Acute ischemic stroke collateral circulation noninvasive assessment method and computing device

Abstract

The embodiment of the invention discloses an acute ischemic stroke collateral circulation noninvasive assessment method and device. The method comprises the steps of obtaining a CTP data set and a DSA data set of different collateral circulation levels, constructing an image segmentation model based on a space-modal attention network and an intracranial artery blood supply partition model based on an atlas, subdividing and extracting different areas of the ischemic side from the angles of perfusion function and vascular structure, and taking a DSA score as a gold standard, combining perfusion function quantitative information and vascular structure quantitative information in each region after ischemic side subdivision to construct a collateral circulation scoring model; and taking perfusion function quantitative information and vascular structure quantitative information of each artery blood supply region, the lateral split region, the infarction core region and the penumbra on the ischemic side as input and model training, and predicting to obtain an individual collateral circulation scoring result. According to the method, the non-invasive evaluation accuracy of collateral circulation can be improved by jointly analyzing the perfusion function and the blood vessel structure information provided by the CTP.

Description

Technical Field

[0001] The present invention relates to the fields of medicine and computer technology, and in particular to a non-invasive assessment method and computing device for collateral circulation of acute ischemic stroke. Background Art

[0002] Stroke is a brain injury caused by blood circulation disorders and has become the leading cause of death and disability in the country. Acute ischemic stroke (AIS) is the most common type of stroke, accounting for about 70% of the total number of patients. When AIS causes a decrease in distal blood perfusion, blood flow can reach the ischemic area through three collateral circulation compensation methods: Wills circle, pial anastomosis branches and local new blood vessels.

[0003] Previous studies have shown that the quality of collateral circulation directly affects the stability of cerebral blood flow in the infarct area, the final size of the infarct, the degree of brain tissue damage during vascular reperfusion, and the hemorrhagic transformation rate after infarction or thrombolysis. With the continuous development of endovascular treatment technology, comprehensive evaluation of cerebral collateral circulation through imaging is an important basis for formulating individualized and precise treatment plans, and has attracted widespread attention from stroke researchers at home and abroad.

[0004] Digital subtraction angiography (DSA) can dynamically and clearly reflect the source of collateral compensation, the speed and coverage of compensatory blood flow, the paths of collateral circulation at all levels, and intuitively measure the degree and range of stenosis of diseased blood vessels. The American Society of Interventional and Therapeutic Neuroradiology / Society of Interventional Radiology (ASITN / SIR) proposed a DSA-based collateral circulation scoring system and recommended the use of a 5-point scale to evaluate cerebral collateral blood flow. After multiple large-scale multicenter clinical controlled studies, it has been proven that the ASITN / SIR scoring system has good consistency and reliability and is recognized as the gold standard for evaluating collateral circulation in acute ischemic stroke. However, DSA is an invasive examination with high costs. Many patients are not suitable for or have no conditions for DSA examination. It is mainly used to complete collateral circulation assessment during interventional thrombectomy surgery. It is difficult to evaluate collateral circulation before treatment and guide treatment decisions. Therefore, non-invasive collateral circulation scoring systems based on CT angiography (CTA) or CT perfusion (CTP) are more often used as alternatives in the emergency treatment process. However, the current scoring systems are mainly based on a certain type of information in the functional or vascular structure of the ischemic side or ischemic area, which is difficult to fully reflect the true level of collateral circulation compensation, and therefore cannot achieve an evaluation efficiency close to the DSA gold standard.

[0005] With the rapid development of imaging technology, CTP can not only quantify perfusion parameters to obtain functional information, but also reconstruct 4D-CTA, and the image quality is close to that of conventional CTA, and then simulate DSA to dynamically observe intracranial blood vessels, obtain the vascular development status of all parts of the brain, and provide information with high temporal and spatial resolution. Therefore, the joint analysis of the perfusion function and vascular structure information provided by CTP can help improve the accuracy of non-invasive assessment of collateral circulation, which may be an ideal solution to replace the gold standard DSA score. Summary of the invention

[0006] In view of the deficiencies of the prior art, an object of the embodiments of the present invention is to provide a non-invasive assessment method and computing device for collateral circulation in acute ischemic stroke.

[0007] To achieve the above objectives, in a first aspect, an embodiment of the present invention provides a method for non-invasively evaluating collateral circulation in acute ischemic stroke, comprising:

[0008] Acquire CTP and DSA datasets at different collateral circulation levels;

[0009] According to the CTP dataset and DSA dataset, an image segmentation model based on spatial-modal attention network and an intracranial artery blood supply zoning model based on atlas were constructed;

[0010] Based on the CTP dataset and DSA dataset, different regions on the ischemic side were subdivided and extracted from the perspectives of perfusion function and vascular structure, and quantitative information on perfusion function and vascular structure in each region after the ischemic side was subdivided was obtained;

[0011] Using DSA score as the gold standard, combined with the quantitative information of perfusion function and vascular structure in each area after the ischemic side was subdivided, a collateral circulation scoring model was constructed.

[0012] The quantitative information of the perfusion function and the quantitative information of the vascular structure of each arterial blood supply area, lateral fissure area, infarct core area and penumbra on the ischemic side are used as input, and the collateral circulation scoring model is trained to predict the individual collateral circulation scoring result.

[0013] As a specific implementation of the present application, an image segmentation model based on a spatial-modal attention network is constructed, specifically:

[0014] The rCBV, rCBF, MTT, Tmax, and MIP-CTA images obtained by CTP reconstruction were used as the CTP dataset;

[0015] Clinical experts manually segmented the images in the CTP dataset to generate labeled data;

[0016] Involving spatial-modal attention networks and constructing composite weighted loss functions;

[0017] Pre-training the spatial-modal attention network using ISLES open source data;

[0018] The pre-trained spatial-modal attention network is retrained using the labeled data.

[0019] As a specific implementation of the present application, a map-based intracranial artery blood supply zoning model is constructed, specifically:

[0020] The 3D CT images of the initial phase in the CTP dataset are used as the floating image dataset;

[0021] Acquire an arterial blood supply zoning map and a reference image from the floating image data set;

[0022] Construct unsupervised registration network and loss function;

[0023] Obtain the image to be partitioned from the CTP dataset;

[0024] Arterial blood supply zoning is performed according to the reference image and the image to be zoned, so as to realize the training of the intracranial arterial blood supply zoning model based on the atlas.

[0025] As a specific implementation of the present application, the arterial blood supply zoning map and the reference image are obtained from the floating image data set, specifically:

[0026] Selecting, from the floating image data set, a ten-division map of intracranial arteries constructed mostly for healthy subjects as an arterial blood supply zoning map; the ten-division map of intracranial arteries includes the anterior, middle, and posterior cerebral artery regions, the pons region, and the cerebellum region on both sides of the left and right sides;

[0027] The image corresponding to the vascular blood supply zoning map is used as the magnetic resonance modality, which is converted into the CT modality through a mature modality migration algorithm and used as a reference image.

[0028] As a specific implementation of the present application, arterial blood supply zoning is performed according to the reference image and the image to be zoned, specifically:

[0029] The deformation field is generated after the image to be partitioned is registered to the reference image. The reversibility of the deformation field is utilized to combine the arterial blood supply partition map with the inverted deformation field to obtain the spatial information of different arterial blood supply areas in the image to be partitioned.

[0030] As a specific implementation of the present application, the quantitative information of the perfusion function and the quantitative information of the vascular structure in each region after the ischemic side is subdivided is obtained, specifically:

[0031] Extract quantitative information of perfusion function from rCBV, rCBF, MTT, and Tmax images of CTP dataset;

[0032] Quantitative analysis was performed using the information ratio of the corresponding partitions between the normal side and the ischemic side to obtain quantitative information on the vascular structure.

[0033] In a second aspect, an embodiment of the present application further provides a computing device, comprising a processor, an input device, an output device and a memory, wherein the processor, input device, output device and memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method of the above-mentioned first aspect.

[0034] The embodiments of the present invention are implemented to study the issues related to intelligent evaluation of collateral circulation, aiming to simultaneously segment the infarct core area, penumbra and lateral fissure area on the ischemic side of stroke through cutting-edge machine learning technology, and zoning the intracranial arterial blood supply, automatically and quantitatively calculate the vascular structure and perfusion function information of different areas on the ischemic side, and then fuse multi-modal information to construct a five-classification model to realize intelligent evaluation of collateral circulation, and verify it in clinical applications. The main advantages include:

[0035] (1) Based on machine learning, a more comprehensive intelligent solution for evaluating collateral circulation is proposed by integrating perfusion function and vascular structure information, breaking through the constraints of incomplete information in previous evaluation systems.

[0036] (2) A neural network model was designed to simultaneously segment the infarct core area and the lateral fissure area of ​​the penumbra on the ischemic side, reflecting the spatial heterogeneity of the changes in perfusion function information on the ischemic side.

[0037] (3) A neural network model was constructed to subdivide the intracranial arterial blood supply area, reflecting the spatial heterogeneity of changes in vascular structure information on the ischemic side.

[0038] (4) Using DSA scores as a reference, a five-classification machine learning model was constructed by integrating the perfusion function and vascular structure quantitative information in different regions of the ischemic side to obtain an intelligent scoring result that fits the DSA gold standard. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the specific implementation of the present invention or the technical solution in the prior art, the drawings required for use in the specific implementation or the description of the prior art are briefly introduced below.

[0040] Figure 1 is a flow chart of a method for non-invasively evaluating collateral circulation in acute ischemic stroke provided by an embodiment of the present invention;

[0041] Figure 2 It is a flow chart of spatial-modal attention network model training;

[0042] Figure 3It is a flow chart of the training of intracranial artery blood supply zoning model based on the atlas;

[0043] Figure 4 It is a block diagram of a computing device. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0046] The inventive concept of the present invention is:

[0047] Given that CTP images can provide rich and complementary perfusion function and vascular structure information, the present invention believes that the joint analysis of perfusion function and vascular structure information can improve the accuracy of collateral circulation assessment. Therefore, accurate segmentation of different areas on the ischemic side to quantitatively extract sufficient perfusion function and vascular structure information is a prerequisite for building an effective collateral circulation scoring model. This project is based on a large number of CTP and DSA data sets including different levels of collateral circulation that can be obtained clinically. Specifically, it constructs an image segmentation model based on a spatial-modal attention network and an intracranial artery blood supply zoning model based on an atlas, and subdivides and extracts different areas on the ischemic side from the perspective of perfusion function and vascular structure; using DSA score as the gold standard, combined with the quantitative information of perfusion function and vascular structure in each area after the ischemic side is subdivided, an intelligent evaluation model is constructed to guide individuals to quickly obtain accurate collateral circulation scores.

[0048] Please refer to Figure 1 , is a non-invasive assessment method for collateral circulation in acute ischemic stroke provided by an embodiment of the present invention, comprising the following steps:

[0049] S1, CTP datasets and DSA datasets at different collateral circulation levels were obtained.

[0050] After collecting, organizing, constructing a certain scale of CTP-DSA image database and performing preprocessing; first, according to the different image characteristics after CTP reconstruction, an effective image segmentation model is studied; secondly, a deep neural network model for intracranial arterial blood supply zoning is studied; then, the perfusion function and vascular structure information in different areas of the ischemic side are automatically and quantitatively calculated, and the DSA five-point scoring result is used as the gold standard to construct a five-classification model to realize the intelligent scoring of collateral circulation.

[0051] Intracranial artery visualization at different phases is an important step in obtaining vascular structure information. This project first uses a deep learning-based vascular extraction model to obtain intracranial artery visualization at the peak phase, which can then be matched to the 4D-CTA image to extract dynamic visualization information to simulate DSA angiography, and then the vascular structure centerline extraction is achieved through the VMTK toolkit.

[0052] To facilitate the subsequent training and testing of image segmentation models, blood supply zoning models, and collateral circulation scoring models, this project divided all data into 70% training sets and 30% test sets.

[0053] S2, based on the CTP dataset and DSA dataset, a spatial-modal attention network-based image segmentation model and an atlas-based intracranial artery blood supply zoning model were constructed.

[0054] 1. Building an image segmentation model based on spatial-modal attention network

[0055] (1) Dataset construction: rCBV, rCBF, MTT, Tmax and MIP-CTA images obtained by CTP reconstruction were used as the dataset.

[0056] (2) Generate annotated data: Comprehensively refer to various images obtained by CTP, and use the manual segmentation of the infarct core area, penumbra, and lateral fissure area by multiple clinical experts as the gold standard to generate annotated data.

[0057] (3) Design of spatial-modal attention network: Based on the U-Net structure, five modal images, rCBV, rCBV, MTT, Tmax and MIP-CTA, are used as inputs. However, the convolution and downsampling operations of each modal image in the downsampling path are independent, so different scale features of each modality are obtained. The spatial-modal attention (SMA) module is embedded between the jump connections, including two branches: spatial attention and modal attention. The spatial attention branch aims to obtain the same spatial weight coefficient as the feature scale by performing operations such as pooling, convolution, and activation on the feature channel. The modal attention branch obtains the modal weight coefficient by performing operations such as pooling, convolution, and activation on the space and each modal feature channel. On this basis, this project will also study and compare the series and parallel combination of the two branches of spatial and modal attention, which is more suitable for image segmentation objectives.

[0058] (4) Constructing a composite weighted loss function: In addition to the Dice loss, the boundary loss is also added to construct a composite weighted loss function according to different weights.

[0059] (5) Model training and optimization: During the training and optimization process, the ISLES open source data is first used to train and construct the spatial-modal attention network as a pre-training model, and then clinical data is used for training to enhance the generalization ability and stability of the model.

[0060] Among them, the spatial-modal attention network model training process is as follows Figure 2 shown.

[0061] 2. Constructing an atlas-based intracranial artery blood supply zoning model

[0062] (1) Construction of floating image dataset: The 3D CT images of the initial phase in the CTP source dataset are used as the floating image dataset.

[0063] (2) Obtaining arterial blood supply zoning maps and reference images: This project selected a map of the ten intracranial arteries constructed from 16 healthy subjects, including the anterior, middle, and posterior cerebral artery regions, the pons region, and the cerebellum region on both sides of the left and right sides. The image corresponding to the map is a magnetic resonance modality, which is converted into a CT modality through a mature modality transfer algorithm and used as a reference image.

[0064] (3) Construct an unsupervised registration network: Generate the deformation field through a fully convolutional network, and then obtain the deformed image through a spatial transformation network. The fully convolutional neural network can refer to the neural network architecture in research plan 2), and the spatial transformation network includes three parts: positioning network, grid generator, and sampler.

[0065] (4) Loss function construction: The loss is mainly based on image similarity loss, combined with the smooth regularization loss of the deformation field to ensure the smoothness of the deformation field.

[0066] (5) Arterial blood supply zoning: After the image to be zoned is registered with the reference image, a deformation field will be generated. The reversibility of the field can then be used to combine the arterial blood supply zoning map with the reversed deformation field to obtain the spatial information of different arterial blood supply areas in the image to be zoned. Using different zoning maps will produce different zoning results. The finer the zoning, the more accurately it can reflect the spatial heterogeneity of the ischemic side information.

[0067] Among them, the training process of the intracranial artery blood supply zoning model based on the atlas is as follows: Figure 3 shown.

[0068] S3, based on the CTP dataset and the DSA dataset, the different regions on the ischemic side are subdivided and extracted from the perspectives of perfusion function and vascular structure, respectively, to obtain the quantitative information of perfusion function and vascular structure in each region after the ischemic side is subdivided.

[0069] After segmentation, different areas on the ischemic side can be directly mapped to the corresponding vascular structure images and perfusion function images, and then the structural and functional quantitative information such as vascular volume, vascular length, development delay time, blood flow, etc. in each area can be calculated. For functional quantitative information, rCBF, rCBV, MTT and Tmax are the quantification of brain tissue blood perfusion, and their absolute values ​​can reflect the perfusion levels of different areas and can be directly extracted from the image.

[0070] However, due to individual differences and the amount of contrast agent used, calculating the absolute value of vascular structure information in different areas on the ischemic side is relatively meaningless. This project intends to use the information ratio of the corresponding partitions of the normal side and the ischemic side to achieve quantitative analysis. This project intends to calculate the peak phase vascular volume ratio, the simultaneous phase vascular volume ratio, and the vascular length ratio. v Calculated by formula (1):

[0071]

[0072] in is a voxel belonging to the vascular structure image development voxel set Φ, I and N represent the ischemic side and the normal side. v The vascular volume ratio of the ischemic side and the normal side at the peak phase can be calculated, and the vascular volume ratio r vg This can be achieved by combining the vascular intensities of different voxels, as shown in formula (2):

[0073]

[0074] in Represents different voxel grayscale values. Vessel length rl It can be calculated by combining the vascular centerline voxel with the average distance of adjacent voxels, as shown in formula (3):

[0075]

[0076] in Represents the average distance between each voxel and its adjacent voxels.

[0077] According to the different areas on the ischemic side and their corresponding perfusion functions and vascular structure information, this project can extract at least 37 types of information (to be determined based on the specific number of phases), as shown in Table 1.

[0078] Table 1. Summary of information in different regions on the ischemic side

[0079]

[0080] S4, using DSA score as the gold standard, combined with the quantitative information of perfusion function and vascular structure in each area after the ischemic side was subdivided, a collateral circulation scoring model was constructed.

[0081] S5, taking the quantitative information of the perfusion function and the quantitative information of the vascular structure of each arterial blood supply area, lateral fissure area, infarct core area and penumbra on the ischemic side as input, and training the collateral circulation scoring model to predict the individual collateral circulation scoring result.

[0082] In this example, the five-point ASTIN / SIR scoring result based on DSA is used as the gold standard, a five-classification machine learning model is designed, and the use of classifiers such as support vector machine (SVM) and random forest is studied. The perfusion function and quantitative information of the vascular structure of each arterial blood supply area, lateral fissure area, infarct core area and penumbra on the ischemic side are used as input to train the model, thereby predicting the individual collateral circulation scoring results, and studying the influence of different quantitative information on the classification accuracy.

[0083] From the above description, it can be seen that the embodiments of the present invention are aimed at studying the issues related to intelligent evaluation of collateral circulation, aiming to simultaneously segment the infarct core area, penumbra and lateral fissure area on the ischemic side of stroke through cutting-edge machine learning technology, and partition the intracranial arterial blood supply, automatically and quantitatively calculate the vascular structure and perfusion function information of different areas on the ischemic side, and then fuse multi-modal information to construct a five-classification model to realize intelligent evaluation of collateral circulation, and verify it in clinical applications. The main advantages include:

[0084] (1) Based on machine learning, a more comprehensive intelligent solution for evaluating collateral circulation is proposed by integrating perfusion function and vascular structure information, breaking through the constraints of incomplete information in previous evaluation systems.

[0085] (2) A neural network model was designed to simultaneously segment the infarct core area and the lateral fissure area of ​​the penumbra on the ischemic side, reflecting the spatial heterogeneity of the changes in perfusion function information on the ischemic side.

[0086] (3) A neural network model was constructed to subdivide the intracranial arterial blood supply area, reflecting the spatial heterogeneity of changes in vascular structure information on the ischemic side.

[0087] (4) Using DSA scores as a reference, a five-classification machine learning model was constructed by integrating the perfusion function and vascular structure quantitative information in different regions of the ischemic side to obtain an intelligent scoring result that fits the DSA gold standard.

[0088] Based on the same inventive concept, an embodiment of the present invention provides a computing device, such as Figure 4 As shown, it may include: one or more processors 101, one or more input devices 102, one or more output devices 103 and a memory 104, wherein the processors 101, the input devices 102, the output devices 103 and the memory 104 are interconnected via a bus 105. The memory 104 is used to store a computer program, wherein the computer program includes program instructions, and the processor 101 is configured to call the program instructions to execute the method of the above method embodiment.

[0089] It should be understood that in the embodiment of the present invention, the processor 101 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0090] The input device 102 may include a keyboard, etc., and the output device 103 may include a display (LCD, etc.), a speaker, etc.

[0091] The memory 104 may include a read-only memory and a random access memory, and provides instructions and data to the processor 101. A portion of the memory 104 may also include a non-volatile random access memory. For example, the memory 104 may also store information on the device type.

[0092] In a specific implementation, the processor 101, input device 102, and output device 103 described in the embodiment of the present invention can execute the implementation method described in the embodiment of the non-invasive assessment method of collateral circulation in acute ischemic stroke provided in the embodiment of the present invention, which will not be repeated here.

[0093] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A non-invasive assessment method for collateral circulation in acute ischemic stroke, characterized in that: include: Acquire CTP and DSA datasets at different collateral circulation levels; According to the CTP dataset and DSA dataset, an image segmentation model based on spatial-modal attention network and an intracranial artery blood supply zoning model based on atlas were constructed; Based on the CTP dataset and DSA dataset, different regions on the ischemic side were subdivided and extracted from the perspectives of perfusion function and vascular structure, and quantitative information on perfusion function and vascular structure in each region after the ischemic side was subdivided was obtained; Using DSA score as the gold standard, combined with the quantitative information of perfusion function and vascular structure in each area after the ischemic side was subdivided, a collateral circulation scoring model was constructed. The quantitative information of the perfusion function and the quantitative information of the vascular structure of each arterial blood supply area, lateral fissure area, infarct core area and penumbra on the ischemic side are used as input, and the collateral circulation scoring model is trained to predict the individual collateral circulation scoring result.

2. The non-invasive assessment method for collateral circulation in acute ischemic stroke according to claim 1, characterized in that: Construct an image segmentation model based on the spatial-modal attention network, specifically: The rCBV, rCBF, MTT, Tmax, and MIP-CTA images obtained by CTP reconstruction were used as the CTP dataset; Clinical experts manually segmented the images in the CTP dataset to generate labeled data; Involving spatial-modal attention networks and constructing composite weighted loss functions; Pre-training the spatial-modal attention network using ISLES open source data; The pre-trained spatial-modal attention network is retrained using the labeled data.

3. The non-invasive assessment method for collateral circulation in acute ischemic stroke according to claim 1, characterized in that: Construct an intracranial arterial blood supply zoning model based on the atlas, specifically: The 3D CT images of the initial phase in the CTP dataset are used as the floating image dataset; Acquire an arterial blood supply zoning map and a reference image from the floating image data set; Construct unsupervised registration network and loss function; Obtain the image to be partitioned from the CTP dataset; Arterial blood supply zoning is performed according to the reference image and the image to be zoned, so as to realize the training of the intracranial arterial blood supply zoning model based on the atlas.

4. The non-invasive assessment method for collateral circulation in acute ischemic stroke according to claim 3, characterized in that: The arterial blood supply zoning atlas and the reference image are obtained from the floating image data set, specifically: Selecting, from the floating image data set, a ten-division map of intracranial arteries constructed mostly for healthy subjects as an arterial blood supply zoning map; the ten-division map of intracranial arteries includes the anterior, middle, and posterior cerebral artery regions, the pons region, and the cerebellum region on both sides of the left and right sides; The image corresponding to the vascular blood supply zoning map is used as the magnetic resonance modality, which is converted into the CT modality through a mature modality migration algorithm and used as a reference image.

5. The non-invasive assessment method for collateral circulation in acute ischemic stroke according to claim 3, characterized in that: The unsupervised registration network generates a deformation field through a fully convolutional network, and then obtains a deformed image through a spatial transformation network; the spatial transformation network includes a positioning network, a grid generator and a sampler.

6. The non-invasive assessment method for collateral circulation in acute ischemic stroke according to claim 3, characterized in that: Arterial blood supply zoning is performed according to the reference image and the image to be zoned, specifically: The deformation field is generated after the image to be partitioned is registered to the reference image. The reversibility of the deformation field is utilized to combine the arterial blood supply partition map with the inverted deformation field to obtain the spatial information of different arterial blood supply areas in the image to be partitioned.

7. The non-invasive assessment method for collateral circulation in acute ischemic stroke according to claim 1, characterized in that: After the ischemic side is subdivided, the quantitative information of perfusion function and vascular structure in each region is obtained, specifically: Extract quantitative information of perfusion function from rCBV, rCBF, MTT, and Tmax images of CTP dataset; Quantitative analysis was performed using the information ratio of the corresponding partitions between the normal side and the ischemic side to obtain quantitative information on the vascular structure.

8. A computing device, characterized in that The invention comprises a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method for non-invasive assessment of collateral circulation in acute ischemic stroke as described in any one of claims 1 to 7.

Images