Automatic detection method and system for abnormalities of superior mesenteric artery

Abstract

The present invention discloses an automatic detection method and system for superior mesenteric artery abnormalities, which belong to the field of medical image processing technology. The method is based on abdominal plain scan CT images. First, the regional positioning and segmentation network is used to automatically identify the superior mesenteric artery region and generate a labeling mask; the precise segmentation and abnormality prediction network is used to further achieve precise segmentation of the superior mesenteric artery, and the characteristics of the superior mesenteric artery are extracted from the segmented region to predict the probability of abnormality or normality of the superior mesenteric artery; finally, the abnormal type diagnosis network is used to classify and differentially diagnose the detected abnormal lesions. The method of the present invention can assist radiologists and clinicians in performing large-scale superior mesenteric artery abnormality detection using abdominal plain scan CT images. It not only has excellent abnormality prediction performance, but can also identify common superior mesenteric artery abnormality classifications, providing technical support for the early diagnosis of superior mesenteric artery abnormalities.

Description

Technical Field

[0001] The present invention relates to a method and system for automatically detecting abnormalities of a superior mesenteric artery, and belongs to the technical field of medical image processing. Background Art

[0002] The superior mesenteric artery, as an important branch of the abdominal aorta, directly nourishes the jejunum, ileum, cecum, ascending colon, and the right half of the transverse colon, maintaining the normal physiological function of the intestine and ensuring the stable operation of the digestive system. When abnormal lesions occur in the superior mesenteric artery, their complexity and severity cannot be ignored. These abnormalities mainly include arterial embolism, vascular wall tears due to dissection, aneurysms with potential rupture risks, and vascular inflammatory reactions caused by vasculitis. In terms of clinical manifestations, abnormalities of the superior mesenteric artery are often diverse and non-specific, which makes the diagnosis of the disease more challenging.

[0003] The application of deep learning models in aortic disease, coronary artery disease, and cervical vertebral artery disease has achieved remarkable results. These studies not only validate the potential of deep learning technology in the diagnosis of vascular diseases, but also demonstrate its advantages in improving diagnostic accuracy, reducing misdiagnosis rates, and shortening diagnostic time. However, despite the high clinical importance of visceral artery disease, the application of deep learning technology in this field is relatively rare, especially research on automatic detection systems for superior mesenteric artery anomalies. Therefore, developing a deep learning-based automatic detection method for superior mesenteric artery anomalies can not only fill the technical gap in this field, but also provide an efficient and accurate diagnostic tool for clinicians, improve the overall diagnosis and treatment of visceral artery disease, and provide patients with better medical services. Summary of the Invention

[0004] In order to achieve early identification and accurate diagnosis of superior mesenteric artery anomalies, the present invention provides an automatic detection method and system for superior mesenteric artery anomalies. The technical solution is as follows:

[0005] The present invention provides a method for automatically detecting abnormalities of the superior mesenteric artery, comprising:

[0006] Step 1: Obtain abdominal CT images;

[0007] Step 2: Using the feature pyramid backbone architecture and the universal convolution path to extract features from the abdominal CT image respectively, and superimposing the features obtained by the two paths to obtain a universal feature map;

[0008] Step 3: Use the target generation network to process the general feature map to generate the superior mesenteric artery RoI target box;

[0009] Step 4: using the target mapping layer, mapping from the general feature map to the target feature map based on the superior mesenteric artery RoI target box;

[0010] Step 5: The target feature map is processed using the mask expert branch to obtain a global feature map, enabling accurate prediction of the superior mesenteric artery region mask. The mask expert branch first expands the dimension of the target feature map through an upsampling convolutional layer, then builds a segmentation backbone architecture and introduces a parallel segmentation task-specific path. The output is the superior mesenteric artery global feature map and the predicted mask.

[0011] Step 6: Based on the target feature map and the global feature map, the superior mesenteric artery is classified using the diagnostic expert branch;

[0012] The diagnostic expert branch takes the fused feature map of the superior mesenteric artery target feature map and the global feature map as input, builds a detection backbone architecture, and introduces a parallel path dedicated to the anomaly type detection task.

[0013] Optionally, the target generation network generates a target box of the superior mesenteric artery region using a 3×3 convolution layer and two 1×1 convolution branches.

[0014] Optionally, the target mapping layer selects corresponding features for the superior mesenteric artery region target frame using bilinear interpolation from the general feature map.

[0015] Optionally, the segmentation backbone architecture of the mask expert branch is based on a ResUnet-like structure, including four downsampling layers and four upsampling layers, each of which has a residual module.

[0016] Optionally, the segmentation task-specific path takes the target feature map as input and maps the channel dimension from c to c through two consecutive 1×1 convolutions. Then, after a global pooling layer, the output is used as the weight of the linear combination of the backbone feature output to assist the mask expert branch in generating segmentation task-specific features.

[0017] Optionally, the detection backbone architecture of the diagnostic expert branch is based on a ResNet-like structure, including four sub-modules, each sub-module has a residual module, and each sub-module includes a convolutional layer, a batch normalization layer, a ReLU activation function, a convolutional layer, a batch normalization layer, a ReLU activation function, a maximum pooling layer and a residual connection layer connected in sequence.

[0018] Optionally, step 6 classifies the superior mesenteric artery into normal, dissection, aneurysm and pseudoaneurysm, thromboembolism, vasculitis, and other abnormalities.

[0019] The present invention provides a system for automatically detecting abnormalities of the superior mesenteric artery, which is applied to the method for automatically detecting abnormalities of the superior mesenteric artery as described above, and comprises:

[0020] An image acquisition module configured to acquire an abdominal CT image;

[0021] a spatial domain universal feature perception architecture configured to extract features from the abdominal CT image using a feature pyramid backbone architecture and a universal convolutional path, and to superimpose features obtained from the two paths to obtain a universal feature map;

[0022] A target generation network is configured to process the universal feature map to generate a superior mesenteric artery RoI target box;

[0023] a target mapping layer, configured to implement mapping from the general feature map to a target feature map based on the superior mesenteric artery RoI target box;

[0024] The mask expert branch is configured to process the target feature map to obtain a global feature map to accurately predict the superior mesenteric artery region mask. The mask expert branch first increases the dimension of the target feature map through an upsampling convolutional layer, then builds a segmentation backbone architecture and introduces a parallel segmentation task-specific path. The output is the superior mesenteric artery global feature map and the predicted mask.

[0025] The diagnostic expert branch is configured to classify the superior mesenteric artery based on the target feature map and the global feature map using the diagnostic expert branch; the diagnostic expert branch uses the fused feature map of the superior mesenteric artery target feature map and the global feature map as input to construct a detection backbone architecture and introduces a parallel path dedicated to the abnormality type detection task.

[0026] The present invention provides an electronic device, comprising a memory and a processor;

[0027] The memory is used to store computer programs;

[0028] The processor is configured to implement any one of the above methods for automatically detecting abnormalities of the superior mesenteric artery when executing the computer program.

[0029] The present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for automatically detecting anomalies of the superior mesenteric artery as described in any one of the above items is implemented.

[0030] The beneficial effects of the present invention are:

[0031] The present invention efficiently detects abnormalities of the superior mesenteric artery based on the input of abdominal plain scan CT images; the present invention utilizes a spatial domain universal feature perception architecture, a target generation network, a target mapping layer, a mask expert branch, and a diagnostic expert branch to construct an automatic detection method for superior mesenteric artery abnormalities, which not only has excellent abnormality prediction performance, but can also identify common superior mesenteric artery abnormality classifications; after verification and expansion of application scenarios, the automatic detection method for superior mesenteric artery abnormalities provided by the present invention can provide technical support for the early diagnosis of large-scale superior mesenteric artery abnormalities. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 A schematic diagram of the basic flow of a method for automatic detection of superior mesenteric artery anomalies based on a multi-task dense neural network provided by one embodiment of the present invention.

[0034] Figure 2 A schematic diagram of the overall structure of a method for automatic detection of superior mesenteric artery anomalies based on a multi-task dense neural network provided by one embodiment of the present invention.

[0035] Figure 3 Schematic diagram of the backbone structure of the spatial domain universal feature perception architecture of the automatic detection method of superior mesenteric artery anomalies based on a multi-task dense neural network provided by one embodiment of the present invention.

[0036] Figure 4 A schematic diagram of the backbone structure of the mask expert branch of the automatic detection method for superior mesenteric artery anomalies based on a multi-task dense neural network provided by one embodiment of the present invention.

[0037] Figure 5 A schematic diagram of the backbone submodule structure of the diagnostic expert branch of the automatic detection method for superior mesenteric artery anomalies based on a multi-task dense neural network provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0038] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0039] Example 1:

[0040] This embodiment provides a method for automatically detecting abnormalities of the superior mesenteric artery, including:

[0041] Step 1: Obtain abdominal CT images;

[0042] Step 2: Use the feature pyramid backbone architecture and the universal convolution path to extract features from the abdominal CT image respectively, and superimpose the features obtained by the two paths to obtain a universal feature map;

[0043] Step 3: Use the target generation network to process the general feature map and generate the superior mesenteric artery RoI target box;

[0044] Step 4: Use the target mapping layer to map the general feature map to the target feature map based on the superior mesenteric artery RoI target box;

[0045] Step 5: The mask expert branch processes the target feature map to obtain a global feature map, enabling accurate prediction of the superior mesenteric artery region mask. The mask expert branch first expands the dimension of the target feature map through an upsampling convolutional layer, then builds a segmentation backbone architecture and introduces a parallel segmentation task-specific path. The output is the superior mesenteric artery global feature map and the predicted mask.

[0046] Step 6: Based on the target feature map and the global feature map, the diagnostic expert branch is used to classify the superior mesenteric artery;

[0047] The diagnostic expert branch takes the fused feature map of the superior mesenteric artery target feature map and the global feature map as input, builds the detection backbone architecture, and introduces a parallel path dedicated to the anomaly type detection task.

[0048] This embodiment efficiently detects abnormalities of the superior mesenteric artery based on the input of abdominal plain scan CT images; utilizes a spatial domain general feature perception architecture, a target generation network, a target mapping layer, a mask expert branch, and a diagnostic expert branch to construct an automatic detection method for superior mesenteric artery abnormalities, which not only has excellent abnormality prediction performance but can also identify common superior mesenteric artery abnormality classifications; after verification and application scenario expansion, the automatic detection method for superior mesenteric artery abnormalities provided in this embodiment can provide technical support for the early diagnosis of large-scale superior mesenteric artery abnormalities.

[0049] Example 2:

[0050] This embodiment provides a method for automatically detecting abnormalities of the superior mesenteric artery. Figure 1-Figure 4 The specific implementation process is as follows:

[0051] Step 1: Obtain abdominal CT images and construct a dataset.

[0052] Step 11: All patients underwent multi-phase CT examination, including plain scan, arterial phase, venous phase, and delayed phase. A GEREVOLUTION ACE 64-row 128-slice spiral CT machine was used to scan the entire abdomen. All patients underwent plain scan followed by enhanced scan.

[0053] Step 12: Load the plain scan CT DICOM image into ITK-SNAP software. Using the corresponding enhanced CT image, accurately label the superior mesenteric artery, which will serve as the label for subsequent network training. All pixels within the region of interest (RoI) are set to 1, and all pixels outside the RoI are set to 0, resulting in a binary brightfield mask image.

[0054] Step 13: Perform spatial interpolation, windowing enhancement, and histogram equalization preprocessing on the abdominal CT image, and construct a training sample set and a test sample set.

[0055] Step 2: Input the abdominal CT image obtained in Step 1 into the spatial domain universal feature perception architecture. Since abnormal lesions of the superior mesenteric artery are usually a small area in CT scans, the spatial domain universal feature perception architecture is used as a preprocessing tool to extract a universal feature map containing the spatial structural characteristics and location information of the superior mesenteric artery, removing irrelevant information for subsequent multi-task analysis.

[0056] The spatial domain universal feature perception architecture is built on top of the feature pyramid backbone architecture and introduces a parallel universal convolutional path. The features of the two paths are superimposed to generate a universal feature map containing the spatial structural features and position information of the superior mesenteric artery, explicitly constructing all inter-task relationships from a global perspective.

[0057] The structure of the general feature perception architecture in the spatial domain is as follows Figure 2 and Figure 3 As shown in Figure 2, the feature pyramid backbone architecture is based on a ResNet-like structure, consisting of four upsampling layers and four downsampling layers. Through 1×1 convolutional layers and 2× upsampling layers, the downsampled features are fused layer by layer with the upsampled features to obtain a feature map with high semantic information. The general convolution path contains a simple 3×3 convolutional layer with a weight matrix of The bias matrix is Where c represents the feature dimension. Since all task features come from the common feature map, it will be optimized simultaneously by the gradients of different tasks, which helps to extract common features in all tasks.

[0058] Step 3: Input the universal feature map into the target generation network to generate the superior mesenteric artery RoI target box.

[0059] Specifically, the target generation network uses a window sliding on the general feature map to generate nine target boxes with preset aspect ratios and areas for each location through a 3×3 convolutional layer. A 1×1 convolutional layer is used to train with Softmax loss to exclude target boxes that exceed the feature map boundary to determine whether the target box actually covers the superior mesenteric artery RoI. Another 1×1 convolutional layer is used to train with SmoothL1 loss to correct the coordinates of the target box belonging to the superior mesenteric artery RoI.

[0060] Step 4: Use the target mapping layer to select the corresponding features for the superior mesenteric artery RoI target box from the general feature map through bilinear interpolation, completing the mapping of the target box from the general feature map to the target feature map. Then, accurate segmentation and anomaly prediction of the superior mesenteric artery are achieved through two branches: the mask expert branch and the diagnosis expert branch. Each branch runs a different task-specific path in parallel, and each path learns different task features.

[0061] Step 5: To more accurately predict the superior mesenteric artery RoI mask, the mask expert branch first expands the dimension of the superior mesenteric artery target feature map through an upsampling convolutional layer, then builds a segmentation backbone architecture and introduces a parallel segmentation task-specific path. The output is the superior mesenteric artery global feature map and the predicted mask.

[0062] Specifically, refer to Figure 4 The segmentation backbone architecture is based on a ResUnet-like structure, consisting of four downsampling layers and four upsampling layers. Each sampling layer has a residual module, and each upsampling layer consists of a convolutional layer, a ReLU activation function, a convolutional layer, a ReLU activation function, a maximum pooling layer, and a residual connection layer connected in sequence. Specifically, the nth downsampling layer The structure can be expressed as:

[0063]

[0064] Each downsampling layer consists of a copy cropping layer, a convolution layer, a ReLU activation function, a convolution layer, a ReLU activation function, a deconvolution layer, a ReLU activation function and a residual connection layer connected in sequence, and the nth upsampling layer f is symmetrical to the nth downsampling layer. Un The structure can be expressed as:

[0065]

[0066] The segmentation task-specific path takes the target feature map as input and maps the channel dimension from c to c through two consecutive 1×1 convolutions. This is followed by a global pooling layer, whose output is used as the weight of the linear combination of the backbone feature outputs to assist the mask expert branch in generating segmentation task-specific features.

[0067] Step 6: The diagnostic expert branch is used to implement superior mesenteric artery diagnostic classification, including six categories: normal superior mesenteric artery, dissection, aneurysm and pseudoaneurysm, thromboembolism, vasculitis, and other abnormalities. The fusion feature map of the superior mesenteric artery target feature map and the global feature map is used as input to construct the detection backbone architecture and introduce a parallel path dedicated to the abnormality type detection task.

[0068] The detection backbone architecture is based on a ResNet-like structure and consists of four submodules. Each submodule has a residual module. Each submodule consists of a convolutional layer, a batch normalization layer, a ReLU activation function, a convolutional layer, a batch normalization layer, a ReLU activation function, a maximum pooling layer, and a residual connection layer connected in sequence. Specifically, refer to Figure 5 , the nth submodule f n The structure can be expressed as:

[0069] X n =ReLU(BN(Conv(ReLU(BN(Conv(f n-1 ))))))

[0070] f n =MaxPool(X n )+f n-1

[0071] The dedicated path for the anomaly type detection task contains a weight matrix And the bias matrix is A 3×3 convolution, followed by a weight matrix of And the bias matrix is 1×1 convolution, where r represents the rank number and r<<c, indicating that these two convolutions are low-rank forms of ordinary convolutions. Low-rank convolution can reduce the computational cost of path training while retaining sufficient model expression capabilities.

[0072] Step 7: Train the automatic detection model for superior mesenteric artery anomalies. The training loss function can be described as:

[0073] L=L box +αL mask +βL cls

[0074] Among them, L box Refers to SmoothL1 loss, L cls Refers to the six-category cross entropy loss, L maskIt refers to the mean of Dice loss and binary cross entropy loss. α and β are weight factors to balance the relationship between losses. The specific formula is as follows:

[0075]

[0076] Among them, T represents the number of categories, y is a 1*T vector, the value of the position corresponding to the true label is 1, and the others are 0, feature a is a T*1 vector, S j It represents the probability that the input sample of the model belongs to the jth category, x is the numerical difference between the predicted box and the real box, v = (v x ,v y ,v w ,v h ) is the coordinate of the real frame, are the coordinates of the predicted box, x, y represent the center coordinates / upper left corner coordinates of the box, w, h represent the width and height respectively; n represents the number of samples, that is, the number of pixels in each image, p and q represent the predicted pixel values and their corresponding ground truth (GT); m = 6.

[0077] Step 8: Input the unlabeled plain scan CT images in the test sample set into the trained superior mesenteric artery abnormality automatic detection model, and output the segmentation mask of the superior mesenteric artery and the classification diagnosis results predicted by the model.

[0078] Example 3:

[0079] This embodiment provides a system for automatically detecting abnormalities of the superior mesenteric artery. The system is applied to the method for automatically detecting abnormalities of the superior mesenteric artery described in the first or second embodiment. The system includes:

[0080] An image acquisition module configured to acquire an abdominal CT image;

[0081] A spatial domain universal feature perception architecture is configured to extract features from abdominal CT images using a feature pyramid backbone architecture and a universal convolutional path, and the features obtained from the two paths are superimposed to obtain a universal feature map;

[0082] The target generation network is configured to process the universal feature map and generate the superior mesenteric artery RoI target box;

[0083] The target mapping layer is configured to implement mapping from the general feature map to the target feature map based on the superior mesenteric artery RoI target box;

[0084] The mask expert branch is configured to process the target feature map to obtain a global feature map, enabling accurate prediction of the superior mesenteric artery region mask. The mask expert branch first expands the dimension of the target feature map through an upsampling convolutional layer, then builds a segmentation backbone architecture and introduces a parallel segmentation task-specific path. The output is the superior mesenteric artery global feature map and the predicted mask.

[0085] The diagnostic expert branch is configured to classify the superior mesenteric artery based on the target feature map and the global feature map. The diagnostic expert branch uses the fused feature map of the superior mesenteric artery target feature map and the global feature map as input to build a detection backbone architecture and introduce a parallel path dedicated to the anomaly type detection task.

[0086] Example 4:

[0087] In order to verify and illustrate the technical effect of the method of the present invention, this example compares the manual and traditional clinical model evaluation methods with the method of the present invention to verify the actual effect of the method.

[0088] The traditional clinical model evaluation method screens out clinical predictive factors by performing univariate and multivariate logistic regression analysis on patients' clinical data, and constructs single logistic regression and nested logistic regression prediction models based on these factors.

[0089] A manual assessment method was used to detect potential SMA anomalies by instructing five vascular radiologists with more than 10 years of experience in vascular radiology to independently review noncontrast abdominal CT images (the entire range of sections for each patient); the window level and width could be freely adjusted, but the radiologists were blinded to any other clinical and radiological information; if at least 1 section was considered to have SMA anomaly, the case was diagnosed with SMA anomaly.

[0090] The comparison results of the method of the present invention with manual and traditional clinical model evaluation methods are shown in Table 1 below:

[0091] Table 1: Comparison results between the present invention and traditional solutions

[0092]

[0093] From the above comparison data, it can be seen that the method of the present invention is superior to the manual and traditional clinical model evaluation methods in terms of accuracy and efficiency. The traditional clinical model evaluation method is only applicable to specific images and cannot be widely promoted. The manual evaluation method and the method of the present invention have a wide range of applications, but the manual evaluation method is inferior to the method of the present invention in other data. The comparison results show that the method of the present invention can quickly, efficiently, accurately and automatically detect abnormalities in the superior mesenteric artery.

[0094] Some steps in the embodiments of the present invention may be implemented using software, and the corresponding software program may be stored in a readable storage medium, such as a CD or a hard disk.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for automatically detecting abnormalities of the superior mesenteric artery, characterized in that: The method comprises: Step 1: Obtain abdominal CT images; Step 2: Using the feature pyramid backbone architecture and the universal convolution path to extract features from the abdominal CT image respectively, and superimposing the features obtained by the two paths to obtain a universal feature map; Step 3: Use the target generation network to process the general feature map to generate the superior mesenteric artery RoI target box; Step 4: using the target mapping layer, mapping from the general feature map to the target feature map based on the superior mesenteric artery RoI target box; Step 5: The target feature map is processed using the mask expert branch to obtain a global feature map, enabling accurate prediction of the superior mesenteric artery region mask. The mask expert branch first expands the dimension of the target feature map using an upsampling convolutional layer, then builds a segmentation backbone architecture and introduces a parallel segmentation task-specific path. The output is the superior mesenteric artery global feature map and the predicted mask. Step 6: Based on the target feature map and the global feature map, the superior mesenteric artery is classified using the diagnostic expert branch; The diagnostic expert branch takes the fused feature map of the superior mesenteric artery target feature map and the global feature map as input, builds a detection backbone architecture, and introduces a parallel path dedicated to the anomaly type detection task.

2. The method for automatically detecting abnormalities of the superior mesenteric artery according to claim 1, characterized in that: The target generation network uses a 3×3 convolution layer and two 1×1 convolution branches to generate the superior mesenteric artery region target box.

3. The method for automatically detecting abnormalities of the superior mesenteric artery according to claim 1, characterized in that: The target mapping layer selects corresponding features for the superior mesenteric artery region target frame from the general feature map through bilinear interpolation.

4. The method for automatically detecting abnormalities of the superior mesenteric artery according to claim 1, characterized in that: The segmentation backbone architecture of the mask expert branch is based on a ResUnet-like structure, including four downsampling layers and four upsampling layers, each with a residual module.

5. The method for automatically detecting abnormalities of the superior mesenteric artery according to claim 1, characterized in that: The segmentation task-specific path takes the target feature map as input and maps the channel dimension from c to t through two consecutive 1×1 convolutions. Then, after a global pooling layer, the output is used as the weight of the linear combination of the backbone feature output to assist the mask expert branch in generating segmentation task-specific features.

6. The method for automatically detecting abnormalities of the superior mesenteric artery according to claim 1, characterized in that: The detection backbone architecture of the diagnostic expert branch is based on a ResNet-like structure, including four submodules, each of which has a residual module, and each submodule includes a convolutional layer, a batch normalization layer, a ReLU activation function, a convolutional layer, a batch normalization layer, a ReLU activation function, a maximum pooling layer and a residual connection layer connected in sequence.

7. The method for automatically detecting abnormalities of the superior mesenteric artery according to claim 1, characterized in that: The step 6 classifies the superior mesenteric artery into normal, dissection, aneurysm and pseudoaneurysm, thromboembolism, vasculitis, and other abnormalities.

8. An automatic detection system for abnormalities of the superior mesenteric artery, characterized in that: The system is applied to the method for automatically detecting abnormalities of the superior mesenteric artery according to any one of claims 1 to 7, and the system comprises: an image acquisition module configured to acquire an abdominal CT image; a spatial domain universal feature perception architecture configured to extract features from the abdominal CT image using a feature pyramid backbone architecture and a universal convolutional path, and to superimpose features obtained from the two paths to obtain a universal feature map; A target generation network is configured to process the universal feature map to generate a superior mesenteric artery RoI target box; a target mapping layer, configured to implement mapping from the general feature map to a target feature map based on the superior mesenteric artery RoI target box; The mask expert branch is configured to process the target feature map to obtain a global feature map to accurately predict the superior mesenteric artery region mask. The mask expert branch first increases the dimension of the target feature map through an upsampling convolutional layer, then builds a segmentation backbone architecture and introduces a parallel segmentation task-specific path. The output is the superior mesenteric artery global feature map and the predicted mask. The diagnostic expert branch is configured to classify the superior mesenteric artery based on the target feature map and the global feature map using the diagnostic expert branch; the diagnostic expert branch uses the fused feature map of the superior mesenteric artery target feature map and the global feature map as input to construct a detection backbone architecture and introduces a parallel path dedicated to the abnormality type detection task.

9. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is configured to implement the method for automatically detecting abnormalities of the superior mesenteric artery according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by the processor, the method for automatically detecting abnormalities of the superior mesenteric artery according to any one of claims 1 to 7 is implemented.

Images