Real scene low-dose CT image denoising method and system

Abstract

This invention discloses a method and system for denoising low-dose CT images in real-world scenarios. The method includes: acquiring a low-dose CT image to be denoised; inputting the low-dose CT image to be denoised into a trained denoising model and outputting a denoised CT image; wherein, the optimized training set acquisition process includes: cropping the low-dose CT image to be optimized and the normal-dose CT image into several image blocks respectively; performing binarization processing on the image blocks to obtain a mask for the low-dose CT image block to be optimized and a mask for the normal-dose CT image block; calculating a common mask based on the image block masks; calculating the similarity between the image blocks based on the pixel values ​​of the common mask; comparing the similarity with a set threshold and selecting several pairs of image blocks with a similarity greater than the set threshold; using each selected pair of image blocks as an optimized sample, and forming a training set from all optimized samples.

Description

Technical Field

[0001] This invention relates to the field of image denoising technology, and in particular to a method and system for denoising low-dose CT images in real-world scenarios. Background Technology

[0002] The statements in this section merely refer to the background art related to this invention and do not necessarily constitute prior art.

[0003] During a computed tomography (CT) scan, the higher the dose, the greater the radiation received by the patient. Radiation dose can affect a patient's health, particularly the risk of hematologic malignancies, which is directly proportional to the cumulative dose received, with an additional relative risk (ERR) of 1.96 / 100mGv. Reducing the CT scan dose introduces additional noise and artifacts, degrading CT image quality and making some tissues and structures unclear, thus affecting the doctor's diagnosis.

[0004] In current deep learning-based low-dose CT reconstruction techniques, supervised learning methods primarily involve adding Poisson noise to the sinusoidal domain of normal-dose CT images to simulate the corresponding low-dose CT sinusoidal domain image. This simulated noise is then projected to obtain the low-dose CT image, and paired data is used for training. However, the distribution of simulated noise differs from that of real clinical scenarios, failing to simulate the complexities of real-world clinical situations. Existing generative adversarial network (GAN)-based methods also use real scanned piglet data for training. However, due to differences between pig and human images, and the inherent fidelity issues in the generative model, the reconstructed image may contain structures absent in the original image or be missing, impacting diagnosis. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for denoising low-dose CT images in real-world scenarios;

[0006] On the one hand, it provides a method for denoising low-dose CT images in real-world scenarios, including:

[0007] A low-dose CT image to be denoised is acquired; the low-dose CT image to be denoised is input into a trained denoising model, and the denoised CT image is output; wherein, the trained denoising model is obtained by training a deep learning model using an optimized training set, and the process of acquiring the optimized training set includes: acquiring a sample set to be optimized, the sample set to be optimized including several pairs of samples to be optimized, each pair of samples to be optimized including: a low-dose CT image to be optimized and a normal-dose CT image;

[0008] The low-dose CT image to be optimized and the normal-dose CT image are cropped into several image blocks respectively; the low-dose CT image block to be optimized and the corresponding normal-dose CT image block are binarized to obtain the low-dose CT image block mask to be optimized and the normal-dose CT image block mask.

[0009] Based on the low-dose CT image block mask to be optimized and the normal-dose CT image block mask, the common mask of the two is calculated; based on the pixel values ​​of the common mask, the similarity between the low-dose CT image block to be optimized and the normal-dose CT image block is calculated.

[0010] The similarity is compared with a set threshold, and several low-dose CT image blocks and normal-dose CT image blocks with similarity greater than the set threshold are selected as optimization samples. Each of the selected low-dose CT image blocks and normal-dose CT image blocks is used as an optimized sample, and all optimized samples are combined into a training set.

[0011] On the other hand, a real-world low-dose CT image denoising system is provided, including:

[0012] The acquisition module is configured to acquire low-dose CT images to be denoised.

[0013] The denoising module is configured to: input the low-dose CT image to be denoised into the trained denoising model, and output the denoised CT image; wherein, the trained denoising model is obtained by training a deep learning model using an optimized training set, and the optimized training set is obtained by: obtaining a sample set to be optimized, the sample set to be optimized including several pairs of samples to be optimized, each pair of samples to be optimized including: a low-dose CT image to be optimized and a normal-dose CT image;

[0014] The low-dose CT image to be optimized and the normal-dose CT image are cropped into several image blocks respectively; the low-dose CT image block to be optimized and the corresponding normal-dose CT image block are binarized to obtain the low-dose CT image block mask to be optimized and the normal-dose CT image block mask.

[0015] Based on the low-dose CT image block mask to be optimized and the normal-dose CT image block mask, the common mask of the two is calculated; based on the pixel values ​​of the common mask, the similarity between the low-dose CT image block to be optimized and the normal-dose CT image block is calculated.

[0016] The similarity is compared with a set threshold, and several low-dose CT image blocks and normal-dose CT image blocks with similarity greater than the set threshold are selected as optimization samples. Each of the selected low-dose CT image blocks and normal-dose CT image blocks is used as an optimized sample, and all optimized samples are combined into a training set.

[0017] Furthermore, an electronic device is also provided, including:

[0018] Memory, used for non-transitory storage of computer-readable instructions; and

[0019] Processor, for executing the computer-readable instructions,

[0020] When the computer-readable instructions are executed by the processor, they perform the method described in the first aspect above.

[0021] In another aspect, a storage medium is also provided for non-transitory storage of computer-readable instructions, wherein when the non-transitory computer-readable instructions are executed by a computer, the instructions of the method described in the first aspect are executed.

[0022] In another aspect, a computer program product is also provided, including a computer program that, when run on one or more processors, is used to implement the method described in the first aspect above.

[0023] The above technical solution has the following advantages or beneficial effects:

[0024] The image denoising method of the present invention can denoise low-dose CT images of patients in real-world scenarios; moreover, it can be combined with other deep learning models to improve the restoration effect of end-to-end training. Attached Figure Description

[0025] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0026] Figure 1 This is a flowchart illustrating the process of obtaining the training set in Example 1.

[0027] Figures 2(a) to 2(d) A comparison is made between the Euclidean distance of Example 1 and the similar distance proposed in this invention;

[0028] Figures 3(a) to 3(e) The motivation for proposing the block data cleanup strategy in Example 1 is as follows. Detailed Implementation

[0029] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] Example 1

[0031] This embodiment provides a method for denoising low-dose CT images in real-world scenarios;

[0032] like Figure 1 As shown, a method for denoising low-dose CT images in real-world scenarios includes:

[0033] S101: Acquire low-dose CT images to be denoised;

[0034] S102: Input the low-dose CT image to be denoised into the trained denoising model, and output the denoised CT image; wherein, the trained denoising model is obtained by training a deep learning model using an optimized training set, such as... Figure 1 As shown, the process of obtaining the optimized training set includes:

[0035] (1): Obtain the sample set to be optimized, which includes several pairs of samples to be optimized, each pair of samples to be optimized including: a low-dose CT image to be optimized and a normal-dose CT image;

[0036] (2): The low-dose CT image to be optimized and the normal-dose CT image are respectively cropped into several image blocks;

[0037] (3): The low-dose CT image block to be optimized and the normal-dose CT image block at the corresponding position are binarized to obtain the mask of the low-dose CT image block to be optimized and the mask of the normal-dose CT image block.

[0038] (4): Based on the low-dose CT image block mask to be optimized and the normal-dose CT image block mask, calculate the common mask of the two;

[0039] (5): Based on the pixel values ​​of the common mask, the similarity between the low-dose CT image block to be optimized and the normal-dose CT image block is calculated;

[0040] (6): Compare the similarity with the set threshold, and select a number of low-dose CT image blocks and normal-dose CT image blocks to be optimized that have a similarity greater than the set threshold; take each of the selected low-dose CT image blocks and normal-dose CT image blocks as an optimized sample, and make all the optimized samples into a training set.

[0041] Further, S101: acquiring the low-dose CT image to be denoised means acquiring the low-dose CT image of the patient participating in the test in a real scene; the low dose refers to a radiation dose less than or equal to 10% of the normal dose.

[0042] Further, in step S102: the low-dose CT image to be denoised is input into the trained denoising model, and the denoised CT image is output. The denoising model is implemented using a Residual Encoder-Decoder Convolutional Neural Network (RED-CNN), or a PAN network or a NAFNet network.

[0043] Further, in step S102: the low-dose CT image to be denoised is input into the trained denoising model, and the denoised CT image is output. The training process of the trained denoising model includes:

[0044] The optimized training set is input into the denoising model to train it. Training is stopped when the total loss function value of the denoising model no longer decreases, and the trained denoising model is obtained.

[0045] Further, in step (1), the sample set to be optimized is obtained, which includes several pairs of samples to be optimized. Each pair of samples to be optimized includes: a low-dose CT image to be optimized and a normal-dose CT image. Each pair of samples to be optimized is a CT image of the same patient, but the radiation dose used on the patient is different. The normal dose is the radiation value set by the doctor, and the low dose is 10% of the normal dose.

[0046] Further, in step (2): the low-dose CT image to be optimized and the normal-dose CT image are respectively cropped into several image blocks. The cropping is to crop both the low-dose CT image to be optimized and the normal-dose CT image according to a set size. After cropping, the position of the low-dose CT image block and the position of the normal-dose CT image block are matched one-to-one according to the same position of the patient's body tissue to which the image block belongs.

[0047] Here, the locations of low-dose CT image blocks are mapped one-to-one with the locations of normal-dose CT image blocks. In other words, two image blocks from the same location in the patient's body tissue are paired. This reduces the computational burden of subsequent similarity calculations; image blocks that are not from the same location in the body tissue do not require similarity calculations.

[0048] Further, in step (3): the low-dose CT image block to be optimized and the corresponding normal-dose CT image block are binarized to obtain the low-dose CT image block mask to be optimized and the normal-dose CT image block mask. The binarization process is:

[0049] Given a threshold T, if the pixel value of pixel i in the low-dose CT image block to be optimized is greater than the threshold T, then the pixel value of pixel i is set to 1; if the pixel value of pixel i in the low-dose CT image block to be optimized is less than or equal to the threshold T, then the pixel value of pixel i is set to 0, thus obtaining the mask of the low-dose CT image block to be optimized.

[0050] If the pixel value of pixel i in a normal dose CT image block is greater than the threshold T, then the pixel value of pixel i is set to 1; if the pixel value of pixel i in a normal dose CT image block is less than or equal to the threshold T, then the pixel value of pixel i is set to 0, thus obtaining a normal dose CT image block mask.

[0051] Further, step (4): Calculating the common mask of the low-dose CT image block mask to be optimized and the normal-dose CT image block mask, specifically including:

[0052] If the pixel value of the mask pixel i of the low-dose CT image block to be optimized is equal to the pixel value of the mask pixel i of the normal-dose CT image block, then the pixel value of the common mask pixel i is 1; otherwise, the pixel value of the common mask pixel i is 0.

[0053] Further, step (5): calculating the similarity between the low-dose CT image block to be optimized and the normal-dose CT image block based on the pixel values ​​of the common mask, specifically includes:

[0054] The ratio of the number of pixels with a value of 1 in the common mask to the total number of pixels in the common mask is calculated, and this ratio is used as the similarity between the low-dose CT image block to be optimized and the normal-dose CT image block.

[0055] It should be understood that this definition of similarity is based on anatomical structure and can better measure the similarity between anatomical structures.

[0056] Figure 3(a) shows a low-dose CT image, Figure 3(b) shows a normal-dose CT image, Figure 3(c) shows an incompletely paired low-dose CT image and a normal-dose CT image, Figure 3(d) shows the training result obtained by training a deep learning model using two incompletely paired CT images, and Figure 3(e) shows the training result obtained by training a deep learning model using two completely paired CT images obtained after data purification strategy. Due to the motion between low-dose CT and normal-dose CT during the scanning process (as shown in Figure 3(c)), the low-dose CT image and the normal-dose CT image are not completely matched. Directly using them for repair will result in motion blur in the repaired image (as shown in Figure 3(d)), affecting the repair effect.

[0057] For this imperfect match, one possible approach is to utilize image nonlocal self-similarity, i.e., for each low-dose CT image patch (64×64), find the most similar patch in the normal-dose CT image (e.g., based on root mean square error (RMSE)). However, this process incurs a huge computational burden; for a 512×512 image, the complexity of calculating the RMSE of the corresponding patch is O(512). 2 ×64 2 The time complexity of sorting the results is O(512). 2 log(512 2 The overall complexity is O(512). 4 ×64 2 Preparing the training data is extremely time-consuming. On a 3.50 GHz Intel i9-10920X CPU, this method requires approximately 45,966.46 hours to process the patient dataset of this invention, which includes 29,278 pairs of 512×512 LDCT and NDCT images, posing a significant challenge to preparing a large-scale training dataset.

[0058] To address the aforementioned challenges, this invention proposes an image patch data cleansing strategy. This strategy only requires filtering the similarity of image patches at corresponding locations. Since pixel-value-based Euclidean distance metrics (such as RMSE) may have issues in measuring image similarity, as shown in Figure 2(a), the RMSE distance is 0.85 lower than that of Figure 2(c), yet they exhibit significant structural differences. Although the RMSE distances between Figure 2(b) and Figure 2(d) are similar, the anatomical structure in Figure 2(b) shows obvious motion; therefore, using Euclidean distance to measure similarity is unreasonable.

[0059] Therefore, this invention proposes a similarity measurement based on a binarized mask to better measure the similarity of anatomical structures. Using an image patch data cleansing strategy can filter the training data (as shown in Figure 3(d)), improving the image restoration effect in terms of detail clarity (such as blood vessels, structural edges, etc.). Furthermore, it significantly reduces the computational complexity to only O(5^12). 2 ×64 2 This method can process the patient dataset of this invention in just 15 minutes on the same Intel i9 CPU. Furthermore, the block data cleansing strategy of this invention can be combined with other CT denoising models and improves the results of restorations using real low-dose and normal-dose CT images.

[0060] The method of this invention, combined with deep learning models RED-CNN, PAN, and NAFNet, improves the clarity of restored details (such as blood vessels and tissue structures) in terms of visual effects, and enhances numerical results in FID, KID, and CLIP-FID metrics. When combined with these three models, the reconstruction results of CT images with low doses (10% of the normal dose) on rabbit and patient datasets outperform current commercial algorithms and representative methods for low-dose CT reconstruction in both visual and numerical terms. This ensures that the invention can reconstruct CT images with low doses (10% of the normal dose), further reducing the dose while assisting doctors in diagnosis.

[0061] Current deep learning-based CT denoising methods primarily use simulated data for paired training. However, the noise distribution in simulated data differs from that in real-world low-dose CT images, failing to simulate the complexities of real clinical scenarios. Furthermore, using real low-dose and normal-dose CT image pairs results in incomplete matching between the two images due to tissue movement caused by patient respiration. Directly utilizing this data for end-to-end learning leads to motion artifacts in the reconstructed data, affecting diagnostic accuracy. A potential solution is based on non-local self-similarity of images, finding the most similar block in the normal-dose CT image for paired training. However, this approach incurs enormous computational costs and dataset preparation time, posing a significant challenge to preparing a large-scale training dataset.

[0062] Meanwhile, due to the influence of noise, directly using Euclidean distance to measure the similarity between low-dose and normal-dose image patches presents problems, resulting in situations where the RMSE is very low but the structures are significantly different. Therefore, this invention proposes to calculate similarity using a binarized mask and to filter training data based on the similarity of corresponding blocks. This significantly reduces the computational burden of preparing the dataset for the aforementioned methods. It also improves both the visual effect and numerical indicators of the repaired images. Combined with existing denoising models, it achieved good results in both visual effect and numerical indicators for repairing CT images where the low dose is 10% of the normal dose. This also ensures that the CT dose can be reduced to 10%, and the repair results for low doses can assist doctors in diagnosis.

[0063] The data cleansing strategy of this invention improves both the visual quality and numerical metrics of incompletely paired CT reconstructed images. Visually, CT reconstructed images using the data cleansing strategy exhibit clearer details, including blood vessels and tissues. Numerically, the method of this invention, combined with RED-CNN, reduces FID, KID, and CLIP-FID metrics by 6%, 16%, and 21%, respectively, on rabbit CT image reconstructing compared to end-to-end training; and by 34%, 45%, and -25%, respectively, on the patient dataset. The PAN model reduces FID, KID, and KID by 6%, 12%, and 10% on rabbit CT data, and by 33%, 47%, and 8%, respectively, on the patient dataset. The NAFNet model reduces FID and KID by -11% and -23%, respectively, on rabbit CT data, while CLIP-FID shows no significant change. On the patient dataset, the three metrics are reduced by 32%, 43%, and 37%, respectively.

[0064] Compared to methods that rely on nonlocal self-similarity to find similar blocks and solve incomplete pairing problems, the method of this invention is simpler to use, computationally simpler, and faster to process. This method searches for the most similar block in a normal-dose CT image for each low-dose CT image block (64×64) (e.g., based on root mean square error (RMSE)). For a 512×512 image, the complexity of calculating the RMSE of the corresponding block is O(512). 2 ×64 2 The time complexity of sorting the results is O(512). 2 log(512 2 The overall complexity is O(512). 4 ×64 2 Preparing training data is a significant time-consuming process. On a 3.50GHz Intel i9-10920X CPU, this method requires approximately 45,966.46 hours to process the patient dataset of this invention, which includes 29,278 pairs of 512×512 LDCT and NDCT images. In contrast, the computational complexity of the method of this invention is only O(512). 2 ×64 2 This method can process the patient dataset of the present invention in just 15 minutes on the same Intel i9 CPU.

[0065] For the restoration of low-dose CT images that are equivalent to normal-dose CT images, the data filtering strategy of this invention, combined with RED-CNN, PAN, and NAFNet, achieves good results in both image quality and numerical metrics compared to other commercial algorithms and CT denoising and image denoising methods. The RMSE distance of Figure 2(a) is 0.85 lower than that of Figure 2(c), but it shows a significant structural difference. Although the RMSE distances between Figure 2(b) and Figure 2(d) are similar, the anatomical structure in Figure 2(b) shows obvious motion. Therefore, using Euclidean distance to measure similarity is unreasonable.

[0066] Example 2

[0067] This embodiment provides a real-world low-dose CT image denoising system;

[0068] A real-world low-dose CT image denoising system, including:

[0069] The acquisition module is configured to acquire low-dose CT images to be denoised.

[0070] The denoising module is configured to: input the low-dose CT image to be denoised into the trained denoising model, and output the denoised CT image; wherein, the trained denoising model is obtained by training a deep learning model using an optimized training set, and the optimized training set is obtained by: obtaining a sample set to be optimized, the sample set to be optimized including several pairs of samples to be optimized, each pair of samples to be optimized including: a low-dose CT image to be optimized and a normal-dose CT image;

[0071] The low-dose CT image to be optimized and the normal-dose CT image are cropped into several image blocks respectively; the low-dose CT image block to be optimized and the corresponding normal-dose CT image block are binarized to obtain the low-dose CT image block mask to be optimized and the normal-dose CT image block mask.

[0072] Based on the low-dose CT image block mask to be optimized and the normal-dose CT image block mask, the common mask of the two is calculated; based on the pixel values ​​of the common mask, the similarity between the low-dose CT image block to be optimized and the normal-dose CT image block is calculated.

[0073] The similarity is compared with a set threshold, and several low-dose CT image blocks and normal-dose CT image blocks with similarity greater than the set threshold are selected as optimization samples. Each of the selected low-dose CT image blocks and normal-dose CT image blocks is used as an optimized sample, and all optimized samples are combined into a training set.

[0074] It should be noted that the acquisition module and the noise reduction module mentioned above correspond to steps S101 to S102 in Embodiment 1. The examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0075] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0076] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0077] Example 3

[0078] This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the method described in Embodiment 1.

[0079] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0080] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0081] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0082] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0083] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0084] Example 4

[0085] This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.

[0086] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for denoising low-dose CT images in real-world scenarios, characterized by: include: Acquire low-dose CT images to be denoised; The low-dose CT image to be denoised is input into the trained denoising model, and the denoised CT image is output. The trained denoising model is obtained by training a deep learning model using an optimized training set. The optimized training set is obtained by acquiring a sample set to be optimized, which includes several pairs of samples to be optimized. Each pair of samples to be optimized includes a low-dose CT image to be optimized and a normal-dose CT image. The low-dose CT image to be optimized and the normal-dose CT image are cropped into several image blocks respectively; the low-dose CT image block to be optimized and the corresponding normal-dose CT image block are binarized to obtain the low-dose CT image block mask to be optimized and the normal-dose CT image block mask. Based on the low-dose CT image block mask to be optimized and the normal-dose CT image block mask, the common mask of the two is calculated, specifically including: If the low-dose CT image block mask pixels to be optimized The pixel value is equal to the mask pixel value of a normal dose CT image block. The pixel value, then the common mask pixels The pixel value is 1; otherwise, the common mask pixel... The pixel value is 0; Based on the pixel values ​​of the common mask, the similarity between the low-dose CT image patch to be optimized and the normal-dose CT image patch is calculated, specifically including: The ratio of the number of pixels with a value of 1 in the common mask to the total number of pixels in the common mask is calculated, and this ratio is used as the similarity between the low-dose CT image block to be optimized and the normal-dose CT image block. The similarity is compared with a set threshold, and several low-dose CT image blocks and normal-dose CT image blocks with similarity greater than the set threshold are selected as optimization samples. Each of the selected low-dose CT image blocks and normal-dose CT image blocks is used as an optimized sample, and all optimized samples are combined into a training set.

2. The method for denoising low-dose CT images in real-world scenarios as described in claim 1, characterized in that, Obtain a sample set to be optimized, which includes several pairs of samples to be optimized. Each pair of samples to be optimized includes: a low-dose CT image to be optimized and a normal-dose CT image. Each pair of samples to be optimized consists of CT images of the same patient, but the radiation doses used on the patient are different.

3. The method for denoising low-dose CT images in real-world scenarios as described in claim 1, characterized in that, The low-dose CT image to be optimized and the normal-dose CT image are each cropped into several image blocks. The cropping process involves cropping both the low-dose CT image to be optimized and the normal-dose CT image to be optimized according to a set size. After cropping, the positions of the low-dose CT image blocks and the positions of the normal-dose CT image blocks are matched one-to-one according to the location of the patient's body tissue to which the image blocks belong.

4. The method for denoising low-dose CT images in real-world scenarios as described in claim 1, characterized in that, The low-dose CT image block to be optimized and the corresponding normal-dose CT image block are binarized to obtain the low-dose CT image block mask and the normal-dose CT image block mask. The binarization process is as follows: Given a threshold T, when pixels in a low-dose CT image block to be optimized... If the pixel value is greater than the threshold T, then the pixel will be... The pixel value is set to 1; when the pixel in the low-dose CT image block to be optimized... If the pixel value is less than or equal to the threshold T, then the pixel... The pixel value is set to 0 to obtain the low-dose CT image block mask to be optimized; When pixels in a normal dose CT image block If the pixel value is greater than the threshold T, then the pixel will be... The pixel value is set to 1; when the pixel in a normal dose CT image block... If the pixel value is less than or equal to the threshold T, then the pixel... The pixel value is set to 0 to obtain a normal dose CT image block mask.

5. A real-world low-dose CT image denoising system, characterized by: include: The acquisition module is configured to acquire low-dose CT images to be denoised. The denoising module is configured to: input the low-dose CT image to be denoised into the trained denoising model, and output the denoised CT image; wherein, the trained denoising model is obtained by training a deep learning model using an optimized training set, and the optimized training set is obtained by: obtaining a sample set to be optimized, the sample set to be optimized including several pairs of samples to be optimized, each pair of samples to be optimized including: a low-dose CT image to be optimized and a normal-dose CT image; The low-dose CT image to be optimized and the normal-dose CT image are cropped into several image blocks respectively; the low-dose CT image block to be optimized and the corresponding normal-dose CT image block are binarized to obtain the low-dose CT image block mask to be optimized and the normal-dose CT image block mask. Based on the low-dose CT image block mask to be optimized and the normal-dose CT image block mask, the common mask of the two is calculated, specifically including: If the low-dose CT image block mask pixels to be optimized The pixel value is equal to the mask pixel value of a normal dose CT image block. The pixel value, then the common mask pixels The pixel value is 1; otherwise, the common mask pixel... The pixel value is 0; Based on the pixel values ​​of the common mask, the similarity between the low-dose CT image patch to be optimized and the normal-dose CT image patch is calculated, specifically including: The ratio of the number of pixels with a value of 1 in the common mask to the total number of pixels in the common mask is calculated, and this ratio is used as the similarity between the low-dose CT image block to be optimized and the normal-dose CT image block. The similarity is compared with a set threshold, and several low-dose CT image blocks and normal-dose CT image blocks with similarity greater than the set threshold are selected as optimization samples. Each of the selected low-dose CT image blocks and normal-dose CT image blocks is used as an optimized sample, and all optimized samples are combined into a training set.

6. An electronic device, characterized in that it comprises: Memory is used to store computer-readable instructions in a non-transitory manner. as well as Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in any one of claims 1-4.

7. A storage medium characterized by being non-transitory. The system stores computer-readable instructions, wherein, when the non-transitory computer-readable instructions are executed by a computer, the instructions of the method according to any one of claims 1-4 are executed.

8. A computer program product, characterized in that, Includes a computer program, which, when run on one or more processors, is used to implement the method described in any one of claims 1-4.

Images