Image generation apparatus, medical apparatus, and storage medium

Abstract

[Problem] To provide a virtual monochromatic X-ray image with reduced image quality degradation. [Solution] A CT system (100) having one or more processors to perform operations including: reconstructing a sequence A (a1 to az) of virtual monochromatic X-ray images at a first energy (50 keV) and a sequence B (b1 to bz) of virtual monochromatic X-ray images at a second energy (70 keV) based on data collected from an imaged subject (112); inputting an input image to a trained model 71, the input image generated based on the reconstructed virtual monochromatic X-ray images at the second energy (70 keV) (b1 to bz), and inferring a sequence C (c1 to cz) of virtual monochromatic X-ray images at the first energy (50 keV) using the trained model 71; and generating a first energy virtual monochromatic X-ray image correction sequence D (d1 to dz) based on the reconstructed sequence A (a1 to az) of virtual monochromatic X-ray images at the first energy (50 (keV)) and the inferred sequence C (c1 to cz) of virtual monochromatic X-ray images at the first energy (50 (keV)).

Description

[Technical Field]

[0001] The present invention relates to an image generation apparatus for generating virtual monochrome X-ray images of an imaging subject, a medical device including the image generation apparatus, and a storage medium containing instructions for generating virtual monochrome X-ray images of an imaging subject. [Background Technology]

[0002] CT scanners are known as non-invasive medical devices for imaging specimens. Because they can acquire tomographic images of specimens in a short scan time, they are widely used in hospitals and other medical facilities.

[0003] A CT scanner generates X-rays by applying a specified voltage to the cathode-anode tube of an X-ray tube. The generated X-rays penetrate the specimen and are detected by a detector. The CT scanner then reconstructs a CT image of the specimen based on the data detected by the detector.

[0004] [List of Citations]

[0005] [Patent Literature]

[0006] Patent Document 1: Japanese Patent Publication No. 6031618 [Summary of the Invention]

[0007] [Technical Issues]

[0008] Single-energy CT (SECT) is a well-known imaging technique used in CT scanners. SECT is a method of obtaining CT images of a specimen by applying a specified voltage (e.g., 120 kVp) to the cathode-anode tube of an X-ray tube to generate X-rays. However, in SECT, CT values ​​can be similar even for different substances, and the identification of different substances can be difficult.

[0009] Therefore, DECT (Dual-Energy CT) technology is being researched and developed. DECT is a technique that uses X-rays in different energy regions to distinguish materials, and DECT-compatible CT devices are commercially available. DECT technology has a wide range of applications and can be used, for example, to reconstruct virtual monochromatic X-ray images at each energy level.

[0010] On the other hand, many medical institutions have not yet installed DECT-compatible CT equipment. Therefore, in order to provide diagnostics using virtual monochrome X-ray images even to medical institutions that do not have DECT-compatible CT equipment, techniques for inferring virtual monochrome X-ray images from images acquired via SECT using deep learning have been researched and developed.

[0011] However, for virtual monochromatic X-ray images reconstructed using DECT technology, there is a problem: when creating low-energy virtual monochromatic X-ray images, the image quality deteriorates due to the inherent characteristics of DECT (e.g., increased noise and deterioration of image texture).

[0012] Therefore, a technique is desired that can provide virtual monochrome X-ray images while reducing image quality degradation.

[0013] [Solution to the problem]

[0014] A first aspect of the present invention is an image generation apparatus comprising one or more processors performing operations including:

[0015] Based on data collected from the imaging subjects, a virtual monochromatic X-ray image of the first energy and a virtual monochromatic X-ray image of the second energy are reconstructed.

[0016] The input image is fed into the training model; the input image is created based on the reconstructed virtual monochromatic X-ray image of the second energy.

[0017] Using a trained model to infer a virtual monochromatic X-ray image of the first energy; and

[0018] Based on the reconstructed virtual monochromatic X-ray image of the first energy and the inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated.

[0019] Furthermore, a second aspect of the present invention is a medical device comprising one or more processors performing operations including:

[0020] Based on data collected from the imaging subjects, a virtual monochromatic X-ray image of the first energy and a virtual monochromatic X-ray image of the second energy are reconstructed.

[0021] The input image is fed into the training model; the input image is created based on the reconstructed virtual monochromatic X-ray image of the second energy.

[0022] Using a trained model to infer a virtual monochromatic X-ray image of the first energy; and

[0023] Based on the reconstructed virtual monochromatic X-ray image of the first energy and the inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated.

[0024] A third aspect of the invention is a storage medium readable by a computer and non-transitory containing one or more instructions executable by one or more processors, wherein...

[0025] One or more instructions instruct one or more processors to perform operations including the following:

[0026] Based on data collected from the imaging subjects, a virtual monochromatic X-ray image of the first energy and a virtual monochromatic X-ray image of the second energy are reconstructed.

[0027] The input image is created based on the reconstructed virtual monochromatic X-ray image of the second energy.

[0028] The input image is fed into the training model, and the training model is used to infer a virtual monochromatic X-ray image of first energy; and

[0029] Based on the reconstructed virtual monochromatic X-ray image of the first energy and the inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated.

[0030] [Advantages of the Invention]

[0031] When a virtual monochromatic X-ray image of a first energy is inferred based on a reconstructed virtual monochromatic X-ray image of a second energy, the inferred virtual monochromatic X-ray image of the first energy tends to be affected by the image quality of the virtual monochromatic X-ray image of the second energy. Therefore, by using an image with better image quality than the reconstructed virtual monochromatic X-ray image of the first energy as the reconstructed virtual monochromatic X-ray image of the second energy, the inferred virtual monochromatic X-ray image of the first energy can inherit the good image quality of the virtual monochromatic X-ray image of the second energy. Considering this, by inferring a virtual monochromatic X-ray image of the first energy based on the reconstructed virtual monochromatic X-ray image of the second energy, a virtual monochromatic X-ray image of the first energy with lower image quality can be inferred from the reconstructed virtual monochromatic X-ray image of the first energy with improved quality.

[0032] Therefore, the present invention has the ability to infer a virtual monochromatic X-ray image of a first energy, which has improved quality for low-quality regions in the reconstructed virtual monochromatic X-ray image of the first energy. Thus, by generating a corrected virtual monochromatic X-ray image of the first energy based on the reconstructed virtual monochromatic X-ray image of the first energy and the inferred virtual monochromatic X-ray image of the first energy, a virtual monochromatic X-ray image of the first energy with improved image quality can be obtained. [Image Description]

[0033] Figure 1 This is a perspective view of the CT system 100 according to implementation scheme 1.

[0034] Figure 2 This is a block diagram of the CT system 100 in this implementation scheme.

[0035] Figure 3 This is a flowchart illustrating the basic idea of ​​a technique for acquiring virtual monochrome X-ray images with improved image quality.

[0036] Figure 4 This is a diagram illustrating the process of acquiring a virtual monochromatic X-ray image with a first energy E1 that has improved quality.

[0037] Figure 5 This is an illustrative diagram illustrating the method for generating the correction sequence D (d1 to dz) of the virtual monochromatic X-ray image at the first energy E1 (=50keV) in step ST4.

[0038] Figure 6 This is an illustration of the method for determining areas of high probability of soft tissue in step ST42.

[0039] Figure 7 This is a diagram illustrating an example of step ST45.

[0040] Figure 8 This is an illustration of the method for determining the weighting coefficients.

[0041] Figure 9 This is a flowchart of step ST45 in implementation scheme 2.

[0042] Figure 10 This is a diagram showing a virtual monochrome X-ray image reconstructed by DECT and a virtual monochrome X-ray image generated by means of the present invention. [Detailed Implementation]

[0043] The following describes embodiments for carrying out the present invention, but the present invention is not limited to the following embodiments.

[0044] Figure 1 This is a perspective view of the CT system 100 in implementation scheme 1. Figure 2 This is a block diagram of the CT system 100 in this embodiment.

[0045] The CT system 100 has a gantry 102. The gantry 102 has an opening through which an imaging subject 112 is conveyed for scanning.

[0046] The rack 102 is equipped with an X-ray tube 104, a filter section 103, a front collimator 105, and a detector array 108.

[0047] When a specified voltage is applied to the cathode-anode tube, the X-ray tube 104 generates X-rays. In embodiment 1, the X-ray tube 104 is an X-ray tube compatible with a fast kV switching system.

[0048] The filter section 103 includes, for example, a planar filter and / or a bowtie filter.

[0049] The precollimator 105 is a component that narrows the X-ray irradiation range so that X-rays are not emitted in unwanted areas.

[0050] The detector array 108 includes a plurality of detector elements 202. The plurality of detector elements 202 detect an X-ray beam 106 emitted from an X-ray tube 104 and passing through an imaging subject 112, such as a patient. Thus, the detector array 108 can acquire projection data for each view.

[0051] Projection data detected by detector array 108 is collected by DAS 214. DAS 214 performs prescribed processing on the collected projection data, including sampling and digital conversion. The processed projection data is sent to computer 216. Computer 216 stores the data from DAS 214 in storage device 218. Storage device 218 includes one or more storage media storing programs and instructions to be executed by a processor. The storage media may be, for example, one or more non-transitory computer-readable storage media. Storage device 218 may include, for example, hard disk drives, floppy disk drives, optical disc read / write (CD-R / W) drives, digital versatile disc (DVD) drives, flash memory drives, and / or solid-state storage drives.

[0052] Processor 216 includes one or more processors. Computer 216 uses one or more processors to output commands and parameters to DAS 214, X-ray controller 210 and / or rack motor controller 212 to control system operations such as data acquisition and / or processing.

[0053] Operator console 220 is linked to computer 216. The operator can input prescribed operator input related to the operation of CT system 100 into computer 216 via operator console 220. Computer 216 receives operator input, including commands and / or scan parameters, via operator console 220 and controls system operation based on this operator input. Operator console 220 may include a keyboard (not shown) or a touchscreen for the operator to specify commands and / or scan parameters.

[0054] X-ray controller 210 controls X-ray tube 104 based on control signals from computer 216. In embodiment 1, X-ray controller 210 can control X-ray tube 104 to emit an X-ray beam from X-ray tube 104 while rapidly switching the voltage applied to the cathode-anode tube of X-ray tube 104 between a first voltage and a second voltage. Additionally, rack motor controller 212 controls rack motors based on control signals from computer 216.

[0055] Figure 2 Only one operator console 220 is shown, but two or more operator consoles can be linked to computer 216.

[0056] In addition, the CT system 100 can also allow multiple remotely positioned monitors, printers, workstations and / or similar devices to be linked via, for example, wired and / or wireless networks.

[0057] In one implementation, for example, the CT system 100 may include or be linked to a Picture Archiving and Communication System (PACS) 224. In a typical implementation, the PACS 224 may be linked to remote systems, such as radiology information systems, hospital information systems, and / or internal or external networks (not shown).

[0058] Computer 216 provides commands to inspection table motor controller 118 to control inspection table 116. Inspection table motor controller 118 can control inspection table 116 based on the received commands. In particular, inspection table motor controller 118 can move inspection table 116 so that the imaging subject 112 is properly positioned within the opening of rack 102.

[0059] As previously described, DAS 214 samples and digitally converts the projection data acquired by detector element 202. Then, image reconstruction unit 230 reconstructs the image using the sampled and digitally converted data. Image reconstruction unit 230 includes one or more processors capable of performing image reconstruction processing. Figure 2 In this design, the image reconstruction unit 230 is shown as a separate structural element from the computer 216, but the image reconstruction unit 230 can be integrated into the computer 216. Furthermore, the computer 216 can also perform one or more functions of the image reconstruction unit 230. Additionally, the image reconstruction unit 230 can be located remotely from the CT system 100 and can be operably connected to the CT system 100 using a wired or wireless network. The computer 216 and the image reconstruction unit 230 serve as an image generation apparatus.

[0060] The image reconstruction unit 230 can store the reconstructed image in the storage device 218. The image reconstruction unit 230 can also send the reconstructed image to the computer 216. The computer 216 can send the reconstructed image and / or patient information to a display 232 communicatively linked to the computer 216 and / or the image reconstruction unit 230.

[0061] The various methods and processes described in this specification may be stored as executable instructions in a non-transitory storage medium within the CT system 100. The executable instructions may be stored on a single storage medium or distributed across multiple storage media. One or more processors located in the CT system 100 execute the various methods, steps, and processes described in this specification according to the instructions stored on the storage medium.

[0062] The CT system 100 is configured as described above.

[0063] The CT system 100 can perform imaging using DECT technology. In this embodiment, the CT system 100 is configured to perform DECT imaging using a fast kV switching method. DECT is a technology that uses X-rays in different energy regions to distinguish materials and has a wide range of applications; for example, it can be used to reconstruct virtual monochromatic X-ray images at each energy. Virtual monochromatic X-ray images have the advantages of suppressing beam hardening artifacts and reducing metal artifacts. Therefore, an increasing number of medical institutions are installing CT systems equipped with DECT technology each year.

[0064] However, some medical institutions have not yet installed DECT-compatible CT equipment. Therefore, in order to provide diagnostics using virtual monochrome X-ray images even to medical institutions that do not have DECT-compatible CT equipment, techniques for inferring virtual monochrome X-ray images from images acquired via SECT using deep learning have been researched and developed.

[0065] In this way, DECT technology is very useful in the diagnosis of patients. On the other hand, DECT technology also involves various challenges. For example, DECT technology can reconstruct virtual monochromatic X-ray images of different energies, but when reconstructing virtual monochromatic X-ray images of lower energies, the principle of DECT leads to increased noise.

[0066] Therefore, the inventors of this application conducted in-depth research and conceived of generating high-quality virtual monochrome X-ray images. The basic concept for obtaining high-quality virtual monochrome X-ray images in this embodiment is described below.

[0067] As described above, techniques for using deep learning to infer virtual monochrome X-ray images have been researched and developed to provide diagnostic services using virtual monochrome X-ray images, even in medical institutions that have not yet installed DECT-compatible CT equipment. For example, a technique has been researched and developed in which a neural network is trained using data collected by DECT to create a trained neural network, and this trained neural network is used to infer virtual monochrome X-ray images. In researching techniques for inferring virtual monochrome X-ray images using trained neural networks, the inventors of this application noted characteristic trends in the image quality of virtual monochrome X-ray images.

[0068] Specifically, the inventors have discovered that virtual monochromatic X-ray images inferred through a trained neural network tend to inherit features of the image quality (e.g., image texture and image noise) of the input image fed into the trained neural network. In this embodiment, this knowledge is used to improve the image quality of the virtual monochromatic X-ray image. References are provided below. Figure 3 This section explains the basic concepts of a method for obtaining virtual monochrome X-ray images with improved image quality.

[0069] Figure 3 This is a flowchart illustrating the basic idea of ​​a technique for acquiring virtual monochrome X-ray images with improved image quality.

[0070] In step ST1, DECT technology is used to scan and image the subject. Scans using DECT technology can be performed, for example, using a fast kV switching method.

[0071] In step ST2, based on the data acquired in step ST1, virtual monochromatic X-ray images with multiple energies Ei (i = 1 to n) are reconstructed. Here, as virtual monochromatic X-ray images with multiple energies Ei (i = 1 to n), we consider sequence A (a1 to az) of virtual monochromatic X-ray images reconstructed at the first energy E1 (i = 1) and sequence B (b1 to bz) of virtual monochromatic X-ray images reconstructed at the second energy E2. The first energy E1 is, for example, 50 (keV), and the second energy E2 is, for example, 70 (keV).

[0072] The sequence A (a1 to az) of virtual monochromatic X-ray images at the first energy E1 is a low-energy virtual monochromatic X-ray image, and its image quality is prone to degradation according to the principles of DECT. On the other hand, among the virtual monochromatic X-ray images reconstructed by DECT at multiple energies Ei (i = 1 to n), there are virtual monochromatic X-ray images at a second energy that exhibit less image quality degradation compared to lower-energy virtual monochromatic X-ray images such as 50 (keV). For example, it is known that a virtual monochromatic X-ray image at the second energy E2 of 70 (keV) has better image quality than a virtual monochromatic X-ray image at the first energy E1 of 50 (keV).

[0073] On the other hand, as described above, the inventors have discovered that virtual monochromatic X-ray images inferred through a trained neural network tend to inherit features of the image quality (e.g., image texture and image noise) of the input image fed into the trained neural network. Therefore, if deep learning techniques are used, in the case of inferring a sequence C (c1 to cz) of a virtual monochromatic X-ray image of first energy E1 based on a sequence B (b1 to bz) of a virtual monochromatic X-ray image of second energy E2 (70 keV) with relatively low image quality degradation, the inferred image sequence C is considered to inherit the image quality of the image sequence B with less image quality degradation. Therefore, regions depicted as having low image quality in the reconstructed sequence A (a1 to az) of the virtual monochromatic X-ray image of first energy E1 can be depicted as regions with improved image quality in the inferred sequence C (c1 to cz) of the virtual monochromatic X-ray image of first energy E1.

[0074] Therefore, in step ST3 of this embodiment, deep learning technology is used to infer the sequence of virtual monochromatic X-ray images C (c1 to cz) of the first energy E1 based on the sequence B (b1 to bz) of the virtual monochromatic X-ray images of the second energy E2 (70keV).

[0075] Next, in step ST4, based on the reconstructed sequence A (a1 to az) of the virtual monochromatic X-ray image of the first energy E1 and the inferred sequence C (c1 to cz) of the virtual monochromatic X-ray image of the first energy E1, a correction sequence D (d1 to dz) of the virtual monochromatic X-ray image of the first energy E1 is generated. Therefore, it is considered that the sequence D (d1 to dz) of the virtual monochromatic X-ray image of the first energy E1 can be obtained with improved quality.

[0076] The inventors have considered obtaining virtual monochromatic X-ray images of first energy with improved quality based on the above principles.

[0077] The following describes the process for obtaining a virtual monochromatic X-ray image with improved quality at a first energy E1 based on the principles described above.

[0078] Figure 4 The procedure for acquiring a virtual monochromatic X-ray image with improved quality at a first energy E1 is shown.

[0079] Additionally, as referenced Figure 3As explained, in this embodiment, since the virtual monochromatic X-ray image of the first energy E1 is inferred from the virtual monochromatic X-ray image of the second energy E2, a training model capable of performing this inference needs to be prepared in advance. Therefore, firstly, the training phase for generating this training model is described below. Furthermore, after describing the training phase, the process for obtaining a high-quality virtual monochromatic X-ray image of the first energy based on the above principles is described.

[0080] ( Training phase )

[0081] During the training phase, the first step is to prepare the original images that form the basis for generating the training data.

[0082] In Implementation Scheme 1, virtual monochromatic X-ray images of different energies Ei (i = 1 to n) obtained by DECT are prepared as the original images.

[0083] In this implementation, we will consider two energies, a first energy E1 (i=1) and a second energy E2 (i=2), as the energy Ei of the virtual monochromatic X-ray image. The first energy E1 is 50 (keV), and the second energy E2 is 70 (keV). Therefore, we prepare a set V of virtual monochromatic X-ray images with the first energy E1 (=50 (keV)) and a set W of virtual monochromatic X-ray images with the second energy E2 (=70 (keV)) as the original image. Alternatively, we can prepare three or more virtual monochromatic X-ray images with energies E1 to En (n≥3) as the original image.

[0084] Next, as Figure 4 As shown, preprocessing is performed on these virtual monochrome X-ray image sets V and W 68.

[0085] The preprocessing 68 includes, for example, image cropping, normalization, image inversion, image rotation, magnification percentage adjustment, and image quality modification. By preprocessing the original image sets V and W, preprocessed virtual monochrome X-ray image sets VA and WA can be obtained. The preprocessed virtual monochrome X-ray image sets VA and WA are used as training data 60 to create a training model.

[0086] Next, training data 60 is used to train neural network 70. Neural network 70 can be, for example, a convolutional neural network. In this embodiment, neural network 70 is trained to output a virtual monochromatic X-ray image with energy of 50 (keV) based on a virtual monochromatic X-ray image with energy of 70 (keV), and a training model 71 is created.

[0087] After creating the trained model 71, an evaluation of the trained model 71 is performed. For example, this evaluation can use a confusion matrix. For example, accuracy can be used as an evaluation metric.

[0088] If the evaluation is good, the trained model 71 described above is used as the model for performing the following inference (step ST3). If the evaluation is poor, the training phase is performed again, for example, by adding more training data.

[0089] The training model 71 is stored in the storage device 218 of the CT system 100. The training model 71 can also be stored on an external storage device accessible to the CT system 100.

[0090] In this implementation, the trained model 71 is used to generate high-quality virtual monochrome X-ray images. See below for reference. Figure 4 The flowchart on the right describes the process of generating high-quality virtual monochrome X-ray images using the trained model 71.

[0091] In step ST1, DECT technology is used to scan and image the subject. Scans using DECT technology can be performed, for example, using a fast kV switching method.

[0092] In step ST2, virtual monochromatic X-ray images of multiple energies Ei (i = 1 to n) are reconstructed based on the data obtained in step ST1.

[0093] The data obtained from the scan is from DAS 214 (see...) Figure 2 The collected data is sent to computer 216 or image reconstruction unit 230. On computer 216 or image reconstruction unit 230, the processor reconstructs virtual monochromatic X-ray images of different energies based on the data obtained from the scan. In embodiment 1, a sequence A (a1 to az) of virtual monochromatic X-ray images of a first energy E1 and a sequence B (b1 to bz) of virtual monochromatic X-ray images of a second energy E2 are reconstructed. Here, the first energy E1 is 50 (keV) and the second energy E2 is 70 (keV), but other energy values ​​can also be used. After reconstructing these virtual monochromatic X-ray image sequences A and B, the process proceeds to step ST3.

[0094] In step ST3, computer 216 performs an inference phase, inferring a sequence C (c1 to cz) of virtual monochromatic X-ray images of the first energy E1 (50 keV) based on a sequence B (b1 to bz) of virtual monochromatic X-ray images of the second energy E2 (=70 keV).

[0095] During the inference phase, computer 216 preprocesses each virtual monochromatic X-ray image b1 to bz of the second energy E2 (=70keV) to generate input images that are input to training model 71.

[0096] Next, computer 216 inputs the input image into training model 71 and uses training model 71 to infer a sequence C (c1 to cz) of virtual monochromatic X-ray images of the first energy E1 (=50keV). After inferring the image sequence C, the process proceeds to step ST4.

[0097] In step ST4, computer 216 generates a correction sequence D (d1 to dz) of the virtual monochromatic X-ray image of the first energy E1 (=50 kev) based on the reconstructed sequence A (a1 to az) of the virtual monochromatic X-ray image of the first energy E1 (=50 kev) and the inferred sequence C (c1 to cz) of the virtual monochromatic X-ray image of the first energy E1 (=50 kev).

[0098] The following is for reference Figure 5 The generation method is explained.

[0099] Figure 5 This is an illustrative diagram illustrating the method for generating the correction sequence D (d1 to dz) of the virtual monochromatic X-ray image at the first energy E1 (=50keV) in step ST4.

[0100] In step ST41, computer 216 extracts a single virtual monochromatic X-ray image aj from the reconstructed sequence A (a1 to az) of the virtual monochromatic X-ray images of the first energy E1 (=50keV). Here, it is assumed that j=1, or in other words, the computer retrieves the virtual monochromatic X-ray image a1. After retrieving the virtual monochromatic X-ray image a1, the process proceeds to step ST42.

[0101] In step ST42, computer 216 identifies possible soft tissue regions in the virtual monochrome X-ray image a1.

[0102] Figure 6 This is an illustration of the method used in step ST42 to identify tissue regions that may be soft tissue.

[0103] Figure 6Two thresholds TH1 and TH2 are shown, which are criteria for determining whether a pixel CT value is a soft tissue CT value. The threshold TH1 represents the lower limit of the CT value of soft tissue, and the threshold TH2 represents the upper limit of the CT value of soft tissue. Therefore, the range G1 where TH1≤v≤TH2 represents the range that may contain soft tissue CT values, and the range G2 where v<TH1 or TH2<v represents the range that is unlikely to contain soft tissue CT values. Soft tissue CT values are usually included in the range of -150 (HU) to 150 (HU). Accordingly, TH1 and TH2 can be determined, for example, TH1=-150 (HU) and TH2=150 (HU).

[0104] The computer 216 compares the CT value v of each pixel of the virtual monochromatic X-ray image a1 with the thresholds TH1 and TH2, so as to determine for each pixel whether the CT value v is included in the range G1 of TH1≤v≤TH2 or included in the range G2 of v<TH1 or TH2<v.

[0105] In addition, for ease of description, let us assume that the CT values v of the pixels P11 to P1a in the region R1 of the virtual monochromatic X-ray image a1 satisfy TH1≤v≤TH2, and the CT values v of the pixels P21 to P2b in the region R2 satisfy v<TH1 or TH2<v.

[0106] Therefore, the computer 216 identifies, from the virtual monochromatic X-ray image a1, the region R1 including the pixels P11 to P1a having the CT value v in the range G1 of TH1≤v≤TH2 as the first region with a high possibility of containing soft tissue. On the other hand, the computer 216 identifies, from the virtual monochromatic X-ray image a1, the region R2 including the pixels P21 to P2b having CT values in the range G2 of v<TH1 or TH2<v as the second region with a low possibility of containing soft tissue.

[0107] Therefore, by performing the processing of step ST42, the pixels P11 to P1a that are more likely to belong to soft tissue and the pixels P21 to P2b that are less likely to belong to soft tissue can be identified in the virtual monochromatic X-ray image a1.

[0108] Next, in step ST43, the computer 216 retrieves the virtual monochromatic X-ray image c1 corresponding to the same cross-sectional image as the virtual monochromatic X-ray image a1 from the inferred sequence C (c1 to cz) of virtual monochromatic X-ray images of the first energy E1 (=50 keV). After the virtual monochromatic X-ray image c1 is retrieved, the process proceeds to step ST44.

[0109] In step ST44, computer 216 identifies regions in virtual monochrome X-ray image c1 that may be soft tissue. Specifically, computer 216 identifies region R3 in virtual monochrome X-ray image c1 that corresponds to region R1 in virtual monochrome X-ray image a1 identified in step ST42 as a third region with a high probability of being soft tissue. Additionally, computer 216 identifies region R4 in virtual monochrome X-ray image c1 that corresponds to region R2 in virtual monochrome X-ray image a1 identified in step ST42 as a fourth region with a low probability of being soft tissue.

[0110] Therefore, by performing the processing in step ST44, it is possible to identify pixels P31 to P3a that are more likely to belong to soft tissue and pixels P41 to P4b that are less likely to belong to soft tissue in the virtual monochrome X-ray image c1.

[0111] Computer 216 then proceeds to step ST45. In step ST45, a corrected virtual monochromatic X-ray image with first energy d1 is generated based on the reconstructed virtual monochromatic X-ray image a1 with first energy E1 (=50keV) and the inferred virtual monochromatic X-ray image c1 with first energy E1 (=50keV). An example of the flow of step ST45 is described below.

[0112] Figure 7 An example of step ST45 is shown.

[0113] Step ST45 includes steps ST451, ST452, and ST453. Each step is described below in turn.

[0114] In step ST451, the weighting coefficients of the CT value v of each pixel in the virtual monochrome X-ray images a1 and c1 are determined.

[0115] Figure 8 This is an explanatory diagram illustrating how the weighting coefficients are determined.

[0116] First, the weighting coefficient W1 of the CT value v of each pixel in region R1 of the reconstructed virtual monochromatic X-ray image a1 of the first energy E1 (=50keV) and the weighting coefficient W3 of the CT value v of each pixel in region R3 of the inferred virtual monochromatic X-ray image c1 of the first energy E1 (=50keV) are explained.

[0117] In the reconstructed virtual monochromatic X-ray image a1 at the first energy E1 (=50keV), pixels P11 to P1a in region R1 may correspond to soft tissue pixels because the CT value v is included in TH1≤v≤TH2. In the low-energy virtual monochromatic X-ray image a1, the image quality of soft tissue tends to deteriorate easily, so region R1 may be depicted with low image quality.

[0118] On the other hand, the inferred virtual monochromatic X-ray image c1 of the first energy E1 (=50keV) inherits the virtual monochromatic X-ray image b1 of the second energy E2 (=70keV) (see Figure 4 Therefore, the image quality of region R3 in the inferred virtual monochromatic X-ray image c1 at the first energy E1 (=50keV) is considered to be better than that of region R1 in the reconstructed virtual monochromatic X-ray image a1 at the first energy E1 (=50keV).

[0119] Therefore, to improve the image quality of the soft tissue region, pixels P11 to P1a in region R1 of the reconstructed virtual monochromatic X-ray image a1 at the first energy E1 (=50keV) are assigned a weighting coefficient W1 to reduce the weighting of the CT value, considering that they are more likely to have lower image quality. On the other hand, considering that pixels P31 to P3a in region R3 of the inferred virtual monochromatic X-ray image c1 at the first energy E1 (=50keV) have better image quality than region R1 of the virtual monochromatic X-ray image a1, a weighting coefficient W3 is assigned to increase the weighting of the CT value.

[0120] Therefore, the weighting coefficient W3 of pixels P31 to P3a in region R3 of the inferred virtual monochromatic X-ray image c1 is assigned a larger value than the weighting coefficient W1 of pixels P11 to P1a in region R1 of the reconstructed virtual monochromatic X-ray image a1. Specifically, under the condition W1 + W3 = 1, the values ​​of W1 and W3 are set such that W3 > W1. For example, W1 = 0.3 and W3 = 0.7. In embodiment 1, W1 + W3 = 1, but the value of W1 + W3 can also be set to a value deviating from "1".

[0121] Next, the weighting coefficient W2 of the CT value v of each pixel in region R2 of the reconstructed virtual monochromatic X-ray image a1 with first energy E1 (=50keV) and the weighting coefficient W4 of the CT value v of each pixel in region R4 of the inferred virtual monochromatic X-ray image c1 with first energy E1 (=50keV) will be explained.

[0122] Since the CT value v of the region R2 in the virtual monochromatic X-ray image a1 falls within the range of v<TH1 or TH2<v, it is considered that air, gas, bone, or the like is depicted in the region R2. Since the region R4 of the virtual monochromatic X-ray image c1 corresponds to the region R2 of the virtual monochromatic X-ray image a1, it is also considered that air, gas, bone, or the like is depicted in the region R4 of the virtual monochromatic X-ray image c1. In general, in regions where the CT value v of air, gas, bone, or the like is far from the CT value of water (0), it can be considered that the reconstructed virtual monochromatic X-ray image a1 at the first energy E1 (=50 keV) is more reliable than the virtual monochromatic X-ray image c1 at the first energy E1 (=50 keV) inferred by a deep learning technique. Therefore, a weighting coefficient W2 with a high weight for CT values is assigned to the pixels P21 to P2b in the region R2 of the reconstructed virtual monochromatic X-ray image a1, and a weighting coefficient W4 is assigned to the pixels P41 to P4b in the region R4 of the inferred virtual monochromatic X-ray image c1. For example, W2=0.7 and W4=0.3. Note that in Embodiment 1, W2+W4=1, but the value of W2+W4 may also be set to a value deviating from "1".

[0123] In this manner, the weighting coefficients W1, W2, W3, and W4 are determined. After the weighting coefficients are determined, the process proceeds to step ST452 (see Figure 7 ).

[0124] In step ST452, the computer 216 weights the CT values v of the reconstructed virtual monochromatic X-ray image a1 at the first energy E1 (=50 keV) with the weighting coefficients W1 and W2, and weights the CT values v of the inferred virtual monochromatic X-ray image c1 at the first energy E1 (=50 keV) with the weighting coefficients W3 and W4. Further, the process proceeds to step ST453, and the computer 216 adds the weighted virtual monochromatic X-ray images a1 and b1. In this way, a corrected virtual monochromatic X-ray image d1 at the first energy E1 (=50 keV) can be generated.

[0125] In the same manner, weighting coefficients are obtained for the other virtual monochromatic X-ray images a2 to az, c2 to cz, and corrected virtual monochromatic X-ray images are generated based on the weighting coefficients. Therefore, as Figure 4 shown, a sequence D (d1 to dz) of virtual monochromatic X-ray images at the first energy E1 with improved quality can be generated.

[0126] Thus, Figure 4 the process flow shown in the above is completed.

[0127] In implementation scheme 1, in step ST2, sequence A (a1 to az) of virtual monochromatic X-ray images of the first energy E1 and sequence B (b1 to bz) of virtual monochromatic X-ray images of the second energy E2 are reconstructed using DECT technology. However, due to the principle of DECT, the low-quality region R1 (see...) Figure 5 It tends to appear in low-energy virtual monochromatic X-ray image sequences A (a1 to az).

[0128] Therefore, in implementation scheme 1, the virtual monochromatic X-ray image is inferred by utilizing the fact that the virtual monochromatic X-ray image inferred from the trained neural network tends to inherit the image quality characteristics (e.g., image texture, image noise) of the input image input to the trained neural network. Specifically, in step ST3 (see... Figure 4 In this study, the sequence C (c1 to cz) of a virtual monochromatic X-ray image of the first energy E1 is inferred based on the sequence B (b1 to bz) of the second energy E2 (70 keV) virtual monochromatic X-ray image, which exhibits relatively less image quality degradation. The inferred sequence C (c1 to cz) of the first energy E1 virtual monochromatic X-ray image inherits the image quality of the sequence B (b1 to bz) of the second energy E2 (70 keV) virtual monochromatic X-ray image. As a result, region R3 (see [reference needed]) in the inferred sequence C (c1 to cz) of the first energy E1 virtual monochromatic X-ray image... Figure 8 The reconstructed virtual monochromatic X-ray image a1 to az contains region R1 (see [reference]) that has a higher energy than the first energy E1. Figure 8 Better image quality.

[0129] Following the inference, in step ST4, a correction sequence D (d1 to dz) of the virtual monochromatic X-ray image of first energy E1 is generated based on the reconstructed sequence A (a1 to az) of the virtual monochromatic X-ray image of first energy E1 and the inferred sequence C (c1 to cz) of the virtual monochromatic X-ray image of first energy E1. Specifically, weighting coefficients W1 and W2 are assigned to the reconstructed sequence A (a1 to az) of the virtual monochromatic X-ray image of first energy E1, such that the weight of the CT value in region R1 is reduced, while the weight of the CT value in region R2 is increased. On the other hand, weighting coefficients W3 and W4 are assigned to the inferred sequence C (c1 to cz) of the virtual monochromatic X-ray image of first energy E1, such that the weight of the CT value in region R3 is increased, while the weight of the CT value in region R4 is decreased. Therefore, a sequence D (d1 to dz) of the virtual monochromatic X-ray image of first energy E1 with improved quality can be obtained.

[0130] In Embodiment 1, in step ST42, in the virtual monochromatic X-ray image a1, the region R1 is identified as a region with high possibility of being soft tissue, and the region R2 is identified as a region with low possibility of being soft tissue. In addition, after step ST42 is executed, the process proceeds to ST44 to identify a region R3 with high possibility of being soft tissue and a region R4 with low possibility of being soft tissue in the virtual monochromatic X-ray image c1. However, step ST44 may be executed first, and step ST42 may be executed after step ST44 is executed. In this case, step ST44 and ST42 are executed as follows.

[0131] In step ST44, the CT value v of each pixel of the virtual monochromatic X-ray image c1 is compared with two thresholds TH1 and TH2 to determine whether the CT value v falls within the range G1 of TH1≤v≤TH2, or falls within the range G2 of v<TH1 or TH2<v. In addition, based on this determination result, a region R3 with high possibility of being soft tissue and a region R4 with low possibility of being soft tissue are identified for the virtual monochromatic X-ray image c1.

[0132] In step ST42, from the virtual monochromatic X-ray image a1, the region R1 corresponding to the region R3 in the virtual monochromatic X-ray image c1 is identified as a region that is likely to be soft tissue. Furthermore, from the virtual monochromatic X-ray image a1, the region R2 corresponding to the region R4 in the virtual monochromatic X-ray image c1 is identified as a region that is unlikely to be soft tissue.

[0133] Therefore, the regions R3 and R4 of the virtual monochromatic X-ray image c1 can be identified first, and the regions R1 and R2 of the virtual monochromatic X-ray image a1 can be identified based on the regions R3 and R4 of the virtual monochromatic X-ray image c1.

[0134] In Embodiment 1, the virtual monochromatic X-ray image of the first energy E1 is inferred based on the virtual monochromatic X-ray image of the second energy E2 which is larger than the first energy E1. However, if the virtual monochromatic X-ray image of the first energy E1 can be inferred with good image quality, the virtual monochromatic X-ray image of the first energy E1 may also be inferred based on the virtual monochromatic X-ray image of the second energy E2 which is smaller than the first energy E1.

[0135] In Embodiment 1, the weighting coefficients W1 to W4 are determined based on the thresholds TH1 and TH2. However, the weighting coefficients W1 to W4 may be determined in another manner. For example, the virtual monochromatic X-ray image a1 and / or b1 may be filtered (low-pass filtering or high-pass filtering), and the weighting coefficients W1 to W4 may be determined based on the filtered virtual monochromatic X-ray image.

[0136] (2) Implementation Plan 2

[0137] Implementation 2 describes an example of generating corrected virtual monochrome X-ray images in a different manner than in Implementation 1.

[0138] In Implementation Scheme 2, except for step ST45, the corrected virtual monochromatic X-ray image is generated using the same process as in Implementation Scheme 1. Therefore, for Implementation Scheme 2, step ST45 is mainly described.

[0139] Figure 9 This is a flowchart of step ST45 in implementation scheme 2.

[0140] In step ST41, computer 216 extracts a single virtual monochromatic X-ray image aj from the reconstructed sequence A (a1 to az) of the virtual monochromatic X-ray images of the first energy E1 (=50keV). Here, it is assumed that j=1, or in other words, the computer retrieves the virtual monochromatic X-ray image a1.

[0141] In step ST43, computer 216 retrieves a virtual monochromatic X-ray image c1 corresponding to the same cross-section of virtual monochromatic X-ray image a1 from the sequence C (c1 to cz) inferred from the virtual monochromatic X-ray image of the first energy E1 (=50keV). After retrieving the virtual monochromatic X-ray image c1, the process proceeds to step ST47.

[0142] In step ST47, computer 216 determines the difference image s1 between virtual monochromatic X-ray images a1 and c1.

[0143] Next, computer 216 identifies the pixel Ps with the largest pixel value (difference value) among the multiple pixels contained in the difference image s1.

[0144] In step ST48, computer 216 identifies pixel Psa corresponding to pixel Ps from among the multiple pixels of the reconstructed virtual monochrome X-ray image a1. Then, the CT value v (=va) of the identified pixel Psa is read out.

[0145] Computer 216 also identifies the pixel Psc corresponding to pixel Ps from among the multiple pixels in the inferred virtual monochrome X-ray image c1. Then, it reads out the CT value v (= vc) of the identified pixel Psc.

[0146] Next, in step ST49, computer 216 corrects the inferred virtual monochromatic X-ray image c1 based on the CT value v (=va) of pixel Psa and the CT value v (=vc) of pixel Psc. For example, gamma correction can be used as the correction method. The gamma (γ) used for gamma correction can be obtained from the CT value va of pixel Psa and the CT value vc of pixel Psc. In this way, the corrected virtual monochromatic X-ray image d1 can be generated.

[0147] In the same manner, differential images are generated for the other virtual monochromatic X-ray images a2 to az and c2 to cz, and corrected virtual monochromatic X-ray images are generated based on the differential images. Therefore, a sequence D(d1 to dz) of virtual monochromatic X-ray images with improved quality at a first energy can be generated (see...). Figure 4 ).

[0148] In implementation scheme 2, a corrected virtual monochromatic X-ray image d1 is generated by determining a difference image s1 and correcting the CT value of each pixel of the inferred virtual monochromatic X-ray image c1 based on the difference image s1. Therefore, the inferred virtual monochromatic X-ray image can be corrected by taking into account the difference in CT values ​​between the inferred virtual monochromatic X-ray image c1 and the reconstructed virtual monochromatic X-ray image a1.

[0149] In implementation scheme 2, a corrected virtual monochromatic X-ray image is generated based on the differential image s1. However, a corrected virtual monochromatic X-ray image d1 can also be generated based on the frequency components by retrieving the low-frequency components of the inferred virtual monochromatic X-ray image c1 and the high-frequency components of the reconstructed virtual monochromatic X-ray image a1.

[0150] The following is a comparison between a virtual monochrome X-ray image reconstructed by DECT and a virtual monochrome X-ray image generated by the method of this invention.

[0151] Figure 10 This is a diagram showing a virtual monochrome X-ray image reconstructed by DECT and a virtual monochrome X-ray image generated by means of the present invention.

[0152] Figure 10 (a) shows a 50 (keV) virtual monochromatic X-ray image reconstructed using DECT, and Figure 10 (b) A virtual monochromatic X-ray image generated by means of the present invention is shown. A comparison of the two images shows that the image quality of the virtual monochromatic X-ray image generated by means of the present invention is improved.

[0153] [List of reference numerals]

[0154] 60. Training Data

[0155] 68. Pretreatment

[0156] 70. Neural Networks

[0157] 71. Training the model

[0158] 100.CT System

[0159] 102. Rack

[0160] 103. Filter Section

[0161] 104. X-ray tube

[0162] 105. Front collimator

[0163] 106. X-ray beam

[0164] 108. Detector Array

[0165] 112. Imaging subject

[0166] 116. Inspection table

[0167] 118. Inspection table motor controller

[0168] 202. Detector Element

[0169] 210. X-ray controller

[0170] 212. Rack motor controller

[0171] 214.DAS

[0172] 216. Computer

[0173] 218. Storage device

[0174] 220. Operator Control Panel

[0175] 224. Image archiving communication system

[0176] 230. Image Reconstruction Unit

[0177] 232. Monitor

Claims

1. An image generating apparatus, the image generating apparatus comprising one or more processors performing operations, the operations including: Based on data collected from the imaging subject using dual-energy CT technology, a virtual monochromatic X-ray image of the first energy and a virtual monochromatic X-ray image of the second energy are reconstructed. An input image, created based on the reconstructed virtual monochromatic X-ray image of the second energy, is fed into the training model. The trained model is used to infer the virtual monochromatic X-ray image of the first energy; as well as Based on the reconstructed virtual monochromatic X-ray image of the first energy and the inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated, wherein generating the corrected virtual monochromatic X-ray image of the first energy includes: The CT value of each pixel in the reconstructed virtual monochromatic X-ray image of the first energy is weighted; The CT value of each pixel in the inferred virtual monochromatic X-ray image of the first energy is weighted; and Based on the weighted reconstructed virtual monochromatic X-ray image of the first energy and the weighted inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated. as well as A first region and a second region are identified from the virtual monochrome X-ray image of the first of the reconstructed virtual monochrome X-ray image and the inferred virtual monochrome X-ray image, wherein the first region includes pixels having CT values ​​in a first range and the second region includes pixels having CT values ​​in a second range. A third region and a fourth region are identified from the virtual monochromatic X-ray image of the second of the reconstructed virtual monochromatic X-ray image and the inferred virtual monochromatic X-ray image, wherein the third region corresponds to the first region and the fourth region corresponds to the second region; A first weighting coefficient is determined for weighting the CT values ​​of the pixels in the first region, a second weighting coefficient is determined for weighting the CT values ​​of the pixels in the second region, a third weighting coefficient is determined for weighting the CT values ​​of the pixels in the third region, and a fourth weighting coefficient is determined for weighting the CT values ​​of the pixels in the fourth region. The first virtual monochromatic X-ray image is weighted using the first weighting coefficient and the second weighting coefficient; The second virtual monochrome X-ray image is weighted using the third and fourth weighting coefficients; as well as A corrected virtual monochromatic X-ray image of the first energy is generated based on the first virtual monochromatic X-ray image weighted by the first weighting coefficient and the second weighting coefficient, and the second virtual monochromatic X-ray image weighted by the third weighting coefficient and the fourth weighting coefficient.

2. The image generation apparatus according to claim 1, wherein the first virtual monochromatic X-ray image is the reconstructed virtual monochromatic X-ray image, and the second virtual monochromatic X-ray image is the inferred virtual monochromatic X-ray image.

3. The image generating apparatus according to claim 1, wherein the first range represents a range of CT values ​​for soft tissue.

4. The image generation apparatus according to claim 3, wherein the first range is the range between a first threshold representing the lower limit of the CT value of soft tissue and a second threshold representing the upper limit of the CT value of soft tissue.

5. The image generation apparatus according to claim 4, wherein the first threshold is a negative value and the second threshold is a positive value.

6. The image generation apparatus of claim 1, wherein the one or more processors perform operations including: Filter the reconstructed virtual monochromatic X-ray image of the first energy and / or the inferred virtual monochromatic X-ray image of the first energy; and The first weighting coefficient and the second weighting coefficient are determined for weighting the CT values ​​of pixels in the reconstructed virtual monochromatic X-ray image of the first energy, and the third weighting coefficient and the fourth weighting coefficient are determined based on the filtered virtual monochromatic X-ray image for weighting the CT values ​​of pixels in the inferred virtual monochromatic X-ray image of the first energy.

7. The image generation apparatus of claim 1, wherein generating the corrected virtual monochromatic X-ray image of the first energy comprises: Generate a difference image between the reconstructed virtual monochromatic X-ray image of the first energy and the inferred virtual monochromatic X-ray image of the first energy; and A corrected virtual monochromatic X-ray image of the first energy is generated based on the differential image.

8. The image generation apparatus of claim 7, wherein generating the corrected virtual monochromatic X-ray image of the first energy based on the difference image comprises: Identify the pixel with the largest difference from a plurality of pixels in the difference image; Identify a first pixel corresponding to the identified pixel among a plurality of pixels in the reconstructed virtual monochromatic X-ray image of the first energy; Identify a second pixel corresponding to the identified pixel among multiple pixels in the inferred virtual monochromatic X-ray image of the first energy; as well as The inferred virtual monochromatic X-ray image of the first energy is corrected based on the CT value of the first pixel and the CT value of the second pixel.

9. The image generation apparatus of claim 8, wherein correcting the inferred virtual monochromatic X-ray image of the first energy comprises: Based on the CT value of the first pixel and the CT value of the second pixel, a virtual monochromatic X-ray image of the first energy is inferred by gamma correction.

10. The image generation apparatus of claim 1, wherein the training model is generated by training a neural network using training data, the training data comprising images obtained by preprocessing a virtual monochromatic X-ray image of the first energy and a virtual monochromatic X-ray image of the second energy.

11. The image generating apparatus according to claim 1, wherein the first energy is lower than the second energy.

12. A medical device comprising one or more processors capable of performing operations, the operations including: Based on data collected from the imaging subject using dual-energy CT technology, a virtual monochromatic X-ray image of the first energy and a virtual monochromatic X-ray image of the second energy are reconstructed. An input image, created based on the reconstructed virtual monochromatic X-ray image of the second energy, is fed into the training model. The trained model is used to infer the virtual monochromatic X-ray image of the first energy; as well as Based on the reconstructed virtual monochromatic X-ray image of the first energy and the inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated, wherein generating the corrected virtual monochromatic X-ray image of the first energy includes: The CT value of each pixel in the reconstructed virtual monochromatic X-ray image of the first energy is weighted; The CT value of each pixel in the inferred virtual monochromatic X-ray image of the first energy is weighted; and Based on the weighted reconstructed virtual monochromatic X-ray image of the first energy and the weighted inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated. as well as A first region and a second region are identified from the virtual monochrome X-ray image of the first of the reconstructed virtual monochrome X-ray image and the inferred virtual monochrome X-ray image, wherein the first region includes pixels having CT values ​​in a first range and the second region includes pixels having CT values ​​in a second range. A third region and a fourth region are identified from the virtual monochromatic X-ray image of the second of the reconstructed virtual monochromatic X-ray image and the inferred virtual monochromatic X-ray image, wherein the third region corresponds to the first region and the fourth region corresponds to the second region; A first weighting coefficient is determined for weighting the CT values ​​of the pixels in the first region, a second weighting coefficient is determined for weighting the CT values ​​of the pixels in the second region, a third weighting coefficient is determined for weighting the CT values ​​of the pixels in the third region, and a fourth weighting coefficient is determined for weighting the CT values ​​of the pixels in the fourth region. The first virtual monochromatic X-ray image is weighted using the first weighting coefficient and the second weighting coefficient; The second virtual monochrome X-ray image is weighted using the third and fourth weighting coefficients; as well as A corrected virtual monochromatic X-ray image of the first energy is generated based on the first virtual monochromatic X-ray image weighted by the first weighting coefficient and the second weighting coefficient, and the second virtual monochromatic X-ray image weighted by the third weighting coefficient and the fourth weighting coefficient.

13. A storage medium readable by a computer and non-transitory containing one or more instructions executable by one or more processors, wherein... The one or more instructions instruct the one or more processors to perform the following operations: Based on data collected from the imaging subject using dual-energy CT technology, a virtual monochromatic X-ray image of the first energy and a virtual monochromatic X-ray image of the second energy are reconstructed. An input image is created based on the reconstructed virtual monochromatic X-ray image of the second energy; The input image is fed into a training model and the training model is used to infer a virtual monochromatic X-ray image of the first energy; as well as Based on the reconstructed virtual monochromatic X-ray image of the first energy and the inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated, wherein generating the corrected virtual monochromatic X-ray image of the first energy includes: The CT value of each pixel in the reconstructed virtual monochromatic X-ray image of the first energy is weighted; The CT value of each pixel in the inferred virtual monochromatic X-ray image of the first energy is weighted; and Based on the weighted reconstructed virtual monochromatic X-ray image of the first energy and the weighted inferred virtual monochromatic X-ray image of the first energy, a corrected virtual monochromatic X-ray image of the first energy is generated. as well as A first region and a second region are identified from the virtual monochrome X-ray image of the first of the reconstructed virtual monochrome X-ray image and the inferred virtual monochrome X-ray image, wherein the first region includes pixels having CT values ​​in a first range and the second region includes pixels having CT values ​​in a second range. A third region and a fourth region are identified from the virtual monochromatic X-ray image of the second of the reconstructed virtual monochromatic X-ray image and the inferred virtual monochromatic X-ray image, wherein the third region corresponds to the first region and the fourth region corresponds to the second region; A first weighting coefficient is determined for weighting the CT values ​​of the pixels in the first region, a second weighting coefficient is determined for weighting the CT values ​​of the pixels in the second region, a third weighting coefficient is determined for weighting the CT values ​​of the pixels in the third region, and a fourth weighting coefficient is determined for weighting the CT values ​​of the pixels in the fourth region. The first virtual monochromatic X-ray image is weighted using the first weighting coefficient and the second weighting coefficient; The second virtual monochrome X-ray image is weighted using the third and fourth weighting coefficients; as well as A corrected virtual monochromatic X-ray image of the first energy is generated based on the first virtual monochromatic X-ray image weighted by the first weighting coefficient and the second weighting coefficient, and the second virtual monochromatic X-ray image weighted by the third weighting coefficient and the fourth weighting coefficient.

Images