Image de-artifacting method, apparatus, device, and storage medium

Abstract

Embodiments of the present application disclose a kind of image deartifact method, device, equipment and storage medium, involve artificial intelligence technical field.The method includes: obtaining the medical image with artifact and contaminated string diagram, and the medical image with artifact and contaminated string diagram are corresponding image for the same object;Based on the medical image with artifact and contaminated string diagram, N stage iteration processing is executed, and the final deartifact medical image corresponding to the medical image with artifact is generated;Wherein, the iteration processing of each stage is used to repair string diagram and deartifact medical image, and N is an integer greater than 1.The present application constructs a joint string diagram and medical image reconstruction model, then proposes an iterative solution algorithm, and expands it into a deep network structure one by one, realizes the reconstruction of the medical image with artifact and contaminated string diagram, and can improve the effect of image deartifact.

Description

[0001] This application claims priority to Chinese patent application No. 202111057910.X filed on September 9, 2021, entitled “Image artifact removal method, device, apparatus and storage medium,” the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The embodiments of the present application relate to the field of artificial intelligence technology, and in particular to an image artifact removal method, apparatus, device, and storage medium. Background Art

[0003] CT (Computed Tomography) images reconstructed based on X-ray projections play a very important role in clinical diagnosis and treatment planning.

[0004] When a patient carries metal objects (such as dentures or hip prostheses) in their body, the CT image will show artifacts such as stripes and shadows due to the absorption of X-rays by the metal, which may affect the clinical diagnosis. In the related art, a dual domain network is provided, which includes a chord graph processing module and a CT image processing model. First, the chord graph is repaired by the chord graph processing module to obtain a repaired chord graph; the repaired chord graph is filtered and projected to obtain a reconstructed CT image, and then the reconstructed CT image is processed by the CT image processing module, and finally a reconstructed artifact-free CT image is output.

[0005] However, the artifact removal effect of the above method is still not ideal. Summary of the Invention

[0006] The embodiments of the present application provide an image artifact removal method, apparatus, device, and storage medium, which can improve the effect of image artifact removal. The technical solution is as follows:

[0007] According to one aspect of an embodiment of the present application, a method for removing image artifacts is provided, the method comprising:

[0008] Acquire a medical image with artifacts and a contaminated chord map, wherein the medical image with artifacts and the contaminated chord map are a set of corresponding images of the same object;

[0009] Based on the artifact-bearing medical image and the contaminated chord diagram, N stages of iterative processing are performed to generate a final artifact-free medical image corresponding to the artifact-bearing medical image; wherein each stage of iterative processing is used to repair the chord diagram and to perform artifact-free processing on the medical image, and N is an integer greater than 1.

[0010] According to one aspect of an embodiment of the present application, a method for training an artifact removal model is provided, the method comprising:

[0011] obtain sample data, the sample data comprising a clean medical image sample and a clean Hounsfield Unit (HU) map sample, the clean medical image sample and the clean HU map sample being a set of corresponding images for a same subject;

[0012] generate, based on the clean medical image sample, a contaminated HU map sample and a medical image sample with artifacts;

[0013] perform, by using the de-artifact model, N-stage iterative processing based on the medical image sample with artifacts and the contaminated HU map sample, to generate a final de-artifact medical image corresponding to the medical image sample with artifacts; wherein each stage of the iterative processing is configured to perform repair processing on the HU map and de-artifact processing on the medical image, and N is an integer greater than 1.

[0014] calculate, according to the clean medical image sample and the final de-artifact medical image, a training loss of the de-artifact model;

[0015] train the de-artifact model based on the training loss.

[0016] According to an aspect of an embodiment of the present application, an image de-artifact device is provided, and the device comprises:

[0017] an image obtaining module configured to obtain a medical image with artifacts and a contaminated HU map, the medical image with artifacts and the contaminated HU map being a set of corresponding images for a same subject;

[0018] a de-artifact module configured to perform N-stage iterative processing based on the medical image with artifacts and the contaminated HU map, to generate a final de-artifact medical image corresponding to the medical image with artifacts; wherein each stage of the iterative processing is configured to perform repair processing on the HU map and de-artifact processing on the medical image, and N is an integer greater than 1.

[0019] According to an aspect of an embodiment of the present application, a training device of a de-artifact model is provided, and the device comprises:

[0020] a sample obtaining module configured to obtain sample data, the sample data comprising a clean medical image sample and a clean HU map sample, the clean medical image sample and the clean HU map sample being a set of corresponding images for a same subject;

[0021] a sample processing module configured to generate, based on the clean medical image sample, a contaminated HU map sample and a medical image sample with artifacts;

[0022] an artifact removal module, configured to employ the artifact removal model and, based on the artifact-bearing medical image sample and the contaminated chord diagram sample, perform N stages of iterative processing to generate a final artifact-free medical image corresponding to the artifact-bearing medical image sample; wherein each stage of iterative processing is used to repair the chord diagram and perform artifact removal processing on the medical image, and N is an integer greater than 1;

[0023] a loss calculation module, configured to calculate a training loss of the artifact removal model based on the clean medical image sample and the final artifact-removed medical image;

[0024] A model training module is used to train the artifact removal model based on the training loss.

[0025] According to one aspect of an embodiment of the present application, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the above-mentioned image artifact removal method, or to implement the above-mentioned artifact removal model training method.

[0026] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the above-mentioned image artifact removal method, or to implement the above-mentioned artifact removal model training method.

[0027] According to one aspect of an embodiment of the present application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the aforementioned image artifact removal method or implement the aforementioned artifact removal model training method.

[0028] The technical solutions provided by the embodiments of the present application may have the following beneficial effects:

[0029] By performing N stages (N is an integer greater than 1) of iterative processing on the artifact-bearing medical image and its corresponding contaminated chord diagram, a final de-artifacted medical image corresponding to the artifact-bearing medical image is generated. The iterative processing in each stage is used to repair the chord diagram and de-artifact the medical image, thereby realizing mutual constraint between the chord diagram and the medical image, making the information interaction between the chord diagram and the medical image more sufficient, improving the effect of de-artifacting processing, and being more conducive to the operation of subsequent clinical tasks.

[0030] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0032] Figure 1 is a schematic diagram of an artifact removal method provided by an embodiment of the present application;

[0033] Figure 2 This is a schematic diagram of an implementation environment for a solution provided by an embodiment of the present application;

[0034] Figure 3 This is a flow chart of an image artifact removal method provided by an embodiment of the present application;

[0035] Figure 4 This is a schematic diagram of the structure of an artifact removal model provided by an embodiment of the present application;

[0036] Figure 5 is a flowchart of an image artifact removal method provided by another embodiment of the present application;

[0037] Figure 6 is a schematic structural diagram of an artifact removal model provided by another embodiment of the present application;

[0038] Figure 7 is a schematic diagram of an artifact removal method provided by another embodiment of the present application;

[0039] Figure 8 is a flowchart of an image artifact removal method provided by another embodiment of the present application;

[0040] Figure 9 is a schematic structural diagram of an artifact removal model provided by another embodiment of the present application;

[0041] Figure 10is a schematic diagram of channel cascading and separation operations provided by one embodiment of the present application;

[0042] Figure 11 is a schematic diagram of channel cascading and separation operations provided by another embodiment of the present application;

[0043] Figure 12 This is a flowchart of a method for training an artifact removal model provided by an embodiment of the present application;

[0044] Figure 13 This is a flowchart of an artifact removal model construction experiment provided by an embodiment of the present application;

[0045] Figure 14 Schematic diagram of experimental results of an artifact removal model provided by one embodiment of the present application;

[0046] Figure 15 This is a block diagram of an image artifact removal device provided by one embodiment of the present application;

[0047] Figure 16 is a block diagram of an image artifact removal device provided by another embodiment of the present application;

[0048] Figure 17 This is a block diagram of a training device for a de-artifacting model provided by one embodiment of the present application;

[0049] Figure 18 is a block diagram of a training device for a de-artifacting model provided by another embodiment of the present application;

[0050] Figure 19 This is a structural block diagram of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0051] The embodiments of the present application will be described in further detail below with reference to the accompanying drawings.

[0052] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0053] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0054] Computer vision (CV) is the science of making machines "see." Specifically, it refers to using cameras and computers to replace the human eye in identifying and measuring objects, and then further processing the images to create images more suitable for human observation or transmission to instruments. As a scientific discipline, computer vision studies related theories and technologies, aiming to build artificial intelligence systems that can extract information from images or multidimensional data. Computer vision technologies typically include image segmentation, image recognition, image semantic understanding, image retrieval, optical character recognition (OCR), video processing, video semantic understanding, video content / action recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, and other technologies. It also includes common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0055] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning through demonstration.

[0056] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, autonomous driving, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0057] like Figure 1As shown, the solution provided by the embodiments of this application involves the fields of computer vision and machine learning. Computer vision and machine learning techniques are used to train a de-artifacting model, which is then used to de-artifact medical images with artifacts and contaminated chord diagrams to obtain a final de-artifacted medical image. This is specifically illustrated by the following embodiments.

[0058] Please refer to Figure 2 , which shows a schematic diagram of an implementation environment of a solution provided by an embodiment of the present application. The implementation environment of the solution can be implemented as a system architecture for image processing. The implementation environment of the solution may include: a terminal 100 and a server 200.

[0059] The terminal 100 can be an electronic device such as a PC (Personal Computer), a tablet computer, a mobile phone, or a medical device. A client running a target application can be installed in the terminal 100. The target application can be a telemedicine application or other application that provides image processing capabilities, such as a medical computing application, a sports and health application, or a life service application, but this application does not limit this. Furthermore, this application does not limit the form of the target application, including but not limited to an App (Application), a mini-program, etc. installed in the terminal 100, or a web page. The terminal 100 can also be a medical device. Optionally, the medical device is used to acquire medical images. Optionally, the medical device can be used to acquire medical images and simultaneously perform artifact removal processing on medical images with artifacts. In one example, the medical device performs artifact removal processing on medical images with artifacts by establishing a connection with a PC. In another example, the medical device can perform artifact removal processing on medical images with artifacts by itself.

[0060] The server 200 may be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The server 200 may be a background server of the target application, used to provide background services for the client of the target application.

[0061] The terminal 100 and the server 200 may communicate with each other via a network, such as a wired or wireless network.

[0062] The image processing method provided in the embodiment of the present application may be performed by a computer device, which refers to an electronic device with data calculation, processing and storage capabilities. Figure 1Taking the illustrated implementation environment as an example, the image artifact removal method can be executed by the terminal 100 (e.g., a client of a target application installed and running on the terminal 100 executes the image artifact removal method), or by the server 200, or by the terminal 100 and the server 200 working together to execute the method, which is not limited in this application. For ease of description, the following method embodiments will only illustrate the steps of the image artifact removal method as performed by a computer device.

[0063] Furthermore, the technical solution of this application can be combined with blockchain technology. For example, some of the data involved in the image artifact removal method disclosed in this application (such as medical images, chord diagrams, etc.) can be stored on the blockchain.

[0064] Optionally, the technical solution provided in this application can be applied in medical scenarios. For example, the medical device is connected to the PC, and the PC is connected to the server. Optionally, the medical device and the PC, and the PC and the server can communicate through a wired connection or a wireless connection, which is not limited in this application. The medical device collects medical images and sends the medical images to the PC, which then uploads the medical images to the server. The server performs artifact removal on the medical images and then sends the artifact-free medical images to the PC. The doctor can make a diagnosis by comparing the artifact-free medical images. Optionally, after receiving the medical images sent by the medical device, the PC directly performs artifact removal on the medical images and then directly displays the artifact-free medical images without uploading to the server.

[0065] Please refer to Figure 3 , which shows a flow chart of an image artifact removal method provided by an embodiment of the present application. The method may include the following steps (210-220):

[0066] Step 210: Acquire a medical image with artifacts and a contaminated chord map, where the medical image with artifacts and the contaminated chord map are a group of corresponding images of the same object.

[0067] The aforementioned medical image with artifacts refers to a medical image contaminated by artifacts, where artifacts refer to signals in the medical image that do not conform to the actual structure and may appear as image deformation, overlap, loss, blur, etc. Optionally, the medical image includes a CT image.

[0068] The contaminated chord diagram is the chord diagram corresponding to the artifact-bearing medical image. A chord diagram is a graph that displays the relationship between data. The contaminated chord diagram is the projection data of the corresponding artifact-bearing medical image.

[0069] It should be noted that in the embodiments provided in this application, the medical image with artifacts can be acquired before the contaminated chord diagram, after the contaminated chord diagram, or at the same time as the contaminated chord diagram, and this application does not limit this.

[0070] In one example, a medical image with artifacts is first acquired, and a corresponding contaminated chord diagram is generated based on the artifact-bearing medical image. Optionally, the corresponding contaminated chord diagram is obtained by forward projecting the artifact-bearing medical image. Forward projection is a computational method that converts a medical image into a corresponding chord diagram using a Radon transform.

[0071] In another example, a contaminated chord diagram is first obtained, and a corresponding medical image with artifacts is generated based on the contaminated chord diagram. Optionally, the corresponding medical image with artifacts is obtained by filtering and then projecting the contaminated chord diagram.

[0072] In another example, a medical image with artifacts and its corresponding contaminated chord graph are acquired simultaneously.

[0073] Optionally, the contaminated chord diagram and the medical image with artifacts may be obtained through a physical device or from a medical image library with artifacts, which is not limited in this application.

[0074] Furthermore, the artifact-bearing medical image and the contaminated chord diagram are a pair of corresponding images of the same subject, and the artifact-bearing medical image and the contaminated chord diagram are a pair of corresponding images of the same body part. For example, the body part can be the head, chest, hand, leg, abdomen, etc., which is not limited in this application.

[0075] Step 220: Based on the artifact-bearing medical image and the contaminated chord diagram, perform N stages of iterative processing to generate a final artifact-free medical image corresponding to the artifact-bearing medical image; wherein each stage of iterative processing is used to repair the chord diagram and perform artifact-free processing on the medical image, and N is an integer greater than 1.

[0076] The above-mentioned final artifact-free medical image refers to the medical image obtained after the artifact-containing medical image is subjected to artifact-free processing. It should be noted that in the above-mentioned N stages of iterative processing, each stage of iterative processing is used to repair the chord diagram and perform artifact-free processing on the medical image, that is, the degree of artifact contamination of the image obtained after processing is less than the degree of artifact contamination of the image before processing. In the embodiment of the present application, the image obtained by the iterative processing of each stage is referred to as a clean medical image and a clean normalized chord diagram, but it should be understood that the clean medical image and the clean normalized chord diagram are not images that are completely free of artifact contamination, including the final artifact-free medical image obtained after the N stages of iterative processing. It may not be a medical image that is completely free of artifacts, but it has fewer artifacts than the initial artifact-containing medical image.

[0077] Exemplarily, as shown in Figure 4 Fig. 1, based on the medical image with artifacts X ma and the contaminated string diagram Y, an N-stage iterative processing is performed to generate the final de-artifact medical image X ma corresponding to the medical image with artifacts. N .

[0078] Optionally, the N-stage iterative processing described above finally outputs, in addition to the final de-artifact medical image corresponding to the medical image with artifacts, also includes the final de-contaminated string diagram corresponding to the medical image with artifacts, which is not limited in the present application.

[0079] Optionally, the iterative processing of each stage can first perform repair processing on the string diagram, and then perform de-artifact processing on the medical image, or first perform de-artifact processing on the medical image, and then perform repair processing on the string diagram, and the present application only takes the example of first performing repair processing on the string diagram, and then performing de-artifact processing on the medical image for each stage of iterative processing.

[0080] In the embodiments of the present application, in the N-stage iterative processing, the repair processing on the string diagram and the de-artifact processing on the medical image are alternately performed, and the N-stage iterative processing is sequentially performed, wherein the output of the i-1th iteration processing is taken as the input of the ith iteration processing, and i is an integer greater than 0.

[0081] In one example, taking N=3 as an example, the first stage first performs repair processing on the string diagram, and then performs de-artifact processing on the medical image, the second stage continues to first perform repair processing on the string diagram, and then performs de-artifact processing on the medical image, and then the third stage still first performs repair processing on the string diagram, and then performs de-artifact processing on the medical image, and the final de-artifact medical image is output after the third stage is completed. Optionally, after the third stage is completed, the final output also includes the final de-contaminated string diagram.

[0082] In another example, taking N=3 as an example, the first stage first performs de-artifact processing on the medical image, and then performs repair processing on the string diagram, the second stage continues to first perform de-artifact processing on the medical image, and then performs repair processing on the string diagram, and then the third stage still first performs de-artifact processing on the medical image, and then performs repair processing on the string diagram, and the final de-artifact medical image is output after the third stage is completed. Optionally, after the third stage is completed, the final output also includes the final de-contaminated string diagram.

[0083] Optionally, the value of N can be set according to the training result of the de-artifact model, or can be set according to experience, which is not limited in the present application.

[0084] Optionally, an artifact removal model is used to implement the above-mentioned N-stage iterative processing. That is, the artifact removal model is used to perform N-stage iterative processing based on artifact-bearing medical image samples and contaminated chord diagram samples to generate a final artifact-free medical image corresponding to the artifact-bearing medical image samples; wherein the iterative processing of each stage is used to repair the chord diagram and perform artifact removal processing on the medical image, and N is an integer greater than 1. In one example, the artifact removal model includes N cascaded artifact removal network modules for performing N-stage iterative processing; each artifact removal network module includes: a first neural network for repairing the chord diagram, and a second neural network for performing artifact removal processing on the medical image.

[0085] In summary, the technical solution provided in the embodiment of the present application generates a final de-artifacted medical image corresponding to the medical image with artifacts by performing N stages (N is an integer greater than 1) of iterative processing on the medical image with artifacts and its corresponding contaminated chord diagram, wherein the iterative processing of each stage is used to repair the chord diagram and de-artifact the medical image, thereby realizing mutual constraints between the chord diagram and the medical image, making the information interaction between the chord diagram and the medical image more sufficient, improving the effect of de-artifacting processing, and being more conducive to the operation of subsequent clinical tasks.

[0086] In addition, this method constructs the de-artifacting model by unfolding the corresponding iterative algorithm one by one, so that each network module of the de-artifacting model has a specific physical meaning. The de-artifacting model has strong physical interpretability, and the entire de-artifacting process is transparent and visible, making it easier for users to understand the role of each module and facilitating the debugging and use of the de-artifacting model.

[0087] Please refer to Figure 5 , which shows a flow chart of a method for removing artifacts provided by another embodiment of the present application. The method may include the following steps (310-360).

[0088] Step 310: Acquire a medical image with artifacts and a contaminated chord map, where the medical image with artifacts and the contaminated chord map are a group of corresponding images of the same object.

[0089] Step 320: Based on the medical image with artifacts and the contaminated chord graph, generate an initialized clean medical image, an initialized clean normalized chord graph, and a normalization factor, wherein the contour feature of the normalization factor is better than the contour feature of the contaminated chord graph.

[0090] In one example, an initialized clean medical image refers to a medical image with artifacts that has been initially de-artifacted, and an initialized clean normalized chord graph refers to a chord graph that has been initially repaired, and is initially processed. Optionally, the initialization process includes channel concatenation and channel separation operations. Channel concatenation involves concatenating images along the channel dimension to obtain a multi-channel tensor as input. Channel separation involves splitting the multi-channel tensor along the channel dimension, with the first channel serving as the image and the remaining channels serving as auxiliary variables.

[0091] In an example, the initialized clean medical image and the initialized clean normalized chord graph can be obtained through the following steps 321 to 323 .

[0092] Step 321: Obtain a reconstructed chord diagram and a reconstructed medical image corresponding to the medical image with artifacts.

[0093] A reconstructed chord graph refers to an image obtained by preliminary repairing a contaminated chord graph; a reconstructed medical image refers to an image obtained by preliminary removing artifacts from a medical image with artifacts.

[0094] Optionally, a linear interpolation algorithm is used to perform preliminary repair on the contaminated chord diagram to obtain a reconstructed chord diagram. Optionally, a reconstructed medical image is obtained based on the reconstructed chord diagram. Exemplarily, the reconstructed chord diagram is filtered and then projected to obtain a reconstructed medical image. The linear interpolation algorithm refers to a method for removing image artifacts, but its operation is simple and it is easy to introduce new artifacts, and it will also cause the loss of tissue structure in the medical image. In an embodiment of the present application, a linear interpolation algorithm is used to perform preliminary repair on the contaminated chord diagram, and then a reconstructed medical image is obtained based on the reconstructed chord diagram, so that the degree of contamination of the medical image with artifacts is reduced, which reduces the workload of the de-artifacting model to a certain extent, and helps to reduce the number of de-artifacting network modules required by the model.

[0095] Step 322: Perform a first proximal operation on the reconstructed chord graph to obtain an initialized clean normalized chord graph. The first proximal operation refers to an operation of repairing the chord graph through a first neural network.

[0096] Step 323: Perform a second proximal operation on the reconstructed medical image to obtain an initialized clean medical image. The second proximal operation refers to an operation of performing artifact removal processing on the medical image through a second neural network.

[0097] In one example, the normalization factor is obtained by normalizing the contaminated chord graph, where the normalization process makes the contour features of the contaminated chord graph more uniform and flat.

[0098] Exemplarily, the normalization factor can be obtained by the following steps 324 to 326. Optionally, the artifact removal model that uses the following steps 324 to 326 to obtain the normalization factor is called InDuDoNet (Interpretability Dual Domain Network).

[0099] Step 324: Generate a priori medical image based on the medical image with artifacts. The priori medical image refers to a medical image having common features of a clean medical image.

[0100] In order to distinguish it from the following embodiments, the "prior medical image" in step 324 here is referred to as the "first prior medical image", that is, step 324 is: generating a first prior medical image based on the medical image with artifacts, and the first prior medical image refers to a medical image having common features of a clean medical image.

[0101] Optionally, the medical image with artifacts is processed by a priori network to generate a first priori medical image. For example, Figure 4 As shown, the medical image X with artifacts is trained by the prior network ma Processing is performed to generate a first priori medical image. Optionally, the medical image with artifacts and the reconstructed medical image are cascaded and input into the priori network, and the first priori medical image is output through the priori network. In the embodiment of the present application, the structure of the priori network is not limited. For example, the priori network is a U-shaped structure. For example, Figure 7 As shown, the medical image with artifacts 620 and the reconstructed medical image 610 are cascaded and input into the priori network 630 to obtain a first priori medical image A.

[0102] Step 325: Fusing the first priori medical image and the reconstructed medical image corresponding to the medical image with artifacts to obtain a fused image.

[0103] The reconstructed medical image refers to an image obtained by performing preliminary de-artifacting on a medical image with artifacts. Optionally, the above fusion process is an addition process. For example, the pixel values ​​at corresponding positions of the prior medical image and the reconstructed medical image are added to obtain a fused image. For example, Figure 7 As shown, the pixel values ​​at corresponding positions of the first priori medical image A and the reconstructed medical image 610 are added together to obtain a fused image B.

[0104] Step 326: Perform forward projection on the fused image to obtain a normalization factor.

[0105] For example, Figure 7 As shown, the fused image B is forward projected to obtain a normalization factor 640.

[0106] It should be noted that the normalization factor can be obtained by forward projecting the fused image, directly forward projecting the medical image with artifacts, or forward projecting the prior medical image, and this application does not limit this. It should be noted that the generation of the normalization factor can be performed before generating the initialized clean medical image and the initialized clean normalized chord diagram, or after generating the initialized clean medical image and the initialized clean normalized chord diagram, or can be performed simultaneously with the generation of the initialized clean medical image and the initialized clean normalized chord diagram, and this application does not limit this.

[0107] For example, Figure 8 As shown, the normalization factor can also be obtained through the following steps 327 to 328. Optionally, the artifact removal model that uses the following steps 327 to 328 to obtain the normalization factor is called InDoDuNet+ (Interpretable Dual Domain Network+).

[0108] Step 327: Generate a second priori medical image based on the medical image with artifacts and the reconstructed medical image corresponding to the medical image with artifacts.

[0109] For example, Figure 6 As shown, according to the medical image X with artifacts ma and the reconstructed medical image X corresponding to the medical image with artifacts LI , generating a second prior medical image.

[0110] The reconstructed medical image refers to an image obtained by preliminary de-artifacting a medical image with artifacts, and the second prior medical image refers to a medical image having common features of clean medical images and unique features of medical images with artifacts.

[0111] Optionally, an initial prior medical image is generated based on the reconstructed medical image, where the initial prior medical image refers to a medical image having common features of a clean medical image; the initial prior medical image and the medical image with artifacts are weightedly fused to obtain a second prior medical image.

[0112] Optionally, an initial prior medical image is generated based on the reconstructed chord diagram and the prior statistical properties of the clean medical image. The reconstructed medical image is combined with the prior statistical properties of the clean medical image and processed using a manual regularization method to obtain the initial prior medical image, thereby enhancing the physical interpretability of the network.

[0113] Optionally, the medical image with artifacts is processed through a weighted network to obtain a weight matrix; and a dot product calculation is performed on the initial prior medical image and the weight matrix to obtain a second prior medical image.

[0114] For example, Figure 9As shown, the reconstructed medical image X obtained by the linear interpolation algorithm LI , combined with the prior statistical characteristics of the clean medical image, an initial prior medical image is generated. The artifact-bearing medical image Xma is processed through the weighted network WNet to obtain a weight matrix. The initial prior medical image and the weight matrix are multiplied to obtain a second prior medical image.

[0115] Step 328: Perform forward projection on the second priori medical image to obtain a normalization factor.

[0116] Compared with the first a priori medical image acquisition method, the second a priori medical image acquisition method uses a lightweight weighted network to process reconstructed medical images and medical images with artifacts, which reduces the number of parameters required by the network and improves the cross-domain generalization potential of the network. In addition, the weighted network has stronger physical interpretability.

[0117] Step 330: For the i-th stage in the N stages, obtain an updated normalized chord graph according to the normalization factor, the clean normalized chord graph obtained at the i-1-th iteration, the clean medical image obtained at the i-1-th iteration, and the contaminated chord graph.

[0118] In one example, an updated normalized chord graph is obtained by calculating the normalization factor, the clean normalized chord graph obtained at the i-1th iteration, the clean medical image obtained at the i-1th iteration, and the contaminated chord graph. Exemplarily, the updated normalized chord graph is calculated using formula (1).

[0119]

[0120] Where, refers to the updated normalized chord diagram; refers to the clean normalized chord diagram obtained at the i-1th iteration; X i-1 refers to the clean medical image obtained at the i-1th iteration; Y refers to the contaminated chord graph; is the normalization factor; η1 is the update step size; P is the Radon transform, i.e., the forward projection operation; α is the summation factor used to balance the consistency of spatial domain and Radon domain data; T r is the metal track in the chord diagram, whose elements are {0,1}, where 1 represents the metal track area; and All operations are performed.

[0121] For example, Figure 7 As shown, an updated normalized chord graph 663 is obtained according to the normalization factor 640 , the clean normalized chord graph 661 obtained at the (i-1)th iteration, the clean medical image 662 obtained at the (i-1)th iteration, and the contaminated chord graph 650 .

[0122] One thing that needs to be explained is that the updated normalized chord diagram obtained in the i-1th iterative processing is used for the i-th iterative processing, but it should be understood that the acquisition process of the above-mentioned updated normalized chord diagram is the same in the iterative processing of each stage, but since the input data of the iterative processing of each stage is different, the updated normalized chord diagram obtained in the iterative processing of each stage is also different.

[0123] In one example, the updated normalized chord diagram can be obtained by the following steps 331 to 332 .

[0124] Step 331: Obtain a first intermediate result according to the normalization factor, the clean normalized chord graph obtained at the (i-1)th iteration, the clean medical image obtained at the (i-1)th iteration, and the contaminated chord graph.

[0125] In one example, the normalization factor, the clean normalized chord diagram obtained at the i-1th iteration, the clean medical image obtained at the i-1th iteration, and the contaminated chord diagram are calculated to obtain a first intermediate result. Exemplarily, the first intermediate result is obtained by calculating using formula (2).

[0126]

[0127] Where, refers to the clean normalized chord diagram obtained at the i-1th iteration; X i-1 refers to the clean medical image obtained at the i-1th iteration; Y refers to the contaminated chord graph; is the normalization factor; η1 is the update step size; P is the Radon transform, i.e., the forward projection operation; α is the summation factor used to balance the consistency of spatial domain and Radon domain data; T r is the metal track in the chord diagram, whose elements are {0,1}, where 1 represents the metal track area; and It should be noted that ⊙ in this application represents dot multiplication calculation, which will not be described in detail in the following text.

[0128] It should be noted that the acquisition process of the above-mentioned first intermediate result is the same in the iterative processing of each stage, but since the input data of the iterative processing of each stage is different, the first intermediate result obtained by the iterative processing of each stage is also different.

[0129] Step 332: Obtain an updated normalized chord diagram based on the clean normalized chord diagram obtained in the (i-1)th iteration and the first intermediate result.

[0130] In one example, the clean normalized chord diagram obtained in the (i-1)th iteration and the first intermediate result are calculated to obtain an updated normalized chord diagram. Exemplarily, the updated normalized chord diagram is calculated using formula (3).

[0131]

[0132] Where, refers to the updated normalized chord diagram; η1 is the update step size; is the partial derivative of formula (2).

[0133] Step 340: Perform a first proximal operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained in the i-th iteration. The first proximal operation is used to repair the chord graph through the first neural network.

[0134] In one example, the first neural network is a first proximal network, which processes the updated normalized chord graph using a proximal gradient technique to obtain a clean normalized chord graph obtained in the i-th iteration.

[0135] Optionally, the first neural network is a residual network composed of T S Residual blocks (Resblocks) (T S is a positive integer), each residual block is composed of a convolutional layer, a batch normalization layer, a ReLU activation layer, a convolutional layer, a batch normalization layer, and a cross-link. For example, the convolution kernel size of the convolutional layer is 3*3, and the stride is 1.

[0136] For example, Figure 7 As shown, the updated normalized chord diagram 663 is input into the first neural network 664, and the clean normalized chord diagram 665 obtained in the i-th iteration is output.

[0137] Optionally, before performing the first proximal operation, the updated normalized chord diagram and the first auxiliary variable obtained in the i-1th iteration are concatenated to obtain an updated normalized chord diagram after the i-th concatenation; then the first proximal operation is performed on the updated chord diagram after the i-th concatenation to obtain a clean normalized chord diagram after the i-th concatenation; the clean normalized chord diagram after the i-th concatenation is subjected to channel splitting, and the image of the first channel is selected as the clean normalized chord diagram obtained in the i-th iteration. In one example, as Figure 10As shown, the updated normalized chord graph and the first auxiliary variable obtained from the i-1th iteration are cascaded, and the updated normalized chord graph obtained after the i-th cascade is input into the first neural network to obtain a clean normalized chord graph after the i-th cascade. The clean normalized chord graph after the i-th cascade is then split into channels, and the first channel is selected as the first channel, the image of the first channel is used as the clean normalized chord graph obtained from the i-th iteration, and the remaining channels are used as the first auxiliary variables obtained from the i-th iteration for the i+1th iteration stage. Exemplarily, if the number of channels of the clean normalized chord graph after the i-th cascade is 33, then the first channel is used as the first channel, the image of the first channel is used as the clean normalized chord graph obtained from the i-th iteration, and the images of the remaining 32 channels are used as the first auxiliary variables obtained from the i-th iteration for the i+1th iteration stage. Optionally, the first channel is any channel of the channels of the clean normalized chord graph after the i-th cascade, which is not limited in this application. Exemplarily, the first channel may be the first channel among the channels of the clean normalized chord graph after the i-th cascade, or may be the last channel among the channels of the clean normalized chord graph after the i-th cascade.

[0138] Step 350: Obtain an updated medical image according to the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the i-th iteration.

[0139] In one example, an updated medical image is obtained by calculating the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the (i)th iteration. Exemplarily, the updated medical image is obtained by calculating formula (4).

[0140]

[0141] Where, Refers to the updated medical image; X i-1 refers to the clean medical image obtained at the i-1th iteration; refers to the clean normalized chord diagram obtained at the i-th iteration; refers to the normalization factor; η2 is the update step size; P refers to the Radon transform, that is, the forward projection operation; P T Refers to the transpose operation of Radon transform; where and Perform the operation.

[0142] For example, Figure 7 As shown, an updated medical image 667 is obtained according to the normalization factor 640, the clean medical image 666 obtained at the (i-1)th iteration, and the clean normalized chord diagram 665 obtained at the i-th iteration.

[0143] It should be noted that the updated medical image obtained by the i-1th iterative processing is used for the i-th iterative processing, but it should be understood that the acquisition process of the above-mentioned updated medical image is the same in the iterative processing of each stage, but since the input data of the iterative processing of each stage is different, the updated medical image obtained by the iterative processing of each stage is also different.

[0144] In one example, the updated medical image can be obtained through the following steps 351 to 352 .

[0145] Step 351: Obtain a second intermediate result according to the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the i-th iteration.

[0146] In one example, the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the (i)th iteration are calculated to obtain a second intermediate result. Exemplarily, the second intermediate result is obtained by calculation using formula (5).

[0147]

[0148] Where, X i-1 refers to the clean medical image obtained at the i-1th iteration; refers to the clean normalized chord diagram obtained at the i-th iteration; is the normalization factor; η2 is the update step size; P is the Radon transform, i.e. the forward projection operation; where and Perform the operation.

[0149] It should be noted that the acquisition process of the above-mentioned second intermediate result is the same in the iterative processing of each stage, but since the input data of the iterative processing of each stage is different, the second intermediate result obtained by the iterative processing of each stage is also different.

[0150] Step 352: Obtain an updated medical image based on the clean medical image obtained in the (i-1)th iteration and the second intermediate result.

[0151] In one example, the clean medical image obtained in the (i-1)th iteration and the second intermediate result are calculated to obtain an updated medical image. Exemplarily, the updated medical image is calculated using formula (6).

[0152]

[0153] Where, refers to the updated medical image; η2 is the update step size; is the partial derivative of formula (5).

[0154] Step 360: Perform a second proximal operation on the updated medical image to obtain a clean medical image obtained in the i-th iteration, where the second proximal operation is used to perform artifact removal processing on the medical image through a second neural network.

[0155] In one example, the second neural network is a second proximal network, which processes the updated medical image using a proximal gradient technique to obtain a clean medical image obtained in the i-th iteration.

[0156] Optionally, the second neural network is a residual network composed of T S Residual blocks (Resblocks) (T S is a positive integer), and each residual block is composed of a convolutional layer, a batch normalization layer, a ReLU activation layer, a convolutional layer, a batch normalization layer, and a cross-link. Exemplarily, the convolution kernel size of the convolutional layer is 3*3, and the stride is 1. Optionally, the structures of the first neural network and the second neural network can be the same or different, and this application does not limit this.

[0157] For example, Figure 7 As shown, the updated medical image 667 is input into the second neural network 668, and the clean medical image 669 obtained in the i-th iteration is output.

[0158] Optionally, before performing the second proximal operation, the updated medical image and the second auxiliary variable obtained in the i-1th iteration are cascaded to obtain the i-th cascaded medical image; then the second proximal operation is performed on the i-th cascaded medical image to obtain the i-th cascaded clean medical image; the i-th cascaded clean medical image is subjected to channel splitting, and the image of the second channel is selected as the clean medical image obtained in the i-th iteration. In one example, Figure 11As shown, the updated medical image and the second auxiliary variable obtained in the (i-1)th iteration are concatenated, and the obtained medical image after the (i-1)th cascade is then input into the second neural network to obtain a clean medical image after the (i)th cascade. The clean medical image after the (i)th cascade is then split into channels, with the first channel serving as the second channel, the image in the second channel serving as the clean medical image obtained in the (i)th iteration, and the remaining channels serving as the second auxiliary variables obtained in the (i)th iteration for use in the (i+1)th iteration. Exemplarily, the number of channels in the clean medical image after the (i)th cascade is 33, of which the first channel is serving as the second channel, the image in the second channel serving as the clean medical image obtained in the (i)th iteration, and the images in the remaining 32 channels serving as the second auxiliary variables obtained in the (i)th iteration for use in the (i+1)th iteration. Optionally, the first channel can be any channel in the channels of the clean medical image after the (i)th cascade, which is not limited in this application. Exemplarily, the first channel can be the first channel in the channels of the clean medical image after the (i)th cascade, or the last channel in the channels of the clean medical image after the (i)th cascade.

[0159] It should be noted that i is a positive integer less than or equal to N. When i = 1, the clean normalized chord diagram obtained at the i-1th iteration is the initialized clean normalized chord diagram, and the clean medical image obtained at the i-1th iteration is the initialized clean medical image; when i = N, the clean medical image obtained at the i-th iteration is used as the final de-artifacted medical image corresponding to the medical image with artifacts.

[0160] To sum up, in the technical solution provided in the embodiment of the present application, a method of cascading input of reconstructed medical images and medical images with artifacts is adopted. Compared with only inputting reconstructed medical images, the data of tissue structure in the complete medical image is supplemented, and to a certain extent, the loss of tissue structure in the medical image is avoided. Compared with only inputting medical images with artifacts, preliminary de-artifacting processing is performed on the medical images with artifacts, which reduces the workload of the de-artifacting model and thus reduces the number of de-artifacting network modules in the de-artifacting model. By combining a weighted network with the prior statistical characteristics of a clean medical image, the network requires fewer parameters, improves the cross-domain generalization ability of the network, and at the same time, the weighted network has stronger physical interpretability.

[0161] Below, the training process of the artifact removal model is introduced and explained through an embodiment. The contents involved in the use of the artifact removal model and the contents involved in the training process correspond to each other, and the two are interconnected. If one side is not explained in detail, you can refer to the description on the other side.

[0162] Please refer to Figure 12It shows a flow chart of a method for training a de-artifacting model provided by an embodiment of the present application. The method may include the following steps (710-750).

[0163] Step 710: Obtain sample data, where the sample data includes clean medical image samples and clean chord graph samples.

[0164] Optionally, the clean medical image samples and the clean chord graph samples are a group of corresponding images of the same object.

[0165] It should be noted that the clean medical image samples can be obtained before the clean chord map samples, or after the clean chord map samples, or simultaneously with the clean chord map samples, and this application does not limit this.

[0166] In one example, a clean medical image sample is obtained, and a corresponding clean chord diagram sample is generated based on the clean medical image sample. Optionally, the corresponding clean chord diagram sample is obtained by forward projecting the clean medical image sample.

[0167] In another example, a clean chord diagram sample is obtained, and a corresponding clean medical image sample is generated based on the clean chord diagram sample. Optionally, the clean chord diagram sample is filtered and then projected to obtain the corresponding clean medical image sample.

[0168] In another example, a clean medical image sample and its corresponding clean chord graph sample are obtained simultaneously.

[0169] Step 720: Generate contaminated chord graph samples and medical image samples with artifacts based on the clean medical image samples.

[0170] In one example, artifact simulation data is first acquired. 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 chordogram sample. Then, based on the contaminated chordogram sample, an artifact-bearing medical image sample is generated. Exemplarily, different types of metal masks are acquired, and metal artifacts are synthesized according to the data simulation process to generate a contaminated chordogram sample. The artifact-bearing medical image sample is then filtered and projected. Optionally, the contaminated projection image is obtained by performing tissue segmentation on the clean medical image sample and then performing multi-level medical image projection.

[0171] Optionally, the value range of the contaminated chord sample is clipped to obtain a clipped contaminated chord sample, limiting the data value of the contaminated chord sample to a certain range. For example, the contaminated chord sample is clipped to a threshold range of [0, 4], then normalized to the range of [0, 1] by dividing by 4, and finally multiplied by 255 to convert to the range of [0, 255].

[0172] Optionally, the cropped contaminated string pattern sample is filtered and projected to obtain a corresponding artifacted medical image sample.

[0173] Optionally, a numerical range of the artifacted medical image sample is cropped to obtain a cropped artifacted medical image sample, and the pixel value of the medical image sample is limited within a certain range. In one example, the artifacted medical image sample is clipped to a threshold value [0, 1] range, and then multiplied by 255 to convert to a [0, 255] range. In another example, the artifacted medical image sample is normalized to a [0, 1] range, and then multiplied by 255 to convert to a [0, 255] range.

[0174] Step 730: using the artifact removal model, based on the artifacted medical image sample and the contaminated string pattern sample, performing N-stage iterative processing to generate a final artifact-removed medical image corresponding to the artifacted medical image sample; wherein each stage of the iterative processing is used for repairing the string pattern and removing artifacts from the medical image, and N is an integer greater than 1.

[0175] In one example, the artifact removal model includes N cascaded artifact removal network modules, each including a first neural network for repairing the string pattern and a second neural network for removing artifacts from the medical image.

[0176] In one example, the final artifact-removed medical image can be obtained by the following steps 731-735.

[0177] Step 731: based on the artifacted medical image sample and the contaminated string pattern sample, generating an initialized clean medical image, an initialized clean normalized string pattern, and a normalization factor, the contour feature of the normalization factor being superior to the contour feature of the contaminated string pattern.

[0178] Step 732: for the i-th stage of the N stages, according to the normalization factor, the clean normalized string pattern obtained by the i-1 iteration, the clean medical image obtained by the i-1 iteration, and the contaminated string pattern sample, obtaining an updated normalized string pattern.

[0179] Step 733: using the first neural network contained in the i-th artifact removal network module, performing a first proximal operation on the updated normalized string pattern to obtain a clean normalized string pattern obtained by the i-th iteration.

[0180] Step 734: according to the normalization factor, the clean medical image obtained by the i-1 iteration, and the clean normalized string pattern obtained by the i-th iteration, obtaining an updated medical image.

[0181] Step 735: Use the second network included in the i-th artifact removal network module to perform a second proximal operation on the updated medical image to obtain a clean medical image obtained in the i-th iteration.

[0182] It should be noted that i is a positive integer less than or equal to N. When i = 1, the clean normalized chord diagram obtained at the i-1th iteration is the initialized clean normalized chord diagram, and the clean medical image obtained at the i-1th iteration is the initialized clean medical image; when i = N, the clean medical image obtained at the i-th iteration is used as the final de-artifacted medical image corresponding to the medical image sample with artifacts.

[0183] It should be noted that in the embodiments of the present application, the clean medical image, clean normalized chord diagram, updated medical image and updated normalized chord diagram obtained in the training method of the artifact removal model are not distinguished in terms of terminology from the clean medical image, clean normalized chord diagram, updated medical image and updated normalized chord diagram obtained in the image artifact removal method, but those skilled in the art should know that the two are not the same.

[0184] Step 740: Calculate the training loss of the artifact removal model based on the clean medical image sample and the final artifact-removed medical image.

[0185] In one example, a training objective function is used to calculate the training loss of the artifact removal model. Optionally, the training objective function is formula (7).

[0186]

[0187] Where, β n is a compromise parameter; γ is the weight used to balance the losses; X gt is a clean medical image sample; Y gt is a clean normalized chord graph sample; M is the metal position in the medical image; is the normalization factor; is the clean normalized chord diagram obtained in the nth iteration; X n The final artifact-free medical image.

[0188] In an exemplary embodiment, β n =0.1 (n is a natural number less than N), β N =1, γ=0.1, N=7, M, X gt and Y gt It is known in training simulation experiments.

[0189] Step 750: Train the artifact removal model based on the training loss.

[0190] In one example, the Adam (Adaptive moment estimation) algorithm is used to update the optimization parameters, including, Step lengths η1 and η2. In each iteration process, the prediction error is calculated and back-propagated to the artifact removal model, the gradient is calculated and the parameters of the artifact removal model are updated.

[0191] In one example, the initial learning rate is set to 2 x 10 -4 , and the learning rate is decayed by 0.5 every 40 iterations, with a total of 100 training iterations. In each iteration process, two neural networks are included in the iterative network module, the CT image size is 416*416, and the chord diagram size is 641*640.

[0192] It should be noted that the above describes the use process and training process of the artifact removal model respectively with different embodiments. The content involved in the use process and the content involved in the training process are corresponding to each other, and are mutually communicated. For example, if one side is not described in detail, the description of the other side can be referred to.

[0193] Based on the above-described artifact removal model, related experiments are designed to verify its artifact removal effect. In the experimental environment, the model is loaded on the cloud system, and the specific process of the experiment is described in detail in the following Figure 13 .

[0194] Step 810: Determine whether the current is a training phase or a test phase.

[0195] Step 820: If the current is in the training phase, obtain the sample data of the artifact removal model.

[0196] Step 830: Iteratively train the artifact removal model.

[0197] Step 840: Train the artifact removal model based on the training loss.

[0198] Step 850: Determine whether the set number of training iterations is reached.

[0199] Step 860: If the set number of training iterations is reached, save the trained artifact removal model; if the set number of training iterations is not reached, continue to train the artifact removal model.

[0200] Step 870: If the current is in the test phase, obtain the artifact medical image.

[0201] Step 880: Process the artifact medical image using the artifact removal model.

[0202] Step 890: Output the final artifact removal medical image.

[0203] For details about the training process, please refer to the embodiment of the training method for the artifact removal model of this application, and this application will not go into details here.

[0204] The idea of ​​the above-mentioned image artifact removal method is inspired by the process of problem derivation, and the process of problem derivation is as follows.

[0205] First, given an observed contaminated chord graph Among them, N b and N p are the number of detectors and projection angles arranged in the medical image acquisition device, respectively. Based on this, the traditional iterative artifact removal method can be expressed by formula (8).

[0206]

[0207] Where X∈R H×W represents a clean medical image (spatial domain), H and W are the height and width of the medical image respectively; P is the Radon transform, i.e., the forward projection operation, T r is the metal trajectory in the chord diagram, whose elements are {0,1}, where 1 represents the metal trajectory area, and ⊙ is the product of the corresponding elements; g(·) is the regularization term, which represents the prior structure of the clean medical image; λ is the trade-off parameter.

[0208] For the mutual learning between the Radon domain and the spatial domain, the spatial domain and the Radon domain are further regularized jointly.

[0209] At this time, formula (8) can be transformed into formula (9).

[0210]

[0211] where S refers to the clean normalized chord graph (Radon domain); α is an additive factor used to balance the data consistency between the spatial domain and the Radon domain; g1(·) and g2(·) are both regularization terms that embed the prior structures of the clean normalized chord graph S and the clean medical image X, respectively.

[0212] Since it is easier to correct the contaminated chord diagram after normalization than to correct it directly, because the normalized contour is more uniform and flat, S can be redefined as formula (10).

[0213]

[0214] Where, is the normalization factor, usually a priori medical image The forward projection is obtained, that is It is obtained through a priori network based on medical images with artifacts; is the clean normalized chord diagram after normalization.

[0215] Substituting formula (10) into formula (9), we can derive formula (11).

[0216]

[0217] In order to obtain artifact-free medical images and clean chord diagrams, formula (11) should be solved.

[0218] For example, the proximal gradient technique is used to alternately update and X. Specifically, in the nth iteration, The update problem of can be solved by solving the equation (11) The quadratic approximation problem of is realized, which can be written as formula (12).

[0219]

[0220] Where, is the updated result calculated at the n-1th iteration, η1 is the update step size,

[0221] For general regularization terms Formula (12) can be written as Formula (13).

[0222]

[0223] Will Substitute into formula (13), and we get The update rule is formula (14).

[0224]

[0225] Where, It is a proximal operator determined by the regularization term g1(·) and can be designed as a network module.

[0226] Similarly, the quadratic approximation problem of X can be written as formula (15).

[0227]

[0228] in,

[0229] Therefore, the update rule of X can be written as formula (16).

[0230]

[0231] Where, It is a proximal operator determined by the regularization term g2(·) and can be designed as a network module.

[0232] Based on formula (14) and formula (16), each iterative sub-step is expanded into a network module, thereby constructing the entire artifact removal model, which makes the artifact removal method have good physical interpretability. At the same time, the Radon domain and the spatial domain learn from each other, making the information interaction between medical images and chord diagrams more sufficient.

[0233] In addition, in the embodiment of the present application, formula (17) and formula (18) are used for initialization processing.

[0234]

[0235]

[0236] Where | is the channel separation operation; concat(·) is the channel cascade operation; S LI The reconstructed chord graph is obtained by repairing the polluted chord graph using the linear interpolation algorithm; X LI According to S LI The reconstructed medical image obtained; Q0 s To initialize the first auxiliary variable; Q0 X To initialize the second auxiliary variable; is the proximal operator determined by the regularization term g1(·); is the proximal operator determined by the regularization term g2(·); K S and K X is the convolution kernel, and the convolution kernel sizes are f s ×f s ×N s ×1 and f x ×f x ×N x ×1. During training, the convolution kernel size is f s ×f s ×N s ×1=f x ×f x ×N x ×1=3*3*32*1.

[0237] For example, this application conducted an experimental comparison on the six artifact removal methods LI, NMAR, CNNMAR, DuDoNet (dual domain network), DSCMAR, DuDoNet++ (dual domain network++) provided by related technologies, and the two artifact removal methods InDuDoNet (interpretable dual domain network) and InDuDoNet+ (interpretable dual domain network+) proposed in the technical solution of this application.

[0238] Please refer to Figure 14Fig. 6 shows a generalization effect comparison chart of the above five de-artifact methods. In Figure 14 It can be seen in Fig. 6 that the generalization effect of InDuDoNet+ is better than that of the other several de-artifact methods.

[0239] As shown in Table 1, InDuDoNet+ requires the least amount of network parameters, which is more conducive to actual deployment and storage.

[0240] Table 1: Comparison table of network parameter amount of de-artifact methods

[0241]

[0242] The following is an apparatus embodiment of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the apparatus embodiments of the present application, please refer to the method embodiments of the present application.

[0243] Please refer to Figure 15 Fig. 9 shows a block diagram of an image de-artifact apparatus provided by an embodiment of the present application. The apparatus has the function of implementing the above image de-artifact method, which can be implemented by hardware or corresponding software executed by hardware. The apparatus 900 can include an image acquisition module 910 and a de-artifact module 920.

[0244] The image acquisition module 910 is configured to acquire an artifact-containing medical image and a contaminated string diagram, the artifact-containing medical image and the contaminated string diagram being a set of corresponding images for the same object.

[0245] The de-artifact module 920 is configured to perform N-stage iterative processing based on the artifact-containing medical image and the contaminated string diagram, to generate a final de-artifact medical image corresponding to the artifact-containing medical image; wherein the iterative processing of each stage is configured to perform repair processing on the string diagram and de-artifact processing on the medical image, and N is an integer greater than 1.

[0246] In some embodiments, as shown in Figure 16 The de-artifact module 920 includes a preprocessing unit 921, a string diagram updating unit 922, a first proximal unit 923, a medical image updating unit 924, and a second proximal unit 925.

[0247] The preprocessing unit 921 is configured to generate an initialized clean medical image, an initialized clean normalized string diagram, and a normalization factor based on the artifact-containing medical image and the contaminated string diagram, the contour feature of the normalization factor being superior to that of the contaminated string diagram.

[0248] In some embodiments, the preprocessing unit 921 is used to obtain a reconstructed chord diagram and a reconstructed medical image corresponding to the medical image with artifacts, where the reconstructed chord diagram refers to an image obtained by preliminary repair of the contaminated chord diagram, and the reconstructed medical image refers to an image obtained by preliminary de-artifacting the medical image with artifacts; a first proximal operation is performed on the reconstructed chord diagram to obtain the initialized clean normalized chord diagram; and a second proximal operation is performed on the reconstructed medical image to obtain the initialized clean medical image.

[0249] In some embodiments, the preprocessing unit 921 is configured to generate a first prior medical image based on the artifact-bearing medical image, where the first prior medical image refers to a medical image having common features of a clean medical image; fuse the first prior medical image with a reconstructed medical image corresponding to the artifact-bearing medical image to obtain a fused image, where the reconstructed medical image refers to an image obtained by preliminary de-artifacting the artifact-bearing medical image; and forward project the fused image to obtain the normalization factor. In some embodiments, the preprocessing unit 921 is configured to concatenate the artifact-bearing medical image and the reconstructed medical image, input the concatenated concatenation into a priori network, and output the first prior medical image through the priori network.

[0250] In some embodiments, the preprocessing unit 921 is used to generate a second prior medical image based on the artifact-bearing medical image and the reconstructed medical image corresponding to the artifact-bearing medical image; wherein the reconstructed medical image refers to an image obtained by preliminary de-artifacting the artifact-bearing medical image, and the second prior medical image refers to a medical image having common features of a clean medical image and unique features of the artifact-bearing medical image; the second prior medical image is forward projected to obtain the normalization factor.

[0251] In some embodiments, the preprocessing unit 921 is used to generate an initial prior medical image based on the reconstructed medical image, where the initial prior medical image refers to a medical image having common features of a clean medical image; and weightedly fuse the initial prior medical image and the medical image with artifacts to obtain the second prior medical image.

[0252] In some embodiments, the preprocessing unit 921 is used to process the artifact-bearing medical image through a weighted network to obtain a weight matrix; and perform a dot product calculation on the initial prior medical image and the weight matrix to obtain the second prior medical image.

[0253] The chord diagram updating unit 922 is configured to obtain, for the i-th stage of the N stages, an updated normalized chord diagram according to the normalization factor, the clean normalized chord diagram obtained at the i-1th iteration, the clean medical image obtained at the i-1th iteration, and the contaminated chord diagram.

[0254] In some embodiments, the chord diagram updating unit 922 is used to obtain a first intermediate result based on the normalization factor, the clean normalized chord diagram obtained at the i-1th iteration, the clean medical image obtained at the i-1th iteration, and the contaminated chord diagram; and obtain the updated normalized chord diagram based on the clean normalized chord diagram obtained at the i-1th iteration and the first intermediate result.

[0255] The first proximal unit 923 is configured to perform a first proximal operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained in the i-th iteration, wherein the first proximal operation is used to repair the chord graph through a first neural network.

[0256] In some embodiments, the first proximal unit 923 is used to cascade the updated normalized chord diagram and the first auxiliary variable obtained in the i-1th iteration to obtain an updated normalized chord diagram after the i-th cascade; perform a first proximal operation on the updated normalized chord diagram after the i-th cascade to obtain a clean normalized chord diagram after the i-th cascade; perform channel splitting on the clean normalized chord diagram after the i-th cascade, and select the image of the first channel as the clean normalized chord diagram obtained in the i-th iteration.

[0257] The medical image updating unit 924 is configured to obtain an updated medical image according to the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the i-th iteration.

[0258] In some embodiments, the medical image updating unit 924 is used to obtain a second intermediate result based on the normalization factor, the clean medical image obtained at the i-1th iteration, and the clean normalized chord diagram obtained at the i-th iteration; and obtain the updated medical image based on the clean medical image obtained at the i-1th iteration and the second intermediate result.

[0259] The second proximal unit 925 is used to perform a second proximal operation on the updated medical image to obtain a clean medical image obtained in the i-th iteration, wherein the second proximal operation is used to remove artifacts from the medical image through a second neural network.

[0260] In some embodiments, the second proximal unit 925 is used to cascade the updated medical image and the second auxiliary variable obtained in the i-1th iteration to obtain an i-th cascaded medical image; perform a second proximal operation on the i-th cascaded medical image to obtain an i-th cascaded clean medical image; perform channel splitting on the i-th cascaded clean medical image, and select the image of the second channel as the clean medical image obtained in the i-th iteration.

[0261] In some embodiments, i is a positive integer less than or equal to N. When i=1, the clean normalized chord diagram obtained in the i-1th iteration is the initialized clean normalized chord diagram, and the clean medical image obtained in the i-1th iteration is the initialized clean medical image; when i=N, the clean medical image obtained in the i-th iteration is used as the final de-artifacted medical image corresponding to the medical image with artifacts.

[0262] Please refer to Figure 17 , which shows a block diagram of a device for training a de-artifact model according to one embodiment of the present application. This device has the function of implementing the aforementioned de-artifact model training method. This function can be implemented by hardware or by hardware executing corresponding software. The device 1100 may include: a sample acquisition module 1110, a sample processing module 1120, a de-artifact module 1130, a loss calculation module 1110, and a model training module 1150.

[0263] The sample acquisition module 1110 is configured to acquire sample data, where the sample data includes clean medical image samples and clean chord diagram samples. The clean medical image samples and the clean chord diagram samples are a group of corresponding images of the same object.

[0264] The sample processing module 1120 is configured to generate a contaminated chord diagram sample and a medical image sample with artifacts based on the clean medical image sample.

[0265] In some embodiments, the sample processing module 1120 is used to obtain artifact simulation data; generate a contaminated projection image based on the artifact simulation data and the clean medical image sample; add noise to the contaminated projection image to generate the contaminated chord diagram sample; and generate the artifact-bearing medical image sample based on the contaminated chord diagram sample.

[0266] The artifact removal module 1130 is configured to employ the artifact removal model to perform N stages of iterative processing based on the artifact-bearing medical image samples and the contaminated chord diagram samples to generate a final artifact-free medical image corresponding to the artifact-bearing medical image samples; wherein each stage of iterative processing is used to repair the chord diagram and perform artifact removal on the medical image, and N is an integer greater than 1.

[0267] In some embodiments, the artifact removal model includes N cascaded artifact removal network modules, each of which includes: a first neural network for repairing the chord graph, and a second neural network for performing artifact removal on the medical image.

[0268] In some embodiments, as Figure 18 As shown, the artifact removal module 1130 includes a pre-processing unit 1131 , a chord diagram updating unit 1132 , a first proximal unit 1133 , a medical image updating unit 1134 and a second proximal unit 1135 .

[0269] The preprocessing unit 1131 is configured to generate an initialized clean medical image, an initialized clean normalized chord diagram, and a normalization factor based on the artifact-bearing medical image and the contaminated chord diagram, wherein the contour feature of the normalization factor is superior to the contour feature of the contaminated chord diagram.

[0270] In some embodiments, the preprocessing unit 1131 is used to obtain a reconstructed chord diagram and a reconstructed medical image corresponding to the medical image with artifacts, where the reconstructed chord diagram refers to an image obtained by preliminary repair of the contaminated chord diagram, and the reconstructed medical image refers to an image obtained by preliminary de-artifacting the medical image with artifacts; a first proximal operation is performed on the reconstructed chord diagram to obtain the initialized clean normalized chord diagram; and a second proximal operation is performed on the reconstructed medical image to obtain the initialized clean medical image.

[0271] In some embodiments, the preprocessing unit 1131 is used to generate a prior medical image based on the medical image with artifacts, where the prior medical image refers to a medical image having common features of a clean medical image; the prior medical image and the reconstructed medical image corresponding to the medical image with artifacts are fused to obtain a fused image, where the reconstructed medical image refers to an image obtained by preliminary de-artifacting the medical image with artifacts; and the fused image is forward projected to obtain the normalization factor.

[0272] In some embodiments, the preprocessing unit 1131 is used to cascade the artifact-bearing medical image and the reconstructed medical image, input the concatenated images into a priori network, and output the priori medical image through the priori network.

[0273] The chord diagram updating unit 1132 is configured to obtain, for the i-th stage of the N stages, an updated normalized chord diagram according to the normalization factor, the clean normalized chord diagram obtained at the i-1th iteration, the clean medical image obtained at the i-1th iteration, and the contaminated chord diagram.

[0274] In some embodiments, the chord diagram updating unit 1132 is used to obtain a first intermediate result based on the normalization factor, the clean normalized chord diagram obtained at the i-1th iteration, the clean medical image obtained at the i-1th iteration, and the contaminated chord diagram; and obtain the updated normalized chord diagram based on the clean normalized chord diagram obtained at the i-1th iteration and the first intermediate result.

[0275] The first proximal unit 1133 is configured to perform a first proximal operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained in the i-th iteration, wherein the first proximal operation is used to repair the chord graph through a first neural network.

[0276] In some embodiments, the first proximal unit 1133 is used to cascade the updated normalized chord diagram and the first auxiliary variable obtained in the i-1th iteration to obtain the updated normalized chord diagram after the i-th cascade; perform a first proximal operation on the updated normalized chord diagram after the i-th cascade to obtain a clean normalized chord diagram after the i-th cascade; perform channel splitting on the clean normalized chord diagram after the i-th cascade, and select the image of the first channel as the clean normalized chord diagram obtained in the i-th iteration.

[0277] The medical image updating unit 1134 is configured to obtain an updated medical image according to the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the (i)th iteration.

[0278] In some embodiments, the medical image updating unit 1134 is used to obtain a second intermediate result based on the normalization factor, the clean medical image obtained at the i-1th iteration, and the clean normalized chord diagram obtained at the i-th iteration; and obtain the updated medical image based on the clean medical image obtained at the i-1th iteration and the second intermediate result.

[0279] The second proximal unit 1135 is used to perform a second proximal operation on the updated medical image to obtain a clean medical image obtained in the i-th iteration, wherein the second proximal operation is used to remove artifacts from the medical image through a second neural network.

[0280] In some embodiments, the second proximal unit 1135 is used to cascade the updated medical image and the second auxiliary variable obtained in the i-1th iteration to obtain the i-th cascaded medical image; perform a second proximal operation on the i-th cascaded medical image to obtain the i-th cascaded clean medical image; perform channel splitting on the i-th cascaded clean medical image, and select the image of the second channel as the clean medical image obtained in the i-th iteration.

[0281] In some embodiments, i is a positive integer less than or equal to N. When i=1, the clean normalized chord diagram obtained in the i-1th iteration is the initialized clean normalized chord diagram, and the clean medical image obtained in the i-1th iteration is the initialized clean medical image; when i=N, the clean medical image obtained in the i-th iteration is used as the final de-artifacted medical image corresponding to the medical image with artifacts.

[0282] The loss calculation module 1140 is used to calculate the training loss of the de-artifacting model based on the clean medical image sample and the final de-artifacted medical image.

[0283] The model training module 1150 is used to train the artifact removal model based on the training loss.

[0284] It should be noted that the apparatus provided in the above embodiments, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0285] Please refer to Figure 18 , which shows a schematic diagram of the structure of a computer device provided in one embodiment of the present application. The computer device can be any electronic device with data calculation, processing and storage functions. The computer device can be used to implement the image artifact removal method provided in the above embodiment, and can also be used to implement the artifact removal model training method provided in the above embodiment. Specifically:

[0286] The computer device 1300 includes a central processing unit (CPU, central processing unit), GPU (graphics processing unit), and FPGA (field programmable gate array) 1301, a system memory 1304 including RAM (random-access memory) 1302 and 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 transmission between various components within the server, and a mass storage device 1307 for storing an operating system 1313, application programs 1314, and other program modules 1315.

[0287] 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 or keyboard, for user input. Both the display 1308 and the input device 1309 are connected to the central processing unit 1301 via 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 variety of other devices, such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller 1310 also provides output to a display screen, printer, or other types of output devices.

[0288] The mass storage device 1307 is connected to the central processing unit 1301 via 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. In other words, the mass storage device 1307 may include a computer-readable medium (not shown) such as a hard disk or a CD-ROM (Compact Disc Read-Only Memory) drive.

[0289] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other solid-state storage technology, CD-ROM, DVD (Digital Video Disc) or other optical storage, tape cassettes, magnetic tape, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that the computer storage media is not limited to the above-mentioned ones. The above-mentioned system memory 1304 and mass storage device 1307 can be collectively referred to as memory.

[0290] According to an embodiment of the present application, the computer device 1300 can also be connected to a remote computer on a network such as the Internet for operation. That is, the computer device 1300 can be connected to the network 1312 via the network interface unit 1311 connected to the system bus 1305. Alternatively, the network interface unit 1311 can be used to connect to other types of networks or remote computer systems (not shown).

[0291] The memory also includes at least one instruction, at least one program, code set or instruction set, which is stored in the memory and configured to be executed by one or more processors to implement the above-mentioned image artifact removal method or artifact removal model training method.

[0292] In an exemplary embodiment, a computer-readable storage medium is also provided, in which at least one instruction, at least one program, a code set or an instruction set is stored. When the at least one instruction, the at least one program, the code set or the instruction set is executed by a processor of a computer device, the at least one instruction, the at least one program, the code set or the instruction set implements the above-mentioned image artifact removal method or the artifact removal model training method.

[0293] Optionally, the computer-readable storage medium may include: ROM (Read-Only Memory), RAM (Random-Access Memory), SSD (Solid State Drives), or an optical disk, etc. Among them, the random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0294] In an exemplary embodiment, a computer program product or computer program is also provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the aforementioned image artifact removal method or the aforementioned artifact removal model training method.

[0295] It should be understood that the term "plurality" used herein refers to two or more. "And / or" describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. The character " / " generally indicates an "or" relationship between the associated objects.

[0296] The above description is merely an exemplary embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for removing image artifacts, characterized in that: The method comprises: Acquire a medical image with artifacts and a contaminated chord map, wherein the medical image with artifacts and the contaminated chord map are a set of corresponding images of the same object; generating a normalization factor based on 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 de-artifacting on the artifact-bearing medical image, and wherein a contour feature of the normalization factor is superior to a contour feature of the contaminated chord diagram; Based on the artifact-bearing medical image, the contaminated chord diagram, and the normalization factor, N stages of iterative processing are performed to generate a final de-artifacted medical image corresponding to the artifact-bearing medical image; the N stages of iterative processing are used to alternately process the chord diagram and the medical image, wherein the iterative processing of each stage sequentially repairs the chord diagram and de-artifacts the medical image, and N is an integer greater than 1.

2. The method according to claim 1, characterized in that The step of performing N stages of iterative processing based on the medical image with artifacts, the contaminated chord diagram, and the normalization factor to generate a final artifact-free medical image corresponding to the medical image with artifacts includes: generating an initialized clean medical image and an initialized clean normalized chord graph based on the artifact-bearing medical image and the contaminated chord graph; For an i-th stage in the N stages, obtaining an updated normalized chord diagram according to the normalization factor, the clean normalized chord diagram obtained at the i-1-th iteration, the clean medical image obtained at the i-1-th iteration, and the contaminated chord diagram; Performing a first proximal operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained in the i-th iteration, wherein the first proximal operation is used to repair the chord graph through a first neural network; Obtaining an updated medical image according to the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the (i)th iteration; performing a second proximal operation on the updated medical image to obtain a clean medical image obtained in an i-th iteration, wherein the second proximal operation is used to perform artifact removal processing on the medical image through a second neural network; Wherein, i is a positive integer less than or equal to N. When i=1, the clean normalized chord diagram obtained in the i-1th iteration is the initialized clean normalized chord diagram, and the clean medical image obtained in the i-1th iteration is the initialized clean medical image; when i=N, the clean medical image obtained in the i-th iteration is used as the final de-artifacted medical image corresponding to the medical image with artifacts.

3. The method according to claim 2, characterized in that Obtaining an updated normalized chord graph according to the normalization factor, the clean normalized chord graph obtained at the (i-1)th iteration, the clean medical image obtained at the (i-1)th iteration, and the contaminated chord graph, comprising: Obtaining a first intermediate result according to the normalization factor, the clean normalized chord graph obtained at the (i-1)th iteration, the clean medical image obtained at the (i-1)th iteration, and the contaminated chord graph; The updated normalized chord diagram is obtained according to the clean normalized chord diagram obtained in the (i-1)th iteration and the first intermediate result.

4. The method according to claim 2, characterized in that Obtaining an updated medical image according to the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the i-th iteration includes: Obtaining a second intermediate result according to the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the i-th iteration; The updated medical image is obtained according to the clean medical image obtained in the (i-1)th iteration and the second intermediate result.

5. The method according to claim 2, characterized in that The performing a first proximal operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained in the i-th iteration includes: cascading the updated normalized chord diagram and the first auxiliary variable obtained in the (i-1)th iteration to obtain an updated normalized chord diagram after the i-th cascade, wherein the first auxiliary variable obtained in the (i-1)th iteration is a normalized chord diagram among the multiple normalized chord diagrams obtained in the (i-1)th iteration, excluding the clean normalized chord diagram obtained in the (i-1)th iteration; Performing a first proximal operation on the updated normalized chord graph after the i-th cascade to obtain a clean normalized chord graph after the i-th cascade; Channel splitting is performed on the clean normalized chord graph after the i-th cascade, and an image of the first channel is selected as the clean normalized chord graph obtained by the i-th iteration.

6. The method according to claim 2, characterized in that The performing a second proximal operation on the updated medical image to obtain a clean medical image obtained in the i-th iteration includes: cascading the updated medical image and the second auxiliary variable obtained in the (i-1)th iteration to obtain an i-th cascaded medical image, wherein the second auxiliary variable obtained in the (i-1)th iteration is a medical image other than the clean medical image obtained in the (i-1)th iteration among the multiple medical images obtained in the (i-1)th iteration; performing a second proximal operation on the i-th cascaded medical image to obtain an i-th cascaded clean medical image; Channel splitting is performed on the clean medical image after the i-th cascade, and the image of the second channel is selected as the clean medical image obtained in the i-th iteration.

7. The method according to claim 2, characterized in that The step of generating an initialized clean medical image and an initialized clean normalized chord graph based on the artifact-bearing medical image and the contaminated chord graph comprises: Obtaining a reconstructed chord diagram and a reconstructed medical image corresponding to the medical image with artifacts, wherein the reconstructed chord diagram refers to an image obtained by performing preliminary repair on the contaminated chord diagram, and the reconstructed medical image refers to an image obtained by performing preliminary de-artifacting on the medical image with artifacts; Performing a first proximal operation on the reconstructed chord graph to obtain the initialized clean normalized chord graph; A second proximal operation is performed on the reconstructed medical image to obtain the initialized clean medical image.

8. The method according to claim 1, characterized in that Generating a normalization factor according to the medical image with artifacts and the reconstructed medical image corresponding to the medical image with artifacts includes: generating a first priori medical image based on the medical image with artifacts, wherein the first priori medical image refers to a medical image having common features of a clean medical image; fusing the first priori medical image and a reconstructed medical image corresponding to the medical image with artifacts to obtain a fused image, wherein the reconstructed medical image is an image obtained by performing preliminary de-artifacting on the medical image with artifacts; Perform forward projection on the fused image to obtain the normalization factor.

9. The method according to claim 8, characterized in that Generating a first priori medical image according to the medical image with artifacts includes: The artifact-bearing medical image and the reconstructed medical image are cascaded and input into a priori network, and the first priori medical image is output through the priori network.

10. The method according to claim 1, characterized in that Generating a normalization factor according to the medical image with artifacts and the reconstructed medical image corresponding to the medical image with artifacts includes: generating a second priori medical image based on 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 de-artifacting on the artifact-bearing medical image, and the second priori medical image is a medical image having common features of clean medical images and unique features of the artifact-bearing medical image; Perform forward projection on the second priori medical image to obtain the normalization factor.

11. The method according to claim 10, characterized in that Generating the second priori medical image according to the medical image with artifacts and the reconstructed medical image corresponding to the medical image with artifacts includes: generating an initial priori medical image based on the reconstructed medical image, wherein the initial priori medical image refers to a medical image having common features of a clean medical image; The initial priori medical image and the medical image with artifacts are weightedly fused to obtain the second priori medical image.

12. The method according to claim 11, characterized in that The step of performing weighted fusion on the initial priori medical image and the medical image with artifacts to obtain the second priori medical image includes: Processing the medical image with artifacts through a weighted network to obtain a weight matrix; Perform a dot product calculation on the initial priori medical image and the weight matrix to obtain the second priori medical image.

13. A method for training an artifact removal model, characterized in that: The method comprises: Acquire sample data, where the sample data includes clean medical image samples and clean chord diagram samples, where the clean medical image samples and the clean chord diagram samples are a group of corresponding images of the same object; Based on the medical image sample, generating a contaminated chord diagram sample and a medical image sample with artifacts; generating a normalization factor based on the artifact-bearing medical image sample and a reconstructed medical image corresponding to the artifact-bearing medical image sample, wherein the reconstructed medical image is an image obtained by performing preliminary artifact removal on the artifact-bearing medical image, and a contour feature of the normalization factor is superior to a contour feature of the contaminated chord diagram; Using an artifact removal model, based on the artifact-bearing medical image sample, the contaminated chord diagram sample, and the normalization factor, N stages of iterative processing are performed to generate a final artifact-free medical image corresponding to the artifact-bearing medical image sample; the N stages of iterative processing are used to alternately process the chord diagram and the medical image, wherein each stage of iterative processing sequentially performs repair processing on the chord diagram and performs artifact removal processing on the medical image, and N is an integer greater than 1; Calculating a training loss of the artifact removal model based on the clean medical image sample and the final artifact-removed medical image; The artifact removal model is trained based on the training loss.

14. The method according to claim 13, characterized in that The artifact removal model includes N cascaded artifact removal network modules, each of which includes: a first neural network for repairing the chord diagram, and a second neural network for performing artifact removal on the medical image; The method further comprises: performing N stages of iterative processing based on the artifact-bearing medical image sample, the contaminated chord diagram sample, and the normalization factor using the artifact removal model to generate a final artifact-free medical image corresponding to the artifact-bearing medical image sample, including: generating an initialized clean medical image and an initialized clean normalized chord graph based on the artifact-bearing medical image sample and the contaminated chord graph sample; For the i-th stage in the N stages, obtaining an updated normalized chord diagram according to the normalization factor, the clean normalized chord diagram obtained at the i-1-th iteration, the clean medical image obtained at the i-1-th iteration, and the contaminated chord diagram sample; Using a first neural network included in an i-th artifact removal network module, perform a first proximal operation on the updated normalized chord graph to obtain a clean normalized chord graph obtained in an i-th iteration; Obtaining an updated medical image according to the normalization factor, the clean medical image obtained at the (i-1)th iteration, and the clean normalized chord diagram obtained at the (i)th iteration; Performing a second proximal operation on the updated medical image using a second neural network included in the i-th artifact removal network module to obtain a clean medical image obtained in an i-th iteration; Wherein, i is a positive integer less than or equal to N. When i=1, the clean normalized chord diagram obtained in the i-1th iteration is the initialized clean normalized chord diagram, and the clean medical image obtained in the i-1th iteration is the initialized clean medical image; when i=N, the clean medical image obtained in the i-th iteration is used as the final de-artifacted medical image corresponding to the artifact-bearing medical image sample.

15. The method according to claim 13, characterized in that The generating of the contaminated chord graph sample and the medical image sample with artifacts based on the clean medical image sample includes: Acquiring artifact simulation data; generating a contaminated projection image according to the artifact simulation data and the clean medical image sample; adding noise to the contaminated projection image to generate the contaminated chord graph sample; The medical image sample with artifacts is generated according to the contaminated chord graph sample.

16. An image artifact removal device, characterized in that: The device comprises: An image acquisition module, configured to acquire a medical image with artifacts and a contaminated chord diagram, wherein the medical image with artifacts and the contaminated chord diagram are a set of corresponding images of the same object; an artifact removal module, configured to generate a normalization factor based on 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 artifact-bearing medical image, and wherein a contour feature of the normalization factor is superior to a contour feature of the contaminated chord diagram; A de-artifacting module is configured to perform N stages of iterative processing based on the artifact-bearing medical image, the contaminated chord diagram, and the normalization factor to generate a final de-artifacted medical image corresponding to the artifact-bearing medical image; the N stages of iterative processing are configured to alternately process the chord diagram and the medical image, wherein each stage of iterative processing sequentially repairs the chord diagram and de-artifacts the medical image, and N is an integer greater than 1.

17. A training device for an artifact removal model, characterized in that: The device comprises: a sample acquisition module, configured to acquire sample data, wherein the sample data includes clean medical image samples and clean chord diagram samples, wherein the clean medical image samples and the clean chord diagram samples are a group of corresponding images of the same object; a sample processing module, configured to generate a contaminated chord diagram sample and a medical image sample with artifacts based on the clean chord diagram sample; an artifact removal module, configured to generate a normalization factor based on 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 artifact-bearing medical image, and wherein a contour feature of the normalization factor is superior to a contour feature of the contaminated chord diagram; an artifact removal module, configured to employ an artifact removal model and, based on the artifact-bearing medical image sample, the contaminated chord diagram sample, and the normalization factor, perform N stages of iterative processing to generate a final artifact-free medical image corresponding to the artifact-bearing medical image sample; the N stages of iterative processing are configured to alternately process the chord diagram and the medical image, wherein each stage of iterative processing sequentially performs repair processing on the chord diagram and artifact removal processing on the medical image, where N is an integer greater than 1; a loss calculation module, configured to calculate a training loss of the artifact removal model based on the clean medical image sample and the final artifact-removed medical image; A model training module is used to train the artifact removal model based on the training loss.

18. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the image artifact removal method as described in any one of claims 1 to 12, or to implement the artifact removal model training method as described in any one of claims 13 to 15.

19. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the image artifact removal method as described in any one of claims 1 to 12, or to implement the artifact removal model training method as described in any one of claims 13 to 15.

20. A computer program product, characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. The processor reads and executes the computer instructions from the computer-readable storage medium to implement the image artifact removal method as described in any one of claims 1 to 12, or the artifact removal model training method as described in any one of claims 13 to 15.

Images