Brain glioma typing prediction method and apparatus, computer device, and storage medium

Abstract

The present application relates to a brain glioma typing prediction method and apparatus, a computer device, and a storage medium. The method comprises: collecting a whole slide image of a tumor patient, and performing patch cutting on the whole slide image to generate patch images; extracting deep feature vectors of the patch images, clustering the deep feature vectors, and grouping the patch images to obtain a set number of patch groups; using all the patch images as a reference data set for training to obtain a reference prediction model, and using patch images in each patch group as a non-reference data set for training to obtain a non-reference prediction model; and screening for a patch group corresponding to the non-reference prediction model of which the accuracy is higher than reference precision, and using the selected patch group for training to obtain a brain glioma typing prediction model. According to embodiments of the present application, the model prediction precision can be ensured, and the model generalization performance can be improved.

Description

Brain glioma classification prediction method, device, computer equipment and storage medium Technical Field

[0001] The present application belongs to the field of medical image processing technology, and in particular relates to a method, apparatus, computer equipment, and storage medium for predicting brain glioma classification. Background Art

[0002] Glioma is the most common malignant brain tumor of the central nervous system, accounting for about 81% of malignant tumors of the central nervous system. It has the "three highs and one low" characteristics of high incidence, high recurrence rate, high disability rate and low survival rate. Glioma cells will produce different subclonal populations during tumor evolution. Each population has significant differences in DNA mutations, mRNA expression, protein translation and post-translational modifications, which leads to differences in tumor cell phenotype, immune microenvironment and response to radiotherapy and chemotherapy. This heterogeneity makes the diagnosis and prognosis assessment of glioma very difficult. In August 2021, the World Health Organization (WHO) released the latest 5th edition of the "Classification of Tumors of the Central Nervous System", which reclassified gliomas into astrocytomas (A), oligodendrogliomas (O) and glioblastomas (GBM). The new classification method integrates important information from multiple levels of magnetic resonance imaging, tissue pathology and molecular markers, and poses new challenges for the precise classification and treatment of gliomas. However, at current clinical levels, the integrated diagnostic methods for gliomas, as defined by the WHO, remain difficult to implement in clinical practice. This is primarily due to two difficulties: First, the limitations of sequencing samples. A single sample can only obtain a small amount of tumor tissue from a specific location within the tumor, failing to characterize the entire tumor. Second, it lacks practical application in routine clinical practice. Biological sequencing platforms, particularly protein sequencing platforms, are extremely scarce, and the cost of testing is prohibitive, making them unsuitable for routine clinical application in gliomas.

[0003] The current main approach for building diagnostic aid models using whole slide images (WSIs) is to use multiple-instance learning (MIL), which treats patches cut from a digital pathology image as different examples. The predictive power of each example is ranked, and the k examples with the highest predictive power are selected to represent the entire WSI for prediction model training. However, because k is typically not very large in the MIL algorithm, meaning only a small number of patches are used to represent the entire WSI, much important cellular information is not incorporated into the final prediction model, limiting the model's accuracy.

[0004] Summary of the Invention

[0005] The present application provides a method, apparatus, computer device and storage medium for predicting brain glioma classification, aiming to solve at least one of the above-mentioned technical problems in the prior art to a certain extent.

[0006] In order to solve the above problems, this application provides the following technical solutions:

[0007] A method for predicting brain glioma classification, comprising:

[0008] Collecting digital pathology images of tumor patients and performing patch cutting on the digital pathology images to generate patch images;

[0009] Extracting deep feature vectors of the patch images using a deep neural network, clustering the deep feature vectors using a K-means clustering algorithm, and grouping the patch images according to the clustering results of the deep feature vectors to obtain a set number of patch groups;

[0010] All patch images are used as a benchmark dataset for training to obtain a benchmark prediction model, the accuracy of the benchmark prediction model is used as the benchmark accuracy, and the patch images in each patch group are used as a non-benchmark dataset for training to obtain a non-benchmark prediction model;

[0011] The accuracy of each non-benchmark prediction model is compared with the benchmark accuracy, and the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy is screened out. The screened patch group is used to train a brain glioma classification prediction model.

[0012] The technical solution adopted in the embodiment of the present application further includes: after collecting the digital pathological images of the tumor patient, the following steps are also included:

[0013] Preprocessing the digital pathology images to obtain a first a-component image of each digital pathology image; wherein the preprocessing method includes: downsampling all digital pathology images according to a set ratio of the initial size, so that the size of the downsampled digital pathology images image is [M, N]; converting the downsampled digital pathology images image from an RGB color space to a Lab color space, and extracting the first a-component image image_a based on the converted digital pathology images image.

[0014] The technical solution adopted in the embodiment of the present application further includes: before performing patch cutting on the digital pathology image, it also includes:

[0015] A global segmentation threshold algorithm is used to calculate the segmentation threshold k of the first a-component image.

[0016] The technical solution adopted in the embodiment of the present application further includes: after performing patch cutting on the digital pathology image, further including:

[0017] Convert the patch image from RGB color space to Lab color space, and extract the second a component image of the patch image;

[0018] Calculate the pixel average of the second a component image of each patch image, and determine whether the pixel average of the second a component image of each patch image is greater than the segmentation threshold k. If it is greater than the segmentation threshold k, retain the patch image; otherwise, eliminate the patch image, and finally obtain n patch images containing tumor tissue.

[0019] The technical solution adopted in the embodiment of the present application also includes: extracting the deep feature vector of the patch image using a deep neural network, clustering the deep feature vector using a K-means clustering algorithm, and grouping the patch image according to the clustering result of the deep feature vector. Specifically,

[0020] The extracted depth feature vectors are clustered using the K-means clustering algorithm to obtain a clustering model. Assuming that the number of clusters of the K-means clustering algorithm is m, the n patch images are clustered into m patch groups according to the clustering results of the depth feature vectors, and the m patch groups are numbered in sequence as the first patch group, the second patch group, ..., the mth patch group.

[0021] The technical solution adopted in the embodiment of the present application also includes: training all patch images as a benchmark data set to obtain a benchmark prediction model, using the accuracy of the benchmark prediction model as the benchmark accuracy, and training the patch images in each patch group as a non-benchmark data set to obtain a non-benchmark prediction model. Specifically,

[0022] Merge all patch images to generate a benchmark dataset, which is named dataset0;

[0023] Merge the patch images in the m patch groups to generate m non-benchmark datasets, and name the m non-benchmark datasets as dataset1, dataset2, ..., datasetm respectively;

[0024] According to dataset0, a deep neural network is used to train a benchmark prediction model model0, and the acc_0 of the benchmark prediction model model0 is used as the benchmark accuracy;

[0025] Based on dataset1, dataset2, ..., datasetm, m non-benchmark prediction models model1, model2, ..., modelm are trained using a deep neural network with the same structure and training parameters.

[0026] The technical solution adopted in the embodiment of the present application also includes: comparing the accuracy of each non-benchmark prediction model with the benchmark accuracy, screening out the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy, and using the screened patch group to train a brain glioma classification prediction model. Specifically,

[0027] Obtain the accuracy of each non-baseline prediction model respectively, compare the accuracy of each non-baseline prediction model with the baseline accuracy of the baseline prediction model, and determine whether the accuracy of each non-baseline prediction model is lower than the baseline accuracy. If it is lower than the baseline accuracy, all patch groups corresponding to the non-baseline prediction model are eliminated; if it is higher than the baseline accuracy, all patch groups corresponding to the non-baseline prediction model are retained;

[0028] Merge all retained patch groups to generate the final dataset dataset_EN;

[0029] The dataset dataset_EN was used to train a deep neural network to obtain the final brain glioma classification prediction model.

[0030] Another technical solution adopted in the embodiment of the present application is: a device for predicting glioma classification, comprising:

[0031] Image processing module: used to collect digital pathology images of tumor patients and perform patch cutting on the digital pathology images to generate patch images;

[0032] Feature extraction and clustering module: uses a deep neural network to extract the deep feature vectors of the patch image, and uses the K-means clustering algorithm to cluster the deep feature vectors. The patch images are grouped according to the clustering results of the deep feature vectors to obtain a set number of patch groups;

[0033] The first model training module is used to train all patch images as a benchmark dataset to obtain a benchmark prediction model, use the accuracy of the benchmark prediction model as the benchmark accuracy, and train the patch images in each patch group as a non-benchmark dataset to obtain a non-benchmark prediction model;

[0034] The second model training module is used to compare the accuracy of each non-benchmark prediction model with the benchmark accuracy, screen out the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy, and use the screened patch group to train to obtain a brain glioma classification prediction model.

[0035] Another technical solution adopted by the embodiment of the present application is: a computer device, the computer device comprising a processor and a memory coupled to the processor, wherein:

[0036] The memory stores program instructions for implementing the method for predicting brain glioma classification;

[0037] The processor is used to execute the program instructions stored in the memory to control the brain glioma classification prediction method.

[0038] Another technical solution adopted in the embodiment of the present application is: a storage medium storing program instructions executable by a processor, wherein the program instructions are used to execute the brain glioma classification prediction method.

[0039] Compared with the prior art, the beneficial effects of the embodiments of the present application are as follows: the brain glioma classification prediction method, device, computer equipment and storage medium of the embodiments of the present application use deep feature vectors for clustering, grouping all patch images, and judging the predictive power of each group of patch images based on the prediction model results of the patch images, thereby achieving the screening of patch images, thereby obtaining a more comprehensive patch image to replace the entire digital pathology image, and using the screened patch images to construct the final brain glioma classification prediction model, which not only ensures the prediction accuracy of the model, but also improves the generalization of the model. The embodiments of the present application avoid expensive gene sequencing, simplify the brain glioma classification prediction process, and are more in line with existing medical conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] FIG1 is a flow chart of a method for predicting glioma classification according to a first embodiment of the present application;

[0041] FIG2 is a flow chart of a method for predicting glioma classification according to a second embodiment of the present application;

[0042] FIG3 is a schematic diagram of the structure of a device for predicting glioma classification according to an embodiment of the present application;

[0043] FIG4 is a schematic diagram of the structure of a computer device according to an embodiment of the present application;

[0044] FIG5 is a schematic diagram of the structure of a storage medium according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0046] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of such features. In the description of this application, "multiple" means at least two, for example, two, three, etc., unless otherwise specifically defined. All directional indications in the embodiments of this application (such as up, down, left, right, front, back...) are only used to explain the relative positional relationship, movement, etc. between the components under a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications also change accordingly. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or computer device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products, or computer devices.

[0047] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0048] Specifically, please refer to FIG1 , which is a flow chart of the method for predicting glioma classification according to the first embodiment of the present application. The method for predicting glioma classification according to the first embodiment of the present application comprises the following steps:

[0049] S100: Collecting digital pathological images of tumor patients, and performing patch cutting on the digital pathological images to generate patch images;

[0050] S110: extracting a deep feature vector of the patch image using a deep neural network, clustering the deep feature vector using a K-means clustering algorithm, and grouping the patch image according to the clustering results of the deep feature vector to obtain a set number of patch groups;

[0051] S120: training all patch images as a benchmark dataset to obtain a benchmark prediction model, taking the accuracy of the benchmark prediction model as the benchmark accuracy, and training the patch images in each patch group as a non-benchmark dataset to obtain a non-benchmark prediction model;

[0052] S130: Compare the accuracy of each non-benchmark prediction model with the benchmark accuracy, screen out the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy, and use the screened patch group to train to obtain a brain glioma classification prediction model.

[0053] Please refer to FIG2 , which is a flow chart of a method for predicting glioma classification according to a second embodiment of the present application. The method for predicting glioma classification according to the second embodiment of the present application comprises the following steps:

[0054] S200: collecting a certain number of digital pathology images of tumor patients, and preprocessing the digital pathology images to obtain a first a component image of each digital pathology image;

[0055] In this step, the digital pathology images come from the patient digital pathology image dataset jointly included in the hospital and TCGA (The Cancer Genome Atlas). The embodiment of the present application does not limit the number of digital pathology images collected. The more images there are, the stronger the generalization ability. Since the formats of digital pathology images scanned by different scanners are not uniform and there are a lot of pixels, the collected digital pathology images need to be preprocessed. Specifically, the preprocessing method of the digital pathology images includes: downsampling all digital pathology images according to a set ratio of the initial size, and the size of the digital pathology image after downsampling is [M, N]; and converting the downsampled digital pathology image image from RGB color space to Lab color space, and extracting the first a component image image_a based on the converted digital pathology image image. Among them, the set ratio of the downsampling is 1 / 10, which can be set according to the actual application scenario.

[0056] S210: Calculating a segmentation threshold k of the first a-component image using a global segmentation threshold algorithm;

[0057] In this step, the global segmentation threshold algorithm is specifically as follows:

[0058] S211: Set a segmentation threshold i, where the initial value of i is set to 0;

[0059] S212: Normalize the first a-component image image_a to obtain a pixel distribution histogram. Based on the pixel distribution histogram, count the proportion w0 of pixels with pixel values ​​between 0 and i in the total M×N pixel count of the first a-component image, count the average grayscale u0 of the foreground pixels, count the proportion w1 of pixels with pixel values ​​between i and 255 in the total M×N pixel count of the first a-component image, and count the average grayscale u1 of the background pixels.

[0060] S213: Calculate the variance of the foreground pixel and the background pixel: g = w0×w1×(u0-u1)×(u0-u1);

[0061] S214: i++, then re-execute S212 until i reaches 256 and the iteration ends;

[0062] S215: Use the i value corresponding to the maximum g as the segmentation threshold K of the first a-component image.

[0063] S220: performing patch cutting on the digital pathology image according to a set pixel size to generate a plurality of patch images, converting the patch images from the RGB color space to the Lab color space, and extracting the second a component image of the patch images;

[0064] In this step, the pixel size is set to 1024×1024, which can be set according to the actual application scenario. Specifically, the patch cutting method for the digital pathology image dataset is as follows: a sliding window with a pixel size of 1024×1024 is designed. The sliding window starts from the upper left corner of each digital pathology image and slides from left to right and from top to bottom in sequence with a sliding step of 1024. After each sliding step, the pixel information covered by the sliding window is extracted as a patch image and saved. Finally, all patch images are converted from RGB color space to Lab color space, and the second a component image of the patch image is extracted.

[0065] S230: Calculate the pixel average of the second a component image of each patch image, filter all patch images according to the pixel average and the segmentation threshold k, remove background patch images, and retain n patch images containing tumor tissue;

[0066] In this step, the patch images are screened by determining whether the pixel average of the second a component image of each patch image is greater than the segmentation threshold k. If it is greater than the segmentation threshold k, it indicates that the patch image contains tumor tissue and is retained. Conversely, if it is less than the segmentation threshold k, it indicates that the patch image is the slide background and is discarded.

[0067] S240: Inputting n patch images into a pre-trained deep neural network in sequence to extract deep feature vectors, clustering the extracted deep feature vectors using a K-means clustering algorithm, and grouping all patch images according to the deep feature vector clustering results to obtain a set number of patch groups;

[0068] In this step, the number of depth feature vectors extracted from each patch image is 2048, that is, n 2048×1 depth feature vectors are extracted from each tumor patient, which can be set according to the actual application scenario. The K-means clustering algorithm is specifically as follows: first, the n 2048×1 depth feature vectors are clustered to obtain a clustering model. Assuming that the number of clusters of the K-means clustering algorithm is set to m, the n patch images are clustered into m patch groups according to the depth feature vector clustering results, and the m patch groups are numbered in sequence as the 1st patch group, the 2nd patch group, ..., the mth patch group. It can be understood that the number of clusters in the embodiment of the present application can be set according to the actual application scenario.

[0069] S250: Merge all patch images into a benchmark dataset, and merge the patch images in each patch group into a non-benchmark dataset, and train a prediction model using the benchmark dataset and each non-benchmark dataset, respectively. The prediction model trained using the benchmark dataset is used as the benchmark prediction model, and the prediction model trained using the non-benchmark dataset is used as the non-benchmark prediction model.

[0070] In this step, the training method of the prediction model specifically includes:

[0071] S251: merge all patch images into a benchmark dataset named dataset0;

[0072] S252: Merge the patch images in the m patch groups respectively to generate m non-benchmark datasets, and name the m non-benchmark datasets as dataset1, dataset2, ..., datasetm respectively;

[0073] S253: Based on dataset0, a deep neural network is used to train a benchmark prediction model model0, and its acc_0 (Accuracy, indicating the accuracy of the model) is used as the benchmark accuracy;

[0074] S254: Based on dataset1, dataset2, ..., datasetm, m non-benchmark prediction models model1, model2, ..., modelm are trained using a deep neural network with the same structure and training parameters.

[0075] S260: Obtain the accuracy of each non-baseline prediction model, compare the accuracy of each non-baseline prediction model with the benchmark accuracy of the benchmark prediction model, and determine whether the accuracy of each non-baseline prediction model is lower than the benchmark accuracy. If so, execute S270; otherwise, execute S280.

[0076] S270: Eliminate all patch groups corresponding to the non-baseline prediction model;

[0077] S280: retain all the patch groups corresponding to the non-baseline prediction model, and merge all the retained patch groups to generate the final dataset dataset_EN;

[0078] S290: Use the dataset dataset_EN to train a deep neural network to obtain the final brain glioma classification prediction model;

[0079] In this step, the network structure and training parameters of the deep neural network are consistent with those of the deep neural network in S250, and will not be repeated here.

[0080] Based on the above, the brain glioma classification prediction method of the second embodiment of the present application uses deep feature vectors for clustering, groups all patch images, and judges the predictive power of each group of patch images based on the prediction model results of the patch images, thereby achieving the screening of patch images, thereby obtaining a more comprehensive patch image to replace the entire digital pathology image, and using the screened patch images to construct the final brain glioma classification prediction model, which not only ensures the prediction accuracy of the model, but also improves the generalization of the model. The embodiment of the present application avoids expensive gene sequencing, simplifies the brain glioma classification prediction process, and is more in line with existing medical conditions.

[0081] Please refer to FIG3 , which is a schematic diagram of the structure of the brain glioma classification prediction device according to an embodiment of the present application. The brain glioma classification prediction device 40 according to the embodiment of the present application comprises:

[0082] Image processing module 41: used for collecting digital pathological images of tumor patients and performing patch cutting on the digital pathological images to generate patch images;

[0083] Feature extraction and clustering module 42: extracts deep feature vectors of the patch images using a deep neural network, clusters the deep feature vectors using a K-means clustering algorithm, and groups the patch images according to the clustering results of the deep feature vectors to obtain a set number of patch groups;

[0084] The first model training module 43 is configured to train all patch images as a benchmark dataset to obtain a benchmark prediction model, use the accuracy of the benchmark prediction model as the benchmark accuracy, and train the patch images in each patch group as a non-benchmark dataset to obtain a non-benchmark prediction model;

[0085] The second model training module 44 is used to compare the accuracy of each non-benchmark prediction model with the benchmark accuracy, screen out the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy, and use the screened patch group to train to obtain a brain glioma classification prediction model.

[0086] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0087] The device provided in the embodiment of the present application can be applied in the aforementioned method embodiment. For details, please refer to the description of the aforementioned method embodiment, which will not be repeated here.

[0088] Please refer to FIG4 , which is a schematic diagram of the computer device structure of an embodiment of the present application. The computer device 50 includes:

[0089] A memory 51 storing executable program instructions;

[0090] a processor 52 connected to the memory 51;

[0091] The processor 52 is used to call the executable program instructions stored in the memory 51 and execute the following steps: collecting digital pathology images of tumor patients, and patch cutting the digital pathology images to generate patch images; using a deep neural network to extract deep feature vectors of the patch images, and using a K-means clustering algorithm to cluster the deep feature vectors, and grouping the patch images according to the clustering results of the deep feature vectors to obtain a set number of patch groups; using all patch images as a benchmark data set to train a benchmark prediction model, using the accuracy of the benchmark prediction model as the benchmark accuracy, and respectively using the patch images in each patch group as a non-benchmark data set to train a non-benchmark prediction model; comparing the accuracy of each non-benchmark prediction model with the benchmark accuracy, screening out patch groups corresponding to non-benchmark prediction models with an accuracy higher than the benchmark accuracy, and using the screened patch groups to train a brain glioma classification prediction model.

[0092] The processor 52 may also be referred to as a CPU (Central Processing Unit). The processor 52 may be an integrated circuit chip having signal processing capabilities. The processor 52 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The general-purpose processor may be a microprocessor or any conventional processor.

[0093] Please refer to Figure 5, which is a schematic diagram of the structure of the storage medium of an embodiment of the present application. The storage medium of the embodiment of the present application stores program instructions 61 that can implement the following steps: collecting digital pathology images of tumor patients, and performing patch cutting on the digital pathology images to generate patch images; using a deep neural network to extract the depth feature vectors of the patch images, and using the K-means clustering algorithm to cluster the depth feature vectors, grouping the patch images according to the clustering results of the depth feature vectors to obtain a set number of patch groups; training all patch images as a benchmark data set to obtain a benchmark prediction model, using the accuracy of the benchmark prediction model as the benchmark accuracy, and respectively training the patch images in each patch group as a non-benchmark data set to obtain a non-benchmark prediction model; comparing the accuracy of each non-benchmark prediction model with the benchmark accuracy, screening out the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy, and using the screened patch group to train a brain glioma classification prediction model. Among them, the program instructions 61 can be stored in the above-mentioned storage medium in the form of a software product, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network computer device, etc.) or a processor to execute all or part of the steps of the various implementation methods of the present application. The aforementioned storage medium includes: various media that can store program instructions, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or a terminal computer device such as a computer, a server, a mobile phone, or a tablet. Among them, the server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0094] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0095] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. The above is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of this application, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for predicting brain glioma typing, characterized in that: include: Collecting digital pathological images of tumor patients, and performing patch cutting on the digital pathological images to generate patch images; Extracting a deep feature vector of the patch image using a deep neural network, clustering the deep feature vector using a K-means clustering algorithm, and grouping the patch image according to the clustering result of the deep feature vector to obtain a set number of patch groups; All patch images are used as a benchmark data set to train a benchmark prediction model, the accuracy of the benchmark prediction model is used as the benchmark accuracy, and the patch images in each patch group are used as a non-benchmark data set to train a non-benchmark prediction model; The accuracy of each non-benchmark prediction model is compared with the benchmark accuracy, and the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy is screened out, and the screened patch group is used to train a brain glioma classification prediction model.

2. The method for predicting brain glioma classification according to claim 1, characterized in that: After collecting the digital pathological images of tumor patients, the method further includes: The digital pathology images are preprocessed to obtain a first a component image of each digital pathology image; wherein the preprocessing method includes: downsampling all digital pathology images according to a set ratio of the initial size, and the size of the digital pathology image after downsampling is [M, N]; converting the downsampled digital pathology image image from an RGB color space to a Lab color space, and extracting a first a component image image_a according to the converted digital pathology image image.

3. The method for predicting brain glioma typing according to claim 2, characterized in that: Before performing patch cutting on the digital pathology image, the method further includes: A global segmentation threshold algorithm is used to calculate the segmentation threshold k of the first a-component image.

4. The method for predicting brain glioma typing according to claim 3, characterized in that: After performing patch cutting on the digital pathology image, the method further includes: Convert the patch image from RGB color space to Lab color space, and extract the second a component image of the patch image; The pixel average value of the second a component image of each patch image is calculated, and it is determined whether the pixel average value of the second a component image of each patch image is greater than the segmentation threshold k. If it is greater than the segmentation threshold k, the patch image is retained; otherwise, the patch image is eliminated, and finally n patch images containing tumor tissue are obtained.

5. The method for predicting brain glioma typing according to claim 4, characterized in that: The method of extracting the deep feature vector of the patch image by using a deep neural network, clustering the deep feature vector by using a K-means clustering algorithm, and grouping the patch image according to the clustering result of the deep feature vector is specifically as follows: The extracted deep feature vector is clustered using the K-means clustering algorithm to obtain a clustering model. Assuming that the number of clustering clusters of the K-means clustering algorithm is m, the n patch images are clustered into m patch groups according to the clustering results of the deep feature vectors, and the m patch groups are numbered sequentially as the first patch group, the second patch group, ..., the mth patch group.

6. The method for predicting brain glioma typing according to claim 5, characterized in that: The method of training all patch images as a benchmark data set to obtain a benchmark prediction model, taking the accuracy of the benchmark prediction model as the benchmark accuracy, and training the patch images in each patch group as a non-benchmark data set to obtain a non-benchmark prediction model is as follows: Merge all patch images to generate a benchmark dataset, which is named dataset0; Merge the patch images in the m patch groups respectively to generate m non-benchmark datasets, and name the m non-benchmark datasets as dataset1, dataset2, ..., datasetm respectively; According to dataset0, a benchmark prediction model model0 is obtained by training a deep neural network, and acc_0 of the benchmark prediction model model0 is used as the benchmark accuracy; According to dataset1, dataset2, ..., datasetm, m non-benchmark prediction models model1, model2, ..., modelm are trained using a deep neural network with the same structure and training parameters.

7. The method for predicting brain glioma typing according to claim 6, characterized in that: The accuracy of each non-benchmark prediction model is compared with the benchmark accuracy, and the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy is screened out, and the brain glioma classification prediction model is obtained by training the screened patch group. Specifically, Obtain the accuracy of each non-benchmark prediction model respectively, compare the accuracy of each non-benchmark prediction model with the benchmark accuracy of the benchmark prediction model, and determine whether the accuracy of each non-benchmark prediction model is lower than the benchmark accuracy. If it is lower than the benchmark accuracy, all the patch groups corresponding to the non-benchmark prediction model are eliminated; if it is higher than the benchmark accuracy, all the patch groups corresponding to the non-benchmark prediction model are retained; Merge all the retained patch groups to generate the final dataset dataset_EN; The dataset dataset_EN is used to train a deep neural network to obtain the final brain glioma classification prediction model.

8. A device for predicting brain glioma classification, characterized in that: include: Image processing module: used to collect digital pathological images of tumor patients, and perform patch cutting on the digital pathological images to generate patch images; Feature extraction and clustering module: extract the deep feature vector of the patch image using a deep neural network, cluster the deep feature vector using a K-means clustering algorithm, group the patch image according to the clustering result of the deep feature vector, and obtain a set number of patch groups; The first model training module is used to train all patch images as a benchmark data set to obtain a benchmark prediction model, use the accuracy of the benchmark prediction model as the benchmark accuracy, and train the patch images in each patch group as a non-benchmark data set to obtain a non-benchmark prediction model; The second model training module is used to compare the accuracy of each non-benchmark prediction model with the benchmark accuracy, screen out the patch group corresponding to the non-benchmark prediction model with an accuracy higher than the benchmark accuracy, and use the screened patch group to train to obtain a brain glioma classification prediction model.

9. A computer device, characterized in that: The computer device includes a processor and a memory coupled to the processor, wherein: The memory stores program instructions for implementing the method for predicting brain glioma classification according to any one of claims 1 to 7; The processor is used to execute the program instructions stored in the memory to control the brain glioma classification prediction method.

10. A storage medium, characterized in that: The method stores program instructions executable by a processor, wherein the program instructions are used to execute the method for predicting brain glioma classification according to any one of claims 1 to 7.

Images