CN115115736A

CN115115736A - Image artifact removing method, device and equipment and storage medium

Info

Publication number: CN115115736A
Application number: CN202210929449.0A
Authority: CN
Inventors: 王红; 郑冶枫; 孟德宇
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-09-09
Filing date: 2022-08-03
Publication date: 2022-09-27
Also published as: CN114283216A

Abstract

The embodiment of the application discloses a method, a device, equipment and a storage medium for removing artifacts from an image, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring an artifact-bearing medical image and a polluted chord map, wherein the artifact-bearing medical image and the polluted chord map are a group of corresponding images aiming at the same object; executing N stages of iterative processing based on the medical image with the artifact and the polluted chord graph to generate a final artifact-removed medical image corresponding to the medical image with the artifact; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1. According to the method, a combined chord graph and medical image reconstruction model is constructed, then an iterative solution algorithm is provided, one of the iterative solution algorithms is correspondingly expanded into a depth network structure, reconstruction of the medical image with the artifact and the polluted chord graph is achieved, and the artifact removing effect of the image can be improved.

Description

Image artifact removing method, device and equipment and storage medium

The present application claims priority from chinese patent application No. 202111057910.X entitled "image deghost method, apparatus, device and storage medium" filed on 09.09.09.2021, which is incorporated herein by reference in its entirety.

Technical Field

The embodiment of the application relates to the technical field of artificial intelligence, in particular to an image artifact removing method, device, equipment and storage medium.

Background

CT (Computed Tomography) images reconstructed on the basis of X-ray projections play a very important role in clinical diagnosis and treatment planning.

When metal objects (such as false teeth and hip joint prostheses) are carried in a patient body, due to the absorption of X-rays by metal, the CT image can present artifacts such as stripes and shadows, and thus clinical diagnosis can be influenced. In the related art, a dual domain network is provided, which includes a chord graph processing module and a CT image processing model. Firstly, repairing the chord graph through a chord graph processing module to obtain a repaired chord graph; and (3) carrying out projection after filtering on the repaired chord graph to obtain a reconstructed CT image, then processing the reconstructed CT image through a CT image processing module, and finally outputting the reconstructed artifact-removed CT image.

However, the above-described deghosting effect in this way is still not ideal.

Disclosure of Invention

The embodiment of the application provides an image artifact removing method, device and equipment and a storage medium, which can improve the artifact removing effect of an image. The technical scheme is as follows:

according to an aspect of an embodiment of the present application, there is provided an image deghost method, including:

acquiring an artifact-bearing medical image and a contaminated chordal map, the artifact-bearing medical image and the contaminated chordal map being a set of corresponding images for a same object;

based on the artifact medical image and the polluted chord graph, executing N stages of iterative processing to generate a final artifact removing medical image corresponding to the artifact medical image; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1.

According to an aspect of the embodiments of the present application, there is provided a training method of a deghost model, the method including:

obtaining sample data, wherein the sample data comprises a clean medical image sample and a clean chord chart sample, and the clean medical image sample and the clean chord chart sample are a group of corresponding images aiming at the same object;

generating a contaminated chordal map sample and an artifact-bearing medical image sample based on the clean medical image sample;

executing N stages of iterative processing based on the artifact-carrying medical image sample and the polluted chord map sample by adopting the artifact removing model to generate a final artifact-removing medical image corresponding to the artifact-carrying medical image sample; the iteration processing of each stage is used for repairing the chord graph and removing artifacts of the medical image, and N is an integer greater than 1;

calculating a training loss of the artifact-removed model from the clean medical image sample and the final artifact-removed medical image;

training the deghost model based on the training loss.

According to an aspect of the embodiments of the present application, there is provided an image deghost apparatus, including:

an image acquisition module for acquiring an artifact-bearing medical image and a contaminated chordal map, the artifact-bearing medical image and the contaminated chordal map being a set of corresponding images for a same object;

an artifact removing module, configured to perform N stages of iterative processing based on the artifact-bearing medical image and the contaminated chord graph, and generate a final artifact-removed medical image corresponding to the artifact-bearing medical image; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1.

According to an aspect of the embodiments of the present application, there is provided an apparatus for training a deghost model, the apparatus including:

a sample acquisition module for acquiring sample data, the sample data comprising a clean medical image sample and a clean chord chart sample, the clean medical image sample and the clean chord chart sample being a set of corresponding images for a same object;

a sample processing module for generating a contaminated chordal map sample and an artifact-bearing medical image sample based on the clean medical image sample;

the artifact removing module is used for executing N stages of iterative processing based on the artifact-carrying medical image sample and the polluted chord map sample by adopting the artifact removing model to generate a final artifact removing medical image corresponding to the artifact-carrying medical image sample; the iterative processing of each stage is used for repairing the chord graph and removing artifacts of the medical image, and N is an integer greater than 1;

a loss calculation module for calculating a training loss of the deghost model based on the clean medical image sample and the final deghost medical image;

a model training module to train the deghost model based on the training loss.

According to an aspect of the embodiments of the present application, there is provided a computer device, including a processor and a memory, where at least one instruction, at least one program, a code set, or a set of instructions is stored in the memory, and the at least one instruction, the at least one program, the code set, or the set of instructions is loaded and executed by the processor to implement the image deghosting method or the training method for the deghosting model.

According to an aspect of embodiments of the present application, there is provided a computer-readable storage medium having at least one instruction, at least one program, a set of codes, or a set of instructions stored therein, which is loaded and executed by a processor to implement the image deghost method or the training method for deghost model.

According to an aspect of embodiments herein, there is provided a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to cause the computer device to execute the image deghosting method or to implement the training method of the deghosting model.

The technical scheme provided by the embodiment of the application can have the following beneficial effects:

the method comprises the steps of carrying out N stages of iterative processing (N is an integer larger than 1) on a medical image with an artifact and a polluted chord graph corresponding to the medical image with the artifact to generate a final artifact-removed medical image corresponding to the medical image with the artifact, wherein the iterative processing of each stage is used for carrying out restoration processing on the chord graph and carrying out artifact-removal processing on the medical image, so that mutual constraint between the chord graph and the medical image is realized, information interaction between the chord graph and the medical image is more sufficient, the artifact-removal processing effect is improved, and the method is more beneficial to operation of subsequent clinical tasks.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a schematic diagram of a deghost method provided by an embodiment of the present application;

FIG. 2 is a schematic illustration of an environment for implementing an embodiment provided by an embodiment of the present application;

FIG. 3 is a flowchart of an image deghost method according to an embodiment of the present application;

FIG. 4 is a schematic structural diagram of a deghost model according to an embodiment of the present application;

FIG. 5 is a flowchart of an image deghost method according to another embodiment of the present application;

FIG. 6 is a schematic structural diagram of a deghost model according to another embodiment of the present disclosure;

FIG. 7 is a schematic diagram of a deghost method provided by another embodiment of the present application;

FIG. 8 is a flowchart of an image deghost method according to another embodiment of the present application;

FIG. 9 is a schematic structural diagram of a deghost model according to another embodiment of the present disclosure;

FIG. 10 is a schematic diagram of a channel cascading and separating operation provided by one embodiment of the present application;

FIG. 11 is a schematic diagram of a channel cascade and separation operation provided by another embodiment of the present application;

FIG. 12 is a flowchart of a deghost model training method provided by an embodiment of the present application;

FIG. 13 is a flowchart of an experiment for constructing a deghost model according to an embodiment of the present disclosure;

FIG. 14 is a graph illustrating experimental results of a deghost model provided in one embodiment of the present application;

FIG. 15 is a block diagram of an image deghost apparatus according to an embodiment of the present application;

fig. 16 is a block diagram of an image deghost apparatus according to another embodiment of the present application;

FIG. 17 is a block diagram of a training apparatus for a deghost model according to an embodiment of the present application;

FIG. 18 is a block diagram of a training apparatus for a deghost model according to another embodiment of the present application;

fig. 19 is a block diagram of a computer device according to an embodiment of the present application.

Detailed Description

Embodiments of the present application will be described in further detail below with reference to the accompanying drawings.

Artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

Computer Vision technology (CV) is a science for researching how to make a machine "look", and more specifically, it refers to that a camera and a Computer are used to replace human eyes to perform machine Vision such as identification and measurement on a target, and further image processing is performed, so that the Computer processing becomes an image more suitable for human eyes to observe or is transmitted to an instrument to detect. As a scientific discipline, computer vision research-related theories and techniques attempt to build artificial intelligence systems that can capture information from images or multidimensional data. Computer vision technologies generally include image segmentation, image Recognition, image semantic understanding, image retrieval, OCR (Optical Character Recognition), video processing, video semantic understanding, video content/behavior Recognition, three-dimensional object reconstruction, 3D (three-dimensional) technology, virtual reality, augmented reality, synchronous positioning, map construction, and other technologies, and further include common biometric technologies such as face Recognition and fingerprint Recognition.

Machine Learning (ML) is a multi-domain cross discipline, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning.

With the research and development of artificial intelligence technology, the artificial intelligence technology is developed and researched in a plurality of fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, unmanned aerial vehicles, robots, smart medical services, smart customer service and the like.

As shown in fig. 1, the scheme provided in the embodiment of the present application relates to the field of computer vision technology and machine learning technology, and a computer vision technology and machine learning technology are used for training to obtain an artifact removal model, and then the artifact removal model is used for performing artifact removal processing on a medical image with an artifact and a contaminated chord graph to obtain a final artifact removal medical image. The following examples are intended to illustrate the details.

Refer to FIG. 2, which illustrates a schematic diagram of an environment for implementing an embodiment of the present application. The embodiment implementation environment can be implemented as a system architecture for image processing. The embodiment implementation environment may include: a terminal 100 and a server 200.

The terminal 100 may be an electronic device such as a PC (Personal Computer), a tablet Computer, a mobile phone, a medical device, and the like. A client running a target application program may be installed in the terminal 100, and the target application program may be a telemedicine application program, or may be other application programs provided with an image processing function, such as a medical computing application program, an exercise health application program, a life service application program, and the like, which is not limited in this application. The form of the target Application is not limited in the present Application, and may include, but is not limited to, an App (Application program) installed in the terminal 100, an applet, and the like, and may be a web page form. The terminal 100 may also be a medical device. Optionally, the medical device is used for acquiring medical images. Optionally, the medical device may be used to acquire medical images while enabling de-artifact processing of the ghosted medical images. In one example, a medical device implements deghost processing of a medical image with artifacts by establishing a connection with a PC. In another example, the medical device may perform de-artifact processing of the ghosted medical image by itself.

The server 200 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud computing services. The server 200 may be a background server of the target application program, and is configured to provide a background service for a client of the target application program.

The terminal 100 and the server 200 may communicate with each other through a network, such as a wired or wireless network.

In the image processing method provided by the embodiment of the application, the execution subject of each step may be a computer device, and the computer device refers to an electronic device with data calculation, processing and storage capabilities. Taking the embodiment environment shown in fig. 1 as an example, the terminal 100 may execute an image deghost method (for example, a client installed and running in the terminal 100 of a target application program executes the image deghost method), the server 200 may execute the image deghost method, or the terminal 100 and the server 200 cooperate with each other to execute the image deghost method, which is not limited in this application. For convenience of explanation, in the following method embodiments, only the implementation subjects of the steps of the image deghost method are described as computer devices.

In addition, the technical scheme of the application can be combined with the block chain technology. For example, the image deghost method disclosed in the present application, wherein some data (such as medical image, chord graph, etc.) involved can be saved on the block chain.

Optionally, the technical scheme provided by the application can be applied in a medical scene. Illustratively, the medical device is connected to a PC, which is connected to a server. Optionally, the medical device may communicate with the PC, the PC may communicate with the server through a wired connection, or may communicate through a wireless connection, which is not limited in this application. The medical equipment collects a medical image and sends the medical image to the PC, the PC uploads the medical image to the server, the server performs artifact removing processing on the medical image, then the artifact-removed medical image is sent to the PC, and a doctor can diagnose according to the artifact-removed medical image. Optionally, after receiving the medical image sent by the medical device, the PC directly performs artifact removal processing on the medical image, and then directly displays the artifact-removed medical image without uploading the image to a server.

Referring to fig. 3, a flowchart of an image deghost method according to an embodiment of the present application is shown. The method comprises the following steps (210-220):

step 210: an artifact-bearing medical image and a contaminated chord map are acquired, the artifact-bearing medical image and the contaminated chord map being a set of corresponding images for the same object.

The medical image with the artifact refers to a medical image contaminated by the artifact, and the artifact refers to a signal which does not conform to an actual structure in the medical image, and can be expressed as image deformation, overlapping, missing, blurring and the like. Optionally, the medical image comprises a CT image.

The polluted chord graph refers to the corresponding chord graph of the medical image with the artifact. A chord graph is a graph that shows the interrelationships between data. The contaminated chord maps are projection data of the medical image with the artifact corresponding to the contaminated chord maps.

It should be noted that, in the embodiments provided in the present application, the artifact-containing medical image may be acquired before the contaminated chord chart, may be acquired after the contaminated chord chart, or may be acquired simultaneously with the contaminated chord chart, which is not limited in the present application.

In one example, an artifact-bearing medical image is first acquired, and a corresponding contaminated chord map is generated from the artifact-bearing medical image. Optionally, a corresponding contaminated chordal map is obtained by forward projecting the artifact-bearing medical image. Forward projection refers to a computational approach to convert medical images into corresponding chords by radon transformation.

In another example, a contaminated chordal map is acquired and a corresponding artifact-bearing medical image is generated from the contaminated chordal map. Optionally, a corresponding medical image with artifacts is obtained by post-filtering projection of the contaminated chordal map.

In another example, an artifact-bearing medical image and its corresponding contaminated chord map are acquired simultaneously.

Optionally, the contaminated chord graph and the artifact-containing medical image may be acquired by a physical device, or may be acquired in a artifact-containing medical image library, which is not limited in this application.

In addition, the artifact-bearing medical image and the contaminated chord map are a set of corresponding images for the same object, and the artifact-bearing medical image and the contaminated chord map are a set of corresponding images for the same body part. For example, the body part may be a head, a chest, a hand, a leg, an abdomen, or the like, which is not limited in the present application.

Step 220: executing N stages of iterative processing based on the medical image with the artifact and the polluted chord graph to generate a final artifact-removed medical image corresponding to the medical image with the artifact; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1.

The final artifact-removed medical image is a medical image obtained by performing artifact-removal processing on the artifact-removed medical image. It should be noted that, in the above-mentioned N stages of iterative processing, each stage of iterative processing is used to perform a repairing process on the chord graph and perform an artifact removing process on the medical image, that is, the degree of artifact contamination of the image obtained after the processing is lighter than the degree of artifact contamination of the image before the processing, in this embodiment of the present application, the image obtained by each stage of iterative processing is referred to as a clean medical image and a clean normalized chord graph, but it should be understood that the clean medical image and the clean normalized chord graph are not images completely not contaminated by artifacts, and the final artifact removing medical image obtained after the N stages of iterative processing are also not medical images completely free of artifacts, but fewer artifacts exist in the initial medical image with artifacts.

Illustratively, as shown in FIG. 4, based on an artifact-bearing medical image X _ma And a contaminated chord graph Y, performing N-stage iterative processing to generate a medical image X with artifacts _ma Corresponding final deghosted medical image X _N 。

Optionally, the final output of the iterative processing in the N stages includes a final decontamination chord map corresponding to the artifact-carrying medical image in addition to the final deghost medical image corresponding to the artifact-carrying medical image, which is not limited in this application.

Optionally, the chord graph may be repaired first and then the medical image may be subjected to artifact removal processing in the iterative processing of each stage, or the medical image may be subjected to artifact removal processing first and then the chord graph may be repaired.

In the embodiment of the application, in the iterative processing of N stages, the repairing processing of the chord graph and the artifact removing processing of the medical image are alternately performed, the iterative processing of the N stages is sequentially performed, wherein the output of the i-1 th iterative processing is used as the input of the i-th iterative processing, and i is an integer greater than 0.

In one example, taking N ═ 3 as an example, in the first stage, the chord graph is repaired first, then the medical image is subjected to artifact removal processing, in the second stage, the chord graph is repaired first, then the medical image is subjected to artifact removal processing, in the third stage, the chord graph is repaired first, then the medical image is subjected to artifact removal processing, and after the third stage is completed, the final artifact-removed medical image is output. Optionally, after the third stage is complete, the final output further includes a final decontaminated chord chart.

In another example, taking N ═ 3 as an example, the first stage performs artifact removal processing on the medical image, then performs restoration processing on the chord graph, the second stage continues to perform artifact removal processing on the medical image, then performs restoration processing on the chord graph, then the third stage still performs artifact removal processing on the medical image, then performs restoration processing on the chord graph, and the final artifact-removed medical image is output after the third stage is completed. Optionally, after the third stage is complete, the final output further includes a final decontaminated chord chart.

Optionally, the value of N may be set according to a training result of the artifact removing model, or may be set according to experience, which is not limited in this application.

Optionally, the above-mentioned N stages of iterative processing are implemented using a deghost model. Namely, an artifact removing model is adopted, and N stages of iterative processing are executed on the basis of the artifact-carrying medical image sample and the polluted chord map sample to generate a final artifact-removing medical image corresponding to the artifact-carrying medical image sample; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1. In one example, the deghost model includes N cascaded deghost network modules for performing N stages of iterative processing; each artifact removal network module comprises: the first neural network is used for carrying out restoration processing on the chord graph, and the second neural network is used for carrying out artifact removing processing on the medical image.

In summary, according to the technical scheme provided by the embodiment of the present application, iterative processing is performed on a medical image with artifacts and a contaminated chord graph corresponding to the medical image with artifacts in N stages (N is an integer greater than 1), so as to generate a final artifact-removed medical image corresponding to the medical image with artifacts, where the iterative processing in each stage is used to perform repairing processing on the chord graph and perform artifact-removing processing on the medical image, thereby implementing mutual constraint between the chord graph and the medical image, enabling information interaction between the chord graph and the medical image to be more sufficient, improving an effect of artifact-removing processing, and being more beneficial to operation of subsequent clinical tasks.

In addition, the method constructs the artifact removing model by corresponding to the iterative algorithm, so that each network module of the artifact removing model has a specific physical meaning, the artifact removing model has strong physical interpretability, and the whole artifact removing process is transparent and visible, so that a user can more easily understand the function of each module, and the debugging and the use of the artifact removing model are facilitated.

Referring to fig. 5, a flowchart of an artifact removing method according to another embodiment of the present application is shown. The method can comprise the following steps (310-360).

Step 310: an artifact-bearing medical image and a contaminated chord map are acquired, the artifact-bearing medical image and the contaminated chord map being a set of corresponding images for the same object.

Step 320: based on the artifact-bearing medical image and the contaminated chordal map, an initialized clean medical image, an initialized clean normalized chordal map and normalization factors are generated, wherein the contour characteristics of the normalization factors are superior to those of the contaminated chordal map.

In one example, the initialized clean medical image refers to a preliminary artifact-removed medical image obtained by initializing a medical image with an artifact; the initialized clean normalized chord graph refers to a preliminarily repaired chord graph obtained by performing initialization processing on the polluted chord graph. Optionally, the initialization process includes channel concatenation and channel separation operations. Channel cascading refers to cascading images according to channel dimensions to obtain a multi-channel tensor as input. The channel separation is to split the multi-channel tensor according to the channel dimension, wherein the first channel is used as an image, and the rest channels are used as auxiliary variables.

In one example, the initialized clean medical image and the initialized clean normalized chord graph described above can be obtained by the following steps 321-323.

Step 321: and acquiring a reconstructed chord graph and a reconstructed medical image corresponding to the medical image with the artifact.

The reconstructed chord graph is an image obtained by performing primary restoration on the polluted chord graph; the reconstructed medical image is an image obtained by performing preliminary artifact removal on a medical image with an artifact.

Optionally, a linear interpolation algorithm is adopted to perform initial repair on the polluted chord graph to obtain a reconstructed chord graph. Optionally, a reconstructed medical image is derived from the reconstructed chord map. Illustratively, post-filtering the reconstructed chordal map results in a reconstructed medical image. The linear interpolation algorithm is an image artifact removing method, but the operation is simple, new artifacts are easily introduced, and meanwhile, the loss of tissue structures in medical images can be caused. In the embodiment of the application, the polluted chord graph is subjected to primary restoration processing through a linear interpolation algorithm, and then the reconstructed medical image is obtained according to the reconstructed chord graph, so that the pollution degree of the medical image with the artifact is reduced, the workload of the artifact removing model is reduced to a certain extent, and the number of artifact removing network modules required by the model is reduced.

Step 322: and executing a first near-end operation on the reconstructed chord graph to obtain an initialized clean normalized chord graph. The first near-end operation refers to an operation of performing a repairing process on the chord graph through the first neural network.

Step 323: a second near-end operation is performed on the reconstructed medical image resulting in an initialized clean medical image. The second near-end operation refers to an operation of performing artifact removal processing on the medical image through a second neural network.

In one example, the normalization factor is obtained by performing normalization processing on the contaminated chordal graph, and the profile characteristics of the contaminated chordal graph are more uniform and flat through the normalization processing.

Illustratively, the normalization factor can be obtained by the following steps 324-326. Alternatively, the artifact removal model using the normalization factors obtained by steps 324-326 as follows is called InDuDoNet (interpretable dual-domain network).

Step 324: a prior medical image is generated from the artifact-bearing medical image, the prior medical image being a medical image having common features of a clean medical image.

For the purpose of distinguishing from the following embodiments, the "prior medical image" in step 324 is referred to herein as a "first prior medical image", that is, step 324 is: a first a priori medical image is generated from the artifact-bearing medical image, the first a priori medical image being a medical image having common features of a clean medical image.

Optionally, the tape artifact medical image is processed by an a priori network to generate a first a priori medical image. Illustratively, as shown in FIG. 4, the artifact-bearing medical image X is provided by an a priori network _ma And processing to generate a first prior medical image. Optionally, the artifact-containing medical image and the reconstructed medical image are concatenated and then input to a prior network, and a first prior medical image is output through the prior network. In the embodiment of the present application, the present application is not limited to the structure of the prior network. Illustratively, the prior network is a U-shaped structure. Illustratively, as shown in fig. 7, after the artifact-carrying medical image 620 is concatenated with the reconstructed medical image 610, it is input into the prior network 630 to obtain a first prior medical image a.

Step 325: and performing fusion processing on the first prior medical image and the reconstructed medical image corresponding to the medical image with the artifact to obtain a fusion image.

The reconstructed medical image is an image obtained by performing preliminary artifact removal on a medical image with an artifact. Optionally, the fusion process is an addition process. Illustratively, the pixel values of the corresponding positions of the prior medical image and the reconstructed medical image are added to obtain a fused image. Illustratively, as shown in fig. 7, the pixel values of the corresponding positions of the first prior medical image a and the reconstructed medical image 610 are added to obtain a fused image B.

Step 326: and carrying out forward projection on the fused image to obtain a normalization factor.

Illustratively, as shown in FIG. 7, the fused image B is forward projected, resulting in a normalization factor 640.

It should be noted that the normalization factor may be obtained by performing forward projection on the fused image, may also be obtained by directly performing forward projection on the artifact-bearing medical image, and may also be obtained by performing forward projection on the prior medical image, which is not limited in this application. It should be noted that the generation of the normalization factor may be performed before the generation of the initialized clean medical image and the initialized clean normalized chord graph, or after the generation of the initialized clean medical image and the initialized clean normalized chord graph, or may be performed in synchronization with the generation of the initialized clean medical image and the initialized clean normalized chord graph, which is not limited in this application.

Illustratively, as shown in FIG. 8, the normalization factor can also be obtained by the following steps 327-328. Alternatively, the deghost model for the normalization factor obtained using steps 327-328 below is called InDoDuNet + (interpretable dual-domain network +).

Step 327: and generating a second prior medical image according to the medical image with the artifact and the reconstructed medical image corresponding to the medical image with the artifact.

Illustratively, as shown in FIG. 6, from an artifact-bearing medical image X _ma Reconstructed medical image X corresponding to medical image with artifact _LI And generating a second prior medical image.

The reconstructed medical image refers to an image obtained by performing preliminary deghost on a medical image with artifacts, and the second prior medical image refers to a medical image having common features of a clean medical image and unique features of the medical image with artifacts.

Optionally, generating an initial prior medical image from the reconstructed medical image, the initial prior medical image being a medical image having common features of a clean medical image; and performing weighted fusion on the initial prior medical image and the medical image with the artifact to obtain a second prior medical image.

Optionally, an initial prior medical image is generated from the reconstructed chord graph in combination with prior statistical properties of the clean medical image. The reconstructed medical image is combined with the prior statistical characteristic of the clean medical image, and the reconstructed medical image is processed in a manual regular mode to obtain an initial prior medical image, so that the physical interpretability of the network is stronger.

Optionally, processing the artifact-bearing medical image through a weighting network to obtain a weighting matrix; and performing point multiplication calculation on the initial prior medical image and the weight matrix to obtain a second prior medical image.

Illustratively, as shown in FIG. 9, a reconstructed medical image X is obtained according to a linear interpolation algorithm _LI And generating an initial prior medical image by combining the prior statistical characteristics of the clean medical image. And processing the medical image with the artifact Xma through a weighting network WNet to obtain a weight matrix, and performing dot product calculation on the initial prior medical image and the weight matrix to obtain a second prior medical image.

Step 328: and carrying out forward projection on the second prior medical image to obtain a normalization factor.

Compared with the acquisition method of the first prior medical image, the acquisition method of the second prior medical image adopts the lightweight weighting network to process the reconstructed medical image and the medical image with the artifact, thereby reducing the parameter quantity required by the network, improving the cross-domain generalization potential of the network, and having stronger physical interpretability of the weighting network.

Step 330: and for the ith stage in the N stages, obtaining an updated normalized chord graph according to the normalization factor, the clean normalized chord graph obtained by the i-1 st iteration, the clean medical image obtained by the i-1 st iteration and the polluted chord graph.

In one example, an updated normalized chord chart is obtained by computing the normalization factor, the clean normalized chord chart from the (i-1) th iteration, the clean medical image from the (i-1) th iteration, and the contaminated chord chart. Illustratively, the updated normalized chord graph is calculated by equation (1).

In the formula (I), the compound is shown in the specification,

means an updated normalized chord graph;

the method is a clean normalized chord graph obtained by the i-1 st iteration; x _i-1 The method refers to a clean medical image obtained by the i-1 st iteration; y is a contaminated chord graph;

refers to a normalization factor; eta ₁ Updating the step length; p refers to radon transform, i.e., forward projection operation; alpha is a summation factor used to balance the consistency of the spatial domain and radon domain data; t is _r The metal track in the chord graph has the elements of {0,1}, wherein 1 represents a metal track area; in the formula

And

all perform an operation.

Illustratively, as shown in FIG. 7, an updated normalized chord graph 663 is derived from the normalization factor 640, the clean normalized chord graph 661 from the i-1 th iteration, the clean medical image 662 from the i-1 th iteration, and the contaminated chord graph 650.

It should be noted that the updated normalized chord graph obtained by the i-1 st iteration is used for the i-th iteration, but it should be understood that the above-mentioned updated normalized chord graph has the same acquisition flow in each stage of the iteration, but the updated normalized chord graph obtained by each stage of the iteration is different because the data input by each stage of the iteration is different.

In one example, the updated normalized chord graph can be obtained by steps 331-332 as follows.

Step 331: and obtaining a first intermediate result according to the normalization factor, the clean normalization chord chart obtained by the i-1 iteration, the clean medical image obtained by the i-1 iteration and the polluted chord chart.

In one example, the normalization factor, the clean normalized chord chart from the (i-1) th iteration, the clean medical image from the (i-1) th iteration, and the contaminated chord chart are computed to obtain a first intermediate result. Illustratively, the first intermediate result is calculated by equation (2).

In the formula (I), the compound is shown in the specification,

refers to a normalization factor; eta ₁ Updating the step length; p is Radon transform, namely forward projection operation; alpha is a summation factor used to balance the consistency of the spatial domain and radon domain data; t is _r The metal track in the chord graph has the elements of {0,1}, wherein 1 represents a metal track area; in the formula

And with

All perform an operation. It should be noted that |, in the present application, represents the point product calculation, which will not be described in detail hereinafter.

It should be noted that, in the iterative processing of each stage, the acquisition flow of the first intermediate result is the same, but the first intermediate result obtained by the iterative processing of each stage is also different because the input data of the iterative processing of each stage is different.

Step 332: and obtaining an updated normalized chord graph according to the clean normalized chord graph obtained by the i-1 iteration and the first intermediate result.

In one example, the clean normalized chord graph and the first intermediate result obtained from the (i-1) th iteration are computed to obtain an updated normalized chord graph. Illustratively, the updated normalized chord graph is calculated by equation (3).

In the formula (I), the compound is shown in the specification,

means an updated normalized chord graph; eta ₁ Updating the step length;

is the partial derivative of equation (2).

Step 340: and executing a first near-end operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained by the ith iteration, wherein the first near-end operation is used for repairing the chord graph through a first neural network.

In one example, the first neural network is a first near-end network, and the updated normalized chord graph is processed by a near-end gradient technique to obtain a clean normalized chord graph obtained by the ith iteration.

Optionally, the first neural network is a residual network, formed by T _S Individual blocks of residual (Resblocks) make up (T) _S Positive integer), each consisting of a convolutional layer, a Batch Normalization layer, a ReLU activation layer, a convolutional layer, a Batch Normalization layer, and a cross-link in that order. Illustratively, the convolution kernel size of the convolution layer is 3 x 3, with a step size of 1.

Illustratively, as shown in FIG. 7, the updated normalized chord graph 663 is input into the first neural network 664, and the clean normalized chord graph 665 from the ith iteration is output.

Optionally, before executing the first near-end operation, cascading the updated normalized chord graph and the first auxiliary variable obtained by the i-1 th iteration to obtain an updated normalized chord graph after the i-th secondary iteration; then, executing a first near-end operation on the updated chord graph after the ith cascade connection to obtain a clean normalized chord graph after the ith cascade connection; and splitting channels of the clean normalized chord graph after the ith cascade connection, and selecting an image of the first channel as the clean normalized chord graph obtained by the ith iteration. In one example, as shown in fig. 10, the updated normalized chord graph and the first auxiliary variable obtained by the i-1 th iteration are cascaded, the obtained updated normalized chord graph after the i-th secondary connection is input to the first neural network, the clean normalized chord graph after the i-th secondary connection is obtained, then the channel splitting is performed on the clean normalized chord graph after the i-th cascade connection, the first channel is selected as the first channel, the image of the first channel is used as the clean normalized chord graph obtained by the i-th iteration, and the remaining channels are used as the first auxiliary variable obtained by the i-th iteration and used in the i + 1-th iteration stage. Exemplarily, if the number of channels of the clean normalized chord graph after the ith cascade is 33, the first channel is used as the first channel, an image of the first channel is used as the clean normalized chord graph obtained by the ith iteration, and images of the remaining 32 channels are used as the first auxiliary variable obtained by the ith iteration for the i +1 th iteration stage. Optionally, the first channel is any one of channels of the clean normalized chord graph after the ith cascade connection, and this is not limited in this application. For example, the first channel may be a first channel in the channels of the clean normalized chord graph after the ith cascade, and may also be a last channel in the channels of the clean normalized chord graph after the ith cascade.

Step 350: and obtaining an updated medical image according to the normalization factor, the clean medical image obtained by the i-1 iteration and the clean normalized chord chart obtained by the i iteration.

In one example, the updated medical image is obtained by computing the normalization factor, the clean medical image obtained from the i-1 th iteration, and the clean normalized chord chart obtained from the i-th iteration. Illustratively, the updated medical image is calculated by equation (4).

In the formula (I), the compound is shown in the specification,

refers to an updated medical image; x _i-1 The method refers to a clean medical image obtained by the i-1 st iteration;

the method is characterized in that a clean normalized chord graph is obtained by the ith iteration;

refers to a normalization factor; eta ₂ Updating the step length; p is Radon transform, namely forward projection operation; p ^T Is a transposition operation of radon transform; in the formula

And

an operation of |.

Illustratively, as shown in FIG. 7, an updated medical image 667 is obtained based on the normalization factor 640, the cleaned medical image 666 from the i-1 th iteration, and the cleaned normalized chord graph 665 from the i-th iteration.

It should be noted that the updated medical image obtained by the i-1 st iteration is used for the i-th iteration, but it should be understood that the above updated medical image has the same acquisition flow in each stage of the iteration, but the updated medical image obtained by each stage of the iteration is different because the data input by each stage of the iteration is different.

In one example, an updated medical image may result from steps 351-352 as follows.

Step 351: and obtaining a second intermediate result according to the normalization factor, the clean medical image obtained by the i-1 iteration and the clean normalized chord chart obtained by the i iteration.

In one example, the normalization factor, the clean medical image from the i-1 th iteration, and the clean normalized chord chart from the i-th iteration are computed to obtain a second intermediate result. Illustratively, the second intermediate result is calculated by equation (5).

In the formula, X _i-1 The method refers to a clean medical image obtained by the i-1 st iteration;

the method is characterized by comprising the following steps of (1) obtaining a clean normalized chord graph by the ith iteration;

refers to a normalization factor; eta ₂ Updating the step length; p is Radon transform, namely forward projection operation; in the formula

And

an operation of |.

It should be noted that, in the iterative processing of each stage, the acquisition flow of the second intermediate result is the same, but the second intermediate result obtained by the iterative processing of each stage is also different because the input data of the iterative processing of each stage is different.

Step 352: and obtaining an updated medical image according to the clean medical image obtained by the i-1 iteration and the second intermediate result.

In one example, the clean medical image and the second intermediate result from the i-1 st iteration are computed to obtain an updated medical image. Illustratively, the updated medical image is calculated by equation (6).

In the formula (I), the compound is shown in the specification,

refers to an updated medical image; eta ₂ Updating the step length;

is the partial derivative of equation (5).

Step 360: and executing a second near-end operation on the updated medical image to obtain a clean medical image obtained by the ith iteration, wherein the second near-end operation is used for performing artifact removing processing on the medical image through a second neural network.

In one example, the second neural network is a second near-end network, and the updated medical image is processed by a near-end gradient technique to obtain a clean medical image obtained by the ith iteration.

Optionally, the second neural network is a residual network, formed by T _S Individual blocks of residual (Resblocks) make up (T) _S Positive integer), each consisting of a convolutional layer, a Batch Normalization layer, a ReLU activation layer, a convolutional layer, a Batch Normalization layer, and a cross-link in that order. Illustratively, the convolution kernel size of the convolution layer is 3 x 3 with a step size of 1. Optionally, the first neural network and the second neural network may have the same structure or different structures, and the present application does not limit this.

Illustratively, as shown in FIG. 7, an updated medical image 667 is input to the second neural network 668, and a clean medical image 669 from the ith iteration is output.

Optionally, before performing a second near-end operation, cascading the updated medical image and a second auxiliary variable obtained by the i-1 st iteration to obtain an i-th secondary-cascaded medical image; then, executing a second near-end operation on the medical image after the ith cascade connection to obtain a clean medical image after the ith cascade connection; and splitting channels of the clean medical image after the ith cascade connection, and selecting an image of a second channel as the clean medical image obtained by the ith iteration. In an example, as shown in fig. 11, an updated medical image and a second auxiliary variable obtained by the i-1 st iteration are cascaded, then the obtained medical image after the i-th secondary cascade is input to a second neural network, a clean medical image after the i-th secondary cascade is obtained, a channel of the clean medical image after the i-th cascade is split, a first channel is used as a second channel, an image in the second channel is used as the clean medical image obtained by the i-th iteration, and the remaining channels are used as second auxiliary variables obtained by the i-th iteration and used in the i + 1-th iteration stage. Illustratively, the number of channels of the clean medical image after the ith cascade is 33, a first channel is taken as a second channel, an image of the second channel is taken as the clean medical image obtained by the ith iteration, and images of the remaining 32 channels are taken as second auxiliary variables obtained by the ith iteration and are used in the (i + 1) th iteration stage. Optionally, the first channel is any one of channels of the clean medical image after the ith concatenation, which is not limited in this application. For example, the first channel may be a first channel of the channels of the clean medical image after the ith concatenation, and may also be a last channel of the channels of the clean medical image after the ith concatenation.

It should be noted that i is a positive integer less than or equal to N, when i is 1, the clean normalized chord graph obtained in the i-1 th iteration is an initialized clean normalized chord graph, and the clean medical image obtained in the i-1 st iteration is an initialized clean medical image; and when i is equal to N, the clean medical image obtained by the ith iteration is used as a final artifact-removed medical image corresponding to the artifact-contained medical image.

To sum up, in the technical solution provided by the embodiment of the present application, a method of cascade input of a reconstructed medical image and a medical image with an artifact is adopted, compared with the case of only inputting a reconstructed medical image, data of an organization structure in a complete medical image is supplemented, and a lack of the organization structure in the medical image is avoided to a certain extent; by adopting the method of combining the weighting network with the prior statistical characteristics of the clean medical image, the quantity of parameters required by the network is less, the cross-domain generalization capability of the network is improved, and the physical interpretability of the weighting network is stronger.

In the following, a description is given of a training procedure of the artifact removal model by an embodiment, and the content involved in the artifact removal model and the content involved in the training procedure are corresponding to each other and intercommunicated with each other, for example, where a detailed description is not given on one side, reference may be made to a description on the other side.

Please refer to fig. 12, which shows a flowchart of a training method of a deghost model according to an embodiment of the present application. The method can include the following steps (710-750).

Step 710: sample data is obtained, and the sample data comprises a clean medical image sample and a clean chord chart sample.

Optionally, the clean medical image sample and the clean chord map sample are a set of corresponding images for the same object.

It should be noted that the clean medical image sample may be acquired before the clean chord chart sample, may be acquired after the clean chord chart sample, or may be acquired simultaneously with the clean chord chart sample, which is not limited in the present application.

In one example, if a clean medical image sample is obtained, a corresponding clean chordal map sample is generated from the clean medical image sample. Optionally, a clean medical image sample is forward projected to obtain a corresponding clean chord map sample.

In another example, a clean chordal map sample is obtained, and a corresponding clean medical image sample is generated from the clean chordal map sample. Optionally, the clean medical image sample is obtained by performing filtered projection on the clean chordal map sample.

In another example, a clean medical image sample and its corresponding clean chord map sample are acquired simultaneously.

Step 720: based on the clean medical image samples, contaminated chordal map samples and artifact-bearing medical image samples are generated.

In one example, artifact simulation data is first obtained, a contaminated projection image is generated based on the artifact simulation data and a clean medical image sample, noise is added to the contaminated projection image to generate a contaminated chordal map sample, and then the medical image sample with the artifact is generated based on the contaminated chordal map sample. Illustratively, different types of metal masks are obtained, metal artifacts are synthesized according to a data simulation process to obtain a polluted chord graph sample, and then the polluted chord graph sample is subjected to post-filtering projection to obtain a medical image sample with the artifacts. Optionally, the contaminated projection image is obtained by performing a multi-level medical image projection after performing tissue segmentation on a clean medical image sample.

Optionally, the numerical range of the contaminated chord graph sample is cut to obtain the cut contaminated chord graph sample, and the data value of the contaminated chord graph sample is limited within a certain range. Illustratively, the contaminated chordal map samples are clipped to a threshold [0,4] range, then normalized to a [0,1] range by dividing by 4, and finally converted to a [0,255] range by multiplying by 255.

Optionally, the clipped contaminated chordal map sample is subjected to post-filter projection to obtain a corresponding medical image sample with the artifact.

Optionally, the numerical range of the artifact-carrying medical image sample is cut to obtain the cut artifact-carrying medical image sample, and the pixel value of the medical image sample is limited within a certain range. In one example, the artifact-bearing medical image sample is clipped to a threshold [0,1] range and then multiplied by 255 to translate to a [0,255] range. In another example, the artifact-bearing medical image sample is normalized to the [0,1] range and then multiplied by 255 to translate to the [0,255] range.

Step 730: adopting a deghost model, executing N stages of iterative processing based on the medical image sample with the artifact and the polluted chord map sample, and generating a final deghost medical image corresponding to the medical image sample with the artifact; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1.

In one example, the deghost model includes N cascaded deghost modules, each deghost module including: the first neural network is used for carrying out restoration processing on the chord graph, and the second neural network is used for carrying out artifact removing processing on the medical image.

In one example, the final de-artifact medical image may result from steps 731-735 as follows.

Step 731: based on the artifact-bearing medical image sample and the contaminated chordal map sample, an initialized clean medical image, an initialized clean normalized chordal map and a normalization factor are generated, the profile characteristics of the normalization factor being superior to the profile characteristics of the contaminated chordal map.

Step 732: and for the ith stage in the N stages, obtaining an updated normalized chord chart according to the normalization factor, the clean normalized chord chart obtained by the i-1 iteration, the clean medical image obtained by the i-1 iteration and the polluted chord chart sample.

Step 733: and executing a first near-end operation on the updated normalized chord chart by adopting a first neural network contained in the ith artifact removing network module to obtain a clean normalized chord chart obtained by the ith iteration.

Step 734: and obtaining an updated medical image according to the normalization factor, the clean medical image obtained by the i-1 iteration and the clean normalized chord graph obtained by the i-th iteration.

Step 735: and executing a second near-end operation on the updated medical image by adopting a second network contained in the ith artifact removing network module to obtain a clean medical image obtained by the ith iteration.

It should be noted that i is a positive integer less than or equal to N, when i is 1, the clean normalized chord graph obtained in the i-1 th iteration is an initialized clean normalized chord graph, and the clean medical image obtained in the i-1 st iteration is an initialized clean medical image; and when i is equal to N, the clean medical image obtained by the ith iteration is used as a final de-artifact medical image corresponding to the artifact medical image sample.

It should be noted that, in the embodiment of the present application, the clean medical image, the clean normalized chord graph, the updated medical image, and the updated normalized chord graph obtained in the training method for removing an artifact model are not worded to be distinguished from the clean medical image, the clean normalized chord graph, the updated medical image, and the updated normalized chord graph obtained in the image artifact removing method, but those skilled in the art should know that the two are not the same.

Step 740: the training loss of the artifact-removed model is calculated from the clean medical image sample and the final artifact-removed medical image.

In one example, a training loss of the deghost model is calculated using a training objective function. Optionally, the training objective function is equation (7).

In the formula, beta _n To compromise the parameters; γ is the weight used to balance the losses; x _gt Is a clean medical image sample; y is _gt Is a clean normalized chordal map sample; m is a metal position in the medical image;

is a normalization factor;

a clean normalized chord graph is obtained for the nth iteration; x _n To a final de-artifact medical image.

In an exemplary embodiment, β _n 0.1(N is a natural number less than N), beta _N ＝1，γ＝0.1，N＝7，M、X _gt And Y _gt Are known from training simulation experiments.

Step 750: the deghost model is trained based on the training loss.

In one example, Adam (Adaptive motion estimation) based algorithms are employed to update solution optimization parameters, including,

step length eta ₁ And η ₂ . In each iteration process, a prediction result error is calculated and propagated back to the artifact removal model, and a gradient is calculated and the parameters of the artifact removal model are updated.

In one example, the initial learning rate is set to 2 × 10 ^-4 Every 40 iterations, the learning rate decays by 0.5, the total training iteration number is 100, and in the process of each iteration, the iteration network module comprises two neural networks, the size of the CT image is 416 x 416, and the size of the chord graph is 641 x 640.

It should be noted that, in the above description, the use process and the training process of the deghost model are respectively described in different embodiments, and the content involved in the use process and the content involved in the training process of the model are corresponding to each other and intercommunicate with each other, for example, where a detailed description is not given on one side, reference may be made to the description on the other side.

Based on the artifact removing model, a relevant experiment is designed to verify the artifact removing effect. In an experimental environment, the model is mounted on a cloud system, and please refer to fig. 13 for a specific flow of the experiment.

Step 810: and judging whether the training stage or the testing stage is currently performed.

Step 820: and if the current training stage is in, acquiring sample data of the artifact removing model.

Step 830: and performing iterative training on the artifact removing model.

Step 840: the deghost model is trained based on the training loss.

Step 850: and judging whether the set training times are reached.

Step 860: if the set training times are reached, storing the artifact removing model after training; and if the set training times are not reached, continuing to train the artifact removing model.

Step 870: and if the current test stage is in, acquiring the medical image with the artifact.

Step 880: and processing the artifact-removed medical image by adopting an artifact-removed model.

Step 890: and outputting the final artifact-removed medical image.

For details of the training process, please refer to the embodiment of the training method for removing the artifact model of the present application, which is not described herein again.

The idea of the above image deghost method is derived from the problem derivation process as follows.

First, an observed contaminated chord chart is given

Wherein N is _b And N _p The number of detectors arranged in the medical image acquisition device and the number of projection angles are respectively. Based on this, the conventional iterative artifact removal method can be expressed by equation (8).

Wherein X ∈ R ^H×W Representing a clean medical image (spatial domain), H and W being the height and width, respectively, of the medical image; p is Radon transform, i.e. forward projection operation, T _r Is a metal trace in the graph, whose elements are {0,1}, wherein 1 indicates a metal trace region, which is the product of the corresponding elements; g (-) is a regularization term representing the prior structure of a clean medical image; λ is a compromise parameter.

And further performing joint regularization on the spatial domain and the Ladong domain for mutual learning of the Ladong domain and the spatial domain.

At this time, equation (8) can be converted into equation (9).

In the formula, S means the dry normalized chord graph (radon domain); α is a summation factor used to balance data consistency for the spatial domain and the radon domain; g ₁ (. and g) ₂ The (·) is a regular term in all,the prior structures of the clean normalized chord graph S and the clean medical image X are embedded, respectively.

Because the contaminated chordal map is corrected after being normalized, it is easier than correcting the contaminated chordal map directly, because the normalized profile is more uniform and flat. Therefore, S can be redefined as formula (10).

In the formula (I), the compound is shown in the specification,

for normalization factors, typically for a priori medical images

Obtained by performing a forward projection, i.e.

Is obtained by a priori network according to the medical image with the artifact;

the normalized clean normalized chord graph is obtained.

By substituting the formula (10) into the formula (9), the formula (11) can be derived.

To obtain the deghost medical image and the clean chord map, equation (11) should be solved.

Illustratively, the near-end gradient technique is used for alternate updating

And X. In particular, in the nth iteration,

can be solved by solving the problem of equation (11) with respect to

The quadratic approximation problem of (a) is implemented, which can be written as equation (12).

In the formula (I), the compound is shown in the specification,

for the updated result, η, obtained in the (n-1) th iteration ₁ In order to update the step size,

for general regularization terms

Equation (12) can be written as equation (13).

Will be provided with

Substituting into the formula (13), solving to obtain

The update rule of (2) is formula (14).

In the formula (I), the compound is shown in the specification,

is composed of a regularization term g ₁ The determined near-end operator can be designed as a network module.

Likewise, the quadratic approximation problem for X can be written as equation (15).

Wherein the content of the first and second substances,

therefore, the update rule of X can be written as formula (16).

In the formula (I), the compound is shown in the specification,

is composed of a regularization term g ₂ The determined near-end operator can be designed as a network module.

Based on the formula (14) and the formula (16), each iteration sub-step is correspondingly expanded into a network module, so that the whole artifact removing model is constructed, the artifact removing method has good physical interpretability, and meanwhile, the radon domain and the spatial domain learn each other, so that the information interaction of the medical image and the chord graph is more sufficient.

In addition, in the embodiment of the present application, the initialization processing is performed by using the formula (17) and the formula (18).

In the formula, | is channel separation operation; concat (·)Performing cascade operation on the channels; s _LI The method comprises the steps of (1) adopting a linear interpolation algorithm to repair a polluted chord graph to obtain a reconstructed chord graph; x _LI Is according to S _LI The resulting reconstructed medical image; q ₀ ^s To initialize a first auxiliary variable; q ₀ ^X To initialize a second auxiliary variable;

is composed of a regularization term g ₁ () the determined near-end operator;

is composed of a regularization term g ₂ () the determined near-end operator; k _S And K _X Is a convolution kernel with a size of f _s ×f _s ×N _s X 1 and f _x ×f _x ×N _x X 1. During the training process, the convolution kernel size is f _s ×f _s ×N _s ×1＝f _x ×f _x ×N _x ×1＝3*3*32*1。

Illustratively, the present application performs experimental comparison on six artifact removing methods provided by the related art, i.e., LI, NMAR, CNNMAR, DuDoNet (dual domain network), DSCMAR, DuDoNet + + (dual domain network + +), and two artifact removing methods proposed by the technical solution of the present application, i.e., inddon (interpretable dual domain network) and inddon + (interpretable dual domain network +).

Please refer to fig. 14, which shows a comparison of the generalization effect of the above five artifact removing methods. As can be seen in fig. 14, the generalization effect of indadonet + is better than that of several other deghost methods.

As shown in table 1, the number of network parameters required by the lndudonet + is minimal, which is more beneficial to actual deployment of storage.

Table 1: network parameter comparison table for artifact removing method

The following are embodiments of the apparatus of the present application that may be used to perform embodiments of the method of the present application. For details which are not disclosed in the embodiments of the apparatus of the present application, reference is made to the embodiments of the method of the present application.

Referring to fig. 15, a block diagram of an image deghost apparatus according to an embodiment of the present application is shown. The device has the function of realizing the image artifact removing method, and the function can be realized by hardware or by hardware executing corresponding software. The apparatus 900 may include: an image acquisition module 910, a deghost module 920.

An image acquisition module 910 for acquiring an artifact-bearing medical image and a contaminated chord map, the artifact-bearing medical image and the contaminated chord map being a set of corresponding images for a same object.

A deghost module 920, configured to perform N stages of iterative processing based on the artifact-bearing medical image and the contaminated chord map, and generate a final deghost medical image corresponding to the artifact-bearing medical image; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1.

In some embodiments, as shown in fig. 16, the deghost module 920 includes a preprocessing unit 921, a chord map updating unit 922, a first proximal unit 923, a medical image updating unit 924, and a second proximal unit 925.

A preprocessing unit 921 for generating an initialized clean medical image, an initialized clean normalized chord graph and a normalization factor based on the artifact-bearing medical image and the contaminated chord graph, the normalization factor having a profile characteristic superior to a profile characteristic of the contaminated chord graph.

In some embodiments, the preprocessing unit 921 is configured to obtain a reconstructed chord chart and a reconstructed medical image corresponding to the medical image with the artifact, where the reconstructed chord chart is an image obtained by performing preliminary repair on the contaminated chord chart, and the reconstructed medical image is an image obtained by performing preliminary artifact removal on the medical image with the artifact; executing a first near-end operation on the reconstructed chord graph to obtain the initialized clean normalized chord graph; performing a second near-end operation on the reconstructed medical image resulting in the initialized clean medical image.

In some embodiments, the preprocessing unit 921 is configured to generate a first a priori medical image from the artifact-bearing medical image, the first a priori medical image referring to a medical image having common features of a clean medical image; performing fusion processing on the first prior medical image and a reconstructed medical image corresponding to the medical image with the artifact to obtain a fused image, wherein the reconstructed medical image is an image obtained by performing preliminary artifact removal on the medical image with the artifact; and carrying out forward projection on the fusion image to obtain the normalization factor. In some embodiments, the preprocessing unit 921 is configured to cascade the artifact-bearing medical image and the reconstructed medical image, input the cascade image to an a priori network, and output the first a priori medical image through the a priori network.

In some embodiments, the preprocessing unit 921 is configured to generate a second a priori medical image from the artifact-bearing medical image and a reconstructed medical image corresponding to the artifact-bearing medical image; wherein the reconstructed medical image is an image obtained by performing preliminary artifact removal on the medical image with the artifact, and the second prior medical image is a medical image having common features of a clean medical image and unique features of the medical image with the artifact; and carrying out forward projection on the second prior medical image to obtain the normalization factor.

In some embodiments, the preprocessing unit 921 is configured to generate an initial prior medical image from the reconstructed medical image, the initial prior medical image being a medical image having common features of a clean medical image; and performing weighted fusion on the initial prior medical image and the medical image with the artifact to obtain the second prior medical image.

In some embodiments, the preprocessing unit 921 is configured to process the medical image with artifacts through a weighting network to obtain a weight matrix; and performing dot product calculation on the initial prior medical image and the weight matrix to obtain the second prior medical image.

And the chord map updating unit 922 is configured to, for the ith stage of the N stages, obtain an updated normalized chord map according to the normalization factor, the clean normalized chord map obtained by the i-1 st iteration, the clean medical image obtained by the i-1 st iteration, and the contaminated chord map.

In some embodiments, the chord map updating unit 922 is configured to obtain a first intermediate result according to the normalization factor, the clean normalized chord map obtained in the i-1 th iteration, the clean medical image obtained in the i-1 th iteration, and the contaminated chord map; and obtaining the updated normalized chord graph according to the clean normalized chord graph obtained by the i-1 iteration and the first intermediate result.

A first near-end unit 923, configured to perform a first near-end operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained by the ith iteration, where the first near-end operation is used to perform a repair process on the chord graph through a first neural network.

In some embodiments, the first near-end unit 923 is configured to cascade the updated normalized chord graph and the first auxiliary variable obtained in the i-1 st iteration to obtain an i-th secondary-cascaded updated normalized chord graph; performing a first near-end operation on the updated normalized chord graph after the ith cascade connection to obtain a clean normalized chord graph after the ith secondary connection; and splitting channels of the clean normalized chord graph after the ith cascade connection, and selecting an image of a first channel as the clean normalized chord graph obtained by the ith iteration.

A medical image updating unit 924, configured to obtain an updated medical image according to the normalization factor, the clean medical image obtained in the i-1 st iteration, and the clean normalized chord chart obtained in the i-th iteration.

In some embodiments, the medical image updating unit 924 is configured to obtain a second intermediate result according to the normalization factor, the clean medical image obtained in the i-1 st iteration, and the clean normalized chord graph obtained in the i-th iteration; and obtaining the updated medical image according to the clean medical image obtained by the i-1 iteration and the second intermediate result.

A second near-end unit 925, configured to perform a second near-end operation on the updated medical image to obtain a clean medical image obtained by the ith iteration, where the second near-end operation is used to perform artifact removal processing on the medical image through a second neural network.

In some embodiments, the second near-end unit 925 is configured to concatenate the updated medical image with the second auxiliary variable obtained in the i-1 st iteration to obtain an i-th sub-concatenated medical image; executing a second near-end operation on the medical image after the ith cascade connection to obtain a clean medical image after the ith cascade connection; and splitting channels of the clean medical image after the ith cascade connection, and selecting an image of a second channel as the clean medical image obtained by the ith iteration.

In some embodiments, i is a positive integer less than or equal to N, when i is 1, the i-1 st iteration resulting clean normalized chord graph is the initialized clean normalized chord graph, and the i-1 st iteration resulting clean medical image is the initialized clean medical image; and when i is equal to N, the clean medical image obtained by the ith iteration is used as a final artifact-removed medical image corresponding to the artifact-contained medical image.

Referring to fig. 17, a block diagram of a deghost model training apparatus according to an embodiment of the present application is shown. The device has the function of implementing the training method for removing the artifact model, and the function can be implemented by hardware or by hardware executing corresponding software. The apparatus 1100 may include: a sample acquisition module 1110, a sample processing module 1120, a deghost module 1130, a loss calculation module 1110, and a model training module 1150.

A sample obtaining module 1110, configured to obtain sample data, where the sample data includes a clean medical image sample and a clean chord chart sample, and the clean medical image sample and the clean chord chart sample are a set of corresponding images for a same object.

A sample processing module 1120 for generating contaminated chordal map samples and artifact-bearing medical image samples based on the clean medical image samples.

In some embodiments, the sample processing module 1120 is configured to obtain artifact simulation data; generating a contaminated projection image from the artifact simulation data and the clean medical image sample; adding noise to the contaminated projection image, generating the contaminated chordal map sample; and generating the medical image sample with the artifact according to the polluted chord map sample.

A de-artifact module 1130, configured to perform N stages of iterative processing based on the artifact-bearing medical image sample and the contaminated chordal map sample by using the de-artifact model, and generate a final de-artifact medical image corresponding to the artifact-bearing medical image sample; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1.

In some embodiments, the deghost model comprises N cascaded deghost network modules, each comprising: the first neural network is used for carrying out restoration processing on the chord graph, and the second neural network is used for carrying out artifact removing processing on the medical image.

In some embodiments, as shown in fig. 18, the deghost module 1130 includes a preprocessing unit 1131, a chord map updating unit 1132, a first proximal unit 1133, a medical image updating unit 1134, and a second proximal unit 1135.

A preprocessing unit 1131, configured to generate an initialized clean medical image, an initialized clean normalized chord graph and a normalization factor based on the artifact-bearing medical image and the contaminated chord graph, where a contour feature of the normalization factor is better than a contour feature of the contaminated chord graph.

In some embodiments, the preprocessing unit 1131 is configured to obtain a reconstructed chord map and a reconstructed medical image corresponding to the artifact-containing medical image, where the reconstructed chord map is an image obtained by performing a preliminary repair on the contaminated chord map, and the reconstructed medical image is an image obtained by performing a preliminary artifact removal on the artifact-containing medical image; executing a first near-end operation on the reconstructed chord graph to obtain the initialized clean normalized chord graph; performing a second near-end operation on the reconstructed medical image resulting in the initialized clean medical image.

In some embodiments, the preprocessing unit 1131 is configured to generate an a priori medical image from the artifact-bearing medical image, the a priori medical image being a medical image having common features of a clean medical image; performing fusion processing on the prior medical image and a reconstructed medical image corresponding to the medical image with the artifact to obtain a fused image, wherein the reconstructed medical image is an image obtained by performing preliminary artifact removal on the medical image with the artifact; and carrying out forward projection on the fusion image to obtain the normalization factor.

In some embodiments, the preprocessing unit 1131 is configured to input the cascaded artifact-containing medical image and the reconstructed medical image into an a priori network, and output the a priori medical image through the a priori network.

And a chord map updating unit 1132, configured to, for an ith stage of the N stages, obtain an updated normalized chord map according to the normalization factor, the clean normalized chord map obtained by the i-1 st iteration, the clean medical image obtained by the i-1 st iteration, and the contaminated chord map.

In some embodiments, the chord map updating unit 1132 is configured to obtain a first intermediate result according to the normalization factor, the clean normalized chord map obtained in the i-1 st iteration, the clean medical image obtained in the i-1 st iteration, and the contaminated chord map; and obtaining the updated normalized chord graph according to the clean normalized chord graph obtained by the i-1 iteration and the first intermediate result.

A first near-end unit 1133, configured to perform a first near-end operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained by the ith iteration, where the first near-end operation is used to perform a repairing process on the chord graph through a first neural network.

In some embodiments, the first near-end unit 1133 is configured to cascade the updated normalized chord graph and the first auxiliary variable obtained by the i-1 th iteration to obtain an i-th secondary-concatenated updated normalized chord graph; executing a first near-end operation on the updated normalized chord graph after the ith cascade connection to obtain a clean normalized chord graph after the ith cascade connection; and splitting channels of the clean normalized chord graph after the ith cascade connection, and selecting an image of a first channel as the clean normalized chord graph obtained by the ith iteration.

A medical image updating unit 1134, configured to obtain an updated medical image according to the normalization factor, the clean medical image obtained in the i-1 st iteration, and the clean normalized chord chart obtained in the i-th iteration.

In some embodiments, the medical image updating unit 1134 is configured to obtain a second intermediate result according to the normalization factor, the clean medical image obtained in the i-1 st iteration, and the clean normalized chord graph obtained in the i-th iteration; and obtaining the updated medical image according to the clean medical image obtained by the i-1 iteration and the second intermediate result.

A second near-end unit 1135, configured to perform a second near-end operation on the updated medical image to obtain a clean medical image obtained by the ith iteration, where the second near-end operation is used to perform artifact removal processing on the medical image through a second neural network.

In some embodiments, the second proximal unit 1135 is configured to concatenate the updated medical image and the second auxiliary variable obtained in the i-1 st iteration to obtain an i-th sub-concatenated medical image; executing a second near-end operation on the medical image after the ith cascade connection to obtain a clean medical image after the ith cascade connection; and splitting channels of the clean medical image after the ith cascade connection, and selecting an image of a second channel as the clean medical image obtained by the ith iteration.

A loss calculation module 1140 for calculating a training loss of the deghost model from the clean medical image samples and the final deghost medical image.

A model training module 1150 for training the deghost model based on the training loss.

It should be noted that, when the apparatus provided in the foregoing embodiment implements the functions thereof, only the division of the functional modules is illustrated, and in practical applications, the functions may be distributed by different functional modules according to needs, that is, the internal structure of the apparatus may be divided into different functional modules to implement all or part of the functions described above. In addition, the apparatus and method embodiments provided by the above embodiments belong to the same concept, and specific implementation processes thereof are described in the method embodiments for details, which are not described herein again.

Referring to fig. 18, a schematic structural diagram of a computer device according to an embodiment of the present application is shown. The computer device may be any electronic device having data computing, processing, and storage functions. The computer device can be used for implementing the image artifact removing method provided in the above embodiment, and can also be used for implementing the training method of the artifact removing model provided in the above embodiment. Specifically, the method comprises the following steps:

the computer device 1300 includes a Central Processing Unit (e.g., a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), etc.) 1301, a system Memory 1304 including a RAM (Random-Access Memory) 1302 and a ROM (Read-Only Memory) 1303, and a system bus 1305 connecting the system Memory 1304 and the Central Processing Unit 1301. The computer device 1300 also includes a basic Input/Output System (I/O System) 1306 for facilitating information transfer between various devices within the server, and a mass storage device 1307 for storing an operating System 1313, application programs 1314 and other program modules 1315.

In some embodiments, the basic input/output system 1306 includes a display 1308 for displaying information and an input device 1309, such as a mouse, keyboard, or the like, for a user to input information. Wherein the display 1308 and input device 1309 are connected to the central processing unit 1301 through an input output controller 1310 connected to the system bus 1305. The basic input/output system 1306 may also include an input/output controller 1310 for receiving and processing input from a number of other devices, such as a keyboard, mouse, or electronic stylus. Similarly, input-output controller 1310 also provides output to a display screen, a printer, or other type of output device.

The mass storage device 1307 is connected to the central processing unit 1301 through a mass storage controller (not shown) connected to the system bus 1305. The mass storage device 1307 and its associated computer-readable media provide non-volatile storage for the computer device 1300. That is, the mass storage device 1307 may include a computer-readable medium (not shown) such as a hard disk or CD-ROM (Compact disk Read-Only Memory) drive.

Without loss of generality, the computer readable media may comprise computer storage media and communication media. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes RAM, ROM, EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash Memory or other solid state Memory technology, CD-ROM, DVD (Digital Video Disc) or other optical, magnetic, or other magnetic storage devices. Of course, those skilled in the art will appreciate that the computer storage media is not limited to the foregoing. The system memory 1304 and mass storage device 1307 described above may be collectively referred to as memory.

The computer device 1300 may also operate as a remote computer connected to a network via a network, such as the internet, according to embodiments of the present application. That is, the computer device 1300 may be connected to the network 1312 through the network interface unit 1311, which is connected to the system bus 1305, or may be connected to other types of networks or remote computer systems (not shown) using the network interface unit 1311.

The memory also includes at least one instruction, at least one program, set of codes, or set of instructions stored in the memory and configured to be executed by the one or more processors to implement the image deghosting method, or the training method of the deghosting model, described above.

In an exemplary embodiment, a computer-readable storage medium is further provided, in which at least one instruction, at least one program, a set of codes, or a set of instructions is stored, which when executed by a processor of a computer device, implements the above-mentioned image deghosting method, or training method of deghosting models.

Optionally, the computer-readable storage medium may include: ROM (Read-Only Memory), RAM (Random-Access Memory), SSD (Solid State drive), or optical disk. The Random Access Memory may include a ReRAM (resistive Random Access Memory) and a DRAM (Dynamic Random Access Memory).

In an exemplary embodiment, a computer program product or computer program is also provided, the computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, so that the computer device executes the image deghost method or the training method of the deghost model.

It should be understood that reference herein to "a plurality" means two or more. "and/or" describes the association relationship of the associated objects, meaning that there may be three relationships, e.g., a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship.

The above description is only exemplary of the present application and should not be taken as limiting the present application, and any modifications, equivalents, improvements and the like that are made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. An image deghost method, comprising:

based on the medical image with the artifact and the polluted chord map, executing N stages of iterative processing to generate a final de-artifact medical image corresponding to the medical image with the artifact; the iterative processing of each stage is used for repairing the chord graph and removing artifacts from the medical image, and N is an integer greater than 1.

2. The method of claim 1, wherein the performing N stages of iterative processing based on the artifact-bearing medical image and the contaminated chordal map to generate a final de-artifact medical image corresponding to the artifact-bearing medical image comprises:

generating an initialized clean medical image, an initialized clean normalized chord map and a normalization factor based on the artifact-bearing medical image and the contaminated chord map, the normalization factor having profile features superior to those of the contaminated chord map;

for the ith stage in the N stages, obtaining an updated normalization chord map according to the normalization factor, the clean normalization chord map obtained by the (i-1) th iteration, the clean medical image obtained by the (i-1) th iteration and the polluted chord map;

executing a first near-end operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained by the ith iteration, wherein the first near-end operation is used for repairing the chord graph through a first neural network;

obtaining an updated medical image according to the normalization factor, the clean medical image obtained by the i-1 iteration and the clean normalization chord chart obtained by the i iteration;

executing a second near-end operation on the updated medical image to obtain a clean medical image obtained by the ith iteration, wherein the second near-end operation is used for performing artifact removing processing on the medical image through a second neural network;

when i is 1, the clean normalized chord graph obtained by the i-1 th iteration is the initialized clean normalized chord graph, and the clean medical image obtained by the i-1 st iteration is the initialized clean medical image; and when i is equal to N, the clean medical image obtained by the ith iteration is used as a final artifact-removed medical image corresponding to the artifact-contained medical image.

3. The method of claim 2, wherein obtaining an updated normalized chord graph from the normalization factor, the clean normalized chord graph from the i-1 th iteration, the clean medical image from the i-1 th iteration, and the contaminated chord graph comprises:

obtaining a first intermediate result according to the normalization factor, a clean normalization chord chart obtained by the i-1 iteration, a clean medical image obtained by the i-1 iteration and the polluted chord chart;

and obtaining the updated normalized chord graph according to the clean normalized chord graph obtained by the i-1 iteration and the first intermediate result.

4. The method according to claim 2, wherein obtaining an updated medical image according to the normalization factor, the clean medical image obtained from the i-1 st iteration, and the clean normalized chord graph obtained from the i-th iteration comprises:

obtaining a second intermediate result according to the normalization factor, the clean medical image obtained by the i-1 iteration and the clean normalization chord map obtained by the i-1 iteration;

and obtaining the updated medical image according to the clean medical image obtained by the i-1 iteration and the second intermediate result.

5. The method of claim 2, wherein performing a first near-end operation on the updated normalized chord graph to obtain a clean normalized chord graph from an i-th iteration comprises:

cascading the updated normalized chord graph and the first auxiliary variable obtained by the i-1 th iteration to obtain an updated normalized chord graph after the i-th secondary iteration;

executing a first near-end operation on the updated normalized chord graph after the ith cascade connection to obtain a clean normalized chord graph after the ith cascade connection;

and splitting channels of the clean normalized chord graph after the ith cascade connection, and selecting an image of a first channel as the clean normalized chord graph obtained by the ith iteration.

6. The method of claim 2, wherein performing a second near-end operation on the updated medical image to obtain an i-th iteration of obtaining a clean medical image comprises:

cascading the updated medical image with a second auxiliary variable obtained by the i-1 st iteration to obtain an i secondary concatenated medical image;

executing a second near-end operation on the medical image after the ith cascade connection to obtain a clean medical image after the ith cascade connection;

and splitting channels of the clean medical image after the ith cascade connection, and selecting an image of a second channel as the clean medical image obtained by the ith iteration.

7. The method of claim 2, wherein generating an initialized clean medical image, an initialized clean normalized chord map based on the artifact-bearing medical image and the contaminated chord map comprises:

acquiring a reconstructed chord chart and a reconstructed medical image corresponding to the medical image with the artifact, wherein the reconstructed chord chart is an image obtained by primarily repairing the polluted chord chart, and the reconstructed medical image is an image obtained by primarily removing the artifact of the medical image with the artifact;

executing a first near-end operation on the reconstructed chord graph to obtain the initialized clean normalized chord graph;

performing a second near-end operation on the reconstructed medical image resulting in the initialized clean medical image.

8. The method of claim 2, wherein the generating the normalization factor comprises:

generating a first prior medical image from the artifact-bearing medical image, the first prior medical image being a medical image having common features of a clean medical image;

performing fusion processing on the first prior medical image and a reconstructed medical image corresponding to the medical image with the artifact to obtain a fused image, wherein the reconstructed medical image is an image obtained by performing preliminary artifact removal on the medical image with the artifact;

and carrying out forward projection on the fusion image to obtain the normalization factor.

9. The method of claim 8, wherein generating a first a priori medical image from the artifact-bearing medical image comprises:

after the medical image with the artifact and the reconstructed medical image are cascaded, the medical image with the artifact and the reconstructed medical image are input into a prior network, and the first prior medical image is output through the prior network.

10. The method of claim 2, wherein generating a normalization factor based on the artifact-bearing medical image comprises:

generating a second prior medical image according to the medical image with the artifact and the reconstructed medical image corresponding to the medical image with the artifact; wherein the reconstructed medical image is an image obtained by performing preliminary artifact removal on the medical image with the artifact, and the second prior medical image is a medical image having common features of a clean medical image and unique features of the medical image with the artifact;

and carrying out forward projection on the second prior medical image to obtain the normalization factor.

11. The method of claim 10, wherein generating the second a priori medical image from the artifact-bearing medical image and a corresponding reconstructed medical image of the artifact-bearing medical image comprises:

generating an initial prior medical image according to the reconstructed medical image, wherein the initial prior medical image is a medical image with common characteristics of a clean medical image;

and performing weighted fusion on the initial prior medical image and the medical image with the artifact to obtain a second prior medical image.

12. The method of claim 11, wherein the weighted fusing of the initial a priori medical image and the artifact-bearing medical image to obtain the second a priori medical image comprises:

processing the medical image with the artifact through a weighting network to obtain a weight matrix;

and performing dot product calculation on the initial prior medical image and the weight matrix to obtain the second prior medical image.

13. A method for training a deghost model, the method comprising:

obtaining sample data, wherein the sample data comprises a clean medical image sample and a clean chord graph sample, and the clean medical image sample and the clean chord graph sample are a group of corresponding images aiming at the same object;

generating a contaminated chordal map sample and an artifact-bearing medical image sample based on the medical image sample;

executing N stages of iterative processing based on the artifact-carrying medical image sample and the polluted chord map sample by adopting the artifact removing model to generate a final artifact-removing medical image corresponding to the artifact-carrying medical image sample; the iterative processing of each stage is used for repairing the chord graph and removing artifacts of the medical image, and N is an integer greater than 1;

training the deghost model based on the training loss.

14. The method of claim 13, wherein the deghost model comprises N cascaded deghost network modules, each comprising: the first neural network is used for carrying out restoration processing on the chord graph, and the second neural network is used for carrying out artifact removing processing on the medical image;

the performing, by using the artifact removing model, N stages of iterative processing based on the artifact-containing medical image sample and the contaminated chordal map sample to generate a final artifact-removed medical image corresponding to the artifact-containing medical image sample, includes:

generating an initialized clean medical image, an initialized clean normalized chord map and a normalization factor based on the artifact-bearing medical image sample and the contaminated chord map sample, the normalization factor having profile characteristics superior to those of the contaminated chord map;

for the ith stage in the N stages, obtaining an updated normalized chord chart according to the normalization factor, the clean normalized chord chart obtained by the i-1 iteration, the clean medical image obtained by the i-1 iteration and the polluted chord chart sample;

executing a first near-end operation on the updated normalized chord chart by adopting a first neural network contained in an ith artifact removing network module to obtain a clean normalized chord chart obtained by the ith iteration;

executing a second near-end operation on the updated medical image by adopting a second neural network contained in the ith artifact removing network module to obtain a clean medical image obtained by the ith iteration;

when i is 1, the clean normalized chord graph obtained by the i-1 th iteration is the initialized clean normalized chord graph, and the clean medical image obtained by the i-1 st iteration is the initialized clean medical image; and when i is equal to N, the clean medical image obtained by the ith iteration is used as a final de-artifact medical image corresponding to the artifact medical image sample.

15. The method of claim 13, wherein generating contaminated chordal map samples and artifact-bearing medical image samples based on the clean medical image samples comprises:

acquiring artifact simulation data;

generating a contaminated projection image from the artifact simulation data and the clean medical image sample;

adding noise to the contaminated projection image, generating the contaminated chordal map sample;

and generating the medical image sample with the artifact according to the polluted chord map sample.

16. An image deghost apparatus, the apparatus comprising:

17. An apparatus for training a deghost model, the apparatus comprising:

a sample processing module for generating contaminated chordal map samples and artifact-bearing medical image samples based on the clean chordal map samples;

the artifact removing module is used for executing N stages of iterative processing based on the artifact-carrying medical image sample and the polluted chord map sample by adopting the artifact removing model to generate a final artifact-removing medical image corresponding to the artifact-carrying medical image sample; the iterative processing of each stage is used for repairing the chord graph and removing artifacts of the medical image, and N is an integer greater than 1;

a model training module to train the deghost model based on the training loss.

18. A computer device comprising a processor and a memory, said memory having stored therein at least one instruction, at least one program, set of codes, or set of instructions, which is loaded and executed by said processor to implement the image deghost method of any one of claims 1 to 12 or to implement the training method of the deghost model of any one of claims 13 to 15.

19. A computer readable storage medium having stored therein at least one instruction, at least one program, a set of codes, or a set of instructions, which is loaded and executed by a processor to implement the image deghost method of any one of claims 1 to 12 or to implement the training method of the deghost model of any one of claims 13 to 15.

20. A computer program product or computer program, characterized in that it comprises computer instructions stored in a computer-readable storage medium, which are read by a processor from the computer-readable storage medium and executed to implement the image deghosting method according to any one of claims 1 to 12 or to implement the training method of the deghosting model according to any one of claims 13 to 15.