Breast cancer slice data analysis method and system based on AI model

Abstract

The embodiment of the invention provides a breast cancer slice data analysis method and system based on an AI model, and relates to the technical field of slice data analysis. The method comprises the following steps: acquiring a breast cancer slice image; performing segmentation processing on the breast cancer slice image to obtain a first image; performing feature extraction processing on the first image through a first algorithm model, and performing first judgment on the first image according to the extracted image features to obtain a second image meeting a first condition; preprocessing the second image, and inputting the processed second image into a pre-trained classification model to obtain a first confidence coefficient; and determining a judgment result of the second image according to the first confidence coefficient. According to the invention, the problem of low data analysis precision is solved, and the effect of improving the data analysis precision is achieved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of slice data analysis, and more particularly, to a method and system for analyzing breast cancer slice data based on an AI model. Background Art

[0002] With the development of computer technology, some computer-aided diagnosis (CAD) systems have been introduced into breast cancer slice analysis. However, when existing CAD systems extract complex features (such as cell morphology, tissue structure, etc.) in breast cancer slice images, the effect is not ideal, and it is difficult to comprehensively capture feature information related to cancer. Especially when facing slice images from different sources and of different qualities, the generalization ability is poor, and misdiagnosis or missed diagnosis is likely to occur.

[0003] Therefore, the prior art cannot meet the requirements of precise analysis. Summary of the Invention

[0004] Embodiments of the present invention provide a method and system for analyzing breast cancer slice data based on an AI model to at least solve the problem of low data analysis accuracy in related technologies.

[0005] According to an embodiment of the present invention, there is provided a method for analyzing breast cancer slice data based on an AI model, including:

[0006] Obtaining a breast cancer slice image;

[0007] Performing segmentation processing on the breast cancer slice image to obtain a first image;

[0008] Performing feature extraction processing on the first image through a first algorithm model, and making a first judgment on the first image according to the extracted image features to obtain a second image meeting the first condition;

[0009] Performing preprocessing on the second image, and inputting the processed second image into a pre-trained classification model to obtain a first confidence level;

[0010] Determining the judgment result of the second image according to the first confidence level.

[0011] In an exemplary embodiment, the making a first judgment on the first image according to the extracted image features to obtain a second image meeting the first condition includes:

[0012] Performing region marking on the first image according to the image features;

[0013] Performing content judgment on the region marking through an ellipse fitting algorithm to determine a first region meeting the first condition;

[0014] Perform a first determination on the first region, and perform weighted segmentation on the first image according to the first determination result to obtain the second image.

[0015] In an exemplary embodiment, the feature extraction process of the first image by the first algorithm model includes:

[0016] Obtain the object features of the target object in the image. The image features include the object features, where the object features include at least one of the morphological features, texture features, arrangement features, and staining features of the target object. The texture features include at least one of roughness and uniformity. The arrangement features include at least one of the arrangement pattern and the tissue structure.

[0017] In an exemplary embodiment, after the breast cancer slice image is segmented to obtain the first image, the method further includes:

[0018] Collect the illumination components of the first image through a preset Gaussian filter to obtain illumination components of different scales;

[0019] Determine the reflection component according to the illumination component;

[0020] Perform image balance transformation processing on the first image based on the reflection component, and perform fusion processing on the result of the image balance transformation according to an adaptive weight, where the adaptive weight is dynamically calculated according to the target object of the first image and a preset image scale; the feature extraction process is based on the result of the fusion processing.

[0021] In an exemplary embodiment, after the illumination components of different scales are collected from the first image through a preset Gaussian filter, the method further includes:

[0022] Construct an illumination component matrix based on the illumination component and the scale information;

[0023] Perform correlation calculation on the illumination component matrix;

[0024] When the correlation calculation result does not meet the second condition, determine that the illumination component is abnormal.

[0025] According to another embodiment of the present invention, there is provided a breast cancer slice data analysis system for an AI model, including:

[0026] An image acquisition module for acquiring breast cancer slice images;

[0027] A segmentation module for segmenting the breast cancer slice image to obtain a first image;

[0028] A feature extraction module, configured to perform feature extraction processing on the first image through a first algorithm model, and perform a first judgment on the first image according to the extracted image features to obtain a second image that meets the first condition;

[0029] A confidence calculation module, configured to preprocess the second image and input the processed second image into a pre-trained classification model to obtain a first confidence;

[0030] A judgment module, configured to determine a judgment result of the second image according to the first confidence.

[0031] In an exemplary embodiment, the performing a first judgment on the first image according to the extracted image features to obtain a second image that meets the first condition includes:

[0032] Performing region marking on the first image according to the image features;

[0033] Performing content judgment on the region marking through an ellipse fitting algorithm to determine a first region that meets the first condition;

[0034] Performing a first judgment on the first region and performing weighted segmentation on the first image according to the first judgment result to obtain the second image.

[0035] In an exemplary embodiment, the performing feature extraction processing on the first image through the first algorithm model includes:

[0036] Obtaining object features of a target object in the image, where the image features include the object features, and the object features include at least one of morphological features, texture features, arrangement features, and staining features of the target object, the texture features include at least one of roughness and uniformity, and the arrangement features include at least one of an arrangement pattern and an organizational structure.

[0037] According to another embodiment of the present invention, there is also provided a computer-readable storage medium, in which a computer program is stored, and the computer program is configured to execute the steps in any one of the above method embodiments when running.

[0038] According to another embodiment of the present invention, there is also provided an electronic device, including a memory and a processor, a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0039] Through the present invention, by identifying images and calculating confidence levels, and by combining multi-scale feature fusion technology, it is possible to comprehensively capture multi-scale features such as cell morphology and tissue structure in breast cancer section images. Compared with traditional methods that only rely on single-scale features, the richness and representativeness of features are significantly improved, thereby improving the accuracy of diagnosis. Therefore, the problem of low data recognition accuracy can be solved, and the effect of improving recognition accuracy can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flowchart of a method for analyzing breast cancer section data based on an AI model according to an embodiment of the present invention;

[0041] Figure 2 is a structural block diagram of a system for analyzing breast cancer section data based on an AI model according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] Next, the technical solutions in the embodiments of the present application will be described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0043] Hereinafter, terms such as "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first", "second", etc. may explicitly or implicitly include one or more of such features. In the description of the present application, unless otherwise specified, the meaning of "a plurality" is two or more.

[0044] In addition, in the present application, orientation terms such as "upper", "lower", "left", and "right" may include but are not limited to being defined relative to the schematic placement of components in the drawings. It should be understood that these directional terms may be relative concepts, and they are used for relative description and clarification, and they may change accordingly with the change of the orientation of the components in the drawings.

[0045] In the present application, unless otherwise clearly defined and limited, the term "connection" should be understood in a broad sense. For example, "connection" may be a fixed connection, a detachable connection, or integrated; it may be directly connected or indirectly connected through an intermediate medium. In addition, the term "coupling" may be a way of electrical connection for signal transmission.

[0046] As used herein, "about", "substantially" or "approximately" includes the stated value and the average value within an acceptable deviation range of the specific value, where the acceptable deviation range is determined by those of ordinary skill in the art considering the measurement being discussed and the errors associated with the measurement of the specific quantity (i.e., the limitations of the measurement system).

[0047] In this embodiment, a method for analyzing breast cancer slice data based on an AI model is provided. Figure 1 It is a flowchart of a method for analyzing breast cancer slice data based on an AI model according to an embodiment of the present invention, as Figure 1 shown, the method includes the following steps:

[0048] Step S11, obtain breast cancer slice images.

[0049] In this embodiment, a digital pathology slice scanner is used to obtain high-resolution breast cancer tissue slice images, and at the same time, the collected images are stored in a server or a local database in a standardized format (such as TIFF or JPEG).

[0050] It should be noted that before collecting slice image data, the slices need to be stained to facilitate image recognition of the slices.

[0051] Step S12, perform segmentation processing on the breast cancer slice images to obtain a first image.

[0052] In this embodiment, cutting the image is to reduce the computing power requirements, improve the recognition efficiency of each sub-region, and at the same time reduce the interference of other image contents and improve the recognition accuracy.

[0053] Among them, perform preprocessing operations such as denoising and contrast enhancement on the image to improve the accuracy of subsequent segmentation; the segmentation processing can use an image segmentation algorithm (such as U-Net or V-Net based on deep learning) to identify and segment the tumor region, where the first image needs to contain the segmented tumor region; it should be noted that when performing segmentation, it is necessary to ensure that each region has a certain target content (such as a suspected tumor region, or muscle cells, etc.) according to a preset segmentation rule, so as to ensure that the relevant regions can be specifically identified subsequently.

[0054] Step S13, perform feature extraction processing on the first image through a first algorithm model, and make a first judgment on the first image according to the extracted image features to obtain a second image that meets the first condition;

[0055] In this embodiment, first identify the images of suspected lesion regions from multiple first images to facilitate subsequent precise recognition of the second image, improving the recognition efficiency and accuracy.

[0056] Among them, a pre-trained deep learning model (such as a convolutional neural network CNN) can be used to extract features from the first image. At this time, according to the extracted features, another deep learning model (such as a support vector machine SVM or a random forest RF) is used for preliminary judgment to identify regions that meet the first condition (such as obvious tumor features); among them, the feature extraction process of the first image by the first algorithm model includes obtaining the object features of the target object in the image, the image features include the object features, and the object features include at least one of the morphological features, texture features, arrangement features, and staining features of the target object. The texture features include at least one of roughness and uniformity, and the arrangement features include at least one of the arrangement pattern and the tissue structure; subsequently, the results are screened to obtain a second image that contains regions with highly suspected tumor features; it is easy to understand that the first condition can be that the image contains suspected abnormal regions, and the abnormal regions can be tumor feature regions, or regions that cannot be recognized by image recognition or are considered to have abnormal image quality.

[0057] Specifically, it can be to judge the regularity of cell arrangement in a certain region. If the regularity of cell arrangement in a certain region suddenly decreases, then further judge the cell - cell connection situation, such as changes in the distribution of desmosomes, etc.; among them, the judgment of regularity can be to calculate the central position of each cell through the geometric center formula, then use principal component analysis (PCA) to calculate the long - axis direction of each cell, then calculate the deviation between the long - axis direction of all cells and the average direction, and then use the standard deviation or variance to calculate the deviation between the long - axis direction of all cells and the average direction, and calculate the regularity of the cell central position according to the Fourier transform or autocorrelation function, and then judge the regularity according to the regularity and direction deviation.

[0058] For example, let θ i be the long - axis direction of the i - th cell, and θ avg be the average value of the long - axis directions of N cells in this region. Then the direction consistency C d can be expressed by formula 1:

[0059]

[0060] At the same time, the central position of the i - th cell is (x i , y i ), and the average position of N cells is (x avg , y avg ). At this time, the position regularity Cp can be expressed by formula 2:

[0061]

[0062] Subsequently, the comprehensive regularity C is expressed by formula 3 as:

[0063] C = α × C d + β × C P (Formula 3)

[0064] Wherein, α and β are the corresponding weights respectively.

[0065] It should be noted that the diversity of cell morphology may affect the calculation of direction consistency, and appropriate adjustment needs to be made according to cell types.

[0066] In particular, the desmosome distribution can be judged by constructing a distribution matrix and calculating the correlation of its distribution matrix. For example, the whole area is divided into a nine-square grid, and then the desmosome distance and density in each nine-square grid are filled into the matrix as matrix elements. Subsequently, the correlation value of the matrix is calculated. If the correlation value is within the preset range, it is judged as normal, otherwise it is judged as abnormal; of course, the distance between desmosomes can also be directly calculated, and judged according to the maximum and minimum values of the distance and the positions of the maximum and minimum desmosomes. For example, when the maximum and minimum values exceed the preset range, and the positions of the desmosomes corresponding to the maximum and minimum values also do not conform to the predetermined range, it is judged as abnormal, and so on.

[0067] It should be noted that to observe desmosomes, the cells need to be stained. After staining, the desmosome area will show a specific color, so that the distribution and quantity of desmosomes can be observed under an optical microscope.

[0068] Step S14, preprocess the second image, and input the processed second image into a pre-trained classification model to obtain a first confidence level;

[0069] In this embodiment, after obtaining the second image, the second image is further preprocessed such as bilateral filtering and normalization, and then the confidence level is calculated by the classification model for the processing result. At this time, if the execution degree is higher than the preset threshold, it is judged as the first state, otherwise it is determined as the second state. The first state can be a positive result or a negative result, which is not limited here.

[0070] Among them, the classification model can be a CNN neural network classification and recognition model, or other models.

[0071] Step S15, determine the judgment result of the second image according to the first confidence level.

[0072] Through the above steps, since the image is marked and the confidence level is calculated, and at the same time combined with the multi-scale feature fusion technology, it can comprehensively capture multi-scale features such as cell morphology and tissue structure in breast cancer section images. Compared with the traditional method that only relies on single-scale features, it significantly improves the richness and representativeness of features, thereby improving the accuracy of diagnosis, solving the problem of low data analysis accuracy, and improving the data analysis accuracy.

[0073] Among them, the execution entity of the above steps may be a base station, a terminal, etc., but is not limited thereto.

[0074] In an optional embodiment, the first determination of the first image according to the extracted image features to obtain a second image that meets the first condition includes:

[0075] Step S131, perform region marking on the first image according to the image features.

[0076] Step S132, perform content determination on the region marking through an ellipse fitting algorithm to determine a first region that meets the first condition.

[0077] Step S133, perform a first determination on the first region and perform weighted segmentation on the first image according to the first determination result to obtain the second image.

[0078] In this embodiment, after determining that a certain image region contains specific features, first identify the features of this region to facilitate subsequent further identification based on the identification; then use the ellipse fitting algorithm to perform identification based on the identification, and then segment the region that needs to be further identified.

[0079] Among them, the ellipse fitting algorithm is adopted because breast tumors often appear as ellipses in ultrasonic images; when performing segmentation, on the basis of a convolutional neural network, a weighting term can be introduced to process features such as edge intensity and texture of different regions; the ellipse fitting algorithm can be algorithms such as the least squares method and the least lower squares method.

[0080] Specifically, image processing algorithms such as edge detection, corner detection, and texture analysis, as well as deep learning models (such as convolutional neural networks), can be used to automatically extract key image features from the first image, and then map the extracted features to each pixel or region of the image to form a feature map; for example, the image can be divided into a high-density region and a low-density region according to the cell density feature, or the image can be divided into a large cell region and a small cell region according to the cell size feature; then, according to the feature map and a preset threshold or rule, perform region marking on the first image. For example, a region with a cell density higher than a certain threshold can be marked as a potential tumor region, or a region with disordered cell arrangement can be marked as an abnormal region. Region marking can be achieved through methods such as binarization, threshold segmentation, and region growing, or by classifying features based on a machine learning classifier, thereby achieving region marking.

[0081] It should be noted that in the existing breast cancer section analysis system, the analysis of staining characteristics mostly focuses on the cell nucleus or the overall staining distribution, and there is a lack of fine analysis of the specific staining of cell membranes (such as HER2 immunohistochemical staining, membrane protein markers such as E-cadherin, etc.); while this embodiment can solve this problem through the following steps based on the first image and the second image:

[0082] First, use the U-Net++ network to perform pixel-level segmentation on the cell membrane. The input is the section image stained with H&E or immunohistochemistry (IHC), and the output is a binary membrane region mask; among them, the pre-trained model is optimized based on a public dataset (such as the HER2Challenge dataset), supports generalization of multiple staining protocols, and introduces a dynamic threshold adaptive module to automatically adjust the segmentation sensitivity according to the staining intensity; then, for the differences in different staining batches, use a color deconvolution algorithm (such as the Macenko method) to separate the DAB (3,3'-diaminobenzidine) staining channel, eliminate background noise, and retain the membrane-specific staining signal.

[0083] Then, based on the existing morphological and texture features, add the following membrane-related features:

[0084] Membrane continuity index (MCI): By calculating the fracture ratio of the edge of the membrane region (specifically ), to quantify the membrane integrity;

[0085] Staining distribution uniformity: Use local binary pattern (LBP) to analyze the distribution of staining particles in the membrane region and extract the uniformity score;

[0086] Membrane-nucleus spatial relationship: Combine the nuclear segmentation results and calculate the statistical distribution (such as mean, variance) of the distance between the membrane and the nucleus, which is used to evaluate cell polarity (such as the loss of polarity in ductal carcinoma).

[0087] Subsequently, through a preset attention mechanism module, dynamically allocate different feature weights. For example, in HER2 analysis, the weight of membrane continuity (MCI) is higher than that of nuclear morphological features.

[0088] Finally, use ResNet-50 in the backbone network to extract global features, and the parallel branch uses a graph convolutional network (GCN) to determine and model the membrane-nucleus spatial relationship, and finally output the combined predicted recurrence risk score (such as the Oncotype DX equivalent value) and the status of membrane markers (such as the probability of HER2 positivity).

[0089] Among them, the traditional HER2 score (0 / 1+ / 2+ / 3+) depends on the subjective judgment of pathologists, especially when the 2+ interpretation requires FISH verification; to improve the accuracy of the HER2 score, in this embodiment, based on image analysis, the relevant images are further processed through the following steps:

[0090] 1. Image preprocessing

[0091] First, separate the DAB staining channel based on color deconvolution (such as the Macenko algorithm) to eliminate batch differences. At the same time, use a Gaussian filter to extract illumination components at different scales, dynamically adjust the reflection component, and eliminate the interference of slice reflection. Among them, when eliminating the reflection interference, it can also be eliminated according to the reflection component corresponding to the illumination component, which is not limited here.

[0092] 2. Cell membrane segmentation and feature extraction

[0093] First, use the aforementioned improved U-Net++ network. The input is the HER2 staining slice image, and the output is the binary mask of the membrane region. At the same time, introduce an edge continuity loss function to reduce membrane breakage and dynamically adjust the segmentation threshold to adapt to the staining intensity difference.

[0094] After that, input continuity (MCR), uniformity (LBP entropy), intensity (Iavg), and polarity (variance) into the multi-task regression network. At the same time, highlight the membrane regions (such as strong staining clusters) that affect the score through gradient-weighted class activation mapping (Grad-CAM). Finally, output a structured result containing sub-item scores such as MCR, LBP entropy, and Iavg, as well as a continuous score from 0 to 10.

[0095] In an optional embodiment, after performing segmentation processing on the breast cancer slice image to obtain a first image, the method further includes:

[0096] Step S15, collect illumination components of the first image through a preset Gaussian filter to obtain illumination components at different scales.

[0097] Step S16, determine the reflection component according to the illumination component.

[0098] Step S17, perform image balance transformation processing on the first image based on the reflection component, and perform fusion processing on the image balance transformation processing result according to an adaptive weight, where the adaptive weight is dynamically calculated according to the target object of the first image and a preset image scale; the feature extraction processing is based on the fusion processing result.

[0099] In this embodiment, the reflection in the image will affect the recognition result of the image. The illumination component reflects the brightness change in the image caused by uneven illumination. Therefore, it is necessary to filter the reflection situation of the image to better extract the reflection component and the target object features in the image and improve the recognition accuracy of the image.

[0100] Among them, for the Gaussian filter, appropriate Gaussian filter parameters need to be selected. The relevant parameters include the size of the filter (such as 3x3, 5x5, etc.) and the standard deviation (σ). Among them, a larger filter and a larger standard deviation can smooth a larger range of illumination changes and are suitable for collecting large-scale illumination components; smaller filters and smaller standard deviations are suitable for collecting small-scale illumination components. The reflection component is the part of the light reflected from the surface of the object in the image and can be obtained by subtracting the illumination component from the original image. This can eliminate the influence of uneven illumination and make the surface features of the object in the image clearer. The specific calculation formula is: reflection component = original image - illumination component. Performing image balance transformation on the reflection component is to adjust the brightness and contrast of the image to make the gray distribution of the image more uniform and the features of the target object more obvious. Commonly used image balance transformation methods include histogram equalization, contrast-limited adaptive histogram equalization (CLAHE), etc.

[0101] It should be noted that when performing dynamic calculation of weights, it can be calculated according to information such as the feature intensity, position, and size of the target object; and the preset image scale determines the distribution ratio of weights at different scales. The calculation of adaptive weights can adopt machine learning methods, such as neural networks, support vector machines, etc., to learn the relationship between weights, target objects, and image scales based on training data.

[0102] The fusion formula can be expressed as: fused image = adaptive weight × result of image balance transformation + (1 - adaptive weight) × original reflection component. Subsequently, when extracting features from the fused image, more accurate and robust features can be extracted. The feature extraction methods can include edge detection, texture analysis, shape feature extraction, etc. For example, using the Canny edge detection algorithm to extract image edge features; using the gray-level co-occurrence matrix (GLCM) to extract texture features of the image; extracting shape features by calculating the area, perimeter, shape factor, etc. of the target object; fusing the extracted multiple features to form a comprehensive feature vector. Feature selection algorithms, such as principal component analysis (PCA), feature selection trees, etc., can be used to optimize the feature vector, remove redundant features, and retain the most representative and discriminative features to provide high-quality feature input for subsequent image classification, object detection, and other tasks.

[0103] In an optional embodiment, after collecting illumination components of different scales from the first image through a preset Gaussian filter, the method further includes:

[0104] Step S151, constructing an illumination component matrix based on the illumination component and scale information;

[0105] Step S152, performing correlation calculation on the illumination component matrix;

[0106] Step S153, when the correlation calculation result does not meet the second condition, it is determined that the illumination component is abnormal.

[0107] In this embodiment, to further ensure the accuracy of image recognition, the filtering of specular reflection can be further refined.

[0108] Specifically, the illumination components of different scales and the corresponding scale information are integrated into an illumination component matrix. The rows of the matrix can represent different scales, and the columns can represent the pixel values or feature values of the illumination components. For example, the i-th row of the matrix can represent the illumination component at scale σi, and each column corresponds to a pixel position in the image; then a component matrix is constructed, where the elements of the correlation matrix represent the correlation intensity between the illumination components of different scales; for example, the (i, j)-th element of the correlation matrix can represent the correlation coefficient between scale σi and scale σj; the second condition can be the threshold range of the correlation coefficients in the correlation matrix, the sparsity requirement of the correlation matrix, the specific structural features of the correlation matrix, etc. For example, the second condition can be set such that the absolute values of all correlation coefficients in the correlation matrix should be greater than a certain threshold, or the correlation matrix should be sparse, that is, most elements are close to 0; or the correlation values of the matrix are within a certain range.

[0109] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases, the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc) and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0110] In this embodiment, a breast cancer slice data analysis system based on an AI model is also provided. This system is used to implement the above embodiments and preferred embodiments, and those that have been described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0111] Figure 2 is a structural block diagram of a breast cancer slice data analysis system based on an AI model according to an embodiment of the present invention, as Figure 2As shown in the figure, the system includes an image acquisition module 21, a segmentation module 22, a feature extraction module 23, a confidence calculation module 24, and a judgment module 25.

[0112] The image acquisition module 21 is used to obtain breast cancer section images. The segmentation module 22 performs segmentation processing on the breast cancer section images to obtain a first image. The feature extraction module 23 performs feature extraction processing on the first image through a first algorithm model, and makes a first judgment on the first image according to the extracted image features to obtain a second image that meets the first condition. The confidence calculation module 24 preprocesses the second image and inputs the processed second image into a pre-trained classification model to obtain a first confidence level. The judgment module 25 determines the judgment result of the second image according to the first confidence level.

[0113] In an optional embodiment, the making a first judgment on the first image according to the extracted image features to obtain a second image that meets the first condition includes: performing region marking on the first image according to the image features, making content judgment on the region marking through an ellipse fitting algorithm to determine a first region that meets the first condition, making a first judgment on the first region, and performing weighted segmentation on the first image according to the first judgment result to obtain the second image.

[0114] In an optional embodiment, the performing feature extraction processing on the first image through a first algorithm model includes: obtaining the object features of the target object in the image, where the image features include the object features, and the object features include at least one of the morphological features, texture features, arrangement features, and staining features of the target object. The texture features include at least one of roughness and uniformity. The arrangement features include at least one of the arrangement pattern and the tissue structure.

[0115] It should be noted that the above-mentioned various modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to this: all the above-mentioned modules are located in the same processor; or, the above-mentioned various modules are respectively located in different processors in any combination form.

[0116] An embodiment of the present invention also provides a computer-readable storage medium, in which a computer program is stored, and the computer program is set to execute the steps in any one of the above method embodiments when running.

[0117] In an exemplary embodiment, the above computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), mobile hard disks, magnetic disks, or optical discs.

[0118] An embodiment of the present invention also provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0119] In an exemplary embodiment, the above electronic device may further include a transmission device and an input / output device. Among them, the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0120] Through the description of the above embodiments, those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above functional modules is used as an example. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0121] In several embodiments provided in the present application, it should be understood that the disclosed device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical or other form.

[0122] The units described as separate components may or may not be physically separated. The components displayed as units may be one physical unit or multiple physical units, that is, they can be located in one place, or they can be distributed to multiple different places. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0123] In addition, each functional unit in various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0124] When an integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiments of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions for causing a device (which can be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0125] The above content is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for analyzing breast cancer slice data based on an AI model, characterized in that, Including: Obtain breast cancer slice images; Perform segmentation processing on the breast cancer slice images to obtain a first image; Perform feature extraction processing on the first image through a first algorithm model, and perform a first judgment on the first image according to the extracted image features to obtain a second image that meets the first condition; Preprocess the second image and input the processed second image into a pre-trained classification model to obtain a first confidence level; Determine the judgment result of the second image according to the first confidence level.

2. The method according to claim 1, wherein The performing a first judgment on the first image according to the extracted image features to obtain a second image that meets the first condition includes: Perform region marking on the first image according to the image features; Perform content judgment on the region marking through an ellipse fitting algorithm to determine a first region that meets the first condition; Perform a first judgment on the first region and perform weighted segmentation on the first image according to the first judgment result to obtain the second image.

3. The method according to claim 1, characterized in that, The performing feature extraction processing on the first image through the first algorithm model includes: Obtain the object features of the target object in the image, the image features include the object features, wherein the object features include at least one of the morphological features, texture features, arrangement features, and staining features of the target object, the texture features include at least one of roughness and uniformity, and the arrangement features include at least one of the arrangement pattern and the tissue structure.

4. The method according to claim 1, characterized in that, After performing segmentation processing on the breast cancer slice images to obtain a first image, the method further includes: Collect illumination components of the first image through a preset Gaussian filter to obtain illumination components of different scales; Determine the reflection component according to the illumination component; Perform image balance transformation processing on the first image based on the reflection component, and perform fusion processing on the image balance transformation processing result according to an adaptive weight, wherein the adaptive weight is dynamically calculated according to the target object of the first image and a preset image scale; the feature extraction processing is based on the fusion processing result.

5. The method according to claim 4, wherein After collecting illumination components of the first image through a preset Gaussian filter to obtain illumination components of different scales, the method further includes: Construct an illumination component matrix based on the illumination component and scale information; Perform correlation calculation on the illumination component matrix; When the correlation calculation result does not meet the second condition, determine that the illumination component is abnormal.

6. A breast cancer slice data analysis system based on an AI model, characterized in that, Including: An image acquisition module for obtaining breast cancer slice images; A segmentation module for performing segmentation processing on the breast cancer slice images to obtain a first image; A feature extraction module for performing feature extraction processing on the first image through a first algorithm model, and performing a first judgment on the first image according to the extracted image features to obtain a second image that meets the first condition; A confidence level calculation module for preprocessing the second image and inputting the processed second image into a pre-trained classification model to obtain a first confidence level; A judgment module, configured to determine a judgment result of the second image according to the first confidence level.

7. The system according to claim 6, characterized in that, The first judgment of the first image according to the extracted image features to obtain a second image meeting the first condition includes: Performing region marking on the first image according to the image features; Performing content judgment on the region marking through an ellipse fitting algorithm to determine a first region meeting the first condition; Performing a first judgment on the first region, and performing weighted segmentation on the first image according to the first judgment result to obtain the second image.

8. The system according to claim 6, characterized in that The feature extraction process of the first image through the first algorithm model includes: Obtaining object features of a target object in the image, where the image features include the object features, and the object features include at least one of morphological features, texture features, arrangement features, and staining features of the target object, the texture features include at least one of roughness and uniformity, and the arrangement features include at least one of an arrangement pattern and an organizational structure.

9. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, where the computer program is configured to execute the method described in any one of claims 1 to 5 when running.

10. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is configured to run the computer program to execute the method described in any one of claims 1 to 5.

Images