A synthesis method and system for synthesizing a CT image based on domain adaptation of an MR image

Abstract

The embodiment of the application provides a synthesis method and system for synthesizing a CT image based on domain adaptation of an MR image, the method comprising determining a target MR image and a target CT image; performing slicing operation along a target slicing direction according to a preset slice size to obtain an MR slice image and a CT slice image, the target slicing direction comprising at least one of a transverse plane direction, a coronal plane direction and a sagittal plane direction; constructing an initial CT image synthesis model, inputting the MR slice image into the initial CT image synthesis model to perform model training, in the training process, projecting the input image into a common feature subspace, reducing the distribution difference of images in different slicing directions through relevant distance measurement; when the model training is completed, obtaining a target CT image synthesis model, performing slicing operation on the obtained to-be-processed MR image, inputting the obtained slice image into the target CT image synthesis model, and obtaining a required synthesized CT image.

Description

Technical Field

[0001] This application relates to the field of multimodal medical imaging technology, and more specifically, to a method and system for synthesizing CT images based on domain adaptation of MR images. Background Technology

[0002] Currently, MR imaging is widely used for patient diagnosis due to its non-ionizing, non-invasive nature and a range of contrast mechanisms. However, MR images do not directly provide electron density information; furthermore, while CT imaging can provide electron density information, it is time-consuming, expensive, potentially harmful to patients, and requires accurate MR / CT registration, thus it cannot be directly used for patient diagnosis.

[0003] In recent years, Generative Adversarial Networks (GANs) have been widely used in image translation tasks due to their excellent realism in image synthesis. GANs possess discriminators that capture information about target distribution and guide the generator to perform a one-time mapping from source to target images. Based on this adversarial mechanism, GAN-based MR-to-CT conversion tasks have reported state-of-the-art results across numerous tasks. Currently, common approaches to MR-to-CT conversion using GANs include: 1. 3D data synthesis, i.e., synthesizing corresponding CT volumes using 3D MR volume data; 2. Synthesizing CT images based on 2D MR images.

[0004] However, existing methods for synthesizing CT images from MR-based 2D images often use one of the cross-sectional, coronal, or sagittal plane datasets for model training. While this ensures good synthesis results from the training dataset, the different data distributions of the datasets from different cross-sections, without considering the different data feature distributions, lead to poor generalization ability of the trained GAN model and poor image synthesis quality. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for synthesizing CT images based on domain adaptation of MR images, which can improve the quality of image synthesis.

[0006] This application also provides a method for synthesizing CT images based on domain adaptation of MR images, including the following steps:

[0007] Identify the target MR image and the target CT image registered with the target MR image;

[0008] For the target MR image and the target CT image, slicing operations are performed along the target slicing direction according to a preset slice size to obtain corresponding MR slice images and CT slice images. The target slicing direction includes at least one of the transverse direction, the coronal direction, and the sagittal direction.

[0009] An initial CT image synthesis model is constructed, and the MR slice image is input into the initial CT image synthesis model for model training. During the training process, the input image is projected into a common feature subspace. By using relevant distance metrics, the distribution differences of images in different slice directions are reduced. At least one of pixel loss and domain feature generation adversarial loss is combined to constrain the model in order to improve the model training accuracy.

[0010] When the model training ends, the corresponding target CT image synthesis model is obtained. After slicing the acquired MR image to be processed, the sliced ​​image is input into the target CT image synthesis model to obtain the required synthesized CT image.

[0011] Secondly, embodiments of this application also provide a synthesis system for synthesizing CT images from MR images based on domain adaptation. The system includes an image acquisition module, an image slicing module, a model training module, and an image synthesis module, wherein:

[0012] The image acquisition module is used to determine the target MR image and the target CT image registered with the target MR image;

[0013] The image slicing module is used to slice the target MR image and the target CT image according to a preset slice size along the target slicing direction to obtain the corresponding MR slice image and CT slice image. The target slicing direction includes at least one of the transverse direction, the coronal direction and the sagittal direction.

[0014] The model training module is used to construct an initial CT image synthesis model and input the MR slice image into the initial CT image synthesis model for model training. During the training process, the input image is projected into a common feature subspace. By using relevant distance metrics, the distribution differences of images in different slice directions are reduced. At least one of pixel loss and domain feature generation adversarial loss is combined to constrain the model in order to improve the model training accuracy.

[0015] The image synthesis module is used to obtain a corresponding target CT image synthesis model when the model training ends. After slicing the acquired MR image to be processed, the sliced ​​image is input into the target CT image synthesis model to obtain the required synthesized CT image.

[0016] Thirdly, embodiments of this application also provide a readable storage medium, which includes a synthesis method program for synthesizing CT images based on domain adaptation of MR images. When the synthesis method program for synthesizing CT images based on domain adaptation of MR images is executed by a processor, it implements the steps of a synthesis method for synthesizing CT images based on domain adaptation of MR images as described in any of the preceding claims.

[0017] As can be seen from the above, the method, system, and readable storage medium for synthesizing CT images from MR images based on domain adaptation provided in this application utilize feature domain adaptation technology and deep learning technology with strong fitting capabilities. By analyzing and verifying MR image information obtained from different slice directions, and determining that there are differences in the data distribution of MR datasets from slices in different directions, the features are projected into a common feature subspace to minimize the distribution differences. This not only effectively improves the image synthesis quality but also enhances the generalization and quality of synthesizing CT images from MR images of different slices.

[0018] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a domain-adaptive MR image-to-CT image synthesis method provided in this application embodiment;

[0021] Figure 2 Flowchart for image preprocessing implementation;

[0022] Figure 3 This is a schematic diagram of the network structure of a CT image synthesis model;

[0023] Figure 4 This is a schematic diagram of the structure of a synthesis system for synthesizing MR images into CT images based on domain adaptation, provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Please refer to Figure 1 , Figure 1 This is a flowchart of a domain-adaptive MR image-to-CT image synthesis method according to some embodiments of this application. The method includes the following steps:

[0027] Step S100: Determine the target MR image and the target CT image registered with the target MR image.

[0028] Step S200: For the target MR image and the target CT image, slice them along the target slice direction according to the preset slice size to obtain the corresponding MR slice image and CT slice image. The target slice direction includes at least one of the transverse direction, the coronal direction and the sagittal direction.

[0029] Step S300: Construct an initial CT image synthesis model and input the MR slice image into the initial CT image synthesis model for model training. During the training process, the input image is projected into a common feature subspace. By using relevant feature distance measurement, the distribution difference of the image in different slice directions is reduced. At least one of pixel loss and domain feature generation adversarial loss is combined to constrain the model in order to improve the model training accuracy.

[0030] In step S400, when the model training ends, the corresponding target CT image synthesis model is obtained. After slicing the acquired MR image to be processed, the sliced ​​image is input into the target CT image synthesis model to obtain the required synthesized CT image.

[0031] As can be seen from the above, the method for synthesizing CT images based on domain adaptation of MR images disclosed in this application utilizes feature domain adaptation technology and deep learning technology with strong fitting capabilities. By analyzing and verifying MR image information obtained from different slice directions, and determining that there are differences in the data distribution of MR datasets from slices in different directions, the features are projected into a common feature subspace to minimize the distribution differences. This not only effectively improves the image synthesis quality, but also enhances the generalization and quality of synthesizing CT images from MR images of different slices.

[0032] In one embodiment, step S100, determining the target MR image and the target CT image registered with the target MR image, includes:

[0033] Step S1000: Acquire the initial MR image and the initial CT image registered with the initial MR image.

[0034] Step S1001: Preprocess the initial MR image and the initial CT image according to a preset data processing method to obtain corresponding preprocessed data. The data processing method includes at least one of the following methods:

[0035] A standardized processing method used to adjust the pixels of a processed image to a preset numerical range.

[0036] Specifically, when acquiring MR and CT images, different protocols result in inconsistent image reading ranges, which may be much larger than the standard pixel range of the image. Therefore, before model training, MR and CT images need to undergo pixel normalization to adjust their pixel values ​​to a preset range. In one embodiment, this preset range can be an integer interval from 0 to 255.

[0037] The N4 bias field correction method is used to adjust regions with uneven display in processed images.

[0038] Specifically, due to changes in the magnetic field, during image visualization, certain image regions may exhibit stronger or weaker display intensity. This is generally caused by variations in the magnetic field, and the mapping of these variations is called the bias field. Therefore, to avoid image display inhomogeneity caused by the bias field, this embodiment requires N4 bias correction processing on the acquired MR images.

[0039] Image denoising methods used to remove noise present in an image.

[0040] Specifically, image denoising refers to the process of reducing noise in digital images. Real-world digital images are often affected by noise interference from imaging equipment and the external environment during digitization and transmission, and are thus called noisy images. Noise is a significant cause of image interference. An image in practical applications may contain various types of noise, which may be generated during transmission or during processing such as quantization. Therefore, to improve image quality and interpretability, noise needs to be identified and filtered out from the input training images before model training.

[0041] In one embodiment, image denoising can be performed based on feature analysis in the wavelet domain and feature analysis in the spatial domain. This application does not limit the specific implementation method.

[0042] A method for further rigid registration of MR images and associated CT images.

[0043] Specifically, in some special cases, the acquired MR images and their corresponding associated CT images may not be perfectly matched. In order to ensure the smooth implementation of subsequent supervised synthesis, the current embodiment requires rigid registration of the MR images and the associated CT images.

[0044] It should be noted that rigid registration refers to seeking a spatial transformation (or a series of spatial transformations) to achieve spatial consistency between corresponding points on an image, thereby further improving the accuracy of image registration.

[0045] A clustering and assimilation method for identifying and preserving common image regions from MR images and associated CT images.

[0046] Step S1002: Based on the obtained preprocessed data, determine the target MR image and the target CT image registered with the target MR image.

[0047] Specifically, the obtained preprocessed data consists of the target MR image to be determined and the target CT image registered with it. The model will then be trained based on the target MR image and the target CT image registered with it.

[0048] It should be noted that the overall image preprocessing workflow can be referenced. Figure 2 ,based on Figure 2 As can be seen, the image preprocessing process will first go through the data reading stage, then the standardization process of MR and CT images, then the N4 bias field correction stage of MR images, then the noise removal stage generated during image acquisition, then the rigid registration stage of MR and CT images, and finally the cropping of the non-shared parts of MR and CT images, retaining only the shared parts of both.

[0049] In the above embodiments, before model training, the acquired initial MR image and the initial CT image registered with it are preprocessed by methods such as image denoising and standardization. By removing factors in the image that affect the accuracy and precision of subsequent processing, the subsequent processing is facilitated and the reliability of subsequent processing is improved.

[0050] In one embodiment, step S200 involves slicing along the target slice direction according to a preset slice size to obtain a corresponding slice image, including:

[0051] Step S2000: Obtain the image size of the slice image covered by the target slice direction, and when it is determined that the image size is smaller than the preset slice size, perform expansion processing along the edge of the slice image to expand the image size of the slice image to the preset slice size.

[0052] Specifically, by examining the image size of the slice image covered by the target slice direction, if it is determined that the image size of the slice image covered by that direction is smaller than a preset slice size, in order to make the training data as close as possible to the test data, it is necessary to perform image augmentation processing along the edge of the image using an appropriate method. In one embodiment, a zero matrix can be used, that is, the image can be augmented using blank pixels.

[0053] Step S2001: When it is determined that the image size is larger than the preset cross-section size, the image size of the cross-section image is reduced to the preset cross-section size while ensuring image clarity.

[0054] Specifically, when reducing the image size of a cross-section image to a preset cross-section size, downsampling can be used to reduce the image size to the preset cross-section size. It should be noted that, to address uneven data distribution, downsampling aims to select a portion of data from the majority set and recombine it with the minority set to form a new dataset. Common downsampling methods include random downsampling, EasyEnsemble, and BalanceCascade (which uses an ensemble learning mechanism to address the information loss problem in traditional random undersampling), etc., and this application does not limit these methods. Downsampling can only be implemented if a preset sampling theorem is satisfied; otherwise, it will cause aliasing of signal components, affecting the sampling effect.

[0055] Step S2002: After determining that the image size of the cross-section image has been adjusted to the preset cross-section size, a slicing operation is performed along the target slicing direction to obtain the corresponding slice image.

[0056] Specifically, when performing a slicing operation, an image cutting frame can be pre-selected, and the scope of this image cutting frame should be able to cover the entire slice area. Then, the position of the image cutting frame is selected, and the slicing operation is performed at that position along the target slice direction to obtain the corresponding sliced ​​image.

[0057] In the above embodiments, the image size of the sliced ​​image covered by the target slice direction is adjusted before the slicing operation, which further ensures that the training data is as close as possible to the test data and improves the training accuracy.

[0058] In one embodiment, the CT image synthesis model includes a feature extraction network and a generator network, wherein:

[0059] The feature extraction network projects the input image into a common feature subspace, and obtains the corresponding target mapping features in the common feature subspace based on the processing method of minimizing the feature distance between images in different slice directions.

[0060] For details, please refer to Figure 3 This is a schematic diagram of the network structure of a CT image synthesis model. When processing target mapping features, it first projects MR slice images from different sections onto a common feature subspace, and then minimizes the distance between the MR data of each section within this feature subspace (see reference for details). Figure 3 middle, To F, This involves establishing a shared implementation process between F and F to mitigate the problem of inconsistent data distribution across different cross sections and improve the quality of synthesized CT data from different datasets.

[0061] The generator network consists of an encoder and a decoder. The encoder encodes the acquired target mapping features to obtain the corresponding feature vectors, and the decoder generates the predicted target image based on the feature vectors.

[0062] For details, please refer to Figure 3 , Figure 3 The G shown in the diagram is the generator network. The encoder extracts key information features to obtain a more refined data representation. The decoder is used to realize the process of reconstructing the original data from the refined data representation. In this way, the main information in the original signal can be reconstructed, and secondary interference information such as noise can be removed.

[0063] The above embodiments, by taking into account different data feature distributions and performing domain adaptation processing, can improve the generalization ability of the model and enhance the generalization and synthesis quality of CT synthesized from MR images of different sections.

[0064] In one embodiment, the CT image synthesis model further includes a domain feature discriminator network, wherein:

[0065] The domain feature discriminator network is used to discriminate the slice orientation of the target mapping features extracted by the feature extraction network, and to determine whether there is a deviation in the slice orientation.

[0066] For details, please refer to Figure 3 , Figure 3 The D shown in the diagram represents the domain feature discriminator network. This network determines whether the extracted features are aligned with the slice direction of the training dataset. If no deviation is found in the slice direction, it outputs a first judgment result indicating no deviation; otherwise, it outputs a second judgment result indicating deviation. It should be noted that the domain feature discriminator network is only used during the training phase.

[0067] In the above embodiments, during the training phase, the domain feature discriminator network can further avoid deviations in the slice direction of the extracted target mapping features, thereby improving the image synthesis quality.

[0068] In one embodiment, pixel loss L pix Determined by the following formula:

[0069] ;

[0070] Among them, y i This indicates the standard image that is actually needed. The target image predicted by the generator network, where i represents the pixel number.

[0071] Domain feature generation adversarial loss L domain Determined by the following formula:

[0072] ;

[0073] Among them, f i D(f) represents the target mapping features obtained through feature extraction network processing. i E represents the discriminator. x This indicates that the mean of the function is taken from the distribution of x.

[0074] Please refer to Figure 4 This application discloses a domain-adaptive MR image to CT image synthesis system 400. The system 400 includes an image processing module 401, an image slicing module 402, a model training module 403, and an image synthesis module 404, wherein:

[0075] Image processing module 401 is used to determine the target MR image and the target CT image registered with the target MR image.

[0076] The image slicing module 402 is used to slice the target MR image and the target CT image according to a preset slice size along the target slicing direction to obtain the corresponding MR slice image and CT slice image. The target slicing direction includes at least one of the transverse direction, the coronal direction and the sagittal direction.

[0077] The model training module 403 is used to construct an initial CT image synthesis model and input MR slice images into the initial CT image synthesis model for model training. During the training process, the input images are projected into a common feature subspace. By using relevant distance metrics, the distribution differences of images in different slice directions are reduced. At least one of pixel loss and domain feature generation adversarial loss is combined to constrain the model in order to improve the model training accuracy.

[0078] The image synthesis module 404 is used to obtain the corresponding target CT image synthesis model when the model training ends. After slicing the acquired MR image to be processed, the sliced ​​image is input into the target CT image synthesis model to obtain the required synthesized CT image.

[0079] In one embodiment, the image processing module is further configured to acquire an initial MR image and an initial CT image registered with the initial MR image; preprocess the initial MR image and the initial CT image according to a preset data processing method to obtain corresponding preprocessed data, wherein the data processing method includes at least one of the following methods: a standardization processing method for adjusting the pixels of the processed image to a preset numerical range; an N4 bias field correction method for adjusting regions with uneven display in the processed image; an image denoising method for removing noise present in the image; a rigid registration method for further rigid registration of the MR image and the associated CT image; a clustering and differentiating method for identifying and retaining common image regions from the MR image and the associated CT image; and determine a target MR image and a target CT image registered with the target MR image based on the obtained preprocessed data.

[0080] In one embodiment, the image slicing module is further configured to obtain the image size of the sliced ​​image covered by the target slicing direction, and when it is determined that the image size is smaller than a preset slice size, to perform expansion processing along the edge portion of the sliced ​​image to enlarge the image size of the sliced ​​image to the preset slice size; when it is determined that the image size is larger than the preset slice size, to reduce the image size of the sliced ​​image to the preset slice size while ensuring image clarity; and when it is determined that the image size of the sliced ​​image has been adjusted to the preset slice size, to perform slicing operation along the target slicing direction to obtain the corresponding sliced ​​image.

[0081] In one embodiment, pixel loss L pix Determined by the following formula:

[0082] ;

[0083] Among them, y i This indicates the standard image that is actually needed. The target image predicted by the generator network, where i represents the pixel number;

[0084] Domain feature generation adversarial loss L domain Determined by the following formula:

[0085] ;

[0086] Among them, f i D(f) represents the target mapping features obtained through feature extraction network processing. i E represents the discriminator. x This indicates that the mean of the function is taken from the distribution of x.

[0087] As can be seen from the above, the MR image synthesis system for synthesizing CT images based on domain adaptation disclosed in this application utilizes feature domain adaptation technology and deep learning technology with strong fitting capabilities. By analyzing and verifying MR image information obtained from different slice orientations, and determining that there are differences in the data distribution of MR datasets from slices in different orientations, the system projects their features into a common feature subspace, thereby minimizing their distribution differences. This not only effectively improves the image synthesis quality but also enhances the generalization and quality of CT synthesis from MR images of different slices.

[0088] This application provides a readable storage medium. When the computer program is executed by a processor, it performs the method in any optional implementation of the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0089] The aforementioned readable storage medium utilizes feature domain adaptation technology and deep learning technology with strong fitting capabilities to analyze and verify MR image information acquired from different slice orientations. When it is determined that there are differences in the data distribution of MR datasets from slices in different orientations, its features are projected into a common feature subspace, thereby minimizing its distribution differences. This not only effectively improves the quality of image synthesis but also enhances the generalization and quality of CT synthesis from MR images of different slices.

[0090] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0091] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0092] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0093] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0094] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for synthesizing MR images into CT images based on domain adaptation, characterized in that, Includes the following steps: Acquire an initial MR image and an initial CT image registered with the initial MR image; The initial MR image and the initial CT image are preprocessed according to a preset data processing method to obtain corresponding preprocessed data, wherein the data processing method includes at least one of the following methods: A standardized processing method used to adjust the pixels of a processed image to a preset numerical range; The N4 bias field correction method is used to adjust regions with uneven display in processed images. Image denoising methods for removing noise from images; A method for further rigid registration of MR images and associated CT images; A clustering and assimilation method for identifying and preserving common image regions from MR images and associated CT images; Based on the obtained preprocessed data, the target MR image and the target CT image registered with the target MR image are determined. For the target MR image and the target CT image, slicing operations are performed along the target slicing direction according to a preset slice size to obtain corresponding MR slice images and CT slice images. The target slicing direction includes at least one of the transverse direction, the coronal direction, and the sagittal direction. An initial CT image synthesis model is constructed, and the MR slice image is input into the initial CT image synthesis model for model training. During the training process, the input image is projected into a common feature subspace. By using relevant feature distance measurement, the distribution difference of the image in different slice directions is reduced. At least one of pixel loss and domain feature generation adversarial loss is combined to constrain the model in order to improve the model training accuracy. When the model training ends, the corresponding target CT image synthesis model is obtained. After slicing the acquired MR image to be processed, the sliced ​​image is input into the target CT image synthesis model to obtain the required synthesized CT image. The CT image synthesis model includes a feature extraction network and a generator network, wherein: The feature extraction network is used to project the input image into a common feature subspace, and in the common feature subspace, obtain the corresponding target mapping features based on the processing method of minimizing the feature distance between images in different slice directions; The generator network consists of an encoder and a decoder. The encoder encodes the acquired target mapping features to obtain a corresponding feature vector, and the decoder generates the predicted target image based on the feature vector.

2. The method according to claim 1, characterized in that, The step of slicing along the target slice direction according to a preset slice size to obtain a corresponding slice image includes: The image size of the slice image covered by the target slice direction is obtained, and when it is determined that the image size is smaller than the preset slice size, the image size of the slice image is expanded along the edge of the slice image to increase the image size of the slice image to the preset slice size. When it is determined that the image size is larger than the preset cross-section size, the image size of the cross-section image is reduced to the preset cross-section size while ensuring image clarity; Once it is determined that the image size of the cross-section image has been adjusted to the preset cross-section size, a slicing operation is performed along the target slicing direction to obtain the corresponding slice image.

3. The method according to claim 1, characterized in that, The CT image synthesis model also includes a domain feature discriminator network, wherein: The domain feature discriminator network is used to discriminate the slice direction of the target mapping features extracted by the feature extraction network, and thereby determine whether there is a deviation in the slice direction.

4. The method according to claim 1, characterized in that, The pixel loss L pix Determined by the following formula: ； Among them, y i This indicates the standard image that is actually needed. The target image predicted by the generator network, where i represents the pixel number; The domain feature generation adversarial loss L domain Determined by the following formula: ； Among them, f i D(f) represents the target mapping features obtained through feature extraction network processing. i E represents the discriminator. x This indicates that the mean of the function is taken from the distribution of x.

5. A synthesis system for synthesizing MR images into CT images based on domain adaptation, characterized in that, The system includes an image processing module, an image slicing module, a model training module, and an image synthesis module, wherein: The image processing module is used to acquire an initial MR image and an initial CT image registered with the initial MR image, and to preprocess the initial MR image and the initial CT image according to a preset data processing method to obtain corresponding preprocessed data. The data processing method includes at least one of the following methods: a standardization processing method for adjusting the pixels of the processed image to a preset numerical range; an N4 bias field correction method for adjusting unevenly displayed areas in the processed image; an image denoising method for removing noise present in the image; a rigid registration method for further rigid registration of the MR image and the associated CT image; a clustering and differentiating method for identifying and retaining common image regions from the MR image and the associated CT image; and, based on the obtained preprocessed data, to determine a target MR image and a target CT image registered with the target MR image. The image slicing module is used to slice the target MR image and the target CT image according to a preset slice size along the target slicing direction to obtain the corresponding MR slice image and CT slice image. The target slicing direction includes at least one of the transverse direction, the coronal direction and the sagittal direction. The model training module is used to construct an initial CT image synthesis model and input the MR slice image into the initial CT image synthesis model for model training. During the training process, the input image is projected into a common feature subspace. By using relevant distance metrics, the distribution differences of images in different slice directions are reduced. At least one of pixel loss and domain feature generation adversarial loss is combined to constrain the model in order to improve the model training accuracy. The image synthesis module is used to obtain a corresponding target CT image synthesis model when the model training ends. After slicing the acquired MR image to be processed, the sliced ​​images are input into the target CT image synthesis model to obtain the required synthesized CT image. The CT image synthesis model includes a feature extraction network and a generator network. The feature extraction network is used to project the input image into a common feature subspace and obtain the corresponding target mapping features in the common feature subspace based on the processing method of minimizing the feature distance between images in different slice directions. The generator network consists of an encoder and a decoder. The encoder is used to encode the acquired target mapping features to obtain the corresponding feature vector. The decoder is used to generate the predicted target image based on the feature vector.

6. The system according to claim 5, characterized in that, The image processing module is further configured to acquire an initial MR image and an initial CT image registered with the initial MR image; preprocess the initial MR image and the initial CT image according to a preset data processing method to obtain corresponding preprocessed data, wherein the data processing method includes at least one of the following methods: a standardization processing method for adjusting the pixels of the processed image to a preset numerical range; an N4 bias field correction method for adjusting unevenly displayed areas in the processed image; an image denoising method for removing noise present in the image; a rigid registration method for further rigidly registering the MR image and the associated CT image; a clustering and differentiating method for identifying and retaining common image regions from the MR image and the associated CT image; and determine a target MR image and a target CT image registered with the target MR image based on the obtained preprocessed data.

7. The system according to claim 5, characterized in that, The image slicing module is also used to obtain the image size of the sliced ​​image covered by the target slicing direction, and when it is determined that the image size is smaller than a preset slice size, to perform expansion processing along the edge portion of the sliced ​​image to expand the image size of the sliced ​​image to the preset slice size; When it is determined that the image size is larger than the preset slice size, the image size of the slice image is reduced to the preset slice size while ensuring image clarity; when it is determined that the image size of the slice image has been adjusted to the preset slice size, a slicing operation is performed along the target slicing direction to obtain the corresponding slice image.

8. A readable storage medium, characterized in that, The readable storage medium includes a method program for synthesizing CT images based on domain adaptation of MR images. When the method program for synthesizing CT images based on domain adaptation of MR images is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 4.

Images