Correction processing method and apparatus for scattering artifact of cone-beam CT image
Through dual contrast learning and unsupervised image style conversion model combined with FDK algorithm, the problems of high hardware cost, large calculation cost and inconsistent anatomical structure after image quality improvement are solved in cone beam CT images scattering artifact correction, achieving efficient and low-cost Image quality is improved and suitable for adaptive radiation therapy.
Patent Information
- Application Number
- PCT/CN2023/137198
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-18
- Filing Date
- 2023-12-07
- Publication Date
- 2024-11-21
AI Technical Summary
The prior art has problems such as high hardware equipment cost, high calculation cost, difficulty in obtaining prior information and data, and inconsistent anatomical structure after image quality improvement, and the method based on image synthesis model is generated. The image lacks fidelity.
Unsupervised image style conversion model using dual contrast learning, converting CBCT images into pCT images, and combining post-processing technology and FDK algorithm to realize scattering artifact correction in the projection domain, generating high-quality synthetic pCTs and subtracting scattering artifacts. Maintain anatomical consistency of the image.
It realizes efficient and low-cost CBCT image scattering artifact correction, significantly improving soft tissue contrast and image quality, without additional prior information or hardware device changes, suitable for half-fan scanning, short correction time and high computing efficiency. Suitable for adaptive radiation therapy.
Smart Images

Figure CN2023137198_21112024_PF_FP_ABST
Abstract
Description
Cone-beam CT image scattering artifact correction processing method and device Technical Field
[0001] The present invention relates to the technical field of image processing, and in particular to a method and device for correcting scattering artifacts in cone-beam CT images. Background Art
[0002] This section is intended to provide a background or context to the embodiments of the invention that are recited in the claims. No statement herein is admitted to be prior art by virtue of its inclusion in this section.
[0003] Scattering artifacts, which are common in cone-beam CT (CBCT) images, often lead to reduced soft tissue contrast and image quality, which can impact adjunctive therapy and disease diagnosis. In recent years, many researchers have proposed methods to correct scattering artifacts in CBCT images to improve image quality. These methods can be broadly categorized as hardware-based, model-based, and deep learning-based.
[0004] The hardware correction method is to add hardware equipment to filter out scattering, that is, to suppress scattering artifacts through beam blockers, anti-scattered light, air gap method, etc. before the X-rays reach the detector.
[0005] Model-based scatter correction uses post-processing techniques to correct for scatter artifacts in scatter-contaminated projection images. These methods typically employ Monte Carlo (MC) techniques to simulate the scatter contribution to CBCT projection data. While MC scatter simulation offers high accuracy, its use is limited by the time-consuming and computationally expensive process of simulating the true paths of all photons in the radiation source. Prior data-based scatter correction methods use deformable image registration (DIR) to convert planning CT (pCT) images into CBCT images to obtain prior information for the corresponding procedure, using this prior information to improve CBCT image quality. Prior data-based scatter artifact correction can yield better corrected images, but this approach requires additional prior patient information. Furthermore, prior information-based correction methods require paired CBCT and planning CT datasets, which can be challenging to obtain. They also rely on precise spatial alignment of CBCT and pCT voxel pairs from the same patient.
[0006] Deep learning methods have been widely applied to CBCT scatter correction. Generative Adversarial Networks (GANs) have been widely used in image artifact correction, significantly improving CBCT imaging quality. In 2018, Kida et al. used mean absolute error as a loss function to train a two-dimensional U-Net to generate high-quality synthetic CT images from pelvic CBCT images. However, anatomical inconsistencies persisted in a few of the results. Liang used cycle-consistent generative adversarial networks (CycleGANs) trained on a dataset of unregistered CBCT images and planning CT images to generate high-quality CBCT images. The mean CT value error in head and neck patient data was 29 HU. In 2021, Zhang et al. used CycleGAN to improve CBCT image quality in pelvic and head and neck patients, acquiring CBCT images on different machines and conducting independent cross-dataset testing. However, the image grayscale mismatch caused by the global training of CycleGANs compromised the accuracy of the generated images. In 2022, Dong et al. proposed using contrastive unsupervised learning to generate high-quality CBCT images. However, CUT is prone to mode collapse during training, which limits its application. Technical issues
[0007] In summary, the existing technologies have the following main problems: 1. Hardware-based correction methods require additional hardware equipment, which is not only cumbersome to operate but also costly. They also filter out some image signals, resulting in the loss of important information in the reconstructed image, thereby reducing its accuracy. 2. Model-based scatter correction methods such as MC have disadvantages such as being time-consuming and computationally expensive. 3. Correction methods based on prior information rely on data that is difficult to obtain and cannot be used as a universal solution for scatter artifact correction. 4. Compared with hardware correction and model-based scatter correction, although scatter artifact correction based on image synthesis models significantly improves CBCT image quality, it is undeniable that CBCT images generated using only image synthesis models are false and lack fidelity, resulting in anatomical geometry inconsistencies before and after the improvement of CBCT image quality. CBCT images generated by traditional analytical methods, while of poor image quality, have high fidelity.
[0008] Therefore, there is an urgent need for a CBCT image processing solution that can overcome the above-mentioned defects, have high processing efficiency, good image quality, high accuracy and low cost. Technical Solutions
[0009] In order to solve the problems existing in the prior art, the present invention proposes a method and device for correcting scattering artifacts in cone-beam CT images.
[0010] In a first aspect of an embodiment of the present invention, a method for correcting scattering artifacts in cone-beam CT images is proposed, comprising:
[0011] Constructing a synthesis network, the synthesis network comprising at least two generators, two discriminators, two CBCT image domain feature extraction units, and two pCT image domain feature extraction units, for converting CBCT images into spCT images; wherein CBCT is cone-beam CT, spCT is synthesized pCT, and pCT is planned CT;
[0012] Forward projecting the CBCT image and the spCT image respectively to obtain the CBCT projection and the spCT projection, and determining the residual projection according to the CBCT projection and the spCT projection;
[0013] extracting scattering artifacts and image details from the residual projection, removing the image details from the residual projection, and retaining the scattering artifacts;
[0014] subtracting the scatter artifact from the CBCT projection to generate a corrected projection;
[0015] The corrected projections are reconstructed to obtain a corrected CBCT image.
[0016] In a second aspect of an embodiment of the present invention, a device for correcting scattering artifacts in cone-beam CT images is provided, comprising:
[0017] A synthesis network construction module, configured to construct a synthesis network comprising at least two generators, two discriminators, two CBCT image domain feature extraction units, and two pCT image domain feature extraction units, configured to convert CBCT images into spCT images; wherein CBCT is cone-beam CT, spCT is synthesized pCT, and pCT is planned CT;
[0018] A forward projection module is used to perform forward projection on the CBCT image and the spCT image respectively to obtain a CBCT projection and a spCT projection, and determine a residual projection according to the CBCT projection and the spCT projection;
[0019] an image processing module, configured to extract scattering artifacts and image details from the residual projection, remove the image details from the residual projection, and retain the scattering artifacts;
[0020] a projection correction module, configured to subtract the scattering artifact from the CBCT projection to generate a corrected projection;
[0021] The image reconstruction module is used to reconstruct the corrected projection to obtain a corrected CBCT image.
[0022] In a third aspect of an embodiment of the present invention, a computer device is proposed, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method for correcting scatter artifacts in cone-beam CT images when executing the computer program.
[0023] In a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, a method for correcting scattering artifacts in cone-beam CT images is implemented.
[0024] In a fifth aspect of an embodiment of the present invention, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, a method for correcting scattering artifacts in cone-beam CT images is implemented. Beneficial effects
[0025] The present invention proposes a method and device for correcting scattering artifacts in cone-beam CT images. Based on an unsupervised image style transfer model using dual contrast learning, the method converts CBCT images into pCT images and generates high-quality synthetic pCTs, which are then forward-projected onto the original CBCT and spCT images. A method combining post-processing techniques with the FDK algorithm is proposed to achieve scattering artifact correction in the projection domain, significantly improving CBCT image quality. The present invention requires no additional prior information, no hardware modifications, and no paired training data. The corrected image reconstruction is time-efficient and computationally efficient, making it suitable for half-fan scanning. The present invention's CBCT scattering artifact correction scheme maintains the fidelity of CBCT images before and after correction, significantly improving soft tissue contrast and image quality, and providing strong technical support for adaptive radiotherapy. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0027] FIG1 is a flow chart of a method for correcting scattering artifacts in cone-beam CT images according to an embodiment of the present invention.
[0028] FIG2 is a schematic diagram of a framework of a method for correcting scattering artifacts in cone-beam CT images according to a specific embodiment of the present invention.
[0029] FIG3 is a schematic diagram of a framework of a synthesis network according to a specific embodiment of the present invention.
[0030] FIG4 is a schematic diagram of axial, sagittal and coronal slice images of CBCT, cCBCT and pCT according to a specific embodiment of the present invention.
[0031] FIG5 is a schematic diagram of the CT number distribution of ROIs selected by CBCT, cCBCT, and pCT according to a specific embodiment of the present invention.
[0032] FIG6 is a schematic diagram of CT number distribution at a certain position of CBCT, cCBCT, and pCT according to a specific embodiment of the present invention.
[0033] FIG. 7 is a schematic diagram of the architecture of a cone-beam CT image scattering artifact correction processing device according to an embodiment of the present invention.
[0034] FIG8 is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. Modes for Carrying Out the Invention
[0035] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make this disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.
[0036] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present disclosure may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.
[0037] According to an embodiment of the present invention, a method and device for correcting scattering artifacts in cone-beam CT images are proposed, which relate to the field of image processing technology.
[0038] The principles and spirit of the present invention are explained in detail below with reference to several representative embodiments of the present invention.
[0039] FIG1 is a flow chart of a method for correcting scatter artifacts in cone-beam CT images according to an embodiment of the present invention. As shown in FIG1 , the method includes:
[0040] Step 101: constructing a synthesis network, wherein the synthesis network comprises at least two generators, two discriminators, two CBCT image domain feature extraction units, and two pCT image domain feature extraction units, for converting CBCT images into spCT images;
[0041] Among them, CBCT is cone-beam CT, spCT is synthesized pCT, and pCT is planning CT;
[0042] Step 102 , forward projecting the CBCT image and the spCT image respectively to obtain a CBCT projection and a spCT projection, and determining a residual projection based on the CBCT projection and the spCT projection;
[0043] Step 103: extracting scattering artifacts and image details from the residual projection, removing the image details from the residual projection, and retaining the scattering artifacts;
[0044] Step 104 , subtracting the scattering artifact from the CBCT projection to generate a corrected projection;
[0045] Step 105: reconstruct the corrected projection to obtain a corrected CBCT image.
[0046] In practical application scenarios, the cone-beam CT image scattering artifact correction processing method works as follows: constructing a synthesis network, which is used to convert CBCT images into spCT images. This synthesis network is improved compared to existing technologies and can improve the processing efficiency and conversion effect of image conversion; forward projecting the CBCT image and spCT image respectively to obtain CBCT projections and spCT projections, and determining residual projections based on the CBCT projections and spCT projections; extracting scattering artifacts and image details from the residual projections, removing the image details from the residual projections while retaining the scattering artifacts; subtracting the scattering artifacts from the CBCT projections to generate corrected projections; and reconstructing the corrected projections to obtain corrected CBCT images. The present invention can achieve CBCT scattering correction by combining an image synthesis model with post-processing techniques and traditional analytical methods in the projection domain, while maintaining the consistency of anatomical structures before and after CBCT image quality improvement.
[0047] In order to explain the above-mentioned cone-beam CT image scattering artifact correction method more clearly, each step is described in detail below.
[0048] In one embodiment, for step 101, the synthesis network includes: a generator G, a generator F, a discriminator D pCT , Discriminator D CBCT , a first CBCT image domain feature extraction unit, a second CBCT image domain feature extraction unit, a first pCT image domain feature extraction unit and a second pCT image domain feature extraction unit.
[0049] Specifically, the generator G is used to generate spCT images from real CBCT images;
[0050] Generator F is used to generate CBCT images from real pCT images;
[0051] Discriminator D pCT Used to distinguish whether the input image comes from the spCT image generated by the generator G or the real pCT image;
[0052] Discriminator D CBCT Used to distinguish whether the input image comes from the CBCT image generated by the generator F or the real CBCT image.
[0053] The first CBCT image domain feature extraction unit and the second CBCT image domain feature extraction unit are used to extract image domain features from the CBCT image;
[0054] The first pCT image domain feature extraction unit and the second pCT image domain feature extraction unit are used to extract image domain features of the spCT image.
[0055] In one embodiment, the loss function of the synthetic network includes at least adversarial loss, PatchNCE loss, and consistency loss; and a total loss function can be determined based on these losses.
[0056] In the synthetic network, the adversarial loss of the generator G is expressed as:
[0057]
[0058] The adversarial loss of the generator F is expressed as:
[0059]
[0060] in, For adversarial losses;
[0061] G and F are generators;
[0062] and is the discriminator;
[0063] CBCT is the CBCT image dataset;
[0064] pCT is the pCT image dataset;
[0065] cbct represents the individual image of CBCT;
[0066] pct represents the individual image of pCT;
[0067] represents the mathematical expectation, where cbct obeys the probability distribution of CBCT;
[0068] Similarly Expressed as a mathematical expectation, where pct obeys the probability distribution of pCT.
[0069] In the synthetic network, the PatchNCE loss of converting CBCT images to pCT images is expressed as:
[0070]
[0071] The PatchNCE loss for converting pCT images to CBCT images is expressed as:
[0072]
[0073] in, is the PatchNCE loss for converting CBCT images to pCT images;
[0074] is the PatchNCE loss for converting pCT images to CBCT images;
[0075] G and F are generators;
[0076] A mapping fitted by a CBCT image domain feature extraction unit, comprising a first CBCT image domain feature extraction unit and a second CBCT image domain feature extraction unit;
[0077] A mapping fitted by a pCT image domain feature extraction unit, comprising a first pCT image domain feature extraction unit and a second pCT image domain feature extraction unit;
[0078] CBCT is the CBCT image dataset;
[0079] pCT is the pCT image dataset;
[0080] cbct represents the individual image of CBCT;
[0081] pct represents the individual image of pCT;
[0082] represents the mathematical expectation, where cbct obeys the probability distribution of CBCT;
[0083] Expressed as mathematical expectation, where pct obeys the probability distribution of pCT;
[0084] N is the number of layers of feature maps;
[0085] Sequence l∈{1,2,3,…,N};
[0086] S lis the number of spatial locations in layer l;
[0087] Sequence s∈{1,2,3,…,S l};
[0088] is the feature of the output image;
[0089] as a positive characteristic;
[0090] A negative characteristic.
[0091] In the synthetic network, the consistency loss is expressed as:
[0092]
[0093] in, is the consistency loss;
[0094] G and F are generators;
[0095] represents the mathematical expectation, where cbct obeys the probability distribution of CBCT;
[0096] Expressed as mathematical expectation, where pct obeys the probability distribution of pCT;
[0097] cbct represents the individual image of CBCT;
[0098] pct represents the individual image of pCT.
[0099] Furthermore, the method further comprises:
[0100] Determining a total loss function according to the adversarial loss, the PatchNCE loss, and the consistency loss, and training the synthetic network according to the total loss function;
[0101] The total loss function is expressed as:
[0102]
[0103] Among them, L is the total loss function; For adversarial losses; 、 for PatchNCE losses; is the consistency loss; 、 、 are the control coefficients of adversarial loss, PatchNCE loss and consistency loss respectively.
[0104] With respect to step 102 , forward projection is performed on the CBCT image and the spCT image respectively to obtain a CBCT projection and a spCT projection, and a residual projection is determined according to the CBCT projection and the spCT projection.
[0105] With respect to step 103 , scattering artifacts and image details are extracted from the residual projection, the image details are removed from the residual projection, and the scattering artifacts are retained.
[0106] In one embodiment, the image details are removed from the residual projection using a low-pass filter, wherein the scattering artifacts are low-frequency signals and the image details are high-frequency signals.
[0107] For step 104 , the scatter artifacts are subtracted from the CBCT projections to generate corrected projections.
[0108] With respect to step 105 , the corrected projection is reconstructed to obtain a corrected CBCT image.
[0109] In one embodiment, the corrected projections are reconstructed using the FDK algorithm to obtain a corrected CBCT image.
[0110] The following describes a method for correcting scattering artifacts in cone-beam CT images with reference to a specific embodiment.
[0111] Refer to Figure 2, which is a schematic diagram of a framework of a method for correcting scattering artifacts in cone-beam CT images according to a specific embodiment of the present invention. The specific process is as follows:
[0112] Build the synthetic network:
[0113] The synthesis network developed in this paper is used to convert CBCT into pCT, generating a synthesized pCT (spCT). To ensure high fidelity of the corrected image, the present invention proposes a method that combines a traditional analytical model with an image synthesis model to address incomplete or inaccurate false information generated by the synthesis network.
[0114] Obtain uncorrected CBCT and obtain spCT (synthesis pCT) through the synthesis network.
[0115] Forward projection:
[0116] The forward projection algorithm is used on CBCT and spCT to obtain CBCT projection P0 (CBCT projection) and spCT projection P respectively. f (synthesis pCT projection).
[0117] By subtracting P from P0 f To generate the residual projection R0 (residual projection), to extract the scattering artifacts and image details in the projection.
[0118] Low-pass filtering:
[0119] Scattering artifacts are low frequency, while image details are high frequency signals. Use a low-pass filter to remove image details (high frequency) from R0 to estimate the scattering artifact R f (scattering projection signal).
[0120] Then, subtract R from P0 f , generate a corrected projection P c (corrected CBCT projection).
[0121] FDK Reconstruction:
[0122] The FDK reconstruction algorithm is used to reconstruct the corrected projection P c Reconstruction is performed to obtain the final corrected CBCT (cCBCT, corrected CBCT).
[0123] The synthesis network is the core improvement of the present invention. Figure 3 shows a schematic diagram of the synthesis network framework of a specific embodiment of the present invention. The symbols in Figure 3 are explained as follows: CBCT is CBCT image; pCT is pCT image; G and F are generators; D CBCT 、D pCT is the discriminator; G enc 、F enc is the encoder; G dec 、F dec For the decoder, separate Embedding CBCT is set for the CBCT image and the generated CBCT image, and separate Embedding pCT is set for the generated pCT image and the real pCT image. CBCT 、H pCT is the image domain feature extraction unit; GAN Loss is the adversarial loss; NCE CBCT Loss, NCE pct Loss is PatchNCE loss.
[0124] Referring to Figure 3, in the synthesis network, the goal of the generator G is to generate synthetic pCT (spCT) images from CBCT images, making them as close to the real pCT as possible.
[0125] Discriminator DpCT The goal is to distinguish spCT images from real pCT images. Ideally, the discriminator D pCT An array of all 0s or all 1s will be output to distinguish sPCT images from real pCT images.
[0126] Generator F and Discriminator D CBCT Similarly, during training, the generator and discriminator need to be continuously optimized alternately to make the spCT image as close as possible to the real pCT image, so that the discriminator has stronger discrimination ability, thereby continuously improving the quality and credibility of the generated images.
[0127] The accuracy of a network directly depends on the design of its corresponding loss function. The DCLGAN network's loss function consists of three components: adversarial loss, PatchNCE loss, and consistency loss. Similar to other GANs, DCLGAN uses an adversarial loss to ensure that the distribution pattern of the generated image spCT is as consistent as possible with that of the target domain image, that is, it is more visually similar to the target domain image.
[0128] The adversarial loss (GAN loss) of the generator G is expressed as:
[0129]
[0130] The adversarial loss (GAN loss) of generator F is similar to that of generator G:
[0131]
[0132] Similar to the CUT model, DCLGAN uses a contrastive estimation framework to maximize the mutual information between the corresponding patches of the input and output. The core idea of contrastive learning is to associate two samples, a "query" and its "positive" example, and contrast them with other samples in the dataset (called "negative"). DCLAGN constructs positive and negative samples by treating patches in the image as objects. An image patch at a certain position n in the generated spCT image is selected as the query sample. ℝ K , where the patch corresponding to the same position n in CBCT is considered as a positive sample ℝ K , the remaining patches in CBCT are regarded as negative samples ℝ N×K Using cross entropy loss, the specific loss mathematical expression is as follows:
[0133]
[0134] Among them, the hyperparameters Used to scale the distance between the query and other samples, with a default value of 0.07.
[0135] Represents query sample v and positive sample v + By minimizing the above cross entropy loss, the present invention can maximize v and v + The mutual information between v and The mutual information between them.
[0136] Unlike CUT, DCLAGN uses two different embedding layers to overcome the problem of using only one embedding layer in CUT, which may not effectively extract features from two independent image domains. Separate Embedding CBCT layers are used for the original CBCT image and the generated CBCT image, and separate Embedding pCT layers are used for the generated pCT image and the real pCT image.
[0137] The present invention uses G enc and H CBCT The Embeding CBCT is combined to extract the CBCT image domain features. Similarly, based on F enc and H pCT Extract PCT image domain features. The two embeddings do not share weights, so they can better learn the specificity of the two image domains. enc (CBCT) selects L layers of feature maps and passes them through one or two layers of MLP projection heads, namely:
[0138]
[0139] in, , Represents the selected lth output feature. The stacked encoder of the present invention The generated features, each feature is actually a patch in the image, , S l Indicates the number of spatial positions in the lth layer. Each time a query is selected, the corresponding feature is found as the positive feature ℝ and other traits as negative traits ℝ , where C l Represents the number of channels in each layer. Different from CUT, the present invention uses different Embedding in the pCT image domain. instead of , that is, using two different embeddings in two different image domains to effectively capture image domain features. So, similarly, the output sPCT is encoded into:
[0140]
[0141] Therefore, the present invention calculates the matching degree of the relevant patches of the input image and the output image. The loss function PatchNCE loss (NCE) based on the multi-layer patch conversion from CBCT to PCT image CBCT Loss) can be expressed as:
[0142]
[0143] In the reverse process: PCT->CBCT, calculate a similar loss (NCE pCT Loss):
[0144]
[0145] The difference from the mapping CBCT->PCT of the generator G is:
[0146]
[0147]
[0148] To ensure structural consistency between CBCT and generated sPCT and avoid unnecessary modifications, we use an identity loss. Unlike CUT, we do not use PatchNCE loss as identity loss to speed up training.
[0149]
[0150] Finally, the total loss function is:
[0151]
[0152] Furthermore, the synthesis network can be used to transform the original CBCT into spCT. Then, the analytical algorithm is combined with the synthesis model to achieve scattering artifact correction. The specific process is: forward project the original CBCT and the pseudo PCT (sPCT) generated by DCLGAN to obtain P0 and P respectively. f . Set P0 and P f Subtraction obtains the residual projection R0 containing image detail signals and scattering artifact signals. Since the image detail part is a high-frequency signal and the scattering artifact is a low-frequency signal, a low-pass filter is used to remove the image detail part to obtain R fSubtract R from the original CBCT projection P0 f , thus obtaining the corrected CBCT projection P c Finally, use the FDK reconstruction algorithm to reconstruct P c Reconstruction is performed to obtain the final corrected CBCT image (cCBCT).
[0153] The proposed CBCT scatter artifact solution addresses the poor image quality and low soft tissue contrast caused by scatter artifacts through steps such as building a synthesis network, forward projection, mean filtering, and FDK reconstruction. The core improvement lies in an unsupervised image style transfer model based on dual contrast learning (DCLGAN), which converts CBCT images into pCT images and generates high-quality synthesized pCT (spCT).
[0154] The present invention forward projects the original CBCT and spCT respectively, and proposes a method combining post-processing technology and FDK algorithm to achieve scattering artifact correction in the projection domain, thereby greatly improving the quality of CBCT images.
[0155] The present invention uses the same experimental parameters to train the CUT and CycleGAN models and performs scattering artifact correction on CBCT images. Experimental results show that the present method is more versatile than CUT, avoids mode collapse, and performs better than CycleGAN.
[0156] Compared with existing methods, the present invention has the following advantages: 1. No additional prior information is required; 2. No hardware changes are required; 3. No paired training data is required; 4. The corrected image reconstruction time is less than 1 minute, with high computational efficiency; 5. It is suitable for half-fan scanning; 6. The CBCT scatter artifact correction scheme proposed in the present invention can maintain the fidelity of CBCT images before and after correction, and can significantly improve the soft tissue contrast and image quality of CBCT.
[0157] In practical application scenarios, the present invention corrects scattering artifacts in CBCT images based on double contrast learning, which can provide key technical support for achieving adaptive radiotherapy.
[0158] It should be noted that although the operations of the method of the present invention are described in a specific order in the above embodiments and drawings, this does not require or imply that these operations must be performed in this specific order, or that all illustrated operations must be performed to achieve the desired results. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.
[0159] In a specific embodiment, the method of the present invention was used to conduct an experimental verification on patient CBCT data, and the experimental results are shown in FIG4 and FIG5.
[0160] Figure 4 shows axial, sagittal, and coronal slices of CBCT, cCBCT, and pCT for patients A and B. As shown in Figure 4, compared with the original CBCT, the scatter artifact-corrected CBCT (cCBCT) not only maintains anatomical consistency but also significantly improves image quality, including soft tissue contrast, spatial uniformity, clarity, and artifact suppression.
[0161] Figure 5 shows the distribution of CT counts for selected ROIs (Regions of Interest) in CBCT, cCBCT, and pCT images. The horizontal axis represents the ROI, and the vertical axis represents the conversion of pixel values to HU (Human Handling Values). Figure 5 clearly shows that the variance of the mean HU (Human Handling Values) for the cCBCT ROI is closer to that of the pCT ROI than that of the CBCT ROI. Although pCT and CBCT differ in structure, cCBCT learns the style of pCT based on the original CBCT structure, achieving high-fidelity, high-contrast image conversion.
[0162] In a specific embodiment, referring to FIG6 , the CT number distribution at a certain position in CBCT, cCBCT, and pCT images is compared.
[0163] Figure 6 shows axial views of CBCT, cCBCT, and pCT from top to bottom on the left. The right figure shows the CT number distribution at the two marked line positions (column 260 and row 312) of CBCT, cCBCT, and pCT. The display window is [-350, 150] HU. The abscissa is position information (Position), and the ordinate is pixel value (Pixel Value) converted to HU. Circle 1 is CBCT, circle 2 is pCT, and circle 3 is CCBCT. In comparison, the processed cCBCT significantly improves image quality.
[0164] After introducing the method according to the exemplary embodiment of the present invention, the cone-beam CT image scattering artifact correction processing device according to the exemplary embodiment of the present invention will be introduced with reference to FIG. 7 .
[0165] The implementation of the cone-beam CT image scatter artifact correction processing device can be referenced to the implementation of the aforementioned method, and any repetitions will not be repeated. The terms "module" or "unit" used below may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.
[0166] Based on the same inventive concept, the present invention also proposes a cone-beam CT image scattering artifact correction processing device, as shown in FIG7 , which includes:
[0167] A synthesis network construction module 710 is used to construct a synthesis network, wherein the synthesis network includes at least two generators, two discriminators, two CBCT image domain feature extraction units, and two pCT image domain feature extraction units, and is used to convert CBCT images into spCT images; wherein CBCT is cone-beam CT, spCT is synthesized pCT, and pCT is planned CT;
[0168] A forward projection module 720 is configured to perform forward projection on the CBCT image and the spCT image to obtain a CBCT projection and a spCT projection, and determine a residual projection based on the CBCT projection and the spCT projection;
[0169] An image processing module 730 is configured to extract scattering artifacts and image details from the residual projection, remove the image details from the residual projection, and retain the scattering artifacts;
[0170] a projection correction module 740 for subtracting the scattering artifacts from the CBCT projections to generate corrected projections;
[0171] The image reconstruction module 750 is configured to reconstruct the corrected projection to obtain a corrected CBCT image.
[0172] It should be noted that while the detailed description above mentions several modules of the apparatus for correcting scatter artifacts in cone-beam CT images, this division is merely exemplary and not mandatory. In practice, according to embodiments of the present invention, the features and functions of two or more modules described above may be embodied in a single module. Conversely, the features and functions of a single module described above may be further divided and embodied by multiple modules.
[0173] Based on the aforementioned inventive concept, as shown in FIG8 , the present invention further proposes a computer device 800 comprising a memory 810, a processor 820, and a computer program 830 stored in the memory 810 and executable on the processor 820. When the processor 820 executes the computer program 830, the aforementioned cone-beam CT image scattering artifact correction processing method is implemented.
[0174] Based on the aforementioned inventive concept, the present invention proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned cone-beam CT image scattering artifact correction processing method.
[0175] Based on the aforementioned inventive concept, the present invention proposes a computer program product, which includes a computer program. When the computer program is executed by a processor, a method for correcting scatter artifacts in cone-beam CT images is implemented.
[0176] The present invention proposes a method and device for correcting scattering artifacts in cone-beam CT images. Based on an unsupervised image style transfer model using dual contrast learning, the method converts CBCT images into pCT images and generates high-quality synthetic pCTs, which are then forward-projected onto the original CBCT and spCT images. A method combining post-processing techniques with the FDK algorithm is proposed to achieve scattering artifact correction in the projection domain, significantly improving CBCT image quality. The present invention requires no additional prior information, no hardware modifications, and no paired training data. The corrected image reconstruction is time-efficient and computationally efficient, making it suitable for half-fan scanning. The present invention's CBCT scattering artifact correction scheme maintains the fidelity of CBCT images before and after correction, significantly improving soft tissue contrast and image quality, and providing strong technical support for adaptive radiotherapy.
[0177] The acquisition, storage, use, and processing of data in the technical solution of this application comply with relevant laws and regulations.
[0178] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatuses, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.
[0179] The present invention is described with reference to the flowcharts and / or block diagrams of the methods and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowcharts and / or one or more blocks in the block diagrams.
[0180] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.
[0181] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.
[0182] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.
Claims
1. A method for correcting scattering artifacts in cone-beam CT images, characterized in that: include: Constructing a synthetic network, wherein the synthetic network comprises at least two generators, two discriminators, two CBCT image domain feature extraction units and two pCT image domain feature extraction units, for converting a CBCT image into a spCT image; wherein CBCT is a cone beam CT, spCT is a synthetic pCT, and pCT is a planned CT; Forward projecting the CBCT image and the spCT image respectively to obtain a CBCT projection and a spCT projection, and determining a residual projection according to the CBCT projection and the spCT projection; Extracting scattering artifacts and image details in the residual projection, removing the image details from the residual projection, and retaining the scattering artifacts; subtracting the scatter artifact from the CBCT projection to generate a corrected projection; The corrected projection is reconstructed to obtain a corrected CBCT image.
2. The method according to claim 1, characterized in that The synthetic network includes: generator G, generator F, discriminator D pCT , Discriminator D CBCT , a first CBCT image domain feature extraction unit, a second CBCT image domain feature extraction unit, a first pCT image domain feature extraction unit and a second pCT image domain feature extraction unit.
3. The method according to claim 2, characterized in that The generator G is used to generate spCT images from real CBCT images; Generator F is used to generate CBCT images from real pCT images; Discriminator D pCT Used to distinguish whether the input image comes from the spCT image generated by the generator G or the real pCT image; Discriminator D CBCT Used to distinguish whether the input image comes from the CBCT image generated by the generator F or the real CBCT image.
4. The method according to claim 2, characterized in that: The first CBCT image domain feature extraction unit and the second CBCT image domain feature extraction unit are used to extract image domain features in the CBCT image; The first pCT image domain feature extraction unit and the second pCT image domain feature extraction unit are used to extract image domain features of the spCT image.
5. The method according to claim 1, characterized in that The loss function of the synthetic network at least includes adversarial loss, PatchNCE loss and consistency loss.
6. The method according to claim 5, characterized in that In the synthetic network, the adversarial loss of the generator G is expressed as: The adversarial loss of the generator F is expressed as: in, For adversarial losses; G and F are generators; and is the discriminator; CBCT is the CBCT image dataset; pCT is the pCT image dataset; cbct represents the individual image of CBCT; pct represents the image individual of pCT; represents the mathematical expectation, where cbct obeys the probability distribution of CBCT; Expressed as a mathematical expectation, where pct obeys the probability distribution of pCT.
7. The method according to claim 5, characterized in that In the synthetic network, the PatchNCE loss of converting CBCT images to pCT images is expressed as: The PatchNCE loss for converting pCT images to CBCT images is expressed as: in, is the PatchNCE loss of converting CBCT images to pCT images; is the PatchNCE loss of converting pCT images to CBCT images; G and F are generators; A mapping fitted by a CBCT image domain feature extraction unit, comprising a first CBCT image domain feature extraction unit and a second CBCT image domain feature extraction unit; A mapping fitted by a pCT image domain feature extraction unit, comprising a first pCT image domain feature extraction unit and a second pCT image domain feature extraction unit; CBCT is the CBCT image dataset; pCT is the pCT image dataset; cbct represents the individual image of CBCT; pct represents the image individual of pCT; represents the mathematical expectation, where cbct obeys the probability distribution of CBCT; It is expressed as a mathematical expectation, where pct follows the probability distribution of pCT; N is the number of layers of feature maps; Sequence l∈{1,2,3,…,N}; S l is the number of spatial locations in layer l; Order s∈{1,2,3,…,S l }; is the feature of the output image; for positive characteristics; A negative characteristic.
8. The method according to claim 5, characterized in that In the synthetic network, the consistency loss is expressed as: in, is the consistency loss; G and F are generators; represents the mathematical expectation, where cbct obeys the probability distribution of CBCT; It is expressed as a mathematical expectation, where pct follows the probability distribution of pCT; cbct represents the individual image of CBCT; pct stands for pCT image entity.
9. The method according to claim 5, characterized in that The method further includes: Determine a total loss function according to the adversarial loss, the PatchNCE loss, and the consistency loss, and train the synthetic network according to the total loss function; The total loss function is expressed as: Among them, L is the total loss function; For adversarial losses; 、 for PatchNCE losses; is the consistency loss; 、 、 They are the control coefficients of adversarial loss, PatchNCE loss and consistency loss respectively.
10. The method according to claim 1, characterized in that Extracting scattering artifacts and image details in the residual projection, removing the image details from the residual projection, and retaining the scattering artifacts, including: The image details are removed from the residual projection using a low-pass filter, wherein the scattering artifacts are low-frequency signals and the image details are high-frequency signals.
11. The method according to claim 1, characterized in that Reconstructing the corrected projection to obtain a corrected CBCT image includes: The corrected projection is reconstructed using the FDK algorithm to obtain a corrected CBCT image.
12. A cone-beam CT image scattering artifact correction processing device, characterized in that: include: A synthesis network construction module, used to construct a synthesis network, wherein the synthesis network comprises at least two generators, two discriminators, two CBCT image domain feature extraction units and two pCT image domain feature extraction units, and is used to convert the CBCT image into a spCT image; wherein CBCT is a cone beam CT, spCT is a synthesized pCT, and pCT is a planned CT; A forward projection module, used to perform forward projection on the CBCT image and the spCT image respectively to obtain a CBCT projection and a spCT projection, and determine a residual projection according to the CBCT projection and the spCT projection; An image processing module, configured to extract scattering artifacts and image details from the residual projection, remove the image details from the residual projection, and retain the scattering artifacts; A projection correction module, configured to subtract the scattering artifact from the CBCT projection to generate a corrected projection; The image reconstruction module is used to reconstruct the corrected projection to obtain a corrected CBCT image.
13. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 11 is implemented.
14. 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 method according to any one of claims 1 to 11 is implemented.
15. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.
Citation Information
Patent Citations
Self-adaptive iteration scattering correction method of cone beam CT
CN104408753A
Blocking optical grating optimization method and device for scattering correction of cone-beam CT (computed tomography) image
CN106408543A
System and method for removing metal artifacts in CBCT image based on double decoders, terminal and medium
CN115272511A
Cone beam CT image scattering artifact correction processing method and device
CN116543070A
Cone-beam CT image enhancement using generative adversarial networks
US20190333219A1
Cited By
Cone beam computed tomography image processing method, electronic equipment and storage medium
CN120876564A
Cone beam computed tomography scattering correction method and system
CN121714295A
CBCT metal artifact suppression method based on unsupervised preoperative prior repair network
CN122265477A
CBCT metal artifact suppression method based on unsupervised preoperative prior repair network
CN122265477B
