An image data augmentation method for cell nucleus segmentation

Abstract

The application relates to an image data augmentation method for cell nucleus segmentation, which comprises the following steps: obtaining a randomly generated noise image, inputting the noise image into a preset denoising model for denoising processing to obtain a target image pair, wherein the target image pair comprises a target image and a corresponding labeled target image, and the preset denoising model is obtained by training based on sample cell images and sample segmentation images. In the above method, since the preset denoising model is obtained by training based on the sample image pair composed of the sample cell images and the sample segmentation images, the target image pair generated by the preset denoising model has certain difference from the sample image pair, and the diversity of the sample data set can be improved. In addition, by generating a random noise image and using the preset denoising model to perform denoising processing on the noise image, more sample pairs can be easily and accurately generated, and the accuracy of the image data augmentation is improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an image data amplification method for cell nucleus segmentation. Background Technology

[0002] Cell nucleus segmentation is a fundamental task in pathological image analysis. Accurate segmentation of cell nuclei helps in the analysis of histopathological images to aid clinical diagnosis and treatment. Currently, cell nucleus segmentation models have achieved good results. However, training cell nucleus segmentation models typically requires a large number of labeled images to form paired pathological images and cell nucleus instance segmentation map samples, which is difficult to collect in practice.

[0003] To address the problem of insufficient data, researchers have proposed using data augmentation methods. Traditional data augmentation methods mainly involve cropping, rotating, and color perturbating existing sample images to obtain new sample images, thereby expanding the training set.

[0004] However, the above-mentioned image data augmentation method suffers from low accuracy. Summary of the Invention

[0005] Therefore, it is necessary to provide an image data amplification method for cell nucleus segmentation that can improve the accuracy of image data amplification and address the aforementioned technical problems.

[0006] In a first aspect, this application provides an image data amplification method for cell nucleus segmentation, the method comprising:

[0007] Obtain a randomly generated noisy image;

[0008] The noisy image is input into a preset denoising model for denoising processing to obtain a target image pair; the target image pair includes the target image and the corresponding labeled target image; the preset denoising model is trained based on the sample cell image and the sample segmentation image.

[0009] In one embodiment, the preset denoising model includes a first denoising model and a second denoising model. A noisy image is input into the preset denoising model for denoising processing to obtain a target image pair, including:

[0010] The noisy image is input into the first denoising model for denoising processing to obtain an intermediate image; the first denoising model is trained based on the sample segmented image.

[0011] The intermediate image and the noisy image are input into the second denoising model for denoising processing to obtain the target image; the second denoising model is trained based on the sample cell image and the sample segmentation image.

[0012] Automatic cell structure generation processing is performed on the intermediate image to obtain the labeled target image.

[0013] In one embodiment, the method further includes:

[0014] Acquire sample cell images and sample segmentation images;

[0015] The first initial diffusion model is trained based on the sample segmented image to obtain the first denoising model; the first initial diffusion model includes a first initial noise-adding model and a first initial denoising model.

[0016] The second initial diffusion model is trained based on the sample cell image and the corresponding sample segmentation image to obtain the second denoising model; the second initial diffusion model includes the second initial noise-adding model and the second initial denoising model.

[0017] In one embodiment, a first initial diffusion model is trained based on the sample segmentation image to obtain a first denoising model, including:

[0018] Automatic cell structure generation processing is performed on the sample segmentation image to obtain the first sample structure image;

[0019] The first sample structure image is input into the first initial noise model for noise addition processing to obtain the first noise sample image;

[0020] The first noisy sample image is input into the first initial denoising model for denoising processing to obtain the first denoised sample image;

[0021] The first loss is determined based on the first denoised sample image and the sample structure image, and the first initial denoising model and the first initial denoising model are trained based on the first loss to obtain the first denoising model.

[0022] In one embodiment, a second initial diffusion model is trained based on sample cell images and corresponding sample segmentation images to obtain a second denoising model, including:

[0023] The sample cell image is input into the second initial noise model for noise addition processing to obtain the second noisy sample image;

[0024] Automatic cell structure generation processing is performed on the sample segmentation image to obtain a second sample structure image;

[0025] The second noise sample image and the second sample structure image are input into the second initial denoising model for denoising processing to obtain the second denoised sample image;

[0026] The second loss is determined based on the second denoised sample image and the sample cell image, and the second initial denoising model and the second initial denoising model are trained based on the second loss to obtain the second denoising model.

[0027] Secondly, this application provides a method for segmenting cell nucleus images, the method comprising:

[0028] Acquire images of cell nuclei;

[0029] The cell nucleus image is input into a preset segmentation model for cell nucleus segmentation to obtain the segmentation result; the preset segmentation model is trained based on the amplified sample dataset; the amplified sample dataset includes target image pairs, which are obtained according to the method of any embodiment of the first aspect described above.

[0030] Thirdly, this application also provides an image data amplification device for cell nucleus segmentation, the device comprising:

[0031] The first acquisition module is used to acquire randomly generated noisy images;

[0032] The amplification module is used to input noisy images into a preset denoising model for denoising processing to obtain target image pairs; the target image pairs include target images and corresponding labeled target images; the preset denoising model is trained based on sample cell images and sample segmentation images.

[0033] Fourthly, this application also provides a segmentation apparatus for cell nucleus images, the apparatus comprising:

[0034] The second acquisition module is used to acquire cell nucleus images;

[0035] The segmentation module is used to input cell nucleus images into a preset segmentation model for cell nucleus segmentation and obtain segmentation results; the preset segmentation model is trained based on an amplified sample dataset; the amplified sample dataset includes target image pairs, which are obtained according to the method of any embodiment of the first aspect described above.

[0036] Fifthly, this application also provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the embodiments of the first and second aspects described above.

[0037] In a sixth aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the embodiments of the first and second aspects described above.

[0038] In a seventh aspect, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the method described in any of the embodiments of the first and second aspects described above.

[0039] The aforementioned image data augmentation method for cell nucleus segmentation acquires randomly generated noisy images and then inputs them into a preset denoising model for denoising processing to obtain target image pairs. Each target image pair includes a target image and its corresponding labeled target image. The preset denoising model is trained based on sample cell images and sample segmentation images. Because the preset denoising model is trained on sample image pairs composed of sample cell images and sample segmentation images, the target image pairs generated by the preset denoising model exhibit a certain degree of difference from the sample image pairs, thus enhancing the diversity of the sample dataset. Furthermore, by generating random noisy images and using the preset denoising model to denoise them, more sample pairs can be easily and accurately generated, thereby improving the accuracy of image data augmentation. Attached Figure Description

[0040] Figure 1 This is an internal structural diagram of a computer device in one embodiment;

[0041] Figure 2 This is a flowchart illustrating an image data amplification method for cell nucleus segmentation in one embodiment;

[0042] Figure 3 This is a flowchart illustrating an image data amplification method for cell nucleus segmentation in another embodiment;

[0043] Figure 4 This is a flowchart illustrating an image data amplification method for cell nucleus segmentation in another embodiment;

[0044] Figure 5 This is a flowchart illustrating an image data amplification method for cell nucleus segmentation in another embodiment;

[0045] Figure 6 This is a schematic diagram illustrating the process of generating a first sample structure image in one embodiment;

[0046] Figure 7 This is a flowchart illustrating an image data amplification method for cell nucleus segmentation in another embodiment;

[0047] Figure 8 This is a flowchart illustrating a method for segmenting cell nucleus images in one embodiment;

[0048] Figure 9 This is a flowchart illustrating an image data amplification method for cell nucleus segmentation in another embodiment;

[0049] Figure 10 This is a structural block diagram of an image data amplification device for cell nucleus segmentation in one embodiment;

[0050] Figure 11 This is a structural block diagram of a cell nucleus image segmentation device in another embodiment. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0052] Cell nucleus segmentation is a fundamental task in pathological image analysis. Accurate segmentation of cell nuclei helps analyze histopathological images to aid clinical diagnosis and treatment. Currently, cell nucleus segmentation using cell nucleus segmentation models has achieved good results. However, training cell nucleus segmentation models typically requires a large number of labeled images to form paired pathological images and cell nucleus instance segmentation map samples, which is difficult to collect in practice. To address the problem of insufficient data, researchers have proposed using data augmentation methods. Traditional data augmentation methods mainly involve cropping, rotating, and color perturbating existing sample images to obtain new sample images, thereby expanding the training set. However, the above-mentioned image data augmentation methods suffer from low accuracy. This application provides an image data augmentation method for cell nucleus segmentation, aiming to solve the above-mentioned technical problems. The following embodiments will specifically illustrate the image data augmentation method for cell nucleus segmentation described in this application.

[0053] The image data amplification method for cell nucleus segmentation provided in this application embodiment can be applied to, for example... Figure 1The computer device shown is a terminal, comprising a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for image data amplification and segmentation of cell nuclei images for cell nucleus segmentation. The display unit of the computer device forms a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0054] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0055] In one embodiment, such as Figure 2 As shown, an image data amplification method for cell nucleus segmentation is provided, which can be applied to... Figure 1 Taking a computer device as an example, the explanation includes the following steps:

[0056] S101, Obtain a randomly generated noisy image.

[0057] Among them, a noisy image is an image composed of random pixel values.

[0058] In this embodiment, when a computer device wants to augment a sample dataset, it first needs to acquire a large number of target image pairs. During the acquisition of these target image pairs, the computer device can generate images composed of random pixel values ​​using uniform distribution, Gaussian distribution, or other probability distributions. A random number generator can be used to generate images composed of random pixel values. Specifically, the size of the noise image can be determined first, then the noise type can be selected, and finally, a random noise image can be generated.

[0059] S102, input the noisy image into the preset denoising model for denoising processing to obtain the target image pair. The preset denoising model is trained based on the sample cell image and the sample segmentation image.

[0060] The preset denoising model is trained based on sample cell images and sample segmentation images. Specifically, it can be trained using a U-Net neural network architecture, or other neural network architectures. The target image pair includes the target image and its corresponding labeled target image. The target image can be a pathological image or a cell image. The labeled target image can be a segmented image of cell nuclei instances, or a segmented image of cell nuclei.

[0061] In this embodiment, the computer device can pre-train an initial denoising model based on sample cell images and sample segmentation images to obtain a preset denoising model. Specifically, the sample segmentation image can first be denoised, then denoised to obtain a denoised sample segmentation image. A first initial denoising model can then be trained based on the denoised sample segmentation image to obtain a first denoising model. Then, the sample cell image can be denoised again, and a second initial denoising model can be trained based on the sample segmentation image and the denoised sample cell image to obtain a second denoising model. Finally, a preset denoising model containing the first and second denoising models is obtained. Optionally, the sample cell image can be pre-processed, and then the preset denoising model can be obtained using the above training method based on the pre-processed sample cell image and sample segmentation image. Optionally, the sample segmentation image can be pre-processed, and then the preset denoising model can be obtained using the above training method based on the sample cell image and the pre-processed sample segmentation image. Optionally, both the sample cell image and the sample segmentation image can be preprocessed first, and then a preset denoising model can be obtained based on the preprocessed sample cell image and the preprocessed sample segmentation image using the above training method. Optionally, the sample cell image can be denoised first, and the sample segmentation image can be perturbed and transformed. Then, the denoised sample cell image and the perturbed and transformed sample segmentation image can be input into the initial denoising model for denoising processing to obtain the denoised sample cell image and the denoised sample segmentation image. The initial denoising model can be trained based on the denoised sample cell image, the denoised sample segmentation image, the sample cell image, and the sample segmentation image to obtain the preset denoising model. Optionally, the sample segmentation image can be denoised first, and the sample cell image can be perturbed and transformed. Then, the denoised sample segmentation image and the perturbed and transformed sample cell image can be input into the initial denoising model for denoising to obtain the denoised sample cell image and the denoised sample segmentation image. The initial denoising model can be trained based on the denoised sample cell image, the denoised sample segmentation image, the sample cell image, and the sample segmentation image to obtain the preset denoising model.

[0062] After the computer device obtains the randomly generated noisy image based on the above steps, it can input the noisy image into the preset denoising model for denoising processing to obtain a target image pair including the target image and the corresponding labeled target image.

[0063] The aforementioned image data augmentation method for cell nucleus segmentation acquires randomly generated noisy images and then inputs them into a preset denoising model for denoising processing to obtain target image pairs. Each target image pair includes a target image and its corresponding labeled target image. The preset denoising model is trained based on sample cell images and sample segmentation images. Because the preset denoising model is trained on sample image pairs composed of sample cell images and sample segmentation images, the target image pairs generated by the preset denoising model have certain differences from the sample image pairs, thus improving the diversity of the sample dataset. Furthermore, by generating random noisy images and using the preset denoising model to denoise them, more sample pairs can be easily and accurately generated, thereby improving the accuracy of image data augmentation.

[0064] In one embodiment, a specific implementation method for obtaining target image pairs is also provided, such as... Figure 3 As shown, the preset denoising model includes a first denoising model and a second denoising model. The step S102 above, "inputting the noisy image into the preset denoising model for denoising processing," includes:

[0065] S201, input the noisy image into the first denoising model for denoising processing to obtain the intermediate image.

[0066] The first noise reduction model is trained based on sample segmented images. Specifically, it can be trained using a neural network based on the U-Net architecture, or it can be trained using a neural network based on other architectures. The intermediate image can be a cell instance segmentation map or a cell structure map corresponding to the cell segmentation map.

[0067] In this embodiment, the computer device can pre-train a first initial denoising model based on a sample segmentation image to obtain a trained first denoising model. The sample segmentation image can be first denoised, then denoised to obtain a denoised sample segmentation image. The first initial denoising model can then be trained based on the denoised sample segmentation image to obtain the first denoising model. Optionally, the sample segmentation image can be pre-processed, and then a preset denoising model can be obtained based on the pre-processed sample segmentation image using the above training method. Optionally, the sample segmentation image can be perturbed and transformed, and then the perturbed and transformed sample segmentation image can be input into the initial denoising model for denoising to obtain a denoised sample segmentation image. The initial denoising model can then be trained based on the denoised sample segmentation image and the sample cell image to obtain a preset denoising model. Then, after obtaining a noise map based on the above steps, the computer device can input the noise map into the first denoising model for denoising. The first denoising model learns sample segmentation features, attempts to remove noise and restore image details, and finally obtains an intermediate image.

[0068] S202, input the intermediate image and the noisy image into the second denoising model for denoising processing to obtain the target image.

[0069] The second noise reduction model is trained based on sample cell images and sample segmentation images. Specifically, it can be trained based on a neural network with a U-Net architecture or on a neural network with other architectures.

[0070] In this embodiment, the computer device can pre-train a second initial denoising model based on sample cell images and sample segmentation images to obtain a trained second denoising model. The sample cell images can be first denoised, and then the second initial denoising model can be trained based on the sample segmentation images and the denoised sample cell images to obtain the second denoising model. Optionally, the sample cell images can be pre-processed, and then a preset denoising model can be obtained using the above training method based on the pre-processed sample cell images and sample segmentation images. Optionally, the sample cell images can be perturbed and transformed first, and then the denoised sample cell images and sample segmentation images can be input into the initial denoising model for denoising processing to obtain denoised sample cell images. The initial denoising model can then be trained based on the denoised sample cell images and sample cell images to obtain a preset denoising model. Then, after obtaining an intermediate image based on the above steps, the computer device can input the intermediate image into the second denoising model for denoising processing. The second denoising model learns sample segmentation features, attempts to remove noise and restore image details, and finally obtains the target image.

[0071] S203, automatically generate cell structures from the intermediate image to obtain the labeled target image.

[0072] The labeled target image can be a cell instance segmentation map or a cell segmentation map.

[0073] In this embodiment, after obtaining the intermediate image based on the above steps, the computer device can use the watershed algorithm or other methods to perform automatic cell structure generation processing, converting the intermediate image into a corresponding labeled target image. In the above embodiment, by inputting the intermediate image and the noisy image together into the second denoising model, more information can be introduced to process the noise. By combining it with the noisy image, the second denoising model can more accurately identify and process the noise, thereby generating mutually matching target image pairs.

[0074] In one embodiment, a specific implementation method for training the first denoising model and the second denoising model is also provided, such as... Figure 4 As shown, the method includes:

[0075] S301, acquire sample cell images and sample segmentation images.

[0076] The sample cell images are actual pathological images. The sample segmentation images are annotated versions of the sample cell images, which can be images with the location and boundaries of cell nuclei marked. Specifically, they can be sample instance segmentation images or cell segmentation images.

[0077] In this embodiment, the computer device can acquire real pathological images and use them as sample cell images. These sample cell images can then be annotated manually or by other means to obtain sample segmentation images. Since the sample cell images are annotated manually at the pixel level, only a small number of sample cell images and corresponding sample segmentation images can be obtained using the above method.

[0078] S302, the first initial diffusion model is trained based on the sample segmentation image to obtain the first noise reduction model.

[0079] The first initial diffusion model includes a first initial noise-adding model and a first initial noise-reducing model. The first initial diffusion model can be a neural network based on the U-Net architecture, or it can be a neural network with other architectures. Similarly, the first initial noise-adding model can be a neural network based on the U-Net architecture, or it can be a neural network with other architectures.

[0080] In this embodiment, after obtaining the sample segmentation image based on the above steps, the computer device can input the sample segmentation image into the first initial noise-adding model in the first initial diffusion model for noise addition processing to obtain a noise-adding result. Then, the noise-adding result is input into the first initial denoising model in the first initial diffusion model for denoising processing to obtain a denoising result. Based on the denoising result and the sample segmentation image, a loss function is designed, and then the loss function is minimized. The parameters of the first denoising model are adjusted through backpropagation and optimization algorithms. After the parameter adjustment is completed, the first denoising model is obtained. Optionally, after obtaining the sample segmentation image based on the above steps, the computer device can first preprocess the sample segmentation image, and then train the first initial diffusion model based on the preprocessed sample segmentation image to obtain the first denoising model.

[0081] S303, the second initial diffusion model is trained based on the sample cell image and the corresponding sample segmentation image to obtain the second noise reduction model.

[0082] The second initial diffusion model includes a second initial noise-adding model and a second initial noise-reducing model. The second initial diffusion model can be a neural network based on the U-Net architecture, or it can be a neural network with other architectures. Similarly, the second initial noise-adding model can be a neural network based on the U-Net architecture, or it can be a neural network with other architectures.

[0083] In this embodiment, after the computer device obtains the sample cell image and the corresponding sample segmentation image based on the above steps, it can input the sample cell image into the second initial noise-adding model in the second initial diffusion model for noise addition processing to obtain the noise-adding result of the sample cell image. Then, the noise-adding result of the sample cell image and the corresponding sample segmentation image are input into the second initial denoising model in the second initial diffusion model for denoising processing to obtain the denoising result. Based on the denoising result and the sample cell image, a loss function is designed, and then the loss function is minimized. The parameters of the second denoising model are adjusted through backpropagation and optimization algorithms. After the parameter adjustment is completed, the second denoising model is obtained. Optionally, after the computer device obtains the sample cell image and the corresponding sample segmentation image based on the above steps, it can first preprocess the sample cell image and the corresponding sample segmentation image, and then train the second initial diffusion model based on the preprocessed sample cell image and the corresponding sample segmentation image to obtain the second denoising model.

[0084] In one embodiment, a specific implementation method for training the first denoising model is also provided, such as... Figure 5 As shown, step S302 above, "training the first initial diffusion model based on the sample segmentation image to obtain the first noise reduction model," includes:

[0085] S401, Automatic cell structure generation processing is performed on the sample segmentation image to obtain the first sample structure image.

[0086] The first sample structure image can be a cell nucleus structure diagram corresponding to a sample instance segmentation image or a cell segmentation image.

[0087] In this embodiment of the application, when it is necessary to generate more pairs of target images (such as pathological images) and labeled target images (such as cell nucleus instance segmentation images), since directly generating labeled target images (such as cell nucleus instance segmentation images) is not feasible, an equivalent first sample structure image (such as a cell nucleus structure image, see [reference]) can be generated. Figure 6The rightmost image in the image (e.g., the cell nucleus structure image). The first sample structural image (e.g., the cell nucleus structure image) is a three-channel image consisting of a cell nucleus semantic segmentation map and a distance transform map, with the same size as the corresponding target image (e.g., the pathological image). The cell nucleus semantic segmentation map is a binary image, where each pixel has a value of 1 or 0, indicating whether the pixel belongs to a certain cell nucleus. The distance transform map includes horizontal and vertical distance transforms, which are obtained by calculating the standardized horizontal and vertical distances from each pixel within a cell nucleus to the center of its respective cell nucleus, respectively. After generating the first sample structural image (e.g., the cell nucleus structure image), the labeled target image (e.g., the cell nucleus instance segmentation image) can be constructed from the first sample structural image (e.g., the cell nucleus structure image) using the watershed algorithm.

[0088] After the computer device obtains the sample segmentation image based on the above steps, it can calculate the first sample structure image from the sample segmentation image through preprocessing, segmentation algorithm, simple mathematical algorithm or neural network model.

[0089] S402, the first sample structure image is input into the first initial noise model for noise addition processing to obtain the first noise sample image.

[0090] The first noise sample image is the image after adding noise to the first sample structure image.

[0091] In this embodiment of the application, after the computer device obtains the first sample structure image based on the above steps, it can input the first sample structure image into the first initial noise model for noise addition processing, such as pixel-based noise addition, filter-based noise model, etc., to obtain the first noise sample image.

[0092] S403, the first noise sample image is input into the first initial denoising model for denoising processing to obtain the first denoised sample image.

[0093] The first denoised sample image is the image after denoising the first noisy sample image.

[0094] In this embodiment of the application, after the computer device obtains the first denoised sample image based on the above steps, it can input the first denoised sample image into the first initial denoising model for denoising processing, for example, denoising processing based on a denoising autoencoder, convolutional neural network, etc., to obtain the first denoised sample image.

[0095] S404, determine the first loss based on the first denoised sample image and the sample structure image, and train the first initial denoising model and the first initial denoising model based on the first loss to obtain the first denoising model.

[0096] In this embodiment of the application, after the computer device obtains the first denoised sample image based on the above steps, it can determine the first loss based on the first denoised sample image and the sample structure image. Then, with the goal of the first loss converging or the first loss reaching the loss threshold, the parameters of the first initial denoising model and the parameters of the first initial denoising model are updated. After updating the parameters, the first denoising model is obtained.

[0097] The process of training the first denoising model is as follows: The first initial diffusion model is used to learn the distribution of the first sample structure image. Let y0 be a first sample structure image sampled from the true distribution q(y). The first initial diffusion model learns q(y) using the maximum likelihood method. The first initial diffusion model defines a forward process (see...). Figure 6 (The arrow in the middle from right to left) and a reverse process (see...) Figure 6 (The arrow in the middle runs from left to right). The forward process performs a T-step Gaussian blur on y0:

[0098]

[0099] Where, ∈ t ~N(0, I), Let be the variance coefficients, for the sequence {y0, ..., y...} T This constitutes a Markov chain and follows a Gaussian distribution:

[0100]

[0101] During the reverse process, due to q(y) t-1 |y t q(y) cannot be directly estimated, therefore it is necessary to obtain q(y) through training the first initial denoising method. t-1 |y t The estimated p) θ (y t-1 |y t (θ represents the network parameters). p θ (y t-1 |y t It also follows a Gaussian distribution:

[0102] p θ (y t-1 |y t )=N(y t-1 μ θ (y t ,t),∑ θ (y t ,t))

[0103] The objective function for training the initial denoising is:

[0104]

[0105] Where, α t =1-β t ,

[0106] After the initial denoising training is completed, a new first sample structure image can be obtained from a point randomly sampled from N(0, I) by gradually denoising in the following way:

[0107]

[0108] In the above embodiments, the first denoising model obtained through training can better remove noise, so that by generating random noise images and denoising the noise images using the first denoising model, more first sample structure images can be generated easily and accurately, providing conditions for the subsequent generation of more sample segmentation images, thereby improving the accuracy of image data augmentation.

[0109] In one embodiment, a specific implementation method for training the second noise reduction model is also provided, such as... Figure 7 As shown, step S303 above, "training the second initial diffusion model based on the sample cell image and the corresponding sample segmentation image to obtain the second noise reduction model," includes:

[0110] S501, input the sample cell image into the second initial noise model for noise addition processing to obtain the second noisy sample image.

[0111] The second noise sample image is the image obtained by adding noise to the second sample cell image.

[0112] In this embodiment of the application, after the computer device obtains the sample cell image based on the above steps, it can input the sample cell image into the second initial noise model for noise addition processing, such as pixel-based noise addition, filter-based noise model, etc., to obtain the second noise sample image.

[0113] S502, automatically generate cell structure from the sample segmentation image to obtain the second sample structure image.

[0114] The second sample structure image can be a cell nuclear structure image corresponding to a sample instance segmentation image or a cell segmentation image, and it can be the same as or different from the first sample structure image.

[0115] In this embodiment of the application, after the computer device obtains the sample segmentation image based on the above steps, it can preprocess the image and calculate the second sample structure image from the sample segmentation image using a segmentation algorithm, a simple mathematical algorithm, or a neural network model.

[0116] S503, input the second noise sample image and the second sample structure image into the second initial denoising model for denoising processing to obtain the second denoised sample image.

[0117] The second noise-reduced sample image is the image obtained by denoising the second noise sample image.

[0118] In this embodiment of the application, after the computer device obtains the second sample structure image and the second noise-reduced sample image based on the above steps, it can input the second noise sample image and the second sample structure image into the second initial noise reduction model for noise reduction processing. For example, noise reduction processing is performed based on a denoising autoencoder, a convolutional neural network, etc., to obtain the second noise-reduced sample image.

[0119] S504, determine the second loss based on the second denoised sample image and the sample cell image, and train the second initial denoising model and the second initial denoising model based on the second loss to obtain the second denoising model.

[0120] In this embodiment of the application, after the computer device obtains the second denoised sample image based on the above steps, it can determine the second loss based on the second denoised sample image and the sample cell image. Then, with the goal of the second loss converging or the second loss reaching the loss threshold, the parameters of the second initial denoising model and the parameters of the second initial denoising model are updated. After updating the parameters, the second denoising model is obtained.

[0121] The process of training the second denoising model is as follows: the sampling process is controlled by a classifier-free approach, so that the generated second denoising sample image and the input second sample structure image maintain an accurate correspondence.

[0122] Let x be the second denoised sample image, ∈ θ (x t ,t) and ∈ θ (x t , t, y) represent the first noise reduction model p θ (x) and the second noise reduction model p θ Noise estimation of (x|y) can be achieved by training the same neural network to fit two models p without classifier guidance. θ (x) and p θ (x|y). Specifically, p can be trained on paired data (x0, y0). θ (x|y), while p θ (x) can be set Obtained by training on x0. Noise estimate ∈ ' without classifier guidance. θ (x t ,t,y) is ∈ θ (x t,t,y) and ∈ θ (x t Linear combinations of t):

[0123] ∈′ θ (x t ,t,y)=(w+1)∈ θ (x t ,t,y)-w∈ θ (x t ,t)

[0124] in, w is a coefficient that controls the strength of the classifier-free guidance.

[0125] In the generation of the second sample structure image, the second noise reduction model uses y t For input, output noise estimate ∈ θ (y t ,t). In the generation of the second denoised sample image, the second denoising model uses x t Given y as input, output noise estimate ∈′ θ (x t (t, y). The neural network generating the second denoised sample image needs to be able to align the second denoised sample image with the corresponding second sample structure image. To achieve this, the second denoising model uses U-Net as its basic framework, consisting of an encoder and a decoder. The encoder is composed of residual blocks, attention modules, and downsampling layers stacked together. Each residual block contains two "group normalization-activation function-1×1 convolutions", and each attention module calculates the self-attention of the input feature map. The decoder is composed of conditional residual blocks, attention modules, and upsampling layers stacked together. Each conditional residual block contains two "conditional fusion-activation function-1×1 convolutions". Among them, the conditional fusion module can effectively fuse the semantic information of the cell kernel structure map with the group normalized feature map. Specifically, the cell kernel structure map y first passes through a 3×3 convolutional layer, then obtains weights w and biases b through two independent 3×3 convolutional layers, and finally performs a linear combination with the group normalized feature map to obtain the semantically fused feature map F′:

[0126] F′=w·GroupNorm(F)+b.

[0127] To verify the effectiveness of data augmentation in this embodiment, experiments were conducted on the MoNuSeg and Kumar datasets. The training sets of each dataset were divided into four subsets, containing 10% / 20% / 50% / 100% of labeled images, respectively. Then, a first denoising model and a second denoising model were trained using samples from each subset, generating a certain number of new samples to augment the original dataset. The effectiveness of data augmentation in this embodiment was evaluated by comparing the performance of the segmentation models trained on the pre- and post-amplification datasets. Two cell kernel segmentation models—Hover-Net and PFF-Net—were trained, and the segmentation model metrics were Dice and AJI.

[0128] The study showcases new samples generated by training the model on a 10% subset of MoNuSeg. It can be seen that the generated cell nuclear structure images (first sample structure image or second sample structure image) and pathological images (second denoised sample images) have similar distribution and texture to the real samples, but are also significantly different from the real samples. This demonstrates that the data augmentation method in this embodiment can generate realistic and diverse samples. At the same time, thanks to the special design of the second denoising model and the classifier-free sampling mechanism, the generated pathological images (second denoised sample images) can accurately correspond to the corresponding cell nuclear structure images (first sample structure image or second sample structure image).

[0129] Taking the Hover-Net model training as an example, for the MoNuSeg dataset, compared to the Hover-Net (baseline model) trained on all labeled images, training Hover-Net on the augmented 10% subset resulted in improvements of 0.9% and 1.3% for Dice and AJI, respectively. Augmenting the 20% and 50% subsets also achieved performance levels similar to the baseline model. Augmenting the dataset to 100% labeled further improved the segmentation metrics compared to the baseline model. Similar observations were observed in experiments on the Kumar dataset and with PFF-Net as the segmentation model. Therefore, by using the data augmentation method in this embodiment to augment datasets with only a small number of labeled images, the performance of the trained segmentation model can reach or surpass that of the baseline model trained on a larger number of labeled datasets.

[0130] In the above embodiments, the trained second denoising model can better remove noise, making it easier and more accurate to generate more sample cell images by subsequently generating random noise images and then using the second denoising model to denoise them, thereby improving the accuracy of image data augmentation.

[0131] In one embodiment, such as Figure 8 As shown, a method for segmenting cell nucleus images is also provided, including the following steps:

[0132] S601, acquire cell nucleus images.

[0133] Among them, the cell nucleus image is the image of the cell nucleus to be segmented.

[0134] In this embodiment of the application, the computer device can acquire cell nucleus images from a suitable data source (such as a cell image database, scientific research experimental results, etc.).

[0135] S602, the cell nucleus image is input into a preset segmentation model for cell nucleus segmentation to obtain the segmentation result. The preset segmentation model is trained based on an amplified sample dataset, which includes target image pairs. The target image pairs are obtained according to the image data amplification method for cell nucleus segmentation as described in any of the embodiments above.

[0136] In this embodiment, the computer device can use the image data augmentation method for cell nucleus segmentation described in any of the above embodiments to obtain an augmented sample dataset including target image pairs (target image and corresponding labeled target image), train the initial segmentation model, and obtain a preset segmentation model. Then, based on the cell nucleus image obtained in the above steps, the computer device can input the cell nucleus image into the preset segmentation model to perform cell nucleus segmentation and obtain the segmentation result.

[0137] In the above embodiments, since the preset cell nucleus segmentation model is trained based on the amplified sample dataset generated by the image data amplification method in the above embodiments, the accuracy of cell nucleus segmentation can be improved by using the preset cell nucleus segmentation model for cell nucleus segmentation.

[0138] In summary, based on all the above embodiments, an image data amplification method for cell nucleus segmentation is also provided, such as... Figure 9 As shown, the method includes:

[0139] S701, train the first initial noise-adding model and the first initial noise-reducing model to obtain the first noise-reducing model.

[0140] S7010, acquire sample cell images and sample segmentation images.

[0141] S7011, automatically generate cell structure from the sample segmentation image to obtain the first sample structure image.

[0142] S7012, the first sample structure image is input into the first initial noise model for noise addition processing to obtain the first noise sample image.

[0143] S7013, the first noise sample image is input into the first initial denoising model for denoising processing to obtain the first denoised sample image.

[0144] S7014, determine the first loss based on the first denoised sample image and the sample structure image, and train the first initial denoising model and the first initial denoising model based on the first loss to obtain the first denoising model.

[0145] S702, train the second initial noise-adding model and the second initial noise-reducing model to obtain the second noise-reducing model.

[0146] S7020, the sample cell image is input into the second initial noise model for noise addition processing to obtain the second noisy sample image.

[0147] S7021, Automatic cell structure generation processing is performed on the sample segmentation image to obtain the second sample structure image.

[0148] S7022, the second noise sample image and the second sample structure image are input into the second initial denoising model for denoising processing to obtain the second denoised sample image.

[0149] S7023, determine the second loss based on the second denoised sample image and sample cell image, and train the second initial denoising model and the second initial denoising model based on the second loss to obtain the second denoising model.

[0150] S703 uses the first and second denoising models to generate a large number of target image pairs.

[0151] S7030, acquire a randomly generated noisy image;

[0152] S7031, input the noisy image into the first denoising model for denoising processing to obtain the intermediate image.

[0153] S7032 inputs the intermediate image and the noisy image into the second denoising model for denoising processing to obtain the target image.

[0154] S7033 performs automatic cell structure generation processing on the intermediate image to obtain the labeled target image.

[0155] S7034, a target image pair is formed by a target image and a target annotation image for each target.

[0156] S704, based on an augmented sample dataset consisting of a large number of target image pairs, trains a preset segmentation model.

[0157] S705, acquire cell nucleus images;

[0158] S706: Input the cell nucleus image into the preset segmentation model to perform cell nucleus segmentation and obtain the segmentation result.

[0159] The methods described in each of the above steps have been described in the foregoing embodiments. For details, please refer to the foregoing descriptions. They will not be repeated here.

[0160] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0161] Based on the same inventive concept, this application also provides an image data amplification device for implementing the image data amplification method for cell nucleus segmentation described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more embodiments of the image data amplification device for cell nucleus segmentation provided below can be found in the limitations of the image data amplification method for cell nucleus segmentation described above, and will not be repeated here.

[0162] In one embodiment, such as Figure 10 As shown, an image data amplification device for cell nucleus segmentation is provided, comprising:

[0163] The first acquisition module 10 is used to acquire randomly generated noisy images;

[0164] The amplification module 11 is used to input the noisy image into the preset denoising model for denoising processing to obtain the target image pair; the target image pair includes the target image and the corresponding labeled target image; the preset denoising model is trained based on the sample cell image and the sample segmentation image.

[0165] In one embodiment, the amplification module 11 includes:

[0166] The first noise reduction unit is used to input the noisy image into the first noise reduction model for noise reduction processing to obtain an intermediate image; the first noise reduction model is trained based on the sample segmented image.

[0167] The second noise reduction unit is used to input the intermediate image and the noisy image into the second noise reduction model for noise reduction processing to obtain the target image; the second noise reduction model is trained based on the sample cell image and the sample segmentation image.

[0168] The annotation unit is used to automatically generate cell structures from the intermediate image to obtain the annotated target image.

[0169] In one embodiment, the image data amplification device for cell nucleus segmentation described above includes:

[0170] The third acquisition module is used to acquire sample cell images and sample segmentation images;

[0171] The first training module is used to train the first initial diffusion model based on the sample segmented image to obtain the first denoising model; the first initial diffusion model includes a first initial noise-adding model and a first initial denoising model.

[0172] The second training module is used to train the second initial diffusion model based on the sample cell image and the corresponding sample segmentation image to obtain the second denoising model; the second initial diffusion model includes the second initial noise-adding model and the second initial denoising model.

[0173] In one embodiment, the first training module includes:

[0174] The first processing unit is used to perform automatic cell structure generation processing on the sample segmentation image to obtain the first sample structure image;

[0175] The second processing unit is used to input the first sample structure image into the first initial noise model for noise processing to obtain the first noise sample image;

[0176] The third processing unit is used to input the first noise sample image into the first initial denoising model for denoising processing to obtain the first denoised sample image.

[0177] The first training unit is used to determine the first loss based on the first denoised sample image and the sample structure image, and to train the first initial denoising model and the first initial denoising model based on the first loss to obtain the first denoising model.

[0178] In one embodiment, the second training module described above includes:

[0179] The fourth processing unit is used to input the sample cell image into the second initial noise model for noise addition processing to obtain the second noisy sample image;

[0180] The fifth processing unit is used to automatically generate cell structures from the sample segmentation image to obtain the second sample structure image.

[0181] The sixth processing unit is used to input the second noise sample image and the second sample structure image into the second initial denoising model for denoising processing to obtain the second denoised sample image;

[0182] The second training unit is used to determine the second loss based on the second denoised sample image and the sample cell image, and to train the second initial denoising model and the second initial denoising model based on the second loss to obtain the second denoising model.

[0183] In one embodiment, such as Figure 11 As shown, a segmentation device for cell nucleus images is provided, comprising:

[0184] The second acquisition module 20 is used to acquire cell nucleus images;

[0185] The segmentation module 21 is used to input the cell nucleus image into a preset segmentation model to segment the cell nucleus and obtain the segmentation result; the preset segmentation model is trained based on the amplified sample dataset; the amplified sample dataset includes target image pairs, which are obtained according to any of the methods in the first aspect above.

[0186] The modules in the image data augmentation device for cell nucleus segmentation described above can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0187] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0188] Obtain a randomly generated noisy image;

[0189] The noisy image is input into a preset denoising model for denoising processing to obtain a target image pair; the target image pair includes the target image and the corresponding labeled target image; the preset denoising model is trained based on the sample cell image and the sample segmentation image.

[0190] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0191] The noisy image is input into the first denoising model for denoising processing to obtain an intermediate image; the first denoising model is trained based on the sample segmented image.

[0192] The intermediate image and the noisy image are input into the second denoising model for denoising processing to obtain the target image; the second denoising model is trained based on the sample cell image and the sample segmentation image.

[0193] Automatic cell structure generation processing is performed on the intermediate image to obtain the labeled target image.

[0194] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0195] Acquire sample cell images and sample segmentation images;

[0196] The first initial diffusion model is trained based on the sample segmented image to obtain the first denoising model; the first initial diffusion model includes a first initial noise-adding model and a first initial denoising model.

[0197] The second initial diffusion model is trained based on the sample cell image and the corresponding sample segmentation image to obtain the second denoising model; the second initial diffusion model includes the second initial noise-adding model and the second initial denoising model.

[0198] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0199] Automatic cell structure generation processing is performed on the sample segmentation image to obtain the first sample structure image;

[0200] The first sample structure image is input into the first initial noise model for noise addition processing to obtain the first noise sample image;

[0201] The first noisy sample image is input into the first initial denoising model for denoising processing to obtain the first denoised sample image;

[0202] The first loss is determined based on the first denoised sample image and the sample structure image, and the first initial denoising model and the first initial denoising model are trained based on the first loss to obtain the first denoising model.

[0203] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0204] The sample cell image is input into the second initial noise model for noise addition processing to obtain the second noisy sample image;

[0205] Automatic cell structure generation processing is performed on the sample segmentation image to obtain a second sample structure image;

[0206] The second noise sample image and the second sample structure image are input into the second initial denoising model for denoising processing to obtain the second denoised sample image;

[0207] The second loss is determined based on the second denoised sample image and the sample cell image, and the second initial denoising model and the second initial denoising model are trained based on the second loss to obtain the second denoising model.

[0208] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0209] Acquire images of cell nuclei;

[0210] The cell nucleus image is input into a preset segmentation model for cell nucleus segmentation to obtain the segmentation result; the preset segmentation model is trained based on the amplified sample dataset; the amplified sample dataset includes target image pairs, which are obtained according to any of the methods in the above embodiments of the image data amplification method for cell nucleus segmentation.

[0211] The computer device provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0212] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0213] Obtain a randomly generated noisy image;

[0214] The noisy image is input into a preset denoising model for denoising processing to obtain a target image pair; the target image pair includes the target image and the corresponding labeled target image; the preset denoising model is trained based on the sample cell image and the sample segmentation image.

[0215] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0216] The noisy image is input into the first denoising model for denoising processing to obtain an intermediate image; the first denoising model is trained based on the sample segmented image.

[0217] The intermediate image and the noisy image are input into the second denoising model for denoising processing to obtain the target image; the second denoising model is trained based on the sample cell image and the sample segmentation image.

[0218] Automatic cell structure generation processing is performed on the intermediate image to obtain the labeled target image.

[0219] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0220] Acquire sample cell images and sample segmentation images;

[0221] The first initial diffusion model is trained based on the sample segmented image to obtain the first denoising model; the first initial diffusion model includes a first initial noise-adding model and a first initial denoising model.

[0222] The second initial diffusion model is trained based on the sample cell image and the corresponding sample segmentation image to obtain the second denoising model; the second initial diffusion model includes the second initial noise-adding model and the second initial denoising model.

[0223] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0224] Automatic cell structure generation processing is performed on the sample segmentation image to obtain the first sample structure image;

[0225] The first sample structure image is input into the first initial noise model for noise addition processing to obtain the first noise sample image;

[0226] The first noisy sample image is input into the first initial denoising model for denoising processing to obtain the first denoised sample image;

[0227] The first loss is determined based on the first denoised sample image and the sample structure image, and the first initial denoising model and the first initial denoising model are trained based on the first loss to obtain the first denoising model.

[0228] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0229] The sample cell image is input into the second initial noise model for noise addition processing to obtain the second noisy sample image;

[0230] Automatic cell structure generation processing is performed on the sample segmentation image to obtain a second sample structure image;

[0231] The second noise sample image and the second sample structure image are input into the second initial denoising model for denoising processing to obtain the second denoised sample image;

[0232] The second loss is determined based on the second denoised sample image and the sample cell image, and the second initial denoising model and the second initial denoising model are trained based on the second loss to obtain the second denoising model.

[0233] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0234] Acquire images of cell nuclei;

[0235] The cell nucleus image is input into a preset segmentation model for cell nucleus segmentation to obtain the segmentation result; the preset segmentation model is trained based on the amplified sample dataset; the amplified sample dataset includes target image pairs, which are obtained according to any of the methods in the above embodiments of the image data amplification method for cell nucleus segmentation.

[0236] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0237] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0238] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0239] Obtain a randomly generated noisy image;

[0240] The noisy image is input into a preset denoising model for denoising processing to obtain a target image pair; the target image pair includes the target image and the corresponding labeled target image; the preset denoising model is trained based on the sample cell image and the sample segmentation image.

[0241] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0242] The noisy image is input into the first denoising model for denoising processing to obtain an intermediate image; the first denoising model is trained based on the sample segmented image.

[0243] The intermediate image and the noisy image are input into the second denoising model for denoising processing to obtain the target image; the second denoising model is trained based on the sample cell image and the sample segmentation image.

[0244] Automatic cell structure generation processing is performed on the intermediate image to obtain the labeled target image.

[0245] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0246] Acquire sample cell images and sample segmentation images;

[0247] The first initial diffusion model is trained based on the sample segmented image to obtain the first denoising model; the first initial diffusion model includes a first initial noise-adding model and a first initial denoising model.

[0248] The second initial diffusion model is trained based on the sample cell image and the corresponding sample segmentation image to obtain the second denoising model; the second initial diffusion model includes the second initial noise-adding model and the second initial denoising model.

[0249] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0250] Automatic cell structure generation processing is performed on the sample segmentation image to obtain the first sample structure image;

[0251] The first sample structure image is input into the first initial noise model for noise addition processing to obtain the first noise sample image;

[0252] The first noisy sample image is input into the first initial denoising model for denoising processing to obtain the first denoised sample image;

[0253] The first loss is determined based on the first denoised sample image and the sample structure image, and the first initial denoising model and the first initial denoising model are trained based on the first loss to obtain the first denoising model.

[0254] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0255] The sample cell image is input into the second initial noise model for noise addition processing to obtain the second noisy sample image;

[0256] Automatic cell structure generation processing is performed on the sample segmentation image to obtain a second sample structure image;

[0257] The second noise sample image and the second sample structure image are input into the second initial denoising model for denoising processing to obtain the second denoised sample image;

[0258] The second loss is determined based on the second denoised sample image and the sample cell image, and the second initial denoising model and the second initial denoising model are trained based on the second loss to obtain the second denoising model.

[0259] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0260] Acquire images of cell nuclei;

[0261] The cell nucleus image is input into a preset segmentation model for cell nucleus segmentation to obtain the segmentation result; the preset segmentation model is trained based on the amplified sample dataset; the amplified sample dataset includes target image pairs, which are obtained according to any of the methods in the above embodiments of the image data amplification method for cell nucleus segmentation.

[0262] The computer program product provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0263] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0264] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0265] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for image data amplification for cell nucleus segmentation, characterized in that, The method includes: Obtain a randomly generated noisy image; The noisy image is input into the first denoising model in the preset denoising model for denoising processing to obtain an intermediate image; the first denoising model is obtained by training a first initial diffusion model based on the sample segmentation image, and the first initial diffusion model includes a first initial noise-adding model and a first initial denoising model; The intermediate image and the noisy image are input into the second denoising model in the preset denoising model for denoising processing to obtain the target image; the second denoising model is obtained by training the second initial diffusion model based on the sample cell image and the corresponding sample segmentation image, and the second initial diffusion model includes the second initial noise-adding model and the second initial denoising model; The intermediate image is automatically annotated to generate an labeled target image; the target image and the corresponding labeled target image constitute a target image pair. The method for training a first initial diffusion model to obtain a first denoising model based on the sample segmentation image includes: performing automatic cell structure generation processing on the sample segmentation image to obtain a first sample structure image; inputting the first sample structure image into the first initial noise-adding model for noise-adding processing to obtain a first noisy sample image; inputting the first noisy sample image into the first initial denoising model for denoising processing to obtain a first denoised sample image; determining a first loss based on the first denoised sample image and the sample structure image, and training the first initial noise-adding model and the first initial denoising model based on the first loss to obtain the first denoising model. The method for training a second initial diffusion model to obtain a second denoising model based on the sample cell image and the corresponding sample segmentation image includes: inputting the sample cell image into the second initial noise-adding model for noise-adding processing to obtain a second noisy sample image; performing automatic cell structure generation processing on the sample segmentation image to obtain a second sample structure image; inputting the second noisy sample image and the second sample structure image into the second initial denoising model for denoising processing to obtain a second denoised sample image; determining a second loss based on the second denoised sample image and the sample cell image, and training the second initial noise-adding model and the second initial denoising model based on the second loss to obtain the second denoising model.

2. A method for segmenting cell nucleus images, characterized in that, The method includes: Acquire images of cell nuclei; The cell nucleus image is input into a preset segmentation model for cell nucleus segmentation to obtain a segmentation result; the preset segmentation model is trained based on an amplified sample dataset; the amplified sample dataset includes target image pairs, which are obtained according to the method described in claim 1.

3. An image data amplification device for cell nucleus segmentation, used to implement the image data amplification method for cell nucleus segmentation as described in claim 1, characterized in that, The device includes: The first acquisition module is used to acquire randomly generated noisy images; An amplification module is used to input the noisy image into a preset denoising model for denoising processing to obtain a target image pair; the target image pair includes a target image and a corresponding labeled target image; the preset denoising model is trained based on sample cell images and sample segmentation images.

4. A method and apparatus for segmenting cell nucleus images, characterized in that, The device includes: The second acquisition module is used to acquire cell nucleus images; A segmentation module is used to input the cell nucleus image into a preset segmentation model for cell nucleus segmentation to obtain a segmentation result; the preset segmentation model is trained based on an amplified sample dataset; the amplified sample dataset includes target image pairs, which are obtained according to the method described in claim 1.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 or 2.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 or 2.

Images