Method and system for synthesizing CT angiography image based on generative adversarial network
By generating CT angiographic images of adversarial networks, the use of contrast agents in traditional CTPA examinations is solved, and safety and efficiency are improved, ensuring spatial alignment and feature fidelity of the image.
Patent Information
- Application Number
- CN202510460906.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-14
- Publication Date
- 2025-07-22
AI Technical Summary
Traditional CTPA examination requires injection of contrast agents, which increases radiation exposure and examination complexity of patients and may trigger allergic reactions, resulting in increased examination costs and time.
The method of synthesizing CT angiographic images by generating adversarial networks is adopted, and CTPA images are synthesized by scanning CT images on a flat basis. Generative adversarial networks, registration networks, spatial transformation networks and multi-scale discriminant networks are used to integrate multi-dimensional information to ensure spatial alignment and feature fidelity of the image.
Without affecting the accuracy of diagnosis, the safety and efficiency of the examination are improved, the use of contrast agents is reduced, and the radiation exposure and examination complexity is reduced.
Smart Images
Figure CN120355807A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical image processing, and particularly relates to a method and system for synthesizing CT angiography images based on a generative adversarial network. Background Art
[0002] CTPA, namely Computed Tomography Pulmonary Angiography, is a medical imaging examination technique specifically used to evaluate the vascular conditions of the pulmonary artery and its branches, and is particularly valuable in the diagnosis of pulmonary embolism. However, traditional CTPA examinations require the injection of contrast agents, which increases the patient's radiation exposure and examination complexity, and may also cause adverse reactions such as contrast agent allergy. In addition, the injection of contrast agents increases the examination cost and time.
[0003] Non-contrast CT, also known as plain CT or non-enhanced CT scan, is a commonly used medical imaging examination technique that is widely used in the examination of various parts of the body, mainly for detecting space-occupying lesions (such as tumors, cysts, hematomas, etc.), vascular diseases, etc. Non-contrast CT examinations do not require the injection of contrast agents (such as iodine agents), so they are relatively simple and fast, and patients do not need to bear the risks such as allergy caused by contrast agents.
[0004] Therefore, a method for synthesizing CTPA images based on non-contrast CT images is explored in order to improve the safety and efficiency of examinations without affecting the diagnostic accuracy. Summary of the Invention
[0005] Aiming at the deficiencies of the above-mentioned prior art, the present invention aims to provide a method and system for synthesizing CT angiography images based on a generative adversarial network, so as to synthesize CTPA images from non-contrast CT images and improve the examination safety and examination efficiency.
[0006] To solve the above problems, the present invention adopts the following technical solutions:
[0007] On the one hand, the present invention provides a method for synthesizing CT angiography images based on a generative adversarial network, including:
[0008] Obtaining paired non-contrast CT sequences and CTPA sequences as source modalities and performing preprocessing to obtain input sources;
[0009] Inputting the input sources into an image synthesis model and training the model using a generative adversarial network. The model training process includes:
[0010] Extract the anatomical structure features in the non-contrast CT image INCCT and the vascular enhancement features in the corresponding contrast-enhanced CTPA image ICTPA from the input source through a generation network to generate a synthetic CTPA image;
[0011] Input the synthetic CTPA image and the corresponding contrast-enhanced CTPA image ICTPA into a registration network to obtain an anatomical structure registration information matrix;
[0012] Input the synthetic CTPA image and the anatomical structure registration information matrix into a spatial transformation network to obtain a registered synthetic CTPA image;
[0013] Extract the vascular regions in the registered synthetic CTPA image and the corresponding contrast-enhanced CTPA image ICTPA respectively through a vascular segmentation model to obtain a pair of segmented vascular segmentation images;
[0014] Input the synthetic CTPA image through a multi-scale discriminative network, calculate the probability that the synthetic CTPA image is the corresponding contrast-enhanced CTPA image ICTPA, and guide the generation network to reduce the difference between the synthetic CTPA image and the corresponding contrast-enhanced CTPA image ICTPA;
[0015] Calculate the probability that the synthetic CTPA image is not the corresponding CTPA image and the probability that the synthetic CTPA image is the corresponding contrast-enhanced CTPA image ICTPA through a multi-scale discriminative network, and feedback to the multi-scale discriminative network to optimize the recognition ability;
[0016] Input the non-contrast CT image into a trained image synthesis model to generate a synthetic CTPA image.
[0017] As an implementable manner, the loss function of the image synthesis model includes the loss of the generation network, the loss of the registration network, the adversarial loss of the generation network, and the unsupervised loss;
[0018] Calculate the sum of the four losses and perform backpropagation to update the parameters of the generation network and the registration network;
[0019] And calculate the loss of the multi-scale discriminative network and update the parameters of the multi-scale network;
[0020] Use the Adam optimizer to update the parameters of the generation network, the registration network, and the multi-scale discriminative network.
[0021] As an implementable manner, the loss function \(L\) of the generation network sm :
[0022]
[0023] where E(·) is the mean of the registration information matrix Tr, and are the horizontal gradient and the vertical gradient respectively;
[0024] The loss function L of the registration network sr is:
[0025]
[0026] where N is the number of samples in each batch for training, and y i is the i-th sample in the enhanced CTPA image ICTPA, is the i-th sample in the registered synthetic CTPA image;
[0027] The loss function L of the adversarial loss of the generation network adv is:
[0028]
[0029] where N is the number of samples in each batch for training, p i and q i are the two outputs of the multi-scale discriminator network, and p i and q i are the probabilities that the calculated synthetic CTPA image is the corresponding enhanced CTPA image ICTPA;
[0030] The loss function L of the unsupervised loss of the generation network nsr is:
[0031]
[0032] where N is the number of samples in each batch for training, and z i is the i-th sample of the vascular segmentation image of the enhanced CTPA image ICTPA in the pair of segmented vascular segmentation images, is the i-th sample of the vascular segmentation image of the registered synthetic CTPA image in the pair of segmented vascular segmentation images.
[0033] As an implementable manner, the loss function L of the multi-scale discriminator network b is:
[0034] L b =(L SynCTPA +L ICTPA ) / 2
[0035] where the formula of L SynCTPA is as follows:
[0036]
[0037] Among them, N is the number of samples in each batch for training, and m i , n i are the two outputs of the multi-scale discriminative network, and m i and n i are used to calculate the probability that the synthesized CTPA image is not the corresponding CTPA image;
[0038] L ICTPA The formula of is as follows:
[0039]
[0040] Among them, N is the number of samples in each batch for training, and s i , t i are the outputs of the multi-scale discriminative network, and s i , t i are used to calculate the probability that the synthesized CTPA image is the corresponding enhanced CTPA image ICTPA.
[0041] As an implementable manner, after the synthesized CTPA image is generated, it is subjected to inverse normalization, smoothing processing, and artifact removal, and then saved as an image in DICOM format.
[0042] On the other hand, the present invention provides a system for synthesizing CT angiography images based on a generative adversarial network, including a data acquisition module, a data processing module, and a data output module;
[0043] The data acquisition module is used to obtain paired non-contrast CT sequences and CTPA sequences as source modalities and perform preprocessing to obtain an input source;
[0044] The data processing module is used to input the input source into an image synthesis model, and train the model using a generative adversarial network. The model training process includes:
[0045] Extract the anatomical structure features in the non-contrast CT image INCCT and the vascular enhancement features in the corresponding enhanced CTPA image ICTPA in the input source through a generative network to generate a synthesized CTPA image;
[0046] Input the synthesized CTPA image and the corresponding enhanced CTPA image ICTPA into a registration network to obtain an anatomical structure registration information matrix;
[0047] Input the synthesized CTPA image and the anatomical structure registration information matrix into a spatial transformation network to obtain a registered synthesized CTPA image;
[0048] The vascular regions in the registered synthetic CTPA image and the corresponding enhanced CTPA image ICTPA are respectively extracted through a vascular segmentation model to obtain a pair of segmented vascular segmentation images;
[0049] Through a multi-scale discriminant network, the synthetic CTPA image is input, and the probability that the synthetic CTPA image is the corresponding enhanced CTPA image ICTPA is calculated to guide the generation network to reduce the difference between the synthetic CTPA image and the corresponding enhanced CTPA image ICTPA;
[0050] Through a multi-scale discriminant network, the probability that the synthetic CTPA image is not the corresponding CTPA image and the probability that the synthetic CTPA image is the corresponding enhanced CTPA image ICTPA are calculated, and the feedback is used to optimize the recognition ability of the multi-scale discriminant network;
[0051] The data output module is used to input the plain CT image into the trained image synthesis model to generate a synthetic CTPA image.
[0052] As an implementable manner, the loss function of the image synthesis model includes the loss of the generation network, the loss of the registration network, the adversarial loss of the generation network, and the unsupervised loss;
[0053] Calculate the sum of the four losses and perform backpropagation to update the parameters of the generation network and the registration network;
[0054] And calculate the loss of the multi-scale discriminant network and update the parameters of the multi-scale network;
[0055] Use the Adam optimizer to update the parameters of the generation network, the registration network, and the multi-scale discriminant network.
[0056] As an implementable manner, the loss function L of the generation network sm :
[0057]
[0058] where E(·) is the mean of the registration information matrix Tr, and are the horizontal gradient and the vertical gradient respectively;
[0059] The loss function L of the registration network sr :
[0060]
[0061] where N is the number of samples in each batch for training, and y i is the i-th sample in the enhanced CTPA image ICTPA, The i-th sample in the registered synthetic CTPA image;
[0062] The loss function L of the adversarial loss of the generation network adv :
[0063]
[0064] where N is the number of samples in each batch for training, p i and q i are the two outputs of the multi-scale discriminator network, p i and q i are the probabilities that the computed synthetic CTPA image is the corresponding enhanced CTPA image ICTPA;
[0065] The loss function L of the unsupervised loss of the generation network nsr :
[0066]
[0067] where N is the number of samples in each batch for training, z i is the i-th sample of the vascular segmentation image of the enhanced CTPA image ICTPA in the pair of segmented vascular segmentation images, is the i-th sample of the vascular segmentation image of the registered synthetic CTPA image in the pair of segmented vascular segmentation images.
[0068] As an implementable manner, the loss function L of the multi-scale discriminator network b :
[0069] L b =(L SynCTPA +L ICTPA ) / 2
[0070] where the formula of L SynCTPA is as follows:
[0071]
[0072] where N is the number of samples in each batch for training, m i , n i are the two outputs of the multi-scale discriminator network, m i and n i are the probabilities that the computed synthetic CTPA image is not the corresponding CTPA image;
[0073] The formula of L ICTPA is as follows:
[0074]
[0075] Among them, N is the number of samples in each batch for training, s i and t i are the outputs of the multi-scale discriminant network, and s i and t i are the probabilities of calculating that the synthesized CTPA image is the corresponding enhanced CTPA image ICTPA.
[0076] As an implementable manner, after generating the synthesized CTPA image, it undergoes inverse normalization, smoothing processing, and artifact removal, and is saved as an image in DICOM format.
[0077] The beneficial effects of the present invention are as follows: By introducing a 3D generation network, a 3D registration network, a 3D spatial transformation network, and a 3D multi-scale discriminant network, the present invention integrates multi-dimensional information, significantly improves the expression ability of the model. The multi-scale discriminant network combines with the weighted mean square error loss function to enhance the game performance, and the registration network ensures the spatial alignment of the synthesized CTPA image and the ICTPA image, thereby realizing more accurate error calculation. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] Figure 1 is a flowchart of a method for synthesizing CT angiography images based on a generative adversarial network according to the present invention.
[0079] Figure 2 is the plain CT image INCCT of the brain of the same patient according to the present invention.
[0080] Figure 3 is the enhanced CTPA image ICTPA of the brain of the same patient according to the present invention.
[0081] Figure 4 is the CTPA image synthesized by the method of the present invention.
[0082] Figure 5 is a schematic diagram of a system for synthesizing CT angiography images based on a generative adversarial network according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0083] The present invention will be further described in detail below with reference to specific embodiments.
[0084] It should be noted that these embodiments are only used to illustrate the present invention, rather than limiting the present invention. Any simple improvement of this method under the premise of the concept of the present invention falls within the scope of protection required by the present invention.
[0085] Referring to Figure 1 , a method for synthesizing CT angiography images based on a generative adversarial network includes:
[0086] S100. Obtain a pair of non-contrast CT sequences and CTPA sequences as source modalities, and after preprocessing, obtain the input source.
[0087] Collect pairs of non-contrast CT and CTPA image data of the same part of the same patient, and screen out the eligible images to include in the standard data.
[0088] Preprocess the screened pairs of non-contrast CT and CTPA image data. The preprocessing includes but is not limited to cropping, rotation, flipping, contrast enhancement, and normalization.
[0089] The input source includes the non-contrast CT image INCCT (see Figure 2 ) and the corresponding contrast-enhanced CTPA image ICTPA (see Figure 3 ).
[0090] S200. Input the input source into the image synthesis model, and use a generative adversarial network to train the model. The model training process includes:
[0091] A. Pass the non-contrast CT image INCCT and the corresponding contrast-enhanced CTPA image ICTPA in the input source through the generative network to extract the anatomical structure features in the non-contrast CT image INCCT and the vascular enhancement features in the contrast-enhanced CTPA image ICTPA, and generate a synthetic CTPA image.
[0092] The synthetic CTPA image retains both the anatomical content of INCCT and the vascular enhancement information of ICTPA.
[0093] B. Input the synthetic CTPA image and the corresponding contrast-enhanced CTPA image ICTPA into the registration network to obtain the anatomical structure registration information matrix.
[0094] Ensure the anatomical position consistency between the synthetic CTPA image and the contrast-enhanced CTPA image ICTPA.
[0095] C. Input the synthetic CTPA image and the anatomical structure registration information matrix into the spatial transformation network to obtain the registered synthetic CTPA image.
[0096] Make the registered synthetic CTPA image aligned with the contrast-enhanced CTPA image ICTPA in the spatial position.
[0097] D. Respectively extract the vascular regions in the registered synthetic CTPA image and the corresponding contrast-enhanced CTPA image ICTPA through the vascular segmentation model to obtain a pair of segmented vascular segmentation images.
[0098] E. Through a multi-scale discriminant network, input the synthesized CTPA image, calculate the probability that the synthesized CTPA image is the corresponding enhanced CTPA image ICTPA, and guide the generation network to reduce the difference between the synthesized CTPA image and the corresponding enhanced CTPA image ICTPA.
[0099] F. Through a multi-scale discriminant network, input the synthesized CTPA image and the corresponding enhanced CTPA image ICTPA, calculate the probability that the synthesized CTPA image is not the corresponding enhanced CTPA image ICTPA and the probability that the synthesized CTPA image is the corresponding enhanced CTPA image ICTPA, and feedback to the multi-scale discriminant network to optimize the recognition ability.
[0100] The loss function of the image synthesis model includes the loss of the generation network (Lsm), the loss of the registration network (Lsr), the adversarial loss of the generation network (Ladv), and the unsupervised loss (Lnsr);
[0101] Calculate the sum of the four losses and perform backpropagation to update the parameters of the generation network and the registration network;
[0102] And calculate the loss of the multi-scale discriminant network (Lb) and update the parameters of the multi-scale network;
[0103] Use the Adam optimizer to update the parameters of the generation network, the registration network, and the multi-scale discriminant network.
[0104] The loss function L of the generation network sm is:
[0105]
[0106] where E(·) is the mean of the registration information matrix Tr, and are the horizontal gradient and the vertical gradient respectively.
[0107] The loss function L of the registration network sr is:
[0108]
[0109] where N is the number of samples in each batch for training, y i is the i-th sample in the enhanced CTPA image ICTPA, is the i-th sample in the registered synthesized CTPA image.
[0110] The loss function L of the adversarial loss of the generation network adv is:
[0111]
[0112] Among them, N is the number of samples in each batch for training, p i and q i are the two outputs of the multi-scale discriminant network, p i and q i is the probability that the calculated synthetic CTPA image is the corresponding enhanced CTPA image ICTPA.
[0113] The loss function L of the unsupervised loss of the generation network nsr is:
[0114]
[0115] Among them, N is the number of samples in each batch for training, z i is the i-th sample of the vascular segmentation image of the enhanced CTPA image ICTPA in the pair of segmented vascular segmentation images, is the i-th sample of the vascular segmentation image of the registered synthetic CTPA image in the pair of segmented vascular segmentation images.
[0116] The loss function L of the multi-scale discriminant network b is:
[0117] L b =(L SynCTPA +L ICTPA ) / 2
[0118] Among them, the formula of L SynCTPA is as follows:
[0119]
[0120] Among them, N is the number of samples in each batch for training, m i , n i are the two outputs of the multi-scale discriminant network, m i and n i is the probability that the calculated synthetic CTPA image is not the corresponding CTPA image.
[0121] The formula of L ICTPA is as follows:
[0122]
[0123] Among them, N is the number of samples in each batch for training, s i , t i are the outputs of the multi-scale discriminant network, s i , t i is the probability that the calculated synthetic CTPA image is the corresponding enhanced CTPA image ICTPA.
[0124] S300. Input the non-contrast CT image into the trained image synthesis model to generate a synthesized CTPA image.
[0125] After generating the synthesized CTPA image, through inverse normalization, smoothing processing, and artifact removal, it is saved as an image in DICOM format (see Figure 4 ).
[0126] See Figure 5 , which is a system for synthesizing CT angiography images based on a generative adversarial network, including a data acquisition module 100, a data processing module 200, and a data output module 300.
[0127] The data acquisition module 100 is used to obtain paired non-contrast CT sequences and CTPA sequences as source modalities, and after preprocessing, obtain the input source.
[0128] The data processing module 200 is used to input the input source into the image synthesis model, and train the model using a generative adversarial network. The model training process includes:
[0129] Extract the anatomical structure features in the non-contrast CT image INCCT and the vascular enhancement features in the contrast-enhanced CTPA image ICTPA from the input source through the generation network to generate a synthesized CTPA image;
[0130] Input the synthesized CTPA image and the corresponding contrast-enhanced CTPA image ICTPA into the registration network to obtain the anatomical structure registration information matrix;
[0131] Input the synthesized CTPA image and the anatomical structure registration information matrix into the spatial transformation network to obtain the registered synthesized CTPA image;
[0132] Extract the vascular regions in the registered synthesized CTPA image and the corresponding contrast-enhanced CTPA image ICTPA respectively through the vascular segmentation model to obtain the segmented vascular segmentation image pair;
[0133] Input the synthesized CTPA image through the multi-scale discriminant network, calculate the probability that the synthesized CTPA image is the corresponding contrast-enhanced CTPA image ICTPA, and guide the generation network to reduce the difference between the synthesized CTPA image and the corresponding contrast-enhanced CTPA image ICTPA;
[0134] Through the multi-scale discriminant network, calculate the probability that the synthesized CTPA image is not the corresponding CTPA image and the probability that the synthesized CTPA image is the corresponding contrast-enhanced CTPA image ICTPA, and feedback to the multi-scale discriminant network to optimize the recognition ability.
[0135] The data output module 300 is used to input the plain CT images into the trained image synthesis model to generate the synthesized CTPA images.
[0136] Among them, the loss function of the image synthesis model includes the loss of the generation network, the loss of the registration network, the adversarial loss of the generation network, and the unsupervised loss;
[0137] Calculate the sum of the four losses and perform backpropagation to update the parameters of the generation network and the registration network;
[0138] And calculate the loss of the multi-scale discriminant network and update the parameters of the multi-scale network;
[0139] Use the Adam optimizer to update the parameters of the generation network, the registration network, and the multi-scale discriminant network.
[0140] Among them, the loss function L of the generation network sm :
[0141]
[0142] Among them, E(·) is the mean value of the registration information matrix Tr, and are the horizontal gradient and the vertical gradient respectively;
[0143] The loss function L of the registration network sr :
[0144]
[0145] Among them, N is the number of samples in each batch used for training, and y i is the i-th sample in the enhanced CTPA image ICTPA, is the i-th sample in the registered synthesized CTPA image;
[0146] The loss function L of the adversarial loss of the generation network adv :
[0147]
[0148] Among them, N is the number of samples in each batch used for training, p i and q i are the two outputs of the multi-scale discriminant network, p i and q i are the probabilities that the calculated synthesized CTPA image is the corresponding enhanced CTPA image ICTPA;
[0149] The loss function L of the unsupervised loss of the generation network nsr :
[0150]
[0151] Among them, N is the number of samples in each batch for training, and z i is the i-th sample of the vascular segmentation image of the enhanced CTPA image ICTPA in the pair of vascular segmentation images after segmentation, and is the i-th sample of the vascular segmentation image of the registered synthetic CTPA image in the pair of vascular segmentation images after segmentation.
[0152] Among them, the loss function L of the multi-scale discriminant network b :
[0153] L b =(L SynCTPA +L ICTPA ) / 2
[0154] Among them, the formula of L SynCTPA is as follows:
[0155]
[0156] Among them, N is the number of samples in each batch for training, m i , n i are the two outputs of the multi-scale discriminant network, and m i and n i are used to calculate the probability that the synthetic CTPA image is not the corresponding CTPA image;
[0157] The formula of L ICTPA is as follows:
[0158]
[0159] Among them, N is the number of samples in each batch for training, s i , t i are the outputs of the multi-scale discriminant network, and s i , t i are used to calculate the probability that the synthetic CTPA image is the corresponding enhanced CTPA image ICTPA.
[0160] After generating the synthetic CTPA image, it is saved as an image in DICOM format after anti-normalization, smoothing processing, and artifact removal.
[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit. Although the present invention has been described by referring to the preferred embodiments of the present invention, those of ordinary skill in the art should understand that various changes can be made to it in form and details without departing from the spirit and scope of the present invention defined by the appended claims.
Claims
1. A method for synthesizing CT angiography images based on a generative adversarial network, characterized in that, Including: Obtaining paired non-contrast CT sequences and CTPA sequences as source modalities, and performing preprocessing to obtain an input source; Inputting the input source into an image synthesis model, and training the model using a generative adversarial network. The model training process includes: Extracting the anatomical structure features in the non-contrast CT image INCCT and the vascular enhancement features in the contrast-enhanced CTPA image ICTPA from the input source through a generative network, and generating a synthetic CTPA image; Inputting the synthetic CTPA image and the corresponding contrast-enhanced CTPA image ICTPA into a registration network to obtain an anatomical structure registration information matrix; Inputting the synthetic CTPA image and the anatomical structure registration information matrix into a spatial transformation network to obtain a registered synthetic CTPA image; Respectively extracting the vascular regions in the registered synthetic CTPA image and the corresponding contrast-enhanced CTPA image ICTPA through a vascular segmentation model to obtain a pair of segmented vascular segmentation images; Through a multi-scale discriminant network, inputting the synthetic CTPA image, calculating the probability that the synthetic CTPA image is the corresponding contrast-enhanced CTPA image ICTPA, and guiding the generative network to reduce the difference between the synthetic CTPA image and the corresponding contrast-enhanced CTPA image ICTPA; Through a multi-scale discriminant network, calculating the probability that the synthetic CTPA image is not the corresponding CTPA image and the probability that the synthetic CTPA image is the corresponding contrast-enhanced CTPA image ICTPA, and feeding back to the multi-scale discriminant network to optimize the recognition ability; Inputting the non-contrast CT image into the trained image synthesis model to generate a synthetic CTPA image.
2. The method for synthesizing CT angiography images based on a generative adversarial network according to claim 1, wherein The loss function of the image synthesis model includes the loss of the generative network, the loss of the registration network, the adversarial loss of the generative network, and the unsupervised loss; Calculating the sum of the four losses and performing backpropagation to update the parameters of the generative network and the registration network; And calculating the loss of the multi-scale discriminant network and updating the parameters of the multi-scale network; Using an Adam optimizer to update the parameters of the generative network, the registration network, and the multi-scale discriminant network.
3. The method for synthesizing CT angiography images based on a generative adversarial network according to claim 2, wherein The loss function L of the generation network sm : where E(·) is the mean of the registration information matrix Tr, and are the horizontal gradient and the vertical gradient, respectively; The loss function L of the registration network sr : where N is the number of samples in each batch for training, and y i is the i-th sample in the ICTPA of the enhanced CTPA image, and is the i-th sample in the registered synthetic CTPA image; The loss function L of the adversarial loss of the generation network adv : Among them, N is the number of samples in each batch for training, p i and q i are the two outputs of the multi-scale discriminative network, p i and q i is the probability that the calculated synthetic CTPA image is the corresponding enhanced CTPA image ICTPA; The loss function L of the unsupervised loss of the generation network nsr : where N is the number of samples in each batch for training, and z i is the i-th sample of the vascular segmentation image ICTPA of the enhanced CTPA image in the pair of segmented vascular segmentation images, and is the i-th sample of the vascular segmentation image of the registered synthetic CTPA image in the pair of segmented vascular segmentation images.
4. The method for synthesizing CT angiography images based on a generative adversarial network according to claim 3, wherein The loss function L of the multi-scale discrimination network b : L b = (L SynCTPA + L ICTPA ) / 2 Among them, L SynCTPA has the following formula: Among them, N is the number of samples in each batch for training, m i , n i are the two outputs of the multi-scale discriminant network, m i and n i are used to calculate the probability that the synthesized CTPA image is not the corresponding CTPA image; L ICTPA The formula is as follows: where N is the number of samples in each batch for training, s i , t i are the outputs of the multi-scale discrimination network, and s i , t i is the probability of calculating the synthesized CTPA image as the corresponding enhanced CTPA image ICTPA.
5. The method for synthesizing CT angiography images based on a generative adversarial network according to claim 4, wherein After generating the synthetic CTPA image, it undergoes denormalization, smoothing processing, and artifact removal, and is saved as an image in DICOM format.
6. A system for synthesizing CT angiography images based on a generative adversarial network, characterized in that, Including a data acquisition module, a data processing module, and a data output module; The data acquisition module is used to obtain paired non-contrast CT sequences and CTPA sequences as source modalities, and perform preprocessing to obtain an input source; The data processing module is used to input the input source into an image synthesis model, and train the model using a generative adversarial network. The model training process includes: Extracting the anatomical structure features in the non-contrast CT image INCCT and the vascular enhancement features in the contrast-enhanced CTPA image ICTPA from the input source through a generative network, and generating a synthetic CTPA image; Input the synthesized CTPA image and the corresponding enhanced CTPA image ICTPA into the registration network to obtain the anatomical structure registration information matrix; Input the synthesized CTPA image and the anatomical structure registration information matrix into the spatial transformation network to obtain the registered synthesized CTPA image; Extract the vascular regions from the registered synthesized CTPA image and the corresponding enhanced CTPA image ICTPA respectively through the vascular segmentation model to obtain the segmented vascular segmentation image pair; Through the multi-scale discriminant network, input the synthesized CTPA image, calculate the probability that the synthesized CTPA image is the corresponding enhanced CTPA image ICTPA, and guide the generation network to reduce the difference between the synthesized CTPA image and the corresponding enhanced CTPA image ICTPA; Through the multi-scale discriminant network, calculate the probability that the synthesized CTPA image is not the corresponding CTPA image and the probability that the synthesized CTPA image is the corresponding enhanced CTPA image ICTPA, and feedback to the multi-scale discriminant network to optimize the recognition ability; The data output module is used to input the plain CT image into the trained image synthesis model to generate the synthesized CTPA image.
7. The system for synthesizing CT angiography images based on a generative adversarial network according to claim 6, wherein The loss function of the image synthesis model includes the loss of the generation network, the loss of the registration network, the adversarial loss of the generation network, and the unsupervised loss; Calculate the sum of the four losses and perform backpropagation to update the parameters of the generation network and the registration network; And calculate the loss of the multi-scale discriminant network and update the parameters of the multi-scale network; Use the Adam optimizer to update the parameters of the generation network, the registration network, and the multi-scale discriminant network.
8. The system for synthesizing CT angiography images based on a generative adversarial network according to claim 7, wherein The loss function L of the generation network sm : where E(·) is the mean value of the registration information matrix Tr, and are the horizontal gradient and the vertical gradient respectively; The loss function L of the registration network sr :[[]] where N is the number of samples in each batch for training, and y i is the i-th sample in the enhanced CTPA image ICTPA, and is the i-th sample in the registered synthetic CTPA image; The loss function L of the adversarial loss of the generation network adv :[[-END]] where N is the number of samples in each batch for training, p i and q i are the two outputs of the multi-scale discrimination network, p i and q i is the probability that the calculated synthetic CTPA image is the corresponding enhanced CTPA image ICTPA; The loss function L of the unsupervised loss of the generation network nsr : where N is the number of samples in each batch for training, and z i is the i-th sample of the vascular segmentation image ICTPA of the enhanced CTPA image in the pair of segmented vascular segmentation images, and is the i-th sample of the vascular segmentation image of the registered synthetic CTPA image in the pair of segmented vascular segmentation images.
9. The system for synthesizing CT angiography images based on a generative adversarial network according to claim 8, wherein The loss function L of the multi-scale discrimination network b : L b = (L SynCTPA + L ICTPA ) / 2 Among them, L SynCTPA has the following formula: Among them, N is the number of samples in each batch for training, m i , n i are the two outputs of the multi-scale discriminant network, m i and n i are used to calculate the probability that the synthesized CTPA image is not the corresponding CTPA image; L ICTPA The formula is as follows: Among them, N is the number of samples in each batch for training, s i , t i are the outputs of the multi-scale discrimination network, s i , t i is the probability of calculating the synthesized CTPA image as the corresponding enhanced CTPA image ICTPA.
10. The system for synthesizing CT angiography images based on a generative adversarial network according to claim 9, wherein After generating the synthesized CTPA image, it undergoes inverse normalization, smoothing processing, and artifact removal, and is saved as an image in DICOM format.