A brain tumor classification method and device, a magnetic resonance device and a readable storage medium

Abstract

The application provides a brain tumor classification method and device, a magnetic resonance equipment and a readable storage medium, and belongs to the field of medical image processing. The method comprises the following steps: inputting the obtained brain tumor image into a trained brain tumor classification network, performing three times of continuous multi-scale dynamic channel sampling feature enhancement on the brain tumor image to obtain a channel sampling feature, and performing visual perception enhancement on the channel sampling feature to obtain a perception enhanced feature; and performing prediction output on the perception enhanced feature to obtain a brain tumor classification result. Through the multi-scale dynamic channel sampling feature enhancement, the brain tumor image can be dynamically sampled at different scale levels, the subtle pathological features of the brain tumor image can be effectively extracted and enhanced, more accurate feature representations can be learned from the integrated subtle pathological features through the visual perception enhancement, the subtle pathological features of the brain tumor image are fully utilized, and the accuracy of brain tumor classification is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a brain tumor classification method, apparatus, magnetic resonance imaging device, and readable storage medium. Background Technology

[0002] Brain tumors are among the most life-threatening diseases affecting humans, occurring in men and women of all ages. Early detection and accurate diagnosis of the type of brain tumor are crucial for developing effective treatment plans.

[0003] Currently, deep learning architectures, especially convolutional neural networks and computer vision, have demonstrated significant performance improvements in medical image processing, effectively enhancing the efficiency of medical work. However, in the task of brain tumor classification, due to the high similarity between different types of brain tumors, distinguishing between them often relies on very subtle pathological changes within the tumor. Existing neural network image recognition methods may fail to capture these subtle pathological changes between different tumors when handling brain tumor classification tasks, resulting in low accuracy in brain tumor classification.

[0004] Therefore, existing technologies have the technical problem of failing to capture subtle pathological changes between different tumors in brain tumor classification tasks, resulting in low accuracy in brain tumor classification. Summary of the Invention

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

[0006] To address the aforementioned problems, this invention provides a method for classifying brain tumors, comprising:

[0007] The acquired brain tumor images are input into a well-trained brain tumor classification network. The brain tumor images are then subjected to three consecutive multi-scale dynamic channel sampling feature enhancements to obtain channel sampling features. Visual perception enhancement is then applied to the channel sampling features to obtain perceptual enhancement features.

[0008] The brain tumor classification result is obtained by predicting the output of the enhanced perception features.

[0009] In one possible implementation, multi-scale dynamic channel sampling feature enhancement includes channel attention convolutional embedding and several multi-scale dynamic channel samplings, while visual perception enhancement includes channel attention convolutional embedding and visual state space enhancement.

[0010] In one possible implementation, the channel attention convolutional embedding includes:

[0011] The input features embedded by the channel attention convolution are divided into first image features and second image features;

[0012] Max pooling is performed on the first image features to obtain global spatial information, and the global spatial information is downsampled and convolved with 1×1 to obtain channel attention features;

[0013] The convolutional embedding features are obtained by performing local information feature convolution and downsampling on the second image features;

[0014] The channel attention features and convolutional embedding features are merged and enhanced by channel shuffling to obtain the output features of the channel attention convolutional embedding.

[0015] In one possible implementation, multi-scale dynamic channel sampling includes:

[0016] The multi-scale sampling features of each dynamic channel are obtained by performing four consecutive dynamic channel samplings on the input features of the multi-scale dynamic channel sampling.

[0017] After feature fusion of multi-scale sampling features and merging with the input features of multi-scale dynamic channel sampling, residual connection features are combined and then nonlinear transformation is performed to obtain the output features of multi-scale dynamic channel sampling.

[0018] In one possible implementation, dynamic channel sampling includes:

[0019] The input features sampled from the dynamic channels are aggregated to obtain a scalar vector;

[0020] The sampling probability vector is determined based on the preset sampling probability predictor and the scalar vector;

[0021] The input features of dynamic channel sampling are obtained by performing feature sampling on the input features of dynamic channel sampling according to the preset sampling strategy and sampling probability vector.

[0022] In one possible implementation, the output features of the dynamic channel sampling are obtained by performing feature sampling on the input features of the dynamic channel sampling according to a preset sampling strategy and a sampling probability vector, including:

[0023] The input features of dynamic channel sampling are divided into several feature subsets, and the sampling probability of each feature subset is determined according to the sampling probability vector.

[0024] If the sampling probability is less than or equal to the preset sampling threshold, feature sampling is performed on the highest activation channel of the feature subset to obtain the sampled output subset.

[0025] If the sampling probability is greater than the preset sampling threshold, average feature sampling is performed on all channels of the feature subset to obtain the sampled output subset.

[0026] The output features of the dynamic channel sampling are obtained by merging all sampled output subsets.

[0027] In one possible implementation, the visual state space is enhanced, including:

[0028] The input features enhanced by the visual state space are segmented into several input feature sub-maps.

[0029] Visual selective cross-scanning is performed on the input feature sub-map to obtain visual state space-enhanced output features.

[0030] On the other hand, the present invention also provides a brain tumor classification device, comprising:

[0031] The image feature extraction unit is used to input the acquired brain tumor image into the 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.

[0032] The classification prediction output unit is used to predict and output brain tumor classification results based on the enhanced perception features.

[0033] On the other hand, the present invention also provides a magnetic resonance imaging device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned brain tumor classification method.

[0034] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described brain tumor classification method.

[0035] The beneficial effects of this invention are as follows: In the brain tumor classification method provided by this invention, the acquired brain tumor image is first input into a fully trained brain tumor classification network. The brain tumor image undergoes three consecutive multi-scale dynamic channel sampling feature enhancements to obtain channel sampling features. These channel sampling features are then enhanced by visual perception to obtain perceptual enhancement features. Finally, the perceptual enhancement features are predicted and output to obtain the brain tumor classification result. This invention, through multi-scale dynamic channel sampling feature enhancement, can dynamically sample tumor images at different scales, effectively extracting and enhancing subtle pathological features of the tumor image. By learning more accurate feature representations from the integrated subtle pathological features through visual perception enhancement, it fully utilizes the subtle pathological features of the tumor image, effectively improving the accuracy of brain tumor classification. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A flowchart illustrating an embodiment of the brain tumor classification method provided by the present invention;

[0038] Figure 2 This is a network structure diagram of the brain tumor classification network according to an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the channel attention convolution embedding process according to an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the multi-scale dynamic channel sampling process according to an embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram of the dynamic channel sampling process according to an embodiment of the present invention;

[0042] Figure 6 This is a flowchart illustrating the preset sampling strategy of an embodiment of the present invention;

[0043] Figure 7 This is a schematic diagram of the visual state space enhancement process according to an embodiment of the present invention;

[0044] Figure 8 This is a graph showing the model test results during the training phase of an embodiment of the present invention;

[0045] Figure 9 This is a graph showing the model classification accuracy test results of an embodiment of the present invention;

[0046] Figure 10 This is a visualization of the model test results in an embodiment of the present invention;

[0047] Figure 11 A schematic diagram of the structure of one embodiment of the brain tumor classification device provided by the present invention;

[0048] Figure 12 This is a schematic diagram of the structure of an embodiment of the magnetic resonance device provided by the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0050] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0051] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0052] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0053] This invention provides a method, apparatus, magnetic resonance imaging device, and readable storage medium for classifying brain tumors, which are described below.

[0054] It should be noted that the brain tumor classification method provided in this embodiment can be applied to brain tumor classification and recognition systems based on magnetic resonance image recognition or CT image recognition of brain tumors, and can also be applied to the identification of subtle pathological features in other parts of the body. The brain tumor classification and recognition system can be a software system running on a terminal device. The terminal device can be a server, tablet computer, laptop computer, personal computer, personal digital assistant, or mobile phone, etc. This application embodiment does not impose any restrictions on the specific type of terminal device.

[0055] Figure 1 This is a flowchart illustrating an embodiment of the brain tumor classification method provided by the present invention, as shown below. Figure 1 As shown, brain tumor classification methods include:

[0056] S101. Input the acquired brain tumor image into the fully 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.

[0057] S102. Predict and output the brain tumor classification results from the enhanced perception features.

[0058] In step S101, the brain tumor image obtained can be, but is not limited to, images captured by various types of medical equipment such as X-ray, ultrasound, or MRI. In this embodiment, subtle pathological features in the brain tumor image are extracted by three consecutive multi-scale dynamic channel sampling feature enhancements and one visual perception enhancement, and the brain tumor type is accurately predicted and classified based on the extracted subtle pathological features.

[0059] Compared with existing technologies, the brain tumor classification method provided in this invention first inputs the acquired brain tumor image into a fully trained brain tumor classification network. The brain tumor image undergoes three consecutive multi-scale dynamic channel sampling feature enhancements to obtain channel sampling features. These channel sampling features are then enhanced with visual perception to obtain perceptually enhanced features. Finally, the perceptually enhanced features are used to predict and output the brain tumor classification result. This invention, through multi-scale dynamic channel sampling feature enhancement, can dynamically sample tumor images at different scales, effectively extracting and enhancing subtle pathological features of the tumor image. By learning more accurate feature representations from the integrated subtle pathological features through visual perception enhancement, it fully utilizes the subtle pathological features of the tumor image, effectively improving the accuracy of brain tumor classification.

[0060] In some embodiments of the present invention, multi-scale dynamic channel sampling feature enhancement includes channel attention convolution embedding and several multi-scale dynamic channel samplings, and visual perception enhancement includes channel attention convolution embedding and visual state space enhancement.

[0061] Specifically, Figure 2 This is a network structure diagram of the brain tumor classification network according to an embodiment of the present invention, such as... Figure 2 As shown, the feature processing in this embodiment is divided into four stages. The first three stages consist of an attention convolutional embedding layer and several multi-scale dynamic channel sampling operations, while the fourth stage consists of an attention convolutional embedding layer and visual state space enhancement. Specifically, the first and second stages perform three multi-scale dynamic channel sampling operations, and the third stage performs nine multi-scale dynamic channel sampling operations.

[0062] In some embodiments of the present invention Figure 3 This is a schematic diagram of the channel attention convolution embedding process according to an embodiment of the present invention, as shown below. Figure 3 As shown, the channel attention convolutional embedding includes:

[0063] S301. Divide the input features embedded by the channel attention convolution into first image features and second image features;

[0064] S302. Perform max pooling on the first image features to obtain global spatial information, and downsample and convolve the global spatial information to obtain channel attention features;

[0065] S303. Perform local information feature convolution and downsampling on the second image features to obtain convolutional embedding features;

[0066] S304. Merge the channel attention features and convolutional embedding features and perform channel shuffling enhancement to obtain the output features of the channel attention convolutional embedding.

[0067] Specifically, in order to extract more effective pathological features from tumor images, the embodiment first performed channel attention convolution embedding in both multi-scale dynamic channel sampling and visual state space enhancement to improve the representation ability of features.

[0068] In the channel attention convolutional embedding, the embodiment first divides the image or feature map from the previous stage into two parts. Combining the channel attention module and convolutional embedding, local and global features are extracted and downsampled respectively. In the channel attention part, the embodiment uses a max-pooling layer to capture global spatial information and downsample the input, then increases the number of channels through a 1×1 convolution. In the convolutional embedding part, the embodiment uses a convolution with a kernel of 7, a stride of 4, and padding of 2 to capture local information, downsample it, and increase the number of channels. Finally, the embodiment merges the features obtained from the channel attention part and the convolutional embedding part to restore the number of channels, and uses a channel shuffling method to enhance the representational power of the merged channels.

[0069] In some embodiments of the present invention Figure 4 This is a schematic diagram of the multi-scale dynamic channel sampling process according to an embodiment of the present invention, as shown below. Figure 4 As shown, multi-scale dynamic channel sampling includes:

[0070] S401. Perform four consecutive dynamic channel samplings on the input features of the multi-scale dynamic channel sampling to obtain the multi-scale sampling features of each dynamic channel sampling output.

[0071] S402. After feature fusion of the multi-scale sampling features and the input features of multi-scale dynamic channel sampling, residual connection features are merged and then nonlinear transformation is performed to obtain the output features of multi-scale dynamic channel sampling.

[0072] Specifically, in each layer's multi-scale dynamic channel sampling, the input features are processed through four sequential dynamic channel samplings. Except for the first dynamic channel sampling, each dynamic channel sampling uses the output of the previous dynamic channel sampling as the input to that layer. In the first dynamic channel sampling, the number of channels remains constant with the number 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 2C channels. A 1×1 convolution is then used to reduce the channel size of the feature map from 2C to C, while weighting and fusing information from different scales. Finally, residual connections are used to combine the obtained features with the initial input features from the multi-scale dynamic channel sampling to alleviate the gradient vanishing problem, and a 1×1 convolutional layer is used to perform a nonlinear transformation on the fused features. Furthermore, each 1×1 convolutional layer is followed by batch normalization and a ReLU activation layer.

[0073] In some embodiments of the present invention Figure 5 This is a schematic diagram of the dynamic channel sampling process according to an embodiment of the present invention, as shown below. Figure 5 As shown, dynamic channel sampling includes:

[0074] S501. Aggregate the information of the input features sampled by the dynamic channel to obtain a scalar vector;

[0075] S502. Determine the sampling probability vector based on the preset sampling probability predictor and the scalar vector;

[0076] S503. Based on the preset sampling strategy and sampling probability vector, feature sampling is performed on the input features of dynamic channel sampling to obtain the output features of dynamic channel sampling.

[0077] In some embodiments of the present invention Figure 6 This is a flowchart illustrating the preset sampling strategy of an embodiment of the present invention, as shown below. Figure 6 As shown, the output features of dynamic channel sampling are obtained by performing feature sampling on the input features of dynamic channel sampling according to a preset sampling strategy and sampling probability vector, including:

[0078] S601. Divide the input features of dynamic channel sampling into several feature subsets, and determine the sampling probability of each feature subset according to the sampling probability vector.

[0079] S602. If the sampling probability is less than or equal to the preset sampling threshold, feature sampling is performed on the highest activation value channel of the feature subset to obtain the sampling output subset.

[0080] S603. If the sampling probability is greater than the preset sampling threshold, average feature sampling is performed on all channels of the feature subset to obtain the sampled output subset.

[0081] S604. Merge all sampled output subsets to obtain the output characteristics of dynamic channel sampling.

[0082] Specifically, in dynamic channel sampling, taking the first dynamic channel sampling as an example, the embodiment first samples the input features... The information from each channel is aggregated into a scalar, forming a one-dimensional scalar vector. :

[0083]

[0084] in, express The One channel, The first Each channel from Dimensionality is reduced to scalar.

[0085] Next, a learnable sampling probability predictor is used. To process vectors To generate sampling probability vectors :

[0086]

[0087]

[0088] in, and It's about weights and biases. It is the Sigmoid activation function.

[0089] Then the implementation will input features Divided into several feature subsets ,in These are grouping parameters. And based on the sampling probability vector... Determine the sampling probability of each subset .

[0090] The implementation then sets the sampling probability of each subset. With preset sampling threshold The sampling probability of a subset is compared. If the sampling probability of a subset is less than or equal to a preset sampling threshold, the channel with the highest activation value in that subset is selected as the output channel for sampling. Otherwise, all channels in the subset are averaged and fused for sampling, as expressed by the formula:

[0091]

[0092] Finally, after completing the sampling process for each feature subset, the output features of the dynamic channel sampling are obtained based on the obtained sampled output subsets. For the entire input feature map, the output can be represented as:

[0093]

[0094] The embodiment uses the above sampling strategy to enable each pixel position to select the required channel combination according to the features of the input data, and effectively extracts subtle pathological features at different scales in the image by executing the sampling strategy multiple times at different scales.

[0095] In some embodiments of the present invention Figure 7 This is a schematic diagram of the visual state space enhancement process according to an embodiment of the present invention, such as... Figure 7 As shown, visual state space enhancement includes:

[0096] S701. The input features enhanced by the visual state space are segmented into several input feature sub-maps.

[0097] S702. Perform visual selective cross-scanning on the input feature sub-map to obtain visual state space-enhanced output features.

[0098] Specifically, in the visual state space enhancement process, the embodiment employs VMamba (Visual State Space Model) to learn and enhance vision. VMamba first segments the input feature image into several input feature sub-maps, and then performs a visually selective cross-scan operation on the input feature sub-maps. The visually selective scan operation can process the input data causally and capture information within the scanned portion of the data in this way, thereby obtaining the contextual information embedded in the input and ensuring the dynamic nature of the weights within the mechanism. However, 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 address this issue, a cross-scan approach is proposed. The cross-scan employs a four-way scanning strategy, i.e., scanning from the four corners of the feature map to their 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 linear computational complexity. The final embodiment, through visual state space enhancement, can learn more accurate feature representations from integrated subtle pathological features.

[0099] To verify the effectiveness of the present invention, the embodiments were experimentally verified.

[0100] First, the embodiment uses a brain tumor dataset containing four types of tumors, including meningiomas, gliomas, pituitary tumors, and tumor-free images. The brain tumor dataset is then preprocessed by cropping all brain tumor images to 224×224 pixels and normalizing each image by dividing by 225 to scale all pixel values ​​to the range [0, 1]. The brain tumor dataset is then divided into training and testing sets in an 8:2 ratio.

[0101] During the experimental phase, the model was implemented and run based on the Linux operating system and the PyTorch 1.13.0 framework. The computer configuration used was: Intel(R) Xeon(R) Gold 6330 CPU @ 2.00GHz, and the GPU model was NVIDIA GeForce RTX 3090 (24 GB). The dataset was loaded by writing code, and the model structure was implemented through code functions.

[0102] During training, the cross-entropy loss function was used to calculate the difference between the predicted distribution and the true label distribution. Parameter updates were performed using the AdamW optimizer with a learning rate of 5e-4, momentum of 0.9, and weight decay of 0.05. Training lasted for 500 epochs with a batch size of 32.

[0103] Figure 8 This is a graph showing the model test results during the training phase of an embodiment of the present invention. Figure 9 This is a graph showing the model classification accuracy test results of an embodiment of the present invention. Figure 10 This is a visualization of the model test results in an embodiment of the present invention. Figures 8 to 10 As shown, during the training process, the model is continuously tested using the test set. The test loss and classification accuracy are as follows: Figure 8 As shown, with the increasing number of training generations, the model's loss gradually decreases until convergence, while the accuracy gradually increases. The output of the model with the best results is as follows. Figure 9 The confusion matrix of the test results is shown, and the classification accuracy of the model for each type of image is calculated to evaluate the actual generalization performance of the model, resulting in the following: Figure 10 The results are shown in the visualization.

[0104] In summary, to capture subtle pathological features and achieve accurate classification of brain tumor images, this invention first inputs the acquired brain tumor images into a well-trained brain tumor classification network. The brain tumor images undergo three consecutive multi-scale dynamic channel sampling feature enhancements to obtain channel sampling features. These channel sampling features are then enhanced with visual perception to obtain perceptually enhanced features. Finally, the perceptually enhanced features are used for prediction output to obtain the brain tumor classification result. This invention, through multi-scale dynamic channel sampling feature enhancement, can dynamically sample tumor images at different scales, effectively extracting and enhancing subtle pathological features of tumor images. By learning more accurate feature representations from the integrated subtle pathological features through visual perception enhancement, this invention fully utilizes the subtle pathological features of tumor images, effectively improving the accuracy of brain tumor classification.

[0105] To better implement the brain tumor classification method in this invention embodiment, based on the brain tumor classification method, correspondingly, as follows: Figure 11 As shown, the present invention also provides a brain tumor classification device, the brain tumor classification device 1100 comprising:

[0106] The image feature extraction unit 1101 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.

[0107] The classification prediction output unit 1102 is used to predict and output brain tumor classification results based on the enhanced perception features.

[0108] The brain tumor classification device 1100 provided in the above embodiments can realize the technical solutions described in the above brain tumor classification method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above brain tumor classification method embodiments, and will not be repeated here.

[0109] like Figure 12 As shown, the present invention also 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 components shown, and more or fewer components may be implemented instead.

[0110] In some embodiments, processor 1201 may be a central processing unit (CPU), microprocessor or other data processing chip, used to run program code stored in memory 1202 or process data, such as a brain tumor classification program that implements the brain tumor classification method of the present invention.

[0111] In some embodiments, memory 1202 may be an internal storage unit of the magnetic resonance imaging (MRI) device 1200, such as a hard disk or memory of the MRI device 1200. In other embodiments, memory 1202 may also be an external storage device of the MRI device 1200, such as a pluggable hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on the MRI device 1200.

[0112] In some embodiments, display 1203 may be an LED display, a liquid crystal display, a touch-screen liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 1203 is used to display information from the magnetic resonance imaging (MRI) device 1200 and to display a visual user interface. Components 1201-1203 of the MRI device 1200 communicate with each other via a system bus.

[0113] In one embodiment, when the processor 1201 executes the brain tumor classification program in the memory 1202, the following steps can be performed:

[0114] The acquired brain tumor images are input into a well-trained brain tumor classification network. The brain tumor images are then subjected to three consecutive multi-scale dynamic channel sampling feature enhancements to obtain channel sampling features. Visual perception enhancement is then applied to the channel sampling features to obtain perceptual enhancement features.

[0115] The brain tumor classification result is obtained by predicting the output of the enhanced perception features.

[0116] It should be understood that when the processor 1201 executes the brain tumor classification program in the memory 1202, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0117] Accordingly, this application also provides a computer-readable storage medium for storing a computer-readable program or instruction. When the program or instruction is executed by a processor, it can implement the steps or functions in the brain tumor classification method provided in the above-described method embodiments.

[0118] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0119] The brain tumor classification method, apparatus, magnetic resonance equipment, and storage medium provided by the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A brain tumor classification method, characterized by, The method comprises the following steps: inputting the acquired brain tumor image into a trained brain tumor classification network, performing multi-scale dynamic channel sampling feature enhancement on the brain tumor image for three times in succession to obtain channel sampling features, performing visual perception enhancement on the channel sampling features to obtain perception enhanced features, wherein the multi-scale dynamic channel sampling feature enhancement comprises channel attention convolution embedding and several times of multi-scale dynamic channel sampling, and the visual perception enhancement comprises channel attention convolution embedding and visual state space enhancement; the multi-scale dynamic channel sampling comprises: performing dynamic channel sampling on the input features of the multi-scale dynamic channel sampling for four times in succession to obtain multi-scale sampling features of the output of each dynamic channel sampling; performing residual connection feature fusion on the multi-scale sampling features and the input features of the multi-scale dynamic channel sampling, and then performing nonlinear transformation to obtain the output features of the multi-scale dynamic channel sampling; the dynamic channel sampling comprises: performing information aggregation on the input features of the dynamic channel sampling to obtain a scalar vector; determining a sampling probability vector according to a preset sampling probability predictor and the scalar vector; and 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; performing prediction output on the perception enhanced features to obtain a brain tumor classification result.

2. The brain tumor classification method of claim 1, wherein, The channel attention convolution embedding comprises: dividing the input features of the channel attention convolution embedding into first image features and second image features; performing max-pooling operation on the first image features to obtain global spatial information, and performing down-sampling and 1x1 convolution on the global spatial information to obtain channel attention features; performing local information feature convolution and down-sampling on the second image features to obtain convolution embedding features; merging the channel attention features and the convolution embedding features and performing channel shuffle enhancement to obtain the output features of the channel attention convolution embedding.

3. The brain tumor classification method of claim 1, wherein, The 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 comprises: dividing the input features of the dynamic channel sampling into several feature subsets, and determining 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, performing feature sampling on the highest activation value channel of the feature subset to obtain a sampling output subset; if the sampling probability is greater than the preset sampling threshold, performing average feature sampling on all channels of the feature subset to obtain a sampling output subset; merging all the sampling output subsets to obtain the output features of the dynamic channel sampling.

4. The brain tumor classification method of claim 1, wherein, The visual state space enhancement comprises: performing image segmentation on the input features of the visual state space enhancement to obtain several input feature subgraphs; performing visual selective cross-scan on the input feature subgraphs to obtain the output features of the visual state space enhancement.

5. A brain tumor classification apparatus characterized by comprising: The method comprises the following steps: The image feature extraction unit is configured to input the acquired brain tumor image into the trained brain tumor classification network, perform continuous three times of multi-scale dynamic channel sampling feature enhancement on the brain tumor image to obtain channel sampling features, perform visual perception enhancement on the channel sampling features to obtain perception enhanced features, wherein the multi-scale dynamic channel sampling feature enhancement comprises channel attention convolution embedding and several times of multi-scale dynamic channel sampling, and the visual perception enhancement comprises channel attention convolution embedding and visual state space enhancement. The multi-scale dynamic channel sampling comprises: performing continuous four times of dynamic channel sampling on input features of the multi-scale dynamic channel sampling to obtain multi-scale sampling features of outputs of each dynamic channel sampling; performing residual connection feature fusion on the multi-scale sampling features and the input features of the multi-scale dynamic channel sampling, and then performing nonlinear transformation to obtain output features of the multi-scale dynamic channel sampling; and the dynamic channel sampling comprises: performing information aggregation on input features of the dynamic channel sampling to obtain a scalar vector; determining a sampling probability vector according to a preset sampling probability predictor and the scalar vector; and 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 output features of the dynamic channel sampling. The classification prediction output unit is configured to perform prediction output on the perception enhanced features to obtain a brain tumor classification result.

6. A magnetic resonance apparatus comprising a processor, a memory, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the brain tumor classification method according to any one of claims 1 to 4.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the brain tumor classification method according to any one of claims 1 to 4.

Images