A self-supervised staining standardization method based on denoising diffusion probability model

Abstract

The application provides a self-supervised staining standardization method based on a denoising diffusion probability model, which is mainly used for solving the problems of low robustness and generalization in the analysis of histopathological images by a computer-aided system and information loss and staining errors in current staining standardization algorithms. Compared with the prior art, the application realizes the staining standardization of various staining style histopathological images by the denoising diffusion probability model and a self-supervised training method, retains the cell morphological structure, improves the accuracy of disease diagnosis in downstream tasks such as classification and segmentation of the computer-aided system, and is beneficial to clinical practice and medical research.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tissue pathology image staining standardization, and particularly relates to a self-supervised staining standardization method based on a denoising diffusion probability model. BACKGROUND

[0002] With the development of computer-aided detection / diagnosis, tissue pathology images become more and more important for the diagnosis and prognosis of cancer. However, due to the differences in staining technology, operation skills and scanner specifications, tissue pathology images exhibit different staining styles. These diverse staining styles reduce the robustness of computer-aided detection / diagnosis algorithms.

[0003] In recent years, staining standardization methods are limited to traditional stain vector decomposition, learning-based generative adversarial networks and their variants. Traditional staining standardization methods usually require medical professionals to screen staining templates, which consumes a lot of time and effort. In learning-based methods, supervised methods require paired images, which are difficult to obtain in clinical applications. Unsupervised methods often have staining errors and cell structure changes, and self-supervised methods mostly map grayscale images to color images, which have poor generalization ability and information loss problems. SUMMARY

[0004] In view of the deficiencies of the existing staining standardization technology, the purpose of the present application is to provide a self-supervised staining standardization method based on a denoising diffusion probability model, which systematically converts the staining normalization task into a self-supervised pixel alignment color mapping, avoids the need for paired data, adapts to various staining styles, and does not need to be retrained to apply to different staining scenarios. At the same time, a resampling algorithm is designed, which has the characteristics of fast speed, high accuracy, and does not destroy the cell morphological structure, etc. It provides a robust and effective solution for the staining standardization of digital tissue pathology images, which is conducive to improving the accuracy of disease diagnosis and subsequent analysis, and promoting clinical practice and medical research.

[0005] The present application achieves the above-mentioned purposes through the following technical solutions:

[0006] A self-supervised staining standardization method based on a denoising diffusion probability model, the method comprising the following steps:

[0007] Obtain different staining tissue pathology images, and divide target staining data sets and other staining data sets;

[0008] Preprocess the obtained other staining data sets, and integrate the extracted staining vectors into a staining matrix database;

[0009] Design a self-supervised staining standardization training strategy, and use the staining vectors in the staining matrix database to perform staining enhancement preprocessing on the target staining data sets;

[0010] A denoising diffusion probability model-based staining standardization model is constructed, and the denoising diffusion probability model is self-supervised trained by using the preprocessed data set;

[0011] It is judged whether the training reaches the expected effect, if yes, the re-normalization sampling strategy is used on the trained model to sample the histopathological image to obtain the staining standardization result, otherwise, the self-supervised staining standardization training strategy is adjusted, and the model is retrained by using the adjusted strategy until the satisfactory standardization effect is reached.

[0012] According to the self-supervised staining standardization method based on the denoising diffusion probability model provided by the application, the obtained histopathological image data set is arranged, including:

[0013] All images in the obtained histopathological image data set are traversed, and the histopathological images with different staining styles are divided;

[0014] A target staining image of a staining style is screened and stored as a data set 0, and other staining images are stored as a data set 1.

[0015] According to the self-supervised staining standardization method based on the denoising diffusion probability model provided by the application, the other staining data set obtained is preprocessed, including:

[0016] For each histopathological image in the data set 1, the RGB color space of the image is mapped to the LAB color space, the image is thresholded on the brightness channel, and the tissue region pixel mask is obtained;

[0017] The RGB color space of the image is mapped to the optical density space, the optical density image is multiplied by the tissue region pixel mask, and the area with almost no staining is removed to highlight the staining range of interest;

[0018] The truncated image is flattened into a vector, and then two perpendicular staining matrices are obtained by using singular value decomposition, and the staining matrix of each image in the data set 1 constitutes a staining matrix database.

[0019] According to the self-supervised staining standardization method based on the denoising diffusion probability model provided by the application, the self-supervised staining standardization training strategy is designed, including:

[0020] A staining matrix is randomly selected from the staining matrix database;

[0021] For the randomly selected staining matrix, a disturbance is added;

[0022] For the perturbated coloring matrix, the K-nearest neighbor algorithm is used to search for samples within Euclidean distance 0.1 in the coloring matrix database to ensure that the number of neighbors found is greater than 5; otherwise, the coloring matrix is ​​returned to the coloring matrix database to select a new coloring matrix for enhancement.

[0023] For the input image, calculate its coloring matrix;

[0024] The coloring matrix of the input image is corrected to a perturbed coloring matrix to obtain a color-enhanced image;

[0025] Adding a Gaussian blur to the stained and enhanced image yields the final stained and enhanced preprocessed image.

[0026] According to the self-supervised coloring normalization method based on a denoising diffusion probability model provided by the present invention, for a randomly selected coloring matrix, a perturbation is added, expressed as the following formula:

[0027] M′=M×a+b

[0028] Where M is a randomly selected coloring matrix, M′ is the coloring matrix after perturbation, a is a random number in the range of 0.3 to 1.7, and b is a random number in the range of -0.7 to 0.7.

[0029] According to the self-supervised staining normalization method based on a denoising diffusion probability model provided by the present invention, the construction of the denoising diffusion probability model includes:

[0030] A forward noise-adding process is constructed by gradually adding noise to the image through a parameterized Markov chain until its distribution approaches a standard Gaussian distribution.

[0031] Construct a UNet network for predicting noise, which receives a noisy image x. t The inputs are the image to be standardized (x) and the current time (t), and the output is the predicted noise in the current state.

[0032] Constructing a backward denoising process, using predicted random noise and noisy image x t The noisy image x at the next time step is obtained through reverse reasoning. t-1 This process continues until all noise is removed, resulting in a stained and normalized image.

[0033] According to the self-supervised colorization normalization method based on a denoising diffusion probability model provided by the present invention, noise is gradually added to the image through a parameterized Markov chain until its distribution approximates a standard Gaussian distribution, as expressed by the following formula:

[0034]

[0035] wherein x0 represents a clean image without added noise, x t is an image with added noise at time t, ∈ is random noise sampled from a normal distribution, is a hyperparameter of the denoising diffusion probabilistic model at time t.

[0036] According to the self-supervised staining standardization method based on the denoising diffusion probabilistic model provided by the application, whether the training meets the expected effect is judged, comprising:

[0037] When comparing the performance of the staining standard image generated by the training model and the real required staining standardization image in terms of structural similarity SSIM, peak signal-to-noise ratio PSNR and learning perceptual image block similarity LPIPS, it is expressed as the following formula:

[0038]

[0039]

[0040]

[0041]

[0042] wherein μ is the mean, σ is the standard deviation, σ xy is the covariance of the image x and the image y, MSE is the mean square error of the two images, MaxValue is the maximum value that the image pixel can take, and is the i-th layer feature extracted by the pre-trained neural network, w l is the corresponding weight vector.

[0043] The staining standardization result is applied to related downstream tasks such as classification and segmentation, and the corresponding indicators of the computer-aided detection / diagnosis system on the staining standardization image are better than those of the image without staining standardization.

[0044] According to the self-supervised staining standardization method based on the denoising diffusion probabilistic model provided by the application, the histopathological image is sampled by using a re-normalization sampling strategy, comprising:

[0045] A certain degree of noise is added to the image to be standardized, and the image is used as the initial image in the backward denoising process;

[0046] By a non-Markov chain method, some steps in the backward denoising process are skipped, so that the staining standardization image of the denoising diffusion probabilistic model is quickly and accurately obtained.

[0047] According to the self-supervised staining standardization method based on the denoising diffusion probability model provided by the application, through a non-Markov chain method, part of the steps in the jump backward denoising process are skipped, and are expressed as the following formula:

[0048]

[0049] Wherein, x prev prev represents a plurality of steps apart from t.

[0050] Therefore, compared with the prior art, the application has the following beneficial effects:

[0051] 1. Automation: Traditional histopathological image staining standardization usually requires professionals to manually select a staining template according to the staining effect, while the application fully utilizes the denoising diffusion probability model to realize the staining standardization of different histopathological images, thereby significantly reducing the burden on human resources and accelerating the entire process of histopathological image analysis by introducing automated processing.

[0052] 2. Accuracy: Through sufficient training and learning by the self-supervised staining standardization training strategy, the application has higher level of ability to standardize histopathological image staining and preserve morphological structure. Compared with the existing method, the application more accurately identifies key information in the histopathological image, thereby providing more reliable staining standardization.

[0053] 3. Efficiency: The application uses an efficient resampling algorithm to solve the slow speed of the denoising diffusion probability model, and the speed is improved by 120 times compared with the initial sampling process of the denoising diffusion probability model, reaching a speed comparable to traditional methods. This real-time analysis process provides strong support for rapid diagnostic decision-making.

[0054] 4. Assistant decision: The application standardizes the staining of different histopathological images of different staining styles, solves the problem of poor generalization ability of the computer-aided detection / diagnosis system for different staining styles of histopathological images, and by combining the analysis results of the computer-aided detection / diagnosis system with the experience and judgment of doctors, the accuracy of disease diagnosis and prognosis analysis and the effect of treatment can be improved, which helps to make more accurate diagnosis and treatment decisions.

[0055] 5. Reduce dependence on paired data: Self-supervised learning can train using unpaired data, thereby reducing dependence on a large number of paired, different staining style histopathological images. Through self-supervised learning, the model can learn the internal rules and characteristics of the data, thereby showing better generalization ability when facing new, unseen data.

[0056] 6. Strong generation capability: the denoising diffusion probability model (DDM) performs well in the field of image generation, and can generate high-quality and realistic images. This generation capability provides a solid foundation for staining standardization, enabling the model to better simulate and restore the target staining style. DDM trains and learns data by gradually adding and removing noise, enabling the model to better adapt to different staining styles of pathological images and improve the accuracy and robustness of staining standardization.

[0057] 7. The present application designs a self-supervised staining standardization training strategy, and dynamically adjusts according to the training effect, so that the model can adapt to different data distribution and staining style, and achieve better standardization effect. After training the model, a re-normalization sampling strategy is used to sample the histopathological images, which can obtain high-precision staining standardization results. This sampling strategy can ensure that the standardized images maintain the original tissue structure while having consistent staining style.

[0058] The present application will be further described in detail below in combination with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0059] Fig. 1 is a flowchart of an embodiment of the self-supervised staining standardization method based on the denoising diffusion probability model of the present application.

[0060] Fig. 2 is a flowchart of the self-supervised staining standardization training strategy of the algorithm model in an embodiment of the self-supervised staining standardization method based on the denoising diffusion probability model of the present application.

[0061] Fig. 3 is a general framework diagram of the algorithm model in an embodiment of the self-supervised staining standardization method based on the denoising diffusion probability model of the present application.

[0062] Fig. 4 is an effect diagram of data set 1 in an embodiment of the self-supervised staining standardization method based on the denoising diffusion probability model of the present application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0064] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. As will be apparent to those of ordinary skill in the art, embodiments described herein can be combinable with other embodiments.

[0065] Referring to Figs. 1 to 4 The application provides a self-supervised staining standardization method based on a denoising diffusion probability model, which comprises the following steps:

[0066] Step S1, different staining histopathological images are obtained, and a target staining data set and other staining data sets are divided;

[0067] Step S2, the obtained other staining data sets are preprocessed, and the extracted staining vectors are integrated into a staining matrix database;

[0068] Step S3, a self-supervised staining standardization training strategy is designed, and the staining vectors in the staining matrix database are used to perform staining enhancement preprocessing on the target staining data set;

[0069] Step S4, a staining standardization model based on a denoising diffusion probability model is constructed, and the denoising diffusion probability model is self-supervised trained by using the preprocessed data set;

[0070] Step S5, it is judged whether the training reaches the expected effect, if yes, step S6 is executed.

[0071] Step S6, a re-normalization sampling strategy is used on the trained model to sample the histopathological images, and a staining standardization result is obtained; otherwise, the self-supervised staining standardization training strategy is adjusted, and the model is retrained using the adjusted strategy until a satisfactory standardization effect is achieved.

[0072] In the above step S1, the obtained histopathological image data set is arranged, including:

[0073] Step S11, all images in the obtained histopathological image data set are traversed, and histopathological images with different staining styles due to dyes, staining procedures, digital scanners, etc. are divided;

[0074] Step S12, a target staining image of a certain staining style is selected and stored as data set 0, and other staining images are stored as data set 1.

[0075] In the above step S2, the obtained other staining data sets are preprocessed, including:

[0076] Step S21, for each histopathological image in dataset 1, map the RGB color space of the image to LAB color space, threshold the image on the lightness channel (e.g. 0.8) to obtain a tissue region pixel mask;

[0077] Step S22, map the RGB color space of the image to optical density space, multiply the optical density image by the tissue region pixel mask to remove areas with little staining to highlight the range of staining of interest;

[0078] The optical density space is the so-called OD value, which conforms to the Lambert-Beer law, and is directly related to the amount of staining substance. The more the amount of measured substance, the higher the OD value, the less the transmitted light, and the darker the photo. The optical density space can better reflect the staining distribution in the histopathological image than the RGB color space. The mathematical expression of the Lambert-Beer law and the conversion of the RGB color space to the optical density space is:

[0079]

[0080] OD = -log 10 (I) (2)

[0081] where A is the absorbance, T is the transmittance ratio (transmittance, transmitted light intensity ratio to incident light intensity), K is the molar absorption coefficient, c is the concentration of the light-absorbing substance, L is the thickness of the absorbing layer, and I is the RGB vector of the image.

[0082] Step S23, flatten the truncated image into a vector, then use singular value decomposition to obtain two perpendicular staining matrices, and the staining matrices of each image in dataset 1 form a staining matrix database.

[0083] In the above step S3, a self-supervised staining standardization training strategy is designed, including:

[0084] Randomly select a staining matrix from the staining matrix database;

[0085] For the randomly selected staining matrix, add perturbation, denoted as the following formula:

[0086] M' = M x a + b (3)

[0087] where M is the randomly selected staining matrix, M' is the staining matrix after adding perturbation, a is a random number ranging from 0.3 to 1.7, and b is a random number ranging from -0.7 to 0.7.

[0088] For the perturbed staining matrix, search for samples within a Euclidean distance of 0.1 in the staining matrix database using the K-Nearest Neighbor algorithm in the staining matrix database, ensure that the number of neighbors searched is greater than 5, otherwise return to the staining matrix database to select a new staining matrix for enhancement;

[0089] For the input image, calculate its staining matrix;

[0090] Correct the staining matrix of the input image to the perturbed staining matrix to obtain the staining enhanced image, which can be represented by the following formula:

[0091]

[0092] Where 99 represents the 99th percentile, M x is the staining matrix of the input image, M' is the staining matrix after random selection and perturbation, M' x is the staining matrix after correction of the input image.

[0093] Add Gaussian blur to the staining enhanced image to obtain the final staining enhanced preprocessed image.

[0094] In the above step S4, the construction of the denoising diffusion probability model includes:

[0095] Construct a forward noise adding process, the purpose of adding noise is to enable the denoising diffusion probability model to generate target data samples from noise, gradually add noise to the image through the parameterized Markov chain until its distribution approaches the standard Gaussian distribution, represented by the following formula:

[0096]

[0097] Where x0 represents a clean image without noise, x t is the image after noise at time t, ∈ is random noise sampled from a normal distribution, is the hyperparameter of the denoising diffusion probability model at time t.

[0098] Construct a UNet network for predicting noise, which receives a noisy image x t , a to-be-standardized image x and an input at the current time t, and outputs the predicted noise

[0099] Construct a backward denoising process, by predicting random noise and noisy image x t , through reverse inference, obtain the noisy image x t-1 at the next time, until the noise is completely removed to obtain the staining standardized image, whose formula is:

[0100]

[0101] where z is random noise sampled from normal distribution, σ t , α t and are hyperparameters of the denoising diffusion probability model at time t, and the next image x t-1 is obtained by gradually updating the image x

[0102] In the above step S5, it is judged whether the training meets the expected effect, including:

[0103] When comparing the performance of the staining standard image generated by the training model with the real required staining standardized image in terms of structural similarity SSIM, peak signal-to-noise ratio PSNR, and learning perceptual image block similarity LPIPS, it should be better than the current existing method, and good results are achieved in visual subjective evaluation, which is expressed in the following formula:

[0104]

[0105]

[0106]

[0107]

[0108] where μ is the mean, σ is the standard deviation, σ xy is the covariance of the image x and the image y, MSE is the mean square error of the two images, MaxValue is the maximum value that the image pixel can take, and are the l-th layer features extracted by the pre-trained neural network, w l is the corresponding weight vector.

[0109] The staining standardization result is applied to related downstream tasks such as classification and segmentation, and the corresponding indicators of the computer-aided detection / diagnosis system after the staining standardization image, such as the accuracy Accuracy used to represent the percentage of the number of samples predicted correctly to the total number of samples and the Dice Similarity Coefficient (DSC) used to evaluate the similarity between two samples, are better than those of the image without staining standardization, which is expressed in the following formula:

[0110]

[0111]

[0112] Wherein, TP is the number of actual positive samples and predicted positive samples; FP is the number of actual negative samples but predicted positive samples; TN is the number of actual negative samples and predicted negative samples; FN is the number of actual positive samples but predicted negative samples, and the sample is an image in the classification task, and the sample is a pixel of the image in the segmentation task.

[0113] Further, in the above step S6, the tissue pathology image is sampled for staining standardization by using a re-normalization sampling strategy, comprising:

[0114] A certain degree of noise is added to the image to be standardized, and the initial image in the backward denoising process is used as the initial image, and the formula is:

[0115]

[0116] Wherein, x N The initial image in the backward denoising process is represented as the initial image, and N represents the initial sampling time step selected by experience value, that is, a certain degree of noise.

[0117] By the non-Markov chain method, some steps in the backward denoising process are skipped, so that the staining standardization image of the denoising diffusion probability model is quickly and accurately obtained, and is represented by the following formula:

[0118]

[0119] Wherein, x prev Prev represents that t can be spaced apart by multiple steps.

[0120] The sampling strategy and the re-planning sampling strategy of the original denoising diffusion probability model are tested by using the respective best parameters, and after testing, the relevant data comparison is shown in the following table (1):

[0121]

[0122] Table (1)

[0123] From the above table (1), it can be seen that after the re-planning sampling strategy of the present application, the SSIM, PSNR and LPIPS are all improved, and the inference speed is improved by 120 times. It is proved that the present application can realize better staining standardization performance and effect with smaller time complexity.

[0124] The staining standardization and the staining standardization not subjected to the staining standardization are tested by using the respective best parameters trained in the breast cancer classification and colorectal gland segmentation tasks, and after testing, the relevant data comparison is shown in the following tables (2) and (3):

[0125]

[0126] Table (2)

[0127]

[0128] Table (3)

[0129] From the above tables (2) (3), it can be seen that after the self-supervised staining standardization method based on the denoising diffusion probability model of the present application, the classification of breast cancer and the segmentation of colorectal glands are improved, which shows that the present application achieves good staining standardization effect, improves the accuracy of disease diagnosis in downstream tasks such as classification and segmentation of computer-aided systems, and is beneficial to clinical practice and promotes medical research.

[0130] In summary, compared with the prior art, the self-supervised staining standardization method based on the denoising diffusion probability model of the present embodiment includes the following steps: first, collect histopathological image data of different staining styles, and divide the images into target staining style part and other staining style part. Then, the obtained other style image dataset is preprocessed, including converting the optical density space, intercepting the staining area of interest, extracting the staining matrix features, and finally forming a staining matrix database for training. In the preprocessing stage of model training, a staining matrix is randomly selected from the staining matrix database, and a disturbance is added. Whether the staining matrix conforms to the actual situation is judged by the K nearest neighbor algorithm. The staining matrix of the input image is corrected to the disturbed staining matrix, and then Gaussian blur is added to obtain the final preprocessing result. In the model construction stage, a denoising diffusion probability algorithm model is used. The overall framework of the model includes a forward noise adding process, a UNet network for predicting noise, and a backward denoising process. After training, the performance of the model is evaluated, including structural similarity SSIM, peak signal-to-noise ratio PSNR, learning perceptual image block similarity LPIPS, and inference time, to determine whether the training achieves the expected effect. Further, the staining standardization result can be applied to related downstream tasks such as classification and segmentation. The computer-aided detection / diagnosis system processes the staining standardized image accordingly, and determines whether the computer system recognition ability has improved. If the effects meet the expectations, the self-supervised staining model based on the denoising diffusion probability model is deployed on the histopathological image processing platform. On the platform, different staining style histopathological images are uploaded, and then the trained model is used to perform staining standardization on the images, and finally the style-unified staining standardized images are obtained. The overall process aims to improve the accuracy of disease detection and diagnosis in downstream tasks such as classification and segmentation of computer-aided systems, promote clinical practice and promote medical research.

[0131] Any technical features in the above embodiments can be combined, and for the sake of brevity, not every possible combination is described above. However, any combination of the technical features described above is considered to be within the scope of the disclosure.

[0132] The above embodiments are merely preferred embodiments of the present application, and cannot be used to limit the protection scope of the present application. Any non-essential changes and replacements made by those skilled in the art based on the present application shall fall within the protection scope of the present application.

Claims

1. A self-supervised coloring normalization method based on a denoising diffusion probability model, characterized in that, The method includes the following steps: Acquire pathological images of different stained tissues and divide the target staining dataset into target staining datasets and other staining datasets; The other acquired staining datasets were preprocessed, and the extracted staining vectors were integrated into a staining matrix database. This included: mapping the RGB color space of each histopathological image in dataset 1 to the LAB color space, truncating the image on the luminance channel to obtain the pixel mask of the tissue region; mapping the RGB color space of the image to the optical density space, multiplying the optical density image by the pixel mask of the tissue region to remove areas with almost no staining, in order to highlight the staining range of interest; flattening the truncated image into a vector, and then using singular value decomposition to obtain two perpendicular staining matrices. The staining matrices of each image in dataset 1 constitute the staining matrix database. A self-supervised coloring standardization training strategy is designed to perform coloring enhancement preprocessing on the target coloring dataset using coloring vectors from a coloring matrix database. This includes: randomly selecting a coloring matrix from the database; adding perturbation to the randomly selected coloring matrix; for the perturbed coloring matrix, using the K-nearest neighbor algorithm to search for samples within a Euclidean distance of 0.1 in the coloring matrix database, ensuring that the number of found neighbors is greater than 5; otherwise, returning to the coloring matrix database to reselect a coloring matrix for enhancement; calculating the coloring matrix of the input image; correcting the coloring matrix of the input image to the perturbed coloring matrix to obtain the color-enhanced image; and adding Gaussian blur to the color-enhanced image to obtain the final color-enhanced preprocessed image. A color normalization model based on a denoising diffusion probability model is constructed, and the denoising diffusion probability model is self-supervised trained using a preprocessed dataset. This includes: constructing a forward noise addition process, which progressively adds noise to the image through a parameterized Markov chain until its distribution approximates a standard Gaussian distribution; and constructing a UNet network for predicting noise, which receives noisy images. Images to be standardized Given the input at the current time t, output the predicted noise in the current state. Construct a backward denoising process, using predicted random noise. and noisy images The image with noise at the next time step is obtained through reverse reasoning. Continue until noise is completely removed to obtain a color-normalized image; Determine whether the training has achieved the expected results. If so, use a re-normalization sampling strategy to sample the histopathological images on the trained model to obtain the staining standardization results. Otherwise, readjust the self-supervised staining standardization training strategy and retrain the model using the adjusted strategy until a satisfactory standardization effect is achieved.

2. The method according to claim 1, characterized in that, The acquired histopathological image dataset was organized, including: Iterate through all images in the acquired histopathological image dataset and classify the histopathological images that produce different staining styles; Select a target stained image with one stained style and store it as dataset 0; store other stained images as dataset 1.

3. The method according to claim 1, characterized in that, For a randomly selected coloring matrix, adding a perturbation is expressed as the following formula: in, It is a randomly selected coloring matrix. It is the coloring matrix after adding perturbation. A random number in the range of 0.3 to 1.

7. A random number in the range of -0.7 to 0.

7.

4. The method according to claim 1, characterized in that, The image is passed through a parameterized Markov chain to progressively add noise until its distribution approximates a standard Gaussian distribution, as expressed by the following formula: in, This represents a clean image without added noise. yes Images that have been subjected to noise at all times. It is random noise sampled from a normal distribution. yes Hyperparameters of the time-denoising diffusion probability model.

5. The method according to claim 1, characterized in that, The determination of whether the training meets the expected results includes: When comparing the performance of the stained standard images generated by the training model with the stained normalized images required by reality in terms of structural similarity (SSIM), peak signal-to-noise ratio (PSNR), and learned perceptual image patch similarity (LPIPS), it can be expressed by the following formula: in, It is the mean. It is the standard deviation. It is an image and images The covariance, MSE is the mean square error of two images. It is the maximum value that the image pixels can take. and It is extracted from the pre-trained neural network. Layer features, It is the corresponding weight vector; When the staining normalization results are applied to relevant downstream tasks, the corresponding indicators of computer-aided detection / diagnosis systems for staining normalized images are better than those for unstaining images.

6. The method according to claim 1, characterized in that, The method of sampling histopathological images using a renormalization sampling strategy includes: Add a certain amount of noise to the image to be standardized and use it as the initial image in the backward denoising process; By using a non-Markov chain method, some steps in the backward denoising process are skipped, thus quickly and accurately obtaining the stained and normalized image of the denoised diffusion probability model.

7. The method according to claim 6, characterized in that: Using the non-Markov chain method, some steps in the skip backward denoising process can be expressed as follows: in, of Indicates and Multiple step sizes are allowed.

Images