Image processing method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the technical field of artificial intelligence, relates to an image processing method and device based on artificial intelligence, computer equipment and a storage medium, and is applied to the field of digital medical treatment. Performing image adjustment processing on the initial pathological image to obtain a pathological image; calling a preset image processing model; wherein the image processing model is a model generated after training a potential diffusion model based on pre-constructed pathological sample data; performing image generation processing on the pathological image based on the image processing model to obtain a reconstructed image; performing optimization processing on the reconstructed image based on a preset fragment type time step tracking strategy to obtain a target pathological image; and performing output processing on the target pathological image. In addition, the target pathological image may be stored in a block chain. According to the application, the definition and the image quality of the generated target pathological image are effectively improved based on the use of the image adjustment processing, so that the efficiency and the accuracy of pathological image analysis are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and the field of digital medicine, and in particular to an image processing method, apparatus, computer equipment and storage medium based on artificial intelligence. Background Art

[0002] In the AI ​​intelligent reading scenario of medical reports, analyzing and understanding the pathological images of patients or users is a crucial step. However, current pathological image analysis technology faces many challenges, the most significant of which is the clarity and quality of pathological images.

[0003] In the existing analysis of pathological images, the image acquisition and processing methods are limited by the technical level and equipment conditions, which often leads to low clarity and poor quality of the obtained images. This situation not only increases the difficulty of image analysis, but also seriously affects the work efficiency of experts. In practical applications, the blur, noise, lack of contrast and other problems of pathological images require experts to spend more time and energy to identify and analyze, which undoubtedly increases their workload, resulting in low efficiency and accuracy of pathological image analysis. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose an image processing method, device, computer equipment and storage medium based on artificial intelligence to solve the technical problem that the existing pathological image acquisition and processing methods will result in low clarity and poor quality of the obtained images due to limitations in technical level and equipment conditions, thereby causing low efficiency and accuracy of pathological image analysis.

[0005] In order to solve the above technical problems, the embodiment of the present application provides an image processing method based on artificial intelligence, which adopts the following technical solutions:

[0006] acquiring an initial pathological image to be processed;

[0007] Performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image;

[0008] Calling a preset image processing model; wherein the image processing model is a model generated by training a preset potential diffusion model based on pre-constructed pathological sample data;

[0009] Performing image generation processing on the pathological image based on the image processing model to obtain a corresponding reconstructed image;

[0010] The reconstructed image is optimized based on a preset fragmentary time step tracking strategy to obtain a corresponding target pathological image;

[0011] The target pathological image is outputted.

[0012] Furthermore, the image processing model includes an encoder and a decoder; the step of performing image generation processing on the pathological image based on the image processing model to obtain a corresponding reconstructed image specifically includes:

[0013] inputting the pathological image into the image processing model;

[0014] The pathological image is converted by the encoder to obtain a corresponding initial latent vector;

[0015] Performing noise addition processing on the initial latent vector to obtain a corresponding target latent vector;

[0016] Based on the decoder, the target latent vector is subjected to a stepwise denoising process to obtain a corresponding output image;

[0017] The output image is used as the reconstructed image.

[0018] Furthermore, the image processing model also includes a fragmented time step tracking module; the step of optimizing the reconstructed image based on the preset fragmented time step tracking strategy to obtain the corresponding target pathological image specifically includes:

[0019] Calling the fragmented time step tracking module;

[0020] dividing the reconstructed image into a plurality of potential patches based on the segmented time step tracking module;

[0021] Performing a separate denoising process on each of the potential plaques to obtain a denoised designated potential plaque;

[0022] Combining all the denoised designated potential patches based on a preset weight calculation strategy to obtain a corresponding combined image;

[0023] The combined image is used as the target pathological image.

[0024] Furthermore, the step of outputting the target pathological image specifically includes:

[0025] Performing image magnification processing on the target pathological image to obtain a corresponding first pathological image;

[0026] Performing image sharpening processing on the first pathological image to obtain a corresponding second pathological image;

[0027] performing super-resolution reconstruction processing on the second pathological image to obtain a corresponding third pathological image;

[0028] The third pathological image is outputted.

[0029] Furthermore, the step of performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image specifically includes:

[0030] Performing a size adjustment process on the initial pathological image to obtain a corresponding first initial pathological image;

[0031] Normalizing the first initial pathological image to obtain a corresponding second initial pathological image;

[0032] Performing format conversion processing on the second initial pathological image to obtain a corresponding third initial pathological image;

[0033] The third initial pathological image is used as the pathological image.

[0034] Furthermore, before the step of calling the preset image processing model, the method further includes:

[0035] obtaining a pre-collected pathological image dataset;

[0036] Performing data segmentation and data description generation processing on the pathological image data set to obtain corresponding pathological sample data;

[0037] calling the potential diffusion model;

[0038] Generate the corresponding target total loss based on the preset reconstruction loss and feature entropy loss;

[0039] Based on the target total loss, the potential diffusion model is trained and evaluated using the pathological sample data to obtain a specified model that meets the preset construction requirements;

[0040] The designated model is used as the image processing model.

[0041] Furthermore, after the step of optimizing the reconstructed image based on the preset fragment-based time step tracking strategy to obtain the corresponding target pathological image, the method further includes:

[0042] Acquire a specified storage strategy corresponding to the target pathological image;

[0043] calling a designated storage medium corresponding to the target pathological image;

[0044] Based on the designated storage strategy, the target pathological image is stored in the designated storage medium.

[0045] In order to solve the above technical problems, the embodiment of the present application further provides an image processing device based on artificial intelligence, which adopts the following technical solution:

[0046] A first acquisition module, used for acquiring an initial pathological image to be processed;

[0047] An adjustment module, used for performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image;

[0048] A first calling module is used to call a preset image processing model; wherein the image processing model is a model generated by training a preset potential diffusion model based on pre-constructed pathological sample data;

[0049] A generation module, used for performing image generation processing on the pathological image based on an image processing model to obtain a corresponding reconstructed image;

[0050] An optimization module, used for optimizing the reconstructed image based on a preset fragmentary time step tracking strategy to obtain a corresponding target pathological image;

[0051] An output module is used to perform output processing on the target pathological image.

[0052] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0053] acquiring an initial pathological image to be processed;

[0054] Performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image;

[0055] Calling a preset image processing model; wherein the image processing model is a model generated by training a preset potential diffusion model based on pre-constructed pathological sample data;

[0056] Performing image generation processing on the pathological image based on the image processing model to obtain a corresponding reconstructed image;

[0057] The reconstructed image is optimized based on a preset fragmentary time step tracking strategy to obtain a corresponding target pathological image;

[0058] The target pathological image is outputted.

[0059] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0060] acquiring an initial pathological image to be processed;

[0061] Performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image;

[0062] Calling a preset image processing model; wherein the image processing model is a model generated by training a preset potential diffusion model based on pre-constructed pathological sample data;

[0063] Performing image generation processing on the pathological image based on the image processing model to obtain a corresponding reconstructed image;

[0064] The reconstructed image is optimized based on a preset fragmentary time step tracking strategy to obtain a corresponding target pathological image;

[0065] The target pathological image is outputted.

[0066] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0067] The present application first obtains the initial pathological image to be processed; then performs image adjustment processing on the initial pathological image to obtain the corresponding pathological image; then calls a preset image processing model; wherein the image processing model is a model generated by training a preset potential diffusion model based on pre-constructed pathological sample data; subsequently, the pathological image is subjected to image generation processing based on the image processing model to obtain a corresponding reconstructed image; further, the reconstructed image is optimized based on a preset fragmented time step tracking strategy to obtain a corresponding target pathological image; finally, the target pathological image is output processed. The present application obtains a pathological image by performing image adjustment processing on the initial pathological image to be processed, and then performs image generation processing on the pathological image based on the image processing model to obtain a reconstructed image, and further optimizes the reconstructed image based on a preset fragmented time step tracking strategy to obtain a target pathological image, thereby effectively improving the clarity and image quality of the generated target pathological image. The subsequent output processing of the target pathological image can effectively improve the efficiency and accuracy of pathological image analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0069] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0070] Figure 2 A flowchart of an embodiment of an artificial intelligence-based image processing method according to the present application;

[0071] Figure 3is a schematic structural diagram of an embodiment of an image processing device based on artificial intelligence according to the present application;

[0072] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0073] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0074] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0075] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0076] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0077] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0078] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to the laptop computer 1011, tablet computer 1012 or mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0079] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0080] It should be noted that the artificial intelligence-based image processing method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based image processing device is generally set in the server / terminal device.

[0081] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0082] Continue to refer Figure 2 , showing a flow chart of an embodiment of an image processing method based on artificial intelligence according to the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted. The image processing method based on artificial intelligence provided in the embodiment of the present application can be applied to any scenario that requires image processing, and the image processing method based on artificial intelligence can be applied to products in these scenarios, for example, image processing in the field of digital medicine. The image processing method based on artificial intelligence comprises the following steps:

[0083] Step S201, obtaining an initial pathological image to be processed.

[0084] In this embodiment, the electronic device (eg, Figure 1The server / terminal device shown in the figure) can obtain the initial pathological image through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or developed in the future. The executor of this application is an image processing system, or simply referred to as a system. The above-mentioned initial pathological image may include a region of interest (RoIs) extracted from a whole slice image (WSI) or a pathological image directly obtained from a database.

[0085] Step S202: performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image.

[0086] In this embodiment, the specific implementation process of performing the image adjustment processing on the initial pathological image to obtain the corresponding pathological image will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0087] Step S203, calling a preset image processing model.

[0088] In this embodiment, the above-mentioned image processing model is a model generated after training a preset potential diffusion model based on pre-constructed pathological sample data. The specific construction process will be further described in detail in subsequent specific embodiments of this application and will not be elaborated here.

[0089] Step S204: performing image generation processing on the pathological image based on the image processing model to obtain a corresponding reconstructed image.

[0090] In this embodiment, the specific implementation process of performing image generation processing on the pathological image based on the image processing model to obtain the corresponding reconstructed image will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0091] Step S205 , optimizing the reconstructed image based on a preset fragmentary time step tracking strategy to obtain a corresponding target pathological image.

[0092] In this embodiment, the above-mentioned specific implementation process of optimizing the reconstructed image based on the preset fragmentary time step tracking strategy to obtain the corresponding target pathological image will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated in detail here.

[0093] Step S206: outputting the target pathological image.

[0094] In this embodiment, the specific implementation process of the above-mentioned output processing of the target pathological image will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated in detail here.

[0095] The present application obtains a pathological image by performing image adjustment processing on the initial pathological image to be processed, and then performs image generation processing on the pathological image based on the image processing model to obtain a reconstructed image, and further optimizes the reconstructed image based on a preset fragmentary time step tracking strategy to obtain a target pathological image, effectively improving the clarity and image quality of the generated target pathological image. The subsequent output processing of the target pathological image can effectively improve the efficiency and accuracy of pathological image analysis.

[0096] In some optional implementations, the image processing model includes an encoder and a decoder; step S204 includes the following steps:

[0097] The pathological image is input into the image processing model.

[0098] In this embodiment, a pre-trained image processing model is loaded. The image processing model is a model including an encoder and a decoder generated after training a preset potential diffusion model based on pre-built pathological sample data. The image processing model includes an encoder ε, a decoder D, and any related weights and parameters. In addition, generation parameters such as time step T, noise addition ratio, weight coefficient, etc. can be set according to the requirements of the high-resolution image to be generated. These parameters usually need to be adjusted according to the performance of the image processing model during training to achieve the best generation effect.

[0099] The pathological image is converted by the encoder to obtain a corresponding initial latent vector.

[0100] In this embodiment, the input pathological image is converted into a latent representation through the encoder ε. This is the first step in the generation process, which converts the pathological image from the pixel space to the latent space and obtains the corresponding initial latent vector.

[0101] Noise is added to the initial latent vector to obtain a corresponding target latent vector.

[0102] In this embodiment, starting from the above initial latent vector, noise is gradually added to the initial latent vector according to the time step T and the noise addition ratio to obtain a series of target latent variables. This process simulates the diffusion process, gradually changing the image from a clear state to a noisy state.

[0103] The target latent vector is subjected to step-by-step denoising based on the decoder to obtain a corresponding output image.

[0104] In this embodiment, after obtaining the final target latent variable, the decoder D is used to decode the target latent variable back to the pixel space to obtain the corresponding output image, that is, the reconstructed image. This process simulates the denoising process, gradually restoring the noisy image to a clear image.

[0105] The output image is used as the reconstructed image.

[0106] The present application inputs the pathological image into the image processing model, then converts the pathological image based on the use of an encoder to obtain an initial latent vector, adds noise to the initial latent vector to obtain a target latent vector, and then gradually denoises the target latent vector based on the use of a decoder, thereby achieving efficient and accurate image generation processing for the pathological image, effectively improving the clarity and image quality of the generated reconstructed image.

[0107] In some optional implementations of this embodiment, the image processing model further includes a fragmentary time step tracking module; step S205 includes the following steps:

[0108] Call the fragmented time-step tracking module.

[0109] In this embodiment, the above-mentioned fragmented time step tracking module is a module included in the image processing model corresponding to the above-mentioned fragmented time step tracking strategy. This module reduces the computational cost by utilizing multi-resolution expert knowledge in latent diffusion models (LDMs). It runs during inference without additional training.

[0110] Specifically, the episodic time-step tracking strategy exploits multi-resolution expert knowledge in latent diffusion models (LDMs) to reduce computational cost and runs during inference without additional training.

[0111] Denoising process: The denoising procedure processes potential patches according to the time step t. is the denoising process of latent diffusion models (LDMs) trained through the pathological alignment framework, processing latent patches x in the image l .c s is texture-guided. The denoising step for each patch is expressed as:

[0112]

[0113] Uneven time steps: Since the denoising process only processes one potential patch x at a time l Operations may lead to uneven time steps for the entire image. This problem is solved by denoising all latent patches in time step T in turn and updating the time step T←T-1.

[0114] Weight calculation: To ensure smooth transitions between overlapping tiles and reduce tiling artifacts, the latent vectors are weighted according to their distance from the patch center, ensuring that the weight map is normalized in the range [0.5, 1]. The weight w formula is:

[0115]

[0116] Among them, p ′ ,q ′ L represents the center of the potential patch in each direction, and p and q represent the horizontal and vertical coordinates of a pixel in the currently processed image patch. p Represents the width of a single latent patch. The final value of the latent vector in the target coordinates is calculated by weighted averaging all the estimated values ​​of the latent vector to prevent visible seams or discontinuities caused by reusing the same image patches or features when synthesizing large images. By using a weighted average method to calculate the final value of the latent vector, smooth transitions between different patches of the image can be ensured, thereby reducing or eliminating these artifacts.

[0117] The specific implementation steps include:

[0118] Synthetic Pathology Overview Image I ref : Reference input generated by a diffusion model trained with pathology knowledge.

[0119] Noise injection: Inject noise δ into the reference image t , create a noise reference image latent image X.

[0120] Latent tile processing: resize the latent image X, divide it into N latent tiles, and combine these tiles using weights w.

[0121] Generating synthetic high-resolution pathological tissue I h : Through the above process, a synthetic high-resolution pathological tissue image free of artifacts is finally generated, which contains spatial context related to cancer.

[0122] The reconstructed image is divided into a plurality of potential patches based on the segmented time step tracking module.

[0123] In this embodiment, the reconstructed image may be divided into N potential patches by using a fragmented time step tracking module. There is no specific limitation on the value of N, which may be determined according to actual processing requirements.

[0124] Each of the potential plaques is subjected to a separate denoising process to obtain a denoised designated potential plaque.

[0125] In this embodiment, an independent denoising network or a shared denoising network can be used to perform denoising on each of the potential patches to obtain a denoised designated potential patch, wherein the time step T can be further updated or other related parameters can be adjusted according to the denoising result and the design of the image processing model.

[0126] All the denoised designated potential patches are combined based on a preset weight calculation strategy to obtain a corresponding combined image.

[0127] In this embodiment, the denoised designated potential patches can be combined using a potential weight map according to the policy content of the weight calculation policy to generate a more natural image. The potential weight map can be a matrix of the same size as the potential patches, where each element represents the weight of the corresponding patch.

[0128] The combined image is used as the target pathological image.

[0129] After performing image generation processing on the pathological image based on the image processing model to obtain a reconstructed image, the present application will intelligently divide the reconstructed image into multiple potential plaques based on the use of a fragmented time step tracking module, and perform individual denoising processing on each potential plaque to obtain a denoised designated potential plaque, and then combine all the denoised designated potential plaques based on a weight calculation strategy, thereby further improving the image naturalness and image quality of the generated target pathological image.

[0130] In some optional implementations, step S206 includes the following steps:

[0131] Perform image magnification processing on the target pathological image to obtain a corresponding first pathological image.

[0132] In this embodiment, the target pathological image can be magnified by using bilinear interpolation, bicubic interpolation or a more advanced method, such as a super-resolution reconstruction method based on deep learning, so as to obtain a corresponding first pathological image.

[0133] Perform image sharpening processing on the first pathological image to obtain a corresponding second pathological image.

[0134] In this embodiment, the image sharpening process may be performed by using a Laplacian operator or other sharpening filters.

[0135] The second pathological image is subjected to super-resolution reconstruction processing to obtain a corresponding third pathological image.

[0136] In this embodiment, the second pathological image may be reconstructed with super-resolution using super-resolution reconstruction technology to obtain a corresponding third pathological image, wherein one or more additional deep neural networks may be used to enlarge the low-resolution image to a high resolution and restore more details.

[0137] The third pathological image is outputted.

[0138] In this embodiment, the generated third pathological image can be saved in a storage device, and displayed and analyzed, thereby completing the output processing of the third pathological image.

[0139] After obtaining the target pathological image, the present application will also intelligently perform image magnification processing, image sharpening processing and super-resolution reconstruction processing on the target pathological image, thereby effectively improving the resolution and quality of the generated third pathological image. Subsequently, the third pathological image is output, which can improve the output intelligence and accuracy of the pathological image.

[0140] In some optional implementations, step S202 includes the following steps:

[0141] The initial pathological image is resized to obtain a corresponding first initial pathological image.

[0142] In this embodiment, an image scaling operation may be performed on the initial pathological image to ensure that the obtained first initial pathological image meets the input size requirement of the image processing model.

[0143] The first initial pathological image is normalized to obtain a corresponding second initial pathological image.

[0144] In this embodiment, normalization is to convert the pixel values ​​of the pathological image into a standard range (usually between 0 and 1), which helps the training and convergence of the model.

[0145] Performing format conversion processing on the second initial pathological image to obtain a corresponding third initial pathological image.

[0146] In this embodiment, the second initial pathological image can be converted into a specific format, such as a PyTorch tensor, a TensorFlow tensor, etc., according to the requirements of the image processing model, so as to obtain a corresponding third initial pathological image.

[0147] The third initial pathological image is used as the pathological image.

[0148] The present application obtains a corresponding first initial pathology image by resizing the initial pathology image; then normalizes the first initial pathology image to obtain a corresponding second initial pathology image; then performs format conversion on the second initial pathology image to obtain a corresponding third initial pathology image; and subsequently uses the third initial pathology image as the pathology image. After obtaining the initial pathology image to be processed, the present application can efficiently and accurately complete the image adjustment processing of the initial pathology image by resizing, normalizing, and format conversion the initial pathology image, thereby effectively ensuring that the generated pathology image meets the input requirements of the image processing model, and ensuring the accuracy and regularity of the generated initial pathology image.

[0149] In some optional implementations of this embodiment, before step S203, the electronic device may further perform the following steps:

[0150] Obtain a pre-acquired pathology image dataset.

[0151] In this embodiment, the BRACS dataset can be selected as the above-mentioned pathological image dataset, wherein the BRACS dataset contains pathological images related to breast cancer extracted from 547 whole slice images (WSI). The dataset contains 4539 regions of interest (RoIs), each of which is labeled with one of seven cancer subtypes: normal (N), pathological benign (PB), normal mammary duct hyperplasia (UDH), flat epithelial atypical hyperplasia (FEA), atypical mammary duct hyperplasia (ADH), ductal carcinoma in situ (DCIS), and invasive carcinoma (IC). In addition, in order to train our generator, the RoIs are divided into 512×512 patches of 64 pixels overlap. In order to solve potential problems related to imbalanced data distribution, the patch-level training data is limited to 6000 patches per class. In addition, overview-level data is extracted by segmenting the RoIs into 2048×2048 large blocks with 256 pixels overlap and then resizing them to 512×512 size.

[0152] The pathological image data set is processed by data segmentation and data description generation to obtain corresponding pathological sample data.

[0153] In this embodiment, the data segmentation process includes: segmenting RoIs into 512×512 patches with 64 pixels overlap to solve the problem of data distribution imbalance and ensure that each class does not exceed 6000 patches. At the same time, RoIs are segmented into 2048×2048 patches with 256 pixels overlap, and then resized to 512×512 size for extracting overview level data.

[0154] The data description generation process includes: using the category information of each image to generate a description text string containing spatial level information such as "whole" and "patch".

[0155] In order to solve the problem of insufficient data annotation, this application proposes a data preparation protocol. The protocol generates a description string c by using the class information of each image. s This description string c s It is automatically generated based on the category information of the image, rather than relying on detailed annotations by pathologists.

[0156] Among them, for the addition of space level: in the description text string c s In , spatial levels such as "whole" and "patch" are added. This means that the description contains not only the category information of the image, but also the spatial information of the image, which helps the generator understand the different parts and overall structure of the image. In addition, for the consideration of inter-class diversity risk: Although this data preparation protocol ensures that each image has a description text, due to c s The generation of is based on limited class information, which may cause the generated images to have risks in terms of inter-class diversity. That is, if the description string c s Too sparse or not detailed enough, the generator may produce images that are similar in categories but lack diversity.

[0157] The latent diffusion model is invoked.

[0158] In this embodiment, the potential diffusion model at least includes an encoder ε and a decoder D, and may also include related network layers, such as a convolutional layer, a fully connected layer, etc. The appropriate network depth and width are selected according to the complexity of the model.

[0159] Among them, latent diffusion models (LDMs) are a kind of generative model, encoder ε and decoder are the key components, which are used to encode the image into the latent representation and decode the latent representation back to the image space respectively.

[0160] Encoding process: The input image I is converted into an embedded value x through a predefined encoder ε 0 , that is, x 0 =ε(I).

[0161] Diffusion process: Between time steps t = 1 and T, the original latent variable x is gradually diffused through the Markov process. 0 At each step t, the latent variable x t Calculated by the following formula:

[0162]

[0163] Among them, w~N(0,I) is a random noise term drawn from a normal distribution with mean 0 and covariance matrix I as the identity matrix. Parameter α t Controls the noise schedule, i.e. the amount of noise added to the data at each step.

[0164] Loss function: loss L of the diffusion model r Defined as:

[0165]

[0166] Among them, σ t is the noise scheduling term, represents the image generation process of the text-guided diffusion model, c s As a conditional vector providing text guidance, x 0 is the image that the diffusion model attempts to reconstruct, is the square of the L2 norm, used to measure the generated image With the original image x 0 The difference between.

[0167] Generate the corresponding target total loss based on the preset reconstruction loss and feature entropy loss.

[0168] In this embodiment, the reconstruction loss L r Including: using mean square error (MSE) or other image reconstruction related loss functions to measure the difference between the generated image and the original image. The above feature entropy loss L e Includes: Designing an algorithm to compare the representation similarity between images with the same description and images with different descriptions to optimize model performance. The target total loss includes: L LDM =L r +βL e . Where L LDM represents the total target loss used for optimization, and β is a weight coefficient used to balance the reconstruction loss L r and feature entropy loss L e The importance of.

[0169] Specifically, Feature Entropy Loss (FEL) is a method used to improve the intra-class generation performance while addressing inter-class diversity when dealing with sparsely labeled pathological image learning tasks.

[0170] The feature entropy loss aims to compare sThe core idea of ​​this method is to improve the consistency within a class and the difference between classes by reducing the distance between images with the same prompt and increasing the distance between images with different prompts.

[0171] An improved cross entropy is used to achieve this goal, and its mathematical form is:

[0172]

[0173] Where z = e(x t ,c s ) is the intermediate block latent variable generated by the encoder, e is the encoder of the denoising U-net, and x t is the input image, c s is the cue; k is the memory latent variable corresponding to each possible cue of the pathological image; τ is the temperature constant used to adjust the model sensitivity; k + is a memory latent variable with the same cue as z; k - is a memory latent variable with a different cue from z.

[0174] The loss function is calculated by calculating z and k + The similarity between them is relative to all k i , to ensure that images with the same prompt have higher representation similarity.

[0175] Update of memory latent variables: Use the exponential moving average (EMA) method to update the memory latent variable k, that is, k n+1 =αk n-1 +(1-α)k n This approach can introduce variance in the comparison target, thus preventing overfitting.

[0176] In addition, using the reconstruction loss (Reconstruction Loss) L r and Feature Entropy Loss L e To train the generator, we can get the total loss formula:

[0177] L LDM =L r +βL e

[0178] Among them, L LDM represents the comprehensive loss function used for optimization, and β is a weight coefficient used to balance the reconstruction loss L r and feature entropy loss L eBy implementing a training scheme based on feature entropy loss, a generator with multi-scale pathology knowledge is trained, which improves the generation capability of high-resolution images.

[0179] Based on the target total loss, the potential diffusion model is trained and evaluated using the pathological sample data to obtain a specified model that meets the preset construction requirements.

[0180] In this embodiment, based on the target total loss, the pathological sample data is used to train and evaluate the latent diffusion model to obtain a specified model that meets the preset construction requirements, including: dividing the pathological image data set into a training set, a validation set, and a test set. Load images and corresponding descriptions from the training set. Input the image into the encoder to obtain a latent representation. Subsequently, between time steps t=1 and T, noise is added to the latent representation through a Markov process to obtain a specified latent representation after adding noise. Use the decoder D to decode the specified latent representation back to the pixel space to obtain a reconstructed image, and calculate the reconstruction loss L r At the same time, calculate the feature entropy loss L e According to the total loss L LDM , update the weights of the encoder ε and decoder D through the back-propagation algorithm. Repeat the above steps until the model achieves satisfactory performance on the validation set or reaches the preset number of iterations. After the training is completed, the trained encoder ε, decoder D, and related weights and parameters are saved to the storage medium to obtain the initial model. Use the test set to evaluate the initial model, including the quality of the reconstructed image, the degree of detail retention, and the similarity with the original image. According to the evaluation results, make necessary adjustments and optimizations to the initial model, and finally obtain a specified model that meets the construction requirements (for example, the quality of the reconstructed image, the degree of detail retention, and the similarity with the original image meet expectations) and serve as the final image processing model.

[0181] The designated model is used as the image processing model.

[0182] The present application performs data segmentation and data description generation processing on a pre-collected pathological image data set to obtain corresponding pathological sample data, and then generates a corresponding target total loss based on a preset reconstruction loss and feature entropy loss. Based on the target total loss, the pathological sample data is used to train and evaluate the potential diffusion model to obtain an image processing model that meets the construction requirements, thereby achieving rapid construction of the image processing model, improving the construction efficiency of the image processing model, and ensuring the model effect of the obtained image processing model.

[0183] In digital pathology, cancer lesions are identified by analyzing the spatial context in pathology images. Because pathology full-slice images are usually characterized by high resolution, small inter-class differences, and sparse labels, it is challenging to synthesize such complex spatial context. Multi-resolution pathology image generators require unified description text to assist the training process. This is because the description text can be used as one of the inputs to the generator to help the model understand the image content and structure it needs to generate. Due to the high expertise requirements of pathologists and the high associated annotation costs, it is unrealistic to provide a detailed and precise description for each pathology image. This means that in practical applications, there is a problem of insufficient data annotation.

[0184] To address these challenges, this application proposes PathUp, a novel diffusion model tailored for synthesizing multi-class high-resolution pathology images.

[0185] Inspired by the fact that the spatial arrangement formed by the collection of tissues and cells in real pathological images can be used as a biomarker for tumor classification, this application proposes a pathological image synthesis pipeline. This application involves a pathological alignment mechanism that integrates multi-class pathological images of different resolutions and texture cues into a diffusion model. To ensure the robust generation of texture-related features, a novel feature entropy loss is introduced. For the synthesis of high-resolution images, the strategy adopted is to split the latent code into overlapping fragments and deploy fragment-based time-step tracking. Therefore, this application mitigates tiling artifacts and effectively bridges the gap between multi-resolution images.

[0186] This application includes latent space patch-wise time-step tracking, which helps generate high-quality images without tiling artifacts. Expert pathology knowledge is integrated into the model through a pathology alignment mechanism. To ensure robust generation of lesion subtype and scale information, a feature entropy loss function is introduced. The effectiveness of our approach is confirmed through qualitative and quantitative evaluations, supplemented by human expert evaluations, demonstrating the authenticity of the generated synthetic data. In addition, the generated images have the potential utility as an augmentation method to improve the performance of downstream tasks such as cancer subtype classification.

[0187] Specifically, the PathUp framework proposed in this application has two key components:

[0188] (a) Patho-align module. This module aligns multiple types of pathological images at different spatial levels and texture descriptions. sIntegrate into the potential diffusion model. The input dimension of the original image I is the image data I of [samples, channels, height, width], where samples represents the number of images in the dataset; channels represents the number of color channels in each image; height represents the vertical size of each image (in pixels); width represents the horizontal size of each image.

[0189] Texture Guidec s As the conditional vector providing text description, the noise w~N(0,I) is generated through the denoising network, the diffusion model (DM) is trained, the parameters such as the potential patch size p, overlapping pixels o, and the potential weight w of the patch are adjusted, and the high-resolution image I is output. h .

[0190] Reconstruction Loss function L r is a measure of the value of the given text condition c s The diffusion model potentially represents With the original latent representation x 0 By minimizing this loss, the decoder can restore the original image on the latent representation with added noise. Through training, the diffusion model learns how to generate high-quality images that match the text description given the latent representation of noise.

[0191] Training feature entropy loss (Entropy Loss) L e , using memory latent to ensure the latent embedding of the same category q Shows a closer distance.

[0192] (b) In order to facilitate the generation of high-resolution images from low-res reference images without additional training, a patch-wise timestep tracking module is proposed. This module solves the problem of uneven timesteps by individually denoising the segmented latent patches in timestep T and updating the timestep T←T-1. It also uses the latent weight map to remove tiling artifacts, resize the latent image X, divide it into N latent patches, and combine these patches with weights w to generate a high-definition image I o .

[0193] In some optional implementations of this embodiment, after step S205, the electronic device may further perform the following steps:

[0194] A specified storage strategy corresponding to the target pathological image is obtained.

[0195] In this embodiment, the above-mentioned designated storage strategy may specifically adopt strategies such as cyclic overwriting or scheduled backup to ensure that the target pathological image to be stored will not be lost due to insufficient storage space.

[0196] A designated storage medium corresponding to the target pathological image is called.

[0197] In this embodiment, there is no specific limitation on the selection of the above-mentioned designated storage medium, and it may include, for example, a local database, a cloud server, a blockchain, etc.

[0198] Based on the designated storage strategy, the target pathological image is stored in the designated storage medium.

[0199] In this embodiment, the target pathological image may be stored in the designated storage medium in a storage manner with a designated storage strategy.

[0200] After obtaining the target pathological image, the present application will intelligently obtain the specified storage strategy corresponding to the target pathological image, and call the specified storage medium corresponding to the target pathological image, and then store and process the target pathological image based on the obtained specified storage strategy and the specified storage medium, thereby ensuring the data security of the target pathological image and ensuring that the target pathological image will not be lost due to insufficient storage space, effectively improving the storage intelligence of the target pathological image.

[0201] In some optional implementations, the user information obtained is subject to the user's consent and complies with relevant laws and policies.

[0202] In addition, the non-Company software tools or components that appear in the embodiments of the present application are merely examples and do not represent actual use.

[0203] In addition, this application proposes a generative model that learns to generate multi-resolution lesion subtypes from pathology images. Pathology alignment is introduced, which combines expert pathology knowledge with multi-class images. A feature entropy loss function is proposed to increase the inter-class diversity of synthetic images. A fragmented time step tracking strategy is proposed within the framework of the latent diffusion model. This strategy enhances the model's ability to generate high-resolution images while solving the tiling artifact problem by utilizing latent weights. Studies have shown that this method is able to generate realistic pathology images of different resolutions. Synthetic pathology images can be used as a data augmentation method, demonstrating the efficacy of augmented data in downstream tasks such as lesion subtype classification.

[0204] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0205] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned target pathology image, the above-mentioned target pathology image can also be stored in a node of a blockchain.

[0206] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0207] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0208] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0209] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0210] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0211] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of an image processing device based on artificial intelligence. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0212] like Figure 3 As shown, the artificial intelligence-based image processing device 300 described in this embodiment includes: a first acquisition module 301, an adjustment module 302, a first calling module 303, a generation module 304, an optimization module 305 and an output module 306. Among them:

[0213] A first acquisition module 301 is used to acquire an initial pathological image to be processed;

[0214] An adjustment module 302 is used to perform image adjustment processing on the initial pathological image to obtain a corresponding pathological image;

[0215] A first calling module 303 is used to call a preset image processing model; wherein the image processing model is a model generated by training a preset potential diffusion model based on pre-constructed pathological sample data;

[0216] A generation module 304 is used to perform image generation processing on the pathological image based on an image processing model to obtain a corresponding reconstructed image;

[0217] An optimization module 305 is used to optimize the reconstructed image based on a preset fragment-based time step tracking strategy to obtain a corresponding target pathological image;

[0218] The output module 306 is used to perform output processing on the target pathological image.

[0219] In some optional implementations of this embodiment, the image processing model includes an encoder and a decoder; the generation module 304 includes:

[0220] An input submodule, used for inputting the pathological image into the image processing model;

[0221] A first conversion submodule, configured to convert the pathological image through the encoder to obtain a corresponding initial latent vector;

[0222] An adding submodule, used for performing noise adding processing on the initial latent vector to obtain a corresponding target latent vector;

[0223] A denoising submodule, configured to perform a step-by-step denoising process on the target latent vector based on the decoder to obtain a corresponding output image;

[0224] The first determining submodule is configured to use the output image as the reconstructed image.

[0225] In some optional implementations of this embodiment, the image processing model further includes a fragmentary time step tracking module; the optimization module 305 includes:

[0226] A calling submodule, used for calling the fragmentary time step tracking module;

[0227] a partitioning submodule, configured to partition the reconstructed image into a plurality of potential patches based on the fragmented time step tracking module;

[0228] A generating submodule, used for performing a separate denoising process on each of the potential plaques to obtain a denoised designated potential plaque;

[0229] A combination submodule, used for combining all the denoised designated potential patches based on a preset weight calculation strategy to obtain a corresponding combined image;

[0230] The second determining submodule is configured to use the combined image as the target pathological image.

[0231] In some optional implementations of this embodiment, the output module 306 includes:

[0232] A first processing submodule, configured to perform image magnification processing on the target pathological image to obtain a corresponding first pathological image;

[0233] A second processing submodule, configured to perform image sharpening processing on the first pathological image to obtain a corresponding second pathological image;

[0234] A third processing submodule is used to perform super-resolution reconstruction processing on the second pathological image to obtain a corresponding third pathological image;

[0235] An output submodule is used to perform output processing on the third pathological image.

[0236] In some optional implementations of this embodiment, the adjustment module 302 includes:

[0237] A fourth processing submodule, configured to resize the initial pathological image to obtain a corresponding first initial pathological image;

[0238] a fifth processing submodule, configured to perform normalization processing on the first initial pathological image to obtain a corresponding second initial pathological image;

[0239] a sixth processing submodule, configured to perform format conversion processing on the second initial pathological image to obtain a corresponding third initial pathological image;

[0240] The third determining submodule is configured to use the third initial pathological image as the pathological image.

[0241] In some optional implementations of this embodiment, the image processing device based on artificial intelligence further includes:

[0242] A second acquisition module is used to acquire a pre-collected pathological image data set;

[0243] A first processing module, used for performing data segmentation and data description generation processing on the pathological image data set to obtain corresponding pathological sample data;

[0244] A second calling module, used for calling the potential diffusion model;

[0245] A construction module is used to generate the corresponding target total loss based on the preset reconstruction loss and feature entropy loss;

[0246] A second processing module is used to train and evaluate the potential diffusion model based on the target total loss using the pathological sample data to obtain a specified model that meets the preset construction requirements;

[0247] A determination module is used to use the specified model as the image processing model.

[0248] In some optional implementations of this embodiment, the image processing device based on artificial intelligence further includes:

[0249] A third acquisition module, used to acquire a specified storage strategy corresponding to the target pathological image;

[0250] A third calling module, used to call a designated storage medium corresponding to the target pathological image;

[0251] A storage module is used to store the target pathological image in the designated storage medium based on the designated storage strategy.

[0252] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0253] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0254] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0255] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of an image processing method based on artificial intelligence, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0256] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the artificial intelligence-based image processing method.

[0257] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0258] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned artificial intelligence-based image processing method.

[0259] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0260] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. An image processing method based on artificial intelligence, characterized in that: The steps include: Acquiring an initial pathological image to be processed; Performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image; Calling a preset image processing model; wherein the image processing model is a model generated by training a preset potential diffusion model based on pre-constructed pathological sample data; Performing image generation processing on the pathological image based on the image processing model to obtain a corresponding reconstructed image; The reconstructed image is optimized based on a preset fragmentary time step tracking strategy to obtain a corresponding target pathological image; The target pathological image is outputted.

2. The image processing method based on artificial intelligence according to claim 1, characterized in that: The image processing model includes an encoder and a decoder; the step of performing image generation processing on the pathological image based on the image processing model to obtain a corresponding reconstructed image specifically includes: inputting the pathological image into the image processing model; The pathological image is converted by the encoder to obtain a corresponding initial latent vector; Performing noise addition processing on the initial latent vector to obtain a corresponding target latent vector; Based on the decoder, the target latent vector is subjected to a stepwise denoising process to obtain a corresponding output image; The output image is used as the reconstructed image.

3. The image processing method based on artificial intelligence according to claim 1, characterized in that: The image processing model also includes a fragmented time step tracking module; the step of optimizing the reconstructed image based on a preset fragmented time step tracking strategy to obtain a corresponding target pathological image specifically includes: Calling the fragmented time step tracking module; dividing the reconstructed image into a plurality of potential patches based on the segmented time step tracking module; Performing a separate denoising process on each of the potential plaques to obtain a denoised designated potential plaque; Combining all the denoised designated potential patches based on a preset weight calculation strategy to obtain a corresponding combined image; The combined image is used as the target pathological image.

4. The image processing method based on artificial intelligence according to claim 1, characterized in that: The step of outputting the target pathological image specifically includes: Performing image magnification processing on the target pathological image to obtain a corresponding first pathological image; Performing image sharpening processing on the first pathological image to obtain a corresponding second pathological image; performing super-resolution reconstruction processing on the second pathological image to obtain a corresponding third pathological image; The third pathological image is outputted.

5. The image processing method based on artificial intelligence according to claim 1, characterized in that: The step of performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image specifically includes: Performing a size adjustment process on the initial pathological image to obtain a corresponding first initial pathological image; Normalizing the first initial pathological image to obtain a corresponding second initial pathological image; Performing format conversion processing on the second initial pathological image to obtain a corresponding third initial pathological image; The third initial pathological image is used as the pathological image.

6. The image processing method based on artificial intelligence according to claim 1, characterized in that: Before the step of calling the preset image processing model, the method further includes: obtaining a pre-collected pathological image dataset; Performing data segmentation and data description generation processing on the pathological image data set to obtain corresponding pathological sample data; calling the potential diffusion model; Generate the corresponding target total loss based on the preset reconstruction loss and feature entropy loss; Based on the target total loss, the potential diffusion model is trained and evaluated using the pathological sample data to obtain a specified model that meets the preset construction requirements; The designated model is used as the image processing model.

7. The image processing method based on artificial intelligence according to claim 1, characterized in that: After the step of optimizing the reconstructed image based on the preset fragment-based time step tracking strategy to obtain the corresponding target pathological image, the method further includes: Acquire a specified storage strategy corresponding to the target pathological image; calling a designated storage medium corresponding to the target pathological image; Based on the designated storage strategy, the target pathological image is stored in the designated storage medium.

8. An image processing device based on artificial intelligence, characterized in that: include: A first acquisition module, used for acquiring an initial pathological image to be processed; An adjustment module, used for performing image adjustment processing on the initial pathological image to obtain a corresponding pathological image; A first calling module is used to call a preset image processing model; wherein the image processing model is a model generated by training a preset potential diffusion model based on pre-constructed pathological sample data; A generation module, used for performing image generation processing on the pathological image based on an image processing model to obtain a corresponding reconstructed image; An optimization module, used for optimizing the reconstructed image based on a preset fragmentary time step tracking strategy to obtain a corresponding target pathological image; An output module is used to perform output processing on the target pathological image.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the artificial intelligence-based image processing method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the artificial intelligence-based image processing method according to any one of claims 1 to 7.

Images