Information processing method and device for intracranial aneurysm hemodynamics prediction, equipment and medium

Abstract

The application discloses an information processing method and device for intracranial aneurysm hemodynamics prediction, equipment and medium, relates to the field of artificial intelligence technology, and includes the following steps: receiving a three-dimensional medical image containing an intracranial aneurysm, and inputting the three-dimensional medical image into a target hemodynamics prediction model to capture comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image; inputting blood flow boundary conditions into the target hemodynamics prediction model, performing low-dimensional mapping on the blood flow boundary conditions, and obtaining a first low-dimensional feature vector; receiving spatial point coordinates and directional distances corresponding to each sampling point in a CFD model formed by the intracranial aneurysm region, performing low-dimensional mapping on the spatial point coordinates and the directional distances, and obtaining a second low-dimensional feature vector; and performing feature fusion on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector through a feature fusion module to predict hemodynamics parameters. The application realizes end-to-end hemodynamics parameter prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an information processing method and device for intracranial aneurysm hemodynamics prediction, equipment and medium. BACKGROUND

[0002] In the medical field, it is crucial to accurately predict the distribution of physical fields (such as pressure field and velocity field, etc.) in complex geometries, for example, hemodynamic analysis of intracranial aneurysms (IAs) is of decisive significance for evaluating their rupture risk and guiding clinical treatment plan. Early detection and risk assessment of unruptured intracranial aneurysms (UIAs) can help doctors take timely intervention measures, which may save patients' lives.

[0003] Currently, the main means to obtain three-dimensional geometric information of patients' blood vessels and aneurysms is clinical medical imaging technology, such as magnetic resonance imaging (MRI) and computed tomography (CT). These imaging data contain rich geometric morphological information. Computational fluid dynamics (CFD) simulation is currently the main tool for studying the hemodynamic characteristics of aneurysms. Through CFD, key parameters such as pressure and velocity in the aneurysm can be calculated. A large number of clinical studies have shown that the traditional CFD method is complex and time-consuming, requiring tedious steps such as geometry reconstruction, mesh generation, and high computational resource requirements, making it difficult to apply to clinical real-time diagnosis. The physical field prediction method based on deep learning has insufficient generalization ability, low prediction accuracy for new geometric shapes, and needs to be adjusted individually for each case, with limited applicability. The neural operator method based on parameterization or point cloud has limited ability to express complex geometry, and the point cloud preprocessing increases the complexity of the process, which is not compatible with the clinical imaging workflow.

[0004] In summary, how to take three-dimensional medical images as input to realize end-to-end fast and accurate prediction of intracranial aneurysm hemodynamics is a technical problem to be solved in the field. SUMMARY

[0005] Therefore, the purpose of the present application is to provide an information processing method and device for intracranial aneurysm hemodynamics prediction, equipment and medium, which can take three-dimensional medical images as input to realize end-to-end fast and accurate prediction of intracranial aneurysm hemodynamics. The specific scheme is as follows:

[0006] In a first aspect, the application discloses an information processing method for intracranial aneurysm hemodynamics prediction, applied to a computer device, comprising:

[0007] receiving a three-dimensional medical image containing an intracranial aneurysm, and inputting the three-dimensional medical image into a target hemodynamics prediction model, so as to capture comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image by a geometric feature extraction module of the target hemodynamics prediction model; wherein the comprehensive morphological features include local morphological features and global anatomical features;

[0008] receiving blood flow boundary conditions, and inputting the blood flow boundary conditions into the target hemodynamics prediction model, so as to perform low-dimensional mapping on the blood flow boundary conditions by a boundary condition encoding module of the target hemodynamics prediction model, to obtain a first low-dimensional feature vector; wherein the first low-dimensional feature vector is a vector representation of the influence of the blood flow boundary conditions on a physical field, and the physical field is a physical field related to hemodynamics;

[0009] receiving spatial point coordinates and directional distances corresponding to each sampling point in a CFD model formed by the intracranial aneurysm region, so as to perform low-dimensional mapping on the spatial point coordinates and the directional distances by a spatial query point processing module of the target hemodynamics prediction model, to obtain a second low-dimensional feature vector; wherein the directional distance is the minimum Euclidean distance between each sampling point and the wall surface of the CFD model of the intracranial aneurysm, and the second low-dimensional feature vector is a position attribute representation of the sampling point and a relationship between each sampling point and the wall surface of the CFD model of the intracranial aneurysm;

[0010] performing feature fusion on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector by a feature fusion module of the target hemodynamics prediction model, so as to predict hemodynamics parameters of each sampling point based on target fused features.

[0011] Optionally, before the receiving a three-dimensional medical image containing an intracranial aneurysm, the method further comprises:

[0012] performing voxelization processing on an original medical image, so as to convert the original medical image into a binary three-dimensional medical image; wherein the voxel value of the intracranial aneurysm region in the three-dimensional medical image is 1, and the voxel value of the non-intracranial aneurysm region is 0.

[0013] Optionally, the geometric feature extraction module of the target hemodynamics prediction model is a three-dimensional computer vision model;

[0014] Correspondingly, the capturing of the comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image by the geometric feature extraction module of the target hemodynamics prediction model comprises:

[0015] The three-dimensional medical image is divided into blocks by the target blood flow dynamics prediction model to obtain a plurality of three-dimensional voxel blocks, and each three-dimensional voxel block is projected to a preset feature vector space to obtain an initial input sequence;

[0016] The initial input sequence is input into the geometric feature extraction module to perform shallow feature extraction on the initial input sequence based on local window self-attention and shift window mechanism by the geometric feature extraction module to obtain local morphological features including vessel diameter, aneurysm neck width, and cyst morphology;

[0017] The initial input sequence is processed by the geometric feature extraction module to perform deep feature extraction on the initial input sequence after the downsampling processing based on local window self-attention and shift window mechanism to obtain global anatomical features including the position of the intracranial aneurysm in the vascular tree and the spatial relationship between the intracranial aneurysm and the surrounding brain tissue.

[0018] Optionally, the boundary condition encoding module of the target blood flow dynamics prediction model is a multi-layer perceptron network structure.

[0019] Correspondingly, the boundary condition encoding module of the target blood flow dynamics prediction model is a multi-layer perceptron network structure.

[0020] The multi-layer perceptron network structure of the target blood flow dynamics prediction model captures the nonlinear relationship between the inlet mass flow parameter and the outlet pressure parameter to obtain the first low-dimensional feature vector.

[0021] Optionally, before receiving the spatial point coordinates and the directional distance corresponding to each sampling point in the CFD model formed by the intracranial aneurysm region, the method further includes:

[0022] Uniformly sampling the region with a voxel value of 1 in the binary three-dimensional medical image to obtain each sampling point;

[0023] Determining the spatial point coordinates corresponding to each sampling point based on a preset spatial coordinate system;

[0024] Obtaining the shortest distance from each sampling point to the wall surface of the CFD model of the intracranial aneurysm, and determining the directional distance from each sampling point to the wall surface of the CFD model based on the current position of each sampling point and the shortest distance;

[0025] Wherein, if the current position of the sampling point is inside the wall surface of the CFD model, the directional distance is the negative shortest distance, and if the current position of the sampling point is on the wall surface of the CFD model, the directional distance is 0.

[0026] Optionally, the spatial query point processing module of the target blood flow dynamics prediction model is a multi-layer perceptron network structure.

[0027] Correspondingly, the low-dimensional mapping of the spatial point coordinates and the directional distance by the spatial query point processing module of the target blood flow dynamics prediction model comprises:

[0028] The spatial position extraction and symbol mapping processing of the spatial point coordinates and the directional distance by the spatial query point processing module of the target blood flow dynamics prediction model comprises obtaining a second low-dimensional feature vector.

[0029] Optionally, the feature fusion of the comprehensive morphological feature, the first low-dimensional feature vector and the second low-dimensional feature vector by the feature fusion module of the target blood flow dynamics prediction model comprises:

[0030] The feature dimension alignment processing and the feature standardization processing of the comprehensive morphological feature, the first low-dimensional feature vector and the second low-dimensional feature vector comprise obtaining a target comprehensive morphological feature, a target first low-dimensional feature vector and a target second low-dimensional feature vector.

[0031] The feature splicing processing of the target comprehensive morphological feature, the target first low-dimensional feature vector and the target second low-dimensional feature vector by the first feature fusion unit of the target blood flow dynamics prediction model comprises obtaining a first fused feature.

[0032] The attention calculation of the target first low-dimensional feature vector and the target second low-dimensional feature vector by the second feature fusion unit of the target blood flow dynamics prediction model based on the cross-attention mechanism comprises obtaining a second fused feature representing the blood flow response of the intracranial aneurysm.

[0033] The fusion processing of the first fused feature and the second fused feature by the third feature fusion unit of the target blood flow dynamics prediction model through the bilinear transformation comprises obtaining a target fused feature.

[0034] In a second aspect, the present application discloses an information processing device for intracranial aneurysm blood flow dynamics prediction, comprising:

[0035] A feature extraction module is configured to receive a three-dimensional medical image containing an intracranial aneurysm, and input the three-dimensional medical image into a target blood flow dynamics prediction model, so as to capture a comprehensive morphological feature of the intracranial aneurysm from the three-dimensional medical image by a geometric feature extraction module of the target blood flow dynamics prediction model; wherein the comprehensive morphological feature comprises a local morphological feature and a global anatomical feature.

[0036] The first mapping module is configured to receive a blood flow boundary condition and input the blood flow boundary condition into the target blood flow dynamics prediction model, so as to perform low-dimensional mapping on the blood flow boundary condition by a boundary condition encoding module of the target blood flow dynamics prediction model, to obtain a first low-dimensional feature vector; wherein the first low-dimensional feature vector is a vector representation of an influence of the blood flow boundary condition on a physical field, and the physical field is a physical field related to blood flow dynamics;

[0037] The second mapping module is configured to receive spatial point coordinates and directional distances corresponding to each sampling point in a CFD model of an intracranial aneurysm region formation, and perform low-dimensional mapping on the spatial point coordinates and the directional distances by a spatial query point processing module of the target blood flow dynamics prediction model, to obtain a second low-dimensional feature vector; wherein the directional distance is a minimum Euclidean distance between each sampling point and a CFD model wall surface of the intracranial aneurysm, and the second low-dimensional feature vector is a position attribute representation of the sampling point and a relationship between each sampling point and the CFD model wall surface of the intracranial aneurysm.

[0038] The fusion prediction module is configured to perform feature fusion on the comprehensive morphological feature, the first low-dimensional feature vector and the second low-dimensional feature vector by a feature fusion module of the target blood flow dynamics prediction model, to predict a blood flow dynamics parameter of each sampling point based on a target fused feature.

[0039] In a third aspect, the present application discloses an electronic device, comprising:

[0040] A memory is configured to save a computer program;

[0041] A processor is configured to execute the computer program to implement the steps of the information processing method for intracranial aneurysm blood flow dynamics prediction disclosed in the foregoing.

[0042] In a fourth aspect, the present application discloses a computer readable storage medium configured to store a computer program; wherein the computer program is executed by a processor to implement the steps of the information processing method for intracranial aneurysm blood flow dynamics prediction disclosed in the foregoing.

[0043] It can be seen that the application discloses an information processing method for intracranial aneurysm hemodynamics prediction, which is applied to a computer device and comprises the following steps: receiving a three-dimensional medical image containing an intracranial aneurysm and inputting the three-dimensional medical image into a target hemodynamics prediction model, so as to capture comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image by means of a geometric feature extraction module of the target hemodynamics prediction model; wherein the comprehensive morphological features comprise local morphological features and global anatomical features; receiving blood flow boundary conditions and inputting the blood flow boundary conditions into the target hemodynamics prediction model, so as to perform low-dimensional mapping on the blood flow boundary conditions by means of a boundary condition coding module of the target hemodynamics prediction model, so as to obtain a first low-dimensional feature vector; wherein the first low-dimensional feature vector is a vector representation of the influence of the blood flow boundary conditions on a physical field, and the physical field is a physical field related to hemodynamics; receiving spatial point coordinates and directional distances corresponding to each sampling point in a CFD model formed by the intracranial aneurysm region, so as to perform low-dimensional mapping on the spatial point coordinates and the directional distances by means of a spatial query point processing module of the target hemodynamics prediction model, so as to obtain a second low-dimensional feature vector; wherein the directional distance is the minimum Euclidean distance between each sampling point and the wall surface of the CFD model of the intracranial aneurysm, and the second low-dimensional feature vector is a position attribute representation of the sampling point and a relationship between each sampling point and the wall surface of the CFD model of the intracranial aneurysm; and performing feature fusion on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector by means of a feature fusion module of the target hemodynamics prediction model, so as to predict hemodynamics parameters of each sampling point based on target fused features. It can be seen that, by receiving a three-dimensional medical image and directly inputting the three-dimensional medical image into a target hemodynamics prediction model, the geometric reconstruction and grid generation steps of a traditional CFD are bypassed, local morphological features and global anatomical features of an aneurysm are captured in layers by means of a geometric feature extraction module, low-dimensional mapping of blood flow parameters is combined with the boundary condition coding module, and the position and vessel wall boundary relationship of a sampling point is represented by means of the spatial query point processing module, so as to realize end-to-end hemodynamics parameter prediction after feature fusion, effectively solve the problems of complex prediction process flow, insufficient generalization ability and poor compatibility with clinical images, and have the advantages of high prediction efficiency, high precision and adaptation to real-time diagnosis requirements. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, brief introductions will be given to the drawings needed in the embodiments or prior art descriptions. Obviously, the drawings in the following description are only embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of the provided drawings.

[0045] Figure 1An information processing method flowchart for intracranial aneurysm hemodynamics prediction disclosed by the present application;

[0046] Figure 2 A neural operator model execution flowchart and a traditional CFD simulation method flowchart disclosed by the present application;

[0047] Figure 3 An overall architecture schematic diagram of a neural operator model based on a DeepONet framework disclosed by the present application;

[0048] Figure 4 An information processing device structure diagram for intracranial aneurysm hemodynamics prediction disclosed by the present application;

[0049] Figure 5 An electronic device structure diagram disclosed by the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0051] In the medical field, it is crucial to accurately predict the distribution of physical fields (such as pressure field and velocity field, etc.) within complex geometries, for example, hemodynamic analysis of intracranial aneurysms (IAs), which is of decisive significance for evaluating their rupture risk and guiding clinical treatment plans. Early detection and risk assessment of unruptured intracranial aneurysms (UIAs) can help doctors take timely intervention measures, which may save patients' lives.

[0052] Currently, the main means to obtain three-dimensional geometric information of patients' blood vessels and aneurysms is clinical medical imaging technology, such as magnetic resonance imaging (MRI) and computed tomography (CT). These imaging data contain rich geometric morphological information. Computational fluid dynamics (CFD) simulation is currently the main tool for studying the hemodynamic characteristics of aneurysms. Through CFD, key parameters such as pressure and velocity within aneurysms can be calculated.

[0053] Take the traditional CFD simulation method as an example. Its process is usually very complex and time-consuming. The first step is often the accurate reconstruction of a three-dimensional geometric model based on medical image data, which requires specialized software and operators, and for aneurysms with complex morphology and high variability, this process is both time-consuming and prone to errors. After the geometry is reconstructed, a high-quality computational mesh needs to be generated, and the quality of the mesh directly affects the accuracy and stability of the simulation results. However, generating a mesh suitable for complex geometry is itself a technical challenge that requires a lot of time and expertise. Then, boundary conditions are set according to clinical data or assumptions. Finally, the Navier-Stokes equation describing fluid motion is solved using numerical algorithms. The entire solving process requires a large amount of computation and needs to be run on high-performance computing resources for hours or even longer to obtain results.

[0054] The main defects of the traditional CFD method are its low efficiency, high cost and complexity. The long calculation time makes it difficult to apply to clinical rapid diagnosis and real-time decision-making scenarios. The high requirements for professional skills and computing resources also limit its popular application.

[0055] Secondly, some deep learning-based physical field prediction methods have also been proposed. For example, some methods try to directly train neural networks to predict physical fields on specific geometric shapes, or use PINNs (Physics-Informed Neural Networks) to solve physical equations. However, such methods often have insufficient generalization ability. They usually need to be trained for specific geometric shapes, and when encountering new geometric shapes that are not in the training set, the prediction accuracy will decrease significantly. For patient-specific aneurysms with high morphological variability, this means that the model is difficult to apply directly and may require time-consuming data preparation or model fine-tuning for each new case.

[0056] In addition, some neural operator methods based on parameterization or point cloud representation try to input geometric information in the form of parameters or point clouds into neural operator models. The disadvantage of such methods is that for geometric bodies like aneurysms with highly complex morphology and difficult to accurately describe with a small number of parameters, parameter-based methods have limited expressive ability. Although point clouds can represent complex shapes more finely, point cloud data is not the original output form of clinical medical images and requires additional preprocessing steps (extracting point clouds from images), which increases the complexity of the process and may lose some information. Therefore, these methods have inconvenience in interfacing with existing clinical image workflows.

[0057] In summary, existing technical solutions have obvious limitations in quickly and accurately predicting the internal physical fields of complex geometries (such as aneurysm hemodynamics) from medical images, especially in terms of efficiency, generalization ability, and compatibility with clinical data workflows.

[0058] According to a large number of clinical studies above, it has been proved that the traditional CFD method has a complex and time-consuming process, needs cumbersome steps such as geometry reconstruction and mesh generation, requires high computing resources, and is difficult to apply to clinical real-time diagnosis. The physical field prediction method based on deep learning has insufficient generalization ability, low prediction accuracy for new geometry, and needs to be adjusted separately for each case, and the applicability is limited. The neural operator method based on parameterization or point cloud has limited ability to express complex geometry, and the point cloud preprocessing increases the complexity of the process, and the compatibility with the clinical image workflow is poor.

[0059] Therefore, the present application provides an information processing scheme for intracranial aneurysm hemodynamics prediction, which can take three-dimensional medical images as input and realize end-to-end fast and accurate prediction of intracranial aneurysm hemodynamics.

[0060] Referring to Figure 1 The embodiment of the present application discloses an information processing method for intracranial aneurysm hemodynamics prediction, applied to a computer device, comprising:

[0061] Step S11: receiving a three-dimensional medical image containing an intracranial aneurysm, and inputting the three-dimensional medical image into a target hemodynamic prediction model, so as to capture the comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image by the geometric feature extraction module of the target hemodynamic prediction model; wherein the comprehensive morphological features include local morphological features and global anatomical features.

[0062] In this embodiment, before receiving the three-dimensional medical image containing the intracranial aneurysm, it further comprises: performing voxelization processing on the original medical image to convert the original medical image into a binary three-dimensional medical image; wherein the voxel value of the intracranial aneurysm region in the three-dimensional medical image is 1, and the voxel value of the non-intracranial aneurysm region is 0. It can be understood that the original medical image (MRI image or CT image containing the region of interest of the intracranial aneurysm) is binary processed to convert the voxel value of the intracranial aneurysm region to 1 and the voxel value of the non-intracranial aneurysm region to 0 to obtain a binary three-dimensional medical image.

[0063] In this embodiment, the target hemodynamic prediction model is a prediction model constructed based on a deep neural operator (DeepONet framework), wherein the geometric feature extraction module of the target hemodynamic prediction model is a three-dimensional SwinTransformer (Swin-T, image feature extraction network used in geometric branch network) architecture;

[0064] Correspondingly, the comprehensive morphological features of the intracranial aneurysm captured from the three-dimensional medical image by the geometric feature extraction module of the target hemodynamic prediction model comprise:

[0065] The three-dimensional medical image is divided into blocks by the target hemodynamic prediction model to obtain a plurality of three-dimensional voxel blocks, and each three-dimensional voxel block is projected to a preset feature vector space to obtain an initial input sequence; the initial input sequence is input into the geometric feature extraction module, so that the geometric feature extraction module performs shallow feature extraction on the initial input sequence based on local window self-attention and shift window mechanism to obtain local morphological features including vessel diameter, aneurysm neck width, and cyst morphology; the initial input sequence is processed by the geometric feature extraction module to perform deep feature extraction on the initial input sequence after the downsampling processing based on local window self-attention and shift window mechanism to obtain global anatomical features including the position feature of the intracranial aneurysm in the blood vessel tree and the spatial relationship between the intracranial aneurysm and the surrounding brain tissue. It can be understood that the geometric feature extraction module respectively performs three-dimensional image blocking, hierarchical feature representation, three-dimensional window division, and three-dimensional window shift, and the specific execution operations are as follows:

[0066] Three-dimensional image blocking: the input three-dimensional medical image (for example, a CT or MRI image with a size of HxWxD) is divided into a series of non-overlapping three-dimensional voxel blocks (referred to as voxel blocks or 3D patches), for example, each small block has a size of 4x4x4 voxels. The voxel values of each three-dimensional small block are flattened and projected to a high-dimensional feature vector through a linear layer (or a 3D convolution layer) to form a series of three-dimensional patch tokens. These tokens constitute the initial input sequence of the SwinTransformer.

[0067] Hierarchical feature representation: the Swin Transformer follows its hierarchical design, which is stacked by multiple SwinTransformer Blocks, each stage contains multiple Swin Transformer Blocks, and a Patch Merging (image splicing) operation is introduced between stages. Patch Merging specifically refers to a three-dimensional Patch Merging operation at the end of each stage. For example, it will combine 2x2x2 adjacent Patch Tokens into a larger Patch Token, while adjusting the feature dimension through a linear layer. It can gradually reduce the spatial resolution of the feature map (the number of voxel blocks) while increasing the receptive field of each Token, thereby extracting more abstract and global features.

[0068] Three-dimensional window partitioning: In each Swin Transformer Block, the feature map (composed of Patch Tokens) is partitioned into a series of non-overlapping fixed-size three-dimensional windows, e.g., each window has a size of 7x7x7 Patch Tokens. Attention mechanism is only computed within these local three-dimensional windows, which greatly reduces the computational complexity and allows the model to handle high-resolution three-dimensional inputs.

[0069] Three-dimensional shift window mechanism: To enable information exchange between different three-dimensional windows without increasing computational complexity, Swin Transformer introduces a shift window mechanism. In consecutive two Swin Transformer Blocks, one Block adopts the regular three-dimensional window partitioning, while the next Block shifts the feature map by half-window size in three dimensions (x, y, z) simultaneously (e.g., if the window size is MxMxM, then shift M / 2xM / 2xM / 2). After shifting, the three-dimensional window partitioning is performed again. In this way, the Patch Tokens originally in different windows may be partitioned into the same window after shifting, thus realizing cross-window information exchange and enhancing the receptive field and modeling ability of the model.

[0070] Swin Transformer (Swin-T) can adapt to different sizes of input images and demonstrate its superiority in processing different resolution data, mainly due to its hierarchical architecture and core patch merging operation, as well as the fixed window-based self-attention mechanism.

[0071] Specifically, Swin Transformer first divides the original input image into a series of small, non-overlapping image blocks (Patches). These image blocks are then flattened and projected into a fixed-dimensional feature vector (called Patch Token). In this way, no matter how large the original image size is, this operation is only to divide the image into more or less fixed-size Patches. For example, a 224x224 image will produce 56x56 4x4 Patches, while a 448x448 image will produce 112x112 4x4 Patches. The model processes the sequence of Patch Tokens, not the original pixels.

[0072] Further, the network structure of Swin Transformer is designed as multiple stages. Between each stage, it introduces a Patch Merging operation. The Patch Merging operation will merge 2x2x2 (3D) adjacent Patch Tokens into one larger Patch Token. At the same time of merging, a linear layer is used to adjust the feature dimension and effectively down-sample the spatial resolution by 2. In this way, this hierarchical and Patch Merging mechanism enables Swin Transformer to gradually build feature representations with different receptive fields like convolutional neural networks. For input images of different sizes, the Patch Merging operation will merge according to the number of initial patches for a corresponding number of times until the final feature map size is reached. This means that no matter how large the input image is, the model can gradually extract multi-scale features in this way.

[0073] Another core of Swin Transformer is its window self-attention. The self-attention computation is only performed within a local, fixed-size window (local window) (e.g., each window contains 7x7 Patch Tokens). It is important to note that the size of the local window is fixed and does not change with the size of the input image. This means that no matter whether the input image is 224x224 or 1024x1024, the amount of self-attention computation within each window is constant. When the input image size changes, the only thing that changes is the total number of local windows, not the computation within each local window. This makes Swin Transformer more computationally efficient and enables it to flexibly handle input images of different sizes because the core attention operation is local and fixed in size.

[0074] The hierarchical design of the Swin Transformer enables it to capture information of different scales, including shallow feature information extraction and deep feature information extraction. The shallow information mainly extracts local and fine detail information in the three-dimensional image, including: vascular morphology information (diameter, wall thickness, branch angle, and micro-irregularities such as early aneurysm wall protrusions), lesion local features (texture, density, and edge sharpness of early micro-lesions), tissue texture (local brain tissue gray scale features and texture details), and micro-structures (local orientation of brain fine blood vessel network and nerve bundle). The deep information is extracted from the global context information and high-level semantic features as the resolution of the feature map decreases with the Patch Merging. Each PatchToken represents an enlarged receptive field, so deep information mainly extracts global context information and high-level semantic features. Specifically, it includes: overall structure information (anatomical structure of the whole brain, symmetry of left and right hemispheres, size and shape of the ventricular system), global relationship between lesions and surrounding tissues (location of aneurysm in the entire vascular tree (e.g., which part of the Willis circle), its compression effect on the surrounding brain tissue, and distance from important neural structures), multi-lesion correlation information (if there are multiple aneurysms or other lesions, their spatial distribution and mutual relationship), and pathological patterns (identify more complex pathological patterns such as diffuse lesions and overall distribution of multiple infarction lesions).

[0075] Step S12: receiving blood flow boundary conditions and inputting the blood flow boundary conditions into the target blood flow hemodynamics prediction model, so as to perform low-dimensional mapping on the blood flow boundary conditions by a boundary condition encoding module of the target blood flow hemodynamics prediction model to obtain a first low-dimensional feature vector; wherein the first low-dimensional feature vector is a vector representation of the influence of the blood flow boundary conditions on a physical field, and the physical field is a physical field related to blood flow hemodynamics.

[0076] In this embodiment, the boundary condition encoding module of the target blood flow hemodynamics prediction model is a multilayer perceptron network structure; accordingly, the low-dimensional mapping of the blood flow boundary condition by the boundary condition encoding module of the target blood flow hemodynamics prediction model to obtain the first low-dimensional feature vector includes: capturing the nonlinear relationship between the inlet mass flow parameter and the outlet pressure parameter by the multilayer perceptron network structure of the target blood flow hemodynamics prediction model to obtain the first low-dimensional feature vector. It can be understood that the blood flow boundary condition is received, wherein the blood flow boundary condition specifically includes: inlet mass flow, outlet pressure, etc., wherein the inlet mass flow determines the overall amplitude of the blood flow velocity field and affects the formation of intratumoral vortex, so the inlet mass flow can be measured by phase-contrast MRI (PC-MRI) to measure the real-time flow of the carrying artery (such as the internal carotid artery), or the standard flow range based on the healthy population is determined, and no specific limitation is made. The outlet pressure constrains the pressure gradient of the blood flow and affects the intratumoral pressure distribution, so it can be determined by the physiological pressure of the distal blood vessel (such as the diastolic pressure of the external carotid artery and the basilar artery, about 60-80 mmHg), or by the central pressure waveform (systolic / diastolic pressure curve) measured clinically, and no specific limitation is made. The above blood flow boundary condition is the information that defines the physical problem "what", such as temperature distribution, fluid velocity, displacement, etc. on a certain boundary. They are functionally inputted and describe the behavior of the physical field on the boundary. In this way, by receiving the above blood flow boundary condition, and then using MLP (Multilayer Perceptron) to map the blood flow boundary condition to a low-dimensional space, the essence of how the blood flow boundary condition affects the entire physical field solution is captured, that is, the first low-dimensional feature vector is obtained.

[0077] Step S13: receiving the spatial point coordinates and the directional distance corresponding to each sampling point in the CFD model of the intracranial aneurysm region formed, and performing low-dimensional mapping on the spatial point coordinates and the directional distance by the spatial query point processing module of the target blood flow hemodynamics prediction model to obtain a second low-dimensional feature vector; wherein the directional distance is the minimum Euclidean distance between each sampling point and the wall surface of the CFD model of the intracranial aneurysm, and the second low-dimensional feature vector is the position attribute representation of the sampling point and the relationship between each sampling point and the wall surface of the CFD model of the intracranial aneurysm.

[0078] In the embodiment, before the spatial point coordinates corresponding to each sampling point in the CFD model of the intracranial aneurysm region formed by the receiving and the directional distance, further comprising: uniformly sampling the region with a voxel value of 1 in the binary three-dimensional medical image to obtain each sampling point; determining the spatial point coordinates corresponding to each sampling point based on a preset spatial coordinate system; obtaining the shortest distance from each sampling point to the wall surface of the CFD model of the intracranial aneurysm, and determining the directional distance from each sampling point to the wall surface of the CFD model based on the current position of each sampling point and the shortest distance; wherein the current position is the sampling point inside the wall surface of the CFD model, then the directional distance is the negative shortest distance, and the current position is the sampling point on the wall surface of the CFD model, then the directional distance is 0. It can be understood that the specific spatial point coordinates (x, y, z) of which the user or the system specifies the physical field parameter to be predicted are obtained, wherein the preset spatial coordinate system is a Cartesian coordinate system, and the sampling points are uniformly sampled in the intracranial aneurysm region (aneurysm and surrounding blood vessel region) according to a fixed interval, and then the signed distance function (SDF, Signed Distance Function, indicating the distance from a spatial point to a geometric boundary) value of the sampling point relative to the wall surface of the geometric body is obtained to obtain the directional distance, and the SDF value is responsible for processing these spatial position information.

[0079] In the embodiment, the spatial query point processing module of the target blood flow dynamics prediction model is a multi-layer perceptron network structure; correspondingly, the spatial point coordinates and the directional distance are mapped to a low-dimensional space by the spatial query point processing module of the target blood flow dynamics prediction model to obtain a second low-dimensional feature vector, comprising: the spatial point coordinates and the directional distance are subjected to spatial position extraction and sign mapping processing by the spatial query point processing module of the target blood flow dynamics prediction model to obtain a second low-dimensional feature vector. It can be understood that the spatial point coordinates and the SDF value are information defining "where" the physical field is, i.e. the spatial position of which the solution is wanted, so that by receiving the spatial point coordinates and the SDF value, the spatial query point processing module constructed by the MLP learns a set of spatial basis functions and maps them to another low-dimensional space to obtain a second low-dimensional vector, i.e. an abstract representation of the spatial query point position attribute, and the spatial query point processing module captures the position of the spatial point in the physical domain, its relative relationship with the geometric boundary and other information.

[0080] Step S14: the feature fusion module of the target blood flow dynamics prediction model performs feature fusion on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector to predict the blood flow dynamics parameters of each sampling point based on the target fused features.

[0081] In this embodiment, the comprehensive morphological features, the first low-dimensional feature vector, and the second low-dimensional feature vector are subjected to feature dimension alignment processing and feature standardization processing to obtain target comprehensive morphological features, target first low-dimensional feature vectors, and target second low-dimensional feature vectors; the first feature fusion unit of the target hemodynamic prediction model is used to perform feature splicing processing on the target comprehensive morphological features, the target first low-dimensional feature vectors, and the target second low-dimensional feature vectors to obtain first fused features; the second feature fusion unit of the target hemodynamic prediction model is used to perform attention calculation on the target first low-dimensional feature vectors and the target second low-dimensional feature vectors based on a cross-attention mechanism to obtain second fused features representing blood flow responses of intracranial aneurysms; and the third feature fusion unit of the target hemodynamic prediction model is used to perform fusion processing on the first fused features and the second fused features through bilinear transformation to obtain target fused features. It can be understood that, since different features are extracted by different modules, there may be a situation that the dimensions do not correspond to each other. Therefore, before feature fusion, the features are subjected to standardization and dimension alignment processing to obtain target comprehensive morphological features, target first low-dimensional feature vectors, and target second low-dimensional feature vectors, respectively, and then enter a primary fusion step of feature splicing, which is as follows:

[0082] Feature splicing: 32-dimensional target comprehensive morphological features, 32-dimensional target first low-dimensional feature vectors, and 32-dimensional target second low-dimensional feature vectors are spliced by channels to form 96-dimensional fused features; then the features are compressed into 96-dimensional vectors through global average pooling; two layers of MLP (96-dimensional to 16-dimensional to 96-dimensional) are used to generate channel weights, and importance coefficients are assigned to each channel of the spliced features (for example, the aneurysm neck geometry feature channel weight is increased by 2.3 times).

[0083] After the primary fusion, a middle-level fusion step of spatial-boundary interaction modeling is entered, which is as follows:

[0084] The SDF value in the spatial feature (target second low-dimensional feature vector) is taken as Query, and the boundary condition feature (target first low-dimensional feature vector) is taken as Key / Value, to calculate the dependence of the boundary distance on the blood flow condition (for example, the point with an SDF value less than 0.5 mm is strengthened in the pressure feature weight); and 32-dimensional interaction features (second fused features) are output, representing blood flow responses near the geometric boundary.

[0085] After the middle-level fusion step, a high-level fusion step of geometric-physical joint encoding is entered, which is as follows:

[0086] The primary fusion features (96-dimensional) and the middle-level interaction features (32-dimensional) are input.

[0087] The 48-dimensional joint feature (target fused feature) is generated by a bilinear transformation, in which the first 24 dimensions encode the constraint of geometric morphology on physical field (e.g. the tumor neck stenosis leads to the increase of flow velocity), and the last 24 dimensions encode the reaction of physical field on geometric boundary (e.g. the deformation trend of tumor wall in high pressure area).

[0088] In this embodiment, the MLP processing flow after fusion is that the 48-dimensional joint feature passes through two layers of MLP (48-dimensional to 64-dimensional to 32-dimensional) to capture the nonlinear coupling information of pressure-velocity-geometry, using LeakyReLU activation, in which, in the first layer of MLP, the tumor neck width in the geometric feature is multiplied by the flow in the boundary condition at the element level to generate the flow velocity amplification factor at the stenosis. Then a learnable linear scaling matrix is introduced to map the 32-dimensional feature to the dimensional space of the physical field parameters, outputting the physical field parameters related to hemodynamics, including the pressure field, the velocity field component and the wall shear stress, wherein the velocity field: outputs a 3-dimensional vector (u, v, w) with the unit of m / s; the pressure field: outputs a scalar with the unit of Pa; the wall shear stress: outputs a scalar with the unit of Pa, and the blood viscosity parameter is embedded through the bias term of the MLP.

[0089] The velocity field is the flow velocity vector distribution of blood at each spatial point in the blood vessel, and its clinical significance is that the high flow velocity area (such as the tumor neck entrance) may intensify the shear force on the blood vessel wall, which is related to the risk of aneurysm rupture; the pressure field is the fluid pressure distribution at each spatial point in the blood vessel, and its clinical significance is that the pressure fluctuation in the tumor may lead to the fatigue of the tumor wall. The wall shear stress field is the tangential force distribution generated by blood flow on the blood vessel wall, and its clinical significance is that the low WSS area (such as the bottom of the tumor) is confirmed to be related to the growth of aneurysm.

[0090] As can be seen, the model directly takes three-dimensional medical images as input, and can directly and one-step predict the internal physical field from the three-dimensional image of complex geometry, completely bypassing the extremely time-consuming and complex intermediate steps such as three-dimensional geometric reconstruction, mesh generation and traditional numerical solution (such as CFD solution) which are indispensable in traditional methods. This “end-to-end” mode greatly simplifies the process and is the key to efficient prediction.

[0091] As shown in Figure 2 , the model is trained by using the training set, and the trained model is used to predict the physical field parameters of the test set. Figure 2The upper half is an information processing method for predicting intracranial aneurysm hemodynamics by CFD method to obtain real hemodynamic labels of three-dimensional images, and the construction and preprocessing of the data set: starting from two representative basic aneurysm geometric models in the real intracranial aneurysm morphology database, in order to construct a diversified training and test data set, a plurality of parameterized continuous deformation operations are performed on the two basic geometric models. These deformations include "deflation" and "inflation" operations simulating the aneurysm growth or atrophy process, a total of 20 different deformations are performed. Through these operations, a total of 20 synthetic aneurysm geometric models with significantly different morphological characteristics are generated. These models show rich diversity in both parent vessel morphology and aneurysm morphology (especially aspect ratio), and can well represent the morphological variability of aneurysms encountered in clinical practice.

[0092] High-quality computational mesh partitioning is performed on the 21 generated aneurysm geometric models to ensure that the mesh quality meets the requirements of CFD simulation. Specifically, the minimum mesh size is set to 0.15 mm, and 10 layers of prismatic layer mesh are generated near the vessel wall to accurately capture the boundary layer flow characteristics. The solver for incompressible flow in the open-source computational fluid dynamics software is used to solve the three-dimensional steady-state Navier-Stokes equation for each aneurysm geometric model. The fluid medium properties for simulation are set to simulate blood characteristics: density of 1050 kg / m³, dynamic viscosity of .

[0093] Boundary conditions are set: a constant mass flow boundary condition is set at the inlet of the blood vessel. In order to simulate different physiological flow conditions, 6 different mass flow values are selected, ranging from 0.001 kg / s to 0.004 kg / s. These flow values are selected to ensure that the flow is in a laminar state, which is consistent with the blood flow characteristics of most intracranial aneurysms. The outlet is set as a zero pressure boundary condition.

[0094] CDF solution process: transient solver is used for simulation, virtual time step is set to , total simulation time is set to 1 second, a total of iterations are performed to ensure that the solution process reaches steady-state convergence. The flow field results (including pressure and velocity vector values at all grid points in the calculation domain) at 1 second are extracted as the real value label data for model training and testing of the present invention.

[0095] As Figure 2As shown in the lower half, the generated aneurysm geometric model is voxelized using medical image processing and visualization software. This step converts the continuous geometric model into a binary image on a discrete three-dimensional voxel grid (i.e., the voxel values within the aneurysm region are 1, and the outside is 0) and generates a high-resolution three-dimensional binary image. These three-dimensional binary images are directly used as the input of the geometric branch (Geo-Branch, geometric feature extraction module) of the neural operator model proposed in the present application. This processing method simulates the process of extracting the aneurysm region of interest (ROI) from clinical MRI / CT scans and performing simple binary processing, and is a key input data form for end-to-end prediction.

[0096] Training set: 40 out of 42 geometric shapes are selected and combined with 4 out of 6 inlet flow rates (0.001, 0.002, 0.003, 0.004 kg / s), resulting in 40x4=160 training samples.

[0097] Test set: The remaining 2 geometric shapes are combined with the remaining 2 inlet flow rates (0.0015, 0.0035 kg / s), resulting in 2x2=4 test samples. This division ensures that the geometric shapes and inlet flow rates in the test set do not appear in the training set, so that the generalization ability of the model can be strictly evaluated.

[0098] Construction and training of the neural operator model (target blood flow hemodynamics prediction model):

[0099] The model architecture is constructed according to the overall functional block diagram shown in Figure 2 The geometric branch (Geo-Branch) uses a three-dimensional Swin Transformer architecture to process the input three-dimensional binary image. The boundary condition branch (BC-Branch, boundary condition encoding module) and the trunk (Trunk, spatial query point processing module) both use MLP networks. The specific network layer number, neuron number, patch size, window size, embedding dimension, and other hyperparameters of the model are shown in Table 1:

[0100] Table 1 Hyperparameter setting table

[0101]

[0102] Loss function definition: the sum of the squared errors (SSE) of the pressure and three velocity components (u, v, w) is used as the total loss function for model training . Where m is the number of training samples. In order to balance the magnitude difference of different physical quantities, different weight coefficients are set for pressure and velocity, where the pressure weight coefficient , the velocity weight coefficient ( , , The following are respectively represented: .

[0103] Sum of Squared Errors (SSE):

[0104] ;

[0105] in, This represents the total number of sampling points within a single category. This represents the true value of the pressure or velocity component. This represents the predicted value of the pressure or velocity component.

[0106] average relative Standard (Mean Relative L2 Norm):

[0107] ;

[0108] in, This indicates the number of categories of hemodynamic parameters, such as different parameter types like flow velocity and pressure, representing the total number of parameter types for which errors need to be calculated; This represents the parameter category index, used to traverse different hemodynamic parameters; Indicates the first Among the hemodynamic parameters, the first one is... The true value of each sampling point This indicates that the target hemodynamic prediction model is effective for the first... Type of parameter, first The predicted value output from each sampling point.

[0109] MNAE Mean Normalized Absolute Error:

[0110] ;

[0111] in, Indicates the first The minimum value of a certain hemodynamic parameter across all sampling points. Indicates the first The maximum value of the hemodynamic parameter at all sampling points. Indicates the first Among the hemodynamic parameters, the first one is... The true value of each sampling point; Indicates the first Among the hemodynamic parameters, the first one is... Predicted values ​​for each sampling point.

[0112] Total loss function:

[0113] ;

[0114] Optimization algorithm: The Adam optimizer is used to iteratively update the model parameters.

[0115] Learning rate setting: Set a fixed learning rate throughout the training process. .

[0116] Training process execution: The model was trained on a computing platform consisting of four high-performance A100-80G GPUs for a total of 40,000 epochs.

[0117] Model evaluation and results analysis:

[0118] Evaluation index selection: The sum of squared errors (SSE), mean relative L2 standard, and mean normalized absolute error (MNAE) are used as the main quantitative evaluation indexes to calculate the errors between the pressure and velocity components (u, v, w) predicted by the model and the actual CFD values.

[0119] Comparative Validation of Geometric Representation Methods: To verify the effectiveness of Image-Based Representation (IBR), a comparative model was built and trained using PointNet++ to extract Point-Based Representation (PBR) based representations. The loss curves of the two models during training and their performance on the test set were compared. The results, along with the MNAE metric, show that the IBR model slightly outperforms the PBR model in training convergence smoothness and stress prediction accuracy. These comparative results strongly support the superiority of the image-based and Swing Transformer-based geometric feature extraction scheme employed in this invention.

[0120] Comparison with CFD results: The pressure fields predicted by the trained neural operator model (as an exemplary implementation) on the training and test sets are compared with the corresponding CFD simulation results to qualitatively evaluate the model's prediction accuracy. Simultaneously, the sum of pressure and velocity (MNAE) on the test set is calculated to quantitatively evaluate the model's prediction accuracy and generalization ability on unseen data.

[0121] Computational efficiency evaluation: The inference time of the neural operator model on an A100-80G GPU or CPU was recorded and compared with the time required for traditional CFD simulation, highlighting the huge advantage of the present invention in terms of computational speed (in the implementation of the present invention using information processing for intracranial aneurysm hemodynamic prediction as an example, the inference time is about 0.3 seconds, which is much faster than CFD).

[0122] Figure 3 This is a schematic diagram of the overall architecture of the neural operator model based on the DeepONet framework proposed in this invention. Figure 3 As shown, this model is built on a stacked DeepONet framework and mainly consists of the following modules: a Geo-Branch network, responsible for receiving 3D image input of complex geometries; a Boundary Condition Branch Network (BC-Branch), responsible for receiving boundary condition information affecting the distribution of the physical field; a Trunk network, responsible for receiving the spatial point coordinates (xyz) and signed distance function (SDF) used to query the physical field parameters; and a Bypass Linear Scaler module, used to scale the final prediction output according to the boundary conditions. This model can achieve end-to-end prediction from 3D image input to physical field (e.g., pressure and velocity) output, without intermediate geometric reconstruction, mesh generation, and traditional numerical solution steps. The final output of the model is the physical field parameters corresponding to the query point under different boundary conditions and geometries.

[0123] Therefore, by directly inputting 3D medical images into the target hemodynamic prediction model, bypassing the geometric reconstruction and mesh generation steps of traditional CFD, the model utilizes a geometric feature extraction module to capture the local morphology and global anatomical features of the aneurysm in layers. Combined with a boundary condition encoding module for low-dimensional mapping of blood flow parameters and a spatial query point processing module for characterizing the relationship between the sampling point location and the vessel wall boundary, end-to-end hemodynamic parameter prediction is achieved after feature fusion. This effectively solves the problems of complex prediction process, insufficient generalization ability, and poor compatibility with clinical images. It has the advantages of high prediction efficiency, high accuracy, and adaptability to the needs of real-time clinical diagnosis.

[0124] Reference Figure 4 As shown, the present invention also discloses an information processing device for hemodynamic prediction of intracranial aneurysms, comprising:

[0125] The feature extraction module 11 is used to receive a three-dimensional medical image containing an intracranial aneurysm and input the three-dimensional medical image into a target hemodynamic prediction model so that the geometric feature extraction module of the target hemodynamic prediction model can capture the comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image; wherein, the comprehensive morphological features include local morphological features and global anatomical features;

[0126] The first mapping module 12 is configured to receive a blood flow boundary condition and input the blood flow boundary condition into the target blood flow dynamics prediction model, so as to perform low-dimensional mapping on the blood flow boundary condition by a boundary condition encoding module of the target blood flow dynamics prediction model, to obtain a first low-dimensional feature vector; wherein the first low-dimensional feature vector is a vector representation of an influence of the blood flow boundary condition on a physical field, and the physical field is a physical field related to blood flow dynamics;

[0127] The second mapping module 13 is configured to receive spatial point coordinates and directional distances corresponding to each sampling point in a CFD model of an intracranial aneurysm region formation, and perform low-dimensional mapping on the spatial point coordinates and the directional distances by a spatial query point processing module of the target blood flow dynamics prediction model, to obtain a second low-dimensional feature vector; wherein the directional distance is a minimum Euclidean distance between each sampling point and a CFD model wall surface of the intracranial aneurysm, and the second low-dimensional feature vector is a position attribute representation of the sampling point and a relationship between each sampling point and a CFD model wall surface boundary of the intracranial aneurysm;

[0128] The fusion prediction module 14 is configured to perform feature fusion on the comprehensive morphological feature, the first low-dimensional feature vector and the second low-dimensional feature vector by a feature fusion module of the target blood flow dynamics prediction model, to predict a blood flow dynamics parameter of each sampling point based on a target fused feature.

[0129] Therefore, by receiving a three-dimensional medical image to directly input a target blood flow dynamics prediction model, bypassing a geometric reconstruction and grid generation step of a traditional CFD, using a geometric feature extraction module to hierarchically capture local morphological and global anatomical features of an aneurysm, combining a boundary condition encoding module to perform low-dimensional mapping on blood flow parameters and a spatial query point processing module to represent a relationship between a position of a sampling point and a blood vessel wall boundary, and performing feature fusion to realize end-to-end blood flow dynamics parameter prediction, the problems of complex prediction process, insufficient generalization ability and poor compatibility with clinical images are effectively solved, and the method has the advantages of high prediction efficiency, high precision and adaptation to real-time diagnosis requirements.

[0130] Further, the embodiment of the present application also discloses an electronic device, Figure 5 is a structural diagram of an electronic device 20 according to an exemplary embodiment, and the content in the figure cannot be considered as any limitation on the use range of the present application.

[0131] Figure 5A structural schematic diagram of an electronic device 20 is provided in the embodiments of the present application. The electronic device 20 can specifically include at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25 and a communication bus 26. The memory 22 is configured to store a computer program, and the processor 21 is configured to load and execute the computer program to implement the related steps in the information processing method for intracranial aneurysm hemodynamics prediction disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in the embodiments of the present application can be specifically an electronic computer.

[0132] In the embodiments of the present application, the power supply 23 is configured to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 is capable of creating a data transmission channel between the electronic device 20 and external devices, and the communication protocol followed by the communication interface 24 can be any communication protocol applicable to the technical solutions of the present application, which is not specifically limited herein; the input / output interface 25 is configured to obtain external input data or output data to the outside, and the specific interface type can be selected according to the specific application needs, which is not specifically limited herein.

[0133] The processor 21 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array) and a PLA (Programmable Logic Array). The processor 21 can also include a main processor and a coprocessor. The main processor is a processor for processing data in a wake-up state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 21 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 can further include an AI (Artificial Intelligence) processor for processing machine learning-related computing operations.

[0134] In addition, the memory 22 as a carrier for resource storage can be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage mode can be temporary storage or permanent storage.

[0135] The operating system 221 is used to manage and control each hardware device on the electronic device 20 and the computer program 222, so as to realize the operation and processing of the processor 21 on the mass data 223 in the memory 22, which can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program capable of completing the information processing method for intracranial aneurysm hemodynamics prediction executed by the electronic device 20 disclosed in any one of the foregoing embodiments, the computer program 222 can further include a computer program capable of completing other specific work. In addition to the data received by the electronic device from the external device, the data 223 can also include the data collected by the self input and output interface 25, etc.

[0136] Further, the application also discloses a computer readable storage medium for storing a computer program; wherein the computer program is executed by a processor to realize the information processing method for intracranial aneurysm hemodynamics prediction disclosed in the foregoing. The specific steps of the method can refer to the corresponding contents disclosed in the foregoing embodiments, which will not be described here.

[0137] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can refer to the method part.

[0138] Those skilled in the art will further appreciate that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or any combination thereof. To clearly illustrate the interchangeability of hardware and software, various components have been described above generally in terms of their functionality, without limitation. The handwiring and software implementations of the examples described herein could be accomplished using any number of microprocessors, microcontrollers, programmable consumption logic devices, application-specific integrated circuits, or general-purpose computers with interconnecting circuits that either run software programs or use opencircuit or other hardware components that are designed to perform the functions described herein. The embodiments described herein can be implemented along with software modules, and the software modules can be stored on any of a variety of non-transitory machine-readable media. A non-transitory machine-readable medium includes any medium that participates in providing instructions to a processor for execution. Such a medium can take many forms, including but not limited to, non-volatile media, volatile media, and transmission media. Non-volatile media includes, for example, optical or magnetic disks and other persistent memory. Volatile media includes dynamic memories, and physical registers. Transmission media includes coaxial cables, copper wires and fiber optic cables, including wires that comprise bus conductors. Transmission media also can also take the form of acoustic or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer-readable media include, for example, a floppy disk, a flexible disk, hard disk, solid-state drive, magnetic tape, or any other magnetic data storage medium, a Compact Disc - Read Only Memory (CD-ROM), any other optical medium, punch cards, paper tape, any other physical medium with patterns of holes, a RAM, a programmable ROM (PROM), an erasable PROM (EPROM), a FLASH-EPROM, any other memory chip or cartridge, a carrier wave, a

[0139] Finally, it should also be noted that, in the present text, relational terms such as first and second and the like can only be used to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between such entities or operations. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by an occurrence of the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0140] The above has carried on the detailed introduction to the scheme provided by the present application, the principle and implementation mode of the present application are described by applying the specific examples in the present text, the above example explanation is only for helping the understanding of the method and core idea of the present application; simultaneously, for the general technical personnel in the field, according to the idea of the present application, there will be the change in the specific implementation mode and application range, and the above-mentioned content should not be understood as the limitation of the present application.

Claims

1. An information processing method for intracranial aneurysm hemodynamics prediction, characterized by, Applied to a computer device, comprising: receiving a three-dimensional medical image containing an intracranial aneurysm, and inputting the three-dimensional medical image into a target blood flow hemodynamics prediction model, so as to capture the comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image by a geometric feature extraction module of the target blood flow hemodynamics prediction model; wherein the comprehensive morphological features include local morphological features and global anatomical features; receiving blood flow boundary conditions, and inputting the blood flow boundary conditions into the target blood flow hemodynamics prediction model, so as to perform low-dimensional mapping on the blood flow boundary conditions by a boundary condition coding module of the target blood flow hemodynamics prediction model, to obtain a first low-dimensional feature vector; wherein the first low-dimensional feature vector is a vector representation of the influence of the blood flow boundary conditions on the physical field, and the physical field is a physical field related to blood flow hemodynamics; receiving spatial point coordinates and directional distances corresponding to each sampling point in the CFD model formed by the intracranial aneurysm region, so as to perform low-dimensional mapping on the spatial point coordinates and the directional distances by a spatial query point processing module of the target blood flow hemodynamics prediction model, to obtain a second low-dimensional feature vector; wherein the directional distance is the minimum Euclidean distance between each sampling point and the wall surface of the CFD model of the intracranial aneurysm, and the second low-dimensional feature vector is a position attribute representation of the sampling point and a relationship between each sampling point and the wall surface of the CFD model of the intracranial aneurysm; performing feature fusion on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector by a feature fusion module of the target blood flow hemodynamics prediction model, to predict blood flow hemodynamics parameters of each sampling point based on target fused features; the feature fusion on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector by the feature fusion module of the target blood flow hemodynamics prediction model, comprising: performing feature dimension alignment processing and feature standardization processing on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector, to obtain target comprehensive morphological features, target first low-dimensional feature vectors and target second low-dimensional feature vectors; performing feature splicing processing on the target comprehensive morphological features, the target first low-dimensional feature vectors and the target second low-dimensional feature vectors by a first feature fusion unit of the target blood flow hemodynamics prediction model, to obtain first fused features; performing attention calculation on the target first low-dimensional feature vectors and the target second low-dimensional feature vectors based on a cross-attention mechanism by a second feature fusion unit of the target blood flow hemodynamics prediction model, to obtain second fused features representing blood flow response of the intracranial aneurysm; performing fusion processing on the first fused features and the second fused features by a third feature fusion unit of the target blood flow hemodynamics prediction model through bilinear transformation, to obtain target fused features.

2. The information processing method for hemodynamics prediction of intracranial aneurysms according to claim 1, characterized in that, before the receiving a three-dimensional medical image containing an intracranial aneurysm, further comprising: The original medical image is voxelized to convert the original medical image into a binary three-dimensional medical image; wherein the voxel value of the intracranial aneurysm region in the three-dimensional medical image is 1, and the voxel value of the non-intracranial aneurysm region is 0.

3. The information processing method for hemodynamics prediction of intracranial aneurysms according to claim 1, characterized in that, The geometric feature extraction module of the target blood flow hemodynamics prediction model is a three-dimensional computer vision model; Correspondingly, the geometric feature extraction module of the target blood flow hemodynamics prediction model captures the comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image, including: The three-dimensional medical image is divided into blocks by the target blood flow hemodynamics prediction model to obtain a plurality of three-dimensional voxel blocks, and each three-dimensional voxel block is projected into a preset feature vector space to obtain an initial input sequence; The initial input sequence is input into the geometric feature extraction module to extract shallow features of the initial input sequence based on local window self-attention and shift window mechanism by the geometric feature extraction module, so as to obtain local morphological features including vessel diameter, aneurysm neck width and sub-capsule shape; The initial input sequence is down-sampled by the geometric feature extraction module to extract deep features of the down-sampled initial input sequence based on local window self-attention and shift window mechanism, so as to obtain global anatomical features including the position of the intracranial aneurysm in the vascular tree and the spatial relationship between the intracranial aneurysm and the surrounding brain tissue.

4. The information processing method for hemodynamics prediction of intracranial aneurysms according to claim 1, characterized by, The boundary condition encoding module of the target blood flow hemodynamics prediction model is a multi-layer perceptron network structure; Correspondingly, the boundary condition encoding module of the target blood flow hemodynamics prediction model maps the blood flow boundary condition to a low-dimensional feature vector, including: The multi-layer perceptron network structure of the target blood flow hemodynamics prediction model captures the nonlinear relationship between the inlet mass flow parameter and the outlet pressure parameter to obtain a first low-dimensional feature vector.

5. The information processing method for hemodynamics prediction of intracranial aneurysms according to claim 2, characterized in that, Before receiving the spatial point coordinates and the directional distance corresponding to each sampling point in the CFD model formed by the intracranial aneurysm region, it further includes: Uniformly sampling the region with a voxel value of 1 in the binary three-dimensional medical image to obtain each sampling point; Determine the spatial point coordinates corresponding to each sampling point based on the preset spatial coordinate system; Obtain the shortest distance from each sampling point to the wall surface of the CFD model of the intracranial aneurysm, and determine the directional distance from each sampling point to the wall surface of the CFD model based on the current position of each sampling point and the shortest distance; Wherein, if the current position of the sampling point is inside the wall surface of the CFD model, the directional distance is the negative shortest distance, and if the current position of the sampling point is on the wall surface of the CFD model, the directional distance is 0.

6. The information processing method for hemodynamics prediction of intracranial aneurysms according to claim 5, characterized in that, The spatial query point processing module of the target blood flow hemodynamics prediction model is a multi-layer perceptron network structure; Correspondingly, the spatial query point processing module of the target blood flow hemodynamics prediction model maps the spatial point coordinates and the directional distance to a second low-dimensional feature vector, including: The spatial query point processing module of the target blood flow dynamics prediction model performs spatial position extraction and symbol mapping processing on the spatial point coordinates and the directional distance to obtain a second low-dimensional feature vector.

7. An information processing apparatus for intracranial aneurysm hemodynamics prediction, characterized by comprising: Comprise: The feature extraction module is used for receiving a three-dimensional medical image containing an intracranial aneurysm, and inputting the three-dimensional medical image into a target blood flow dynamics prediction model, so as to capture comprehensive morphological features of the intracranial aneurysm from the three-dimensional medical image by a geometric feature extraction module of the target blood flow dynamics prediction model; wherein the comprehensive morphological features include local morphological features and global anatomical features; The first mapping module is used for receiving blood flow boundary conditions and inputting the blood flow boundary conditions into the target blood flow dynamics prediction model, so as to perform low-dimensional mapping on the blood flow boundary conditions by a boundary condition encoding module of the target blood flow dynamics prediction model to obtain a first low-dimensional feature vector; wherein the first low-dimensional feature vector is a vector representation of the influence of the blood flow boundary conditions on a physical field, and the physical field is a physical field related to blood flow dynamics; The second mapping module is used for receiving spatial point coordinates and directional distances corresponding to each sampling point in a CFD model formed by an intracranial aneurysm region, so as to perform low-dimensional mapping on the spatial point coordinates and the directional distances by a spatial query point processing module of the target blood flow dynamics prediction model to obtain a second low-dimensional feature vector; wherein the directional distance is the minimum Euclidean distance between each sampling point and the wall surface of the CFD model of the intracranial aneurysm, and the second low-dimensional feature vector is a position attribute representation of the sampling point and a relationship between each sampling point and the wall surface of the CFD model of the intracranial aneurysm; The fusion prediction module is used for performing feature fusion on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector by a feature fusion module of the target blood flow dynamics prediction model, so as to predict blood flow dynamics parameters of each sampling point based on target fused features. The fusion prediction module is specifically configured to perform feature dimension alignment processing and feature standardization processing on the comprehensive morphological features, the first low-dimensional feature vector and the second low-dimensional feature vector to obtain target comprehensive morphological features, target first low-dimensional feature vectors and target second low-dimensional feature vectors; perform feature splicing processing on the target comprehensive morphological features, the target first low-dimensional feature vectors and the target second low-dimensional feature vectors by a first feature fusion unit of the target blood flow dynamics prediction model to obtain first fused features; perform attention calculation on the target first low-dimensional feature vectors and the target second low-dimensional feature vectors by a second feature fusion unit of the target blood flow dynamics prediction model based on a cross-attention mechanism to obtain second fused features representing blood flow response of the intracranial aneurysm; and perform fusion processing on the first fused features and the second fused features by a third feature fusion unit of the target blood flow dynamics prediction model through bilinear transformation to obtain target fused features.

8. An electronic device, comprising: Comprise: Memory for saving computer programs; A processor for executing the computer program to implement the steps of the information processing method for intracranial aneurysm hemodynamics prediction according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A computer program product for storing a computer program; wherein the computer program is executed by a processor to implement the steps of the information processing method for intracranial aneurysm hemodynamics prediction according to any one of claims 1 to 6.

Images