Pathological image classification method and equipment based on multi-view depth Gaussian representation learning, and medium

Abstract

The invention relates to a pathological image classification method and equipment based on multi-view depth Gaussian representation learning, and a medium. The method comprises the following steps: acquiring a pathological image data set; capturing different view features of the pathological image by using a deep neural network; constructing a multi-view depth Gaussian process model, performing modeling representation on different view features of the pathological image through an embedded convolution kernel, and converting the different view features of the pathological image into embedded representation in the same space to obtain multi-view common features; wherein a Bayesian training method is adopted to train the multi-view depth Gaussian process model; and inputting the multi-view public features into a pre-trained classification network, and outputting a classification result. Compared with the prior art, the method has the advantage of accurate and reliable classification.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a pathological image classification method, device and medium based on multi-view deep Gaussian representation learning. Background Art

[0002] Accurate pathological classification results are crucial for assisting in the selection of subsequent cancer treatment options.

[0003] Pathologists need to examine and process a large number of pathological sections and perform manual classification every day. The manual classification process has the following problems: (1) Manual interpretation is time-consuming and inefficient, and long-term classification work can easily cause fatigue, greatly reducing the accuracy of case image classification; (2) The classification of some special pathological images only relies on qualitative morphological standards and lacks unified standards. Therefore, it is highly dependent on the pathologist's knowledge reserves and clinical experience, and has a certain degree of subjectivity.

[0004] In recent years, with the rapid development of digital technology, digital pathology applications have emerged. This involves using digital slide scanners to digitize slides into full-slide images and computer-assisted pathology image classification, greatly improving the efficiency of case image classification.

[0005] However, when analyzing pathology images, color inconsistencies can occur due to variations in color concentration, the use of different dye chemicals, and differences in scanners from different vendors. Therefore, color normalization is necessary to account for these variations. However, color normalization can result in the loss of a significant amount of useful information in the image, significantly impacting the accuracy of classification results. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a pathological image classification method, device and medium based on multi-view deep Gaussian representation learning with high classification accuracy.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] According to a first aspect of the present invention, a pathological image classification method based on multi-view deep Gaussian representation learning is provided, comprising:

[0009] Obtain a pathology image dataset;

[0010] Using deep neural networks to capture different view features of pathological images;

[0011] A multi-view deep Gaussian process model is constructed. The features of different views of the pathology image are modeled and represented by embedding convolution kernels. The features of different views of the pathology image are converted into embedded representations in the same space to obtain multi-view common features. The multi-view deep Gaussian process model is trained using a Bayesian training method.

[0012] Input the multi-view common features into the pre-trained classification network and output the classification results.

[0013] Preferably, the multi-view deep Gaussian process model includes multiple Gaussian process mapping layers embedded with convolution kernels, and the different view features of the pathological image are modeled and represented by the embedded convolution kernels, and the different view features of the pathological image are converted into embedded representations in the same space to obtain multi-view common features, which are expressed as:

[0014]

[0015] Where: is a shared latent variable; is the deep Gaussian process output corresponding to view v; X ＜v,h-1＞ is a latent variable unique to view v; X ＜v＞ Input data for the model corresponding to view v; is the model output data, which contains the common features of all view input data; H is the modeling depth of the common features; H v Represents the feature modeling depth of view v; V is the total number of views; f is the Gaussian process mapping layer function corresponding to the embedded convolution kernel, ε is the bias, and σ is the variance.

[0016] Preferably, the multi-view deep Gaussian process model training process specifically includes:

[0017] 1) Establish the joint probability distribution of latent variables and observed variables:

[0018]

[0019] Where: represents the set of latent variables; p(A|B) represents the probability of event A occurring under the condition that event B occurs; For events Probability of occurrence;

[0020] 2) Based on the joint probability distribution of the latent variable and the observed variable, the variational distribution is used to approximate the posterior distribution of the latent variable; wherein the posterior distribution expression of the latent variable is:

[0021]

[0022] 3) The KL divergence is used to measure the matching degree between the variational distribution and the posterior distribution. The KL divergence is minimized by maximizing the lower bound of the evidence, and the optimization objective function is constructed to perform the optimization solution.

[0023] Preferably, the KL divergence is used to measure the degree of matching between the variational distribution and the posterior distribution, and the optimization objective function is constructed by maximizing the lower bound of the evidence to minimize the KL divergence. The objective function expression is:

[0024]

[0025] Where: E Z is the expectation of the latent variable set Z; p(X ＜1＞ ,…,X ＜V＞ , Z) is the joint probability distribution of latent variables and observed variables; Q is the variational distribution; g ＜v＞ is the expectation of the Gaussian distribution involving only v views; r ＜v,h＞ is the expectation of all variable densities involving v views h layers; q(X ＜v,h＞ ) is the variational posterior distribution involving v views h layers; H[q] represents the entropy of distribution q; is the expectation of the density of all variables involved in the h layer; KL(q||p) is the KL divergence between distribution p and distribution q, which characterizes the matching relationship between distribution p and distribution q.

[0026] Preferably, based on the optimization objective function, a gradient descent algorithm is used to optimize the training of the multi-view deep Gaussian process model.

[0027] Preferably, in the multi-view deep Gaussian process model, the Gaussian process mapping layer function f embedded in the convolution kernel is specifically:

[0028]

[0029] Where: GP(0,k f (x i ,x j )) is zero mean and the convolution kernel is k f Prior of k f (x i ,x j ) is the image x i and image x j The extended convolution kernel between g is the basic kernel function; P is the image x i and image x j The number of image blocks in ; For image x i The p-th image block, For image x j The pth 'image blocks.

[0030] Preferably, after the multi-view deep Gaussian process model is trained, the representation vector of the model root node is taken as the final learned multi-view feature.

[0031] Preferably, the classification network is a support vector machine classifier.

[0032] According to a second aspect of the present invention, an electronic device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, any one of the methods described above is implemented.

[0033] According to a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, any one of the methods described above is implemented.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) The present invention combines multi-view learning and deep Gaussian process to construct a multi-view deep Gaussian process model. By embedding convolution kernels to model the features of different views of pathological images, it can not only extract the unique features of different views based on the differences between views, but also extract the common features between views, thereby satisfying the consistency and complementarity principles in multi-view learning to learn a more comprehensive representation and improve the accuracy of pathological image classification.

[0036] (2) Based on the diversity and differences of data from different views, the present invention constructs network structures of different depths for different views, so that intrinsic information can be extracted from different views and integrated into a comprehensive representation. It can effectively and adaptively balance the consistency and complementarity between different views, avoid the information loss caused by the existing color regularization scheme, and improve the efficiency and accuracy of case image classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of the method of the present invention;

[0038] Figure 2 is a probabilistic graphical model of deep Gaussian processes;

[0039] Figure 3 This is a framework diagram of the children's pathology image diagnosis model based on multi-view deep Gaussian representation learning. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0041] Example

[0042] like Figure 1 As shown, this embodiment provides a pathological image classification method based on multi-view deep Gaussian representation learning, which includes the following steps:

[0043] S1. Obtain pathological image dataset;

[0044] S2, using deep neural networks to capture different view features of pathological images;

[0045] S3. Construct a multi-view deep Gaussian process model, model and represent the different view features of the pathological image by embedding convolution kernels, convert the different view features of the pathological image into embedded representations in the same space, and obtain multi-view common features; wherein, the multi-view deep Gaussian process model is trained using a Bayesian training method;

[0046] S4. Input the multi-view common features into the pre-trained classification network and output the classification results.

[0047] Next, the method of this embodiment is introduced in detail.

[0048] S1. Obtain a pathological image dataset.

[0049] This example selects a pathological image dataset of children's tumors. The specific acquisition process includes:

[0050] Digital images of the whole pathology slides obtained by scanning the pathology slides with a digital slide scanner;

[0051] Digital images of whole pathology slides were digitized using a digitizer at a set magnification;

[0052] Each histology slide image is segmented into multiple non-overlapping image blocks of a set pixel size to obtain a pathology image dataset.

[0053] S2. Use deep neural network to capture different view features of pathological images.

[0054] In this embodiment, the original pathological image is preprocessed, including using different deep neural networks as feature extractors to extract different features, including texture, grayscale, shape features, etc., which are used as different views for subsequent multi-view learning to make up for the lack of information caused by single view features.

[0055] S3. Construct a multi-view deep Gaussian process model, model and represent the different view features of the pathological image by embedding convolution kernels, convert the different view features of the pathological image into embedded representations in the same space, and obtain multi-view common features; among them, the Bayesian training method is used to train the multi-view deep Gaussian process model.

[0056] The multi-view deep Gaussian process model constructed in this embodiment is different from the traditional deep Gaussian model. The traditional deep Gaussian process (see Figure 2 ) is generally suitable for processing single-view data. For multi-view data, the only option is to concatenate the different views into a single view. However, this operation ignores the unique statistical properties of each view, resulting in a meaningless synthesized view and difficulty in exploring the underlying nonlinear relationships between different views. It also increases the dimensionality of the feature space, making learning difficult.

[0057] like Figure 3 As shown, the multi-view deep Gaussian process model constructed in the embodiment can not only extract unique features of different views based on the differences between views, but also extract common features between views, thereby satisfying the consistency and complementarity principles in multi-view learning to learn a more comprehensive representation.

[0058] The Deep Gaussian Process (DGP) is a deep extension of the Gaussian Process (GP), where each layer of the network is a Gaussian process mapping. The Multi-View DGP model uses this DGP to transform the features of different views of pathology images into an embedded representation in the same space. Based on the diversity and differences in data from different views, network structures of varying depths are constructed for different views. This allows for the extraction of intrinsic information from different views, fusion into a comprehensive representation, and adaptive balance of consistency and complementarity between views.

[0059] Specifically, the multi-view deep Gaussian process model constructed in this embodiment includes multiple Gaussian process mapping layers with embedded convolution kernels. The different view features of the pathological image are modeled and represented by the embedded convolution kernels. The different view features of the pathological image are converted into embedded representations in the same space to obtain multi-view common features, which are expressed as:

[0060]

[0061] Where: is a shared latent variable; is the deep Gaussian process output corresponding to view v; X ＜v,h-1＞ is a latent variable unique to view v; X ＜v＞ Input data for the model corresponding to view v; is the model output data, which contains the common features of all view input data; H is the modeling depth of the common features; H v Represents the feature modeling depth of view v; V is the total number of views; f is the Gaussian process mapping layer function corresponding to the embedded convolution kernel, ε is the bias, and σ is the variance.

[0062] Among them, the Gaussian process mapping layer function f embedded in the convolution kernel is specifically:

[0063]

[0064] Where: GP(0,k f (x i ,x j )) is zero mean and the convolution kernel is k f Prior of k f (x i ,x j ) is the image x i and image x j The extended convolution kernel between g is the basic kernel function; P is the image x i and image x j The number of image blocks in ; For image x i The p-th image block, For image x j The pth ' image blocks.

[0065] In this embodiment, the training process of the multi-view deep Gaussian process model specifically includes:

[0066] 1) Establish the joint probability distribution of latent variables and observed variables:

[0067]

[0068] Where: represents the set of latent variables; p(A|B) represents the probability of event A occurring under the condition that event B occurs; For events Probability of occurrence;

[0069] 2) Based on the joint probability distribution of the latent variable and the observed variable, a variational distribution is used to approximate the posterior distribution of the latent variable, thereby cleverly transforming the posterior inference problem into an optimization problem for solution. This approach has good convergence and scalability, making it more suitable for solving large-scale approximate inference problems. The posterior distribution of the latent variable is expressed as:

[0070]

[0071] 3) The KL divergence is used to measure the matching degree between the variational distribution and the posterior distribution. The KL divergence is minimized by maximizing the lower bound of the evidence, and the optimization objective function is constructed to perform the optimization solution.

[0072] The posterior distribution is approximated using a variational distribution Q(Z;λ), where λ is a parameter of the variational distribution that is optimized to achieve the best approximation. The Kullback-Leibler (KL) divergence is used to measure the degree of match between the variational distribution and the posterior distribution. KL=0 indicates a perfect match. Minimizing the KL divergence is equivalent to maximizing the Evidence Lower Bound (ELBO). This lower bound is derived from logp(X) according to Jensen's inequality. <1> ,Λ,X <v>< / v> ) is derived, which is specifically described as:

[0073]

[0074] By derivation, the objective function is written in the form of expectation about a known distribution, and the objective expression is:

[0075]

[0076] Where: E Z is the expectation of the latent variable set Z; p(X ＜1＞ ,…,X ＜V＞ , Z) is the joint probability distribution of latent variables and observed variables; Q is the variational distribution; g ＜v＞ is the expectation of the Gaussian distribution involving only v views; r ＜v,h＞ is the expectation of all variable densities involving v views h layers; q(X ＜v,h＞ ) is the variational posterior distribution involving v views h layers; H[q] represents the entropy of distribution q; is the expectation of the density of all variables involved in the h layer; KL(q||p) is the KL divergence between distribution p and distribution q, which characterizes the matching relationship between distribution p and distribution q.

[0077] Based on the optimization objective function, the gradient descent algorithm is used to optimize the training of the multi-view deep Gaussian process model. The specific process includes:

[0078] 1) Determine the network structure based on the input data and randomly initialize the model parameters, including the number of network layers H, H v Hyperparameters such as , the dimension of the latent space Q H , batch size, maximum number of iterations;

[0079] 2) Update the model parameters using the gradient descent algorithm;

[0080] 3) When the objective function remains unchanged after the training reaches the maximum number of iterations, the model training is terminated.

[0081] After the multi-view deep Gaussian process model is trained, the representation vector of the model root node is taken as the final learned multi-view feature.

[0082] Specifically, for each batch of the training set, the following operations are performed: calculating the parameters of the variational posterior distribution, calculating the mean of the model root node, and calculating the feature representation of the final output based on the mean of the model root node.

[0083] The deep Gaussian process model selected in this embodiment is a Bayesian probability model based on the kernel method, which is more robust and interpretable than the deep model. At the same time, the kernel function in the model is an effective way to solve the problem of nonlinear model analysis, and can discover and learn the relationship between data. The kernel function is designed for specific application problems. It is easy to integrate prior knowledge related to the problem and has better anti-overfitting control to ensure generalization performance. The mean field approximation is also selected to make the posterior distribution more simple and effective. In this way, a stable training process can be obtained.

[0084] S4. Input the multi-view common features into the pre-trained classification network and output the classification results.

[0085] In this embodiment, the classification network uses a support vector machine classifier. After the model training is completed, the representation vector of the model root node is taken as the final learned multi-view representation. The learned multi-view representation is input into the pre-trained support vector machine classifier for classification, and the classification result is output.

[0086] In this example, the classification is illustrated using international neuroblastoma pathology images. Peripheral neuroblastoma is divided into four basic forms: neuroblastoma, mixed ganglioneuroblastoma, ganglioneuroma, and nodular ganglioneuroblastoma. Neuroblastoma can be further divided into three subtypes: undifferentiated, poorly differentiated, and differentiated.

[0087] Five experienced pathologists manually annotated the pathology image data based on the downstream analyses to be performed. For each tumor type, the original pathology images were divided into 300 non-overlapping 768×768 sub-slices. Each 768×768 sub-slice was further divided into 144 non-overlapping 64×64 sub-slices. Therefore, a total of 43,200 pathology slices were obtained for each tumor type.

[0088] In the pathology image representation learning task, this example first extracts different features of the pathology image as different views. These images belong to different modalities and have significant differences in feature representation. After training and testing on the dataset, the diagnostic performance of this example is reported in Table 1.

[0089] Table 1 Pathological image classification results of different methods on the test dataset

[0090]

[0091] Table 1 shows the pathological image classification results of the method proposed in this invention and some multi-view methods on the test dataset, among which the multi-view deep Gaussian process model MvDGPID (Multi-view Deep Gaussian Pathology Image Diagnosis) is the method of this invention, CCA (Canonical Correlation Analysis) is a classic multi-view method, KCCA (Kernel CCA) is the kernel method extension version of CCA, DCCA (Deep CCA) is the deep extension version of CCA, VCCA (Deep Variational CCA) is the deep variational extension version of CCA, MULPP (Multi-view Uncorrelated Locality Preserving Projection) considers the preservation of local feature maps between different views, AE 2 -Nets (Autoencoder in Autoencoder Networks) integrates different views through autoencoder networks and finally learns a common potential representation.

[0092] The results show that the pathology classification results of the proposed model are better than those of all the comparison methods. At the same time, the more view features are used, the more accurate the pathology classification results are.

[0093] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0094] Many components in a device are connected to the I / O interface, including: input units, such as a keyboard and mouse; output units, such as various types of displays and speakers; storage units, such as magnetic disks and optical disks; and communication units, such as network cards, modems, and wireless communication transceivers. The communication unit allows the device to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks.

[0095] The processing unit performs the various methods and processes described above, such as methods S1 to S4. For example, in some embodiments, methods S1 to S4 can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S4 described above can be performed. Alternatively, in other embodiments, the CPU can be configured to execute methods S1 to S4 by any other appropriate means (for example, by means of firmware).

[0096] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), and the like.

[0097] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0098] In the context of the present invention, machine-readable medium can be a tangible medium that can contain or store a program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0099] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A pathological image classification method based on multi-view deep Gaussian representation learning, characterized in that: include: Obtain a pathology image dataset; Using deep neural networks to capture different view features of pathological images; A multi-view deep Gaussian process model is constructed. The features of different views of the pathology image are modeled and represented by embedding convolution kernels. The features of different views of the pathology image are converted into embedded representations in the same space to obtain multi-view common features. The multi-view deep Gaussian process model is trained using a Bayesian training method. Input the multi-view common features into the pre-trained classification network and output the classification results.

2. A pathological image classification method based on multi-view deep Gaussian representation learning according to claim 1, characterized in that: The multi-view deep Gaussian process model includes multiple Gaussian process mapping layers embedded with convolution kernels. The different view features of the pathological image are modeled and represented by the embedded convolution kernels. The different view features of the pathological image are converted into embedded representations in the same space to obtain the multi-view common features, which are expressed as: X ＜v,h-1＞ =f ＜v,h＞ (X ＜v,h＞ )+e ＜v,h＞ ,he ＜v,h＞ ~N(0,(σ ＜v,h＞ ) 2 I),h∈{2,…,H v } (3) X ＜v＞ =f ＜v,1＞ (X ＜v,1＞ )+e ＜v,1＞ ,he ＜v,1＞ ~N(0,(σ ＜v,1＞ ) 2 I),v∈{1,…,V} (4) Where: is a shared latent variable; is the deep Gaussian process output corresponding to view v; X ＜v,h-1＞ is a latent variable unique to view v; X ＜v＞ Input data for the model corresponding to view v; is the model output data, which contains the common features of all view input data; H is the modeling depth of the common features; H v Represents the feature modeling depth of view v; V is the total number of views; f is the Gaussian process mapping layer function corresponding to the embedded convolution kernel, ε is the bias, and σ is the variance.

3. A pathological image classification method based on multi-view deep Gaussian representation learning according to claim 2, characterized in that: The multi-view deep Gaussian process model training process specifically includes: 1) Establish the joint probability distribution of latent variables and observed variables: Where: represents the set of latent variables; p(A|B) represents the probability of event A occurring under the condition that event B occurs; For events Probability of occurrence; 2) Based on the joint probability distribution of the latent variable and the observed variable, the variational distribution is used to approximate the posterior distribution of the latent variable; wherein the posterior distribution expression of the latent variable is: 3) The KL divergence is used to measure the degree of match between the variational distribution and the posterior distribution. The KL divergence is minimized by maximizing the lower bound of the evidence representation, and the optimization objective function is constructed to perform the optimization solution.

4. A pathological image classification method based on multi-view deep Gaussian representation learning according to claim 3, characterized in that: The KL divergence is used to measure the matching degree between the variational distribution and the posterior distribution. The optimization objective function is constructed by maximizing the lower bound of the evidence to minimize the KL divergence. The objective function expression is: Where: E Z is the expectation of the latent variable set Z; p(X ＜1＞ ,…,X ＜V＞ , Z) is the joint probability distribution of latent variables and observed variables; Q is the variational distribution; g ＜v＞ is the expectation of the Gaussian distribution involving only v views; r ＜v,h＞ is the expectation of all variable densities involving v views h layers; q(X ＜v,h＞ ) is the variational posterior distribution involving v views h layers; H[q] represents the entropy of distribution q; is the expectation of the density of all variables involved in the h layer; KL(q||p) is the KL divergence between distribution p and distribution q, which characterizes the matching relationship between distribution p and distribution q.

5. A pathological image classification method based on multi-view deep Gaussian representation learning according to claim 3, characterized in that: Based on the optimization objective function, the gradient descent algorithm is used to optimize the training of the multi-view deep Gaussian process model.

6. A pathological image classification method based on multi-view deep Gaussian representation learning according to claim 3, characterized in that: In the multi-view deep Gaussian process model, the Gaussian process mapping layer function f embedded in the convolution kernel is specifically: Where: GP(0,k f (x i ,x j )) is zero mean and the convolution kernel is k f Prior of k f (x i ,x j ) is the image x i and image x j The extended convolution kernel between g is the basic kernel function; P is the image x i and image x j The number of image blocks in ; For image x i The p-th image block, For image x j The pth ' image blocks.

7. A pathological image classification method based on multi-view deep Gaussian representation learning according to claim 3, characterized in that: After the multi-view deep Gaussian process model is trained, the representation vector of the model root node is taken as the final learned multi-view feature.

8. A pathological image classification method based on multi-view deep Gaussian representation learning according to claim 1, characterized in that: The classification network is a support vector machine classifier.

9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images