprocessing the three-dimensional data to obtain magnetic resonance imaging data; Inputting the magnetic resonance imaging data into a trained image classification model, and outputting an image category through the image classification model; wherein the image classification model includes the extracted spatial dimension features and channel dimension features of the brain neural image; The image classification model includes several stage modules, three-dimensional deconvolution blocks, average pooling layers and fully connected layers, and each stage module includes two three-dimensional residual modules and a three-dimensional attention shuffling module; Inputting the magnetic resonance imaging data into a trained image classification model, and outputting image categories through the image classification model comprises: inputting the magnetic resonance imaging data into a plurality of stage modules in sequence; For each stage module, the two 3D residual modules in each stage module output initial features, and the initial features are input into the 3D attention shuffling module to obtain the stage module output features; The output feature of the last stage module is used as the first feature; Input the first feature into the average pooling layer to obtain the second feature; The second feature is input into a fully connected layer to obtain an image category.

2. The image classification method based on magnetic resonance imaging data according to claim 1, characterized in that: The three-dimensional attention shuffling module includes a global average pooling layer, a gating mechanism, an activation function, a group normalization layer, and a splicing layer.

3. The image classification method based on magnetic resonance imaging data according to claim 1, characterized in that: Inputting the magnetic resonance imaging data into a trained image classification model, and outputting the image category through the image classification model further comprises: The second feature is input into the three-dimensional deconvolution block to obtain predicted voxel information.

4. The image classification method based on magnetic resonance imaging data according to claim 1, characterized in that: The initial features are input into the three-dimensional attention shuffling module to obtain the output features of the stage module, including: Dividing the initial features into several groups to obtain several sub-features; For each sub-feature, each sub-feature is divided into two parts to obtain a first sub-feature and a second sub-feature, the first sub-feature is sequentially input to a global average pooling layer, a gating mechanism, and an activation function to obtain a third sub-feature, the first sub-feature is sequentially input to a group normalization layer, a gating mechanism, and an activation function to obtain a fourth sub-feature, the third sub-feature and the fourth sub-feature are input to a splicing layer to obtain a merged sub-feature corresponding to each sub-feature; wherein the third sub-feature includes a channel feature; and the fourth sub-feature includes a spatial feature; Merge all merged sub-features corresponding to all sub-features to obtain the stage module output feature.

5. The image classification method based on magnetic resonance imaging data according to claim 1, characterized in that: The training method of the image classification model includes: Get training data and labels; Inputting the training data into an initial image classification model to obtain model output data; Obtaining a loss function based on the label and the model output data; wherein the loss function includes a self-supervised learning loss function and a classification loss function; Based on the loss function, the initial image classification model is trained to obtain a trained image classification model.

6. An intelligent terminal, characterized in that: The device comprises a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, and the one or more programs include being used to execute the method according to any one of claims 1 to 5.

7. A non-transitory computer-readable storage medium, characterized in that When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method according to any one of claims 1 to 5.

Images