Method and device for acquiring aortic dissection images based on non-enhanced CT images

Abstract

The present application discloses a method and device for obtaining aortic dissection images based on non-enhanced CT images. The method for obtaining aortic dissection images based on non-enhanced CT images includes: obtaining original image information; obtaining a preprocessing feature matrix according to the original image information; inputting the preprocessing feature matrix into a two-stage cascaded calcification inward migration recognition network to obtain a calcification inward migration recognition image; inputting the preprocessing feature matrix into an intimal flap recognition network to obtain an intimal flap recognition image; and inputting the preprocessing feature matrix into a pseudo-CTA image generation network to obtain a pseudo-CTA image. Through the method for obtaining aortic dissection images based on non-enhanced CT images of the present application, the layer and location where the dissection is located can be clearly indicated, assisting doctors in quickly confirming whether the NCCT-related layer contains a suspected dissection.

Description

Technical Field

[0001] The present application relates to the technical field of image recognition, and in particular, to a method for obtaining aortic dissection images based on non-enhanced CT images and an apparatus for obtaining aortic dissection images based on non-enhanced CT images. Background Art

[0002] Aortic dissection refers to the entry of blood in the aortic lumen from the tear of the aortic intima into the aortic media, causing the separation of the media and extending along the long axis of the aorta to form a state of separation between the true and false cavities of the aortic wall. This disease is rare, with an annual incidence of one in 100,000 to one in 200,000, and the peak age is 50-70 years old, with a male-female ratio of about 2-3:1. 65%-70% die of cardiac tamponade, arrhythmia, etc. in the acute phase, so early diagnosis and treatment are very necessary.

[0003] The diagnosis of aortic dissection mainly relies on imaging examinations. Commonly used clinical screening methods include transthoracic echocardiogram (TTE), CT angiography (CTA), and magnetic resonance imaging (MRI). CT angiography (CTA) is a non-invasive and effective vascular imaging technique widely used in the diagnosis of vascular diseases such as aortic aneurysm and dissection. However, CTA requires intravascular injection of iodine contrast agent (ICA), which poses risks to patients with iodine allergy and renal insufficiency. In addition, compared with non-contrast CT (NCCT), CTA is expensive and time-consuming, especially in grass-roots areas lacking technology and support, and the application of 24-hour CTA is limited.

[0004] Identifying dissection from non-contrast CT is difficult for doctors with limited experience at the grass-roots level and is prone to missed diagnosis. Therefore, it is meaningful to develop a method for automatically detecting high-risk signs of dissection on NCCT from non-contrast CT (NCCT) images, which can reduce the missed diagnosis of aortic dissection.

[0005] Therefore, it is desirable to have a technical solution to solve or at least mitigate the above deficiencies of the prior art. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for obtaining aortic dissection images based on non-enhanced CT images to at least solve one of the above technical problems.

[0007] The present invention provides the following solutions:

[0008] According to one aspect of the present invention, there is provided a method for obtaining aortic dissection images based on non-enhanced CT images, the method for obtaining aortic dissection images based on non-enhanced CT images comprising:

[0009] Obtaining original image information;

[0010] Preprocess the original image information to obtain a preprocessed feature matrix;

[0011] Obtain a two - stage cascaded calcification inward migration recognition network, an intimal flap recognition network, and a pseudo - CTA image generation network;

[0012] Input the preprocessed feature matrix into the two - stage cascaded calcification inward migration recognition network to obtain a calcification inward migration recognition image;

[0013] Input the preprocessed feature matrix into the intimal flap recognition network to obtain an intimal flap recognition image;

[0014] Input the preprocessed feature matrix into the pseudo - CTA image generation network to obtain a pseudo - CTA image.

[0015] Optionally, the preprocessing of the original data to obtain a preprocessing matrix includes:

[0016] Perform operations such as windowing and voxel spacing normalization on the original data.

[0017] Optionally, the two - stage cascaded neural network includes:

[0018] The first - stage NCCT aortic segmentation network, and the first - stage NCCT aortic segmentation network adopts the nnUnet architecture;

[0019] The second - stage calcification inward migration segmentation network, and the output of the first - stage NCCT aortic segmentation network is used as the input of the second - stage calcification inward migration segmentation network; where

[0020] For the z - direction convolution kernel of the network backbone layer of the two - stage cascaded neural network, the downsampling step size is changed to 1.

[0021] Optionally, the intimal flap recognition network includes a 3D segmentation network and a 2D segmentation network.

[0022] Optionally, the pseudo - CTA image generation network includes a generator, a discriminator, and a registration network. The original image information passes through the generator and is converted into a pseudo - CTA image. The original image information and the pseudo - CTA image are respectively sent to the discriminator to obtain results.

[0023] This application also provides an aortic dissection image acquisition device based on non - enhanced CT images, characterized in that the aortic dissection image acquisition device based on non - enhanced CT images includes:

[0024] An original image information acquisition module, and the original image information acquisition module is used to acquire original image information;

[0025] Preprocessing feature matrix acquisition module, which is used to preprocess the original image information to obtain a preprocessing feature matrix;

[0026] Network acquisition module, which is used to acquire a two-stage cascaded calcification inward displacement recognition network, an intimal flap recognition network, and a pseudo-CTA image generation network;

[0027] Calcification inward displacement recognition image acquisition module, which is used to input the preprocessing feature matrix into the two-stage cascaded calcification inward displacement recognition network to obtain a calcification inward displacement recognition image;

[0028] Intimal flap recognition image acquisition module, which is used to input the preprocessing feature matrix into the intimal flap recognition network to obtain an intimal flap recognition image;

[0029] Pseudo-CTA image acquisition module, which is used to input the preprocessing feature matrix into the pseudo-CTA image generation network to obtain a pseudo-CTA image.

[0030] Through the method for acquiring aortic dissection images based on non-enhanced CT images of the present application, the layer and location where the dissection is located can be clearly indicated, assisting doctors to quickly confirm whether the NCCT-related layer contains a suspected dissection. Description of the Drawings

[0031] Figure 1 It is a schematic flowchart of the method for acquiring aortic dissection images based on non-enhanced CT images in an embodiment of the present application.

[0032] Figure 2 It is a schematic structural diagram of an electronic device in an embodiment of the present application.

[0033] Figure 3 It is a schematic diagram of the aortic image of a patient with positive dissection in the prior art.

[0034] Figure 4 It is a schematic structural diagram of the two-stage cascaded calcification inward displacement recognition network of the method for acquiring aortic dissection images based on non-enhanced CT images of the present application.

[0035] Figure 5 It is a schematic structural diagram of the intimal flap recognition network of the method for acquiring aortic dissection images based on non-enhanced CT images of the present application.

[0036] Figure 6 It is a schematic structural diagram of the pseudo-CTA image generation network of the method for acquiring aortic dissection images based on non-enhanced CT images of the present application. Detailed Description of the Invention

[0037] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0038] Figure 1 It is a schematic flowchart of a method for obtaining an aortic dissection image based on non-enhanced CT images in an embodiment of the present application.

[0039] As Figure 1 shown, the method for obtaining an aortic dissection image based on non-enhanced CT images includes:

[0040] Step 1: Obtain original image information;

[0041] Step 2: Preprocess the original image information to obtain a preprocessing feature matrix;

[0042] Step 3: Obtain a two-stage cascaded calcification inward displacement recognition network, an intimal flap recognition network, and a pseudo-CTA image generation network;

[0043] Step 4: Input the preprocessing feature matrix into the two-stage cascaded calcification inward displacement recognition network to obtain a calcification inward displacement recognition image;

[0044] Step 5: Input the preprocessing feature matrix into the intimal flap recognition network to obtain an intimal flap recognition image;

[0045] Step 6: Input the preprocessing feature matrix into the pseudo-CTA image generation network to obtain a pseudo-CTA image.

[0046] Through the method for obtaining an aortic dissection image based on non-enhanced CT images of the present application, the layer and position where the dissection is located can be clearly prompted, assisting doctors to quickly confirm whether the NCCT-related layer contains a suspected dissection.

[0047] In this embodiment, the preprocessing of the original data to obtain a preprocessing matrix includes:

[0048] Perform operations such as windowing and voxel spacing normalization on the original data.

[0049] In this embodiment, the windowing operation refers to first setting the Hu value range of the original plain CT image to a specific window width and window level (in this patent, the mediastinal window width is 500 and the window level is 50) to enhance the contrast of the image within a specific gray scale range.

[0050] Voxel spacing normalization refers to scaling the original plain scan CT image to a uniform voxel spacing. This patent uses a pixel spacing of 1 mm (x, y directions) and a layer spacing of 5 mm (z direction) for thick-slice CT.

[0051] See also Figure 4 In this embodiment, the two-stage cascade neural network includes a first-stage NCCT aorta segmentation network and a second-stage calcification inward migration segmentation network, wherein,

[0052] The first-level NCCT aorta segmentation network adopts the nnUnet architecture;

[0053] The output of the first-level NCCT aorta segmentation network is used as the input of the second-level calcification inward migration segmentation network;

[0054] The downsampling step size of the z-direction convolution kernel of the backbone layer of the two-stage cascade neural network is changed to 1.

[0055] In this embodiment, based on NCCT, this application performs large-scale voxel-level calcification inward migration annotation and trains a 3D segmentation network. The network architecture is as follows: Figure 4 As shown in the figure, aortic calcification inward migration detection is implemented using a deep learning-based segmentation network. This network consists of a two-stage cascade: the first stage is the NCCT aortic segmentation network, and the second stage is the calcification inward migration segmentation network. In this cascaded network architecture, the first-stage output of the aortic contour provides supporting information for the second-stage network to determine inward migration, effectively reducing false positives for calcification inward migration.

[0056] The calcification inward migration segmentation network uses a two-stage cascade structure, rather than a single stage directly following each calcification. This advantage lies in the fact that the output of the first-stage NCCT aortic segmentation network provides the second-stage network with important aortic lumen distance information, making it easier for the second-stage network to identify whether the calcification is close to the aortic wall or within the aortic lumen. Conversely, the second-stage network needs to learn both the morphological characteristics of the calcification and the range constraints of the aortic lumen, which is not conducive to calcification inward migration detection.

[0057] The first-level network uses the nnUnet architecture, and the second-level network uses the ConvNeXt architecture. For thick-slice input images, the following improvements are made: the z-direction convolution kernel in the network's stem layer has its downsampling stride changed to 1. This allows the network's stem layer to downsample only in the x and y axes, not in the z direction. This fully preserves z-direction information and avoids oversampling in thick-slice images, which can obscure calcified lesions.

[0058] In this embodiment, the endocardial sheet recognition network includes a 3D segmentation network and a 2D segmentation network.

[0059] In this embodiment, the present application relies on paired NCCT and CTA data to label a large number of dissection intimal flaps and trains a 3D and 2D fusion segmentation network. The network architecture is as follows Figure 5 shown. Different from calcification, it is difficult to directly identify the intimal flap on NCCT images. Therefore, it is necessary to collect paired NCCT and CTA data of the same patient for intimal flap labeling.

[0060] To distinguish real intimal flaps from various artifacts and false positive signs caused by bone volume effects, the network needs to refer to distant image context information. Making full use of the image continuity of adjacent slices is beneficial to accurately distinguish real intimal flaps from false positive intimal flaps. In this case, the 3D network has more advantages than the 2D network (axial slices). On the other hand, considering that thick-slice (about 5 mm) plain CT is widely used clinically, and its slice spacing is significantly larger than the pixel spacing of axial slices. To achieve accurate segmentation of the axial contour of the intimal flap, it is necessary to learn the detailed features of high-resolution axial slice images. Considering the above two aspects, this patent uses a 3D + 2D fusion model to achieve intimal flap segmentation.

[0061] In this embodiment, the pseudo-CTA image generation network includes a generator and a discriminator. The real NCCT image passes through the "generator" and is converted into a "pseudo-CTA" image. The "real CTA" and "pseudo-CTA" are respectively sent into the discriminator, and through adversarial loss, the "authenticity" of the pseudo-CTA image is improved. Due to relative displacement caused by reasons such as patient breathing, paired CTA and NCCT cannot correspond pixel by pixel. To use the "pseudo" CTA generated by pixel-by-pixel supervision of the CTA image, a registration network is introduced. At the same time, a "registration loss" is used to constrain the deformation field of the registration transformation to make the deformation field as smooth as possible. After registering the "pseudo-CTA" image to the real CTA, a pixel-by-pixel loss function can be used to supervise the training of the generator network and the registration network, such as the L1 and L2 loss functions.

[0062] Specifically, the discriminator of the present application consists of 4 convolutional layers, 1 fully connected layer, and 1 pooling layer. Both the generator and the registration network are U-shaped networks (U-Net). Among them, the generator contains 2 encoding paths, 1 bottleneck layer, 2 decoding paths, and 1 output layer. The registration network is a ResUnet network structure, which contains 7 encoding paths, 1 bottleneck layer, 7 decoding paths, and 1 output layer.

[0063] In this embodiment, the U-Net network includes the following parts:

[0064] (1) Contracting Path: It consists of a series of convolutional layers and max-pooling layers. Each convolutional block contains two 3x3 convolutional layers, and each convolutional layer is followed by a ReLU activation function. After each convolutional block, there is a 2x2 max-pooling layer for downsampling (usually with a stride of 2). The contracting path gradually reduces the spatial dimension of the feature map while increasing the number of channels of the feature map.

[0065] (2) Bottleneck Layer: Located at the bottom of the U-shaped network, connecting the contracting path and the expansive path. It contains two 3x3 convolutional layers and a ReLU activation function, without pooling operations.

[0066] The number of channels of the feature map here is further increased compared to the contracting path, and the features are more abundant.

[0067] (3) Expansive Path: Symmetric to the contracting path, it consists of a series of upsampling layers and convolutional layers. Each upsampling block first performs a 2x2 upsampling (transposed convolution) operation to restore the spatial dimension. After upsampling, the feature map is concatenated (skip connection) with the corresponding feature map of the contracting path in the channel dimension. The concatenated feature map is processed by two 3x3 convolutional layers, and each convolutional layer is followed by a ReLU activation function.

[0068] (4) Output Layer: The final convolutional layer, using 1x1 convolution to reduce the number of channels to the required number of classes. The finally output feature map has the same spatial dimension as the input image, and each pixel corresponds to a class probability. The present invention designs and trains a pseudo-CTA image generation network relying on paired NCCT and CTA data. In real NCCT images, the imaging features of the dissection are usually not obvious. The pseudo-CTA highlights the imaging signs of aortic dissection visible on NCCT in a form similar to that of angiography.

[0069] The training process of the pseudo-CTA image generation network. The network architecture is as Figure 6As shown in the figure. The real NCCT images are converted into "pseudo-CTA" images through the "generator". Based on the idea of the generative adversarial network, the "real CTA" and "pseudo-CTA" are respectively fed into the discriminator, and through the adversarial loss, the "authenticity" of the pseudo-CTA images is improved. Due to relative displacements caused by reasons such as patient breathing, the paired CTA and NCCT do not correspond pixel by pixel. In order to use the CTA images to supervise the generation of "pseudo" CTA pixel by pixel, a registration network is introduced. At the same time, the "registration loss" is used to constrain the deformation field of the registration transformation, making the deformation field as smooth as possible. For the "pseudo-CTA" images registered to the real CTA, pixel-by-pixel loss functions can be used to supervise the training of the generator network and the registration network, such as L1 and L2 loss functions.

[0070] In the pseudo-CTA generation module, taking the recognition results of "NCCT calcification inward shift" and "NCCT intimal flap" as the constraint conditions of the generator has the following advantages: (1) Compared with unsupervised image conversion training, "NCCT calcification inward shift" and "NCCT intimal flap" are discriminative networks, and supervised training is adopted, and the reliability and interpretability of their detection signs are stronger. (2) The generator can combine the results of the above models and the NCCT images, making the generated images more conform to the morphology of real CTA.

[0071] The present application also provides an aortic dissection image acquisition device based on non-enhanced CT images. The aortic dissection image acquisition device based on non-enhanced CT images includes an original image information acquisition module, a preprocessing feature matrix acquisition module, a network acquisition module, a calcification inward shift recognition image acquisition module, an intimal flap recognition image acquisition module, and a pseudo-CTA image acquisition module, wherein,

[0072] The original image information acquisition module is used to acquire original image information;

[0073] The preprocessing feature matrix acquisition module is used to preprocess the original image information to obtain a preprocessing feature matrix;

[0074] The network acquisition module is used to acquire a two-stage cascaded calcification inward shift recognition network, an intimal flap recognition network, and a pseudo-CTA image generation network;

[0075] The calcification inward shift recognition image acquisition module is used to input the preprocessing feature matrix into the two-stage cascaded calcification inward shift recognition network to obtain a calcification inward shift recognition image;

[0076] The intimal flap recognition image acquisition module is used to input the preprocessing feature matrix into the intimal flap recognition network to obtain an intimal flap recognition image;

[0077] The pseudo-CTA image acquisition module is used to input the preprocessing feature matrix into the pseudo-CTA image generation network to obtain a pseudo-CTA image.

[0078] Figure 2 It is a block diagram of an electronic device structure provided by one or more embodiments of the present invention.

[0079] As Figure 2 shown, the present application also discloses an electronic device, including: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus; a computer program is stored in the memory. When the computer program is executed by the processor, the processor executes the steps of the method for obtaining aortic dissection images based on non-enhanced CT images.

[0080] The present application also provides a computer-readable storage medium, which stores a computer program executable by an electronic device. When the computer program runs on the electronic device, it can implement the steps of the method for obtaining aortic dissection images based on non-enhanced CT images.

[0081] The communication bus mentioned in the above electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0082] The electronic device includes a hardware layer, an operating system layer running on the hardware layer, and an application layer running on the operating system. The hardware layer includes hardware such as a Central Processing Unit (CPU), a Memory Management Unit (MMU), and a memory. The operating system can be any one or more computer operating systems that implement the control of the electronic device through a process. For example, the Linux operating system, the Unix operating system, the Android operating system, the iOS operating system, or the windows operating system, etc. And in the embodiments of the present invention, the electronic device can be a handheld device such as a smart phone or a tablet computer, or an electronic device such as a desktop computer or a portable computer. The embodiments of the present invention do not particularly limit this.

[0083] The execution subject of the electronic device control in the embodiments of the present invention can be an electronic device, or a functional module in the electronic device that can call and execute a program. The electronic device can obtain the firmware corresponding to the storage medium, and the firmware corresponding to the storage medium is provided by the supplier. The firmware corresponding to different storage media can be the same or different, which is not limited herein. After the electronic device obtains the firmware corresponding to the storage medium, it can write the firmware corresponding to the storage medium into the storage medium. Specifically, it burns the firmware corresponding to the storage medium into the storage medium. The process of burning the firmware into the storage medium can be implemented by the prior art and will not be elaborated in the embodiments of the present invention.

[0084] The electronic device can also obtain the reset command corresponding to the storage medium, and the reset command corresponding to the storage medium is provided by the supplier. The reset commands corresponding to different storage media can be the same or different, which is not limited herein.

[0085] At this time, the storage medium of the electronic device is the storage medium written with the corresponding firmware. The electronic device can respond to the reset command corresponding to the storage medium in the storage medium written with the corresponding firmware. Thus, the electronic device resets the storage medium written with the corresponding firmware according to the reset command corresponding to the storage medium. The process of resetting the storage medium according to the reset command can be implemented by the prior art and will not be elaborated in the embodiments of the present invention.

[0086] For the convenience of description, when describing the above device, various units and modules are described separately according to their functions. Of course, when implementing the present application, the functions of each unit and module can be implemented in the same or multiple software and / or hardware.

[0087] Those skilled in the art of the present technology can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the art in the field to which the present invention belongs. It should also be understood that those terms defined in a general dictionary, such as those, should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless specifically defined.

[0088] For the method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the embodiments of the present invention are not limited by the described action sequence, because according to the embodiments of the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0089] As can be seen from the description of the above embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An aortic dissection image acquisition method based on non-enhanced CT images, characterized in that, The method for obtaining aortic dissection images based on non-enhanced CT images includes: Obtaining original image information; Preprocessing the original image information to obtain a preprocessing feature matrix; Obtaining a two-stage cascaded calcification inward migration recognition network, an intimal flap recognition network, and a pseudo-CTA image generation network; Inputting the preprocessing feature matrix into the two-stage cascaded calcification inward migration recognition network to obtain a calcification inward migration recognition image; Inputting the preprocessing feature matrix into the intimal flap recognition network to obtain an intimal flap recognition image; Inputting the preprocessing feature matrix into the pseudo-CTA image generation network to obtain a pseudo-CTA image; The preprocessing of the original data to obtain a preprocessing matrix includes: Performing windowing and voxel spacing normalization operations on the original data; The two-stage cascaded neural network includes: The first-stage NCCT aortic segmentation network, and the first-stage NCCT aortic segmentation network adopts the nnUnet architecture; The second-stage calcification inward migration segmentation network, and the output of the first-stage NCCT aortic segmentation network serves as the input of the second-stage calcification inward migration segmentation network; among them, For the z-direction convolutional kernel of the network backbone layer of the two-stage cascaded neural network, the downsampling step size is changed to 1; Using the registration loss to constrain the registration transformation deformation field, for the pseudo-CTA image after registration to the real CTA, using a per-pixel loss function to supervise the training of the generator network and the registration network; In the pseudo-CTA generation module, the NCCT calcification inward migration and NCCT intimal flap recognition results are used as the constraint conditions for the generator.

2. The method for acquiring aortic dissection images based on non-enhanced CT images according to claim 1, wherein, The intimal flap recognition network includes a 3D segmentation network and a 2D segmentation network.

3. The method for acquiring aortic dissection images based on non-enhanced CT images according to claim 2, wherein The pseudo-CTA image generation network includes a generator, a discriminator, and a registration network. The original image information passes through the generator and is converted into a pseudo-CTA image. The original image information and the pseudo-CTA image are respectively sent to the discriminator to obtain the results.

4. An aortic dissection image acquisition device based on non-enhanced CT images, characterized in that, The device for obtaining aortic dissection images based on non-enhanced CT images includes: An original image information acquisition module, which is used to acquire original image information; A preprocessing feature matrix acquisition module, which is used to preprocess the original image information to obtain a preprocessing feature matrix; A network acquisition module, which is used to acquire a two-stage cascaded calcification inward migration recognition network, an intimal flap recognition network, and a pseudo-CTA image generation network; A calcification inward migration recognition image acquisition module, which is used to input the preprocessing feature matrix into the two-stage cascaded calcification inward migration recognition network to obtain a calcification inward migration recognition image; An intimal flap recognition image acquisition module, which is used to input the preprocessing feature matrix into the intimal flap recognition network to obtain an intimal flap recognition image; A pseudo-CTA image acquisition module, which is used to input the preprocessing feature matrix into the pseudo-CTA image generation network to obtain a pseudo-CTA image; The preprocessing of the original data to obtain a preprocessing matrix includes: Perform windowing and voxel spacing normalization operations on the original data; The two-stage cascaded neural network includes: The first-stage NCCT aortic segmentation network, which adopts the nnUnet architecture; The second-stage calcification inward migration segmentation network, and the output of the first-stage NCCT aortic segmentation network is used as the input of the second-stage calcification inward migration segmentation network; among them, For the z-direction convolutional kernel of the network backbone layer of the two-stage cascaded neural network, the downsampling step size is changed to 1; Use the registration loss to constrain the registration transformation deformation field. For the pseudo-CTA image after registration to the real CTA, use a pixel-by-pixel loss function to supervise the training of the generator network and the registration network; In the pseudo-CTA generation module, use the NCCT calcification inward migration and NCCT intimal flap recognition results as the constraint conditions for the generator.

Images