Image conversion processing method and device, electronic equipment and storage medium

Abstract

The invention discloses an image conversion processing method and device, electronic equipment and a storage medium. The method comprises the steps of inputting a to-be-processed image and random noise into an image conversion model, and performing feature splicing processing on the to-be-processed image and the random noise based on a feature splicing module to obtain a to-be-processed spliced feature image; wherein the to-be-processed image is a magnetic resonance slice image containing a target part; carrying out primary denoising processing on the to-be-processed spliced feature image based on a denoising module to obtain a target image which is a computed tomography slice image containing a target part; wherein the image conversion model is obtained through training according to the sample magnetic resonance slice image and the target noise feature image, and the target noise feature image is determined after multiple times of noise adding processing are carried out on the sample tomography slice image based on the first model. According to the invention, the purpose of converting the magnetic resonance slice image into the tomography slice image is achieved, and the accuracy and stability of image conversion processing are ensured.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image conversion processing method, device, electronic equipment and storage medium. Background Art

[0002] Magnetic resonance imaging (MRI) images are mainly used to display soft tissues, while the imaging results of bone structure and calcified tissue are relatively poor. Computed tomography (CT) images can clearly display bone structure, stones, and calcified tissue. In order to give full play to the advantages of MRI images and CT images and save image acquisition costs, MRI images can be converted into CT images. In this way, the soft tissue information displayed by MRI and the bone structure and calcified tissue information displayed by CT images can be combined to meet more clinical needs.

[0003] At present, MRI images are mainly converted to CT images through Generative Adversarial Networks (GANs). However, when the above network performs image conversion, as the number of training times increases, there may be a problem of mode collapse, and the accuracy and stability of the model training results cannot be guaranteed, thus affecting the accuracy of the image conversion results. Summary of the invention

[0004] The present invention provides an image conversion processing method, device, electronic device and storage medium, which achieve the purpose of converting magnetic resonance slice images into tomography slice images and ensure the accuracy and stability of image conversion processing.

[0005] According to one aspect of the present invention, an image conversion processing method is provided. The method is processed by an image conversion model. The image conversion model at least includes a feature splicing module and a denoising module. The method includes:

[0006] The image to be processed and the random noise are input into a pre-trained image conversion model, so as to perform feature splicing processing on the image to be processed and the random noise through a feature splicing module to obtain a spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing a target part;

[0007] Performing a denoising process on the spliced ​​feature image to be processed based on the denoising module to obtain a target image corresponding to the image to be processed, wherein the target image is a computed tomography slice image containing the target part;

[0008] The image conversion model is trained based on sample magnetic resonance slice images and target noise feature images, and the target noise feature images are determined after multiple noise addition processes are performed on the sample tomographic slice images based on a pre-trained first model.

[0009] According to another aspect of the present invention, an image conversion processing device is provided. The device is processed by an image conversion model. The image conversion model at least includes a feature splicing module and a denoising module. The device includes:

[0010] The feature image determination module is used to input the image to be processed and the random noise into a pre-trained image conversion model, so as to perform feature splicing processing on the image to be processed and the random noise through the feature splicing module to obtain a spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing the target part;

[0011] A target image determination module is used to perform a denoising process on the spliced ​​feature image to be processed based on the denoising module to obtain a target image corresponding to the image to be processed, wherein the target image is a computed tomography slice image containing the target part;

[0012] The image conversion model is trained based on sample magnetic resonance slice images and target noise feature images, and the target noise feature images are determined after multiple noise addition processes are performed on the sample tomographic slice images based on a pre-trained first model.

[0013] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by at least one processor. The computer program is executed by at least one processor so that the at least one processor can execute the image conversion processing method of any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the image conversion processing method of any embodiment of the present invention when executed.

[0018] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, wherein the computer program, when executed by a processor, implements the image conversion processing method according to any embodiment of the present invention.

[0019] The technical solution of the embodiment of the present invention is to input the image to be processed and the random noise into the pre-trained image conversion model, so as to perform feature splicing processing on the image to be processed and the random noise according to the feature splicing module, and obtain the spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing the target part; based on the denoising module, the spliced ​​feature image to be processed is subjected to a denoising process once, and a target image corresponding to the image to be processed is obtained, and the target image is a computed tomography slice image containing the target part; wherein the image conversion model is obtained by training based on the sample magnetic resonance slice image and the target noise feature image, and the target noise feature image is determined after multiple noise addition processes are performed on the sample tomography slice image based on the pre-trained first model. The present invention solves the problem of mode collapse caused by image conversion by generating adversarial networks in the prior art, and achieves the purpose of converting magnetic resonance slice images into tomography slice images. By performing a denoising process on the spliced ​​feature image once, the computational overhead is reduced and the efficiency of image conversion is improved. The trained image conversion model is obtained based on the sample magnetic resonance slice image and the target noise feature image, and the accuracy and stability of the image conversion result are guaranteed.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 is a flow chart of an image conversion processing method provided by an embodiment of the present invention;

[0023] Figure 2 is a flow chart of an image conversion model training method provided by an embodiment of the present invention;

[0024] Figure 3 is an example diagram of an image conversion model training method provided by an embodiment of the present invention;

[0025] Figure 4 is a structural schematic diagram of an image conversion processing device provided by an embodiment of the present invention;

[0026] Figure 5 It is a schematic diagram of the structure of an electronic device for implementing the image conversion processing method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] Embodiment 1

[0030] Figure 1 This is a flowchart of an image conversion method provided by the first embodiment of the present invention. This embodiment is applicable to the case of converting a magnetic resonance slice image into a tomography slice image. The method can be executed by an image conversion device. The image conversion device can be implemented in the form of hardware and / or software. The image conversion device can be configured in an electronic device such as a mobile phone, a computer or a server. Figure 1 As shown, the method includes:

[0031] S110, inputting the image to be processed and the random noise into a pre-trained image conversion model, so as to perform feature splicing processing on the image to be processed and the random noise through a feature splicing module to obtain a spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing the target part.

[0032] The image to be processed may be a magnetic resonance slice image currently acquired and containing a target part. The target part may be a part that needs to be detected currently. For example, the target part may be the abdomen. The random noise may be a random Gaussian noise that obeys a preset Gaussian distribution. The image conversion model may be a model for converting a magnetic resonance slice image containing a target part into a computed tomography slice image. Optionally, the image conversion model may be a diffusion model. The image conversion model includes at least a feature splicing module and a denoising module.

[0033] The feature splicing module is used to perform feature splicing processing on the input image and the noise image. In this embodiment, the feature splicing module is used to splice the image to be processed and the noise image corresponding to the random noise. Correspondingly, the image obtained after the splicing process is the spliced ​​feature image to be processed. That is, the spliced ​​feature image to be processed can be an image obtained by performing feature splicing of the image to be processed and the noise image corresponding to the random noise. The denoising module of the image conversion model is used for image denoising processing.

[0034] Specifically, an image of a target part is collected by a magnetic resonance device to obtain a magnetic resonance image containing the target part. The magnetic resonance image is subjected to two-dimensional slicing processing to obtain a magnetic resonance slice image. The magnetic resonance slice image containing the target part is used as an image to be processed. Random noise that obeys a preset Gaussian distribution is obtained, and a noise image corresponding to the random noise is determined. The image to be processed and the noise image corresponding to the random noise are input into a pre-trained image conversion model, so that the image to be processed and the noise image corresponding to the random noise are subjected to feature splicing processing by a feature splicing module of the image conversion model to obtain a spliced ​​feature image to be processed.

[0035] S120, performing a denoising process on the spliced ​​feature image to be processed based on the denoising module to obtain a target image corresponding to the image to be processed, where the target image is a computed tomography slice image containing the target part.

[0036] Among them, the denoising module of the image conversion model is used to perform image denoising on the spliced ​​feature image to be processed.

[0037] Specifically, a denoising process is performed on the spliced ​​feature image to be processed through the denoising module of the image conversion model to obtain a computer tomography slice image containing the target part corresponding to the image to be processed, that is, the target image.

[0038] In an embodiment of the present invention, the image conversion model is trained based on sample magnetic resonance slice images and target noise feature images, and the target noise feature images are determined after multiple noise addition processes are performed on the sample tomographic slice images based on a pre-trained first model.

[0039] Among them, the sample magnetic resonance slice image can be a pre-acquired magnetic resonance slice image containing a target part. The first model can be a pre-trained model, which is used to output a computed tomography slice image corresponding to the magnetic resonance slice image based on the input magnetic resonance slice image. The first model may include an image denoising module, a feature splicing module and multiple denoising modules. The image denoising module is used to perform multiple image denoising processes on the input sample tomography slice image to obtain a target noise feature image. The sample tomography slice image can be a pre-acquired computed tomography slice image containing a target part. The feature splicing module is used to perform feature splicing processing on the noisy feature image and the input magnetic resonance slice image. Multiple denoising modules are used to perform multiple image denoising processes on the spliced ​​feature image.

[0040] Specifically, random noise that obeys a preset Gaussian distribution is obtained. Based on the random noise, the sample tomography slice image is subjected to multiple image noise processing according to the image noise module of the pre-trained first model to obtain a target noise feature image. The image conversion model to be trained is trained according to the target noise feature image and the sample magnetic resonance slice image to obtain a trained image conversion model. Based on this, the input image to be processed and the random noise can be processed according to the image conversion model to obtain a target image corresponding to the image to be processed.

[0041] The technical solution of this embodiment is to input the image to be processed and the random noise into the pre-trained image conversion model, and perform feature splicing processing on the image to be processed and the random noise according to the feature splicing module to obtain the spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing the target part; based on the denoising module, a denoising process is performed on the spliced ​​feature image to be processed to obtain a target image corresponding to the image to be processed, and the target image is a computed tomography slice image containing the target part; wherein the image conversion model is obtained by training based on the sample magnetic resonance slice image and the target noise feature image, and the target noise feature image is determined after multiple noise addition processing is performed on the sample tomography slice image based on the pre-trained first model. The present invention solves the problem of mode collapse caused by image conversion by generating adversarial networks in the prior art, and achieves the purpose of converting magnetic resonance slice images into tomography slice images. By performing a denoising process on the spliced ​​feature image, the computational overhead is reduced and the efficiency of image conversion is improved. The trained image conversion model is obtained based on the sample magnetic resonance slice image and the target noise feature image, and the accuracy and stability of the image conversion result are guaranteed.

[0042] Embodiment 2

[0043] Figure 2: is a flowchart of an image conversion model training method provided by the second embodiment of the present invention. This embodiment is based on the above embodiment. Before performing image conversion on the magnetic resonance slice image based on the image conversion model, it is necessary to train the image conversion model. The specific implementation method can refer to the technical solution of this embodiment. Among them, the technical terms that are the same or corresponding to the above embodiment are not repeated here. Figure 2 As shown, the method includes:

[0044] S210, obtaining a plurality of training samples, wherein the training samples include: a sample magnetic resonance slice image including a target part and a sample tomographic slice image corresponding to the sample magnetic resonance slice image.

[0045] The training samples may be composed of sample magnetic resonance slice images containing the target part and corresponding sample tomography slice images containing the target part. For example, the training samples may be composed of abdominal sample magnetic resonance (MRI) slice images and corresponding abdominal sample computed tomography (CT) slice images.

[0046] Specifically, before training the training image conversion model, multiple training samples can be obtained to train the model based on the training samples. In order to improve the accuracy of the model, as many and rich training samples as possible can be obtained. Multiple sample magnetic resonance images and sample tomographic images containing the target part are obtained. And the sample magnetic resonance images are subjected to data enhancement processing to obtain the sample magnetic resonance images after data enhancement. Optionally, the data enhancement processing may include image rotation, image translation, image elastic transformation, and image noise addition. It should be noted that if the data enhancement processing involves the movement or position change of the target part, the sample tomographic image is processed accordingly.

[0047] Since the sample magnetic resonance image is a three-dimensional image, the sample magnetic resonance image after data enhancement is subjected to two-dimensional slicing to obtain the sample magnetic resonance slice image after slicing. And the continuous sample magnetic resonance slice images are randomly extracted for model training. Correspondingly, the sample tomographic slice image is obtained by the above method. Based on this, a large number of rich training samples are constructed.

[0048] Exemplarily, taking the target part as the abdomen as an example, the sample magnetic resonance slice image containing the target part is the abdominal sample MRI slice image, and the sample tomography slice image containing the target part is the abdominal sample CT slice image. Obtain N training samples, wherein each training sample includes an abdominal sample MRI slice image and an abdominal sample CT slice image. Divide all training samples into a training set, a validation set, and a test set in a ratio of 8:1:1. Among them, the training set is used to train the model; the validation set is used to determine the best performing model from the trained models; and the test set is used to evaluate the model output results. It can be understood that an abdominal sample MRI image and a corresponding abdominal sample CT image are obtained. The abdominal sample MRI image is subjected to data enhancement processing such as image rotation, translation, and elastic transformation to obtain an abdominal sample MRI image after data enhancement. And if the data enhancement involves a change in the position of the abdominal part, the abdominal sample CT image is transformed accordingly.

[0049] Perform two-dimensional slicing on the abdominal sample MRI image after data enhancement, and randomly extract several consecutive abdominal sample MRI slice images. For example, three abdominal sample MRI slice images can be extracted. Perform two-dimensional slicing on the abdominal sample CT image, and obtain several consecutive abdominal sample CT slice images. For example, taking the i-th training sample as an example, the abdominal sample MRI slice image is recorded as The abdominal sample CT slice image is recorded as

[0050] S220: Input the training sample into a pre-trained first model to obtain a target noise feature image corresponding to the sample tomographic slice image and a first output image corresponding to the sample magnetic resonance slice image.

[0051] The first model includes an image denoising module, a feature splicing module and a plurality of denoising modules. The target noise feature image can be a noise image obtained by performing multiple noise additions on the sample tomographic slice image. The first output image can be a computed tomographic slice image obtained by performing image conversion on the sample magnetic resonance slice image.

[0052] Specifically, the sample magnetic resonance slice image and the sample tomographic slice image are input into the first model, and the sample tomographic slice image is subjected to multiple image noise processing by the image denoising module of the first model to obtain a target noise feature image. The target noise feature image and the sample magnetic resonance slice image are subjected to feature splicing processing in the channel dimension by the feature splicing module of the first model to obtain a spliced ​​feature image. The spliced ​​feature image is subjected to multiple denoising processing by the multiple denoising modules of the first model to obtain a first output image.

[0053] Exemplarily, in combination with the above example, taking random noise as ε (i) Take 2 as an example to illustrate. and Input into the first model, and the image noise module of the first model is used to Perform forward noise processing, that is, use random noise ε (i) right Perform T consecutive image noise processing to obtain the target noise feature image For the target noise feature image and Perform feature concatenation processing on the channel dimension. For example, The corresponding feature size is 3*256*256, The corresponding feature size is 3*256*256, so the feature size of the spliced ​​feature image is 6*256*256. The spliced ​​feature image is subjected to multiple reverse denoising processes to obtain the first output image.

[0054] In an embodiment of the present invention, before obtaining the target noise feature image according to the trained first model, it is necessary to train the first model. The first model includes an image denoising module, a feature splicing module and a plurality of denoising modules. The training method of the first model may be: inputting a sample magnetic resonance slice image and a sample tomographic slice image into the first model to be trained, so as to perform multiple image denoising processes on the sample tomographic slice image according to random noise through the image denoising module, and obtain a denoised feature image after each image denoising; for multiple denoised feature images, perform feature splicing processes on the current denoised feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be used; for multiple spliced ​​feature images to be used, perform denoising processes on the current spliced ​​feature image to be used based on the denoising module, and determine the actual predicted noise, wherein the actual predicted noise is the noise required to denoise the previous spliced ​​feature image to be used obtained through the current spliced ​​feature image to be used; based on the actual predicted noise and the random noise, determine the loss value of the first model to be trained; based on the loss value, update the model parameters of the first model to be trained to obtain the trained first model.

[0055] The noise-added characteristic image may be a characteristic image obtained after image noise addition processing is performed on the sample tomographic slice image each time. The spliced ​​characteristic image to be used may be an image obtained after feature splicing of the noise-added characteristic image and the sample magnetic resonance slice image. The actual predicted noise may be the noise required to be denoised for obtaining the previous spliced ​​characteristic image to be used through the current spliced ​​characteristic image to be used. The loss value may be the difference between the actual predicted noise and the random noise.

[0056] Specifically, the sample magnetic resonance slice image and the sample tomographic slice image are input into the first model to be trained, and the sample tomographic slice image is subjected to multiple image denoising processes through the image denoising module of the first model to obtain a noisy feature image after each denoising. For multiple noisy feature images, the current noisy feature image and the sample magnetic resonance slice image are subjected to feature splicing process to obtain a spliced ​​feature image to be used. Based on this, multiple spliced ​​feature images to be used are obtained. For multiple spliced ​​feature images to be used, the current feature splicing image to be used is denoised according to the current denoising module to determine the noise required for denoising the previous spliced ​​feature image to be used, that is, the actual predicted noise. Based on the actual predicted noise and the random noise, the loss value of the first model to be trained is determined.

[0057] When using the loss value to correct the model parameters in the first model to be trained, the convergence of the loss function can be used as a training goal, such as whether the training error is less than the preset error, or whether the error change tends to be stable, or whether the current number of iterations is equal to the preset number. If the detection reaches the convergence condition, such as the training error of the loss function is less than the preset error, or the error change trend tends to be stable, it indicates that the training of the first model to be trained is completed, and the iterative training can be stopped at this time. If it is detected that the convergence condition is not met at present, other training samples can be further obtained to continue training the first model to be trained until the training error of the loss function is within the preset range. When the training error of the loss function reaches convergence, the trained first model can be obtained.

[0058] For example, in combination with the above examples, and The image is input into the first model to be trained for training, and the image noise module of the first model to be trained is used to generate random noise ε (i) right Perform forward noise processing, that is, use random noise ε (i) right Perform T consecutive image denoising processes to obtain the denoised feature image after each denoising. Determine the noise feature image after the Tth noise addition The corresponding noise image, that is, the target noise feature image Will and Perform feature splicing processing on the channel dimension to obtain the spliced ​​feature image to be used. Perform denoising processing on the current feature splicing image to be used according to the denoising module of the first model to be trained to determine the actual predicted noise required to denoise the previous spliced ​​feature image to be used. calculate and ε (i)The mean square error loss between them is used to obtain the loss value. The model parameters of the first model to be trained are updated according to the loss value to obtain the trained first model. It should be noted that multiple trained first models can be obtained according to the above method, and the multiple trained first models are verified by the training samples in the verification set to obtain the first model with the best model performance.

[0059] In an embodiment of the present invention, the method for determining the first output image may specifically be: performing multiple image denoising processes on the sample tomographic scan slice image to obtain a target noise feature image; performing feature splicing processing on the target noise feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be input; and performing denoising processing on the input spliced ​​feature image in sequence based on multiple denoising modules to obtain a first output image.

[0060] The feature image to be input for feature splicing can be understood as an image obtained by feature splicing the target noise feature image and the sample magnetic resonance slice image.

[0061] Specifically, the sample tomographic slice image is subjected to multiple image denoising processes through the trained image denoising module of the first model to obtain a target noise feature image. The target noise feature image and the sample magnetic resonance slice image are subjected to feature splicing processes through the feature splicing module of the first model to obtain a spliced ​​feature image to be input. For multiple denoising modules of the first model, the input feature splicing image is subjected to denoising processes through the current denoising module to determine the current predicted noise. The input feature splicing feature image is subjected to denoising processes through the current predicted noise to obtain a spliced ​​feature image to be input after one denoising. Based on the above denoising process, the input feature splicing feature image is subjected to denoising processes in turn through multiple denoising modules until all denoising modules perform image denoising processes to obtain a first output image.

[0062] For example, in conjunction with the above examples, see Figure 3 , Figure 3 middle The feature size of is h×w×d, and the feature size of the target noise feature image is h×w×d. and Input into the first model, and the image noise module of the first model is used to Perform forward noise processing, that is, use random noise ε (i) right Perform T consecutive image noise processing to obtain the target noise feature image For the target noise feature image and Perform feature splicing processing on the channel dimension to obtain the input splicing feature image. Process the input splicing feature image through the current denoising module to obtain the current predicted noise use Perform denoising on the input splicing feature image to obtain the input splicing feature image after denoising By analogy, multiple denoising modules are used to perform multiple denoising processes on the input spliced ​​feature image to obtain the first output image

[0063] S230: Input the sample magnetic resonance slice image and the target noise feature image into the image conversion model to be trained to obtain a second output image.

[0064] The image conversion model to be trained includes a feature splicing module, a denoising module and a discriminator module. The second output image may be a computed tomography slice image output by the image conversion model to be trained.

[0065] Specifically, the sample magnetic resonance slice image and the target noise feature image are input into the image conversion model to be trained, and the sample magnetic resonance slice image and the target noise feature image are subjected to feature splicing processing by the feature splicing module of the image conversion model to be trained to obtain the spliced ​​feature image to be input. The input spliced ​​feature image is subjected to denoising processing by the denoising module of the image conversion model to be trained to obtain a second output image.

[0066] In an embodiment of the present invention, a method for determining the second output image may be: performing feature splicing processing on the target noise feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be input; and performing an image denoising process on the input spliced ​​feature image based on a denoising module to obtain a second output image.

[0067] Specifically, feature splicing is performed on the target noise feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be input, and image denoising is performed on the spliced ​​feature image to be input by a denoising module to obtain a second output image.

[0068] For example, in conjunction with the above examples, see Figure 3 . The target noise feature image And sample MRI slice images The input is used in the image conversion model to be trained, and the sample magnetic resonance slice image is processed by the feature splicing module of the image conversion model to be trained. and target noise feature image Perform feature splicing processing to obtain the input splicing feature image. Perform a denoising process on the input splicing feature image through the denoising module of the image conversion model to be trained to obtain the second output image.

[0069] S240: Determine an image generation loss and an image discrimination loss of the image conversion model to be trained based on the first output image and the second output image.

[0070] The image generation loss may be the loss corresponding to the feature concatenation module and the denoising module of the image conversion model to be trained. The image discrimination loss may be the loss corresponding to the discriminator module of the image conversion model to be trained.

[0071] Specifically, according to the difference between the first output image and the second output image, the image generation loss of the image conversion model to be trained is determined. The first output image and the second output image are discriminated by the discriminator module of the image conversion model to be trained to obtain a discrimination result. According to the discrimination result, the image discrimination loss of the image conversion model to be trained is determined.

[0072] In an embodiment of the present invention, the image generation loss includes a first loss, a second loss, a third loss and a fourth loss. The method for determining the image generation loss may be: determining the first loss based on the first output image and the second output image; performing image denoising processing on the first output image according to random noise to obtain the first output image after denoising, and using the denoising module in the first model to denoise the first output image after denoising to obtain the third output image; performing image denoising processing on the second output image according to random noise to obtain the second output image after denoising, and using the denoising module in the first model to denoise the second output image after denoising to obtain the fourth output image; determining the second loss based on the third output image and the fourth output image; performing discriminative processing on the first output image and the second output image based on the discriminator module to obtain the first discriminative result, and determining the third loss based on the first discriminative result; performing discriminative processing on the third output image and the fourth output image based on the discriminator module to obtain the second discriminative result, and determining the fourth loss based on the second discriminative result.

[0073] The first loss is used to characterize the degree of difference between the first output image and the second output image.

[0074] The third output image may be an image obtained by denoising the first output image after adding noise.

[0075] The fourth output image may be an image obtained by denoising the second output image after adding noise. The second loss is used to characterize the degree of difference between the third output image and the fourth output image. The first discrimination result may be the result of discriminating the first output image as true and the second output image as true. The third loss may be a loss corresponding to the first discrimination result. The second discrimination result may be the result of discriminating the third output image as true and the fourth output image as true. The fourth loss may be a loss corresponding to the second discrimination result.

[0076] Specifically, the first loss is determined according to the degree of difference between the first output image and the second output image. The first output image is subjected to denoising by random noise to obtain the first output image after denoising. The first output image after denoising is subjected to noise prediction and denoising according to the denoising module of the trained first model to obtain the third output image. The second output image is subjected to denoising by random noise to obtain the second output image after denoising. The second output image after denoising is subjected to noise prediction and denoising according to the denoising module of the first model to obtain the fourth output image. The second loss is determined according to the degree of difference between the third output image and the fourth output image.

[0077] The first output image and the second output image are processed according to the discriminator module of the image conversion model to be trained, and the third loss corresponding to the first output image is determined when the first output image is judged to be true and the second output image is judged to be true. The third output image and the fourth output image are processed according to the discriminator module of the image conversion model to be trained, and the fourth loss corresponding to the third output image is determined when the third output image is judged to be true and the fourth output image is judged to be true.

[0078] Optionally, the first loss, the second loss, the third loss and the fourth loss are added to obtain the image generation loss of the image conversion model to be trained, so as to correct the model parameters of the feature splicing module and the denoising module of the image conversion model to be trained based on the image generation loss.

[0079] Exemplarily, in combination with the above example, the first output image is calculated using the mean square error function and the second output image The first loss L_1 between . Using random noise ε (i) For the first output image Perform noise processing to obtain a first output image after noise addition. Perform noise prediction and noise reduction processing on the first output image after noise addition according to the noise reduction module of the first model to obtain a third output image. Using random noise ε (i) For the second output image The noise is added to obtain a second output image after the noise is added. The noise is predicted and denoised according to the denoising module of the first model on the second output image after the noise is added to obtain a fourth output image. Calculate the third output image using the mean square error function and the fourth output image The second loss between them is L_2.

[0080] The first output image Input to the discriminator module of the image conversion model to be trained, predict its true loss, and convert the second output image Input it into the discriminator module of the image conversion model to be trained, and predict its true loss. The above loss is recorded as the third loss L_3.

[0081] The third output image Input to the discriminator module of the image conversion model to be trained, predict its true loss, and convert the fourth output image Input it into the discriminator module of the image conversion model to be trained, and predict its true loss. The above loss is recorded as the fourth loss L_4.

[0082] The first loss L_1, the second loss L_2, the third loss L_3 and the fourth loss L_4 are added together to obtain the image generation loss, so as to correct the model parameters of the image conversion model to be trained based on the image generation loss.

[0083] In an embodiment of the present invention, the method for determining the image discrimination loss can be: based on the discriminator module, the first output image and the second output image are discriminated and processed to obtain a third discrimination result, and the third output image and the fourth output image are discriminated and processed to obtain a fourth discrimination result; according to the third discrimination result and the fourth discrimination result, the image discrimination loss corresponding to the discriminator module is determined.

[0084] The third discrimination result may be a result of discriminating the first output image as true and the second output image as false. The fourth discrimination result may be a result of discriminating the third output image as true and the fourth output image as false. The image discrimination loss may be a loss corresponding to the third discrimination result and the fourth discrimination result.

[0085] Specifically, the first output image and the second output image are processed according to the discriminator module of the image conversion model to be trained, and the corresponding loss is determined when the first output image is judged to be true and the second output image is judged to be false. Also, the third output image and the fourth output image are processed according to the discriminator module of the image conversion model to be trained, and the corresponding loss is determined when the third output image is judged to be true and the fourth output image is judged to be false. The image discrimination loss is obtained according to the above two losses.

[0086] Exemplarily, in combination with the above example, the first output image Input to the discriminator module of the image conversion model to be trained, predict its true loss, and convert the second output image Input to the discriminator module of the image conversion model to be trained, predict its false loss, and convert the third output image Input to the discriminator module of the image conversion model to be trained, predict its true loss, and convert the fourth output image Input it into the discriminator module of the image conversion model to be trained, predict its false loss, and get the image discrimination loss based on the above loss.

[0087] S250, updating the model parameters in the image conversion model to be trained according to the image generation loss and the image discrimination loss to obtain a trained image conversion model to be trained.

[0088] Specifically, when the model parameters in the image conversion model to be trained are corrected using the image generation loss and the image discrimination loss, the convergence of the loss function can be used as a training goal, such as whether the training error is less than the preset error, or whether the error change tends to be stable, or whether the current number of iterations is equal to the preset number. If the detection reaches the convergence condition, such as the training error of the loss function is less than the preset error, or the error change trend tends to be stable, it indicates that the training of the image conversion model to be trained is completed, and the iterative training can be stopped at this time. If it is detected that the convergence condition is not met at present, other training samples can be further obtained to continue training the image conversion model to be trained until the training error of the loss function is within the preset range. When the training error of the loss function reaches convergence, the trained image conversion model can be obtained, that is, at this time, after the image to be processed including the target part and random noise are input into the image conversion model, the target image can be accurately obtained.

[0089] S260: Remove the discriminator module in the trained image conversion model to be trained to obtain the image conversion model.

[0090] Specifically, the discriminator module in the trained image conversion model to be trained is eliminated to obtain the image conversion model, so as to realize the conversion processing between the magnetic resonance slice image and the computer tomography slice image through the image conversion model.

[0091] The technical solution of this embodiment trains the image conversion model to be trained by using the target noise feature image and the sample magnetic resonance slice image output by the first model to obtain a trained image conversion model, shortens the processing process of the image conversion model, reduces the computational overhead of the image conversion process, and speeds up the prediction speed of the image conversion model. Multiple losses are determined by the first output image and the second output image, and the model parameters of the image conversion model to be trained are corrected according to the multiple losses, thereby ensuring the robustness of the image conversion model and improving the conversion accuracy and stability of the image conversion model.

[0092] Embodiment 3

[0093] Figure 4 FIG. 1 is a schematic diagram of the structure of an image conversion processing device provided by Embodiment 3 of the present invention. Figure 4 As shown, the device includes: a feature image determination module 310 and a target image determination module 320.

[0094] The feature image determination module 310 is used to input the image to be processed and the random noise into a pre-trained image conversion model, so as to perform feature splicing processing on the image to be processed and the random noise through the feature splicing module to obtain a spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing the target part; the target image determination module 320 is used to perform a denoising process on the spliced ​​feature image to be processed based on the denoising module to obtain a target image corresponding to the image to be processed, and the target image is a computed tomography slice image containing the target part; wherein the image conversion model is trained based on the sample magnetic resonance slice image and the target noise feature image, and the target noise feature image is determined after multiple noise addition processes are performed on the sample tomography slice image based on the pre-trained first model.

[0095] The technical solution of this embodiment is to input the image to be processed and the random noise into the pre-trained image conversion model, and perform feature splicing processing on the image to be processed and the random noise according to the feature splicing module to obtain the spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing the target part; based on the denoising module, a denoising process is performed on the spliced ​​feature image to be processed to obtain a target image corresponding to the image to be processed, and the target image is a computed tomography slice image containing the target part; wherein the image conversion model is obtained by training based on the sample magnetic resonance slice image and the target noise feature image, and the target noise feature image is determined after multiple noise addition processing is performed on the sample tomography slice image based on the pre-trained first model. The present invention solves the problem of mode collapse caused by image conversion by generating adversarial networks in the prior art, and achieves the purpose of converting magnetic resonance slice images into tomography slice images. By performing a denoising process on the spliced ​​feature image, the computational overhead is reduced and the efficiency of image conversion is improved. The trained image conversion model is obtained based on the sample magnetic resonance slice image and the target noise feature image, and the accuracy and stability of the image conversion result are guaranteed.

[0096] On the basis of the above embodiment, optionally, the device further includes: an image conversion model training module, which includes: a sample acquisition unit, which is used to acquire multiple training samples, wherein the training samples include: a sample magnetic resonance slice image containing a target part and a sample tomographic slice image corresponding to the sample magnetic resonance slice image; a first output image determination unit, which is used to input the training sample into a pre-trained first model to obtain a target noise feature image corresponding to the sample tomographic slice image and a first output image corresponding to the sample magnetic resonance slice image; a second output image determination unit, which is used to input the sample magnetic resonance slice image and the target noise feature image into the image conversion model to be trained to obtain a second output image; a loss determination unit, which is used to determine the image generation loss and the image discrimination loss of the image conversion model to be trained based on the first output image and the second output image; a model parameter updating unit, which is used to update the model parameters in the image conversion model to be trained according to the image generation loss and the image discrimination loss to obtain a trained image conversion model to be trained; an image conversion model determination unit, which is used to eliminate the discriminator module in the trained image conversion model to be trained to obtain an image conversion model.

[0097] Optionally, the first model includes an image denoising module, a feature splicing module and a plurality of denoising modules, and the device also includes a first model training module, which includes: an image denoising unit, used to input a sample magnetic resonance slice image and a sample tomographic slice image into the first model to be trained, so as to perform multiple image denoising processes on the sample tomographic slice image according to random noise through the image denoising module to obtain a noisy feature image after each image denoising; a feature splicing unit, used to perform feature splicing processing on the current noisy feature image and the sample magnetic resonance slice image for multiple noisy feature images to obtain A spliced ​​feature image to be used; a predicted noise determination unit, used to perform denoising on a current spliced ​​feature image to be used based on a denoising module for a plurality of spliced ​​feature images to be used, and determine an actual predicted noise, wherein the actual predicted noise is the noise required to be denoised for obtaining a previous spliced ​​feature image to be used through a current spliced ​​feature image to be used; a loss value determination unit, used to determine a loss value of a first model to be trained based on the actual predicted noise and random noise; a first model determination unit, used to update the model parameters of the first model to be trained based on the loss value, and obtain a trained first model.

[0098] Optionally, the first output image determination unit is used to perform multiple image noise addition processes on the sample tomographic scan slice image to obtain a target noise feature image; perform feature splicing processing on the target noise feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be input; and perform denoising processing on the input spliced ​​feature image in sequence based on multiple denoising modules to obtain a first output image.

[0099] Optionally, the second output image determination unit is used to perform feature splicing processing on the target noise feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be input; and perform an image denoising process on the input spliced ​​feature image based on the denoising module to obtain a second output image.

[0100] Optionally, the image generation loss includes a first loss, a second loss, a third loss and a fourth loss. The loss determination unit includes: an image generation loss determination subunit, which is used to determine the first loss based on the first output image and the second output image; perform image denoising on the first output image according to random noise to obtain the first output image after denoising, and use the denoising module in the first model to denoise the first output image after denoising to obtain the third output image; perform image denoising on the second output image according to random noise to obtain the second output image after denoising, and use the denoising module in the first model to denoise the second output image after denoising to obtain the fourth output image; determine the second loss based on the third output image and the fourth output image; perform discriminative processing on the first output image and the second output image based on the discriminator module to obtain a first discriminative result, and determine the third loss based on the first discriminative result; perform discriminative processing on the third output image and the fourth output image based on the discriminator module to obtain a second discriminative result, and determine the fourth loss based on the second discriminative result.

[0101] Optionally, the loss determination unit includes an image discrimination loss determination subunit, which is used to perform discrimination processing on the first output image and the second output image based on the discriminator module to obtain a third discrimination result, and to perform discrimination processing on the third output image and the fourth output image to obtain a fourth discrimination result; based on the third discrimination result and the fourth discrimination result, determine the image discrimination loss corresponding to the discriminator module.

[0102] The image conversion processing device provided in the embodiment of the present invention can execute the image conversion processing method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0103] Embodiment 4

[0104] Figure 51 is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0105] like Figure 5 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0106] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0107] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs the various methods and processes described above, such as an image conversion processing method.

[0108] In some embodiments, the image conversion processing method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the image conversion processing method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the image conversion processing method in any other appropriate manner (e.g., by means of firmware).

[0109] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0110] The computer program for implementing the image conversion processing method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program can be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as an independent software package, or completely on a remote machine or server.

[0111] Embodiment 5

[0112] Embodiment 5 of the present invention further provides a computer-readable storage medium, the computer-readable storage medium stores computer instructions, the computer instructions are used to enable a processor to execute an image conversion processing method, the method is processed by an image conversion model, the image conversion model includes at least a feature splicing module and a denoising module, the method includes:

[0113] The image to be processed and the random noise are input into a pre-trained image conversion model, so as to perform feature splicing processing on the image to be processed and the random noise through a feature splicing module to obtain a spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing a target part; based on the denoising module, a denoising processing is performed on the spliced ​​feature image to be processed to obtain a target image corresponding to the image to be processed, and the target image is a computed tomography slice image containing the target part; wherein the image conversion model is trained based on a sample magnetic resonance slice image and a target noise feature image, and the target noise feature image is determined after multiple noise addition processing is performed on the sample tomography slice image based on the pre-trained first model.

[0114] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0115] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0116] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0117] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0118] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0119] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An image conversion processing method, characterized in that: The method is processed by an image conversion model, the image conversion model at least includes a feature splicing module and a denoising module, and the method includes: Inputting the image to be processed and the random noise into a pre-trained image conversion model, so as to perform feature splicing processing on the image to be processed and the random noise through the feature splicing module to obtain a spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing a target part; Performing a denoising process on the to-be-processed spliced ​​feature image based on the denoising module to obtain a target image corresponding to the to-be-processed image, wherein the target image is a computed tomography slice image containing the target part; The image conversion model is trained based on sample magnetic resonance slice images and target noise feature images, and the target noise feature images are determined after multiple noise addition processes are performed on sample tomographic slice images based on a pre-trained first model.

2. The method according to claim 1, characterized in that The image conversion model to be trained also includes a discriminator module, and the method further includes: Acquire a plurality of training samples, wherein the training samples include: a sample magnetic resonance slice image including the target part and a sample tomographic slice image corresponding to the sample magnetic resonance slice image; Inputting the training sample into a pre-trained first model to obtain a target noise feature image corresponding to the sample tomographic slice image and a first output image corresponding to the sample magnetic resonance slice image; Inputting the sample magnetic resonance slice image and the target noise feature image into the image conversion model to be trained to obtain a second output image; Determining an image generation loss and an image discrimination loss of the image conversion model to be trained based on the first output image and the second output image; The model parameters in the image conversion model to be trained are updated according to the image generation loss and the image discrimination loss to obtain a trained image conversion model to be trained; The discriminator module in the trained image conversion model to be trained is eliminated to obtain an image conversion model.

3. The method according to claim 2, characterized in that The first model includes an image denoising module, a feature splicing module and a plurality of denoising modules, and the method further includes: Inputting the sample magnetic resonance slice image and the sample tomographic slice image into the first model to be trained, so as to perform multiple image noise addition processes on the sample tomographic slice image according to random noise through the image noise addition module, and obtaining a noisy feature image after each image noise addition; For the plurality of the noise-added feature images, feature splicing processing is performed on the current noise-added feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be used; For the plurality of splicing feature images to be used, based on the denoising module, a denoising process is performed on the current splicing feature image to be used to determine an actual predicted noise, wherein the actual predicted noise is the noise required to be denoised for obtaining the previous splicing feature image to be used through the current splicing feature image to be used; Determining a loss value of the first model to be trained based on the actual prediction noise and the random noise; Based on the loss value, the model parameters of the first model to be trained are updated to obtain a trained first model.

4. The method according to claim 2, characterized in that: The step of inputting the training sample into a pre-trained first model to obtain a target noise feature image corresponding to the sample tomographic slice image and a first output image corresponding to the sample magnetic resonance slice image comprises: Performing multiple image noise addition processes on the sample tomographic scan slice image to obtain a target noise characteristic image; Performing feature splicing processing on the target noise feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be input; Based on multiple denoising modules, denoising processing is performed on the input splicing feature image in sequence to obtain a first output image.

5. The method according to claim 2, characterized in that: The step of inputting the sample magnetic resonance slice image and the target noise feature image into the image conversion model to be trained to obtain a second output image comprises: Performing feature splicing processing on the target noise feature image and the sample magnetic resonance slice image to obtain a spliced ​​feature image to be input; Based on the denoising module, an image denoising process is performed on the input splicing feature image to obtain a second output image.

6. The method according to claim 2, characterized in that The image generation loss includes a first loss, a second loss, a third loss, and a fourth loss, and determining the image generation loss of the image conversion model to be trained based on the first output image and the second output image includes: determining the first loss based on the first output image and the second output image; Performing image noise processing on the first output image according to the random noise to obtain a first output image after noise is added, and performing denoising processing on the first output image after noise is added using a denoising module in the first model to obtain a third output image; Performing image noise processing on the second output image according to the random noise to obtain a noisy second output image, and performing denoising processing on the noisy second output image using a denoising module in the first model to obtain a fourth output image; determining the second loss based on the third output image and the fourth output image; Performing discriminative processing on the first output image and the second output image based on the discriminator module to obtain a first discriminative result, and determining the third loss according to the first discriminative result; The third output image and the fourth output image are discriminated based on the discriminator module to obtain a second discriminant result, and the fourth loss is determined according to the second discriminant result.

7. The method according to claim 6, characterized in that Determining the image discrimination loss of the image conversion model to be trained based on the first output image and the second output image includes: performing discrimination processing on the first output image and the second output image based on the discriminator module to obtain a third discrimination result, and performing discrimination processing on the third output image and the fourth output image to obtain a fourth discrimination result; An image discrimination loss corresponding to the discriminator module is determined according to the third discrimination result and the fourth discrimination result.

8. An image conversion processing device, characterized in that: The device is processed by an image conversion model, the image conversion model at least includes a feature splicing module and a denoising module, and the device includes: A feature image determination module is used to input the image to be processed and the random noise into a pre-trained image conversion model, so as to perform feature splicing processing on the image to be processed and the random noise through the feature splicing module to obtain a spliced ​​feature image to be processed; wherein the image to be processed is a magnetic resonance slice image containing a target part; A target image determination module, configured to perform a denoising process on the to-be-processed spliced ​​feature image based on the denoising module to obtain a target image corresponding to the to-be-processed image, wherein the target image is a computed tomography slice image containing the target part; The image conversion model is trained based on sample magnetic resonance slice images and target noise feature images, and the target noise feature images are determined after multiple noise addition processes are performed on sample tomographic slice images based on a pre-trained first model.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the image conversion processing method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the image conversion processing method according to any one of claims 1 to 7 when executed.

Images