Method and device for training detection model for detecting vascular adhesion and storage medium

Abstract

The invention discloses a method and equipment for training a detection model for detecting vascular adhesion and a storage medium. The method comprises the following steps: acquiring multi-modal image data, wherein the multi-modal image data comprises digital subtraction angiography data, CT angiography data and magnetic resonance angiography data; performing blood vessel segmentation based on the multi-modal image data to obtain a graph structure of a blood vessel; according to the graph structure, graph structure features of a single blood vessel and target blood vessel features are extracted, and whether the blood vessel passes through an adhesion blood vessel is marked; constructing training data based on the graph structure features of the single blood vessel, the target blood vessel features and the labeling result; and inputting the training data into a detection model for vascular adhesion detection to obtain a detection result, so as to train the detection model for vascular adhesion detection. By means of the scheme, accurate positioning and efficient detection of vascular adhesion can be improved.

Description

Technical Field

[0001] The present application generally relates to the technical field of vascular adhesion detection, and more specifically, to a method for training a detection model for detecting vascular adhesion, a method for detecting vascular adhesion, a device, and a computer-readable storage medium. Background Art

[0002] Digital subtraction angiography ("DSA") is the gold standard for diagnosing vascular diseases. In its images, intracranial blood vessels present a tree-like structure. However, due to volume effects and limitations of imaging technology, vascular adhesions often occur after vascular segmentation, leading to the formation of ring structures. In existing technologies, skeletonization algorithms are usually used to extract centerlines and construct vascular maps after vascular segmentation, but the ring structures formed by adhesions will significantly affect the efficiency and stability of subsequent processing such as path tracing, and may even cause the algorithm to crash. Traditional methods rely on manual labeling or simple threshold segmentation, which makes it difficult to accurately detect small target adhesions. In addition, the model is highly complex, and the training and deployment costs are high.

[0003] In view of this, there is an urgent need to provide a solution for training a detection model for detecting vascular adhesions in order to improve the accurate positioning and efficient detection of vascular adhesions. Summary of the Invention

[0004] In order to at least solve one or more of the technical problems mentioned above, the present application proposes a solution for training a detection model for detecting vascular adhesion in multiple aspects.

[0005] In a first aspect, the present application provides a method for training a detection model for detecting vascular adhesions, comprising: acquiring multimodal image data, wherein the multimodal image data includes digital subtraction angiography data, CT angiography data, and magnetic resonance angiography data; performing vascular segmentation based on the multimodal image data to obtain a graph structure of the blood vessels; extracting graph structural features and target vessel features of a single blood vessel based on the graph structure, and marking whether the vessel passes through an adhesion vessel; constructing training data based on the graph structural features, target vessel features, and marking results of the single blood vessel; and inputting the training data into a detection model to perform vascular adhesion detection to obtain a detection result, thereby training the detection model for vascular adhesion detection.

[0006] In some embodiments, the graph structure features include at least node information, edge information, and blood vessel information at the node, and the blood vessel information at the node includes at least one or more of the radius of the node, the grayscale average value or the standard deviation within the radius area.

[0007] In some other embodiments, the target blood vessel feature is extracted by the following operation: a target node in the single blood vessel is set to a target pixel value, and the remaining nodes are set to 0, so as to extract the target blood vessel feature.

[0008] In some further embodiments, it further includes: judging whether the target node passes through an adhesion vessel according to the node information and the edge information; and marking the target node as passing through an adhesion vessel and not passing through an adhesion vessel based on the judgment result to obtain the marking result.

[0009] In some further embodiments, the method further includes: removing the distal blood vessel of the single blood vessel; and constructing the training data based on the graph structure features, target blood vessel features and annotation results of the single blood vessel after removing the distal blood vessel.

[0010] In some further embodiments, the detection model includes an encoder and a classifier, and the training data is input into the detection model to perform vascular adhesion detection to obtain a detection result, so as to train the detection model for vascular adhesion detection: based on the training data, the encoder is used to perform an encoding operation to obtain an encoding result; and the classifier is used to perform a classification operation on the encoding result to obtain a detection result of vascular adhesion detection, so as to train the detection model for vascular adhesion detection.

[0011] In some further embodiments, the detection model comprises a graph network model.

[0012] In a second aspect, the present application provides a method for detecting vascular adhesions, comprising: acquiring multimodal image data, wherein the multimodal image data includes digital subtraction angiography data, CT angiography data, and magnetic resonance angiography data; performing vascular segmentation based on the multimodal image data to acquire a graph structure of the blood vessels; extracting graph structure features and target vessel features of a single blood vessel based on the graph structure; and inputting the graph structure features and target vessel features of the single blood vessel into a detection model trained according to one or more embodiments of the first aspect to perform vascular adhesion detection to obtain a detection result.

[0013] In a third aspect, the present application provides a device comprising: a processor; and a memory storing computer instructions for training a detection model for detecting vascular adhesion, which, when executed by the processor, implement multiple embodiments of the aforementioned first aspect; or storing computer instructions for detecting vascular adhesion, which, when executed by the processor, implement one embodiment of the aforementioned second aspect.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium storing computer program instructions for training a detection model for detecting vascular adhesions, which, when executed by one or more processors, implement multiple embodiments of the aforementioned first aspect; or storing computer instructions for detecting vascular adhesions, which, when executed by a processor, implement one embodiment of the aforementioned second aspect.

[0015] Through the scheme for training the detection model for detecting vascular adhesions provided above, the embodiment of the present application collects multimodal imaging data, covering vascular structure information of different imaging principles, thereby improving the adaptability of the model to complex scenarios; based on the graph structure generated by vascular segmentation, it effectively retains the position, connection relationship and attribute characteristics of vascular nodes, providing a structured data basis for adhesion analysis; by extracting the graph structure features and target vessel features of a single blood vessel and annotating adhesions, multi-dimensional feature fusion and accurate label generation are achieved, ensuring that the model learns the geometric and grayscale abnormal patterns of the adhesion area; the constructed training data is combined with the graph structure features and the annotation results, and the model parameters are optimized through end-to-end training, ultimately achieving accurate positioning and efficient detection of vascular adhesions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and other objects, features and advantages of the exemplary embodiments of the present application will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0017] Figure 1 is an exemplary flowchart illustrating a method for training a detection model for detecting vascular adhesion according to an embodiment of the present application;

[0018] Figure 2 is an exemplary flowchart illustrating an overall process of training a detection model for detecting vascular adhesion according to an embodiment of the present application;

[0019] Figure 3 is an exemplary schematic diagram showing a partial blood vessel segment according to an embodiment of the present application;

[0020] Figure 4 is an exemplary flowchart illustrating a method for detecting vascular adhesion according to an embodiment of the present application;

[0021] Figure 5 is a block diagram showing an exemplary structure of a device according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0023] It should be understood that the terms "include" and "comprising" used in the description and claims of this application indicate the presence of described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0024] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this specification and claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" as used in this specification and claims refers to any and all possible combinations of one or more of the associated listed items, including and including these combinations.

[0025] As used in this specification and claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0026] The specific implementation of the present application will be described in detail below with reference to the accompanying drawings.

[0027] Figure 1 FIG. 1 is an exemplary flow chart illustrating a method 100 for training a detection model for detecting vascular adhesion according to an embodiment of the present application. Figure 1 As shown in , the method 100 includes: step S101: acquiring multimodal image data, wherein the multimodal image data includes digital subtraction angiography data, CT angiography data, and magnetic resonance angiography data; step S102: performing blood vessel segmentation based on the multimodal image data to obtain a graph structure of the blood vessels; step S103: extracting graph structure features and target vessel features of a single blood vessel based on the graph structure, and marking whether it passes through an adhesion vessel; step S104: constructing training data based on the graph structure features, target vessel features, and marking results of the single blood vessel; and step S105: inputting the training data into a detection model to perform vascular adhesion detection to obtain a detection result, thereby training the detection model for vascular adhesion detection.

[0028] First, in step S101, multimodal imaging data is collected. The multimodal imaging data may include digital subtraction angiography data, CT angiography data, and magnetic resonance angiography data. Among them, digital subtraction angiography data is the gold standard for screening and diagnosis of various vascular diseases, which can provide details of the vascular lumen; while CT angiography data and magnetic resonance angiography data can also be used for vascular disease screening, follow-up, etc., respectively supplementing the spatial relationship between blood vessels and surrounding tissues and displaying hemodynamic characteristics. By using three modal data for the training of the detection model, the embodiment of the present application can cover the vascular structure information of different imaging principles and improve the adaptability of the model to complex scenes.

[0029] Next, in step S102, vessel segmentation is performed based on the multimodal imaging data to obtain a vascular graph structure. In some embodiments, vessel segmentation can be performed using image processing methods or deep learning methods. Such image processing methods may include, but are not limited to, image edge detection and image morphology. Deep learning methods may be based on existing trained segmentation models for vessel segmentation, and this application imposes no restrictions on this. Using the segmented vessels, a network consisting of nodes and edges can be generated using, for example, the sknw library, thereby obtaining a vascular graph structure.

[0030] Based on the obtained vascular graph structure, in step S103, graph structural features and target vessel features of individual vessels are extracted from the graph structure, and whether the graph passes through adhesion vessels is noted. In some embodiments, the graph structural features may include, but are not limited to, node information, edge information, and vessel information at the node. Node vessel information may include, but is not limited to, one or more of the node radius, the grayscale mean within the radius, or the standard deviation. Node information may include, for example, the coordinate position of the vessel center point in image space; edge information may include the connectivity between nodes, the edge length of the vessel segment, and directional information to aid in determining the vessel's course.

[0031] In some embodiments, the target vessel features can be extracted by the following operation: the target node in a single vessel is set to the target pixel value (for example, set to 1), and the remaining nodes are set to 0 to extract the target vessel features. That is, a binary mask of the target vessel is extracted. In some embodiments, it can be determined whether the target node passes through an adhesion vessel based on the node information and edge information, and the target node is marked as passing through an adhesion vessel and not passing through an adhesion vessel based on the judgment result to obtain a labeling result. That is, two types of labels are formed, one type passes through an adhesion vessel, and its value can be set to 1; the other type does not pass through an adhesion vessel, and its value can be set to 0. In some implementation scenarios, all paths of a single vessel are traversed through the graph structure (nodes, edges, attributes) generated based on vessel segmentation. If a node has multiple paths connected to the same end point, it is determined to be a ring structure, that is, there is an adhesion path.

[0032] Furthermore, in step S104, training data is constructed based on the graph structure features of a single vessel, the target vessel features, and the annotation results. For example, assume that the graph structure of a single vessel is denoted as G, the graph structure features are denoted as feature1, the target vessel features are denoted as feature2, and the annotation result indicates that the vessel passes through an adhesion vessel, with a value of 1. In this scenario, the training data [[G, feature1, feature2], 1] can be obtained. Similarly, if the annotation result indicates that the vessel does not pass through an adhesion vessel, with a value of 0, the training data [[G, feature1, feature2], 0] can be obtained.

[0033] In some implementations, the distal vessels of a single vessel can be removed. Training data can be constructed based on the graph structure characteristics of the removed vessels, the target vessel characteristics, and the annotation results. This allows the vessel segments in the neighborhood near the target node to be retained, while redundant branches further distally are removed to construct a streamlined subgraph, improving training efficiency.

[0034] As you can understand, the distal end of a single vessel is considered distal if it's more than three edges away from the target node. As an example, let's denote a single vessel after removing the distal vessels as G_box, and the corresponding graph structure features and target vessel features as feature1_box and feature2_box, respectively. The annotation result is "passing through an adhesion vessel" with a value of 1. In this scenario, we obtain training data [[G_box, feature1_box, feature2_box], 1]. Alternatively, if the annotation result is "passing through an adhesion vessel" with a value of 0, we obtain training data [[G_box, feature1_box, feature2_box], 0].

[0035] Finally, at step S105, the training data is input into the detection model to perform vascular adhesion detection to obtain a detection result, thereby training the detection model for vascular adhesion detection. In some embodiments, the detection model may include a graph network model. In other embodiments, the detection model may include an encoder and a classifier. Specifically, based on the training data, an encoder is used to perform an encoding operation to obtain an encoding result, and then a classifier is used to classify the encoding result to obtain a vascular adhesion detection result, thereby training the detection model for vascular adhesion detection.

[0036] In an exemplary scenario, if the encoder is denoted as encoder and the classifier is denoted as clf, the encoder can first perform an encoding operation to obtain the encoding result feature = encoder(G, feature1, feature2). For the scenario where distal blood vessels are removed, feature = encoder(G_box, feature1_box, feature2_box). Next, the classifier performs a classification operation pred = clf(feature) on the encoding result and calculates a cross-entropy loss function, such as Loss = CrossEntroy(label, pred), for backpropagation to update the parameters of the encoder and classifier clf to train the detection model for vascular adhesion detection. Label represents the label value.

[0037] Combined with the above description, it can be seen that the multimodal imaging data acquisition (digital subtraction angiography data, CT angiography data and magnetic resonance angiography data) of the embodiment of the present application covers vascular information of different imaging principles, thereby improving the adaptability of the model to complex scenes. By adopting, for example, the sknw library to generate a vascular graph containing node positions, radii, edge connection relationships and grayscale attributes, a structured data basis is provided for subsequent analysis, and graph structure features (such as node information, edge information, vascular attributes) and target vascular features (such as binary masks) are extracted and adhesions are marked. The geometric and grayscale anomalies of the adhesion area are accurately captured through multi-dimensional feature fusion, and accurate label generation improves data consistency. Furthermore, by constructing training data or streamlining training data (such as removing distal blood vessels), and adopting, for example, cross entropy loss to optimize the graph network model, end-to-end efficient feature learning is achieved, while ensuring detection accuracy and reducing model complexity, supporting real-time clinical applications.

[0038] Figure 2 FIG. 1 is an exemplary flow chart showing the overall process of training a detection model for detecting vascular adhesion according to an embodiment of the present application. Figure 2As shown in FIG, in step S201, multimodal image data is acquired, where the multimodal image data includes digital subtraction angiography data, CT angiography data, and magnetic resonance angiography data. Next, in step S202, vessel segmentation is performed based on the multimodal image data to obtain a vessel graph structure. Specifically, vessel segmentation can be performed using image processing methods or deep learning methods, and the vessel graph structure can be generated using, for example, the sknw library. In step S203, distal vessels are removed, and in step S204, the graph structure features, target vessel features, and annotation results of the single vessel after the distal vessels are removed are extracted to obtain training data. Furthermore, the training data is input into a detection model, and in step S205, an encoder is used to perform an encoding operation to obtain an encoding result. In step S206, a classifier is used to classify the encoding result, and in step S207, a loss function (e.g., cross-entropy loss) is calculated to train the detection model for vascular adhesion detection. The trained detection model is then generated in step S208.

[0039] Figure 3 FIG is an exemplary schematic diagram showing a portion of a blood vessel segment according to an embodiment of the present application. Figure 3 Figure 1 shows a partial vascular segment graph X. When training a detection model for vascular adhesions, the segmentation is first performed on this graph. Then, a vascular graph structure can be generated using, for example, the sknw library. The figure only illustrates the nodes in the graph structure, such as points 1, 2, 3, 4, 5, and 6. By connecting the corresponding nodes according to the graph structure, corresponding individual vascular segments can be represented, such as vascular segments 1-3-6 and vascular segment 5-2-1.

[0040] Next, the graph structure features and target vessel features of individual vessels are extracted based on the graph structure, and whether or not they pass through adhesion vessels is annotated. Taking vessel segment 1-3-6 as an example, the node information, edge information, node radius, and grayscale mean or standard deviation within the radius region from its graph structure G are extracted as graph structure feature 1. Furthermore, target vessel feature 2 is extracted, with the corresponding nodes in this target vessel feature assigned a value of 1 and all other nodes assigned a value of 0. Furthermore, since vessel segment 1-3-6 is an adhesion vessel, its annotated value is 1, resulting in the training data [[G, feature1, feature2], 1]. Similarly, since vessel segment 5-2-1 does not pass through adhesion vessels and has a value of 0, the training data [[G', feature1', feature2'], 0] is obtained. As mentioned above, we can construct training data [[G_box,feature1_box,feature2_box],1] by removing distal blood vessels, input it into the detection model, use the encoder to perform encoding operations to obtain encoding results, and then use the classifier to perform classification operations on the encoding results to obtain the detection results of vascular adhesion detection, so as to train the detection model of vascular adhesion detection.

[0041] Figure 4 FIG. 4 is an exemplary flow chart illustrating a method 400 for detecting vascular adhesion according to an embodiment of the present application. Figure 4 As shown in , the method 400 includes: step S401: acquiring multimodal image data, wherein the multimodal image data includes digital subtraction angiography data, CT angiography data, and magnetic resonance angiography data; step S402: performing blood vessel segmentation based on the multimodal image data to obtain the graph structure of the blood vessels; step S403: extracting the graph structure features and target vessel features of a single blood vessel based on the graph structure; step S404: inputting the graph structure features and target vessel features of a single blood vessel into the trained detection model to perform vascular adhesion detection to obtain a detection result. For more details on the extraction of the graph structure features and target vessel features of a single blood vessel and the training of the detection model, please refer to the above Figure 1 、 Figure 2 According to the detection result, if the classification operation pred is 1, it means that there is adhesion at the middle endpoint, and other values indicate that there is no adhesion.

[0042] In some embodiments, when detecting vascular adhesions, the distal end of a single vessel can be removed. The graph structural features of the single vessel after the distal end is then input into a trained detection model along with the target vessel features to perform vascular adhesion detection and obtain a detection result. Similarly, based on the detection result, if the classification operation pred is 1, adhesion is present at the intermediate endpoint; any other value indicates the absence of adhesion.

[0043] Figure 5 is an exemplary structural block diagram of a device 500 according to an embodiment of the present application. It is understood that the device 500 may include the apparatus of the embodiment of the present application, and the device implementing the solution of the present application may be a single device (such as a computing device) or a multifunctional device including various peripheral devices.

[0044] like Figure 5 As shown in , the device of the present application may also include a central processing unit ("CPU") 511, which may be a general-purpose CPU, a dedicated CPU, or other information processing and program execution unit. Furthermore, the device 500 may also include a large-capacity memory 512 and a read-only memory ("ROM") 513, wherein the large-capacity memory 512 may be configured to store various types of data, including various multimodal data, graph structure features, target blood vessel features and annotation results, algorithm data, intermediate results, and various programs required to run the device 500. The ROM 513 may be configured to store data and instructions required for power-on self-test of the device 500, initialization of various functional modules in the system, basic input / output drivers of the system, and booting the operating system.

[0045] Optionally, the device 500 may also include other hardware platforms or components, such as the illustrated tensor processing unit ("TPU") 514, graphics processing unit ("GPU") 515, field programmable gate array ("FPGA") 516, and machine learning unit ("MLU") 517. It will be appreciated that while various hardware platforms or components are shown in the device 500, these are merely exemplary and non-restrictive, and those skilled in the art may add or remove corresponding hardware as needed. For example, the device 500 may include only a CPU, associated storage devices, and interface devices to implement the method for training a detection model for detecting vascular adhesions or the method for detecting vascular adhesions described herein.

[0046] In some embodiments, to facilitate the transmission and interaction of data with an external network, the device 500 of the present application further includes a communication interface 518, which can be connected to a local area network / wireless local area network ("LAN / WLAN") 505 via the communication interface 518, and then connected to a local server 506 or the Internet ("Internet") 507 via the LAN / WLAN. Alternatively or additionally, the device 500 of the present application can also be directly connected to the Internet or a cellular network via the communication interface 518 based on wireless communication technology, such as third generation ("3G"), fourth generation ("4G"), or fifth generation ("5G") wireless communication technology. In some application scenarios, the device 500 of the present application can also access a server 508 and a database 509 on an external network as needed to obtain various known algorithms, data, and modules, and can remotely store various data, such as various data or instructions for presenting multimodal data, graph structure features, target vessel features, and annotation results.

[0047] The peripheral devices of the device 500 may include a display device 502, an input device 503, and a data transmission interface 504. In one embodiment, the display device 502 may include, for example, one or more speakers and / or one or more visual displays, which are configured to train the detection model for detecting vascular adhesions of the present application or to provide voice prompts and / or image video displays for detecting vascular adhesions. The input device 503 may include, for example, other input buttons or controls such as a keyboard, a mouse, a microphone, a gesture capture camera, etc., which are configured to receive audio data input and / or user instructions. The data transmission interface 504 may include, for example, a serial interface, a parallel interface, or a universal serial bus interface ("USB"), a small computer system interface ("SCSI"), a serial ATA, a FireWire ("FireWire"), a PCI Express, and a high-definition multimedia interface ("HDMI"), etc., which are configured for data transmission and interaction with other devices or systems. According to the solution of the present application, the data transmission interface 504 can receive multimodal data collected from CT or MRI equipment and transmit data or results including multimodal data or various other types to the device 500.

[0048] The CPU 511, mass storage 512, ROM 513, TPU 514, GPU 515, FPGA 516, MLU 517, and communication interface 518 of the device 500 of the present application can be interconnected via a bus 519 and can interact with peripheral devices via the bus. In one embodiment, the CPU 511 can control other hardware components in the device 500 and its peripheral devices via the bus 519.

[0049] Combination of the above Figure 5The devices that can be used to implement the present application are described. It should be understood that the device structures or architectures herein are merely exemplary, and the implementation methods and implementation entities of the present application are not limited thereto, but can be modified without departing from the spirit of the present application.

[0050] According to the above description in combination with the accompanying drawings, those skilled in the art can also understand that the embodiments of the present application can also be implemented by software programs. Therefore, the present application also provides a computer-readable storage medium, which stores computer-readable instructions for training a detection model for detecting vascular adhesions or detecting vascular adhesions. When the computer-readable instructions are executed by one or more processors, they can be used to implement the present application in combination with the accompanying drawings. Figure 1 The method for training the detection model for detecting vascular adhesion described in the present application; or realizing the method of ... Figure 4 Methods for detecting vascular adhesions are described.

[0051] It should be noted that although the operations of the present method are described in a particular order in the accompanying drawings, this does not require or imply that the operations must be performed in that particular order, or that all of the operations shown must be performed to achieve the desired results. Rather, the steps depicted in the flowcharts may be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into a single step, and / or a single step may be broken down into multiple steps.

[0052] It should be understood that when the terms "first," "second," "third," and "fourth," etc., are used in the claims, specification, and drawings of this application, they are only used to distinguish different objects, rather than to describe a specific order. The terms "comprise" and "comprising" used in the specification and claims of this application indicate the presence of the 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.

[0053] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this specification and claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" as used in this specification and claims refers to any and all possible combinations of one or more of the associated listed items, including and including these combinations.

[0054] Although the implementation methods of this application are as described above, the contents are only examples adopted to facilitate understanding of this application and are not intended to limit the scope and application scenarios of this application. Any technician in the technical field described in this application can make any modifications and changes in the form and details of implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be based on the scope defined by the attached claims.

[0055] In addition, the collection and acquisition of various data in this application complies with relevant laws and regulations and is authorized by the data provider. Any organization or individual that needs to obtain external data must obtain authorization in accordance with the law and ensure data security. They must not illegally collect, use, process, or transmit unauthorized or unprotected data, nor illegally buy, sell, provide, or disclose unauthorized or unprotected data.

Claims

1. A method for training a detection model for detecting vascular adhesions, comprising: Acquiring multimodal imaging data, wherein the multimodal imaging data includes digital subtraction angiography data, CT angiography data, and magnetic resonance angiography data; Performing blood vessel segmentation based on the multimodal image data to obtain a graph structure of the blood vessels; extracting the graph structure features of a single blood vessel and the target blood vessel features according to the graph structure, and marking whether it passes through an adhesion vessel; constructing training data based on the graph structure features of the single blood vessel, the target blood vessel features, and the annotation results; The training data is input into a detection model to perform vascular adhesion detection to obtain a detection result, so as to train the detection model for vascular adhesion detection.

2. The method according to claim 1, wherein the graph structure features include at least node information, edge information, and blood vessel information at the node, and the blood vessel information at the node includes at least one or more of the radius of the node, the grayscale average value within the radius area, or the standard deviation.

3. The method according to claim 2, wherein the target blood vessel feature is extracted by: The target node in the single blood vessel is set to a target pixel value, and the remaining nodes are set to 0, so as to extract the target blood vessel feature.

4. The method according to claim 2, further comprising: Determine whether the target node passes through an adhesion vessel based on node information and edge information; Based on the judgment result, the target node is marked as passing through the adhesion blood vessel and not passing through the adhesion blood vessel to obtain the marking result.

5. The method according to claim 1, further comprising: removing the distal blood vessel of the single blood vessel; The training data is constructed based on the graph structure features of a single blood vessel after removing the distal blood vessels, the target blood vessel features and the annotation results.

6. The method according to claim 1 or 5, wherein the detection model comprises an encoder and a classifier, and the training data is input into the detection model to perform vascular adhesion detection to obtain a detection result, so as to train the detection model for vascular adhesion detection: Based on the training data, using the encoder to perform an encoding operation to obtain an encoding result; The classifier is used to perform a classification operation on the encoding result to obtain a detection result of vascular adhesion detection, so as to train a detection model for vascular adhesion detection. The method of claim 6 , wherein the detection model comprises a graph network model.

8. A method for detecting vascular adhesion, comprising: Acquiring multimodal imaging data, wherein the multimodal imaging data includes digital subtraction angiography data, CT angiography data, and magnetic resonance angiography data; Performing blood vessel segmentation based on the multimodal image data to obtain a graph structure of the blood vessels; Extracting graph structure features of a single blood vessel and target blood vessel features according to the graph structure; and The graph structural features of the single blood vessel and the target blood vessel features are input into the detection model trained according to the method according to any one of claims 1 to 7 to perform vascular adhesion detection to obtain a detection result.

9. A device comprising: processor; as well as A memory storing computer instructions for training a detection model for detecting vascular adhesion, which, when executed by a processor, implement the method according to any one of claims 1 to 7; or storing computer instructions for detecting vascular adhesion, which, when executed by a processor, implement the method according to claim 8.

10. A computer-readable storage medium storing computer program instructions for training a detection model for detecting vascular adhesions, wherein the computer program instructions, when executed by one or more processors, implement the method according to any one of claims 1 to 7; or storing computer program instructions for detecting vascular adhesions, wherein the computer program instructions, when executed by one or more processors, implement the method according to claim 8.

Images