Image denoising method and device based on generative artificial intelligence technology

Abstract

The present invention provides an image denoising method and device based on generative artificial intelligence technology, comprising: inputting a medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised; inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; the noise type recognition model is obtained after training based on samples of medical images to be denoised and predetermined noise classification labels, and the samples of medical images to be denoised are obtained by adding noise based on clear medical image samples of different types; the image denoising model is obtained after training based on clear medical image samples of different types, samples of medical images to be denoised and predetermined noise classification labels, and can simultaneously meet the efficiency and accuracy requirements of denoising of different types of medical images.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image denoising method and device based on generative artificial intelligence technology. Background Art

[0002] Medical imaging plays an important role in the diagnosis and treatment of various diseases. However, interpreting medical images is challenging even for professional clinicians because they are often affected by noise and artifacts, which hinder the accurate identification and analysis of diseases. Deep learning has become a powerful tool for image analysis and has shown certain potential in denoising different medical images.

[0003] Existing deep learning-based medical image denoising solutions can usually only denoise a certain type of medical image (such as X-ray images or ultrasound images). However, hospitals produce a wide variety of medical images, and existing deep learning-based medical image denoising solutions cannot simultaneously meet the efficiency and accuracy requirements of denoising different types of medical images. Summary of the invention

[0004] The present invention provides an image denoising method and device based on generative artificial intelligence technology, which are used to solve the problem that the existing technology cannot simultaneously meet the efficiency and accuracy requirements of denoising of different types of medical images.

[0005] In one aspect, the present invention provides an image denoising method based on generative artificial intelligence technology, comprising:

[0006] Inputting the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised;

[0007] Inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network;

[0008] The noise type recognition model is obtained after training based on the medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on different types of clear medical image samples; the image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels.

[0009] In an optional embodiment of the present application, the step of generating the medical image sample to be denoised includes:

[0010] Get clear medical image samples of different types;

[0011] Based on a pre-established mapping relationship table between medical image types and noise types, noise adding operations are performed on different types of clear medical image samples to obtain different types of medical image samples to be denoised.

[0012] In an optional embodiment of the present application, the image denoising model includes a generator and a discriminator, and the generator includes:

[0013] Shared feature extraction unit, multi-branch denoising unit and output unit;

[0014] The shared feature extraction unit is used to receive the medical image to be denoised, extract the underlying features of the medical image to be denoised and generate a feature map corresponding to the medical image to be denoised;

[0015] The multi-branch denoising unit is used to enable multiple target denoising branches to perform denoising processing on the feature map corresponding to the medical image to be denoised in parallel based on the noise classification result corresponding to the medical image to be denoised;

[0016] The output unit is used to integrate the outputs of the target denoising branches to generate a denoised medical image corresponding to the medical image to be denoised.

[0017] In an optional embodiment of the present application, integrating the outputs of the target denoising branches to generate a denoised medical image corresponding to the medical image to be denoised specifically includes:

[0018] The outputs of the target denoising branches are weighted and summed based on the weights of the target denoising branches to obtain a denoised medical image corresponding to the medical image to be denoised.

[0019] In an optional embodiment of the present application, the weight of the target denoising branch is obtained based on the final loss and performance index of each denoising branch after the multi-branch denoising unit is trained.

[0020] In an optional embodiment of the present application, the training process of the multi-branch denoising unit includes:

[0021] Training the multi-branch denoising unit based on the medical image samples to be denoised and the predetermined noise classification labels;

[0022] At the end of each training cycle, the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit are calculated, and whether the training termination condition is reached based on the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit;

[0023] If yes, stop training; otherwise, update the weights of each denoising branch and continue with subsequent training.

[0024] In an optional embodiment of the present application, updating the weights of each denoising branch specifically includes:

[0025] Determine the independent loss of each current denoising branch;

[0026] Based on the independent loss of each current denoising branch, the weight of each denoising branch is updated using the inverse proportional weight method.

[0027] In a second aspect, the present invention further provides an image denoising device based on generative artificial intelligence technology, comprising:

[0028] A noise classification module, used for inputting the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised;

[0029] An image denoising module, used for inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network;

[0030] The noise type recognition model is obtained after training based on the medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on different types of clear medical image samples; the image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels.

[0031] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the image denoising method based on generative artificial intelligence technology as described above is implemented.

[0032] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the image denoising method based on generative artificial intelligence technology as described above.

[0033] The present invention provides an image denoising method and device based on generative artificial intelligence technology, which inputs a medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised; inputs the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; the noise type recognition model is obtained after training based on medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on clear medical image samples of different types; the image denoising model is obtained after training based on clear medical image samples of different types, the medical image samples to be denoised and predetermined noise classification labels, and can simultaneously meet the efficiency and accuracy requirements of denoising of different types of medical images. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A schematic diagram of a process of an image denoising method based on generative artificial intelligence technology provided by the present invention;

[0036] Figure 2 A schematic diagram of the generation process of the medical image samples to be denoised provided by the present invention;

[0037] Figure 3 A schematic diagram of the training process of the multi-branch denoising unit provided by the present invention;

[0038] Figure 4 A structural block diagram of an image denoising device based on generative artificial intelligence technology provided by the present invention;

[0039] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0041] The specific implementation process of the image denoising method and device based on generative artificial intelligence technology in an embodiment of the present invention is described below.

[0042] Figure 1 A flowchart of an image denoising method based on generative artificial intelligence technology provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method may include:

[0043] Step 101: input the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised.

[0044] Step 102, inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network;

[0045] The noise type recognition model is obtained after training based on the medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on different types of clear medical image samples; the image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels.

[0046] Specifically, it can be understood that the medical image to be denoised can be any type of medical image, including but not limited to X-ray images, computed tomography (CT) images, magnetic resonance images (MRI), ultrasonic images, positron emission tomography (PET), nuclear medicine images, optical coherence tomography (OCT) images, digital subtraction angiography (DSA) images and electrophysiological images. The present application embodiment has found through research that the noise types corresponding to medical images include but are not limited to Gaussian noise, salt and pepper noise, shot noise, quantization noise, low-frequency noise, periodic noise, artifacts, and high-frequency noise. Although there are differences in the noise types corresponding to different types of medical images, the denoising methods for the same type of noise can be reused. Based on this, the image denoising method based on generative artificial intelligence technology proposed in the embodiment of the present application can perform targeted denoising operations based on the noise type, which can not only improve the accuracy of the denoising operation, but also does not need to distinguish the type of medical image. Based on this concept, the embodiment of the present application first needs to input the medical image to be denoised into the noise type recognition model to obtain the noise classification result corresponding to the medical image to be denoised, so as to perform targeted denoising operations later.

[0047] It is understandable that the noise type recognition model is obtained after training based on the medical image samples to be denoised and the predetermined noise classification labels. It is worth noting that, considering that the existing medical image acquisition involves patient privacy issues, resulting in insufficient samples for training, the medical image samples to be denoised described in the embodiment of the present application are obtained by adding noise to clear medical image samples of different types. That is, the embodiment of the present application only needs to obtain a small number of clear medical images of different types to generate a large number of training samples. More specifically, Figure 2 A schematic diagram of the generation process of the medical image samples to be denoised provided by the present invention is shown in FIG. Figure 2 As shown, the step of generating the medical image sample to be denoised includes:

[0048] Step 201, obtaining different types of clear medical image samples;

[0049] Step 202 : Based on a pre-established mapping relationship table between medical image types and noise types, noise adding operations are performed on different types of clear medical image samples to obtain different types of medical image samples to be denoised.

[0050] It is understandable that, since the sample data volume of different types of clear medical images (i.e., noise-free medical images) is small, it can be directly obtained from a public medical image database or a database of a cooperative hospital, which greatly reduces the difficulty of obtaining training samples. At the same time, the embodiment of the present application pre-establishes a mapping relationship table between medical image types and noise types based on the mapping relationship between medical image types and noise types. Based on this, after obtaining different types of clear medical image samples, the embodiment of the present application can perform noise addition operations on different types of clear medical image samples to obtain different types of medical image samples to be denoised. Specifically, different types of noise models can be used to implement random noise addition operations to obtain a sufficient number of different types of medical image samples to be denoised, thereby maximizing the adequacy of the noise type recognition model and the image denoising model training process and the accuracy of the trained model.

[0051] The image denoising model preferably uses a conditional generative adversarial network, which has good performance in image tasks such as image generation and style transfer. Specifically, the image denoising model includes a generator and a discriminator, and the generator includes:

[0052] Shared feature extraction unit, multi-branch denoising unit and output unit;

[0053] The shared feature extraction unit is used to receive the medical image to be denoised, extract the underlying features of the medical image to be denoised and generate a feature map corresponding to the medical image to be denoised;

[0054] The multi-branch denoising unit is used to enable multiple target denoising branches to perform denoising processing on the feature map corresponding to the medical image to be denoised in parallel based on the noise classification result corresponding to the medical image to be denoised;

[0055] The output unit is used to integrate the outputs of the target denoising branches to generate a denoised medical image corresponding to the medical image to be denoised.

[0056] Based on this, the embodiment of the present application can perform targeted denoising operations on various types of noise in the denoised medical image in parallel, which can not only improve the denoising efficiency, but also ensure the accuracy of the denoising result. Specifically, the output of the target denoising branch is integrated to generate a denoised medical image corresponding to the denoised medical image, which specifically includes:

[0057] The outputs of the target denoising branches are weighted and summed based on the weights of the target denoising branches to obtain a denoised medical image corresponding to the medical image to be denoised.

[0058] That is, assuming there are M target denoising branches, the denoised image generated by each target denoising branch is Output i , where i=1,2,…,M. Assign a weight w to each branch i , the sum of the weights should be 1. The final denoised image Final_Output can be obtained by weighted average calculation, that is:

[0059] .

[0060] It is worth noting that the weight of the target denoising branch is obtained based on the final loss and performance index of each denoising branch after the multi-branch denoising unit is trained. Specifically, after the training is completed, the final loss of each denoising branch on the validation set needs to be calculated first. Assume that there are N denoising branches, marked as Loss1, Loss2, ..., Loss N , then for the denoising branch i, the weight calculation formula corresponding to its final loss is:

[0061] ;

[0062] In addition to the loss, it is also necessary to calculate the performance indicators of each denoising branch (such as PSNR, SSIM). Assume that P1, P2, ..., P M Represents the performance index of each denoising branch. First, each performance index needs to be standardized. Assuming that the higher the performance index, the better, the standardized performance index of the denoising branch i can be calculated using the following formula:

[0063] ;

[0064] Among them, Pmin and P max are the minimum and maximum values ​​of all denoising branch performance indicators, respectively. Based on this, for denoising branch i, the weight calculation formula corresponding to its performance indicator is:

[0065] .

[0066] Based on this, the final weight can be determined by combining the loss and performance indicators to balance the impact of the two. The specific weight calculation formula for the denoising branch i is:

[0067] .

[0068] Among them, the hyperparameter α (between 0 and 1) is used to balance the impact of loss and performance indicators. The value of the hyperparameter α can be selected using existing methods, such as grid search or Bayesian optimization, etc., and the embodiment of the present application is not specifically limited here.

[0069] Based on the above weight setting scheme, the embodiment of the present application can maximize the denoising performance.

[0070] The following is an introduction to the training process of the multi-branch denoising unit in the embodiment of the present application. Figure 3 A schematic diagram of the training process of the multi-branch denoising unit provided by the present invention is shown in FIG. Figure 3 As shown, the training process of the multi-branch denoising unit includes:

[0071] Step 301, training the multi-branch denoising unit based on the medical image samples to be denoised and the predetermined noise classification labels;

[0072] Step 302, at the end of each training cycle, calculating the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit, and judging whether the training termination condition is reached based on the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit;

[0073] Step 303: If yes, stop training; otherwise, update the weights of each denoising branch and continue subsequent training.

[0074] It can be understood that the multi-branch denoising unit is trained based on the medical image samples to be denoised and the predetermined noise classification labels, that is, the medical image samples to be denoised are input into the trained shared feature extraction unit to obtain the feature map corresponding to the medical image to be denoised, and then the multi-branch denoising unit is trained based on the feature map corresponding to the medical image to be denoised and the predetermined noise classification labels. The shared feature extraction unit can adopt an existing image feature extraction network, which is not specifically limited in the embodiments of the present application.

[0075] It is worth noting that in order to maximize the performance of the trained multi-branch denoising unit, the embodiment of the present application sets two loss functions, namely, the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit. Specifically, the independent loss can be calculated using the mean square error. The combined loss is the weighted combination of the independent losses of all branches to form a comprehensive loss function. The calculation formula of the combined loss is as follows:

[0076] ;

[0077] Among them, w i is the weight of each denoising branch.

[0078] The training termination condition is that the change of the independent loss and the combined loss in several consecutive training cycles is less than a preset threshold. Based on this, it can be ensured that each denoising branch and the entire multi-branch denoising unit converge. If the training termination condition is not met, the embodiment of the present application updates the weights of each denoising branch and continues the subsequent training. Specifically, the updating of the weights of each denoising branch specifically includes:

[0079] Determine the independent loss of each current denoising branch;

[0080] Based on the independent loss of each current denoising branch, the weight of each denoising branch is updated using the inverse proportional weight method.

[0081] More specifically, the specific calculation formula for updating the weights of each denoising branch using the inverse proportional weight method is:

[0082] ;

[0083] Based on this, it is possible to ensure that the branch with smaller loss will obtain a larger weight and thus occupy a more important position in the combined loss.

[0084] The solution provided by the present invention is to input the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised; input the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; the noise type recognition model is obtained after training based on medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on clear medical image samples of different types; the image denoising model is obtained after training based on clear medical image samples of different types, the medical image samples to be denoised and predetermined noise classification labels, and can simultaneously meet the efficiency and accuracy requirements of denoising of different types of medical images.

[0085] Figure 4A structural block diagram of an image denoising device based on generative artificial intelligence technology is provided for an embodiment of the present invention, such as Figure 4 As shown, the system includes:

[0086] The noise classification module 401 is used to input the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised;

[0087] An image denoising module 402 is used to input the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network;

[0088] The noise type recognition model is obtained after training based on the medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on different types of clear medical image samples; the image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels.

[0089] The solution provided by the present invention is as follows: the medical image to be denoised is input into a noise type recognition model through a noise classification module 401 to obtain a noise classification result corresponding to the medical image to be denoised; the image denoising module 402 inputs the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; the noise type recognition model is obtained after training based on the medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on different types of clear medical image samples; the image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels, and can simultaneously meet the efficiency and accuracy requirements of denoising of different types of medical images.

[0090] In an optional embodiment of the present application, the image denoising device based on generative artificial intelligence technology further includes a medical image sample generation module to be denoised, and the medical image sample generation module to be denoised is used to perform the following operations:

[0091] Get clear medical image samples of different types;

[0092] Based on a pre-established mapping relationship table between medical image types and noise types, noise adding operations are performed on different types of clear medical image samples to obtain different types of medical image samples to be denoised.

[0093] In an optional embodiment of the present application, the image denoising model includes a generator and a discriminator, and the generator includes:

[0094] Shared feature extraction unit, multi-branch denoising unit and output unit;

[0095] The shared feature extraction unit is used to receive the medical image to be denoised, extract the underlying features of the medical image to be denoised and generate a feature map corresponding to the medical image to be denoised;

[0096] The multi-branch denoising unit is used to enable multiple target denoising branches to perform denoising processing on the feature map corresponding to the medical image to be denoised in parallel based on the noise classification result corresponding to the medical image to be denoised;

[0097] The output unit is used to integrate the outputs of the target denoising branches to generate a denoised medical image corresponding to the medical image to be denoised.

[0098] In an optional embodiment of the present application, integrating the outputs of the target denoising branches to generate a denoised medical image corresponding to the medical image to be denoised specifically includes:

[0099] The outputs of the target denoising branches are weighted and summed based on the weights of the target denoising branches to obtain a denoised medical image corresponding to the medical image to be denoised.

[0100] In an optional embodiment of the present application, the weight of the target denoising branch is obtained based on the final loss and performance index of each denoising branch after the multi-branch denoising unit is trained.

[0101] In an optional embodiment of the present application, the image denoising device based on generative artificial intelligence technology further includes a multi-branch denoising unit training module, and the multi-branch denoising unit training module is used to perform the following operations:

[0102] Training the multi-branch denoising unit based on the medical image samples to be denoised and the predetermined noise classification labels;

[0103] At the end of each training cycle, the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit are calculated, and whether the training termination condition is reached based on the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit;

[0104] If yes, stop training; otherwise, update the weights of each denoising branch and continue with subsequent training.

[0105] In an optional embodiment of the present application, updating the weights of each denoising branch specifically includes:

[0106] Determine the independent loss of each current denoising branch;

[0107] Based on the independent loss of each current denoising branch, the weight of each denoising branch is updated using the inverse proportional weight method.

[0108] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor (processor) 501, a communication interface (Communications Interface) 502, a memory (memory) 503 and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504. The processor 501 can call the logic instructions in the memory 503 to execute an image denoising method based on generative artificial intelligence technology, which includes: inputting the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised; inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; the noise type recognition model is obtained after training based on medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on different types of clear medical image samples; the image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels.

[0109] In addition, the logic instructions in the above-mentioned memory 503 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0110] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the image denoising method based on generative artificial intelligence technology provided by the above-mentioned methods, and the method includes: inputting the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised; inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; the noise type recognition model is obtained after training based on medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on different types of clear medical image samples; the image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels.

[0111] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the image denoising method based on generative artificial intelligence technology provided by the above-mentioned methods, the method comprising: inputting the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised; inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; the noise type recognition model is obtained after training based on medical image samples to be denoised and predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise based on different types of clear medical image samples; the image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels.

[0112] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0113] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution can essentially or contribute to the prior art in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the image denoising method based on generative artificial intelligence technology described in each embodiment or some parts of the embodiment.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An image denoising method based on generative artificial intelligence technology, characterized in that: include: Inputting the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised; Inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; The noise type recognition model is obtained after training based on the medical image samples to be denoised and the predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise to clear medical image samples of different types; The image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels; The image denoising model includes a generator and a discriminator, and the generator includes: Shared feature extraction unit, multi-branch denoising unit and output unit; The shared feature extraction unit is used to receive the medical image to be denoised, extract the underlying features of the medical image to be denoised and generate a feature map corresponding to the medical image to be denoised; The multi-branch denoising unit is used to enable multiple target denoising branches to perform denoising processing on the feature map corresponding to the medical image to be denoised in parallel based on the noise classification result corresponding to the medical image to be denoised; The output unit is used to integrate the outputs of the target denoising branches to generate a denoised medical image corresponding to the medical image to be denoised.

2. The image denoising method based on generative artificial intelligence technology according to claim 1, characterized in that: The step of generating the medical image sample to be denoised comprises: Get clear medical image samples of different types; Based on a pre-established mapping relationship table between medical image types and noise types, noise adding operations are performed on different types of clear medical image samples to obtain different types of medical image samples to be denoised.

3. The image denoising method based on generative artificial intelligence technology according to claim 2, characterized in that: The step of integrating the outputs of the target denoising branches to generate a denoised medical image corresponding to the medical image to be denoised specifically includes: The outputs of the target denoising branches are weighted and summed based on the weights of the target denoising branches to obtain a denoised medical image corresponding to the medical image to be denoised.

4. The image denoising method based on generative artificial intelligence technology according to claim 3, characterized in that: The weight of the target denoising branch is obtained based on the final loss and performance index of each denoising branch after the multi-branch denoising unit is trained.

5. The image denoising method based on generative artificial intelligence technology according to claim 4, characterized in that: The training process of the multi-branch denoising unit includes: Training the multi-branch denoising unit based on the medical image samples to be denoised and the predetermined noise classification labels; At the end of each training cycle, the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit are calculated, and whether the training termination condition is reached based on the independent loss of each denoising branch and the combined loss of the multi-branch denoising unit; If yes, stop training; otherwise, update the weights of each denoising branch and continue with subsequent training.

6. The image denoising method based on generative artificial intelligence technology according to claim 5, characterized in that: The updating of the weights of each denoising branch specifically includes: Determine the independent loss of each current denoising branch; Based on the independent loss of each current denoising branch, the weight of each denoising branch is updated using the inverse proportional weight method.

7. An image denoising device based on generative artificial intelligence technology, characterized in that: include: A noise classification module, used for inputting the medical image to be denoised into a noise type recognition model to obtain a noise classification result corresponding to the medical image to be denoised; An image denoising module, used for inputting the medical image to be denoised and the noise classification result corresponding to the medical image to be denoised into an image denoising model to obtain a denoised medical image corresponding to the medical image to be denoised; the image denoising model is a conditional generative adversarial network; The noise type recognition model is obtained after training based on the medical image samples to be denoised and the predetermined noise classification labels, and the medical image samples to be denoised are obtained by adding noise to clear medical image samples of different types; The image denoising model is obtained after training based on different types of clear medical image samples, the medical image samples to be denoised and predetermined noise classification labels; The image denoising model includes a generator and a discriminator, and the generator includes: Shared feature extraction unit, multi-branch denoising unit and output unit; The shared feature extraction unit is used to receive the medical image to be denoised, extract the underlying features of the medical image to be denoised and generate a feature map corresponding to the medical image to be denoised; The multi-branch denoising unit is used to enable multiple target denoising branches to perform denoising processing on the feature map corresponding to the medical image to be denoised in parallel based on the noise classification result corresponding to the medical image to be denoised; The output unit is used to integrate the outputs of the target denoising branches to generate a denoised medical image corresponding to the medical image to be denoised.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the image denoising method based on generative artificial intelligence technology as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the image denoising method based on generative artificial intelligence technology as described in any one of claims 1 to 6 is implemented.

Images