A cerebral aneurysm detection model establishment method, device and equipment and a storage medium

Abstract

This invention relates to the field of model building and discloses a method, apparatus, device, and storage medium for building a cerebral aneurysm detection model. The method includes: acquiring a fusion map of cerebral artery data and inputting it into a graph convolutional model to obtain graph convolutional features; acquiring vascular preprocessing results and vascular knowledge fusion data and inputting them into an encoder for decoupling to obtain decoupled features; fusing the graph convolutional features and decoupled features and inputting the feature fusion result into a decoder; and using vascular spatial annotation to supervise the training of the detection model to obtain a cerebral aneurysm detection model. This invention processes the fusion map of cerebral artery data to obtain corresponding graph convolutional features, processes the vascular preprocessing results and vascular knowledge fusion data, establishes a cerebral aneurysm detection model using an encoding and decoding framework, constructs a knowledge-guided multi-scale network model, and uses the cerebral aneurysm detection model to detect cerebral aneurysms, thereby improving the accuracy of cerebral aneurysm detection results.

Description

Technical Field

[0001] This invention relates to the field of model building, specifically to a method, apparatus, device, and storage medium for building a brain aneurysm detection model. Background Technology

[0002] Intracranial aneurysms are among the most common cerebrovascular diseases. They are highly prevalent and high-risk, with insidious onset, making early detection and intervention crucial. Currently, improving network structures for aneurysm detection is a major research focus. As a typical example of small-target detection, aneurysm detection benefits from effective optimization methods through network structure improvement.

[0003] In existing technologies, SENet channel attention mechanisms are added to the high-dimensional feature convolutional layers after downsampling in 3DU-Net. Residual networks and dual-channel attention modules can also be incorporated into 3DU-Net. Furthermore, GLIA-Net (a segmentation network based on global localization) can be proposed using a multi-scale detection method. However, the knowledge-guided networks in these detection methods only introduce single-concept knowledge, resulting in low accuracy in cerebral aneurysm detection. Summary of the Invention

[0004] In view of this, the present invention provides a method, apparatus, device and storage medium for establishing a cerebral aneurysm detection model, in order to improve the accuracy of cerebral aneurysm detection.

[0005] In a first aspect, the present invention provides a method for establishing a cerebral aneurysm detection model, wherein the detection model includes a graph convolutional model, an encoder, and a decoder, and the method includes:

[0006] A fused image of cerebral artery data is obtained and input into the graph convolution model to obtain graph convolution features;

[0007] The vascular preprocessing results and vascular knowledge fusion data are obtained, and the vascular preprocessing results and vascular knowledge fusion data are input into the encoder for decoupling to obtain decoupled features;

[0008] The graph convolutional features and decoupled features are fused, and the result of the feature fusion is input into the decoder. The detection model is trained under supervision using vascular spatial annotation to obtain a cerebral aneurysm detection model.

[0009] This invention takes multi-level fused data as input, processes the fusion map of cerebral artery data to obtain the graph convolution features corresponding to the fusion map of cerebral artery data, processes the vascular preprocessing results and vascular knowledge fusion data, establishes a cerebral aneurysm detection model with encoding and decoding as the framework, constructs a multi-scale network model guided by graph convolution knowledge, and then uses the cerebral aneurysm detection model to detect cerebral aneurysms, thereby improving the accuracy of cerebral aneurysm detection results.

[0010] In an optional implementation, before inputting the fused cerebral artery data map into the graph convolution model, the method further includes:

[0011] Using semantic cross-entropy as the loss function, a parallel convolutional network is trained to obtain a graph convolutional model.

[0012] This invention trains a parallel convolutional network to obtain a graph convolutional model, which facilitates the processing of fusion maps of cerebral artery data.

[0013] In an optional implementation, before inputting the vascular preprocessing results and vascular knowledge fusion data into the encoder, the method further includes:

[0014] Acquire cerebral artery segmentation data and vascular structure layering;

[0015] The receptive field of the encoder is established with the vascular centerline as the axis, based on the cerebral artery segmentation data and vascular structure layering.

[0016] This invention constructs a multi-scale vascular morphology receptive field in a multi-layered vascular space based on cerebral artery segmentation data and vascular structure layering, so as to input the vascular preprocessing results and vascular knowledge fusion data into the encoder for processing.

[0017] In one alternative implementation, the scale of the receptive field includes local blood vessels, blood vessel segments, blood vessel zones, whole blood vessels, and the entire image.

[0018] This invention sets the scale of the receptive field to observe local features of the data while grasping global information, thereby enabling a more comprehensive analysis of vascular data images.

[0019] In one optional implementation, the encoder has a multi-branch structure. The step of inputting the vascular preprocessing results and vascular knowledge fusion data into the encoder for decoupling to obtain decoupling features includes:

[0020] The vascular preprocessing results and vascular knowledge fusion data are input into the encoder branch to obtain the corresponding hierarchical features;

[0021] Based on the spatial attention mechanism, the hierarchical features are transmitted, and the hierarchical features are decoupled into multiple branches to obtain decoupled features.

[0022] This invention uses a spatial attention mechanism to transmit hierarchical features, which can enhance the effective transmission of hierarchical features. It utilizes an encoder to decouple data into multiple branches, thereby achieving a long connection mechanism for decoupling hierarchical features, thus decoupling the data to facilitate the processing of decoupled features.

[0023] In one optional implementation, the decoder employs a dual-branch path, comprising a supervisory branch and an instance branch, wherein the supervisory branch is obtained as follows:

[0024] Based on the result of feature fusion, the supervision path is fed back for supervision, resulting in a supervision branch.

[0025] This invention obtains the supervisory branch of the decoder by performing feedback supervision on the supervisory path, and then processes the data processed by the encoder through the supervisory branch.

[0026] In one optional implementation, the step of inputting the result of feature fusion into the decoder and training the detection model using multi-scale annotation as supervision to obtain a cerebral aneurysm detection model includes:

[0027] The result of feature fusion is input into the convolutional layer of the supervision branch, and the features processed by the supervision branch are input into the instance branch.

[0028] By utilizing vascular spatial annotation and minimizing the vascular spatial annotation loss function, a supervised training method is used to train the detection model, resulting in a cerebral aneurysm detection model.

[0029] This invention establishes a brain aneurysm detection model based on a knowledge-guided neural network by inputting the features of the supervised branch into the instance branch for processing.

[0030] Secondly, the present invention provides a device for establishing a cerebral aneurysm detection model, wherein the detection model includes a graph convolution model, an encoder, and a decoder, and the device includes:

[0031] The input module is used to acquire a fused image of cerebral artery data and input the fused image of cerebral artery data into the graph convolution model to obtain graph convolution features;

[0032] The decoupling module is used to acquire vascular preprocessing results and vascular knowledge fusion data, and input the vascular preprocessing results and vascular knowledge fusion data into the encoder for decoupling to obtain decoupling features;

[0033] The training module is used to fuse the graph convolutional features and decoupled features, and input the feature fusion result into the decoder. The detection model is trained under supervision using vascular spatial annotation to obtain a cerebral aneurysm detection model.

[0034] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method for establishing a cerebral aneurysm detection model as described in the first aspect or any corresponding embodiment thereof.

[0035] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the method for establishing a cerebral aneurysm detection model according to the first aspect or any corresponding embodiment described above. Attached Figure Description

[0036] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0037] Figure 1 This is a flowchart illustrating the method for establishing a cerebral aneurysm detection model according to an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the cerebral aneurysm detection model according to an embodiment of the present invention;

[0039] Figure 3 This is a flowchart illustrating another method for establishing a cerebral aneurysm detection model according to an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the encoder structure of the cerebral aneurysm detection model according to an embodiment of the present invention;

[0041] Figure 5 This is a flowchart illustrating another method for establishing a cerebral aneurysm detection model according to an embodiment of the present invention;

[0042] Figure 6 This is a flowchart illustrating the specific application of the cerebral aneurysm detection model according to an embodiment of the present invention.

[0043] Figure 7 This is a structural block diagram of a brain aneurysm detection model establishment device according to an embodiment of the present invention;

[0044] Figure 8 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] As a typical small-target detection study, cerebral aneurysm detection is currently a hot research topic. Improving network structure to realize aneurysm features is a current research hotspot. Due to the limitation of the amount of 3D image data, improving the learning efficiency of the network is an effective optimization method.

[0047] By adding SENet channel attention mechanism to the high-dimensional feature convolutional layer after downsampling of 3D U-Net, a sensitivity of 90.8% and an average of 2.49 false positives per case were achieved on a single-center dataset, and a sensitivity of 82.5% and an average of 0.88 false positives per case were achieved on a multi-center dataset. This significantly improved the detection performance while maintaining a relatively low false positive rate.

[0048] Currently, there are few studies on this type of optimization method in the field of TOF-MRA cerebral aneurysm detection, but there are several excellent research results in CTA cerebral aneurysm detection, which also uses leukemia sequences. In 3D U-Net, a residual network and a dual-channel attention module are added simultaneously, with the two channels representing channel attention and spatial attention respectively, and weighted by averaging. This significantly improves the model's detection performance, achieving a sensitivity of 97.3% and an average of 0.34 false positives per case. Adding an attention module to the ResNet-18 network, and similarly utilizing dense nonlinear convolutions and residual multi-kernel ensemble blocks to focus the network on optimizing the aneurysm target, achieves a sensitivity of 97.5% and an average of 13.8 false positives per case.

[0049] A multi-scale detection method, GLIA-Net, was proposed. It utilizes a global localization network containing global risk probability information and a high-resolution local segmentation network to combine global localization priors with local features, generating refined 3D segmentation results with a sensitivity of 96.2% and an average of 4.38 false positives per case. While attention-based and multi-scale network structures improve network performance to some extent, it is evident that existing networks have not yet achieved a deep understanding of the relationship between cerebral aneurysms and normal blood vessels, thus encountering bottlenecks in improving sensitivity and suppressing false positives.

[0050] The multi-level, related, and continuous feature information in aneurysm detection, such as the aneurysm's location, shape, size, and other basic information, as well as its differences relative to the vessel wall, the vessel segment it belongs to, and the course of the vessel it is located in, are all components of the conceptual knowledge of cerebral arteries. How to enable the network to "understand" and "recognize" the above conceptual knowledge in the convolutional feature space is a major problem that needs to be solved.

[0051] One approach involves introducing conceptual knowledge into convolutional neural networks (CNNs) through an intermediate medium. Several studies have employed methods such as first converting knowledge into a graph using image features, and then concatenating these features with a graph convolutional network (GCNN) to achieve the introduction of conceptual knowledge. For example, after discovering the effect of vascular connectivity on vascular segmentation networks, a graph attention network targeting connectivity was trained using the segmentation results to build a graph, and this was backpropagated to UNet training to achieve targeted optimization of UNet's connections. Edge features were used to build a topological graph, and the fusion of graph network features and CNN features improved the segmentation effect of retinal vessels. Furthermore, the concept of shape was used as a separate branch on top of CNNs to optimize the network's target segmentation and prediction. This method can build graphs and introduce knowledge into the network in a relatively explicit way, but the knowledge it relies on is still relatively limited and constrained by image representation methods.

[0052] Currently, methods that directly incorporate conceptual features into convolutional networks have been preliminarily studied, and are widely used in the field of zero-shot semantic segmentation. In zero-shot semantic segmentation tasks, features are divided into image space and semantic space. After dividing the image into blocks using object detection algorithms, graph convolutional networks are used to extract features from the graph features formed by each block and to globally associate the blocks. This enables the same network model to automatically classify targets while detecting them, proving the feasibility of fusing graph convolutional features with convolutional features in the feature space dimension. Following this line of thought, Hybrid Knowledge Routed Modules are proposed, which fuse image features, block classification semantic graph features, and block internal content semantic graph features to achieve highly consistent target detection in images.

[0053] Few-Shot Image Recognition, which integrates CNN features with knowledge transfer modules, directly converts textual knowledge into features using graphs and fuses them with CNN features to achieve the recognition of unlabeled samples. Building upon these studies, the knowledge introduced by graph convolutional networks, containing only a single label definition, is considered too simplistic. Therefore, a method based on dense graphs is proposed, enhancing the hierarchy of the knowledge graph and enabling the introduction of multiple labeled conceptual knowledge, thus improving the performance of graph networks in zero-shot tasks.

[0054] In summary, given the current bottlenecks in optimizing aneurysm detection networks using methods such as attention, introducing conceptual knowledge through graph convolution offers a solution. However, current knowledge-guided networks only introduce single conceptual knowledge; effective methods for introducing multi-layered, interconnected, and complex conceptual features into convolutional networks are still lacking.

[0055] According to an embodiment of the present invention, a method for establishing a cerebral aneurysm detection model is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0056] This embodiment provides a method for establishing a cerebral aneurysm detection model, which can be used on mobile terminals. Figure 1 This is a flowchart of a method for establishing a cerebral aneurysm detection model according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0057] Step S101: Obtain the fusion map of cerebral artery data and input the fusion map of cerebral artery data into the graph convolution model to obtain graph convolution features.

[0058] In this embodiment of the invention, the cerebral artery data fusion map can be obtained by performing relevant examinations on the patient, such as transcranial color Doppler ultrasound or neck ultrasound. The graph convolutional network aims at aneurysm detection, such as... Figure 2 As shown, the fusion map of cerebral artery data is used as input to obtain the corresponding graph convolution features.

[0059] Step S102: Obtain the vascular preprocessing results and vascular knowledge fusion data, and input the vascular preprocessing results and vascular knowledge fusion data into the encoder for decoupling to obtain decoupled features.

[0060] In this embodiment of the invention, the vascular preprocessing result can be the optimal vascular preprocessing result, and the vascular knowledge fusion data can be multi-level vascular knowledge fusion data. The encoder and decoder serve as the backbone network, and the input of the backbone network is the optimal vascular preprocessing result and the multi-level vascular knowledge fusion data. During the encoding stage, the input of the network is decoupled into multiple branches to obtain decoupled features.

[0061] Step S103: The graph convolutional features and decoupled features are fused, and the result of the feature fusion is input into the decoder. The detection model is trained under supervision using vascular spatial annotation to obtain the cerebral aneurysm detection model.

[0062] In this embodiment of the invention, the graph convolutional features obtained by the graph convolutional model and the decoupled features obtained by the encoder are fused one-to-one. Multi-scale annotation is used as supervision, and the fused features are input into the decoder. The detection model is trained under supervision using vascular spatial annotation. Under the premise of realizing the fusion and introduction of knowledge from outside the network, a cerebral aneurysm detection neural network model that can be guided by multi-dimensional knowledge is constructed, resulting in a cerebral aneurysm detection model based on knowledge-guided neural network, thereby achieving accurate detection of cerebral aneurysms.

[0063] It should be noted that, as Figure 2 As shown, a multi-supervised cerebral aneurysm detection model is constructed by adding graph annotation supervision to the graph convolution model, multi-scale annotation supervision to the feature fusion, fusion annotation supervision to the supervision branch of the decoder, and blood vessel annotation supervision to the instance branch.

[0064] The cerebral aneurysm detection model uses a 3D fully convolutional neural network with an encoder-decoder framework as its backbone. It establishes a multi-channel input using multi-level fused data and constructs a multi-scale vascular morphology receptive field within a multi-level vascular space using typical morphological representations of 3D vascular structural features. A multi-branch, multi-supervisory mechanism is designed using the obtained optimal knowledge representation method. A parallel graph convolutional aneurysm detection network is employed to establish a dimension-wise, layer-wise knowledge weighted fusion and supervision-guided mechanism. The overall network structure is utilized to address issues such as multi-scale feature decoupling and graph convolution-convolution feature fusion, ultimately forming a multi-scale convolutional network model of vascular space guided by graph convolutional knowledge.

[0065] The method for establishing a cerebral aneurysm detection model provided in this embodiment takes multi-level fused data as input, processes the fused cerebral artery data to obtain the graph convolutional features corresponding to the fused cerebral artery data, processes the vascular preprocessing results and vascular knowledge fused data, and establishes a cerebral aneurysm detection model with encoding and decoding as the framework. This model constructs a multi-scale network model guided by graph convolutional knowledge, and then uses the cerebral aneurysm detection model to detect cerebral aneurysms, thereby improving the accuracy of cerebral aneurysm detection results.

[0066] This embodiment provides a method for establishing a cerebral aneurysm detection model, which can be used in the aforementioned mobile terminal. Figure 3 This is a flowchart of a method for establishing a cerebral aneurysm detection model according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0067] Step S301: Using semantic cross-entropy as the loss function, train the parallel convolutional network to obtain the graph convolutional model.

[0068] In this embodiment of the invention, the graph convolutional model employs a parallel convolutional network to extract key semantic knowledge across multiple dimensions. Using semantic cross-entropy as the loss function and aneurysm detection as the objective, the convolutional network is trained to obtain a trained graph convolutional model. This graph convolutional model, obtained by training the parallel convolutional network, facilitates the processing of cerebral artery data fusion maps.

[0069] Step S302: Obtain the fusion map of cerebral artery data and input the fusion map of cerebral artery data into the graph convolution model to obtain graph convolution features.

[0070] Please see details Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0071] Step S303: Obtain cerebral artery segmentation data and vascular structure layering.

[0072] Step S304: Using the vascular centerline as the axis, establish the receptive field of the encoder based on the cerebral artery segmentation data and vascular structure layering.

[0073] In this embodiment of the invention, the receptive field is the operator region extracted and calculated by the encoder for each convolutional layer. For example... Figure 4 As shown, the receptive field of the encoder in the cerebral aneurysm detection model is established based on cerebral artery segmentation data and vascular structure layering. It is a multi-scale receptive field in the vascular space with the vascular centerline analysis as the axis. The receptive field scale includes local blood vessels, vascular segments, vascular zones, overall blood vessels, and the entire image.

[0074] Specifically, such as Figure 4 As shown, semantic cross-entropy supervision is added to the graph convolution model, standard dimensional supervision is added to the feature fusion, fused feature supervision is added to the supervised branch of the decoder, and detection task target supervision is added to the instance branch of the decoder to establish a multi-supervised model.

[0075] By constructing a multi-scale vascular morphology receptive field in a multi-layered vascular space based on cerebral artery segmentation data and vascular structure layering, the preprocessing results of blood vessels and vascular knowledge fusion data are input into the encoder for processing.

[0076] Step 305: Obtain the vascular preprocessing results and vascular knowledge fusion data, and input the vascular preprocessing results and vascular knowledge fusion data into the encoder for decoupling to obtain decoupled features.

[0077] Please see details Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0078] In some optional implementations, the encoder has a multi-branch structure, and step S305 includes:

[0079] Step S3051: Input the vascular preprocessing results and vascular knowledge fusion data into the encoder branch to obtain the corresponding hierarchical features.

[0080] Step S3052: Based on the spatial attention mechanism, the hierarchical features are transmitted and decoupled into multiple branches to obtain decoupled features.

[0081] In this embodiment of the invention, the encoder is a multi-branch long-connection structure, and continuous residual blocks are used to suppress the influence of other branches. Data is input into the encoder branches to obtain the hierarchical features corresponding to the vascular preprocessing results and the vascular knowledge fusion data.

[0082] Since the hierarchical features after multi-branch processing are shallow features, the spatial attention mechanism can be used to pass hierarchical features, which can enhance the shallow features and form a long connection mechanism for decoupling hierarchical features, so as to achieve multi-branch decoupling of hierarchical features and obtain decoupled features.

[0083] By using spatial attention mechanisms to transmit hierarchical features, the effective transmission of hierarchical features can be enhanced. Encoders are used to decouple data into multiple branches, thereby achieving long-connection mechanisms for decoupling hierarchical features, which decouples the data and facilitates the processing of decoupled features.

[0084] This embodiment provides a method for establishing a cerebral aneurysm detection model, which can be used on the aforementioned mobile terminals, such as mobile phones and tablets. Figure 5 This is a flowchart of a method for establishing a cerebral aneurysm detection model according to an embodiment of the present invention, such as... Figure 5 As shown, the process includes the following steps:

[0085] Step S501: Obtain the fused image of cerebral artery data and input the fused image of cerebral artery data into the graph convolution model to obtain graph convolution features.

[0086] Please see details Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0087] Step S502: Obtain the vascular preprocessing results and vascular knowledge fusion data, and input the vascular preprocessing results and vascular knowledge fusion data into the encoder for decoupling to obtain decoupled features.

[0088] Please see details Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0089] Step S503: The graph convolutional features and decoupled features are fused, and the result of the feature fusion is input into the decoder. The detection model is trained under supervision using multi-scale annotation to obtain the cerebral aneurysm detection model.

[0090] In some optional implementations, step S503 above includes:

[0091] Step S5031: Input the feature fusion result into the convolutional layer of the supervised branch, and input the features processed by the supervised branch into the instance branch.

[0092] Step S5032: The detection model is trained using the vascular spatial labeling loss function as supervision to obtain the cerebral aneurysm detection model.

[0093] In this embodiment of the invention, the encoder employs a dual-branch path, specifically including a supervisory branch and an instance branch. Each branch utilizes a swim-transformer structure to extract the spatial continuity features of blood vessels. The features extracted using the swim-transformer structure are continuous, making them easier to analyze compared to scattered features.

[0094] Specifically, the supervised path is supervised based on the feature fusion result, resulting in a supervised branch. First, the feature fusion result is input into convolutional layers of different dimensions in the supervised branch, and then the features processed by the supervised branch are input into the instance branch. Simultaneously, decoupled features obtained through multi-branch decoupling of hierarchical features are also input into each convolutional layer in the instance branch. With the goal of minimizing the vascular spatial annotation loss function, the detection model is trained under supervision to obtain a cerebral aneurysm detection model, ultimately achieving accurate detection of cerebral aneurysms using a knowledge-guided neural network based on TOF-MRA.

[0095] The method for establishing a cerebral aneurysm detection model provided in this embodiment establishes a cerebral aneurysm detection model based on a knowledge-guided neural network by inputting the features of the supervised branch processing into the instance branch for processing.

[0096] It should be noted that in specific applications, such as Figure 6As shown, the process involves acquiring segmented cerebral arteries, manually labeled data, and cerebral aneurysm knowledge data of the target object, which serve as inputs for the cerebral aneurysm detection model. Manual labeling of aneurysms is performed to establish the cerebral aneurysm detection model. Suspected regions are detected, cerebral aneurysms are identified, and the detection results are obtained, thus achieving cerebral aneurysm detection.

[0097] This embodiment also provides a device for establishing a cerebral aneurysm detection model. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0098] This embodiment provides a device for establishing a cerebral aneurysm detection model, such as... Figure 7 As shown, the device includes:

[0099] Input module 701 is used to acquire a fused image of cerebral artery data and input the fused image of cerebral artery data into a graph convolution model to obtain graph convolution features.

[0100] The decoupling module 702 is used to acquire the vascular preprocessing results and vascular knowledge fusion data, and input the vascular preprocessing results and vascular knowledge fusion data into the encoder for decoupling to obtain decoupling features.

[0101] Training module 703 is used to fuse graph convolutional features and decoupled features, and input the feature fusion result into the decoder. The detection model is trained under supervision using vascular spatial annotation to obtain a cerebral aneurysm detection model.

[0102] In some alternative embodiments, the device further includes:

[0103] Module 704 is obtained, which is used to train a parallel convolutional network with semantic cross-entropy as the loss function to obtain a graph convolutional model.

[0104] In some alternative embodiments, the device further includes:

[0105] The acquisition module 705 is used to acquire cerebral artery segmentation data and vascular structure layering.

[0106] Module 706 is established to create the receptive field of the encoder based on cerebral artery segmentation data and vascular structure layering, with the vascular centerline as the axis.

[0107] In some alternative implementations, the decoupling module 702 includes:

[0108] The first obtaining unit is used to input the vascular preprocessing results and vascular knowledge fusion data into the encoder branch to obtain the corresponding hierarchical features.

[0109] The second unit is used to transmit hierarchical features based on the spatial attention mechanism, and to decouple the hierarchical features through multiple branches to obtain decoupled features.

[0110] In some alternative implementations, the training module 703 includes:

[0111] The input unit is used to transmit hierarchical features based on the spatial attention mechanism, and to decouple the hierarchical features through multiple branches to obtain decoupled features.

[0112] The supervised training unit is used to supervise the training of the detection model by using vascular spatial annotation with the goal of minimizing the vascular spatial annotation loss function, so as to obtain a cerebral aneurysm detection model.

[0113] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0114] The cerebral aneurysm detection model establishment device in this embodiment is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0115] This invention also provides a computer device having the above-described features. Figure 7 The device shown is for establishing a cerebral aneurysm detection model.

[0116] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 8 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 8 Take a processor 10 as an example.

[0117] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0118] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0119] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0120] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0121] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 8 Taking the example of a connection between China and Israel via a bus.

[0122] The input device 30 can receive input numeric or character information, and generate key signal inputs related to user settings and function control of the computer device.

[0123] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0124] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for establishing a cerebral aneurysm detection model, characterized in that, The detection model includes a graph convolutional model, an encoder, and a decoder; the method includes: A fused image of cerebral artery data is obtained and input into the graph convolution model to obtain graph convolution features; The vascular preprocessing results and vascular knowledge fusion data are obtained, and the vascular preprocessing results and vascular knowledge fusion data are input into the encoder for decoupling to obtain decoupled features; The graph convolutional features and decoupled features are fused together, and the result of the feature fusion is input into the decoder. The detection model is trained under supervision using vascular spatial annotation to obtain a cerebral aneurysm detection model. Before inputting the vascular preprocessing results and vascular knowledge fusion data into the encoder, the method further includes: Acquire cerebral artery segmentation data and vascular structure layering; Using the vascular centerline as the axis, the receptive field of the encoder is established based on the cerebral artery segmentation data and vascular structure layering. The receptive field of the encoder is established based on the cerebral artery segmentation data and vascular structure layering. It is a multi-scale receptive field in vascular space with the vascular centerline analysis as the axis. The receptive field scale includes local blood vessels, blood vessel segments, blood vessel zones, overall blood vessels, and the entire image. Semantic cross-entropy supervision is added to the graph convolution model, standard dimensional supervision is added to the feature fusion, fusion feature supervision is added to the supervised branch of the decoder, and detection task target supervision is added to the instance branch of the decoder to establish a multi-supervised model.

2. The method according to claim 1, characterized in that, Before inputting the fused cerebral artery data map into the graph convolution model, the method further includes: Using semantic cross-entropy as the loss function, a parallel convolutional network is trained to obtain a graph convolutional model.

3. The method according to claim 1, characterized in that, The encoder has a multi-branch structure. The preprocessing results of blood vessels and the fusion data of blood vessel knowledge are input into the encoder for decoupling to obtain decoupling features, including: The vascular preprocessing results and vascular knowledge fusion data are input into the encoder branch to obtain the corresponding hierarchical features; Based on the spatial attention mechanism, the hierarchical features are transmitted, and the hierarchical features are decoupled into multiple branches to obtain decoupled features.

4. The method according to claim 1, characterized in that, The decoder employs a dual-branch path, which includes a supervisory branch and an instance branch. The supervisory branch is obtained as follows: Based on the result of feature fusion, the supervision path is fed back for supervision, resulting in a supervision branch.

5. The method according to claim 4, characterized in that, The process of inputting the feature fusion result into the decoder and training the detection model using multi-scale annotation as supervision to obtain a cerebral aneurysm detection model includes: The result of feature fusion is input into the convolutional layer of the supervision branch, and the features processed by the supervision branch are input into the instance branch. By utilizing vascular spatial annotation and minimizing the vascular spatial annotation loss function, a supervised training method is used to train the detection model, resulting in a cerebral aneurysm detection model.

6. A device for establishing a cerebral aneurysm detection model, characterized in that, The detection model includes a graph convolutional model, an encoder, and a decoder; the device includes: The input module is used to acquire a fused image of cerebral artery data and input the fused image of cerebral artery data into the graph convolution model to obtain graph convolution features; The decoupling module is used to acquire vascular preprocessing results and vascular knowledge fusion data, and input the vascular preprocessing results and vascular knowledge fusion data into the encoder for decoupling to obtain decoupling features; The training module is used to fuse the graph convolutional features and decoupled features, and input the feature fusion result into the decoder. The detection model is trained under supervision using vascular spatial annotation to obtain a cerebral aneurysm detection model. The device also includes: The acquisition module is used to acquire cerebral artery segmentation data and vascular structure layering. A module is established to create the receptive field of the encoder based on the cerebral artery segmentation data and vascular structure layering, with the vascular centerline as the axis. The receptive field of the encoder is a multi-scale receptive field in vascular space based on the cerebral artery segmentation data and vascular structure layering, with the vascular centerline analysis as the axis. The receptive field scale includes local blood vessels, blood vessel segments, blood vessel partitions, overall blood vessels, and the entire image. Semantic cross-entropy supervision is added to the graph convolution model, standard dimensional supervision is added to the feature fusion, fusion feature supervision is added to the supervised branch of the decoder, and detection task target supervision is added to the instance branch of the decoder to establish a multi-supervised model.

7. A computer device, characterized in that, include: The system includes a memory and a processor, which are interconnected and the memory stores computer instructions. The processor executes the computer instructions to perform the method for establishing a cerebral aneurysm detection model as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the method for establishing a cerebral aneurysm detection model according to any one of claims 1 to 5.

Images