Brain tumor classification method and device, magnetic resonance equipment and readable storage medium

Abstract

The invention provides a brain tumor classification method and device, magnetic resonance equipment and a readable storage medium, and belongs to the field of medical image processing. The method comprises the following steps: inputting an obtained brain tumor image into a brain tumor classification network which is completely trained, carrying out continuous three-time multi-scale dynamic channel sampling feature enhancement on the brain tumor image to obtain channel sampling features, and carrying out visual perception enhancement on the channel sampling features to obtain perception enhancement features; and predicting and outputting the perception enhancement features to obtain a brain tumor classification result. According to the invention, through multi-scale dynamic channel sampling feature enhancement, dynamic channel sampling can be carried out on the tumor image at different scale levels, the tiny pathological features of the tumor image can be effectively extracted and enhanced, and more accurate feature representation is learned from the integrated tiny pathological features through visual perception enhancement. The subtle pathological features of the tumor image are fully utilized, and the accuracy of brain tumor classification is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and particularly to a method and apparatus for classifying brain tumors, a magnetic resonance device, and a readable storage medium. Background Art

[0002] Brain tumors are one of the most life-threatening diseases to humans, occurring in both men and women of different age groups. Early detection and accurate diagnosis of the type of brain tumor are crucial for formulating effective treatment plans.

[0003] Currently, deep learning architectures, especially convolutional neural networks and computer vision, have shown significant performance improvements in the field of medical image processing, which can effectively assist in improving the efficiency of medical work. However, in the task of brain tumor classification, due to the high similarity between different types of brain tumors, differentiating different types of brain tumors often relies on very subtle pathological changes in the brain tumors. Existing neural network image recognition methods may not be able to capture these subtle pathological changes between different tumors when dealing with the brain tumor classification task, resulting in a low accuracy of brain tumor classification.

[0004] Therefore, there is a technical problem in the prior art that it is difficult to capture the subtle pathological changes between different tumors in the brain tumor classification task, resulting in a low accuracy of brain tumor classification. Summary of the Invention

[0005] In view of this, it is necessary to provide a method and apparatus for classifying brain tumors, a magnetic resonance device, and a readable storage medium to solve the technical problem in the prior art that it is difficult to capture the subtle pathological changes between different tumors, resulting in a low accuracy of brain tumor classification.

[0006] To solve the above problems, on the one hand, the present invention provides a method for classifying brain tumors, including: Inputting the obtained brain tumor image into a trained brain tumor classification network, performing continuous three - scale dynamic channel sampling feature enhancement on the brain tumor image to obtain channel sampling features, and performing visual perception enhancement on the channel sampling features to obtain perception enhancement features; Performing prediction output on the perception enhancement features to obtain the brain tumor classification result.

[0007] In a possible implementation manner, the multi - scale dynamic channel sampling feature enhancement includes channel attention convolution embedding and several times of multi - scale dynamic channel sampling, and the visual perception enhancement includes channel attention convolution embedding and visual state space enhancement.

[0008] In a possible implementation manner, the channel attention convolution embedding includes: Divide the input features embedded with channel attention convolution into first image features and second image features; Perform max pooling operation on the first image features to obtain global spatial information, and perform downsampling and 1×1 convolution on the global spatial information to obtain channel attention features; Perform local information feature convolution and downsampling on the second image features to obtain convolution embedded features; Merge the channel attention features and the convolution embedded features and perform channel shuffle enhancement to obtain the output features of the channel attention convolution embedding.

[0009] In a possible implementation, multi-scale dynamic channel sampling includes: Perform continuous four times of dynamic channel sampling on the input features of the multi-scale dynamic channel sampling to obtain multi-scale sampling features output by each dynamic channel sampling; Perform feature fusion on the multi-scale sampling features, perform residual connection feature merging with the input features of the multi-scale dynamic channel sampling, and then perform non-linear transformation to obtain the output features of the multi-scale dynamic channel sampling.

[0010] In a possible implementation, dynamic channel sampling includes: Aggregate the input features of the dynamic channel sampling to obtain a scalar vector; Determine a sampling probability vector according to a preset sampling probability predictor and the scalar vector; Perform feature sampling on the input features of the dynamic channel sampling according to a preset sampling strategy and the sampling probability vector to obtain the output features of the dynamic channel sampling.

[0011] In a possible implementation, performing feature sampling on the input features of the dynamic channel sampling according to a preset sampling strategy and the sampling probability vector to obtain the output features of the dynamic channel sampling includes: Divide the input features of the dynamic channel sampling into several feature subsets, and determine the sampling probability of each feature subset according to the sampling probability vector; If the sampling probability is less than or equal to a preset sampling threshold, perform feature sampling on the highest activation value channels of the feature subset to obtain a sampled output subset; If the sampling probability is greater than the preset sampling threshold, perform average feature sampling on all channels of the feature subset to obtain a sampled output subset; Merge all the sampled output subsets to obtain the output features of the dynamic channel sampling.

[0012] In a possible implementation, visual state space enhancement includes: Perform image segmentation on the input features of the visual state space enhancement to obtain several input feature subgraphs; Perform visual selective cross-scanning on the input feature subgraphs to obtain the output features of the visual state space enhancement.

[0013] On the other hand, the present invention also provides a brain tumor classification device, including: An image feature extraction unit, configured to input the acquired brain tumor image into a trained brain tumor classification network, perform continuous three - scale dynamic channel sampling feature enhancement on the brain tumor image to obtain channel sampling features, and perform visual perception enhancement on the channel sampling features to obtain perception enhancement features; A classification prediction output unit, configured to perform prediction output on the perception enhancement features to obtain a brain tumor classification result.

[0014] On the other hand, the present invention also provides a magnetic resonance device, including a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above - mentioned brain tumor classification method is implemented.

[0015] On the other hand, the present invention also provides a computer - readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above - mentioned brain tumor classification method is implemented.

[0016] The beneficial effects of the present invention are as follows: In the brain tumor classification method provided by the present invention, first, the acquired brain tumor image is input into a trained brain tumor classification network, continuous three - scale dynamic channel sampling feature enhancement is performed on the brain tumor image to obtain channel sampling features, and visual perception enhancement is performed on the channel sampling features to obtain perception enhancement features; then, prediction output is performed on the perception enhancement features to obtain a brain tumor classification result. Through multi - scale dynamic channel sampling feature enhancement, dynamic channel sampling can be performed on tumor images at different scale levels, and fine pathological features of tumor images can be effectively extracted and enhanced. Through visual perception enhancement, more accurate feature representations can be learned from the integrated fine pathological features, making full use of the fine pathological features of tumor images, and effectively improving the accuracy of brain tumor classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other accompanying drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flowchart of an embodiment of the brain tumor classification method provided by the present invention; Figure 2 It is a network structure diagram of the brain tumor classification network in the embodiment of the present invention; Figure 3 It is a schematic flowchart of channel attention convolution embedding in the embodiment of the present invention; Figure 4 Schematic flowchart of multi-scale dynamic channel sampling according to an embodiment of the present invention; Figure 5 Schematic flowchart of dynamic channel sampling according to an embodiment of the present invention; Figure 6 Schematic flowchart of a preset sampling strategy according to an embodiment of the present invention; Figure 7 Schematic flowchart of visual state space enhancement according to an embodiment of the present invention; Figure 8 Graph of model test results in the training phase according to an embodiment of the present invention; Figure 9 Graph of model classification accuracy test results according to an embodiment of the present invention; Figure 10 Visualization display graph of model test results according to an embodiment of the present invention; Figure 11 Schematic structural diagram of an embodiment of a brain tumor classification device provided by the present invention; Figure 12 Schematic structural diagram of an embodiment of a magnetic resonance device provided by the present invention. Detailed implementation manners

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise specified, the meaning of "a plurality of" is two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The descriptions such as "first" and "second" involved in the embodiments of the present invention are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Therefore, the technical features defined with "first" and "second" may explicitly or implicitly include at least one such feature.

[0022] References to "embodiments" in this specification mean that a particular feature, structure, or characteristic described in connection with the embodiments can be included in at least one embodiment of the invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive of other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0023] The present invention provides a method, apparatus, magnetic resonance device, and readable storage medium for classifying brain tumors, which will be described separately below.

[0024] It should be noted that the brain tumor classification method provided in this embodiment can be applied to a brain tumor classification and recognition system for magnetic resonance image recognition or CT image recognition of brain tumors, and can also be applied to the recognition of other types of subtle pathological features in other parts. The brain tumor classification and recognition system can be a software system running on a terminal device, and the terminal device can be a server, a tablet computer, a laptop computer, a personal computer, a personal digital assistant, or a mobile phone, etc. The specific type of the terminal device is not limited in the embodiments of this application.

[0025] Figure 1 It is a schematic flowchart of an embodiment of the brain tumor classification method provided by the present invention. As Figure 1 shown, the brain tumor classification method includes: S101: Input the obtained brain tumor image into a trained brain tumor classification network, perform continuous three - time multi - scale dynamic channel sampling feature enhancement on the brain tumor image to obtain channel sampling features, and perform visual perception enhancement on the channel sampling features to obtain perception enhancement features; S102: Perform prediction output on the perception enhancement features to obtain the brain tumor classification result.

[0026] Among them, the brain tumor image obtained in step S101 can be, but is not limited to, images captured and collected by various types of medical devices such as X - ray, ultrasonic, or nuclear magnetic resonance. The embodiment extracts the subtle pathological features in the brain tumor image through continuous three - time multi - scale dynamic channel sampling feature enhancement and one - time visual perception enhancement, and accurately predicts and classifies the brain tumor type according to the extracted subtle pathological features.

[0027] Compared with the prior art, the brain tumor classification method provided by the embodiments of the present invention first inputs the obtained brain tumor image into a trained brain tumor classification network, performs continuous three - scale dynamic channel sampling feature enhancement on the brain tumor image to obtain channel sampling features, and performs visual perception enhancement on the channel sampling features to obtain perception enhancement features; then performs prediction output on the perception enhancement features to obtain the brain tumor classification result. Through multi - scale dynamic channel sampling feature enhancement, the present invention can perform dynamic channel sampling on tumor images at different scale levels, effectively extract and enhance the subtle pathological features of tumor images, learn more accurate feature representations from the integrated subtle pathological features through visual perception enhancement, make full use of the subtle pathological features of tumor images, and effectively improve the accuracy of brain tumor classification.

[0028] In some embodiments of the present invention, the multi - scale dynamic channel sampling feature enhancement includes channel attention convolution embedding and several times of multi - scale dynamic channel sampling, and the visual perception enhancement includes channel attention convolution embedding and visual state space enhancement.

[0029] Specifically, Figure 2 is the network structure diagram of the brain tumor classification network of the embodiments of the present invention. As Figure 2 shown, the feature processing process of the embodiment is divided into four stages. The first three stages consist of an attention convolution embedding layer and several times of multi - scale dynamic channel sampling, and the fourth stage consists of an attention convolution embedding layer and visual state space enhancement. Among them, the first and second stages perform three times of multi - scale dynamic channel sampling, and the third stage performs nine times of multi - scale dynamic channel sampling.

[0030] In some embodiments of the present invention, Figure 3 is the flow schematic diagram of the channel attention convolution embedding of the embodiments of the present invention. As Figure 3 shown, the channel attention convolution embedding includes: S301: Divide the input feature of the channel attention convolution embedding into a first image feature and a second image feature; S302: Perform a max - pooling operation on the first image feature to obtain global spatial information, perform down - sampling and 1×1 convolution on the global spatial information to obtain channel attention features; S303: Perform local information feature convolution and down - sampling on the second image feature to obtain convolution embedding features; S304: Combine the channel attention features and the convolution embedding features and perform channel shuffle enhancement to obtain the output feature of the channel attention convolution embedding.

[0031] Specifically, in order to extract more effective pathological features of tumor images, the embodiments perform channel attention convolution embedding first in both multi - scale dynamic channel sampling and visual state space enhancement to improve the feature representation ability.

[0032] In the channel attention convolution embedding, in the embodiment, the image or feature map from the previous stage is first divided into two parts. Combining the channel attention module and convolution embedding, local and global features are respectively extracted and downsampled. In the channel attention part, the embodiment uses a max-pooling layer to capture global spatial information and downsample the input, and then increases the number of channels through a 1×1 convolution; in the convolution embedding part, the embodiment uses a convolution with a kernel size of 7, a stride of 4, and a padding of 2 to capture local information and downsample while increasing the number of channels. Finally, the embodiment merges the features obtained from the channel attention part and the convolution embedding part to restore the number of channels, and enhances the representation ability of the channels by using the channel shuffle method for the merged channels.

[0033] In some embodiments of the present invention, Figure 4 is a schematic flowchart of multi-scale dynamic channel sampling according to an embodiment of the present invention, as Figure 4 shown, the multi-scale dynamic channel sampling includes: S401. Continuously perform four dynamic channel samplings on the input features of the multi-scale dynamic channel sampling to obtain multi-scale sampling features of each dynamic channel sampling output; S402. After fusing the multi-scale sampling features, perform residual connection feature merging with the input features of the multi-scale dynamic channel sampling, and then perform a non-linear transformation to obtain the output features of the multi-scale dynamic channel sampling.

[0034] Specifically, in the multi-scale dynamic channel sampling of each layer, the input features are processed through four consecutive dynamic channel samplings. Except for the first dynamic channel sampling, the output of the previous dynamic channel sampling is used as the input of this layer in each dynamic channel sampling. In the first dynamic channel sampling, the number of channels remains the same as the number of channels C of the input features. In the subsequent second to fourth dynamic channel samplings, the number of channels is reduced to C / 2, C / 4, and C / 4 respectively. Finally, the features obtained from each layer are fused to obtain features with a channel number of 2C, and a 1×1 convolution is used to reduce the channel size of the feature map from 2C to C, while weighting and fusing information of different scales. Finally, the obtained features are combined with the features initially input to the multi-scale dynamic channel sampling through a residual connection to alleviate the problem of gradient disappearance, and a 1×1 convolution layer is used to perform a non-linear transformation on the fused features. In addition, batch normalization processing and ReLU activation layers are followed after each 1×1 convolution layer.

[0035] In some embodiments of the present invention, Figure 5 is a schematic flowchart of the dynamic channel sampling according to an embodiment of the present invention, as Figure 5 shown, the dynamic channel sampling includes: S501. Aggregate the input features sampled by the dynamic channel to obtain a scalar vector; S502. Determine the sampling probability vector according to the preset sampling probability predictor and the scalar vector; S503. Perform feature sampling on the input features sampled by the dynamic channel according to the preset sampling strategy and the sampling probability vector to obtain the output features of the dynamic channel sampling.

[0036] In some embodiments of the present invention, Figure 6 is a flowchart of the preset sampling strategy of the embodiments of the present invention. As Figure 6 shown, performing feature sampling on the input features sampled by the dynamic channel according to the preset sampling strategy and the sampling probability vector to obtain the output features of the dynamic channel sampling includes: S601. Divide the input features sampled by the dynamic channel into several feature subsets, and determine the sampling probability of each feature subset according to the sampling probability vector; S602. If the sampling probability is less than or equal to the preset sampling threshold, perform feature sampling on the channel with the highest activation value of the feature subset to obtain the sampled output subset; S603. If the sampling probability is greater than the preset sampling threshold, perform average feature sampling on all channels of the feature subset to obtain the sampled output subset; S604. Combine all the sampled output subsets to obtain the output features of the dynamic channel sampling.

[0037] Specifically, in the dynamic channel sampling, taking the first dynamic channel sampling as an example, the embodiment first aggregates the information of each channel of the input feature into a scalar to form a one-dimensional scalar vector :

[0038] Among them, represents the th channel of and reduces the th channel from to a scalar.

[0039] Next, use a learnable sampling probability predictor to process the vector to generate the sampling probability vector :

[0040]

[0041] Among them, and are the weights and biases, is the Sigmoid activation function.

[0042] Then, the embodiment divides the input features into several feature subsets , where is the grouping parameter. And according to the sampling probability vector , the sampling probability of each subset is determined .

[0043] Then, the embodiment compares the sampling probability of each subset with the preset sampling threshold . If the sampling probability of a certain subset is less than or equal to the preset sampling threshold, then the channel with the highest activation value in the subset is selected as the output channel for sampling; otherwise, all channels in the subset are averaged and fused for sampling. The formula is expressed as:

[0044] Finally, after completing the sampling process of each feature subset, the output feature of dynamic channel sampling is obtained according to the obtained sampling output subsets. For the entire input feature map, the output can be expressed as:

[0045] Through the above sampling strategy, the embodiment enables each pixel position to select the required channel combination according to the characteristics of the input data, and effectively extracts the subtle pathological features at different scales in the image by executing the sampling strategy multiple times at different scales.

[0046] In some embodiments of the present invention, Figure 7 is a schematic flowchart of visual state space enhancement according to an embodiment of the present invention. As Figure 7 shown, visual state space enhancement includes: S701. Image segmentation is performed on the input features of visual state space enhancement to obtain several input feature subgraphs; S702. Visual selective cross-scanning is performed on the input feature subgraphs to obtain the output features of visual state space enhancement.

[0047] Specifically, during the visual state space enhancement process, the embodiment uses VMamba (Visual State Space Model) to enhance visual learning. VMamba first divides the input features into several input feature subgraphs by image segmentation, and then performs visual selective cross-scanning operations on the input feature subgraphs. The visual selective scanning operation can process the input data in a causal manner and capture the information within the scanned part of the data in this way, achieving the acquisition of context information embedded in the input and ensuring the dynamic nature of the internal weights of this mechanism. At the same time, considering the non-causal nature of visual data, directly applying this strategy to patched and flattened images will inevitably lead to a limited receptive field. To solve this problem, a cross-scanning method is proposed. The cross-scanning adopts a four-way scanning strategy, that is, from the four corners of the feature map to the relative positions, ensuring that each element in the feature map integrates information from all other positions in different directions, thereby generating a global receptive field without increasing the linear computational complexity. Finally, through visual state space enhancement, the embodiment can learn more accurate feature representations from the integrated subtle pathological features.

[0048] To verify the effectiveness of the present invention, the embodiment conducts experimental verification on the solution of the present invention.

[0049] First, the embodiment uses a brain tumor data set containing four categories, including meningioma, glioma, pituitary tumor, and tumor-free images. Then, the brain tumor data set is preprocessed. All brain tumor images are cropped to 224×224 pixels, and each image is normalized by dividing all pixel values by 225 to scale them to the range [0, 1]. The brain tumor data set is divided into a training set and a test set in a ratio of 8:2.

[0050] In the experimental stage, the implementation and operation of the model are based on the Linux operating system and the Pytorch 1.13.0 framework. The computer configuration used is: Intel(R) Xeon(R) Gold 6330 CPU @ 2.00GHz, and the GPU model is NVIDIA GeForce RTX 3090 (24 GB). The loading of the data set is implemented by writing code, and the model structure is implemented through code functions.

[0051] During the training process, the cross-entropy loss function is used to calculate the difference between the predicted distribution and the true label distribution. The parameter update adopts the AdamW optimizer, the learning rate is set to 5e-4, the momentum is 0.9, and the weight decay is 0.05. Train for 500 epochs with a batch size of 32.

[0052] Figure 8 This is the model test result graph in the training stage of the embodiment of the present invention. Figure 9This is the graph of the model classification accuracy test results for the embodiments of the present invention. Figure 10 This is the visualization display graph of the model test results for the embodiments of the present invention. As Figures 8 to 10 shown, during the training process, the test set is continuously used to test the model. The loss and classification accuracy of the test are as Figure 8 shown. As the number of training epochs increases continuously, the loss of the model gradually decreases to convergence, and the accuracy gradually increases. After obtaining the model with the best results, the confusion matrix of the test results as Figure 9 shown is output, and the classification accuracy of the model for various types of images is calculated to evaluate the actual generalization performance of the model, and the visualization test results as Figure 10 shown are obtained.

[0053] In summary, in order to capture subtle pathological features and achieve accurate classification of brain tumor images, the present invention first inputs the obtained brain tumor images into a trained brain tumor classification network, performs continuous three - scale dynamic channel sampling feature enhancement on the brain tumor images to obtain channel sampling features, and performs visual perception enhancement on the channel sampling features to obtain perception - enhanced features; then predicts and outputs the perception - enhanced features to obtain the brain tumor classification results. Through multi - scale dynamic channel sampling feature enhancement, the present invention can perform dynamic channel sampling on tumor images at different scale levels, effectively extract and enhance the subtle pathological features of tumor images, learn more accurate feature representations from the integrated subtle pathological features through visual perception enhancement, make full use of the subtle pathological features of tumor images, and effectively improve the accuracy of brain tumor classification.

[0054] To better implement the brain tumor classification method in the embodiments of the present invention, correspondingly, as Figure 11 shown, the present invention also provides a brain tumor classification device. The brain tumor classification device 1100 includes: An image feature extraction unit 1101, configured to input the obtained brain tumor images into a trained brain tumor classification network, perform continuous three - scale dynamic channel sampling feature enhancement on the brain tumor images to obtain channel sampling features, and perform visual perception enhancement on the channel sampling features to obtain perception - enhanced features; A classification prediction output unit 1102, configured to predict and output the perception - enhanced features to obtain the brain tumor classification results.

[0055] The above - mentioned brain tumor classification device 1100 provided by the above - mentioned embodiments can implement the technical solutions described in the above - mentioned brain tumor classification method embodiments. The specific implementation principles of the above - mentioned modules or units can be referred to the corresponding content in the above - mentioned brain tumor classification method embodiments, and will not be elaborated here.

[0056] As Figure 12As shown, the present invention also correspondingly provides a magnetic resonance device 1200, which includes a processor 1201, a memory 1202, and a display 1203. Figure 8 Only some components of the magnetic resonance device 1200 are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0057] In some embodiments, the processor 1201 can be a central processing unit (CPU), a microprocessor, or other data processing chips, and is used to run the program code stored in the memory 1202 or process data, such as the brain tumor classification program for implementing the brain tumor classification method in the present invention.

[0058] In some embodiments, the memory 1202 can be an internal storage unit of the magnetic resonance device 1200, such as the hard disk or memory of the magnetic resonance device 1200. In other embodiments, the memory 1202 can also be an external storage device of the magnetic resonance device 1200, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the magnetic resonance device 1200.

[0059] In some embodiments, the display 1203 can be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (organic light-emitting diode) toucher, etc. The display 1203 is used to display the information of the magnetic resonance device 1200 and to display a visual user interface. The components 1201-1203 of the magnetic resonance device 1200 communicate with each other through a system bus.

[0060] In one embodiment, when the processor 1201 executes the brain tumor classification program in the memory 1202, the following steps can be implemented: Input the obtained brain tumor image into a well-trained brain tumor classification network, perform continuous three-time multi-scale dynamic channel sampling feature enhancement on the brain tumor image to obtain channel sampling features, and perform visual perception enhancement on the channel sampling features to obtain perception enhancement features; Perform prediction output on the perception enhancement features to obtain a brain tumor classification result.

[0061] It should be understood that when the processor 1201 executes the brain tumor classification program in the memory 1202, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the corresponding method embodiments above.

[0062] Accordingly, an embodiment of the present application further provides a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the programs or instructions are executed by a processor, the steps or functions in the brain tumor classification method provided by the above-mentioned method embodiments can be implemented.

[0063] Those skilled in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The computer program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0064] The above has introduced in detail the brain tumor classification method, device, magnetic resonance device, and storage medium provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for classifying brain tumors, characterized in that: include: The acquired brain tumor image is input into a well-trained brain tumor classification network, the brain tumor image is subjected to three consecutive multi-scale dynamic channel sampling feature enhancements to obtain channel sampling features, and the channel sampling features are subjected to visual perception enhancement to obtain perception enhancement features; The perception enhancement features are predicted and output to obtain a brain tumor classification result.

2. The brain tumor classification method according to claim 1, characterized in that: The multi-scale dynamic channel sampling feature enhancement includes channel attention convolution embedding and several times of multi-scale dynamic channel sampling, and the visual perception enhancement includes channel attention convolution embedding and visual state space enhancement.

3. The brain tumor classification method according to claim 2, characterized in that: The channel attention convolution embedding includes: Dividing the input features of the channel attention convolution embedding into a first image feature and a second image feature; Performing a maximum pooling operation on the first image feature to obtain global spatial information, and performing downsampling and 1×1 convolution on the global spatial information to obtain a channel attention feature; Performing local information feature convolution and downsampling on the second image feature to obtain a convolution embedding feature; The channel attention feature and the convolution embedding feature are merged and channel shuffle enhancement is performed to obtain the output feature of the channel attention convolution embedding.

4. The brain tumor classification method according to claim 2, characterized in that: The multi-scale dynamic channel sampling includes: Performing four consecutive dynamic channel sampling on the input features of the multi-scale dynamic channel sampling to obtain multi-scale sampling features of each dynamic channel sampling output; The multi-scale sampling features are feature fused, and then residual connection features are combined with the input features of the multi-scale dynamic channel sampling, and then nonlinear transformation is performed to obtain the output features of the multi-scale dynamic channel sampling.

5. The brain tumor classification method according to claim 4, characterized in that: The dynamic channel sampling comprises: Aggregating the input features of the dynamic channel sampling to obtain a scalar vector; Determine a sampling probability vector according to a preset sampling probability predictor and the scalar vector; Feature sampling is performed on the input features of the dynamic channel sampling according to a preset sampling strategy and the sampling probability vector to obtain output features of the dynamic channel sampling.

6. The brain tumor classification method according to claim 5, characterized in that: The step of performing feature sampling on the input features of the dynamic channel sampling according to the preset sampling strategy and the sampling probability vector to obtain the output features of the dynamic channel sampling includes: Dividing the input features of the dynamic channel sampling into a plurality of feature subsets, and determining the sampling probability of each of the feature subsets according to the sampling probability vector; If the sampling probability is less than or equal to a preset sampling threshold, feature sampling is performed on the channel with the highest activation value of the feature subset to obtain a sampling output subset; If the sampling probability is greater than a preset sampling threshold, average feature sampling is performed on all channels of the feature subset to obtain a sampling output subset; All of the sample output subsets are combined to obtain the output features of the dynamic channel sampling.

7. The brain tumor classification method according to claim 1, characterized in that: The visual state space enhancement includes: Performing image segmentation on the visual state space enhanced input features to obtain a plurality of input feature subgraphs; The input feature subgraph is subjected to visual selective cross scanning to obtain the output feature enhanced in the visual state space.

8. A brain tumor classification device, characterized in that: include: An image feature extraction unit is used to input the acquired brain tumor image into a well-trained brain tumor classification network, perform three consecutive multi-scale dynamic channel sampling feature enhancements on the brain tumor image to obtain channel sampling features, and perform visual perception enhancement on the channel sampling features to obtain perception enhancement features; The classification prediction output unit is used to predict and output the perception enhancement feature to obtain a brain tumor classification result.

9. A magnetic resonance device comprising a processor, a memory and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the brain tumor classification method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the brain tumor classification method according to any one of claims 1 to 7 is implemented.

Images