Microscopic image enhancement and restoration method and system for tumor pathology

Abstract

The invention discloses a microscopic image enhancement and restoration method and system for tumor pathology, and relates to the technical field of medical imaging. Comprising the following steps: S1, acquiring a normalized value of a microscopic image, comparing the normalized value with a preset normalized value, and determining a final microscopic image; s2, determining an artifact area and an artifact type corresponding to the final microscopic image, and processing the artifact area; and S3, optimizing the microscopic image after the artifact region processing, and reconstructing a three-dimensional blood vessel structure in the microscopic image to obtain an optimized microscopic image. According to the invention, artifact identification is carried out on the preprocessed microscopic image through the artifact detection model, so that tiny noise existing in the microscopic image can be reduced, and the identification efficiency and the identification result can be improved, so that the artifact area in the microscopic image can be better identified, and the identification accuracy of the microscopic image is improved. And the recovery efficiency of the artifact region is further improved.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a method and system for enhancing and restoring microscopic images of tumor pathology. Background Art

[0002] With the development of medical technology, the role of pathology in disease diagnosis and treatment has become increasingly prominent. Especially for complex diseases such as cancer, accurate pathological analysis is a prerequisite for formulating effective treatment plans. However, traditional pathology relies on doctors observing tissue sections through a microscope, which is not only time-consuming but also easily affected by subjective factors.

[0003] Digital pathology has become an important research field. Digital microscopes can convert pathological images into electronic formats for easy storage, transmission and analysis. However, these images are often affected by various factors (such as noise and blur), which can reduce image quality and affect the judgment of pathologists. In addition, with the advancement of high-resolution imaging technology, how to efficiently process large amounts of data has become a new challenge.

[0004] The Chinese invention patent with publication number CN112017131A discloses a method, device and computer-readable storage medium for removing metal artifacts from CT images, including the steps of: collecting images containing metal artifacts and images without metal artifacts, processing to obtain corresponding projection data; reconstructing the two generated projection data respectively to generate images containing metal artifacts and images without artifacts, further processing to obtain initial artifact suppression images and data integrity maps; designing neural networks and loss functions, and the neural network outputs images after artifact removal, etc. The invented method for removing metal artifacts from CT images utilizes the correlation between metal artifacts and surrounding structures, as well as the relationship with tissue components, and overcomes the problem that existing image domain technologies cannot integrate data integrity information in metal scanning data into the image domain processing process, thereby improving the accuracy of artifact correction and the quality of CT images.

[0005] In the process of processing pathological images by the above-mentioned and similar artifact removal methods, pathological images usually contain complex noise components, such as random noise, speckle noise, etc. These noises will seriously affect the image quality, resulting in the loss of key details, thus affecting the accuracy of diagnosis. At the same time, due to the physical limitations of the sample preparation process or the limitations of equipment performance, the acquired images often have varying degrees of blur, which further increases the difficulty of pathological feature recognition. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for microscopic image enhancement and restoration for tumor pathology to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for microscopic image enhancement and restoration for tumor pathology, comprising: S1: Image preprocessing: preprocessing the microscopic image to obtain a normalized value of the microscopic image, and comparing the normalized value with a preset normalized value, and determining a final microscopic image according to the comparison result; S2: Image artifact processing: using an artifact detection model, the final microscopic image is identified, an artifact region and an artifact type corresponding to the final microscopic image are determined, and the artifact region is processed according to the artifact type; S3: Optimizing the microscopic image: optimizing the microscopic image after the artifact area processing through multi-scale analysis, and reconstructing the three-dimensional vascular structure in the microscopic image through phase variance to obtain an optimized microscopic image; Processing the artifact area includes: S2.1: Determine artifact sorting: Identify artifact regions in the final microscopic image using an artifact detection model, determine the type and size corresponding to each artifact region, and sort all the artifact regions according to the type and size; S2.2: performing artifact repair: repairing, removing and marking the artifact areas according to the sorting of the artifact areas; S2.3: Perform artifact detection: Detect the microscopic image after artifact processing to determine whether all artifact areas in the microscopic image are repaired. When all artifact areas are repaired, execute the next step. Otherwise, repeat steps M1-S2.3 until all artifact areas are repaired.

[0008] Furthermore, the final microscopic image is determined, including: S1.1: Image processing: smoothing the microscopic image by using a Gaussian filter, processing the frequency of the microscopic image by using a histogram equalization, normalizing each pixel value of the microscopic image by using a linear normalization, and obtaining a normalized value corresponding to the microscopic image; S1.2: Setting a preset normalization value: Processing the microscopic image through the CNN model to obtain the preset normalization value, specifically: ; in: is the preset normalized value, is the original pixel value of the microscopic image at position (x, y), is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, is a convolutional neural network, are all trainable parameters of the convolutional neural network, is the index of the sequence number; S1.3: Determine the processed image: compare the normalized value obtained with the preset normalized value, obtain the difference between the normalized value and the preset normalized value, and compare the difference with the preset difference. When the difference is not greater than the preset difference, the microscopic image corresponding to the difference is the processed image. Otherwise, repeat steps S1.1 and S1.3 until the difference obtained is not greater than the preset difference.

[0009] Furthermore, the formula for obtaining all trainable parameters of the convolutional neural network is as follows: ; in: are the values ​​of all trainable parameters at the t+1th iteration, are the values ​​of all trainable parameters at the tth iteration, is the learning rate, is the loss function, is the gradient of the loss function with respect to all trainable parameters; The formula for obtaining the loss function is as follows: ; in: is the loss function, is the width of the microscopic image, is the height of the microscopic image, is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, is the preset normalized value, is the original pixel value of the microscopic image at position (x, y), are all trainable parameters of the convolutional neural network, is the standard pixel value of the i-th image at position (x, y), The index of the sequence number.

[0010] Furthermore, all the artifact areas are sorted, including: M1: Determine the artifact area: determine the artifact probability corresponding to each pixel through the artifact detection model, and convert the artifact probability into a binary image. Process the artifact area through the binary image and the connected domain marker; M2: Determine the type of artifact: determine the type of each artifact area according to the characteristics of the artifact area, including circular or nearly circular artifacts, strip-shaped artifacts and other artifacts; M3: Determine the artifact ranking: Determine the priority of the artifact area according to the area and the artifact type of the artifact area.

[0011] Furthermore, the artifact area is processed, including: M1.1: Determine the initial artifact area: Obtain the artifact probability corresponding to each pixel in the microscopic image through the artifact detection model, specifically: ; in: is the probability that the pixel at position (x, y) belongs to an artifact, is the U-Net network structure, is the original pixel value of the microscopic image at position (x, y), is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, are all trainable parameters of the U-Net model, is the index of the sequence number; M1.2: Obtaining a binary image: Converting the artifact probability corresponding to each pixel in the microscopic image into a binary image, and determining the artifact area based on the binary image, specifically: ; in: is the output binary image, is the probability that the pixel at position (x, y) belongs to an artifact, is the preset threshold; M1.3: Determine the connected area: mark the connected domain of the artifact area, obtain the connected domain label map and the area of ​​each connected domain, and compare the connected domain area with the minimum area threshold to determine the final artifact area.

[0012] Furthermore, the final artifact area is determined, including: N1: Connected domain labeling: label each artifact region using a connected domain labeling algorithm to obtain a label map corresponding to the microscopic image; N2: Get the area of ​​the connected domain: Get the area of ​​the connected domain according to the number of pixels in each connected domain, specifically: ; in: is the area of ​​the i-th connected domain, is the set of all pixel positions of the i-th connected domain, is the index of the sequence number, is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image; N3: Remove small areas: compare the area of ​​the connected domain with the minimum area threshold. When the area of ​​the connected domain is smaller than the minimum area threshold, the area of ​​the connected domain is deleted; otherwise, the area of ​​the connected domain is retained. N4: Merge adjacent artifact regions: Compare the minimum distance between two adjacent connected domain regions with the preset distance. When the minimum distance is less than the preset distance, merge the two adjacent connected domain regions corresponding to the minimum distance. Otherwise, do not merge the two adjacent connected domain regions. The formula for obtaining the minimum distance between two adjacent connected domain regions is as follows: ; in: Connected domain and connected domain The minimum distance between is the set of all pixel positions of the i-th connected domain, is the set of all pixel positions of the jth connected domain, is the horizontal coordinate of the pixel in the i-th connected domain, is the horizontal coordinate of the pixel in the jth connected domain, is the vertical coordinate of the pixel in the i-th connected domain, is the vertical coordinate of the pixel in the jth connected domain, , The index of the sequence number.

[0013] Furthermore, the minimum area threshold is determined as follows: According to the connected domain area, a connected domain area histogram is obtained, and two main peaks are determined from the connected domain area histogram, and the valley value between the two main peaks is the minimum area threshold.

[0014] Furthermore, the optimized microscopic images are obtained, including: S3.1: Obtaining a fused microscopic image: smoothing the microscopic image after the artifact area processing by Gaussian smoothing, downsampling the smoothed microscopic image by downsampling to obtain a reduced microscopic image, and fusing microscopic images at different scales to obtain a fused microscopic image; S3.2: Reconstructing a three-dimensional vascular structure: Obtaining B-scan data of the fused microscopic image, and determining the phase change of adjacent time points according to each B-scan position, and constructing a three-dimensional vascular structure according to the phase change.

[0015] A microscopic image enhancement and restoration system for tumor pathology uses any of the microscopic image enhancement and restoration methods for tumor pathology described above.

[0016] Compared with the prior art, the present invention has the following beneficial effects: Firstly, the present invention uses an artifact detection model to perform artifact recognition on the preprocessed microscopic image, which can not only reduce the tiny noise existing in the microscopic image itself, but also improve the recognition efficiency and recognition results, so that the artifact area in the microscopic image can be better recognized, and the type and size of the artifact area are recognized at the same time, and the artifact area is sorted according to the recognition result, and the corresponding artifact repair, removal and marking are performed according to the sorting result, thereby further improving the restoration efficiency of the artifact area; Second: After the present invention preliminarily determines the artifact probability through the artifact detection model, the artifact probability is converted into a binary image, and the corresponding artifact area is determined based on the binary image. At the same time, the identified artifact area is further processed through the connected domain marking, thereby improving the accuracy of artifact area identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the process of the microscopic image enhancement and restoration method of the present invention; Figure 2 is the connected domain area histogram in the present invention; Figure 3 It is a schematic diagram of the process of determining the artifact area in the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] In the process of processing pathological images by existing artifact removal methods, since pathological images usually contain complex noise components, such as random noise, speckle noise, etc., these noises will seriously affect the image quality, resulting in the loss of key details, thereby affecting the accuracy of diagnosis. At the same time, due to the physical limitations in the sample preparation process or the limitations of equipment performance, the acquired images often have varying degrees of blurring, which further increases the difficulty of identifying pathological features. The technical solution of the present application identifies the pre-processed microscopic image through the constructed artifact detection model, determines the artifact area, and sorts, repairs and detects the identified artifact area, and then detects the restored artifact area. At the same time, when the detection results show that all artifacts have been removed, the microscopic image after the artifact area is restored is optimized through multi-scale analysis, and the three-dimensional vascular structure is reconstructed through phase variance to obtain the processed microscopic image. In other words, not only can the tiny noise existing in the microscopic image itself be reduced, but also the recognition efficiency and recognition results can be improved, so that the artifact area in the microscopic image can be better recognized.

[0020] refer to Figure 1-Figure 3 This embodiment provides a microscopic image enhancement and restoration method for tumor pathology, and the microscopic image enhancement and restoration method includes the following steps: Step S1: Image preprocessing. That is, preprocess the microscopic image, that is, after image denoising, image enhancement and image normalization, obtain the normalized value corresponding to the microscopic image, and compare the obtained normalized value with the preset normalized value, and determine the final processed microscopic image according to the comparison result. The details are as follows: Step S1.1: Image processing. That is, the microscopic image is smoothed by a Gaussian filter so that the central pixel value of the microscopic image remains at the same value. Then, the frequency of each gray level of the microscopic image is maintained within a preset range by histogram equalization. At the same time, each pixel value of the image is normalized by linear normalization. It is worth noting that the image denoising, image enhancement and image normalization processing in this embodiment are all existing conventional technical means, so they will not be specifically explained in this embodiment.

[0021] Step S1.2: Setting a preset normalization value. That is, the microscopic image is processed by the CNN model to obtain a preset normalization value corresponding to the microscopic image, specifically: ; in: is the preset normalized value, is the original pixel value of the microscopic image at position (x, y), is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, is a convolutional neural network, are all trainable parameters of the convolutional neural network, The index of the sequence number.

[0022] Furthermore, all trainable parameters of a convolutional neural network The formula for obtaining is as follows: ; in: are the values ​​of all trainable parameters at the t+1th iteration, are the values ​​of all trainable parameters at the tth iteration, is the learning rate, is the loss function, is the gradient of the loss function with respect to all trainable parameters.

[0023] Furthermore, the loss function The formula for obtaining is as follows: ; in: is the loss function, is the width of the microscopic image, is the height of the microscopic image, is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, is the preset normalized value, is the original pixel value of the microscopic image at position (x, y), are all trainable parameters of the convolutional neural network, is the standard pixel value of the i-th image at position (x, y), The index of the sequence number.

[0024] Step S1.3: Determine the processed image. That is, the microscopic image is used as the input of the CNN model in step S1.2, and the corresponding preset normalized value is obtained by output. At the same time, the microscopic image is processed according to step S1.1, and the image normalized value corresponding to the microscopic image is obtained. Further, the image normalized value is compared with the preset normalized value, and the difference between the image normalized value and the preset normalized value is obtained. At the same time, the difference size is compared with the preset difference. When the obtained difference is not greater than the preset difference, the obtained microscopic image is the final microscopic image. Otherwise, steps S1.1 and S1.3 are repeated until the obtained difference is not greater than the preset difference. It is worth noting that the preset difference can be set according to the actual needs in actual use, so it will not be specifically explained in this embodiment.

[0025] Step S2: Image artifact processing. That is, the microscopic image preprocessed in step S1.3 is identified through the established artifact detection model, and the artifact area in the microscopic image is determined therefrom. At the same time, the artifact area is identified, the type corresponding to each artifact area is determined, and the artifact area is processed according to its corresponding artifact type. The details are as follows: Step S2.1: Determine the artifact sorting. That is, identify the artifact areas in the microscopic image through the established artifact detection model, determine the type and size corresponding to each artifact area, and sort all the artifact areas according to the determined type and size. The details are as follows: Step M1: Determine the artifact area. That is, through the constructed artifact detection model, determine the artifact probability corresponding to each pixel, and convert each artifact probability into a binary image. Then, the artifact area is processed through the binary image and connected domain marking. The details are as follows: Step M1.1: Determine the artifact probability. That is, through the artifact detection model, obtain the probability that each pixel in the microscopic image is an artifact, specifically: ; in: is the probability that the pixel at position (x, y) belongs to an artifact, is the U-Net network structure, is the original pixel value of the microscopic image at position (x, y), is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, are all trainable parameters of the U-Net model, The index of the sequence number.

[0026] That is to say, the probability that each pixel in the microscopic image belongs to an artifact can be obtained according to the artifact detection model, and the probability of the artifact area corresponding to the area where each pixel is located can be determined according to the obtained probability.

[0027] In the specific implementation process, a 256*256 grayscale image is used as the input of the artifact detection model, and its pixel range is [0,1]. Furthermore, after being processed by the artifact detection model, it can output a probability map of size 256*256, where each pixel value represents the probability that the corresponding position belongs to an artifact. Specifically, the pixel value at position (100,150) is 0.85, which means that the probability that the position (100,150) belongs to the artifact area is 85%.

[0028] Step M1.2: Obtain a binary image. That is, convert the probability image obtained in step M1.1 into a binary image, specifically: ; in: is the output binary image, is the probability that the pixel at position (x, y) belongs to an artifact, is the preset threshold.

[0029] That is, the binary image corresponding to the probability that each position belongs to the artifact area is obtained, and the artifact area is determined according to the size of the binary image. When it is 0, it indicates that the pixel does not belong to the artifact area, otherwise, it indicates that the pixel belongs to the artifact area.

[0030] In the specific implementation process, some pixel values ​​and corresponding binary image values ​​of a probability map of size 5*5 are shown in the following Table 1, specifically Table 1.

[0031] Table 1: Some pixel values ​​and binary image values ​​corresponding to the probability map

[0032] Furthermore, the preset threshold in this embodiment It is set to 0.5, so pixels greater than 0.5 are marked as 1, that is, the area where the pixels marked as 1 are located is the artifact area, and pixels not greater than 0.5 are marked as 0, that is, the area where the pixels marked as 0 are located is not the artifact area.

[0033] Step M1.3: Determine the connected area. That is, according to the artifact area determined in step M1.2, mark each artifact area as a connected domain to obtain a connected domain label map. At the same time, according to the connected domain label map, obtain the area of ​​each connected domain, and determine the final artifact area by comparing the area of ​​the connected domain with the minimum area threshold. The details are as follows: Step N1: Connected domain labeling. That is, a label is set for each artifact region determined in step M1.2 through a connected domain labeling algorithm (such as breadth-first search, depth-first search, and union-find), so that each artifact region corresponds to a connected domain number, that is, the corresponding label map is obtained.

[0034] Step N2: Obtain the area of ​​the connected domain. That is, obtain the number of pixels in each connected domain, and use the obtained number of pixels as the area of ​​the corresponding connected domain, specifically: ; in: is the area of ​​the i-th connected domain, is the set of all pixel positions of the i-th connected domain, is the index of the sequence number, is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image.

[0035] That is, by traversing the label map obtained in step N1, the number of times each label appears is determined, and the number of pixels corresponding to each label is recorded through a hash table or an array.

[0036] Step N3: Remove small areas. That is, compare the area of ​​the connected domain obtained in step N2 with the minimum area threshold. When the area of ​​the connected domain is smaller than the minimum area threshold, the area of ​​the connected domain is deleted, otherwise the area of ​​the connected domain is retained.

[0037] In this embodiment, the corresponding connected domain area histogram is obtained according to the areas of all connected domains obtained in step N2. At the same time, two main peaks are determined according to the connected domain area histogram, and the valley value between the two main peaks is used as the minimum area threshold.

[0038] In the specific implementation process, a connected domain area data set is set, specifically: [3, 5, 5, 7, 8, 8, 8, 10, 12, 14, 15, 15, 15, 18, 20, 22, 25, 30, 30, 30, 35, 40, 45, 50, 60]. At the same time, according to the connected domain area data set, the corresponding connected domain area histogram is obtained, such as Figure 2 Reference Figure 2 It can be seen that the smaller clusters are 5, 8 and 15, and the larger cluster is 30. That is to say, the two main peak areas are the smaller cluster (5, 8, 15) and the larger cluster (30). Further, in this embodiment, the minimum area threshold is set to 25. That is, the areas of the connected domains deleted are 3, 5, 5, 7, 8, 8, 8, 10, 12, 14, 15, 15, 15, 18, 20 and 22. The areas of the connected domains retained are 25, 30, 30, 30, 35, 40, 45, 50 and 60.

[0039] Step N4: Merge adjacent artifact regions. That is, obtain the minimum distance between two adjacent connected domain regions, and compare the obtained minimum distance with a preset distance (the preset distance can be specifically set according to actual needs, so it is not specifically explained in this embodiment). When the minimum distance is less than the preset distance, the two adjacent connected domain regions corresponding to the minimum distance are merged. Otherwise, the two adjacent connected domain regions are not merged.

[0040] In this embodiment, the minimum distance between two adjacent connected domain areas is specifically: ; in: Connected domain and connected domain The minimum distance between is the set of all pixel positions of the i-th connected domain, is the set of all pixel positions of the jth connected domain, is the horizontal coordinate of the pixel in the i-th connected domain, is the horizontal coordinate of the pixel in the jth connected domain, is the vertical coordinate of the pixel in the i-th connected domain, is the vertical coordinate of the pixel in the jth connected domain, , The index of the sequence number.

[0041] In the specific implementation process, the pixel positions contained in the first connected domain are: (1,1), (1,2) and (2,1). The pixel positions contained in the second connected domain are: (4,4), (4,5) and (5,4). Then the distances between each pair of pixels are: 4.24, 5 and 5. In other words, the minimum distance between the first connected domain and the second connected domain is 4.24. Furthermore, in the process of merging connected domain regions, 4.24 is compared with the preset distance.

[0042] Step M2: Determine the type of artifacts. That is, according to the features (such as shape and texture) corresponding to the connected domain area determined in step N4, determine the type of each artifact area, including circular or nearly circular artifacts, stripe artifacts and other artifacts.

[0043] Step M3: Determine the artifact ranking. That is, determine the priority of the artifact area according to the area size of the artifact area and the corresponding artifact type. Further, according to the artifact type determined in step M2, the artifact area is sorted in the order of circular or nearly circular artifacts, stripe artifacts and other artifacts. Then, all artifact areas in each artifact type are sorted in descending order according to the artifact size.

[0044] Step S2.2: Perform artifact repair. That is, according to the artifact order determined in step M3, the artifact area is repaired, removed and marked. Specifically, the artifact area is repaired by an interpolation method based on neighboring pixel information to predict and fill in missing pixel values, and the artifact area is set to the background color by a binarization operation, or the artifact area is smoothed by an edge detection algorithm. Furthermore, by marking, manual review is directly performed or specific processing is performed according to actual conditions.

[0045] Step S2.3: Perform artifact detection. That is, the microscopic images processed in step S2.2 are detected to determine whether the artifact areas in each microscopic image are repaired. When it is detected that the artifact areas in the microscopic image are not repaired, steps M1 to S2.3 are repeated until all artifact areas are repaired. Otherwise, the next step is executed.

[0046] Step S3: Optimize the microscopic image. That is, optimize the microscopic image repaired in step S2.3 through multi-scale analysis, and reconstruct the three-dimensional vascular structure in the microscopic image through phase variance to obtain an optimized microscopic image. The details are as follows: Step S3.1: Obtain a fused microscopic image. That is, the microscopic image repaired in step S2.3 is smoothed by Gaussian smoothing, and then the smoothed microscopic image is downsampled by downsampling to obtain a reduced microscopic image. Furthermore, microscopic images at different scales are fused to obtain a fused microscopic image. It is worth noting that the Gaussian smoothing, downsampling and fusion processing in this embodiment are all existing conventional technical means, so they will not be explained in detail in this embodiment.

[0047] Step S3.2: Reconstruct the three-dimensional vascular structure. That is, continuous B-scan data is obtained through phase variance optical coherence tomography angiography. At the same time, according to each B-scan position, the phase change of adjacent time points is determined, and the blood flow signal is identified therefrom. Furthermore, all phase variance maps are combined to form a complete three-dimensional vascular structure.

[0048] This embodiment also provides a microscopic image enhancement and restoration system for tumor pathology. The microscopic image enhancement and restoration system uses a microscopic image enhancement and restoration method for tumor pathology described in the above embodiment.

[0049] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is limited by the attached embodiments and their equivalents.

Claims

1. A method for microscopic image enhancement and restoration for tumor pathology, characterized in that: Included are: S1: Image preprocessing: preprocessing the microscopic image to obtain a normalized value of the microscopic image, and comparing the normalized value with a preset normalized value, and determining a final microscopic image according to the comparison result; S2: Image artifact processing: using an artifact detection model, the final microscopic image is identified, an artifact region and an artifact type corresponding to the final microscopic image are determined, and the artifact region is processed according to the artifact type; S3: Optimizing the microscopic image: optimizing the microscopic image after the artifact area processing through multi-scale analysis, and reconstructing the three-dimensional vascular structure in the microscopic image through phase variance to obtain an optimized microscopic image; Processing the artifact area includes: S2.1: Determine artifact sorting: Identify artifact regions in the final microscopic image using an artifact detection model, determine the type and size corresponding to each artifact region, and sort all the artifact regions according to the type and size; S2.2: performing artifact repair: repairing, removing and marking the artifact areas according to the sorting of the artifact areas; S2.3: Perform artifact detection: Detect the microscopic image after artifact processing to determine whether all artifact areas in the microscopic image are repaired. When all artifact areas are repaired, execute the next step. Otherwise, repeat steps M1-S2.3 until all artifact areas are repaired.

2. A method for microscopic image enhancement and restoration for tumor pathology according to claim 1, characterized in that: Determine the final microscopic image, including: S1.1: Image processing: smoothing the microscopic image by using a Gaussian filter, processing the frequency of the microscopic image by using a histogram equalization, normalizing each pixel value of the microscopic image by using a linear normalization, and obtaining a normalized value corresponding to the microscopic image; S1.2: Setting a preset normalization value: Processing the microscopic image through the CNN model to obtain the preset normalization value, specifically: ； in: is the preset normalized value, is the original pixel value of the microscopic image at position (x, y), is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, is a convolutional neural network, are all trainable parameters of the convolutional neural network, is the index of the sequence number; S1.3: Determine the processed image: compare the normalized value obtained with the preset normalized value, obtain the difference between the normalized value and the preset normalized value, and compare the difference with the preset difference. When the difference is not greater than the preset difference, the microscopic image corresponding to the difference is the processed image. Otherwise, repeat steps S1.1 and S1.3 until the difference obtained is not greater than the preset difference.

3. A method for microscopic image enhancement and restoration for tumor pathology according to claim 2, characterized in that: The formula for obtaining all trainable parameters of the convolutional neural network is as follows: ； in: are the values ​​of all trainable parameters at the t+1th iteration, are the values ​​of all trainable parameters at the tth iteration, is the learning rate, is the loss function, is the gradient of the loss function with respect to all trainable parameters; The formula for obtaining the loss function is as follows: ； in: is the loss function, is the width of the microscopic image, is the height of the microscopic image, is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, is the preset normalized value, is the original pixel value of the microscopic image at position (x, y), are all trainable parameters of the convolutional neural network, is the standard pixel value of the i-th image at position (x, y), The index of the sequence number.

4. The method for microscopic image enhancement and restoration for tumor pathology according to claim 1, characterized in that: Sort all the artifact areas, including: M1: Determine the artifact area: determine the artifact probability corresponding to each pixel through the artifact detection model, and convert the artifact probability into a binary image. Process the artifact area through the binary image and the connected domain marker; M2: Determine the type of artifact: determine the type of each artifact area according to the characteristics of the artifact area, including circular or nearly circular artifacts, strip-shaped artifacts and other artifacts; M3: Determine the artifact ranking: Determine the priority of the artifact area according to the area and the artifact type of the artifact area.

5. A method for microscopic image enhancement and restoration for tumor pathology according to claim 4, characterized in that: Processing of artifact areas includes: M1.1: Determine the initial artifact area: Obtain the artifact probability corresponding to each pixel in the microscopic image through the artifact detection model, specifically: ； in: is the probability that the pixel at position (x, y) belongs to an artifact, is the U-Net network structure, is the original pixel value of the microscopic image at position (x, y), is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image, are all trainable parameters of the U-Net model, is the index of the sequence number; M1.2: Obtaining a binary image: Converting the artifact probability corresponding to each pixel in the microscopic image into a binary image, and determining the artifact area based on the binary image, specifically: ； in: is the output binary image, is the probability that the pixel at position (x, y) belongs to an artifact, is the preset threshold; M1.3: Determine the connected area: mark the connected domain of the artifact area, obtain the connected domain label map and the area of ​​each connected domain, and compare the connected domain area with the minimum area threshold to determine the final artifact area.

6. A method for microscopic image enhancement and restoration for tumor pathology according to claim 5, characterized in that: Determine the final artifact area, including: N1: Connected domain labeling: label each artifact region using a connected domain labeling algorithm to obtain a label map corresponding to the microscopic image; N2: Get the area of ​​the connected domain: Get the area of ​​the connected domain according to the number of pixels in each connected domain, specifically: ； in: is the area of ​​the i-th connected domain, is the set of all pixel positions of the i-th connected domain, is the index of the sequence number, is the horizontal coordinate of the pixel in the image, is the vertical coordinate of the pixel in the image; N3: Remove small areas: compare the area of ​​the connected domain with the minimum area threshold. When the area of ​​the connected domain is smaller than the minimum area threshold, the area of ​​the connected domain is deleted; otherwise, the area of ​​the connected domain is retained. N4: Merge adjacent artifact regions: Compare the minimum distance between two adjacent connected domain regions with the preset distance. When the minimum distance is less than the preset distance, merge the two adjacent connected domain regions corresponding to the minimum distance. Otherwise, do not merge the two adjacent connected domain regions. The formula for obtaining the minimum distance between two adjacent connected domain regions is as follows: ； in: Connected domain and connected domain The minimum distance between is the set of all pixel positions of the i-th connected domain, is the set of all pixel positions of the jth connected domain, is the horizontal coordinate of the pixel in the i-th connected domain, is the horizontal coordinate of the pixel in the jth connected domain, is the vertical coordinate of the pixel in the i-th connected domain, is the vertical coordinate of the pixel in the jth connected domain, , The index of the sequence number.

7. A method for microscopic image enhancement and restoration for tumor pathology according to claim 6, characterized in that: The method for determining the minimum area threshold is specifically as follows: According to the connected domain area, a connected domain area histogram is obtained, and two main peaks are determined from the connected domain area histogram, and the valley value between the two main peaks is the minimum area threshold.

8. The method for microscopic image enhancement and restoration for tumor pathology according to claim 1, characterized in that: Get optimized microscopy images, including: S3.1: Obtaining a fused microscopic image: smoothing the microscopic image after the artifact area processing by Gaussian smoothing, downsampling the smoothed microscopic image by downsampling to obtain a reduced microscopic image, and fusing microscopic images at different scales to obtain a fused microscopic image; S3.2: Reconstructing a three-dimensional vascular structure: Obtaining B-scan data of the fused microscopic image, and determining the phase change of adjacent time points according to each B-scan position, and constructing a three-dimensional vascular structure according to the phase change.

9. A microscopic image enhancement and restoration system for tumor pathology, characterized in that: A microscopic image enhancement and restoration method for tumor pathology as described in any one of claims 1 to 8 is used.

Images