CT image denoising method and system, electronic equipment and storage medium

Abstract

The invention discloses a CT image denoising method and system, electronic equipment and a storage medium, and the method comprises the steps: obtaining an unlabeled normal dose high-resolution CT image, and reducing the image resolution by a preset multiple to obtain an unlabeled normal dose low-resolution CT image; cascading the unconditional diffusion model and the conditional diffusion model to construct a cascade diffusion model; the unconditional diffusion model is trained by using an unlabeled normal dose low-resolution CT image, and the trained model is used for generating a low-resolution normal dose CT image from a random noise image; the conditional diffusion model is trained by using an unlabeled normal-dose high-resolution CT image, and the trained model generates a high-resolution normal-dose CT image under the condition of an image generated by the unconditional diffusion model; and combining the trained cascade diffusion model with a maximum posteriori framework to construct a denoising model, inputting a low-dose CT image, and iteratively solving a plurality of maximum posteriori problems by combining prior provided by the cascade diffusion model so as to realize denoising of the low-dose CT image.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a CT image denoising method, system, electronic equipment and storage medium. Background Art

[0002] Pelvic segmentation has always been an essential step in the clinical diagnosis and surgical planning of pelvic bone diseases, and the processing of CT images is a basic work in this step. Low-dose computed tomography (CT) has attracted increasing attention because it significantly reduces radiation dose. Although low-dose CT can reduce the risk of patients being exposed to X-rays, it usually results in poor quality reconstructed images with significant noise, which hinders accurate diagnosis. At present, various algorithms have been developed to improve the image quality of low-dose CT, including sinusoidal filtering, iterative reconstruction, and image denoising. Among them, low-dose CT image denoising is the most widely used because it does not involve the reconstruction process and can be applied to different CT imaging systems, which is convenient for doctors to conduct subsequent surgical planning and postoperative effect evaluation.

[0003] The success of supervised deep learning algorithms relies heavily on large amounts of paired training data, however, obtaining such data in clinical settings is challenging due to the principle of "as low as reasonably achievable". Therefore, it is crucial to develop methods that can exploit the high performance of deep neural networks without requiring large amounts of labeled data.

[0004] Some existing methods try to solve this problem using different strategies. One class of methods relies on training data obtained through special acquisition schemes. Another class of methods further relaxes the conditions of training data, but still requires a large number of low-dose CT images for training.

[0005] By exploring existing data and basic common sense, the following problems exist in low-dose CT image denoising: 1. There is currently no dataset for the metal artifact removal task, making it difficult to train conventional deep learning models.

[0006] 2. Supervised deep learning methods usually require paired low-dose and normal-dose CT images for training, and paired data are difficult to obtain in clinical settings.

[0007] 3. A patient’s CT sequence images have a depth of more than 300 images, the starting and ending positions of the artifacts are uncertain, and there is a certain similarity between adjacent anatomical structures and noise, which will cause the generative model to mistakenly generate redundant structures.

[0008] 4. The traditional generative network has the problem of poor generation effect when there is insufficient labeled data and no gain for the segmentation model. Summary of the invention

[0009] In view of the deficiencies in the prior art, the present invention provides a CT image denoising method, system, electronic device and storage medium. The denoising method only uses normal-dose CT images to train a cascade diffusion model. The trained cascade diffusion model can gradually generate realistic low-resolution normal-dose CT images and high-resolution normal-dose CT images from random noise. The prior of the diffusion model is used in combination with a maximum a posteriori framework to achieve zero-sample denoising of low-dose CT images.

[0010] The first object of the present invention is to provide a CT image denoising method, comprising: Obtaining unlabeled normal-dose high-resolution CT images from a public data set, and reducing the resolution of the unlabeled normal-dose high-resolution CT images by a preset multiple to obtain unlabeled normal-dose low-resolution CT images; Constructing a cascade diffusion model, wherein the cascade diffusion model is composed of an unconditional diffusion model and a conditional diffusion model cascaded; The unconditional diffusion model is trained with unlabeled normal-dose low-resolution CT images, and the trained unconditional diffusion model is used to generate low-resolution normal-dose CT images from random noise images; The conditional diffusion model is trained with unlabeled normal-dose high-resolution CT images, and the trained conditional diffusion model is used to generate high-resolution normal-dose CT images with low-resolution normal-dose CT images generated by the unconditional diffusion model as conditions; The trained cascade diffusion model is combined with the maximum a posteriori framework to construct a denoising model. The low-dose CT image is input into the denoising model. The denoising model combines the prior provided by the cascade diffusion model and iteratively solves multiple maximum a posteriori problems to achieve denoising of the input low-dose CT image.

[0011] As a further improvement of the present invention, the training of the unconditional diffusion model includes: Initialize the model parameters, use the unlabeled normal-dose low-resolution CT image as the model forward process input, and gradually add Gaussian noise to the input image at each time step to obtain a series of noisy images until the input image is completely converted into a random noise image; The model inverse process starts with a random noise image, predicts the noise that should be removed in the current image at each time step, gradually removes the noise of the noisy image and restores the input image, and updates the model parameters; The variational bound that minimizes the negative log-likelihood is used as the loss function. The model parameters are continuously adjusted through the optimization of the loss function. When the model performance no longer improves or reaches the preset number of iterations, the training is stopped to obtain the trained unconditional diffusion model.

[0012] As a further improvement of the present invention, the training of the conditional diffusion model includes: Initialize the model parameters, use the unlabeled normal-dose high-resolution CT image as the model forward process input, and gradually add Gaussian noise to the input image at each time step to obtain a series of noisy images until the input image is completely converted into a random noise image; The inverse process of the model starts from a random noise image, and takes the low-resolution normal-dose CT image generated by the unconditional diffusion model as a condition. At each time step, the noise that should be removed in the current image is predicted to gradually remove the noise of the noisy image and restore the input image, and the model parameters are updated. The variational bound that minimizes the negative log-likelihood is used as the loss function. The model parameters are continuously adjusted by optimizing the loss function. When the model performance no longer improves or reaches the preset number of iterations, the training is stopped to obtain the trained conditional diffusion model.

[0013] As a further improvement of the present invention, the inverse process of the trained conditional diffusion model is combined with the maximum a posteriori framework to construct a denoising model, and the low-dose CT image is input as the likelihood into the inverse process of the trained conditional diffusion model. Combined with the prior of the trained conditional diffusion model, continuous iteration is performed to achieve denoising of the input low-dose CT image.

[0014] As a further improvement of the present invention, in the iterative denoising process, the denoised image obtained in each iteration is expressed as: ; in, is the time step; is the time step Noise image at ; is the time step The hyperparameters at For is the conditional mean; is the time step The denoised image obtained by iterating at 10000 times; For the time step Noise level at the time The condition of the conditional diffusion model, i.e., the low-resolution normal-dose CT image generated by the unconditional diffusion model; For The conditional diffusion model with condition t at time step The denoised image obtained by iterating .

[0015] As a further improvement of the present invention, in the iterative denoising process, an adaptive strategy is used to adjust the hyperparameters to adapt to images with different noise levels; The adaptive strategy is to use the standard deviation of all noises in the image as the noise level to dynamically determine the hyperparameters; The formula for adjusting hyperparameters using an adaptive strategy is: ; in, is the hyperparameter adjusted by the adaptive strategy, , All parameters are manually selected. For the The standard deviation of all pixels in the image obtained by iteration.

[0016] As a further improvement of the present invention, the inverse process of the trained conditional diffusion model is accelerated by interval sampling to reduce the number of iterations.

[0017] The second object of the present invention is to provide a CT image denoising system, which is used to implement the above denoising method, comprising: An image acquisition module, which acquires an unlabeled normal-dose high-resolution CT image from a public data set, and reduces the resolution of the unlabeled normal-dose high-resolution CT image by a preset multiple to obtain an unlabeled normal-dose low-resolution CT image; A cascade diffusion model construction module, used to construct a cascade diffusion model, wherein the diffusion model is composed of an unconditional diffusion model and a conditional diffusion model cascaded; A cascade diffusion model training module, used for training the cascade diffusion model, comprising an unconditional diffusion model training submodule and a conditional diffusion model training submodule; the unconditional diffusion model training submodule uses an unlabeled normal-dose low-resolution CT image to train the unconditional diffusion model, and the trained unconditional diffusion model is used to generate a low-resolution normal-dose CT image from a random noise image; the conditional diffusion model training submodule uses an unlabeled normal-dose high-resolution CT image to train the conditional diffusion model, and the trained conditional diffusion model is used to generate a high-resolution normal-dose CT image with the low-resolution normal-dose CT image generated by the unconditional diffusion model as a condition; A denoising model building module, which is used to combine the trained cascade diffusion model with the maximum a posteriori framework to build a denoising model; The image denoising module is used to input the low-dose CT image into the denoising model. The denoising model combines the prior provided by the cascade diffusion model and iteratively solves multiple maximum a posteriori problems to achieve denoising of the input low-dose CT image.

[0018] The third object of the present invention is to provide an electronic device, comprising at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the program is executed by the processing unit, the processing unit performs the above-mentioned denoising method.

[0019] A fourth object of the present invention is to provide a storage medium storing a computer program executable by an electronic device, wherein when the program runs on the electronic device, the electronic device executes the above-mentioned denoising method.

[0020] Compared with the prior art, the present invention has the following beneficial effects: When training the diffusion model, no paired normal-dose and low-dose CT images are required, nor are any low-dose CT images required as training sets. Only normal-dose CT images are used to train the cascade diffusion model. The trained cascade diffusion model can gradually generate realistic low-resolution normal-dose CT images and high-resolution normal-dose CT images from random noise.

[0021] By combining the cascade diffusion model with the maximum a posteriori framework and utilizing the prior of the diffusion model, multiple maximum a posteriori problems are iteratively solved to achieve zero-sample denoising of low-dose CT images, thereby improving the quality of low-dose CT images.

[0022] The denoising method provided in the present application utilizes a cascade diffusion model and a maximum a posteriori framework technique to train and generate artifact-free low-dose pelvic CT images without a control data set, so as to improve the segmentation effect of 3D pelvic CT images during and after surgery. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a denoising algorithm provided by the present invention; Figure 2 It is a schematic diagram of the reverse process of the unconditional diffusion model; Figure 3 It is a schematic diagram of the reverse process of the conditional diffusion model; Figure 4 Schematic diagram of the denoising process of the denoising model; Figure 5 A schematic diagram of adjusting hyperparameters using an adaptive strategy; Figure 6 The present invention provides a flow chart of the denoising system. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] The present invention is further described in detail below in conjunction with the accompanying drawings: See also Figure 1 This embodiment provides a CT image denoising method, comprising: Obtaining unlabeled normal-dose high-resolution CT images from a public data set, and reducing the resolution of the unlabeled normal-dose high-resolution CT images by a preset multiple to obtain unlabeled normal-dose low-resolution CT images; Constructing a cascade diffusion model, wherein the cascade diffusion model is composed of an unconditional diffusion model and a conditional diffusion model cascaded; The unconditional diffusion model is trained with unlabeled normal-dose low-resolution CT images, and the trained unconditional diffusion model is used to generate low-resolution normal-dose CT images from random noise images; The conditional diffusion model is trained with unlabeled normal-dose high-resolution CT images, and the trained conditional diffusion model is used to generate high-resolution normal-dose CT images with low-resolution normal-dose CT images generated by the unconditional diffusion model as conditions; The trained cascade diffusion model is combined with the maximum a posteriori framework to construct a denoising model. The low-dose CT image is input into the denoising model. The denoising model combines the prior provided by the cascade diffusion model and iteratively solves multiple maximum a posteriori problems to achieve denoising of the input low-dose CT image.

[0026] Given a set of normal-dose high-resolution CT images , ,Will The resolution is reduced Get low-resolution images at normal doses , get the image pair ,in , and based on this, two models are defined.

[0027] The first model is a model with positive process parameters and reverse process parameters An unconditional diffusion model for generating low-resolution normal-dose CT images from random noise.

[0028] The second diffusion model is the conditional diffusion model, which uses the normal-dose low-resolution CT image generated by the unconditional diffusion model as a condition to achieve normal-dose CT image super-resolution. Its forward process parameters are , the reverse process parameters are .

[0029] By cascading two diffusion models, a normal-dose low-resolution CT image is first generated from random noise, and then the generated normal-dose low-resolution CT image is used as a condition in the conditional diffusion model to generate a normal-dose high-resolution CT image.

[0030] When training the unconditional diffusion model: Will As the input data of the model, it is used to learn the distribution characteristics and generation process of the data, where It is an unlabeled normal-dose low-resolution CT image, which is usually clean data without noise pollution. for data distribution.

[0031] The forward process of the unconditional diffusion model: step by step to input data Add Gaussian noise to get a series of gradually noisy data , until Completely converted into random noise data .

[0032] In the forward process, each time step (from 1 to ), are based on the data of the previous time step Generate data for the current time step , the forward process of the unconditional diffusion model It is defined as a Markov chain (Formula 1), and the formula is: ; in, is the time step Noisy data at is the time step Noisy data at is the predefined noise variance, is the identity matrix, is a Gaussian distribution.

[0033] The reverse process of the unconditional diffusion model: from random noise data At the beginning, the noise that should be removed in the current image is predicted at each time step to gradually remove the noise of the noisy image and finally restore the original input data .

[0034] The inverse process of the unconditional diffusion model is defined as having learnable parameters (reverse process parameters) of the Markov chain (Formula 2), the reverse process is a series of conditional probability distributions To achieve this, the formula is: ; in, is the mean, usually predicted by U-nets; For the time step The noise level at For the time step The variance of the noise level when Defined as , , Represents the time step , The forward process parameters of ; the training loss is obtained by minimizing the variational bound of the negative log-likelihood get.

[0035] By optimizing the loss function, the parameters of the unconditional diffusion model are continuously adjusted. When the model performance no longer improves or reaches the preset number of iterations, the training is stopped to obtain the trained unconditional diffusion model for generating normal-dose low-resolution CT images from random noise images.

[0036] When training the conditional diffusion model: Will As the input data of the model, it is used to learn the distribution characteristics and generation process of the data, where It is an unlabeled normal-dose high-resolution CT image (resolution is 512*512), which is usually clean data without noise pollution. for data distribution.

[0037] The forward process of the conditional diffusion model: step by step to input data Add Gaussian noise to get a series of gradually noisy data , until Completely converted into random noise data , which is the same as the forward process of the unconditional diffusion model.

[0038] The reverse process of the conditional diffusion model: from random noise data At the beginning, the low-resolution normal-dose CT image generated by the unconditional diffusion model is used as a condition to predict the noise that should be removed in the current image at each time step, so as to gradually remove the noise of the noisy image and finally restore the original input data. .

[0039] The reverse process of the conditional diffusion model is defined as (Formula 3): ; in, is the condition of the conditional diffusion model, i.e., the low-resolution normal-dose CT images generated by the unconditional diffusion model; the training loss is obtained by minimizing the variational bound of the negative log-likelihood get.

[0040] By optimizing the loss function, the parameters of the conditional diffusion model are continuously adjusted. When the model performance no longer improves or reaches the preset number of iterations, the training is stopped to obtain the trained normal-dose low-resolution CT images generated by the unconditional diffusion model, and then a conditional diffusion model of normal-dose high-resolution CT images is generated.

[0041] Two diffusion models are cascaded to generate normal-dose CT images. The first model is an unconditional diffusion model, which generates normal-dose low-resolution CT images from random noise. The generation process is as follows: Figure 2 As shown; the second model is a conditional diffusion model, which generates a high-resolution normal-dose CT image of size 512×512 based on the normal-dose low-resolution CT image generated by the unconditional diffusion model. The generation process is as follows Figure 3 shown.

[0042] When training the diffusion model, no paired normal-dose and low-dose CT images are required, nor are any low-dose CT images required as training sets. Only normal-dose CT images are used to train the cascade diffusion model, and the trained cascade diffusion model can gradually generate realistic low-resolution normal-dose CT images and high-resolution normal-dose CT images from random noise. The entire model only uses unlabeled normal-dose CT images for unsupervised training, effectively solving the problem that the existing technology requires paired low-dose CT images and normal-dose CT images for training, and paired images are difficult to obtain.

[0043] After the cascade diffusion model training is completed, the inverse process of the trained conditional diffusion model is combined with the maximum a posteriori framework to construct a denoising model. The low-dose CT image is input as the likelihood into the inverse process of the trained conditional diffusion model. Combined with the prior of the trained conditional diffusion model, the input low-dose CT image is denoised through continuous iteration. The denoising process is as follows: Figure 4 shown.

[0044] In the image denoising problem, the noisy image is modeled as follows (Formula 4): ; in, is the input noisy image, This is the bottom clean image we want to get.

[0045] To achieve denoising of low-dose CT images, a well-trained diffusion model is used to capture the prior distribution of the underlying clean image. The diffusion model uses the forward process parameters and reverse process parameters , accurately capture The goal is to use the diffusion prior to effectively Perform noise reduction.

[0046] The specific methods are outlined as follows: In diffusion models, the time step Image According to the forward process, Generate, Image The calculation formula (Formula 5) is: ; in, Obtained from a normal distribution.

[0047] In order to introduce a noisy image into the input At each time step The likelihood of From the input noisy image Generate a denoised image at the current time step , time step Image The calculation formula (Formula 6) is: ; Substitute Equation 3 and Equation 4 into Equation 5 and set ,for The following equation is obtained (Formula 7): ; Formula 7 shows that at each time step , there is a new noise denoising problem.

[0048] because along with decreases with the increase of hour, , so we can use as initialization; then for , we can transform the diffusion prior Incorporate into the Maximum A Posteriori (MAP) framework to solve , the calculation formula (Formula 8) is: ; in, is the trade-off time step The probability and prior coefficients at . In the inverse process of the diffusion model, Follow the mean and the diagonal covariance matrix Gaussian distribution.

[0049] when , , we get formula 9: ; when , , indicating that the final generation step of diffusion is deterministic, and we get formula 10: ; Formula 9 is a convex optimization problem with a closed-form solution. By solving 9 and combining it with Formula 10, we can derive The solution (Formula 11): ; Equation 11 illustrates the denoising process using a pre-trained unconditional diffusion model.

[0050] Similarly, if the pre-trained diffusion model is As a condition, The solution becomes (Formula 12): ; in, is the time step; is the time step Noise image at ; is the time step The hyperparameters at For is the conditional mean; is the time step The denoised image obtained by iterating at 10000 times; For the time step Noise level at the time The condition of the conditional diffusion model, i.e., the low-resolution normal-dose CT image generated by the unconditional diffusion model; For The conditional diffusion model with condition t at time step The denoised image obtained by iterating .

[0051] Formula 12 illustrates the denoising process using the pre-trained conditional diffusion model, that is, in this embodiment, the normal dose low-resolution CT image generated by the unconditional diffusion model is used as a condition, and the expression of the denoised image obtained at each iteration in the iterative denoising process.

[0052] The inverse process of the trained conditional diffusion model is combined with the maximum a posteriori framework. The prior of the conditional diffusion model is used to iteratively solve multiple maximum a posteriori problems to achieve zero-sample denoising of low-dose CT images, thereby improving the quality of low-dose CT images.

[0053] During the iterative denoising process, an adaptive strategy is used to adjust the hyperparameters (i.e., coefficients that weigh the likelihood and the prior) to balance the likelihood and the prior in MAP estimation to accommodate images with different noise levels.

[0054] Considering that different regions of low-dose CT images exhibit different levels of noise, the adaptive strategy estimates the standard deviation of all noise in the image as the noise level to dynamically determine the hyperparameters; The formula for adjusting hyperparameters using an adaptive strategy is: ; in, is the hyperparameter adjusted by the adaptive strategy, , All parameters are manually selected. For the The process of refining hyperparameters is as follows: Figure 5 shown.

[0055] By using dynamic coefficients to resume the denoising process at intermediate time steps, the denoising model can adaptively process low-dose CT images with different noise levels.

[0056] In the denoising process, the inverse process of the trained conditional diffusion model is accelerated by interval sampling to reduce the number of iterations and speed up the denoising process of low-dose CT images.

[0057] Specifically, we start from The length extracted is subsequence of , then execute from arrive The sampling of , the denoised image is expressed as: ; This non-uniform sampling method is used because the likelihood term plays a more prominent role in smaller time steps, and denser sampling can better utilize the input of low-dose CT images, resulting in more accurate denoising results.

[0058] The denoising method provided in this embodiment uses a cascade diffusion model and a maximum a posteriori framework technology to train and generate artifact-free low-dose pelvic CT images without a control data set (paired low-dose CT images and normal-dose CT images), so as to improve the segmentation effect of 3D pelvic CT images during and after surgery.

[0059] See also Figure 6 This embodiment provides a CT image denoising system, which is used to implement the above denoising method, including: An image acquisition module, which acquires an unlabeled normal-dose high-resolution CT image from a public data set, and reduces the resolution of the unlabeled normal-dose high-resolution CT image by a preset multiple to obtain an unlabeled normal-dose low-resolution CT image; A cascade diffusion model construction module, used to construct a cascade diffusion model, wherein the diffusion model is composed of an unconditional diffusion model and a conditional diffusion model cascaded; A cascade diffusion model training module, used for training the cascade diffusion model, comprising an unconditional diffusion model training submodule and a conditional diffusion model training submodule; the unconditional diffusion model training submodule uses an unlabeled normal-dose low-resolution CT image to train the unconditional diffusion model, and the trained unconditional diffusion model is used to generate a low-resolution normal-dose CT image from a random noise image; the conditional diffusion model training submodule uses an unlabeled normal-dose high-resolution CT image to train the conditional diffusion model, and the trained conditional diffusion model is used to generate a high-resolution normal-dose CT image with the low-resolution normal-dose CT image generated by the unconditional diffusion model as a condition; A denoising model building module, which is used to combine the trained cascade diffusion model with the maximum a posteriori framework to build a denoising model; The image denoising module is used to input the low-dose CT image into the denoising model. The denoising model combines the prior provided by the cascade diffusion model and iteratively solves multiple maximum a posteriori problems to achieve denoising of the input low-dose CT image.

[0060] The denoising system provided in this embodiment uses a cascade diffusion model and a maximum a posteriori framework technology to train and generate artifact-free low-dose pelvic CT images without a control data set (paired low-dose CT images and normal-dose CT images), so as to improve the segmentation effect of 3D pelvic CT images during and after surgery.

[0061] This embodiment provides an electronic device, including at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the program is executed by the processing unit, the processing unit performs the above-mentioned denoising method.

[0062] This embodiment provides a storage medium storing a computer program executable by an electronic device. When the program runs on the electronic device, the electronic device executes the above-mentioned denoising method.

[0063] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A CT image denoising method, characterized in that: include: Obtaining unlabeled normal-dose high-resolution CT images from a public data set, and reducing the resolution of the unlabeled normal-dose high-resolution CT images by a preset multiple to obtain unlabeled normal-dose low-resolution CT images; Constructing a cascade diffusion model, wherein the cascade diffusion model is composed of an unconditional diffusion model and a conditional diffusion model cascaded; The unconditional diffusion model is trained with unlabeled normal-dose low-resolution CT images, and the trained unconditional diffusion model is used to generate low-resolution normal-dose CT images from random noise images; The conditional diffusion model is trained with unlabeled normal-dose high-resolution CT images, and the trained conditional diffusion model is used to generate high-resolution normal-dose CT images with low-resolution normal-dose CT images generated by the unconditional diffusion model as conditions; The trained cascade diffusion model is combined with the maximum a posteriori framework to construct a denoising model. The low-dose CT image is input into the denoising model. The denoising model combines the prior provided by the cascade diffusion model and iteratively solves multiple maximum a posteriori problems to achieve denoising of the input low-dose CT image.

2. A CT image denoising method according to claim 1, characterized in that: The training of the unconditional diffusion model includes: Initialize the model parameters, use the unlabeled normal-dose low-resolution CT image as the model forward process input, and gradually add Gaussian noise to the input image at each time step to obtain a series of noisy images until the input image is completely converted into a random noise image; The model inverse process starts with a random noise image, predicts the noise that should be removed in the current image at each time step, gradually removes the noise of the noisy image and restores the input image, and updates the model parameters; The variational bound that minimizes the negative log-likelihood is used as the loss function. The model parameters are continuously adjusted through the optimization of the loss function. When the model performance no longer improves or reaches the preset number of iterations, the training is stopped to obtain the trained unconditional diffusion model.

3. A CT image denoising method according to claim 1, characterized in that: The training of the conditional diffusion model includes: Initialize the model parameters, use the unlabeled normal-dose high-resolution CT image as the model forward process input, and gradually add Gaussian noise to the input image at each time step to obtain a series of noisy images until the input image is completely converted into a random noise image; The inverse process of the model starts from a random noise image, and takes the low-resolution normal-dose CT image generated by the unconditional diffusion model as a condition. At each time step, the noise that should be removed in the current image is predicted to gradually remove the noise of the noisy image and restore the input image, and the model parameters are updated. The variational bound that minimizes the negative log-likelihood is used as the loss function. The model parameters are continuously adjusted by optimizing the loss function. When the model performance no longer improves or reaches the preset number of iterations, the training is stopped to obtain the trained conditional diffusion model.

4. A CT image denoising method according to claim 1, characterized in that: The inverse process of the trained conditional diffusion model is combined with the maximum a posteriori framework to construct a denoising model. The low-dose CT image is input as the likelihood into the inverse process of the trained conditional diffusion model. Combined with the prior of the trained conditional diffusion model, it is continuously iterated to achieve denoising of the input low-dose CT image.

5. A CT image denoising method according to claim 4, characterized in that: In the iterative denoising process, the denoised image obtained at each iteration is expressed as: ； in, is the time step; is the time step Noise image at ; is the time step The hyperparameters at For is the conditional mean; is the time step The denoised image obtained by iterating at 10000 times; For the time step Noise level at the time The condition of the conditional diffusion model, i.e., the low-resolution normal-dose CT image generated by the unconditional diffusion model; For The conditional diffusion model with condition t at time step The denoised image obtained by iterating .

6. A CT image denoising method according to claim 4 or 5, characterized in that: In the iterative denoising process, an adaptive strategy is used to adjust the hyperparameters to adapt to images with different noise levels; The adaptive strategy is to use the standard deviation of all noises in the image as the noise level to dynamically determine the hyperparameters; The formula for adjusting hyperparameters using an adaptive strategy is: ； in, is the hyperparameter adjusted by the adaptive strategy, , All parameters are manually selected. For the The standard deviation of all pixels in the image obtained by iteration.

7. A CT image denoising method according to claim 6, characterized in that: Speed ​​up the inverse process of the trained conditional diffusion model by sampling at intervals to reduce the number of iterations.

8. A CT image denoising system, used to implement the denoising method according to any one of claims 1 to 7, characterized in that: include: An image acquisition module, which acquires an unlabeled normal-dose high-resolution CT image from a public data set, and reduces the resolution of the unlabeled normal-dose high-resolution CT image by a preset multiple to obtain an unlabeled normal-dose low-resolution CT image; A cascade diffusion model construction module, used to construct a cascade diffusion model, wherein the diffusion model is composed of an unconditional diffusion model and a conditional diffusion model cascaded; A cascade diffusion model training module, used for training the cascade diffusion model, comprising an unconditional diffusion model training submodule and a conditional diffusion model training submodule; the unconditional diffusion model training submodule uses an unlabeled normal-dose low-resolution CT image to train the unconditional diffusion model, and the trained unconditional diffusion model is used to generate a low-resolution normal-dose CT image from a random noise image; the conditional diffusion model training submodule uses an unlabeled normal-dose high-resolution CT image to train the conditional diffusion model, and the trained conditional diffusion model is used to generate a high-resolution normal-dose CT image with the low-resolution normal-dose CT image generated by the unconditional diffusion model as a condition; A denoising model building module, which is used to combine the trained cascade diffusion model with the maximum a posteriori framework to build a denoising model; The image denoising module is used to input the low-dose CT image into the denoising model. The denoising model combines the prior provided by the cascade diffusion model and iteratively solves multiple maximum a posteriori problems to achieve denoising of the input low-dose CT image.

9. An electronic device, characterized in that: The method comprises at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the program is executed by the processing unit, the processing unit executes the denoising method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: It stores a computer program executable by an electronic device. When the program runs on the electronic device, the electronic device executes the denoising method as described in any one of claims 1 to 7.

Images