A method, system and apparatus for x-ray three-dimensional scanning imaging

Abstract

The application relates to the technical field of medical intelligent imaging, in particular to an X-ray three-dimensional scanning imaging method, system and device. The method comprises the following steps: acquiring a denoising network training data set, the denoising network training data set comprising a plurality of groups of denoising training data, wherein one group of denoising training data comprises a corresponding relationship between an original noisy DR image, noise parameters, sensor parameters and a denoised DR image; the noise parameters are used for representing noise-related data acquired when the original noisy DR image is acquired; the sensor parameters are used for representing performance data of a sensor for acquiring the original noisy DR image and motion data acquired by the sensor in a three-dimensional scanning process; an initial denoising network model is trained by using the denoising network training data set to obtain a target denoising network model; a to-be-processed DR image is input into the target denoising network model to obtain a first DR image, and the noise in the first DR image is smaller than that in the to-be-processed DR image.

Description

Technical Field

[0001] This application relates to the field of medical intelligent imaging technology, and in particular to a method and system for X-ray three-dimensional scanning imaging. Background Art

[0002] Medical imaging technology is of vital importance to the advancement of healthcare and has paved a new path for research in medicine and the natural sciences. The earliest medical imaging devices were analog X-ray machines, used by clinicians to observe the internal structures of the human body and becoming the first radiological imaging devices used in clinical practice. In the decades that followed, radiological imaging devices such as computed tomography (CT), computed radiography (CR), and digital radiography (DR) were introduced and became indispensable diagnostic equipment in hospitals.

[0003] Digital radiography (DR) is a medical X-ray imaging technique developed by combining computer-aided digital image processing with radiology. In recent years, with the advancement of science and technology, X-ray imaging has also seen tremendous progress. From the perspective of imaging principles, it has evolved from analog film (analog X-ray machines), IP panels (CR machines), and finally digital flat-panel detectors (DR). Compared to analog X-ray machines and CR, DR offers advantages such as faster imaging speed, wider dynamic range, lower radiation dose, and a higher degree of automation.

[0004] However, DR currently still has the problem of large noise caused by ultra-low dose during three-dimensional dynamic reconstruction, resulting in low image quality of three-dimensional dynamic DR images, affecting doctors' visual judgment and experience. Summary of the Invention

[0005] In order to improve the image quality of three-dimensional dynamic DR images and enhance the doctor's visual judgment and experience, this application provides a method and system for X-ray three-dimensional scanning imaging, which adopts the following technical solutions:

[0006] In a first aspect, an embodiment of the present application discloses a method for three-dimensional X-ray scanning imaging, comprising:

[0007] Obtain a denoising network training data set, wherein the denoising network training data set includes multiple sets of denoising training data, wherein one set of denoising training data includes a correspondence between an original noisy DR image, noise parameters, sensor parameters, and a denoised DR image; the noise parameters are used to represent noise-related data obtained when obtaining the original noisy DR image; and the sensor parameters are used to represent performance data of the sensor used to obtain the original noisy DR image and motion data obtained by the sensor during the DR three-dimensional scanning process;

[0008] Using the denoising network training data set to train an initial denoising network model to obtain a target denoising network model; the target denoising network model includes motion estimation parameters of the DR generated based on the training, and the motion estimation parameters of the DR are used to represent the error value between the actual motion position of the DR during the three-dimensional scanning process and the preset motion position;

[0009] The DR image to be processed is input into the target denoising network model to obtain a first DR image, wherein the noise in the first DR image is smaller than the noise in the DR image to be processed.

[0010] By adopting the above technical solution, a denoising network training data set with the correspondence between the original noisy DR image, noise parameters, sensor parameters and the denoised DR image is obtained, which can be used to train the initial denoising network model to establish a target denoising network model with denoising function, and then process the DR image to be processed to obtain a first DR image with less noise than the noise in the DR image to be processed, thereby improving the image quality of the DR image and enhancing the doctor's visual judgment and experience.

[0011] Optionally, after obtaining the processed DR image, the method further includes:

[0012] Using the attention module and the perceptual loss module to train the target denoising network model to obtain a first image network model;

[0013] The first DR image is input into the first image network model to obtain a second DR image; wherein the resolution of the second DR image is greater than the resolution of the first DR image.

[0014] By adopting the above technical solution, by using the attention module and the perceptual loss module to train the target denoising network model, and inputting the first DR image into the first image network model, the second DR image can remove frame images with higher similarity compared to the first DR image, reduce image redundancy and increase image resolution.

[0015] Optionally, inputting the first DR image into the target image network model to obtain the second DR image specifically includes:

[0016] Aligning and selecting features between adjacent frames in the first DR image to obtain a feature-processed image;

[0017] Adjacent frames of the feature-processed image are fused to obtain a second DR image, which is a three-dimensional image.

[0018] By adopting the above technical solution, by aligning the features between adjacent frames in the first DR image, it is possible to select images with large differences between adjacent frames, thereby obtaining a feature-processed image with low redundancy, and by fusing the feature-processed images between adjacent frames, a second DR image in a three-dimensional state can be obtained, which can facilitate doctors to diagnose lesions.

[0019] Optionally, the attention module in the first image network model aligns and selects features between adjacent frames in the first DR image to obtain a feature-processed image, specifically including:

[0020] Extracting features from each frame of the first DR image;

[0021] The features in adjacent frame images whose feature similarity reaches a preset similarity value are aligned, otherwise they are discarded to obtain the feature-processed image.

[0022] By adopting the above technical solution, by first extracting features from each frame of the first DR image and performing alignment and selection based on feature similarity, the redundancy of the first DR image can be further reduced.

[0023] Optionally, after inputting the first DR image into the first image network model to obtain the second DR image, the method further includes:

[0024] Using three low dynamic range images with adjacent frame numbers of underexposed, normally exposed, and overexposed to train the first image network model, obtaining a second image network model;

[0025] The second DR image is input into the second image network model to obtain the third DR image; wherein the dynamic range of the third DR image is greater than the dynamic range of the second DR image.

[0026] By adopting the above technical solution, by using three low dynamic range images with adjacent frame numbers of underexposure, normal exposure and overexposure to train the first image network model, and inputting the second DR image into the second image network model, it is possible to process the low dynamic image, i.e., the second DR image, into a high dynamic image, i.e., the third DR image.

[0027] In a second aspect, another embodiment of the present application discloses a system for three-dimensional X-ray scanning imaging, comprising:

[0028] A denoising network training data set acquisition module is used to acquire a denoising network training data set, wherein the denoising network training data set includes multiple sets of denoising training data, wherein one set of denoising training data includes a correspondence between an original noisy DR image, noise parameters, sensor parameters, and a denoised DR image; the noise parameters are used to represent noise-related data obtained when acquiring the original noisy DR image; the sensor parameters are used to represent performance data of the sensor used to acquire the original noisy DR image and motion data acquired by the sensor during the DR three-dimensional scanning process;

[0029] A target denoising network model establishment module is used to train an initial denoising network model using the denoising network training data set to obtain a target denoising network model; the target denoising network model includes motion estimation parameters of the DR generated based on the training, and the motion estimation parameters of the DR are used to represent the error value between the actual motion position of the DR during the three-dimensional scanning process and the preset motion position;

[0030] The first DR image acquisition module is used to input the DR image to be processed into the target denoising network model to obtain a first DR image, wherein the noise in the first DR image is smaller than the noise in the DR image to be processed.

[0031] By adopting the above technical solution, first, through the denoising network training data set acquisition module, a denoising network training data set with the corresponding relationship between the original noisy DR image, noise parameters, sensor parameters and denoised DR image can be obtained. Secondly, through the target denoising network model establishment module, the initial denoising network model is trained with the denoising network training data set to establish a target denoising network model with denoising function. Thirdly, through the first DR image acquisition module, the DR image to be processed is input into the target denoising network model for processing, and a first DR image with less noise than the noise in the DR image to be processed can be obtained, thereby improving the image quality of the DR image to enhance the doctor's visual judgment and experience.

[0032] Optionally, a system for three-dimensional X-ray scanning imaging further includes: a first image network model establishment module, configured to train the target denoising network model using an attention module and a perceptual loss module to obtain a first image network model;

[0033] The second DR image acquisition module is used to input the first DR image into the first image network model to obtain a second DR image; wherein the resolution of the second DR image is greater than the resolution of the first DR image.

[0034] By adopting the above-mentioned technical solution, through the first image network model establishment module, the attention module and the perceptual loss module can be used to train the target denoising network model to obtain the first image network model. Through the second DR image acquisition module, the first DR image can be input into the first image network model, so that the second DR image removes frame images with higher similarity compared to the first DR image, reduces image redundancy and increases image resolution.

[0035] Optionally, the X-ray three-dimensional scanning imaging system further includes: a second image network model establishment module, configured to train the first image network model using three low dynamic range images with adjacent frames of underexposure, normal exposure, and overexposure to obtain a second image network model;

[0036] The third DR image acquisition module is used to input the second DR image into the second image network model to obtain the third DR image; wherein the dynamic range of the third DR image is greater than the dynamic range of the second DR image.

[0037] By adopting the above technical solution, a module is established through the second image network model, and three low-dynamic range images with adjacent frame numbers of underexposure, normal exposure and overexposure are used to train the first image network model. The second DR image is input into the second image network model through the third DR image acquisition module. This can realize the processing of the low-dynamic image, namely the second DR image, into a high-dynamic image, namely the third DR image, thereby increasing the clarity of the image and avoiding the problem of inconsistent brightness between adjacent frame images in the image during playback, which causes flickering.

[0038] In a third aspect, another embodiment of the present application discloses an X-ray three-dimensional scanning imaging device, comprising: a memory and a processor, wherein the memory is used to store a computer program; and the processor is used to implement the steps of an X-ray three-dimensional scanning imaging method as described in any one of the above items when executing the computer program.

[0039] In a fourth aspect, another embodiment of the present application discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for three-dimensional X-ray scanning imaging as described in any one of the above items are implemented.

[0040] In summary, this application includes at least one of the following beneficial technical effects:

[0041] 1. By obtaining a denoising network training dataset with the correspondence between the original noisy DR image, noise parameters, sensor parameters and the denoised DR image, it can be used to train the initial denoising network model to establish a target denoising network model with denoising function, and then process the DR image to be processed to obtain a first DR image with less noise than the noise in the DR image to be processed, thereby improving the image quality of the DR image and enhancing the doctor's visual judgment and experience.

[0042] 2. By using the attention module and the perceptual loss module to train the target denoising network model, the first DR image is input into the first image network model, so that the second DR image can remove the frame images with higher similarity compared to the first DR image, reduce image redundancy and increase image resolution.

[0043] 3. By using three low-dynamic-range images with adjacent frame numbers of underexposure, normal exposure, and overexposure to train the first image network model, and inputting the second DR image into the second image network model, it is possible to process the low-dynamic image, i.e., the second DR image, into a high-dynamic image, i.e., the third DR image. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A schematic diagram of the usage scenario of practical X-ray 3D scanning imaging;

[0045] Figure 2 A schematic flow chart of a method for three-dimensional X-ray scanning imaging disclosed in one embodiment of the present application;

[0046] Figure 3 This is a schematic diagram of the process after step S30 in the X-ray three-dimensional scanning imaging method disclosed in another embodiment of the present application;

[0047] Figure 4 This is a schematic diagram of the process after step S30 in the X-ray three-dimensional scanning imaging method disclosed in another embodiment of the present application;

[0048] Figure 5 This is a module diagram of an X-ray three-dimensional scanning imaging system disclosed in another embodiment of the present application.

[0049] Figure 6 This is a schematic structural diagram of an X-ray three-dimensional scanning imaging device disclosed in another embodiment of the present application.

[0050] Description of reference numerals:

[0051] 10. Denoising network training data set acquisition module; 20. Target denoising network model establishment module; 30. First DR image acquisition module; 40. First image network model establishment module; 50. Second DR image acquisition module; 60. Second image network model establishment module; 70. Third DR image acquisition module; 80. Memory; 90. Processor. DETAILED DESCRIPTION

[0052] The present application is further described in detail below with reference to the accompanying drawings.

[0053] See also Figure 1 , which is a schematic diagram of the scene of a patient undergoing X-ray 3D photography. The X-ray emitting device can be set to a double-column or suspended type to scan the patient's body parts at different angles to obtain a series of images with angle information. Through algorithm reconstruction, a coronal tomographic image, that is, a 3D dynamic image, can be obtained to solve the problem of unnecessary obstructions when shooting from a single side (such as the front of the chest cavity), such as overlapping occlusions of organs. By obtaining DR images at different angles, organ imaging at other angles can be obtained. For example, the lungs shot at a 45-degree angle from above can be used to complete the lungs and heart blocked by the sternum and ribs at a 30-degree angle, thereby forming a 3D image for doctors to diagnose lesions in different parts of the body.

[0054] It is explained here that DR imaging uses three-dimensional images and uses a small radiation dose, which can reduce damage to the human body. Although the single-sided penetration is weaker than that of CT scanning, it can achieve the effect of three-dimensional scanning by combining multi-angle scanning and three-dimensional dynamic reconstruction through algorithms. However, during three-dimensional dynamic reconstruction, there is still a problem of large noise caused by ultra-low dose, which affects the image quality of three-dimensional dynamic DR images. This problem is solved in this application through the following embodiments.

[0055] [First embodiment]

[0056] In the first embodiment of the present application, in order to improve the image quality of three-dimensional dynamic DR images and enhance the visual diagnosis and experience of doctors, a method of three-dimensional X-ray scanning imaging is specifically disclosed. Figure 2 , specifically including the following steps:

[0057] S10: Obtain denoising network training data set;

[0058] Among them, the denoising network training data set may include multiple groups of denoising training data, one group of denoising training data may include an original noisy DR image, noise parameters, sensor parameters and the corresponding relationship between a denoised DR image; the noise parameters are used to represent the noise-related data obtained when obtaining the original noisy DR image; the sensor parameters are used to represent the performance data of the sensor for obtaining the original noisy DR image and the motion data obtained by the sensor of the DR device during the three-dimensional scanning process. The motion data may include motion speed, direction, acceleration and inertia, etc.

[0059] S20: Using the denoising network training data set to train an initial denoising network model to obtain a target denoising network model;

[0060] Among them, the target denoising network model may include the motion estimation parameters of the DR generated based on training. The motion estimation parameters of the DR are used to represent the error value between the actual motion position of the DR in the three-dimensional scanning process and the preset motion position. Regarding the error value, it can be that the motion position of the DR in the three-dimensional scanning process is inconsistent with the preset motion position, that is, the rational position due to inertia factors, and the difference is also the error value. Of course, the factors that form the error value are not limited here.

[0061] S30: inputting the DR image to be processed into the target denoising network model to obtain a first DR image;

[0062] The noise in the first DR image is smaller than the noise in the DR image to be processed.

[0063] For example, when there are two denoising network training data sets A and B, the denoising training data in A only includes the correspondence between the original noisy DR image and the denoised DR image, and the denoising training data in B includes the correspondence between the original noisy DR image, noise parameters, sensor parameters and the denoised DR image. The initial denoising network models are trained for A and B respectively, and the A denoising network model and the B denoising network model are obtained respectively. Then, a section of unprocessed DR image is input into the A denoising network model and the B denoising network model respectively, and the A image processed by the A denoising network model and the B image processed by the B denoising network model are obtained, where the noise of the B image is smaller than the noise in the A image and the noise of the unprocessed DR image respectively.

[0064] See also Figure 3 In another embodiment, after step S30, the method further includes:

[0065] S40: Using the attention module and the perceptual loss module to train the target denoising network model to obtain a first image network model;

[0066] Among them, the attention module can be understood as enhancing attention on certain specific features. When training the network model, it can be specifically used to perform attention training on the original DR image based on the set specific features; the perceptual loss module can be understood as being used to determine the content loss information of the frame image in the image. When training the network model, it can be specifically used to perform content loss training on the frame image in the original DR image based on the set high-resolution image features.

[0067] S50: inputting the first DR image into the first image network model to obtain a second DR image;

[0068] Among them, the first DR image is processed by the first image network model to remove frame images with high similarity, reduce image redundancy and increase image resolution.

[0069] In another embodiment, step S50 further specifically includes:

[0070] S51: aligning and selecting features between adjacent frames in the first DR image to obtain a feature-processed image;

[0071] Among them, the alignment and selection processing mentioned above can be used to extract features from each frame image in the first DR image, so as to align the features in adjacent frame images whose feature similarity reaches a preset similarity value, otherwise, they are discarded to obtain a feature-processed image, thereby reducing the redundancy of the image.

[0072] S52: Fusing adjacent frames of the feature-processed image to obtain a second DR image, where the second DR image is a three-dimensional image.

[0073] Among them, the second DR image is a three-dimensional image, which can be understood as a multi-angle scan based on the DR device. The scanned images are fused to obtain a three-dimensional image of the human body, which is convenient for doctors to diagnose lesions.

[0074] See also Figure 4 In another embodiment, after step S50 or step S52, the method further includes:

[0075] S60: Using three low dynamic range images with adjacent frame numbers, namely, underexposed, normally exposed, and overexposed, to train the first image network model to obtain a second image network model;

[0076] Among them, underexposed, normally exposed and overexposed images can be selected manually based on the exposure intensity.

[0077] S70: Inputting the second DR image into the second image network model to obtain the third DR image; wherein the dynamic range of the third DR image is greater than the dynamic range of the second DR image.

[0078] Among them, after the second DR image is input into the second image network model for processing, the low-dynamic image, that is, the second DR image, can be processed into a high-dynamic image, that is, the third DR image. The high-dynamic image can be understood as a high-dynamic image (HDRI) played back in each frame, which helps the human eye to identify the image and increase the detail or clarity of each frame. It can not only solve the problem of limited dynamic range caused by equipment reasons, but also avoid the problem of inconsistent brightness and darkness between adjacent frames in the image during playback, resulting in flickering.

[0079] In summary, the first embodiment of the present invention discloses a method for X-ray three-dimensional scanning imaging, which can be used to train an initial denoising network model to establish a target denoising network model with a denoising function by obtaining a denoising network training data set having a corresponding relationship between an original noisy DR image, noise parameters, sensor parameters and a denoised DR image, and then process the DR image to be processed to obtain a first DR image with less noise than the noise in the DR image to be processed, thereby improving the image quality of the DR image to enhance the doctor's visual judgment and experience; by using an attention module and a perceptual loss module to train the target denoising network model, the first DR image is input into the first image network model, so that the second Compared with the first DR image, the DR image removes frame images with high similarity, reduces image redundancy and increases image resolution; by using three low-dynamic range images with adjacent frames of underexposure, normal exposure and overexposure to train the first image network model, and inputting the second DR image into the second image network model, the low-dynamic image, namely the second DR image, can be processed into a high-dynamic image, namely the third DR image, which helps the human eye to identify the image and increases the detail or clarity of each frame image. It can not only solve the problem of limited dynamic range caused by equipment reasons, but also avoid the problem of inconsistent brightness and darkness between adjacent frame images in the image during playback, resulting in flickering.

[0080] [Second embodiment]

[0081] In the second embodiment of the present application, a system for three-dimensional X-ray scanning imaging is disclosed. Figure 5 The system includes: a denoising network training data set acquisition module 10, a target denoising network model establishment module 20, a first DR image acquisition module 30, a first image network model establishment module 40, a second DR image acquisition module 50, a second image network model establishment module 60 and a third DR image acquisition module 70.

[0082] Among them, the denoising network training data set acquisition module 10 is used to acquire a denoising network training data set, wherein the denoising network training data set includes multiple groups of denoising training data, wherein one group of denoising training data includes a correspondence between an original noisy DR image, noise parameters, sensor parameters, and a denoised DR image; the noise parameters are used to represent noise-related data obtained when acquiring the original noisy DR image; the sensor parameters are used to represent performance data of the sensor that acquires the original noisy DR image and motion data acquired by the sensor during the DR three-dimensional scanning process;

[0083] A target denoising network model establishment module 20 is configured to train an initial denoising network model using the denoising network training data set to obtain a target denoising network model; the target denoising network model includes motion estimation parameters of the DR generated based on the training, and the motion estimation parameters of the DR are used to represent the error value between the actual motion position of the DR during the three-dimensional scanning process and the preset motion position;

[0084] The first DR image acquisition module 30 is used to input the DR image to be processed into the target denoising network model to obtain a first DR image, in which the noise in the first DR image is smaller than the noise in the DR image to be processed.

[0085] A first image network model building module 40 is used to train the target denoising network model using an attention module and a perceptual loss module to obtain a first image network model;

[0086] The second DR image acquisition module 50 is used to input the first DR image into the first image network model to obtain a second DR image; wherein the resolution of the second DR image is greater than that of the first DR image.

[0087] A second image network model building module 60 is configured to train the first image network model using three low dynamic range images with adjacent frames of underexposure, normal exposure, and overexposure to obtain a second image network model;

[0088] The third DR image acquisition module 70 is used to input the second DR image into the second image network model to obtain the third DR image; wherein the dynamic range of the third DR image is greater than the dynamic range of the second DR image.

[0089] It should be noted that the X-ray 3D scanning imaging method implemented by the X-ray 3D scanning imaging system disclosed in the second embodiment of this application is similar to that of the first embodiment, and therefore will not be described in detail here. Optionally, the various modules and other operations or functions described above in this embodiment are respectively for implementing the methods in the aforementioned embodiments.

[0090] [Third embodiment]

[0091] In the third embodiment of the present application, an X-ray three-dimensional scanning imaging device is disclosed. Figure 6 The device includes a memory 80 and a processor 90. The memory 80 is used to store a computer program. The processor 90 is used to execute the computer program to implement the steps of the X-ray 3D scanning imaging method described in the first embodiment. For details, please refer to the above description and will not be described in detail here. The technical effects of the X-ray 3D scanning imaging device provided in this embodiment in actual application are the same as those of the X-ray 3D scanning imaging method in the first embodiment.

[0092] [Fourth embodiment]

[0093] In the fourth embodiment of the present application, a computer-readable storage medium is disclosed. The computer-readable storage medium is, for example, a non-volatile memory, such as: magnetic media (such as hard disks, floppy disks, and tapes), optical media (such as CDROMs and DVDs), magneto-optical media (such as optical disks), and hardware devices specially constructed to store and execute computer-executable instructions (such as read-only memory (ROM), random access memory (RAM), flash memory, etc.). A computer program is stored on the computer-readable storage medium. The computer-readable storage medium can be used by one or more processors or processing devices to execute the computer program to implement the method of X-ray three-dimensional scanning imaging in the aforementioned embodiment.

[0094] In addition, it can be understood that the aforementioned embodiments are merely exemplary descriptions of the present invention. Under the premise that the technical features do not conflict, the structures do not contradict, and the purpose of the present invention is not violated, the technical solutions of the various embodiments can be arbitrarily combined and used in combination.

[0095] In the several embodiments provided by the present invention, it should be understood that the disclosed methods, systems, and devices can be implemented in other ways. For example, the modules included in the system described above are merely schematic, and the division of modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0096] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0097] In addition, the functional units / modules in various embodiments of the present invention may be integrated into a single processing unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated into a single unit / module. The aforementioned integrated units / modules may be implemented in the form of hardware or hardware plus software functional units / modules.

[0098] The above-mentioned integrated unit / module implemented in the form of a software functional unit / module can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium and includes a number of instructions for causing one or more processors of a computer device (which can be a personal computer, server, or network device, etc.) to execute some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc., various media that can store program code.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for three-dimensional X-ray scanning imaging, characterized in that: include: Obtaining a denoising network training data set, wherein the denoising network training data set includes multiple groups of denoising training data, wherein one group of denoising training data includes a correspondence between an original noisy DR image, noise parameters, sensor parameters, and a denoised DR image; The noise parameters are used to represent noise-related data obtained when acquiring the original noisy DR image; the sensor parameters are used to represent performance data of the sensor used to acquire the original noisy DR image and motion data acquired by the sensor during the DR three-dimensional scanning process; Using the denoising network training data set to train an initial denoising network model to obtain a target denoising network model; the target denoising network model includes motion estimation parameters of the DR generated based on the training, and the motion estimation parameters of the DR are used to represent the error value between the actual motion position of the DR during the three-dimensional scanning process and the preset motion position; The DR image to be processed is input into the target denoising network model to obtain a first DR image, wherein the noise in the first DR image is smaller than the noise in the DR image to be processed.

2. The method according to claim 1, characterized in that After inputting the DR image to be processed into the target denoising network model to obtain the first DR image, the method further includes: Using the attention module and the perceptual loss module to train the target denoising network model to obtain a first image network model; The first DR image is input into the first image network model to obtain a second DR image; wherein the resolution of the second DR image is greater than the resolution of the first DR image, and the redundancy of the second DR image is less than the redundancy of the first DR image.

3. The method according to claim 2, characterized in that Inputting the first DR image into the first image network model to obtain the second DR image specifically includes: Aligning and selecting features between adjacent frames in the first DR image to obtain a feature-processed image; Adjacent frames of the feature-processed image are fused to obtain a second DR image, which is a three-dimensional image.

4. The method according to claim 3, characterized in that The attention module in the first image network model aligns and selects features between adjacent frames in the first DR image to obtain a feature-processed image, specifically including: Extracting features from each frame of the first DR image; The features in adjacent frame images whose feature similarity reaches a preset similarity value are aligned, otherwise they are discarded to obtain the feature-processed image.

5. The method according to claim 2 or 4, characterized in that After inputting the first DR image into the first image network model to obtain a second DR image, the method further includes: Using three low dynamic range images with adjacent frame numbers of underexposed, normally exposed, and overexposed to train the first image network model, obtaining a second image network model; The second DR image is input into the second image network model to obtain a third DR image; wherein the dynamic range of the third DR image is greater than the dynamic range of the second DR image.

6. An X-ray three-dimensional scanning imaging system, characterized in that: include: A denoising network training data set acquisition module is used to acquire a denoising network training data set, wherein the denoising network training data set includes multiple sets of denoising training data, wherein one set of denoising training data includes a correspondence between an original noisy DR image, noise parameters, sensor parameters, and a denoised DR image; The noise parameters are used to represent noise-related data obtained when acquiring the original noisy DR image; the sensor parameters are used to represent performance data of the sensor used to acquire the original noisy DR image and motion data acquired by the sensor during the DR three-dimensional scanning process; A target denoising network model establishment module is used to train an initial denoising network model using the denoising network training data set to obtain a target denoising network model; the target denoising network model includes motion estimation parameters of the DR generated based on the training, and the motion estimation parameters of the DR are used to represent the error value between the actual motion position of the DR during the three-dimensional scanning process and the preset motion position; The first DR image acquisition module is used to input the DR image to be processed into the target denoising network model to obtain a first DR image, wherein the noise in the first DR image is smaller than the noise in the DR image to be processed.

7. The system according to claim 6, characterized in that Also includes: A first image network model building module is used to train the target denoising network model using an attention module and a perceptual loss module to obtain a first image network model; The second DR image acquisition module is used to input the first DR image into the first image network model to obtain a second DR image; wherein the resolution of the second DR image is greater than the resolution of the first DR image.

8. The system according to claim 7, characterized in that Also includes: a second image network model building module, configured to train the first image network model using three low dynamic range images with adjacent frame numbers of underexposed, normally exposed, and overexposed to obtain a second image network model; The third DR image acquisition module is used to input the second DR image into the second image network model to obtain a third DR image; wherein the dynamic range of the third DR image is greater than the dynamic range of the second DR image.

9. An X-ray three-dimensional scanning imaging device, characterized in that: include: A memory and a processor, wherein the memory is used to store a computer program; The processor is configured to implement the steps of the X-ray three-dimensional scanning imaging method according to any one of claims 1 to 5 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the X-ray three-dimensional scanning imaging method according to any one of claims 1 to 5 are implemented.

Images