Bladder cancer individualized diagnosis and treatment method based on multi-mode magnetic resonance image fusion

Abstract

The invention discloses a bladder cancer individualized diagnosis and treatment method based on multi-modal magnetic resonance image fusion and a related device. The method comprises the following steps: acquiring multi-modal image data of a bladder tumor of a patient; extracting quantitative image features related to staging and grading of the bladder cancer according to the multi-modal image data; dynamic curve characteristics of dynamic contrast enhanced magnetic resonance imaging DCE-MRI are extracted according to the quantitative image characteristics, the dynamic curve characteristics comprise a rising slope, peak arrival time and an area under a curve AUC, and heterogeneity of the dynamic curve characteristics in different areas of the bladder cancer tumor is analyzed to form a DCE-MRI characteristic set; the muscle layer infiltration depth and range of the bladder cancer tumor and the surrounding tissue invasion condition are quantitatively evaluated, and the individualized resection depth and range and the operation method of the transurethral bladder tumor are predicted. According to the method, the multi-modal magnetic resonance images are fused, the quantitative image features and the dynamic curve features are extracted, the muscle layer infiltration depth and the surrounding tissue invasion condition of the bladder cancer tumor are quantitatively evaluated, and an individualized diagnosis and treatment scheme is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing and tumor auxiliary diagnosis and treatment, and in particular to a method and device for individualized diagnosis and treatment of bladder cancer based on multimodal magnetic resonance image fusion, as well as a computing device. Background Art

[0002] Bladder cancer is one of the common malignant tumors of the urinary system. Its diagnosis and treatment decisions depend on information such as the tumor stage, grade, and depth of muscle invasion. Accurate assessment of the above information is crucial for the formulation of individualized treatment plans.

[0003] Existing imaging diagnostic methods, such as CT and MRI, rely primarily on the physician's subjective experience in film reading for evaluation. This is subject to high subjectivity and poor reproducibility, making it difficult to accurately determine the depth and extent of myometrial invasion, thus limiting diagnostic accuracy. Cystoscopy and biopsy are the gold standard for diagnosing bladder cancer, but they are invasive procedures with certain risks and can only sample visible lesions, failing to comprehensively assess the tumor status of the entire bladder. VI-RADS (Vesical Imaging-Reporting and Data System) is used to assess the risk of myometrial invasion in bladder cancer, but this scoring method only provides a crude risk stratification and cannot meet the needs of personalized diagnosis and treatment. Although some studies have attempted to use radiomics methods to extract quantitative features of tumors, the lack of multimodal imaging information and the inadequate characterization of internal tumor heterogeneity make it difficult to accurately predict tumor invasive behavior, thus affecting the development of personalized diagnosis and treatment plans.

[0004] To address the above problems, the present invention integrates multimodal magnetic resonance imaging and extracts quantitative image features and dynamic curve features to quantitatively evaluate the depth of myometrial invasion and surrounding tissue invasion of bladder cancer tumors, so as to provide personalized diagnosis and treatment guidance for bladder cancer. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method and apparatus for individualized diagnosis and treatment of bladder cancer based on multimodal magnetic resonance imaging fusion, as well as a computing device.

[0006] According to one aspect of the present invention, a method for personalized diagnosis and treatment of bladder cancer based on multimodal magnetic resonance imaging fusion is provided, comprising:

[0007] Acquiring multimodal imaging data of a patient's bladder tumor, wherein the multimodal imaging data includes T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI);

[0008] extracting quantitative image features related to bladder cancer staging and grading based on the multimodal image data, wherein the quantitative image features include a full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features;

[0009] Extracting dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features, wherein the dynamic curve features include rising slope, peak time, and area under the curve (AUC), and analyzing the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set;

[0010] Based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score, the depth and range of muscle invasion of the bladder cancer tumor and the invasion of surrounding tissues are quantitatively evaluated, and the individualized resection depth, range and surgical method of transurethral bladder tumors are predicted.

[0011] In an optional manner, the full-volume apparent diffusion coefficient histogram is reconstructed based on a variational framework diffusion tensor;

[0012] In the diffusion-weighted imaging (DWI) data, a high-order diffusion tensor field is obtained by solving a constrained optimization problem, and the constrained optimization condition is:

[0013]

[0014] Where D is the second-order diffusion tensor field; Ω is the tumor area; is the spatial gradient of the tensor field; λ is the regularization parameter; E i (D) is the diffusion coding signal prediction model at the i-th voxel; y i is the measured signal value at the i-th voxel; N is the number of samples.

[0015] In an optional manner, the heterogeneity of the dynamic curve characteristics is obtained by solving a spatiotemporal coupled partial differential equation model, wherein the spatiotemporal coupled partial differential equation model is established according to the division of the tumor region into angiogenesis zone, necrosis zone, and infiltration zone;

[0016] The expression of the space-time coupled partial differential equation model is:

[0017]

[0018] Among them, C v is the contrast agent concentration in the vascular area; C n is the contrast agent concentration in the necrotic area; θ is the cell density in the infiltrated area; D v is the diffusion coefficient; k trans and k epis the permeability parameter; β is the necrosis rate; γ is the invasion driving coefficient; α and δ are the cell spreading and response parameters, respectively; C p is the contrast agent concentration in plasma or blood; C v is the contrast agent concentration in the vascular area; t is time.

[0019] In an optional manner, the muscle layer invasion depth is calculated based on a multi-physics coupling model, wherein the multi-physics coupling model models the bladder wall as a hyperelastic material and tumor growth is considered as a mass transfer process in a porous medium;

[0020] Among them, the coupling equations of the multi-physics field coupling model are:

[0021]

[0022] Where u is the displacement field; σ is the Piola-Kirchhoff stress tensor; c is the tumor growth factor concentration; D c is the diffusion coefficient; v is the velocity field; S c is the source term; W is the strain energy function; F is the deformation gradient; p is the hydrostatic pressure.

[0023] In an optional manner, the predicting of individualized resection depth, extent, and surgical method for transurethral bladder tumors further comprises:

[0024] discretizing the three-dimensional structure of the bladder into a voxel grid, wherein the state space of the voxel grid contains parameters of tumor location, invasion depth, and surrounding tissue type;

[0025] The definition of surgical actions includes superficial resection, deep resection, extended resection, precise dissection and electrocoagulation hemostasis. Each surgical action corresponds to different resection depth and range parameters.

[0026] Define a reward function based on tumor clearance rate, normal tissue preservation rate, and risk of surgical complications;

[0027] Multimodal imaging data, voxel grids, surgical actions, and reward functions are input into the DDPG deep deterministic policy gradient network model to output an optimal surgical action sequence; wherein the optimal surgical action sequence includes individualized resection depth, resection range, and surgical method; the individualized resection depth is the three-dimensional resection boundary based on the tumor base, and the resection range is the resection angle parameter in the sagittal, coronal, and transverse planes; the surgical method includes transurethral bladder tumor resection, laser resection, and corresponding operating parameters.

[0028] In an optional manner, extracting dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features further includes:

[0029] determining diffusion characteristics of the tumor region according to the full-volume apparent diffusion coefficient histogram, and dividing the tumor region into different subregions;

[0030] identifying heterogeneous regions within the tumor region based on the texture features, determining the boundary and shape of the tumor based on the morphological features, and calculating the average signal intensity and standard deviation of the tumor region based on the first-order statistical features;

[0031] performing time series analysis on the DCE-MRI data based on the heterogeneous region, the boundary and shape of the tumor, and the average signal intensity and standard deviation of the tumor region to obtain dynamic curve characteristics of each subregion;

[0032] The heterogeneity of the dynamic curve characteristics of each subregion in different areas of bladder cancer tumors was analyzed by using a spatiotemporal coupled partial differential equation model to form a DCE-MRI feature set.

[0033] In an optional manner, the quantitative evaluation of the depth and extent of muscle invasion of the bladder cancer tumor and the surrounding tissue invasion based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score further includes:

[0034] Parse the text information in the bladder imaging report to obtain the tumor location, morphological description, and surrounding anatomical structure relationship; convert the VI-RADS score into a quantitative risk probability matrix and perform weighted fusion with the imaging omics features to obtain a fusion feature vector;

[0035] The fused feature vector is input as an initial condition into the multi-physics field coupling model, and the mechanical erosion boundaries of the tumor on the bladder wall mucosa, muscularis, and serosa are calculated by solving the coupling equation of the deformation gradient field and the tumor growth factor concentration. A three-dimensional geometric grid is constructed based on the tumor morphological characteristics, and the level set method is used to track the invasion front.

[0036] The probabilistic heat map of potential invasion pathways is predicted by combining the distribution of angiogenic areas in the DCE-MRI feature set. The anatomical coordinates of surrounding tissue invasion are determined by calculating the spatial topological relationship between the heat map gradient field and the bladder anatomy.

[0037] Multi-parameter regression analysis was performed between the quantitative assessment results and the postoperative pathology gold standard to establish a calibration curve. This generated a three-dimensional visualization report and a structured assessment document containing a color-coded map of the myometrial invasion depth, a range probability heat map, and the risk level of surrounding tissues. The structured assessment document included the T stage confidence interval, the millimeter value of the invasion depth, the type of surrounding tissue invasion, and the corresponding VI-RADS score adjustment coefficient.

[0038] In an optional manner, analyzing the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set further includes:

[0039] Based on the anisotropy index of the full-volume apparent diffusion coefficient histogram, the K-means++ clustering algorithm was used to divide the tumor into three sub-regions: high diffusion area, medium diffusion area, and low diffusion area;

[0040] Based on the entropy value of the gray-level co-occurrence matrix in the texture features, the boundaries of the three sub-regions were optimized so that the matching degree between the sub-region division and the microstructural heterogeneity reached more than 85%.

[0041] In each subregion, the DCE-MRI time series was processed using a Savitzky-Golay filter to calculate the slope, time to peak, and AUC values, and to establish a dynamic feature vector space. The non-Gaussianity of the feature distribution was evaluated using kernel density estimation, and a feature probability density map was generated.

[0042] The parameters of the spatiotemporal coupled partial differential equation model of angiogenesis, necrosis and invasion areas were correlated and mapped with the sub-regional eigenvectors;

[0043] The finite element method is used to solve the partial differential equations to obtain the spatiotemporal evolution matrix of the contrast agent concentration in various subregions. The concentration gradient field is extracted as a quantitative indicator of heterogeneity, the anisotropy fraction FA value of the gradient field is calculated, and a correlation curve between the FA value and the malignancy of the tumor is established. The dynamic curve features, texture entropy values, morphological compactness index and first-order statistical features are spliced together to obtain a multidimensional tensor. The principal component features of the multidimensional tensor are extracted using the sparse principal component analysis method, so that the cumulative variance contribution rate exceeds 90% and a DCE-MRI feature set is formed.

[0044] According to another aspect of the present invention, a device for personalized diagnosis and treatment of bladder cancer based on multimodal magnetic resonance imaging fusion is provided, comprising:

[0045] A multimodal imaging data acquisition module is used to acquire multimodal imaging data of a patient's bladder tumor, wherein the multimodal imaging data includes T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI);

[0046] a quantitative image feature extraction module, configured to extract quantitative image features related to bladder cancer staging and grading based on the multimodal image data, wherein the quantitative image features include a full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features;

[0047] A DCE-MRI dynamic curve feature extraction module is used to extract dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features, wherein the dynamic curve features include rising slope, peak time, and area under the curve (AUC), and analyze the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set;

[0048] The muscle invasion assessment and personalized prediction module is used to quantitatively assess the muscle invasion depth, range and surrounding tissue invasion of the bladder cancer tumor based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score, and predict the personalized resection depth, range and surgical method of transurethral bladder tumors.

[0049] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0050] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion.

[0051] According to the solution provided by the present invention, multimodal imaging data of a patient's bladder tumor is obtained, the multimodal imaging data including T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI); quantitative imaging features related to bladder cancer staging and grading are extracted based on the multimodal imaging data, the quantitative imaging features including full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features; dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) are extracted based on the quantitative imaging features, the dynamic curve features including rising slope, peak time, and area under the curve (AUC), and the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor is analyzed to form a DCE-MRI feature set; based on the DCE-MRI feature set, bladder imaging report, and VI-RADS data score, the depth and range of muscle layer invasion and surrounding tissue invasion of the bladder cancer tumor are quantitatively evaluated, and the individualized resection depth, range, and surgical method of transurethral bladder tumors are predicted. The present invention integrates multimodal magnetic resonance imaging and extracts quantitative image features and dynamic curve features to quantitatively evaluate the depth of myometrial invasion of bladder cancer tumors and the invasion of surrounding tissues, realize personalized diagnosis and treatment plans, and significantly improve the diagnostic accuracy and treatment effect of bladder cancer. Specifically, by extracting the full-volume apparent diffusion coefficient histogram, texture features, morphological features and first-order statistical features, the microstructure and heterogeneity of the tumor are more accurately described. By analyzing the dynamic curve features of DCE-MRI and combining it with a spatiotemporal coupled partial differential equation model, the growth and invasion behavior of the tumor can be more accurately evaluated. The bladder wall is modeled as a hyperelastic material through a multi-physics field coupling model, and tumor growth is regarded as a mass transfer process in a porous medium. The growth and invasion process of the tumor is more realistically simulated, thereby more accurately predicting the depth and range of myometrial invasion. By inputting multimodal imaging data, voxel grids, surgical actions and reward functions into the DDPG deep deterministic policy gradient network model to output the optimal surgical action sequence, the accuracy of the surgery is improved. By quantitatively evaluating the depth and extent of myometrial invasion of bladder cancer tumors and the invasion of surrounding tissues, and generating a three-dimensional visual report and structured assessment document including a color-coded map of myometrial invasion depth, a range probability heat map, and the risk level of surrounding tissues, it provides clinicians with intuitive decision support.

[0052] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0054] Figure 1 A schematic diagram showing a process of a personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion according to an embodiment of the present invention is shown;

[0055] Figure 2 A schematic diagram of an MRI image of non-muscle invasive bladder cancer (patient, male, 72 years old) according to an embodiment of the present invention is shown;

[0056] Figure 3 A schematic diagram of an MRI image of non-muscle invasive bladder cancer (patient, female, 56 years old) according to an embodiment of the present invention is shown;

[0057] Figure 4 A schematic diagram of an MRI image of non-muscle invasive bladder cancer (patient, male, 64 years old) according to an embodiment of the present invention is shown;

[0058] Figure 5 A schematic diagram of an MRI image of non-muscle invasive bladder cancer (patient, male, 59 years old) according to an embodiment of the present invention is shown;

[0059] Figure 6 A schematic diagram of delineating regions of interest on T2WI and ADC images according to an embodiment of the present invention is shown;

[0060] Figure 7 A schematic diagram of the framework of a personalized bladder cancer diagnosis and treatment device based on multimodal magnetic resonance imaging fusion according to an embodiment of the present invention is shown;

[0061] Figure 8 A schematic structural diagram of a computing device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0062] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0063] Figure 1 FIG2 shows a flow chart of a method for personalized diagnosis and treatment of bladder cancer based on multimodal magnetic resonance imaging fusion according to an embodiment of the present invention. Specifically, Figure 1 As shown, the following steps are included:

[0064] Step S101 : acquiring multimodal imaging data of a patient's bladder tumor, wherein the multimodal imaging data includes T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI).

[0065] In this embodiment, T2WI provides anatomical structural information, DWI reflects cell density and water molecule diffusion, and DCE-MRI reflects tumor angiogenesis and hemodynamic characteristics, such as Figure 2-Figure 5 As shown in Figure 2, the multimodal imaging data above reveals tumor characteristics from different dimensions, improving the accuracy of assessments of tumor staging, grade, extent of invasion, and treatment response. Furthermore, multimodal imaging improves the detection rate of different types of tumor lesions, especially in early stages or when the lesions are small. For example, DWI is more sensitive to areas with high cell density, while DCE-MRI is more sensitive to areas with rich blood vessels.

[0066] Table 1 shows the imaging data of a 55-year-old male patient who presented with gross hematuria and was suspected of bladder cancer.

[0067] Table 1

[0068]

[0069]

[0070] like Figure 6 As shown, by comprehensively analyzing multimodal imaging data, doctors can more accurately determine the stage, grade, and extent of tumor invasion, thereby formulating personalized treatment plans. For example, if T2WI shows that the tumor has invaded the myometrium, DWI shows a high density of tumor cells, and DCE-MRI shows that the tumor has a rich blood supply, it indicates that the tumor is highly malignant and may require radical cystectomy.

[0071] Step S102 : extracting quantitative image features related to bladder cancer staging and grading based on the multimodal image data, wherein the quantitative image features include a full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features.

[0072] In this example, the full-volume apparent diffusion coefficient histogram reflects cell density and microstructure, texture features reflect internal tumor heterogeneity, morphological features reflect tumor shape and size, and first-order statistical features reflect regional signal intensity distribution. These quantitative features are correlated with bladder cancer staging and grade and are used to predict tumor aggressiveness, metastasis risk, and treatment response to aid clinical decision-making.

[0073] In an optional manner, the full-volume apparent diffusion coefficient histogram is reconstructed based on a variational framework diffusion tensor;

[0074] In the diffusion-weighted imaging (DWI) data, a high-order diffusion tensor field is obtained by solving a constrained optimization problem, and the constrained optimization condition is:

[0075]

[0076] Where D is the second-order diffusion tensor field; Ω is the tumor area; is the spatial gradient of the tensor field; λ is the regularization parameter; E i (D) is the diffusion coding signal prediction model at the i-th voxel; y i is the measured signal value at the i-th voxel; N is the number of samples.

[0077] In this embodiment, the high-order diffusion tensor field can be used to extract more quantitative imaging features (such as the full-volume apparent diffusion coefficient histogram), thereby more comprehensively describing the tumor microstructure and heterogeneity. Constrained optimization is used to ensure the stability of the diffusion tensor field and reduce the effects of noise and artifacts.

[0078] Step S103: extracting dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features, wherein the dynamic curve features include rising slope, peak time, and area under the curve (AUC), and analyzing the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set.

[0079] In this example, extracting dynamic curve features from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) allows for more precise quantification of tumor blood supply and metabolic activity, providing more detailed tumor characteristics. Analyzing the heterogeneity of dynamic curve features across different tumor regions reveals complex structural and functional differences within the tumor, helping to identify distinct tumor subregions (e.g., angiogenesis, necrosis, and invasion).

[0080] In an optional manner, the heterogeneity of the dynamic curve characteristics is obtained by solving a spatiotemporal coupled partial differential equation model, wherein the spatiotemporal coupled partial differential equation model is established according to the division of the tumor region into angiogenesis zone, necrosis zone, and infiltration zone;

[0081] The expression of the space-time coupled partial differential equation model is:

[0082]

[0083] Among them, C v is the contrast agent concentration in the vascular area; C n is the contrast agent concentration in the necrotic area; θ is the cell density in the infiltrated area; D v is the diffusion coefficient; k trans and kep is the permeability parameter; β is the necrosis rate; γ is the invasion driving coefficient; α and δ are the cell spreading and response parameters, respectively; C p is the contrast agent concentration in plasma or blood; C v is the contrast agent concentration in the vascular area; t is time.

[0084] In this embodiment, by distinguishing between angiogenesis areas, necrotic areas, and infiltration areas, the complex microenvironment inside the tumor can be simulated more accurately, capturing the differences in contrast agent absorption and metabolism in different areas. The dynamic curve characteristics of DCE-MRI are linked to the biological processes of the tumor microenvironment. By simulating the diffusion, penetration, and metabolism of contrast agents in different areas, the dynamic curve characteristics are more comprehensively described and the heterogeneity of the tumor is quantified. Not only angiogenesis and necrosis are taken into account, but also the changes in cell density in the infiltration area, so that the model can predict the growth and infiltration behavior of the tumor, thereby helping to assess the aggressiveness of the tumor. By solving a set of partial differential equations, the distribution of contrast agent concentration in time and space is obtained and a visual image is generated, which facilitates doctors to understand the heterogeneous characteristics of the tumor. DCE-MRI scans of bladder cancer patients were performed to obtain dynamic imaging data as shown in Table 2.

[0085] Table 2

[0086]

[0087] The curve for the angiogenic area (Area 1) in Table 2 shows a rapid rise and fall, the curve for the necrotic area (Area 2) is very flat, and the curve for the invasive area (Area 3) shows a slow rise and continuous accumulation. Calculation of the concentration gradient field revealed that the gradient was most pronounced at the tumor edge (invasive zone), indicating the direction of tumor invasion. Finally, a comparison of the model's predictions with the patient's pathological findings revealed that the predicted tumor invasion depth was generally consistent with the pathological findings. This demonstrates that the spatiotemporal coupled partial differential equation model can effectively simulate the heterogeneity of bladder cancer tumors and aid in assessing tumor invasiveness.

[0088] In an optional manner, extracting dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features further includes:

[0089] determining diffusion characteristics of the tumor region according to the full-volume apparent diffusion coefficient histogram, and dividing the tumor region into different subregions;

[0090] identifying heterogeneous regions within the tumor region based on the texture features, determining the boundary and shape of the tumor based on the morphological features, and calculating the average signal intensity and standard deviation of the tumor region based on the first-order statistical features;

[0091] performing time series analysis on the DCE-MRI data based on the heterogeneous region, the boundary and shape of the tumor, and the average signal intensity and standard deviation of the tumor region to obtain dynamic curve characteristics of each subregion;

[0092] The heterogeneity of the dynamic curve characteristics of each subregion in different areas of bladder cancer tumors was analyzed by using a spatiotemporal coupled partial differential equation model to form a DCE-MRI feature set.

[0093] In this example, a multi-physics coupling model and a spatiotemporal coupled partial differential equation model enable more accurate prediction of tumor growth and invasion, thereby reducing uncertainty and lowering surgical risk. The DDPG deep deterministic policy gradient network model outputs the optimal surgical action sequence, further improving surgical efficiency and effectiveness.

[0094] In an optional manner, analyzing the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set further includes:

[0095] Based on the anisotropy index of the full-volume apparent diffusion coefficient histogram, the K-means++ clustering algorithm was used to divide the tumor into three sub-regions: high diffusion area, medium diffusion area, and low diffusion area;

[0096] Based on the entropy value of the gray-level co-occurrence matrix in the texture features, the boundaries of the three sub-regions were optimized so that the matching degree between the sub-region division and the microstructural heterogeneity reached more than 85%.

[0097] In each subregion, the DCE-MRI time series was processed using a Savitzky-Golay filter to calculate the slope, time to peak, and AUC values, and to establish a dynamic feature vector space. The non-Gaussianity of the feature distribution was evaluated using kernel density estimation, and a feature probability density map was generated.

[0098] The parameters of the spatiotemporal coupled partial differential equation model of angiogenesis, necrosis and invasion areas were correlated and mapped with the sub-regional eigenvectors;

[0099] The finite element method is used to solve the partial differential equations to obtain the spatiotemporal evolution matrix of the contrast agent concentration in various subregions. The concentration gradient field is extracted as a quantitative indicator of heterogeneity, the anisotropy fraction FA value of the gradient field is calculated, and a correlation curve between the FA value and the malignancy of the tumor is established. The dynamic curve features, texture entropy values, morphological compactness index and first-order statistical features are spliced together to obtain a multidimensional tensor. The principal component features of the multidimensional tensor are extracted using the sparse principal component analysis method, so that the cumulative variance contribution rate exceeds 90% and a DCE-MRI feature set is formed.

[0100] In this embodiment, by combining the anisotropy index of the full-volume apparent diffusion coefficient histogram (ADC) and the K-means++ clustering algorithm, the tumor was divided into three sub-regions of high, medium, and low diffusion, reflecting the differences in cell density and tissue structure within the tumor, which helps to identify regions of different malignancy. The sub-region boundaries were optimized using texture features (gray-level co-occurrence matrix entropy), further improving the match between the sub-region division and the tumor microstructure. A match of more than 85% ensured the accuracy of subsequent analysis. The non-Gaussianity of the feature distribution was evaluated using the kernel density estimation method to more comprehensively understand the statistical characteristics of the features. The spatiotemporal coupled partial differential equation model of the angiogenesis zone, necrotic zone, and invasive zone was associated with the sub-region feature vector, linking the imaging features with the biological processes of the tumor and improving the biological interpretability of the features. By solving the system of partial differential equations, the spatiotemporal evolution matrix of the sub-region contrast agent concentration was obtained and the anisotropy fraction FA value of the concentration gradient field was calculated (the FA value reflects the uniformity of the contrast agent diffusion within the tumor and is an effective indicator for quantifying tumor heterogeneity). The principal component features of the multidimensional tensor are extracted by the sparse principal component analysis method, which can effectively reduce the data dimension while retaining more than 90% of the variance contribution rate.

[0101] Step S104, based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score, quantitatively evaluate the muscle invasion depth and range of the bladder cancer tumor and the invasion of surrounding tissues, and predict the individualized resection depth, range and surgical method of transurethral bladder tumors.

[0102] In this embodiment, the DCE-MRI feature set, bladder imaging report (tumor macromorphology and anatomical relationship) and VI-RADS score are combined to improve the accuracy of the assessment of the depth and extent of myometrial invasion and surrounding tissue invasion.

[0103] In an optional manner, the muscle layer invasion depth is calculated based on a multi-physics coupling model, wherein the multi-physics coupling model models the bladder wall as a hyperelastic material and tumor growth is considered as a mass transfer process in a porous medium;

[0104] Among them, the coupling equations of the multi-physics field coupling model are:

[0105]

[0106] Where u is the displacement field; σ is the Piola-Kirchhoff stress tensor; c is the tumor growth factor concentration; D c is the diffusion coefficient; v is the velocity field; S c is the source term; W is the strain energy function; F is the deformation gradient; p is the hydrostatic pressure.

[0107] Traditional methods ignore the mechanical effects of tumor growth on the bladder wall. This implementation, however, takes into account the hyperelastic properties of the bladder wall, enabling more accurate simulation of bladder wall deformation and stress distribution during tumor growth. By treating tumor growth as a mass transfer process in a porous medium, it simulates processes such as tumor cell diffusion, migration, and proliferation, more closely resembling the actual tumor growth mechanism. By coupling a set of equations to link tumor growth with the mechanical behavior of the bladder wall, this method achieves multi-physics interaction simulation, thereby improving the accuracy of invasion depth assessment.

[0108] For example, the Mooney-Rivlin model is used to describe the hyperelastic properties of the bladder wall, with C10 = 0.2 MPa and C01 = 0.1 MPa. -9 m 2 / s, tumor cell proliferation rate = 10 -6 The three-dimensional geometric structure of the bladder was discretized into a tetrahedral mesh, with a mesh size of 0.5 mm in the tumor area and a mesh size of 1 mm in the bladder wall area. The initial tumor growth factor concentration in the tumor area was 1. COMSOL Multiphysics was used to solve the coupled equations to simulate the tumor growth process, and the mechanical erosion boundary of the tumor on the bladder wall was calculated. The erosion depth of the mucosal layer was 3 mm, the erosion depth of the muscle layer was 6 mm, and the serosal layer was not eroded. The calculation results were visualized to generate a color-coded map of the muscle layer invasion depth. Doctors use this map to evaluate the tumor invasion depth and formulate individualized surgical plans.

[0109] In this embodiment, the prediction of individualized resection depth, range, and surgical method for transurethral bladder tumors further includes:

[0110] discretizing the three-dimensional structure of the bladder into a voxel grid, wherein the state space of the voxel grid contains parameters of tumor location, invasion depth, and surrounding tissue type;

[0111] The definition of surgical actions includes superficial resection, deep resection, extended resection, precise dissection and electrocoagulation hemostasis. Each surgical action corresponds to different resection depth and range parameters.

[0112] Define a reward function based on tumor clearance rate, normal tissue preservation rate, and risk of surgical complications;

[0113] Multimodal imaging data, voxel grids, surgical actions, and reward functions are input into the DDPG deep deterministic policy gradient network model to output an optimal surgical action sequence; wherein the optimal surgical action sequence includes individualized resection depth, resection range, and surgical method; the individualized resection depth is the three-dimensional resection boundary based on the tumor base, and the resection range is the resection angle parameter in the sagittal, coronal, and transverse planes; the surgical method includes transurethral bladder tumor resection, laser resection, and corresponding operating parameters.

[0114] In an optional manner, the quantitative evaluation of the depth and extent of muscle invasion of the bladder cancer tumor and the surrounding tissue invasion based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score further includes:

[0115] Parse the text information in the bladder imaging report to obtain the tumor location, morphological description, and surrounding anatomical structure relationship; convert the VI-RADS score into a quantitative risk probability matrix and perform weighted fusion with the imaging omics features to obtain a fusion feature vector;

[0116] The fused feature vector is input as an initial condition into the multi-physics field coupling model, and the mechanical erosion boundaries of the tumor on the bladder wall mucosa, muscularis, and serosa are calculated by solving the coupling equation of the deformation gradient field and the tumor growth factor concentration. A three-dimensional geometric grid is constructed based on the tumor morphological characteristics, and the level set method is used to track the invasion front.

[0117] The probabilistic heat map of potential invasion pathways is predicted by combining the distribution of angiogenic areas in the DCE-MRI feature set. The anatomical coordinates of surrounding tissue invasion are determined by calculating the spatial topological relationship between the heat map gradient field and the bladder anatomy.

[0118] Multi-parameter regression analysis was performed between the quantitative assessment results and the postoperative pathology gold standard to establish a calibration curve. This generated a three-dimensional visualization report and a structured assessment document containing a color-coded map of the myometrial invasion depth, a range probability heat map, and the risk level of surrounding tissues. The structured assessment document included the T stage confidence interval, the millimeter value of the invasion depth, the type of surrounding tissue invasion, and the corresponding VI-RADS score adjustment coefficient.

[0119] In this example, for example, a patient's bladder cancer MRI image shows: "A mass approximately 3 cm x 4 cm with irregular morphology is visible on the posterior bladder wall, located near the right ureteral opening." The VI-RADS score is 4. The DCE-MRI feature set shows uneven distribution of angiogenesis within the tumor, with higher vascular density near the muscular layer of the bladder wall. The tumor's location (posterior wall), size (3 cm x 4 cm), morphology (irregular), and relationship to the ureteral opening are extracted from the bladder imaging report. A VI-RADS score of 4 is converted into a myometrial invasion risk probability matrix (assuming a score of 4 corresponds to a 60% probability of myometrial invasion). The extracted radiomics features, the risk probability matrix converted from the VI-RADS score, and the DCE-MRI vascular richness are weighted and fused to form a fused feature vector. This fused feature vector is used as the initial condition and fed into a multiphysics coupling model to calculate the mechanical erosion boundaries of the tumor on each layer of the bladder wall. The calculated results show a 4 mm depth of erosion into the myometrium. A three-dimensional geometric mesh is constructed based on the tumor's shape, and the level set method is used to track the tumor's invasion front within the mesh. The most likely tumor invasion path was predicted by combining the distribution of angiogenic zones revealed by the DCE-MRI feature set. Due to the higher vascular density near the muscularis of the bladder wall, the tumor was predicted to most likely invade deep into the muscularis. The gradient field of the invasion path probability heat map was calculated to determine the most likely direction of tumor invasion. The risk of surrounding tissue invasion was determined by combining the bladder anatomy. The calculated results showed that the tumor was close to the right ureteral opening, posing a high risk of ureteral invasion. The model-predicted myometrial invasion depth (4 mm) and ureteral invasion risk were calibrated against historical data. A visual report was generated, including: a color-coded myometrial invasion depth map showing the depth of tumor invasion into the myometrium; a range probability heat map showing the probability of tumor invasion into periascaptic tissues; and a surrounding tissue risk grading, with the ureter labeled as high risk. A structured assessment document was generated, including: T stage: T2; myometrial invasion depth: 4 mm; surrounding tissue invasion: possible invasion of the right ureteral opening; and VI-RADS score adjustment factor: 0.9.

[0120] According to the solution provided by the present invention, multimodal imaging data of a patient's bladder tumor is obtained, the multimodal imaging data including T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI); quantitative imaging features related to bladder cancer staging and grading are extracted based on the multimodal imaging data, the quantitative imaging features including full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features; dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) are extracted based on the quantitative imaging features, the dynamic curve features including rising slope, peak time, and area under the curve (AUC), and the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor is analyzed to form a DCE-MRI feature set; based on the DCE-MRI feature set, bladder imaging report, and VI-RADS data score, the depth and range of muscle layer invasion and surrounding tissue invasion of the bladder cancer tumor are quantitatively evaluated, and the individualized resection depth, range, and surgical method of transurethral bladder tumors are predicted. The present invention integrates multimodal magnetic resonance imaging and extracts quantitative image features and dynamic curve features to quantitatively evaluate the depth of myometrial invasion of bladder cancer tumors and the invasion of surrounding tissues, realize personalized diagnosis and treatment plans, and significantly improve the diagnostic accuracy and treatment effect of bladder cancer. Specifically, by extracting the full-volume apparent diffusion coefficient histogram, texture features, morphological features and first-order statistical features, the microstructure and heterogeneity of the tumor are more accurately described. By analyzing the dynamic curve features of DCE-MRI and combining it with a spatiotemporal coupled partial differential equation model, the growth and invasion behavior of the tumor can be more accurately evaluated. The bladder wall is modeled as a hyperelastic material through a multi-physics field coupling model, and tumor growth is regarded as a mass transfer process in a porous medium. The growth and invasion process of the tumor is more realistically simulated, thereby more accurately predicting the depth and range of myometrial invasion. By inputting multimodal imaging data, voxel grids, surgical actions and reward functions into the DDPG deep deterministic policy gradient network model to output the optimal surgical action sequence, the accuracy of the surgery is improved. By quantitatively evaluating the depth and extent of myometrial invasion of bladder cancer tumors and the invasion of surrounding tissues, and generating a three-dimensional visual report and structured assessment document including a color-coded map of myometrial invasion depth, a range probability heat map, and the risk level of surrounding tissues, it provides clinicians with intuitive decision support.

[0121] Figure 7 The following is a schematic diagram of a framework of a personalized bladder cancer diagnosis and treatment device based on multimodal magnetic resonance imaging fusion according to an embodiment of the present invention. The personalized bladder cancer diagnosis and treatment device based on multimodal magnetic resonance imaging fusion includes:

[0122] A multimodal imaging data acquisition module 710 is configured to acquire multimodal imaging data of a patient's bladder tumor, wherein the multimodal imaging data includes T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI);

[0123] a quantitative image feature extraction module 720 for extracting quantitative image features related to bladder cancer staging and grading based on the multimodal image data, wherein the quantitative image features include a full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features;

[0124] A DCE-MRI dynamic curve feature extraction module 730 is configured to extract dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features, wherein the dynamic curve features include rising slope, time to peak, and area under the curve (AUC), and analyze the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set;

[0125] The muscle invasion assessment and personalized prediction module 740 is used to quantitatively assess the muscle invasion depth, range and surrounding tissue invasion of the bladder cancer tumor based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score, and predict the personalized resection depth, range and surgical method of transurethral bladder tumors.

[0126] Figure 8 The schematic diagram of the structure of the computing device embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0127] like Figure 8 As shown, the computing device may include: a processor 802 , a communications interface 804 , a memory 806 , and a communication bus 808 .

[0128] Processor 802, communication interface 804, and memory 806 communicate with each other via communication bus 808. Communication interface 804 is used to communicate with other devices, such as client devices or other server network elements. Processor 802 is used to execute program 810, which specifically performs the steps described in the embodiment of the personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion.

[0129] Specifically, the program 810 may include program codes, which include computer operation instructions.

[0130] Processor 802 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a computing device may be processors of the same type, such as one or more CPUs, or processors of different types, such as one or more CPUs and one or more ASICs.

[0131] The memory 806 is used to store the program 810. The memory 806 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0132] According to the solution provided by the present invention, multimodal imaging data of a patient's bladder tumor is obtained, the multimodal imaging data including T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI); quantitative imaging features related to bladder cancer staging and grading are extracted based on the multimodal imaging data, the quantitative imaging features including full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features; dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) are extracted based on the quantitative imaging features, the dynamic curve features including rising slope, peak time, and area under the curve (AUC), and the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor is analyzed to form a DCE-MRI feature set; based on the DCE-MRI feature set, bladder imaging report, and VI-RADS data score, the depth and range of muscle layer invasion and surrounding tissue invasion of the bladder cancer tumor are quantitatively evaluated, and the individualized resection depth, range, and surgical method of transurethral bladder tumors are predicted. The present invention integrates multimodal magnetic resonance imaging and extracts quantitative image features and dynamic curve features to quantitatively evaluate the depth of myometrial invasion of bladder cancer tumors and the invasion of surrounding tissues, realize personalized diagnosis and treatment plans, and significantly improve the diagnostic accuracy and treatment effect of bladder cancer. Specifically, by extracting the full-volume apparent diffusion coefficient histogram, texture features, morphological features and first-order statistical features, the microstructure and heterogeneity of the tumor are more accurately described. By analyzing the dynamic curve features of DCE-MRI and combining it with a spatiotemporal coupled partial differential equation model, the growth and invasion behavior of the tumor can be more accurately evaluated. The bladder wall is modeled as a hyperelastic material through a multi-physics field coupling model, and tumor growth is regarded as a mass transfer process in a porous medium. The growth and invasion process of the tumor is more realistically simulated, thereby more accurately predicting the depth and range of myometrial invasion. By inputting multimodal imaging data, voxel grids, surgical actions and reward functions into the DDPG deep deterministic policy gradient network model to output the optimal surgical action sequence, the accuracy of the surgery is improved. By quantitatively evaluating the depth and extent of myometrial invasion of bladder cancer tumors and the invasion of surrounding tissues, and generating a three-dimensional visual report and structured assessment document including a color-coded map of myometrial invasion depth, a range probability heat map, and the risk level of surrounding tissues, it provides clinicians with intuitive decision support.

[0133] Those skilled in the art will appreciate that modules in the devices of the embodiments may be adaptively modified and deployed in one or more devices different from the embodiments. Modules, units, or components in the embodiments may be combined into a single module, unit, or component, and furthermore, they may be divided into multiple submodules, subunits, or subcomponents. All features disclosed in this specification (including the accompanying claims, abstract, and drawings), as well as all processes or units of any method or device disclosed therein, may be combined in any combination, except where at least some of such features and / or processes or units are mutually exclusive. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that provides the same, equivalent, or similar purpose. Furthermore, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the present invention and to form different embodiments. For example, in the claims below, any of the claimed embodiments may be used in any combination. The present invention may be implemented using hardware comprising a number of different elements and using a suitably programmed computer. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be understood as limiting the order of execution.

Claims

1. A personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion, characterized in that: include: Acquiring multimodal imaging data of a patient's bladder tumor, wherein the multimodal imaging data includes T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI); extracting quantitative image features related to bladder cancer staging and grading based on the multimodal image data, wherein the quantitative image features include a full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features; Extracting dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features, wherein the dynamic curve features include rising slope, peak time, and area under the curve (AUC), and analyzing the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set; Based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score, the depth and range of muscle invasion of the bladder cancer tumor and the invasion of surrounding tissues are quantitatively evaluated, and the individualized resection depth, range and surgical method of transurethral bladder tumors are predicted.

2. The personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion according to claim 1, characterized in that: The full-volume apparent diffusion coefficient histogram is reconstructed based on the variational framework diffusion tensor; In the diffusion-weighted imaging (DWI) data, a high-order diffusion tensor field is obtained by solving a constrained optimization problem, and the constrained optimization condition is: Where D is the second-order diffusion tensor field; Ω is the tumor area; is the spatial gradient of the tensor field; λ is the regularization parameter; E i (D) is the diffusion coding signal prediction model at the i-th voxel; y i is the measured signal value at the i-th voxel; N is the number of samples.

3. The personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion according to claim 1, characterized in that: The heterogeneity of the dynamic curve characteristics is obtained by solving a spatiotemporal coupled partial differential equation model, wherein the spatiotemporal coupled partial differential equation model is established according to the division of the tumor region into angiogenesis zone, necrosis zone and infiltration zone; The expression of the space-time coupled partial differential equation model is: Among them, C v is the contrast agent concentration in the vascular area; C n is the contrast agent concentration in the necrotic area; θ is the cell density in the infiltrated area; D v is the diffusion coefficient; k trans and k ep is the permeability parameter; β is the necrosis rate; γ is the invasion driving coefficient; α and δ are the cell spreading and response parameters, respectively; C p is the contrast agent concentration in plasma or blood; C v is the contrast agent concentration in the vascular area; t is time.

4. The personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion according to claim 1, characterized in that: The muscle layer invasion depth is calculated based on a multi-physics coupling model, in which the bladder wall is modeled as a hyperelastic material and tumor growth is considered as a mass transfer process in a porous medium; Among them, the coupling equations of the multi-physics field coupling model are: Where u is the displacement field; σ is the Piola-Kirchhoff stress tensor; c is the tumor growth factor concentration; D c is the diffusion coefficient; v is the velocity field; S c is the source term; W is the strain energy function; F is the deformation gradient; p is the hydrostatic pressure.

5. The personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion according to claim 1, characterized in that: The prediction of individualized resection depth, extent, and surgical approach for transurethral bladder tumors further includes: discretizing the three-dimensional structure of the bladder into a voxel grid, wherein the state space of the voxel grid contains parameters of tumor location, invasion depth, and surrounding tissue type; The definition of surgical actions includes superficial resection, deep resection, extended resection, precise dissection and electrocoagulation hemostasis. Each surgical action corresponds to different resection depth and range parameters. Define a reward function based on tumor clearance rate, normal tissue preservation rate, and risk of surgical complications; Multimodal imaging data, voxel grids, surgical actions, and reward functions are input into the DDPG deep deterministic policy gradient network model to output an optimal surgical action sequence; wherein the optimal surgical action sequence includes individualized resection depth, resection range, and surgical method; the individualized resection depth is the three-dimensional resection boundary based on the tumor base, and the resection range is the resection angle parameter in the sagittal, coronal, and transverse planes; the surgical method includes transurethral bladder tumor resection, laser resection, and corresponding operating parameters.

6. The personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion according to claim 1, characterized in that: The extracting of dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features further comprises: determining diffusion characteristics of the tumor region according to the full-volume apparent diffusion coefficient histogram, and dividing the tumor region into different subregions; identifying heterogeneous regions within the tumor region based on the texture features, determining the boundary and shape of the tumor based on the morphological features, and calculating the average signal intensity and standard deviation of the tumor region based on the first-order statistical features; performing time series analysis on the DCE-MRI data based on the heterogeneous region, the boundary and shape of the tumor, and the average signal intensity and standard deviation of the tumor region to obtain dynamic curve characteristics of each subregion; The heterogeneity of the dynamic curve characteristics of each subregion in different areas of bladder cancer tumors was analyzed by using a spatiotemporal coupled partial differential equation model to form a DCE-MRI feature set.

7. The personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion according to claim 4, characterized in that: The quantitative evaluation of the depth and extent of muscle invasion and surrounding tissue invasion of the bladder cancer tumor based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score further includes: Parse the text information in the bladder imaging report to obtain the tumor location, morphological description, and surrounding anatomical structure relationship; convert the VI-RADS score into a quantitative risk probability matrix and perform weighted fusion with the imaging omics features to obtain a fusion feature vector; The fused feature vector is input as an initial condition into the multi-physics field coupling model, and the mechanical erosion boundaries of the tumor on the bladder wall mucosa, muscularis, and serosa are calculated by solving the coupling equation of the deformation gradient field and the tumor growth factor concentration. A three-dimensional geometric grid is constructed based on the tumor morphological characteristics, and the level set method is used to track the invasion front. The probabilistic heat map of potential invasion pathways is predicted by combining the distribution of angiogenic areas in the DCE-MRI feature set. The anatomical coordinates of surrounding tissue invasion are determined by calculating the spatial topological relationship between the heat map gradient field and the bladder anatomy. Multi-parameter regression analysis was performed between the quantitative assessment results and the postoperative pathology gold standard to establish a calibration curve. This generated a three-dimensional visualization report and a structured assessment document containing a color-coded map of the myometrial invasion depth, a range probability heat map, and the risk level of surrounding tissues. The structured assessment document included the T stage confidence interval, the millimeter value of the invasion depth, the type of surrounding tissue invasion, and the corresponding VI-RADS score adjustment coefficient.

8. The personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion according to claim 1, characterized in that: Analyzing the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set further includes: Based on the anisotropy index of the full-volume apparent diffusion coefficient histogram, the K-means++ clustering algorithm was used to divide the tumor into three sub-regions: high diffusion area, medium diffusion area, and low diffusion area; Based on the entropy value of the gray-level co-occurrence matrix in the texture features, the boundaries of the three sub-regions were optimized so that the matching degree between the sub-region division and the microstructural heterogeneity reached more than 85%. In each subregion, the DCE-MRI time series was processed using a Savitzky-Golay filter to calculate the slope, time to peak, and AUC values, and to establish a dynamic feature vector space. The non-Gaussianity of the feature distribution was evaluated using kernel density estimation, and a feature probability density map was generated. The parameters of the spatiotemporal coupled partial differential equation model of angiogenesis, necrosis and invasion areas were correlated and mapped with the sub-regional eigenvectors; The finite element method is used to solve the partial differential equations to obtain the spatiotemporal evolution matrix of the contrast agent concentration in various subregions. The concentration gradient field is extracted as a quantitative indicator of heterogeneity, the anisotropy fraction FA value of the gradient field is calculated, and a correlation curve between the FA value and the malignancy of the tumor is established. The dynamic curve features, texture entropy values, morphological compactness index and first-order statistical features are spliced together to obtain a multidimensional tensor. The principal component features of the multidimensional tensor are extracted using the sparse principal component analysis method, so that the cumulative variance contribution rate exceeds 90% and a DCE-MRI feature set is formed.

9. A personalized diagnosis and treatment device for bladder cancer based on multimodal magnetic resonance imaging fusion, characterized in that: include: A multimodal imaging data acquisition module is used to acquire multimodal imaging data of a patient's bladder tumor, wherein the multimodal imaging data includes T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI); a quantitative image feature extraction module, configured to extract quantitative image features related to bladder cancer staging and grading based on the multimodal image data, wherein the quantitative image features include a full-volume apparent diffusion coefficient histogram, texture features, morphological features, and first-order statistical features; A DCE-MRI dynamic curve feature extraction module is used to extract dynamic curve features of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) based on the quantitative image features, wherein the dynamic curve features include rising slope, peak time, and area under the curve (AUC), and analyze the heterogeneity of the dynamic curve features in different regions of the bladder cancer tumor to form a DCE-MRI feature set; The muscle invasion assessment and personalized prediction module is used to quantitatively assess the muscle invasion depth, range and surrounding tissue invasion of the bladder cancer tumor based on the DCE-MRI feature set, bladder imaging report and VI-RADS data score, and predict the personalized resection depth, range and surgical method of transurethral bladder tumors.

10. A computing device comprising: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned personalized diagnosis and treatment method for bladder cancer based on multimodal magnetic resonance imaging fusion.

Images