Adult type diffuse glioma nuclear magnetic resonance image classification diagnosis method

Abstract

The invention relates to a nuclear magnetic resonance image classification diagnosis method for adult diffuse glioma. The method comprises the following specific steps: step 1, respectively acquiring T2-fluid attenuation inversion recovery T2-FLAIR and contrast enhancement T1 weighted imaging CE-T1WI magnetic resonance images of glioma labeled by a doctor according to histological and genomics information; step 2, carrying out data preprocessing operation on the T2-FLAIR image data and the CE-T1WI image data; step 3, each pair of T2-FLAIR and CE-T1WI images after preprocessing is sent to a decision tree diagnosis model, and an Attention-DenseNet model is used for learning and extracting the characteristics of the T2-FLAIR and CE-T1WI images; 4, obtaining a prediction result based on a plurality of independently trained Attention-DenseNet models, fusing the outputs of the plurality of models by adopting a voting mechanism or a weighted average strategy, and generating a final classification decision; and 5, sending the obtained final classification to a platform, and making final diagnosis based on a classification result.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image data processing, and in particular to a method for classifying and diagnosing adult diffuse glioma nuclear magnetic resonance images. Background Art

[0002] Glioma is the most common primary malignant tumor of the central nervous system in adults. In 2016, the fourth revised classification of central nervous system (CNS) tumors of the World Health Organization (WHO) (hereinafter referred to as the fourth revised classification) clarified for the first time the importance of IDH mutation status (presence or absence of mutations in IDH1 and IDH2 genes) in the diagnosis and classification of diffuse gliomas, providing an important reference for the comprehensive diagnosis and classification of diffuse gliomas. It is well known that IDH wild-type diffuse astrocytomas are more aggressive and have a worse prognosis than IDH mutant diffuse astrocytomas, indicating that IDH mutant and IDH wild-type diffuse astrocytomas have significant differences in clinical characteristics and biological behaviors. It is worth noting that the prognosis and survival of patients with IDH mutant astrocytomas of the same histological grade are also different, and the prognosis of low-grade IDH mutant astrocytomas may be similar to that of high-grade IDH mutant astrocytomas. This suggests that in addition to IDH and histological grade, there are other molecular markers that can provide strong prognostic information for diffuse gliomas. Therefore, the fifth edition of the WHO Classification of Central Nervous System Tumors (hereinafter referred to as the fifth edition of the new classification) published in 2021 once again clarified the status and value of genetic biomarkers in the diagnosis and grading of gliomas, emphasizing the combination of histological morphological diagnosis and genetic biomarkers for grading, and that genetic biomarkers are superior to histological morphology in terms of evidence.

[0003] In the existing technology, radiomics and deep learning studies have predicted the histological grade and genetic biomarkers of gliomas and achieved satisfactory results. However, the biological behavior of gliomas does not rely on a single gene biomarker, and there are interactions and connections between different genes. There are also a few studies that predict multiple gene biomarkers at the same time, but they do not combine histological grade and genetic biomarkers for comprehensive diagnosis and classification of gliomas, which limits the transformation of research results into clinical practice. According to the fifth edition of the new classification, adult diffuse gliomas are divided into three categories: astrocytoma, IDH mutation; olive branch glioma, IDH mutation and 1p / 19q deletion; and glioblastoma, IDH wild type. It is worth noting that astrocytoma, IDH-wild type, WHO grade 2-3 is not included in this new classification, which we believe is worthy of discussion and improvement. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a method for reclassifying adult diffuse glioma based on magnetic resonance imaging and an artificial intelligence decision tree of a deep learning network, which can combine histological grading and genetic biomarkers to perform comprehensive diagnosis and classification of glioma. The specific technical solution is as follows:

[0005] A method for classifying and diagnosing adult diffuse glioma using magnetic resonance imaging, comprising the following specific steps:

[0006] Step 1: Obtain T2-fluid attenuated inversion recovery (T2-FLAIR) and contrast-enhanced T1-weighted imaging (CE-T1WI) magnetic resonance images of gliomas labeled by doctors based on histological and genomic information;

[0007] Step 2: Perform data preprocessing on T2-FLAIR and CE-T1WI image data;

[0008] Step 3: Each pair of preprocessed T2-FLAIR and CE-T1WI images is sent to the decision tree diagnosis model, and the Attention-DenseNet model learns and extracts the features of T2-FLAIR and CE-T1WI images;

[0009] Step 4: Obtain prediction results based on multiple independently trained Attention-DenseNet models, and use a voting mechanism or weighted average strategy to fuse the outputs of multiple models to generate the final classification decision;

[0010] Step 5: Send the final classification to the platform and make a final diagnosis based on the classification results.

[0011] Preferably, in step one, the histological grades labeled on the T2-fluid attenuated inversion recovery (T2-FLAIR) and contrast-enhanced T1-weighted imaging (CE-T1WI) magnetic resonance images of gliomas based on histological and genomic information include: IDH, 1p / 19q, CDKN2A / B, epidermal growth factor receptor (EGFR), and TERT.

[0012] As a priority: the step 2 is specifically as follows;

[0013] 2.1. Co-register the images to align images of different modalities and reduce positional deviation;

[0014] 2.2. Perform offset field correction to eliminate uneven brightness in the image and ensure image uniformity;

[0015] 2.3. Then, the image is anisotropically resampled to unify the spatial resolution to facilitate subsequent feature extraction and classification;

[0016] 2.4. Then crop the image and extract the target area to reduce the amount of calculation and focus on the lesion area.

[0017] As a preference: in the step 2.3,

[0018] The image is adjusted to a uniform size using the resampling method. The expression for image resampling is shown in Formula 1:

[0019]

[0020] Among them, I new (x, y) is the pixel value of the image at the new size, I is the original image, W orig ,H orig is the width and height of the original image, W new ,H new The width and height of the new image size.

[0021] Since tumors vary in size and deep learning models require uniform input dimensions, we cropped MR images based on the tumor lesion center, which was obtained by matrix multiplication between the MR image volume and the mask data. This approach preserves the actual size and margin information of the tumor, as previous radiomics literature has shown that tumor size and margins are key factors in predicting tumor genotype. During the training phase, data augmentation was performed on each MR image to improve the generalization ability of the model, including vertical flipping, horizontal flipping, translational rotation, random rotation, addition of Gaussian noise, and projection transformation. To avoid inconsistent voxel resolution caused by MR images from different devices, each training image was subjected to random anisotropy before input into the network.

[0022] Preferably, the Attention-DenseNet model structure includes a first input channel for inputting images of the T2-FLAIR modality, which is connected in sequence to: a first Dense Block, a first Attention layer, a second DenseBlock, a second Attention layer, a third Dense Block, a third Attention layer, a pooling layer, and a linear layer; a second input channel for inputting images of the CE-T1WI modality, which is connected in sequence to: a fourth Dense Block, a fourth Attention layer, a fifth Dense Block, a fifth Attention layer, a sixth Dense Block, a sixth Attention layer, a pooling layer, and a linear layer; the first input channel and the second input channel have multiple cross-channel connections between the Attention layer and the pooling layer, respectively, for feature fusion; the features output by the pooling layer are connected by a splicing operation before the linear layer for the final prediction result output.

[0023] Preferably, the expression of the Dense Block is as shown in Formula 2:

[0024] H l+1 =H l +F l (H l , W l ) (2)

[0025] Among them, H l represents the input feature map of the lth layer in DenseBlock, H l+1 represents the output feature map of the l+1th layer, F l Represents the nonlinear transformation function of the lth layer, including convolution operation and activation function, W l is the convolution kernel weight of the lth layer.

[0026] Preferably, the Attention-DenseNet model architecture uses a self-attention mechanism to weight the feature map, and the self-attention weight calculation formula is:

[0027]

[0028] Among them, Q is the query matrix, K is the key matrix, V is the value matrix, K T is the transpose of the key matrix, d k is the dimension of the key matrix.

[0029] Preferably, the Attention-DenseNet model architecture adopts the Focal Loss loss function to deal with the sample imbalance problem. The expression of the loss function is:

[0030] FL(p t )=-α t (1-p t ) γ log(p t ) (4)

[0031] Among them, p t is the predicted probability of the positive class, α t is the importance coefficient of positive and negative classes, and γ is the parameter for adjusting the weight of easy-to-classify samples.

[0032] Preferably, the final classification decision is a voting-based ensemble method for ensemble Attention-DenseNet predictions from different folds.

[0033] The beneficial effects of the present invention are as follows: by constructing a decision tree diagnostic platform based on deep learning, adult diffuse gliomas are effectively and automatically classified into five subtypes, thereby improving the accuracy and efficiency of diagnosis. The platform combines genetic biomarkers and histological grading to assist clinicians in developing personalized treatment plans, reduce human errors, adapt to the latest WHO classification standards, and promote the standardization and scientificization of tumor classification. The scheme of the present invention significantly improves the ability of neuroradiologists in diagnosing and classifying adult diffuse gliomas, and provides strong technical support for precision medicine.

[0034] First of all, the research motivation of this invention stems from the limitations of current adult diffuse glioma classification methods in clinical applications. Existing classification methods rely too much on a single histological feature and cannot fully combine genetic biomarkers, resulting in insufficient classification accuracy and affecting the effectiveness of clinical treatment decisions. To solve this problem, the present invention introduces artificial intelligence and deep learning technology, and designs an artificial intelligence decision tree diagnostic platform (DTDP) based on magnetic resonance imaging and deep convolutional neural network (CNN), aiming to achieve comprehensive classification and accurate diagnosis of adult diffuse glioma.

[0035] The present invention uses a deep learning network Attention-DenseNet to extract genetic and histological features from patient MRI images. This strategy effectively overcomes the shortcomings of traditional methods that are difficult to capture complex tumor features, allowing the classification model to better identify and classify different types of gliomas. Through extensive training and verification, the model of the present invention performs well in classification accuracy and greatly improves the accuracy of glioma classification.

[0036] The decision tree structure adopted by the present invention connects multiple convolutional neural network models in series and in parallel to form a comprehensive diagnostic platform. This design enables different genetic biomarkers and histological grades to be organically combined in the same system, thereby achieving a more comprehensive diagnosis. The design of the decision tree structure takes into account the diversity of adult diffuse gliomas, making the classification results more in line with actual clinical needs and improving the practicality and accuracy of diagnosis.

[0037] In the present invention, a series of classification tasks and constraints are set to ensure that the diagnostic model can give the best classification results. The key to this step is that by reasonably setting the classification tasks and constraints, the model can effectively avoid common problems such as overfitting, thereby providing more stable and reliable classification results. The setting of constraints not only ensures the accuracy of the classification results, but also improves the generalization ability of the model, making the present invention more valuable for practical clinical applications.

[0038] The innovation of this invention is also reflected in the comprehensive evaluation of different gene biomarkers. Through the comprehensive analysis of gene markers such as IDH, 1p / 19q, CDKN2A / B, EGFR, TERT, etc., researchers can deeply understand the molecular characteristics of glioma and provide a scientific basis for subsequent clinical treatment and personalized medicine. The quantitative analysis in this process provides strong data support for further optimization of the classification model, making the diagnostic platform more instructive and effective in practical applications.

[0039] The present invention also verifies the developed diagnostic platform by introducing an independent test data set, ensuring the consistency and accuracy of the model on different data sets. The introduction of this step greatly enhances the reliability and adaptability of the research. Through independent verification, the robustness and reliability of the developed diagnostic platform in practical applications can be ensured, providing a reliable tool for clinicians.

[0040] In summary, the present invention successfully solves the current problems in the classification of adult diffuse gliomas through innovative research methods and comprehensive technical means. By introducing deep learning models, designing decision tree structures, rationally setting classification tasks and constraints, and comprehensively evaluating gene biomarkers, the present invention has achieved remarkable results in the classification research of adult diffuse gliomas, providing useful experience and methods for in-depth exploration of related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Shown is the data preprocessing workflow of the present invention.

[0042] Figure 2 Shown is the Attention-DenseNet model training flow chart of the present invention.

[0043] Figure 3 Shown is a flow chart of the voting-based integration method of the present invention.

[0044] Figure 4 Shown is a workflow diagram of the platform for reclassifying adult diffuse glioma based on magnetic resonance imaging and an artificial intelligence decision tree of a deep learning network of the present invention.

[0045] Figure 5 Shown is a model architecture diagram of the platform for reclassifying adult diffuse glioma based on magnetic resonance imaging and an artificial intelligence decision tree of a deep learning network.

[0046] Figure 6 It is a flow chart of the five-fold cross-validation experiment of the present invention.

[0047] Figure 7 This is a decision curve analysis diagram obtained by performing five-fold cross validation on different data sets of the present invention. DETAILED DESCRIPTION

[0048] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0049] The embodiment of the present application provides a method for reclassifying adult diffuse gliomas based on an artificial intelligence decision tree based on magnetic resonance imaging and deep learning networks, combined with genetic biomarkers and histological grading, to assist clinicians in developing personalized treatment plans, reduce human errors, adapt to the latest WHO classification standards, and promote the standardization and scientificization of tumor classification. This innovation significantly improves the ability of neuroradiologists in diagnosing and classifying adult diffuse gliomas, and provides strong technical support for precision medicine.

[0050] The technical solution in the embodiment of the present application is to solve the above technical problems, and the specific technical solution is as follows:

[0051] In view of the problem of multimodal data preprocessing, the embodiment of the present invention proposes the following Figure 1 The data preprocessing workflow shown. This embodiment collects 210 T2-fluid attenuated inversion recovery (T2-FLAIR) and contrast-enhanced T1-weighted imaging (CE-T1WI) magnetic resonance images of gliomas from the First Affiliated Hospital of Chongqing Medical University and the Cancer Imaging Archives (TCIA), as well as genomic and histological information from the Cancer Genome Atlas (TCGA) and the First Affiliated Hospital of Chongqing Medical University. The patient's data was randomly divided into a training set and a validation set at a ratio of approximately 4:1. The training set is used to train the network, and the validation set is used to select the best performing model. Finally, an independent test data set consisting of 56 patients was prepared for patient-level multi-classification performance evaluation of the model. In the image preprocessing stage, this embodiment first aligns the multimodal magnetic resonance images to ensure image alignment between different modalities and eliminate errors caused by patient motion or different imaging angles. Then, the intensity inhomogeneity in the image is reduced by bias field correction, eliminating brightness changes caused by magnetic field inhomogeneity. In addition, anisotropic resampling technology is used to resample the image to a uniform resolution to eliminate the resolution inconsistency caused by differences in equipment or scanning parameters, such as

[0052] As shown in formula 1:

[0053]

[0054] Among them, I new (x, y) is the pixel value of the image at the new size, I is the original image, W orig ,H origis the width and height of the original image, W new ,H new The width and height of the new image size.

[0055] Finally, the image was cropped to focus on the region of interest (ROI), thereby reducing irrelevant information and improving the computational efficiency and processing accuracy of the model. These preprocessing steps provide a stable and reliable data basis for subsequent image analysis and diagnostic decisions.

[0056] Reference Figure 2-Figure 5 , a method for reclassifying adult diffuse gliomas based on magnetic resonance imaging and an artificial intelligence decision tree with a deep learning network, including:

[0057] Construction and training phase of the Attention-DenseNet model:

[0058] In the construction and training phase of the Attention-DenseNet model, such as Figure 2 As shown in the figure, the Attention-DenseNet model is based on the DenseNet-121 architecture, and a self-attention mechanism is added to enhance the model's ability to capture complex pathological features. The model training process involves extracting features from magnetic resonance imaging (MRI) images, using image data from two sequences: contrast-enhanced T1-weighted imaging (CE-T1WI) and T2-fluid-attenuated inversion recovery (T2-FLAIR). The DenseNet-121 architecture is trained on preprocessed MRI images, and retains the image's detailed information through layer-by-layer connections, while avoiding the information loss problem that may occur in deep networks.

[0059] Attention-DenseNet further strengthens the fusion of different sequence features by introducing a self-attention mechanism after each dense block in the DenseNet architecture, so that the model can more accurately capture the relationship and feature differences between different tumor lesions. The DenseNet architecture is shown in Formula 2:

[0060] H l+1 =H l +F l (H l , W l ) (2)

[0061] Among them, H l represents the input feature map of the lth layer in DenseBlock, H l+1 represents the output feature map of the l+1th layer, F l Represents the nonlinear transformation function of the lth layer, including convolution operation and activation function, Wl is the convolution kernel weight of the lth layer.

[0062] The attention mechanism is shown in Formula 3:

[0063]

[0064] Among them, Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the dimension of the key matrix,

[0065] K T is the transpose of the key matrix.

[0066] During the training process, the model uses the Adam optimizer, the initial learning rate is set to 1e-4, and the cross entropy loss function and the Focal Loss function are used to process the unbalanced data. The application of the Focal Loss function is particularly effective in dealing with the problem of class imbalance, allowing Attention-DenseNet to better handle complex multi-classification tasks, as shown in Formula 4:

[0067] FL(p t )=-α t (1-p t ) γ log(p t ) (4)

[0068] Among them, p t is the predicted probability of the positive class, α t is the importance coefficient of positive and negative classes, and γ is the parameter for adjusting the weight of easy-to-classify samples.

[0069] Voting-based ensemble approach phase:

[0070] After the Attention-DenseNet model was trained and validated, a voting-based ensemble method was further adopted, such as Figure 3 As shown. This ensemble method improves the classification accuracy and robustness of the model by integrating the prediction results of multiple independent models. Specifically, the data set is randomly divided into five groups, each group of data is used as a validation set in turn, and the remaining data is used as a training set for five rounds of training. An independent model is generated in each round of training. Through this 5-fold cross-validation method, the impact of the uneven segmentation of the data set on the model performance can be effectively reduced. Subsequently, the prediction results of different models are integrated into the final classification result through a voting mechanism. This ensemble method effectively improves the stability and accuracy of the final diagnosis by weighted averaging the prediction scores of multiple models, especially when dealing with a variety of glioma pathological characteristics, showing strong adaptability and broad application prospects.

[0071] Modeling decision tree diagnostic platform stage:

[0072] Based on the Attention-DenseNet model and voting integration method, the present invention further constructs an artificial intelligence decision tree diagnosis platform, such as Figure 4 As shown. The platform combines the advantages of deep learning models and implements the reclassification process of adult diffuse glioma in the form of a decision tree. In this decision tree platform, each node corresponds to a specific classification task, such as binary classification of IDH mutation status, 1p / 19q deletion status, CDKN2A / B homologous deletion status, epidermal growth factor receptor (EGFR) status, and telomerase reverse transcriptase promoter (TERTp) status. These classification tasks are performed by the trained Attention-DenseNet model, and these nodes are connected in series and parallel to form a complete decision tree diagnostic process. First, the decision tree divides patients into IDH mutant and IDH wild type according to the IDH mutation status, and then further classifies patients according to other genes and histological grades. Ultimately, patients were automatically classified into five different subtypes, namely oligodendroglioma, IDH mutant astrocytoma (WHO grade 2-3), IDH mutant astrocytoma (WHO grade 4), IDH wild-type astrocytoma (WHO grade 2-3), and glioblastoma (IDH wild-type). Figure 5 As shown in the figure, the decision tree diagnostic platform of the present invention achieves the reclassification of adult diffuse glioma by accurately predicting different classification tasks. The platform not only performs well in multiple classification tasks, but also achieves significant improvement in the overall diagnostic performance of the model.

[0073] Reference Figure 6-Figure 7 , which is the five-fold cross-validation experimental process of the present invention:

[0074] like Figure 6 As shown in the figure. Specifically, the original data set is randomly divided into five subsets, each of which is used as a validation set, and the remaining subsets are used as training sets, and five trainings are performed in sequence. In each round of training, the model will optimize parameters based on the data of the training set, and the model performance will be evaluated on the validation set. Through five rounds of cross-validation, the model bias caused by uneven data segmentation can be effectively reduced, and more robust classification results can be obtained. The models obtained from each training are retained and used in the subsequent integration method to form the final decision model.

[0075] like Figure 7As shown, in order to further evaluate the practical application effect of the model on different data sets, the present invention performs a decision curve analysis based on five-fold cross validation. Decision curve analysis (DCA) is an evaluation method that can take into account the preferences of patients or decision makers, and is used to measure the actual benefits of classification models at different thresholds. Specifically, the present invention performs five-fold cross validation on different data sets, and draws a decision curve based on the results of each cross validation. By observing the shape and trend of the decision curve, the performance of the model under different threshold settings can be evaluated, and then the optimal application scenario of the model can be determined. The results show that the Attention-DenseNet model shows good practicality in the prediction of multiple gene biomarkers such as IDH, 1p / 19q, CDKN2A / B, EGFR, and TERT. In particular, for the classification tasks of IDH status and histological grading (WHO 2-3 and WHO 4), the decision curve of the model shows a higher diagnostic benefit, proving its value in practical clinical applications.

[0076] In this embodiment, in order to comprehensively evaluate the proposed Attention-DenseNet model, five indicators are used, including accuracy (ACC), area under the curve (AUC), sensitivity (Sen), specificity (Spe), and F1 score (F1). All indicators are defined as follows:

[0077]

[0078]

[0079] In classification tasks, ACC and AUC are commonly used indicators to evaluate the overall performance of the model. Sen refers to the proportion of correctly predicted positive categories. Spe is the proportion of correctly predicted negative samples among all samples that are actually negative. The F1 score is the harmonic mean of precision (Pre) and recall (Sen). For all indicators, the closer the score is to 1, the better the classification result.

[0080] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products, electronic devices. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0081] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0084] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0085] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0086] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

Claims

1. A method for classifying and diagnosing adult diffuse glioma using magnetic resonance imaging, characterized in that: The specific steps are: Step 1: Obtain T2-fluid attenuated inversion recovery (T2-FLAIR) and contrast-enhanced T1-weighted imaging (CE-T1WI) magnetic resonance images of gliomas labeled by doctors based on histological and genomic information; Step 2: Perform data preprocessing on T2-FLAIR and CE-T1WI image data; Step 3: Each pair of preprocessed T2-FLAIR and CE-T1WI images is sent to the decision tree diagnosis model, and the Attention-DenseNet model learns and extracts the features of T2-FLAIR and CE-T1WI images; Step 4: Obtain prediction results based on multiple independently trained Attention-DenseNet models, and use a voting mechanism or weighted average strategy to fuse the outputs of multiple models to generate the final classification decision; Step 5: Send the final classification to the platform and make a final diagnosis based on the classification results.

2. The method for classifying and diagnosing adult diffuse glioma using magnetic resonance imaging according to claim 1, characterized in that: In the step 1, the histological grades labeled on the T2-fluid attenuated inversion recovery (T2-FLAIR) and contrast-enhanced T1-weighted imaging (CE-T1WI) magnetic resonance images of gliomas based on histological and genomic information include: IDH, 1p / 19q, CDKN2A / B, epidermal growth factor receptor (EGFR), and TERT.

3. The method for classification and diagnosis of adult diffuse glioma by nuclear magnetic resonance imaging according to claim 1, characterized in that: The step 2 specifically comprises: 2.

1. Co-register the images to align images of different modalities and reduce positional deviation; 2.

2. Perform offset field correction to eliminate uneven brightness in the image and ensure image uniformity; 2.

3. Then, the image is anisotropically resampled to unify the spatial resolution to facilitate subsequent feature extraction and classification; 2.

4. Then crop the image and extract the target area to reduce the amount of calculation and focus on the lesion area.

4. The method for classifying and diagnosing adult diffuse glioma using magnetic resonance imaging according to claim 3, characterized in that: In step 2.3, The image is adjusted to a uniform size using the resampling method. The expression for image resampling is shown in Formula 1: Among them, I new (x, y) is the pixel value of the image at the new size, I is the original image, W orig ,H orig is the width and height of the original image, W new ,H new The width and height of the new image size.

5. The method for classification and diagnosis of adult diffuse glioma by nuclear magnetic resonance imaging according to claim 1, characterized in that: The Attention-DenseNet model structure includes a first input channel for inputting images of the T2-FLAIR modality, and sequentially connected: a first Dense Block, a first Attention layer, a second Dense Block, a second Attention layer, a third Dense Block, a third Attention layer, a pooling layer, and a linear layer; The second input channel is used to input images of the CE-T1WI modality, which are sequentially connected to the fourth Dense Block, the fourth Attention layer, the fifth Dense Block, the fifth Attention layer, the sixth Dense Block, the sixth Attention layer, the pooling layer, and the linear layer; The first input channel and the second input channel have multiple cross-channel connections between the attention layer and the pooling layer, respectively, for feature fusion; The features output by the pooling layer are connected through a concatenation operation before the linear layer for the final prediction result output.

6. The method for classification and diagnosis of adult diffuse glioma by nuclear magnetic resonance imaging according to claim 5, characterized in that: The expression of the Dense Block is shown in Formula 2: H l+1 =H l +F l (H l ,W l ) (2) Among them, H l represents the input feature map of the lth layer in DenseBlock, H l+1 represents the output feature map of the l+1th layer, F l Represents the nonlinear transformation function of the lth layer, including convolution operation and activation function, W l is the convolution kernel weight of the lth layer.

7. The method for classification and diagnosis of adult diffuse glioma by nuclear magnetic resonance imaging according to claim 5, characterized in that: The Attention-DenseNet model architecture uses a self-attention mechanism to weight the feature map. The self-attention weight calculation formula is: Among them, Q is the query matrix, K is the key matrix, V is the value matrix, K T is the transpose of the key matrix, d k is the dimension of the key matrix.

8. The method for classification and diagnosis of adult diffuse glioma by nuclear magnetic resonance imaging according to claim 5, characterized in that: The Attention-DenseNet model architecture uses the Focal Loss loss function to deal with the sample imbalance problem. The expression of the loss function is: FL(p t )=-a t (1-p t ) γ log(p t ) (4) Among them, p t is the predicted probability of the positive class, α t is the importance coefficient of positive and negative classes, and γ is the parameter for adjusting the weight of easy-to-classify samples.

9. The method for classifying and diagnosing adult diffuse glioma using magnetic resonance imaging according to claim 1, characterized in that: The final classification decision is a voting-based ensemble approach for ensemble Attention-DenseNet predictions from different folds.

Images