Image processing system, method, storage medium and computer program product

Through image segmentation and sinusoidal image processing technology, artifacts in CT images are identified and corrected to generate high-quality target CT images, solving the problems of image quality reduction and low efficiency caused by artifact correction in the prior art.

CN118710565BActive Publication Date: 2024-12-06YOFO MEDICAL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411197611.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-12-06
Estimated Expiration
2044-08-29

AI Technical Summary

Technical Problem

The existing CT image processing technology can easily lead to blurring of non-artifact areas during artifact correction, which is difficult to meet image quality requirements, and the iterative reconstruction technology is inefficient, and the image risks generated by neural network correction methods are uncontrollable.

Method used

The masked image is obtained through image segmentation, a first target sine map and a second target sine map are generated, and the pixel value of the target area is set, and the image correction model is recognized and corrected, and a target CT image is generated to reduce artifacts.

Benefits of technology

Effectively reduce artifacts, improve image quality, ensure the clarity of non-mask areas, avoid newly formed image defects, and improve artifact correction efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118710565B_ABST
    Figure CN118710565B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of medical imaging technology, and specifically provides an image processing system, method, readable storage medium and computer program product, wherein the system obtains a mask image in a CT image of a subject by image segmentation, generates a first target sinogram of the mask image and a second target sinogram of the CT image of the subject, sets the pixel value of the coordinate point in the target area of ​​the second target sinogram to 1, corrects the pixel value of the coordinate point in the target area of ​​the second target sinogram set by the pixel value of the coordinate point through a trained image correction model to obtain h third target sinograms, and generates a target CT image based on the h third target sinograms. According to the present invention, artifacts in CT images can be eliminated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to an image processing system, an image processing method, a readable storage medium and a computer program product. Background Art

[0002] Artifacts are some defects that may exist in CT images, such as metal artifacts or calcification artifacts caused by the scanned object containing metal or calcification during the scanning process. Artifacts will reduce the image quality of the reconstructed CT image, making it difficult for the CT image to correctly display some parts of the scanned object.

[0003] In the related art, when correcting an artifact, it will have a certain impact on other images near the artifact, causing some non-artifact areas in the image obtained after the artifact correction to be blurred, resulting in image quality that is difficult to meet requirements. Summary of the invention

[0004] In order to solve at least one of the above technical problems, the present invention provides an image processing system, method, storage medium and computer program product.

[0005] A first aspect of the present invention provides an image processing system, comprising: a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following process: obtaining a mask image in a CT image of a subject by image segmentation, wherein the mask image is used to represent part or all of a first object in the subject corresponding to the CT image of the subject that can cause artifacts; generating a first target sinusoidal graph of the mask image and a second target sinusoidal graph of the CT image of the subject, wherein the number of the first target sinusoidal graph and the number of the second target sinusoidal graph are both h, where h>1, and the number of the h first target sinusoidal graphs and the h second target sinusoidal graphs is one according to the scanning position. One to one correspondence; setting the pixel values ​​of the coordinate points of the target area in the second target sinusoidal image to the target value, the area range of the target area is the same as the image area range corresponding to the first object in the first target sinusoidal image, and the target value enables the target area to be recognized by the trained image correction model; correcting the pixel values ​​of the coordinate points of the target area in the second target sinusoidal image by the image correction model to obtain h third target sinusoidal images; and generating a target CT image based on the h third target sinusoidal images, the pixel value correction makes the non-mask image area in the target CT image contain fewer artifacts than the non-mask image area of ​​the subject CT image.

[0006] The second aspect of the present invention proposes an image processing method, comprising: obtaining a mask image in a CT image of a subject by image segmentation, wherein the mask image is used to represent part or all of a first object in the subject corresponding to the CT image of the subject that can cause artifacts; generating a first target sinusoidal graph of the mask image and a second target sinusoidal graph of the CT image of the subject, wherein the number of the first target sinusoidal graph and the number of the second target sinusoidal graph are both h, h>1, and the h first target sinusoidal graphs and the h second target sinusoidal graphs correspond one-to-one according to scanning positions; and converting pixels of coordinate points in a target area in the second target sinusoidal graph into pixels of coordinate points. The value is set to a target value, the area range of the target area is the same as the image area range corresponding to the first object in the first target sinusoidal graph, and the target value enables the target area to be recognized by the trained image correction model; the pixel value correction of the coordinate points of the target area in the second target sinusoidal graph is performed by the image correction model to obtain h third target sinusoidal graphs; and a target CT image is generated based on the h third target sinusoidal graphs, and the pixel value correction makes the non-mask image area in the target CT image contain less artifacts than the non-mask image area of ​​the subject CT image.

[0007] A third aspect of the present invention provides a readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the method described in any of the above embodiments.

[0008] A fourth aspect of the present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it is used to implement at least the method described in any of the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The accompanying drawings illustrate exemplary embodiments of the present invention and together with the description serve to explain the principles of the present invention. These drawings are included to provide a further understanding of the present invention and are incorporated in and constitute a part of this specification.

[0010] Figure 1 A schematic diagram showing application scenarios of image processing systems according to some embodiments of the present invention is shown.

[0011] Figure 2 A flowchart of using an image processing device to perform artifact correction according to some embodiments of the present invention is shown.

[0012] Figure 3 A flowchart of a data preparation process according to some embodiments of the present invention is shown.

[0013] Figure 4A flowchart of the CNN model training process of some embodiments of the present invention is shown.

[0014] Figure 5 A flowchart of applying a model to perform artifact correction according to some embodiments of the present invention is shown.

[0015] Figure 6 A schematic diagram showing a first target sinusoidal diagram according to some embodiments of the present invention is shown.

[0016] Figure 7 shows the generated Figure 6 Schematic diagram of the corresponding second target sinusoidal diagram.

[0017] Figure 8 Shows Figure 7 Schematic diagram after pixel setting.

[0018] Fig. 9 A schematic diagram of a third target sinusoidal graph obtained after image restoration using an image correction model in some embodiments of the present invention is shown.

[0019] Fig.10 A third target projection image showing image discontinuity after image correction according to some embodiments of the present invention is shown.

[0020] Fig.11 A third target projection image after image continuity improvement according to some embodiments of the present invention is shown.

[0021] Fig.12 A schematic diagram of an image processing system using a hardware implementation of a processing system according to some embodiments of the present invention is shown.

[0022] Fig.13 A schematic flow chart of an image processing method according to some embodiments of the present invention is shown. DETAILED DESCRIPTION

[0023] The present invention will be further described in detail below in conjunction with the accompanying drawings and implementations. It is to be understood that the specific implementations described herein are only used to explain the relevant content, rather than to limit the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings.

[0024] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0025] Unless otherwise specified, the exemplary embodiments / embodiments shown will be understood as exemplary features that provide various details of some ways in which the technical concept of the present invention can be implemented in practice. Therefore, unless otherwise specified, the features of the various embodiments / embodiments can be combined, separated, interchanged and / or rearranged without departing from the technical concept of the present invention.

[0026] The terms used herein are for the purpose of describing specific embodiments, rather than being restrictive. As used herein, unless the context clearly indicates otherwise, the singular forms "one (kind, person)" and "said (the)" are also intended to include plural forms. In addition, when the terms "comprise" and / or "include" and their variations are used in this specification, it is explained that there are stated features, integral bodies, steps, operations, parts, assemblies and / or their groups, but it is not excluded that there are or add one or more other features, integral bodies, steps, operations, parts, assemblies and / or their groups. It should also be noted that, as used herein, the terms "substantially", "approximately" and other similar terms are used as approximate terms and not as degree terms, so that they are used to explain the inherent deviations of the measured values, calculated values ​​and / or provided values ​​that will be recognized by those of ordinary skill in the art.

[0027] When performing CBCT (cone beam CT), artifacts may exist in the captured CBCT images. For example, when performing oral CBCT, if there are metal objects such as metal teeth inside the mouth, metal artifacts will exist in the obtained oral CBCT images. For example, when performing breast CBCT, since the source tube voltage of breast CBCT is usually small (about 40kV), the X-ray penetration ability of breast CBCT is weaker than that of oral CBCT. If there are calcification points inside the breast, the calcification points in the breast may produce artifacts similar to metal artifacts in the obtained breast CBCT images.

[0028] For CBCT images with artifacts, artifact correction can be performed based on iterative reconstruction technology, but this method requires the scanned object to be completely within the field of view, and the iterative reconstruction technology requires multiple reconstructions, which makes the amount of calculation very large, resulting in low efficiency of artifact correction. Artifact correction can also be performed by pre-training the neural network and directly inputting the CBCT image into the neural network, but the risk of images generated in this way is uncontrollable. The newly generated image content may be indistinguishable from real artifact content, making the original non-artifact area blurred. While correcting the original artifact, new unclear areas will be generated, forming new image defects.

[0029] Figure 1A schematic diagram of an application scenario of an image processing system of some embodiments of the present invention is shown. In this application scenario, an image processing device 10 and a display 20 for displaying an image may be included. The image processing device 10 and the display 20 may communicate and transmit data in a wired or wireless manner. The image processing device 10 may be used to process projection images and perform artifact repair on CT images, and the display 20 may be used to display the repaired CT image. The CT image may be a CBCT image. In the present invention, the image processing system may be different types of electronic devices, for example, the image processing system may be a tablet computer, a laptop computer, a desktop computer, a self-service machine or other device.

[0030] Figure 2 A flowchart of using an image processing device to perform artifact correction according to some embodiments of the present invention is shown. Figure 2 The detailed description is for a clearer understanding of the technical solution of the present invention and should not be considered as a limitation of the protection scope of the present invention. In the process of implementing the technical idea of ​​the present invention, one or some steps may be omitted, or other alternatives may be adopted.

[0031] See also Figure 2 The method of performing artifact correction and elimination by the image processing device 10 includes three parts: the first part is data preparation, the second part is model training, and the third part is model application.

[0032] Figure 3 FIG. 1 is a flowchart showing a data preparation process according to some embodiments of the present invention. Figure 3 ,In the data preparation stage, actual clinical data in clinical ,application scenarios can be acquired, and the acquired data can be ,processed accordingly to obtain training data.

[0033] The acquired data may include two types of data, wherein the first type of data includes pre-stored scan data (i.e., data of the first sample projection image), corresponding bright field image data, and geometric parameters obtained by scanning some subjects. The second type of data includes data of sample CT images of some subjects that have been reconstructed. Both the projection image in the first type of data and the sample CT image in the second type of data are images that do not contain artifacts.

[0034] The object to be inspected may be a tooth, a breast, or any other object that can be scanned. A bright field image is a projection image obtained when no object is present, which is equivalent to an image obtained by photographing the air. Only one bright field image may be taken. A bright field image is used to represent the intensity of X-rays when they do not pass through any object to be inspected. During the reconstruction process, the amount of X-ray absorption can be determined by using the bright field image and the projection image obtained by scanning the object, so that image reconstruction can be performed based on this. Geometric parameters are used to describe the configuration of the scanning device in the process of scanning the object to obtain projection image data. Geometric parameters may include: the distance between the source and the detector, the offset angle, the detection surface size of the detector, and so on. The projection image obtained by scanning the object and the corresponding bright field image have the same corresponding geometric parameters.

[0035] For the first type of data, assume that when scanning the subject, the width of the projection image collected by the detector is w0, the height is h0, and the number of images is k0, that is, the scan data is k0 w0*h0 first sample projection image data. First, the data of each image in the k0 w0*h0 first sample projection images is processed according to formula (1) to obtain k0 w0*h0 second sample projection image data. Formula (1) is: .in, is the value of the i-th position point in the first sample projection image, is the value of the i-th position point in the corresponding bright field image, is the value of the i-th position point in the second sample projection image. The image size of the second sample projection image is the same as that of the first sample projection image, and the number of images is also the same.

[0036] Then, the data of the k0 w0*h0 second sample projection images are rearranged to obtain h0 w0*k0 first sample sinogram (sinogram, sinograph for short) data. The rearrangement method is to arrange the same row of data of the k0 second sample projection images from top to bottom (in the image height direction) according to the scanning order when scanning to obtain each second sample projection image, to obtain h0 w0*k0 sinograms, and to sort these h0 sinograms according to the sequence number of the row when they are generated, to obtain sorted h0 w0*k0 sinograms. For example, the three second sample projection images p1, p2 and p3 with 4 columns and 5 rows (i.e., w0=4, h0=5) are rearranged. The coordinate data of the first row of the three second sample projection images can be first obtained and arranged from top to bottom in the scanning order of the three projection images (the order of p1, p2, p3) to obtain the first sinusoidal image p1'. Then, the coordinate data of the second row of the three second sample projection images can be obtained and arranged from top to bottom in the above scanning order to obtain the second sinusoidal image p2'. And so on. The 5 rows of coordinate data form a total of 5 sinusoidal images p1'~p5' with 4 columns and 3 rows. Compared with the second sample projection image, the number of columns of the sample sinusoidal image has not changed, the number of rows has become 3 rows, and the number of images has become 5.

[0037] For the second type of data, the sample CT image data is forward projected according to the corresponding geometric parameters to obtain simulated projection data. The simulated projection data corresponds to: the projection data that can obtain the sample CT image data by reconstruction is processed according to the above formula (1) to obtain the data of k1 w1*h1 projection images. In other words, if the first sample projection image is reconstructed to obtain the sample CT image P1, the simulated projection data obtained by forward projecting the sample CT image P1 is the data of the second sample projection image. After obtaining the simulated projection data, the simulated projection data is rearranged in the manner of rearranging the data of the second sample projection image to obtain the sinusoidal graph data. Assuming that the simulated projection data is the data of k1 w1*h1 projection images, after arrangement, the data of h1 w1*k1 second sample sinusoidal graphs are obtained. The first sample sinusoidal graph and the second sample sinusoidal graph belong to the same sample sinusoidal graph and together constitute the training data in the data preparation stage, thereby completing the data preparation process.

[0038] Figure 4 FIG. 1 is a flowchart showing the CNN model training process of some embodiments of the present invention. Figure 4 In the model training phase, the model is trained using the training data obtained in the data preparation phase to obtain a trained model that can be used for artifact correction and elimination.

[0039] Taking a convolutional neural network (CNN) model such as a UNet network model as an example, during the model training process, the position of the first two-dimensional region is first determined in part or all of the sample sinusoidal graphs contained in the training data, and the position of the first region in the sinusoidal graph of the training data can be randomly set. Then, the pixel values ​​of the position points in the first region of these sample sinusoidal graphs are set to a preset value (the preset value can be 1), thereby covering the image content in the first region and obtaining a first sample image. The shape of the first region can include preset regions such as polygons and circles and object regions of some pre-stored preset objects in the sinusoidal graph, and can also include regions of some other shapes.

[0040] The preset object may be an object that can cause artifacts. Taking the preset object as a metal object as an example, the shape of the first area includes the metal object area in the metal object sinusoidal map. Among them, the metal object area is roughly arched. The generation method of the metal object sinusoidal map can be implemented in the above-mentioned way of processing the second data, specifically: obtaining a metal object CT image obtained by scanning and reconstructing a known metal object in advance, and then identifying the metal object area (three-dimensional area) in the metal object CT image and binarizing it, that is, setting the value of each position point in the metal object area to 1, and setting the value of each position point in the non-metallic object area to 0, to obtain a metal object binary image. Then forward project the metal object binary image to obtain a metal object projection map. Then, the metal object projection map is rearranged as described above to obtain a metal object sinusoidal map. In the metal object sinusoidal map, the value of the metal object area (two-dimensional arched area) is 1, and the value of other areas is 0, and the two-dimensional arched metal object area is taken as the first area.

[0041] It can be understood that, for each sinogram in the training data, there may be only one first region in the sinogram, or there may be multiple first regions. Each first sample image forms a sample image set.

[0042] Input a first sample image in the sample image set into the UNet model to obtain the image output by the UNet model, compare the output image with the original projection image without the first region content coverage and calculate the loss, and adjust the UNet model parameters based on the calculated loss value until the training goal is reached (the loss converges to meet certain requirements or reaches a certain number of training times). Then input another first sample image in the sample image set into the UNet model and repeat the above process until all the first sample images in the sample image set are used to train the UNet model, thereby completing the training of the UNet model.

[0043] The model trained in the model training phase can also be a generative network model. Taking the generative adversarial network model (GAN model) as an example, the sample sinusoidal graph in the training data is used as the second sample image. During the model training process, a Gaussian noise image that obeys the normal distribution can be randomly generated. The noise image and the second sample image (the image content of the first area is not covered) are used to train the GAN model (including the generator and the discriminator), so that the GAN model can output a sinusoidal graph with an image effect that meets the requirements. Among them, the noise image is mainly used as the input of the generator in the GAN model, and the second sample image is mainly used as a comparison basis for the discriminator to identify. In addition, the generative network model can also be a diffusion model or other models.

[0044] Figure 5 FIG. 1 is a flowchart showing the process of applying the model to perform artifact correction in some embodiments of the present invention. Figure 5 In the model application stage, the data of the CT image to be corrected for artifacts is obtained as the data of the subject CT image, the metal area or calcification area in the subject CT image is segmented to obtain the image of the mask area, and the trained model is used to correct and eliminate the artifacts in the mask area of ​​the subject CT image, and the data of the target CT image after the artifacts are corrected or even completely eliminated is obtained. The model used in the model application stage can be a CNN model such as a UNet model, or a generative network model such as a GAN model or a diffusion model.

[0045] After scanning the target object and obtaining the scan data, the scan data is reconstructed to obtain the data of the CT image of the subject. If the target object contains a first object such as metal or calcification point, the CT image of the subject will contain artifacts due to the presence of the first object. Then, the metal object area or calcification point area in the CT image of the subject is identified and segmented by threshold segmentation to obtain a three-dimensional mask image (image of the mask area). The mask image is forward projected to obtain a first target projection map. Each first target projection map is binarized, and in the binarized projection map, the pixel values ​​of the metal area and the calcification point area are target values ​​(the target value is 1). The binarized projection maps are rearranged in the above-mentioned data rearrangement method to obtain a first target sinusoidal map. In addition, the scan data (i.e., projection map data) of the target object is processed according to the above-mentioned formula (1) to obtain a second target projection map. The second target projection map is rearranged in the above-mentioned data rearrangement method to obtain a second target sinusoidal map. The image size and number of images of the first target sinusoidal graph and the second target sinusoidal graph are the same, the image size is w*k, and the number of images is h.

[0046] Figure 6 A schematic diagram showing a first target sinusoidal diagram according to some embodiments of the present invention is shown. Figure 7 shows the generated Figure 6 Schematic diagram of the corresponding second target sinusoidal diagram. Figure 6 and Figure 7 , Figure 6 The white area in the figure is the representation of the metal object in the mouth alone in the sinusoidal diagram. Figure 6 One of the plurality of first target sinusoidal graphs formed for the first target projection graph. Figure 7 The sine diagram of the target object is shown in FIG. 1 , in which the area A1 formed by the white sine lines represents the metal object, and the other areas are the bone, soft tissue and other parts of the target object. Figure 6 and Figure 7 It can be seen that the shape of the white area in the first target sinusoidal graph is basically the same as the shape of the white arched area A1 in the second target sinusoidal graph, because both represent the metal object in the target object. The representation area of ​​the metal object in the target object in the sinusoidal graph is determined by the above-mentioned threshold segmentation, forward projection, binarization and data rearrangement operations, thereby correctly segmenting the area content belonging to the metal object from the first target sinusoidal graph. Among them, the area formed by the coordinate points whose coordinate values ​​in the first target sinusoidal graph are the target values ​​is called the target area, which corresponds to the representation area of ​​the above-mentioned metal object in the sinusoidal graph. In the area other than the target area in the first target sinusoidal graph, the pixel value of the coordinate is 0.

[0047] Figure 8 Shows Figure 7 Schematic diagram after pixel setting. Figure 8 , area A2 is the target area in the second target sinusoidal graph belonging to the corresponding first target sinusoidal graph, and the pixel value of the coordinates in area A2 is set. The pixel value can be set to 1, so that the area representing the metal object in the first target sinusoidal graph covers the corresponding area A2 in the corresponding second target sinusoidal graph, thereby Figure 8 It can be seen that area A1 is basically covered. Since the pixel values ​​in area A2 are set, it is equivalent to missing the data in area A2 in the second target sinusoidal graph, and the pixel values ​​(colors) of area A2 are clearly different from those of other sinusoidal lines so that they can be recognized by the image correction model.

[0048] Fig. 9 FIG. 1 is a schematic diagram showing a third target sinusoidal graph obtained after image restoration using an image correction model in some embodiments of the present invention. Fig. 9 , the covered second target sinusoidal graph is input into the trained network model to obtain the third target sinusoidal graph output by the network model. h second target sinusoidal graphs are input h times to obtain h third target sinusoidal graphs.

[0049] The data of the h third target sinusoidal images are reversely arranged to obtain the third target projection image. The reverse arrangement is the inverse process of the above rearrangement, specifically: the same row of data in each target sinusoidal image is arranged from top to bottom according to the order of the target sinusoidal image, and k data of the third target projection image with an image size of w*h are obtained. For example, for the 5 target sinusoidal graphs g1'~g5' with 4 columns and 3 rows (i.e., w=4, k=3) are inversely rearranged, the first row of data of g1'~g5' can be obtained first, and arranged from top to bottom in the order of g1', g2', g3', g4', g5' to obtain the first third target projection graph g1', and then the second row of data of g1'~g5' can be obtained, and arranged from top to bottom in the order of g1', g2', g3', g4', g5' to obtain the second third target projection graph g2', and so on. The 3 rows of coordinate data form 3 third target projection graphs g1~g3 with 4 columns and 5 rows. Compared with the third target sinusoidal graph, the number of columns of the third target projection graph has not changed, the number of rows has become 5, and the number of images has become 3.

[0050] The obtained third target projection image is reconstructed to obtain the data of the target CT image. Then, the mask image obtained by the threshold segmentation is used to cover the corresponding area in the target CT image, and the metal object area and the calcification area are restored. The original metal object area and the calcification area are used to cover the metal area and the calcification area in the target CT image, so that the texture generated by the network model in the metal area and the calcification area is covered and restored to the original metal area and the calcification area in the CT image, so that the target CT image not only reflects the effect of artifact correction and artifact elimination, but also can retain the original content and existence of the target object.

[0051] If a generative network model is used to perform artifact correction during the model application stage, the image correction process of the model can be optimized, thereby reducing or even eliminating the image discontinuity in the artifact correction area in the obtained third target projection image. Fig.10 The third target projection diagram of some embodiments of the present invention is shown after image correction with image discontinuity. Fig.10 , area A3 (the area pointed by the arrow) represents the transverse texture existing in the first object area. The continuity between different rows of data in A3 is poor, so the image content has the characteristics of transverse texture, and the continuity between area A3 and the area outside A3 is also poor. In the reconstructed CT image, the clarity of the image content in the corresponding area will be poor. The reason for the appearance of these transverse textures is that the continuity between the multiple sinusoidal graphs used to generate the projection image is weak. The weak continuity between the sinusoidal graphs makes the continuity between different rows of data in the projection image formed by the same row of data also weak.

[0052] Therefore, before reconstructing the third target projection image, all the third target projection images can be subjected to one-dimensional Gaussian filtering in the image height direction, thereby improving the inter-layer continuity in each third target projection image. When the image correction model uses a UNet model, a generative adversarial model (GAN model), a diffusion model, or other types of models, the continuity of the image content can be improved by filtering.

[0053] The following optimization is performed on the image correction process of the diffusion model: after a total of J second target sinusoidal graphs set by the coordinate point pixel values ​​are input into the diffusion model, the diffusion model will iterate during the image correction process, and for the jth (1 ≤ j ≤ J) second target sinusoidal graph currently input , perform t times of reverse denoising from t to 0 to obtain the third target sinusoidal graph Among them, for the first second target sinusoidal graph , and perform t iterations according to the conventional reverse process of the diffusion model, and finally obtain the output of the diffusion model For the 2nd to Jth second target sinusoidal graphs For each second target sinusoidal graph in t iterations, when the denoising result is obtained, the gradient image between the image of the denoising result and the third target sinusoidal graph obtained last is calculated first, and then the difference image between the image of the denoising result and the gradient image is calculated, and the difference image is used as the iteration result of the current iteration and the input of the next iteration.

[0054] The above difference image can be obtained by formula (2): It is expressed as follows. Where j is the serial number of the second target sinusoidal graph that the diffusion model is currently iterating. i is the current iteration number of the diffusion model for the jth second target sinusoidal graph, 0 ≤ i ≤ t. is the denoising result of the i-th iteration of the j-th second target sinusoidal graph according to the conventional inverse process of the diffusion model. is the j-1th third target sinusoidal graph output by the diffusion model. α is a preset coefficient, which can be set to 0.01. Represents the j-1th third target sinusoidal graph and denoising result The gradient image between . It is the output of the diffusion model for the i-th iteration of the j-th second target sinusoidal graph, and it is also the input of the diffusion model for the i-1-th iteration of the j-th second target sinusoidal graph. is the output of the last iteration of the jth second target sinusoidal graph, which can be written as , which is the third target sinusoid corresponding to the jth second target sinusoid. After completing all t iterations of the jth second target sinusoid, the third target sinusoid is obtained. After completing the iteration of all J second target sinusoidal graphs, all J third target sinusoidal graphs are obtained. It can be understood that, for the diffusion model, since the image continuity has been improved by the above formula (2) during the iteration process, the third target projection image may not be smoothed or may be smoothed to further improve the image continuity.

[0055] Fig.11 FIG. 4 shows a third target projection image after image continuity improvement according to some embodiments of the present invention. Fig.11 Since the third target projection image is smoothed and filtered, or since the gradient changes between the sinusoidal images are considered in the process of iteratively generating the sinusoidal images, the gradient changes between the previous sinusoidal images are subtracted when the current sinusoidal image is generated, thereby enhancing the continuity between adjacent sinusoidal images. This makes the data of different rows in the first object area in the formed projection image more coherent, and also makes the data between the first object area and the non-first object area more coherent, thereby reducing the horizontal texture and improving the image quality and clarity.

[0056] Fig.12 A schematic diagram of an image processing system using a hardware implementation of a processing system according to some embodiments of the present invention is shown. Fig.12 The image processing system 1000 provided by the present invention includes a processor 1100 and a memory 1200. The processor 1100 may be a central processing unit (CPU). The processor 1100 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0057] The system 1000 can be implemented using a bus architecture. The bus architecture can include any number of interconnecting buses and bridges, depending on the specific application and overall design constraints of the hardware. The bus 1300 connects various circuits including one or more processors 1100, memory 1200, and / or hardware modules together. The bus 1300 can also connect various other circuits 1400 such as peripherals, voltage regulators, power management circuits, external antennas, etc.

[0058] The bus 1300 may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. A bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one connecting line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0059] The memory 1200 can be used as a non-transitory computer-readable storage medium, and can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions of the computer program in the embodiment of the present invention. The processor 1100 executes various functional applications and data processing of the image processing system and implements the image processing method by running the non-transitory software programs, instructions and modules stored in the memory 1200.

[0060] The memory 1200 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created by the processor 1100, such as projection images, mask images, CT images, sinograms, binary images, data of image correction models, etc. In addition, the memory 1200 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 1200 may optionally include a memory remotely arranged relative to the processor 1100, and these remote memories may be connected to the processor 1100 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0061] The memory 1200 stores a computer program. When the computer program is executed by the processor 1100, the processor 1100 may perform the following steps to implement the image processing method.

[0062] S110 , obtaining a mask image in the CT image of the subject by image segmentation, wherein the mask image is used to represent part or all of the first object in the subject corresponding to the CT image of the subject that can cause artifacts. Step S110 can be implemented by the image segmentation module 1002 .

[0063] S130, generating a first target sinogram of the mask image and a second target sinogram of the subject CT image. The number of the first target sinogram and the number of the second target sinogram are both h, h>1, and the h first target sinograms and the h second target sinograms correspond one to one according to the scanning position. Step S130 can be implemented by the sinogram generation module 1004.

[0064] S150, setting the pixel values ​​of the coordinate points in the target area in the second target sinusoidal graph to target values. The area range of the target area is the same as the image area range corresponding to the first object in the first target sinusoidal graph, and the target value enables the target area to be recognized by the trained image correction model. Step S150 can be implemented by the value modification module 1006.

[0065] S170 , correcting the pixel values ​​of the coordinate points of the target area in the second target sinusoidal graph by using the trained image correction model to obtain h third target sinusoidal graphs. Step S170 may be implemented by the pixel value correction module 1008 .

[0066] S190, generating a target CT image based on the h third target sinusoidal images. The pixel value correction makes the non-mask image area in the target CT image contain fewer artifacts than the non-mask image area in the subject CT image. Step S190 can be implemented by the CT image generation module 1010.

[0067] According to the image processing system proposed in the embodiment of the present invention, firstly, by performing image segmentation on the CT image of the subject to obtain the CT image of the first object that can cause artifacts, that is, to obtain the mask image, and respectively convert the mask image and the CT image of the subject into sinograms to obtain a first target sinogram corresponding to the mask image and a second target sinogram corresponding to the CT image of the subject, in the second target sinogram, determine the area corresponding to the image content of the first object in the first target sinogram as the target area, and uniformly set the pixel values ​​of the coordinate points of the target area to the target values, so that the image content of the target area is different from the image content of the non-target area in the second target sinogram, so that it is easy to be identified, and input the second target sinogram into the artifact correction model, the artifact correction model can adjust the pixel value of the image content of the target area to achieve pixel value correction, and the pixel value corrected image output by the artifact correction model is the third target sinogram, and the target area after pixel value correction can make the artifacts in the area of ​​the first object in the obtained target CT image be at least partially eliminated when the target CT image is generated based on the third target sinogram, thereby at least partially alleviating the artifacts caused by the first object.

[0068] Compared with directly identifying the first object from each projection image, this method utilizes the characteristics that all objects in the sinusoidal image are represented by sinusoidal regions, and at the same time utilizes the characteristics that the coordinate pixel values ​​of the first object and the coordinate pixel values ​​of non-first objects are somewhat different, and uses the sinusoidal image of the mask image to accurately locate the area of ​​the first object, and uses the part of the sinusoidal image that is not blocked by the first object area to restore the data of the first object area, so as to achieve the removal of image artifacts and image restoration, especially when the real shape of the first object is irregular, it can better reflect the better artifact removal effect of this embodiment. In addition, this embodiment can be applied to correct artifacts caused by calcification points in breast CBCT (cone beam CT), can also be applied to correct metal artifacts in oral CBCT, and can also be applied to correct artifacts caused by objects in other CT or CBCT images.

[0069] Taking breast CBCT as an example, the CT image of the subject can be a breast CBCT image. The subject lies prone on the scanning bed, and the breast sags normally due to gravity. The radiation source and the detector rotate horizontally around the breast under the scanning bed and scan the breast at the same time to obtain the scanning data of the breast. The breast CBCT image of the subject is obtained by reconstructing the scanning data of the breast. The reconstruction method can use the filtered back projection algorithm (FBP algorithm) or other reconstruction algorithms.

[0070] After obtaining the breast CBCT image, if there are calcification points in the breast, the calcification points may cause artifacts, that is, the first object may include the calcification points. First, the breast CBCT image is segmented, and the image content of the calcification points is segmented from the breast CBCT image to obtain a mask image.

[0071] The image segmentation method may be threshold segmentation. Calcification points have a stronger ability to absorb X-rays than other substances such as soft tissues. Therefore, the pixel values ​​of calcification points and other tissue structures in the breast are different in CT data. The image can be segmented by a preset numerical threshold. The area formed by the coordinate points whose pixel values ​​are less than the numerical threshold is the calcification point area, that is, the mask area, and the image of the mask area is the mask image. The mask image is also a three-dimensional CT image, which only contains the image content of the calcification points, but not the image content of other substances such as soft tissue in the breast. In this case, the mask image is the calcification point image. The size of the mask image can be the same as the image size of the subject CT image (breast CBCT image), and the pixel value of the non-calcification point area in the mask image can be 0.

[0072] The sinusoidal graph of the calcification point image (first target sinusoidal graph) and the sinusoidal graph of the breast CBCT image (second target sinusoidal graph) are generated respectively. The generated first target sinusoidal graph and second target sinusoidal graph can be referred to Figure 6 and Figure 7 The generation order of the first target sinusoidal graph and the second target sinusoidal graph can be arbitrary, and the number of the two is the same, both h. Figure 1 One-to-one correspondence, the first target sinusoidal graph and the corresponding second target sinusoidal graph correspond to each other in terms of scanning angle and image content, that is, in the second target sinusoidal graph and the first target sinusoidal graph that correspond to each other, the second target sinusoidal graph contains part or content of the first target sinusoidal graph. The second target sinusoidal graph after numerical setting can be referred to Figure 8 .

[0073] The target area is the calcification area in the first target sinusoidal graph, that is, the area to be image repaired. The non-calcification area is the non-target area. Since the first target sinusoidal graph corresponds to the second target sinusoidal graph, the coordinate pixel values ​​in the corresponding target area in the second target sinusoidal graph can be adjusted, for example, the pixel values ​​in the target area are set to a uniform target value, so that the pixel values ​​in the target area become non-valid values ​​and are obviously different from the sinusoidal areas of non-calcification points, so that the target area can be easily recognized by the model. The target value can be 1.

[0074] The h second target sinusoids are sequentially input into the trained image correction model to obtain h third target sinusoids as output. The third target sinusoids can be found in Fig. 9 .from Fig. 9 It can be seen that the image correction model supplements data and repairs the image in area A2, supplements and repairs the sine lines of the non-calcified point objects in area A2, and makes the sine lines of the non-calcified point objects that are blocked and discontinuous and incomplete by the calcified point sine line area become continuous and complete. Due to the data supplementation and image repair of the image correction model, the third target sinogram at this time has approached the sinogram image of the subject without calcification points.

[0075] A target CT image is generated based on the h third target sinusoidal graphs, and the process of generating a CT image from a sinusoidal graph is equivalent to the inverse process of generating a sinusoidal graph from a CT image. Since image restoration is performed in the h third target sinusoidal graphs, the image quality of the generated target CT image is higher than that of the initial subject CT image, especially the target CT image contains fewer artifacts than the subject CT image, and the effective information in the image is more comprehensive and complete.

[0076] In another embodiment, taking oral CBCT as an example, the CT image of the subject may be an oral CBCT image. The radiation source and the detector rotate horizontally around the head of the scanned subject and scan simultaneously to obtain scanning data of the head. The oral CBCT image of the subject is obtained by reconstructing the scanning data of the head. The reconstruction method may adopt a filtered back projection algorithm (FBP algorithm) or other reconstruction algorithms.

[0077] After obtaining the oral CBCT image, if there is a metal object in the teeth, the metal object may cause artifacts, that is, the first object may include a metal object, such as a metal tooth. First, the oral CBCT image is segmented, and the image content of the metal object is segmented from the oral CBCT image to obtain a mask image. The image segmentation method may be threshold segmentation. The size of the mask image (metal object CT image) may be the same as the image size of the subject CT image (oral CBCT image), and the pixel value of the non-metal object area in the mask image may be 0. The coordinate pixel values ​​in the corresponding target area in the second target sinusoidal image are adjusted so that the metal object area in the second target sinusoidal image can be easily recognized by the model. The second target sinusoidal image is input into the trained image correction model to obtain a third target sinusoidal image. The target CT image is generated by the third target sinusoidal image. At this time, the target CT image is an oral CBCT image, and compared with the original oral CBCT image, the target CT image contains fewer artifacts, higher image quality, and more comprehensive and complete effective information in the image.

[0078] In step S130, the method of generating the first target sinogram of the mask image may include the following steps: using the mask image as the CT image to be processed, performing the CT image processing step to obtain h first target sinograms.

[0079] In step S130, the second target sinogram of the CT image of the subject may be generated by the following method 1 or method 2. Method 1: With the CT image of the subject as the CT image to be processed, the CT image processing step is performed to obtain h second target sinograms. Method 2: The scan data corresponding to the CT image of the subject is obtained, and with the scan data corresponding to the CT image of the subject as the scan data to be processed, the scan data processing step is performed to obtain h second target sinograms.

[0080] The mask image and the subject CT image are both three-dimensional images, and the method of generating the sinusoidal graphs of both can be achieved by executing the CT image processing step. Since the subject CT image is reconstructed by scanning data, another method can be used when generating the sinusoidal graph of the subject CT image, that is, by executing the scanning data processing step. In the above two methods of generating the subject CT image, the first method is to use the subject CT image itself as the starting data and start processing, and the second method is to directly use the scanning data of the subject CT image as the starting data and start processing.

[0081] If the CT image of the subject is generated recently and the scan data of the CT image of the subject is retained in the database, either method 1 or method 2 can be selected to generate the sinusoidal graph. If the CT image of the subject is generated a long time ago, since the retention time of the scan data of the CT image of the subject is usually shorter than the retention time of the CT image of the subject, it is necessary to generate the sinusoidal graph by method 1. The above-mentioned CT image processing steps and scan data processing steps are explained below.

[0082] The CT image processing step may include the following sub-steps: forward projecting the CT image to be processed to obtain a plurality of projection images; and taking the plurality of projection images obtained by the forward projection as the projection images to be processed, executing the first image transformation step to obtain data of a plurality of sinusoidal images. The number of images of the plurality of projection images obtained by the forward projection may be the same as the number of projection images corresponding to the CT image to be processed. The method for determining the image area range corresponding to the first object in the first target sinusoidal image may include the following steps: before obtaining the first target sinusoidal image by executing the first image transformation step, binarizing the plurality of first target projection images corresponding to the mask image obtained by the forward projection; and after executing the first image transformation step on the binarized first target projection image to obtain the first target sinusoidal image, determining the image area range corresponding to the first object in the first target sinusoidal image by a preset threshold.

[0083] The image size of the first target sinogram can be the same as that of the second target sinogram, but the effective content in the first target sinogram only includes the image content of the first object. Therefore, in order to facilitate the determination of the image area range of the first object in the first target sinogram, in the process of using the CT image processing step to generate the first target sinogram of the mask image, when the mask image is used as the CT image to be processed to perform the front-projection to obtain multiple first target projection images, the first target projection image is binarized. The binarized first target projection image contains the image content of the first object. Then, the first image transformation step is performed using the binarized first target projection image as the projection image to be processed to obtain the first target sinogram of the mask image. Since the first object has a stronger ability to absorb X-rays than other objects in the subject, a preset numerical threshold is used to judge the pixel values ​​of each position point in the first target sinogram, identify the position points with pixel values ​​higher than the numerical threshold, and form the area range of the first object, that is, Figure 6 The white area in the

[0084] The scanning data processing step may include the following sub-steps: The scan data to be processed is calculated to obtain data of multiple projection images, wherein x is the value of a position point in the scan data to be processed, L is the value of a corresponding position point in a bright field image corresponding to the scan data to be processed, and y is the value of the corresponding position point in the obtained projection image; and using the above multiple projection images as the projection images to be processed, executing the first image transformation step to obtain data of multiple sinusoidal images.

[0085] In the scanning data processing step, each position point in each projection image in the scan data to be processed is calculated according to the above formula (1), so as to obtain a corresponding number of projection images. The projection image data obtained according to the above formula (1) in the scanning data processing step is the same as the projection image data obtained by forward projection in the CT image processing step, so the same sinusoidal graph can be obtained by executing the same first image transformation step. If k1 third projection images with a width (i.e., the number of columns) of w1 and a height (i.e., the number of rows) of h1 are obtained by forward projection, then h1 sinusoidal graphs with a width of w1 and a height of k1 are obtained by executing the first image transformation step. Similarly, if k2 projection images with a width (i.e., the number of columns) of w2 and a height (i.e., the number of rows) of h2 are obtained by the above formula (1), then h2 sinusoidal graphs with a width of w2 and a height of k2 are obtained by executing the first image transformation step.

[0086] The first image transformation step may include the following sub-steps: arranging the data of the same row position of the multiple projection images to be processed in sequence along a preset direction according to a corresponding scanning order. The scanning order is the order in which the multiple projection images to be processed are scanned and generated, and the number of sinusoidal graphs obtained after the arrangement is the same as the number of rows of the projection images to be processed, and the number of projection images to be processed is the same as the number of rows of the sinusoidal graphs obtained after the arrangement. The above-mentioned preset direction may be the image height direction. The number of sinusoidal graphs obtained after the arrangement corresponds to the height of the projection images to be processed, and the height of the sinusoidal graphs obtained after the arrangement corresponds to the number of projection images to be processed.

[0087] The CT image processing step and the scan data processing step both use the first image transformation step to finally obtain a sinogram. The image height direction is the vertical direction, and the Y axis is the vertical direction. The negative semi-axis direction of the Y axis can be used as the preset direction, that is, each row of data is arranged from top to bottom to obtain a sinogram. The sinogram obtained after arrangement can be referred to Figure 6 and Figure 7 , thereby obtaining the first target sinusoidal graph and the second target sinusoidal graph. The scanning order is the order in which the projection images to be processed are generated by scanning, and is usually sorted according to the change of the scanning angle.

[0088] In step S190, the method of generating a target CT image based on h third target sinusoidal graphs may include the following steps: generating k third target projection images based on the h third target sinusoidal graphs, k>1; and reconstructing the k third target projection images to obtain a target CT image. Among them, the method of generating k third target projection images based on h third target sinusoidal graphs may include the following steps: taking the h third target sinusoidal graphs as the sinusoidal graphs to be processed, executing the second image transformation step to obtain data of the k third target projection images. The second image transformation step may include the following sub-steps: arranging the data of the same row position of multiple sinusoidal graphs to be processed in sequence along a preset direction according to the order of the above-mentioned multiple sinusoidal graphs to be processed. Among them, the order of the above-mentioned multiple sinusoidal graphs to be processed corresponds to the order of the rows of the projection images used when generating the above-mentioned multiple sinusoidal graphs to be processed.

[0089] After obtaining the third target sinusoidal graph, it needs to be restored to a CT image before reading the image. The process of generating a projection image based on a sinusoidal graph is the inverse of the process of generating a sinusoidal graph based on a projection image, that is, the second image transformation step is the inverse of the first image transformation step. The preset direction in the second image transformation step is the same as the preset direction in the first image transformation step, and both can be the negative semi-axis direction of the Y axis.

[0090] In step S190, before reconstructing the k third target projection images to obtain the target CT image, the third target projection images are first subjected to one-dimensional filtering in a preset direction. The preset direction may be the image height direction, and the one-dimensional smoothing filtering improves the inter-layer continuity of the projection images in the image height direction.

[0091] After executing step S190, the processor 1100 may also execute step S191: replace the corresponding area in the target CT image with the image of the mask area in the subject CT image. Step S191 may also be implemented by the CT image generation module 1010. Step S191 is used to restore the image content in the first object area in the target CT image. Since in the target CT image obtained after the image correction model performs data supplementation, in addition to correcting the artifacts in some non-mask image areas, the image content in the mask image area is also corrected, which makes the image content in the mask image area difficult to identify. Therefore, the mask image can be used to cover the corresponding position in the target CT image, so that the target CT image is more realistic and accurate.

[0092] In step 170, pixel value correction is performed by a pre-trained image correction model. The training process of the image correction model may include the following steps: generating a sample sinusoidal graph; and training the image correction model based on the sample sinusoidal graph to obtain a trained image correction model. The method of generating the sample sinusoidal graph may include the following steps: obtaining pre-stored scan data, and using the pre-stored scan data as the scan data to be processed, executing the scan data processing step to obtain a plurality of sample sinusoidal graphs; or obtaining a pre-stored CT image, and using the pre-stored CT image as the CT image to be processed, and executing the CT image processing step to obtain a plurality of sample sinusoidal graphs. The method of generating the sample sinusoidal graph used for training is roughly the same as the method of generating the second target sinusoidal graph described above. All the sample sinusoidal graphs may be obtained by processing the pre-stored scan data, or may be obtained by processing the pre-stored CT image, or may be obtained by processing a part of the sample sinusoidal graphs by processing the pre-stored scan data, and another part of the sample sinusoidal graphs by processing the pre-stored CT image. The pre-stored scan data may be obtained by scanning some other subjects in the past, and the pre-stored CT images may be obtained by scanning and reconstructing some other subjects in the past. The scan data processing steps and the CT image processing steps are as described above and will not be described in detail here.

[0093] Image correction models can use different types of network models. For example, a convolutional neural network model (CNN model) can be used, or a generative network model, such as a generative adversarial network model (GAN model) or a fractional diffusion model can be used. By using a network model to repair an image, compared to using interpolation or other methods, it can avoid the formation of secondary artifacts and effectively ensure the quality of the repaired image.

[0094] If the image correction model is a convolutional neural network model, then in the training process of the image correction model, the step of training the image correction model based on the sample sinusoidal graph may include the following sub-steps: setting the pixel value of the coordinate point in the sample sinusoidal graph located in the first area to a preset value to obtain a first sample image, wherein the preset value enables the first area to be recognized by the image correction model; and training the image correction model based on the first sample image to obtain a trained image correction model. The position of the first area in the sample sinusoidal graph may be randomly determined. The shape of the first area may include one or more of the following: a polygon, a circle, and the shape of the area of ​​a preset object in the sinusoidal graph. Both the preset value and the target value may be 1.

[0095] The convolutional neural network model can be a UNet model. Taking breast CBCT as an example, the first area is the calcification point area in the sample sinusoidal image, that is, the area to be image repaired. The first sample image is a sinusoidal image, and there is a first area in the first sample image whose position and shape are randomly set, and the pixel value of the position point in the first area is 1, so that the pixel value in the first area becomes a non-valid value, so that the first area is obviously different from the non-first area (non-calcification point sinusoidal area) in the first sample image. The first sample image may include a polygonal first area, a circular first area, or multiple first areas at the same time. The first area can also be an arched area formed by multiple sinusoidal lines, that is, the area shape of the preset object in the sinusoidal image, so as to make the shape of the first area close to the shape of the target area such as the real calcification point and metal object.

[0096] In the process of training the UNet model, the output of the UNet model is compared with the sample sinusoidal graph without the first region, and the loss function is calculated, so as to adjust the parameters of the UNet model and realize the training of the UNet model.

[0097] The area of ​​the preset object in the sinusoidal graph can be obtained by: generating a sinusoidal graph of the preset object; and identifying the area corresponding to the preset object in the sinusoidal graph of the preset object. The method of generating the sinusoidal graph of the preset object may include the following steps: scanning the preset object, using the obtained scan data as the scan data to be processed, and executing the scan data processing step to obtain multiple sinusoidal graphs of the scanned object.

[0098] The preset object may be an object with a shape similar to that of the calcification point or the metal object, or may be the metal object itself. For example, the preset object may be a metal tooth. The sinusoidal graph of the preset object is generated by scanning the preset object alone, and then calculating according to the above formula (1) through the scanning data processing step, and performing image transformation on the calculation result to obtain multiple sinusoidal graphs.

[0099] If the image correction model is a generative adversarial network model, then during the training process of the image correction model, the step of training the image correction model based on the sample sinusoidal graph may include the following sub-steps: training the image correction model based on the noise graph and the sample sinusoidal graph, the input of the generator of the image correction model includes the noise graph, and the input of the discriminator of the image correction model includes the sample sinusoidal graph.

[0100] The generative adversarial network model (GAN model) includes a generator and a discriminator. A Gaussian noise image that follows a normal distribution is input into the generator to obtain the generator output. The discriminator discriminates the generator output based on the sample sinusoidal graph. The training is performed through the mutual confrontation between the generator and the discriminator in an iterative process.

[0101] If the image correction model is a diffusion model, then in step S170, the image correction model corrects the pixel values ​​of the coordinate points of the target area in the second target sinusoidal graph, which may include the following steps: based on the i-1th iteration result of the jth second target sinusoidal graph Perform the i-th reverse denoising to get the denoising result , 1 ≤ j ≤ J, J is the number of the second target sinusoidal graph; and based on the formula Get the i-th iteration result of the j-th second target sinusoidal graph , and the result of the i-th iteration As the input of the i+1th inverse denoising of the jth second target sinusoidal graph, is the preset coefficient, is the j-1th third target sinusoidal graph obtained after image correction by the image correction model, Represents the j-1th third target sinusoidal graph and denoising result The gradient image between .

[0102] Assuming that there are 100 second target sinusoids that need to be corrected, When image correction is performed, t iterations are performed from t to 0 according to the conventional reasoning process of the diffusion model. After t iterations are completed, the diffusion model outputs the first third target sinusoidal diagram From the 2nd target sinusoidal diagram Initially, the diffusion model uses formula (2) to enhance the image continuity between the current second target sinusoidal graph and the last obtained third target sinusoidal graph in each iteration of t iterations. Formula (2) is: That is, in the second target sinusoidal graph During the t iterations, it is necessary to first perform reverse denoising based on the results of the previous iteration to obtain the denoising results of this (i-th) iteration process. , then based on and the 1st third target sinusoidal graph The gradient image between the two is used to obtain the iterative result of this (i-th) iteration process. In the next (i+1th) iteration, continue based on Get the i+1th denoising result , then based on and the 1st third target sinusoidal graph The gradient image between the two is used to obtain the iterative result of the i+1th iterative process. . And so on until t iterations are completed, and the iteration result of the tth iteration is obtained , that is, the second third target sinusoidal diagram is obtained Then the image correction of the third second target sinusoidal graph is started, and the second third target sinusoidal graph is used in the iterative process. To calculate the gradient image, we finally get the third target sinusoidal image The same process is repeated until all J third target sinusoidal images are obtained. In the third target projection image formed by the third target sinusoidal image, the continuity between different rows of data in the image content of the first object area is good, the phenomenon of horizontal stripes is eliminated, and the image quality of the target CT image finally reconstructed is improved.

[0103] Fig.10 , one of the third target projection images obtained after the diffusion model is iterated according to the conventional reasoning process, that is, the third target projection image obtained when the image gradient is not subtracted according to the above formula during the iteration process, wherein the area A3 pointed by the arrow has horizontal texture. Fig.11 The third target projection image obtained by the diffusion model in the iterative process by calculating the image gradient to improve the image continuity is shown in Fig.10At the position corresponding to region A3 in [description], it can be seen that the continuity of the image in the vertical direction is better, the horizontal texture is eliminated, and the recognizability of the image content is improved.

[0104] It can be understood that the operation can be performed on only some of the second target sine diagrams according to the above formula (2). For example, the operation is performed on m second target sine diagrams corresponding to the first object region among J second target sine diagrams according to the above formula (2), where m < J. For the remaining J - m second target sine diagrams, the denoising result of the i-th time can be directly used as the result of this iteration and the input of the (i + 1)-th iteration to start the (i + 1)-th iteration process, without first performing the operation according to formula (2) and then using the operation result as the result of this iteration. The value of m and the position of m in J can be determined by the position and length of the first object in the height direction of the CT image of the subject. Assume J = 100, the height of the first object in the CT image of the subject is one-fifth of the height of the CT image of the subject and is located in the middle position. Then, the operation can be performed on the 40 - 60th second target sine diagrams (a total of 20, which is one-fifth of 100) according to the above formula (2), that is, m = 20, and the interval of m is located in the middle of J.

[0105] Fig.13 The flowchart of the image processing method according to some embodiments of the present invention is shown. Refer to Fig.13 The present invention provides an image processing method M100, which may include the following steps S110, step S130, step S150, step S170, and step S190.

[0106] S110, obtaining a mask image in the CT image of the subject through image segmentation. Among them, the mask image is used to represent all or part of the first object in the subject corresponding to the CT image of the subject that can cause artifacts.

[0107] S130, generating a first target sine diagram of the mask image and a second target sine diagram of the CT image of the subject. Among them, the number of the first target sine diagrams and the second target sine diagrams is h, h > 1, and the h first target sine diagrams and the h second target sine diagrams correspond one by one according to the scanning position.

[0108] S150, setting the pixel values of the coordinate points in the target region of the second target sine diagram to the target value. Among them, the region range of the target region is the same as the region range of the image region corresponding to the first object in the first target sine diagram, and the target value enables the target region to be recognized by the image correction model.

[0109] S170, correcting the pixel values of the coordinate points in the target region of the second target sine diagram through the trained image correction model to obtain h third target sine diagrams.

[0110] S190, generating a target CT image based on the h third target sinusoidal images, wherein the pixel value correction makes the artifacts contained in the non-mask image area in the target CT image less than the artifacts contained in the non-mask image area in the subject CT image.

[0111] It should be noted that the details not disclosed in the image processing method M100 of this embodiment may refer to the details disclosed in the image processing system 1000 of the above-mentioned embodiment proposed by the present invention, and will not be described in detail here.

[0112] The present invention also provides a readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, it is used to implement the method of any of the above embodiments. For the purpose of this specification, "readable storage medium" can be any device that can contain, store, communicate, propagate or transmit a program for use in an instruction execution system, device or equipment or in combination with these instruction execution systems, devices or equipment. More specific examples of readable storage media (non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and editable read-only memory (EPROM or flash memory), an optical fiber device, and a portable read-only memory (CDROM).

[0113] The present invention also provides a computer program product. The method of the present invention can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instruction is loaded and executed, the process or function of the present invention is executed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user device, a core network device, an OAM or other programmable device.

[0114] The computer program or instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; it may also be an optical medium, such as a digital video disk; it may also be a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both volatile and non-volatile types of storage media.

[0115] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0116] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0117] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0118] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "specific examples", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments / methods or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments / methods or examples described in this specification and the features of the different embodiments / methods or examples, without contradiction.

[0119] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0120] It should be understood by those skilled in the art that the above embodiments are only for the purpose of clearly illustrating the present invention, and are not intended to limit the scope of the present invention. For those skilled in the art, other changes or modifications may be made based on the above disclosure, and these changes or modifications are still within the scope of the present invention.

Claims

1. An image processing system, characterized in that: include: A processor and a memory, wherein the memory stores a computer program. When the computer program is executed by the processor, the processor performs the following process: Obtaining a mask image in the subject CT image by image segmentation, wherein the mask image is used to represent part or all of a first object in the subject corresponding to the subject CT image that can cause artifacts; Taking the mask image as the CT image to be processed, executing the CT image processing step to obtain h first target sinusoidal images, where h>1; Taking the subject CT image as the CT image to be processed, executing the CT image processing step to obtain h second target sinograms, or acquiring scanning data corresponding to the subject CT image, and taking the scanning data corresponding to the subject CT image as the scanning data to be processed, executing the scanning data processing step to obtain h second target sinograms, wherein the h first target sinograms and the h second target sinograms correspond to each other one by one according to scanning positions; Setting pixel values ​​of coordinate points in a target area in the second target sinusoidal graph to a target value, wherein the area range of the target area is the same as the image area range corresponding to the first object in the first target sinusoidal graph, and the target value enables the target area to be recognized by a trained image correction model; Correcting the pixel values ​​of the coordinate points of the target area in the second target sinusoidal graph by using the image correction model to obtain h third target sinusoidal graphs; and generating a target CT image based on the h third target sinusoidal images, wherein the pixel value correction makes the non-mask image region in the target CT image contain fewer artifacts than the non-mask image region in the subject CT image; The CT image processing step includes: forward projecting the CT image to be processed to obtain a plurality of projection images; and taking the plurality of projection images obtained by the forward projection as the projection images to be processed, executing a first image transformation step to obtain data of a plurality of sinusoidal images; The first image transformation step includes: arranging the data of the same row position of the plurality of projection images to be processed in sequence along a preset direction according to a corresponding scanning order, wherein the scanning order is the order in which the plurality of projection images to be processed are scanned and generated, and the number of sinusoidal images obtained after the arrangement is the same as the number of rows of the projection images to be processed, and the number of the projection images to be processed is the same as the number of rows of the sinusoidal image obtained after the arrangement; The image correction model is a diffusion model, and the image correction model corrects the pixel values ​​of the coordinate points of the target area in the second target sinusoidal image, including: Based on the i-1th iteration result of the jth second target sinusoidal graph Perform the i-th reverse denoising to get the denoising result , 1 ≤ j ≤ J, J is the number of the second target sinusoidal graphs; and Based on the formula Get the i-th iteration result of the j-th second target sinusoidal graph , and the i-th iteration result As the input of the i+1th inverse denoising of the jth second target sinusoidal graph, is the preset coefficient, is the j-1th third target sinusoidal graph obtained after image correction by the image correction model, Represents the j-1th third target sinusoidal graph and the denoising result The gradient image between .

2. The image processing system according to claim 1, characterized in that Generating a target CT image based on the h third target sinusoidal images comprises: Generate k third target projection images based on the h third target sinusoidal images, k>1; and The k third target projection images are reconstructed to obtain a target CT image.

3. The image processing system according to claim 2, characterized in that: Generating k third target projection images based on the h third target sinusoidal images comprises: The h third target sinusoidal graphs are used as the sinusoidal graphs to be processed, and the second image transformation step is performed to obtain data of k third target projection graphs.

4. The image processing system according to claim 3, characterized in that: The second image transformation step comprises: The data at the same row position of the plurality of sinusoidal graphs to be processed are sequentially arranged along a preset direction according to the order of the plurality of sinusoidal graphs to be processed, and the order of the plurality of sinusoidal graphs to be processed corresponds to the order of the rows of the projection graph used when generating the plurality of sinusoidal graphs to be processed.

5. The image processing system according to claim 2, characterized in that: Before reconstructing the k third target projection images to obtain the target CT image, one-dimensional filtering is first performed on the third target projection images in a preset direction.

6. The image processing system according to claim 1, characterized in that: After generating the target CT image based on the h third target sinusoidal images, the processor further performs the following process: The mask image is used to replace the corresponding area in the target CT image.

7. The image processing system according to claim 1, characterized in that: The training process of the image correction model includes: generating a sample sinogram; and The image correction model is trained based on the sample sinusoidal graph to obtain a trained image correction model.

8. The image processing system according to claim 7, characterized in that: Generates a sample sinogram, including: Acquiring pre-stored scan data, and using the pre-stored scan data as scan data to be processed, executing a scan data processing step to obtain a plurality of sample sinusoidal graphs; and / or, A pre-stored CT image is acquired, and the pre-stored CT image is used as the CT image to be processed, and a CT image processing step is performed to obtain a plurality of sample sinograms.

9. The image processing system according to claim 1, characterized in that: The number of the multiple projection images obtained by forward projection is the same as the number of projection images corresponding to the CT image to be processed.

10. The image processing system according to claim 1, characterized in that: The method for determining the image area range corresponding to the first object in the first target sinusoidal graph includes: Before obtaining the first target sinusoidal graph by executing the first image transformation step, a plurality of first target projection graphs corresponding to the mask image obtained by forward projection are binarized; and after obtaining the first target sinusoidal graph by executing the first image transformation step on the binarized first target projection graphs, an image area range corresponding to the first object in the first target sinusoidal graph is determined by a preset threshold.

11. The image processing system according to claim 1 or 8, characterized in that: The scanning data processing step comprises: By formula Calculating the scan data to be processed to obtain data of multiple projection images, wherein f is the value of a position point in the scan data to be processed, L is the value of a corresponding position point in the bright field image corresponding to the scan data to be processed, and y is the value of the corresponding position point in the obtained projection image; and The plurality of projection images obtained by calculating the scan data to be processed are used as the projection images to be processed, and the first image transformation step is performed to obtain the data of a plurality of sinusoidal images.

12. The image processing system according to claim 1, characterized in that: The preset direction is the image height direction.

13. The image processing system according to claim 1 or 12, characterized in that: The number of sinusoidal graphs obtained after arrangement corresponds to the height of the projection graph to be processed, and the height of the sinusoidal graphs obtained after arrangement corresponds to the number of the projection graphs to be processed.

14. The image processing system according to claim 1, characterized in that: The image segmentation method is threshold segmentation.

15. The image processing system according to claim 1 or 2, characterized in that: The first object includes a metal object and / or a calcification point.

16. The image processing system according to claim 1 or 2, characterized in that: The size of the mask image is the same as the image size of the object CT image.

17. An image processing method, characterized in that: include: Obtaining a mask image in the subject CT image by image segmentation, wherein the mask image is used to represent part or all of a first object in the subject corresponding to the subject CT image that can cause artifacts; Taking the mask image as the CT image to be processed, executing the CT image processing step to obtain h first target sinusoidal images, where h>1; Taking the subject CT image as the CT image to be processed, executing the CT image processing step to obtain h second target sinograms, or acquiring scanning data corresponding to the subject CT image, and taking the scanning data corresponding to the subject CT image as the scanning data to be processed, executing the scanning data processing step to obtain h second target sinograms, wherein the h first target sinograms and the h second target sinograms correspond to each other one by one according to scanning positions; Setting pixel values ​​of coordinate points in a target area in the second target sinusoidal graph to a target value, wherein the area range of the target area is the same as the image area range corresponding to the first object in the first target sinusoidal graph, and the target value enables the target area to be recognized by a trained image correction model; Correcting the pixel values ​​of the coordinate points of the target area in the second target sinusoidal graph by using the image correction model to obtain h third target sinusoidal graphs; and generating a target CT image based on the h third target sinusoidal images, wherein the pixel value correction makes the non-mask image region in the target CT image contain fewer artifacts than the non-mask image region in the subject CT image; The CT image processing step includes: forward projecting the CT image to be processed to obtain a plurality of projection images; and taking the plurality of projection images obtained by the forward projection as the projection images to be processed, executing a first image transformation step to obtain data of a plurality of sinusoidal images; The first image transformation step includes: arranging the data of the same row position of the plurality of projection images to be processed in sequence along a preset direction according to a corresponding scanning order, wherein the scanning order is the order in which the plurality of projection images to be processed are scanned and generated, and the number of sinusoidal images obtained after the arrangement is the same as the number of rows of the projection images to be processed, and the number of the projection images to be processed is the same as the number of rows of the sinusoidal image obtained after the arrangement; The image correction model is a diffusion model, and the image correction model corrects the pixel values ​​of the coordinate points of the target area in the second target sinusoidal image, including: Based on the i-1th iteration result of the jth second target sinusoidal graph Perform the i-th reverse denoising to get the denoising result , 1 ≤ j ≤ J, J is the number of the second target sinusoidal graphs; and Based on the formula Get the i-th iteration result of the j-th second target sinusoidal graph , and the i-th iteration result As the input of the i+1th inverse denoising of the jth second target sinusoidal graph, is the preset coefficient, is the j-1th third target sinusoidal graph obtained after image correction by the image correction model, Represents the j-1th third target sinusoidal graph and the denoising result The gradient image between .

18. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which is used to implement the method according to claim 17 when executed by a processor.

19. A computer program product, characterized in that The computer program product comprises a computer program, which is used to implement at least the method according to claim 17 when being executed by a processor.