Cancer prognostic marker mining method, device, server and storage medium

Abstract

The embodiment of the present invention provides a method, device, server and storage medium for mining cancer prognosis markers. The commonly used concept of directly using a deep learning model to predict the cancer prognosis risk index is abandoned. The distribution relationship between tissues and tissues included in the pathological slice image is characterized based on multi-category tissue spatial distribution data, so that the attribution method and the trained prognosis model can be used to determine the high-contribution area of ​​prognosis in the multi-category tissue spatial distribution data, and potential markers are determined based on the high-contribution area of ​​prognosis. The potential markers characterize: at least one category of tissue and the distribution relationship between each tissue, so that a mathematical model with clear pathological meaning can be established based on the potential markers, and the patient's prognosis can be predicted with high precision based on the mathematical model of the potential markers, overcoming the difficulty of deep learning in clinical application, and realizing an interpretable and generalizable precision medical prognosis based on deep learning.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of information processing technology, and in particular to a method, device, server and storage medium for mining cancer prognosis markers. Background Art

[0002] Artificial intelligence technology, represented by deep learning, can analyze complex information sources in a data-driven way and has been widely used in various complex image processing tasks. A deep neural network can be regarded as a high-dimensional function with a large number of learnable parameters. On a specific data set, a deep neural network is trained in a back-propagation manner, continuously updating its learnable parameters and continuously narrowing the gap between the function output and the target output, thereby realizing the function of data-based learning.

[0003] In recent years, with the gradual improvement of artificial intelligence theory and the rapid development of deep learning technology, it has become possible to predict the cancer stage and prognosis of the patient based only on the microscopic digital pathology section images stained with hematoxylin and eosin (H&E). Most of the current deep learning-based patient prognosis methods use digital pathology section images as input and a single risk index as output to achieve the prognosis of the patient, and achieve a prognosis accuracy close to or even exceeding that of professional pathologists. However, the results given by these methods are only an abstract risk index. Due to the current limited interpretability and generalization capabilities of deep learning, this single indicator obtained by the black box is difficult to be accepted clinically. Therefore, it is difficult to apply this single risk index prediction method to clinical practice.

[0004] In summary, there is an urgent need to propose a new solution for cancer prognosis using deep learning methods. Summary of the invention

[0005] Embodiments of the present invention provide a method, device, server and storage medium for mining cancer prognosis markers to achieve interpretable and generalizable precision medicine prognosis based on deep learning.

[0006] A first aspect of an embodiment of the present invention provides a method for mining cancer prognosis markers, the method comprising:

[0007] Processing the sample pathological slice image to obtain the sample multi-category tissue spatial distribution data, wherein the sample multi-category tissue spatial distribution data represents: the multiple categories of tissues included in the sample pathological slice image and the distribution relationship of the multiple categories of tissues;

[0008] The preset model is trained using the sample multi-category tissue spatial distribution data and the corresponding prognostic index labels to obtain a prognostic model;

[0009] According to the prognostic model and the sample multi-category tissue spatial distribution data, a two-dimensional contribution heat map of the sample multi-category tissue spatial distribution data is obtained, wherein the two-dimensional contribution heat map represents: an area in the sample multi-category tissue spatial distribution data that has a high contribution to the prognostic index;

[0010] A potential marker is determined according to the two-dimensional contribution heat map and the spatial distribution data of the multi-category tissues of the sample, wherein the potential marker represents: at least one category of tissues and the distribution relationship between the tissues.

[0011] Optionally, the sample pathological slice image is processed to obtain the sample multi-category tissue spatial distribution data, including:

[0012] Slice the sample pathological slice image to obtain multiple small block images;

[0013] Classifying the tissues included in the plurality of small-block images respectively, and marking the tissue categories of the small-block images;

[0014] According to the spatial position information of multiple small-block images, the tissue categories of the multiple small-block images are spliced ​​to form sample multi-category tissue spatial distribution data.

[0015] Optionally, the method further comprises:

[0016] Testing the potential markers, and comparing the prognostic significance of the potential markers with existing prognostic markers;

[0017] Potential markers that show significant results during the screening test are used as new markers.

[0018] Optionally, testing the potential marker comprises:

[0019] The COX multivariate regression analysis method was used to obtain the risk ratio and significance of the potential markers.

[0020] Optionally, determining potential markers according to the two-dimensional contribution heat map and the sample multi-category tissue spatial distribution data includes:

[0021] Superimposing the two-dimensional contribution heat map with the multi-category tissue spatial distribution data of the sample to obtain a superimposed image;

[0022] Potential markers are determined based on the superimposed images, where the potential markers represent: at least one type of tissue and the distribution relationship between the tissues.

[0023] Optionally, the preset model includes: a convolution layer for processing multi-channel input images and a batch normalization layer for stabilizing prediction results; the loss function of the preset model includes: prognosis accuracy loss and regularization loss.

[0024] A second aspect of an embodiment of the present invention provides a cancer prognosis marker mining device, the device comprising:

[0025] A sample processing module is used to process the sample pathological slice image to obtain the sample multi-category tissue spatial distribution data, wherein the sample multi-category tissue spatial distribution data represents: the multiple categories of tissues included in the sample pathological slice image and the distribution relationship of the multiple categories of tissues;

[0026] A training module is used to train a preset model using sample multi-category tissue spatial distribution data and corresponding prognostic index labels to obtain a prognostic model;

[0027] An attribution module, for obtaining a two-dimensional contribution heat map of the multi-category tissue spatial distribution data of the sample according to the prognostic model and the multi-category tissue spatial distribution data of the sample, wherein the two-dimensional contribution heat map represents: an area in the multi-category tissue spatial distribution data of the sample that has a high contribution to the prognostic index;

[0028] A potential marker determination module is used to determine potential markers based on the two-dimensional contribution heat map and the sample multi-category tissue spatial distribution data.

[0029] Optionally, the sample processing module includes:

[0030] A sub-slicing module is used to slice the sample pathological slice image into multiple small pieces of images;

[0031] A classification submodule, used to classify the tissues included in the plurality of small-block images respectively, and mark the tissue categories of the small-block images;

[0032] The splicing submodule is used to splice the tissue categories of multiple small-block images according to the spatial position information of the multiple small-block images to form sample multi-category tissue spatial distribution data.

[0033] Optionally, the device further comprises:

[0034] A testing module, used to test the potential markers and compare the prognostic significance of the potential markers with existing prognostic markers;

[0035] The screening module is used to screen potential markers that show significant performance during the test process as new markers.

[0036] Optionally, the inspection module is specifically used to:

[0037] The risk ratio and significance of the potential markers were obtained using the COX multifactor regression analysis device.

[0038] Optionally, the potential marker determination module comprises:

[0039] A superposition submodule, used for superimposing the two-dimensional contribution heat map with the sample multi-category tissue spatial distribution data to obtain a superimposed image;

[0040] The determination submodule is used to determine potential markers based on the superimposed image, wherein the potential markers represent: at least one type of tissue and the distribution relationship between each tissue.

[0041] Optionally, the preset model includes: a convolution layer for processing multi-channel input images and a batch normalization layer for stabilizing prediction results; the loss function of the preset model includes: prognosis accuracy loss and regularization loss.

[0042] A third aspect of an embodiment of the present invention provides a server, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps of the method described in the first aspect of the present invention.

[0043] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method described in the first aspect of the present invention.

[0044] The cancer prognosis marker mining method provided by the embodiment of the present invention abandons the common concept of directly using a deep learning model to predict the cancer prognosis risk index. First, the embodiment of the present invention proposes to process the sample pathological section image to obtain the sample multi-category tissue spatial distribution data, and the sample multi-category tissue spatial distribution data represents: the sample pathological section image includes multiple categories of tissues and the distribution relationship of the multiple categories of tissues, so as to use the sample multi-category tissue spatial distribution data and the corresponding prognosis index label to train a preset model to obtain a prognosis model. The prognosis model can predict the prognosis index corresponding to the input data based on the multiple categories of tissues included in the input data and the distribution relationship of the multiple categories of tissues.

[0045] Furthermore, the embodiment of the present invention proposes that, based on the attribution method, according to the prognostic model and the sample multi-category tissue spatial distribution data, a two-dimensional contribution heat map of the sample multi-category tissue spatial distribution data is obtained, and the two-dimensional contribution heat map characterizes the area with high contribution to the prognostic index in the sample multi-category tissue spatial distribution data. Thus, the embodiment of the present invention can use the attribution method to determine the prognostic high contribution area in the multi-category tissue spatial distribution data, and determine potential markers based on the prognostic high contribution area.

[0046] Therefore, the embodiment of the present invention can mine potential markers that dominate the cancer based on pathological section images. In the embodiment of the present invention, the distribution relationship between tissues and tissues included in the marker is characterized based on multi-category tissue spatial distribution data, so that a mathematical model with clear pathological meaning can be established, and the patient's prognosis can be predicted with high precision based on the mathematical model of the marker, overcoming the difficulty of deep learning in clinical application, and realizing an interpretable and generalizable precision medicine prognosis based on deep learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0048] Figure 1 is a flow chart of a cancer prognosis marker mining method according to an embodiment of the present invention;

[0049] Figure 2 is a flow chart of another method for mining cancer prognosis markers according to an embodiment of the present invention;

[0050] Figure 3 It is a structural block diagram of a cancer prognosis marker mining device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Reference Figure 1 , showing a flow chart of a cancer prognosis marker mining method according to an embodiment of the present invention. A cancer prognosis marker mining method disclosed in an embodiment of the present invention is applied to a server and may specifically include the following steps:

[0053] S101, processing the sample pathological slice image to obtain the sample multi-category tissue spatial distribution data.

[0054] The sample multi-category tissue spatial distribution data represents: the multiple categories of tissues included in the sample pathological section image and the distribution relationship between the multiple categories of tissues.

[0055] In the embodiment of the present invention, the sample pathological section image is a microscopic digital pathological section image stained with hematoxylin-eosin.

[0056] In the embodiment of the present invention, the pathological slice image can be converted into multi-category tissue spatial distribution data by utilizing the categories to which each tissue included in the sample pathological slice image belongs and the distribution relationship between each tissue.

[0057] Specifically, in an embodiment of the present invention, a pathological section image can be segmented to obtain a plurality of small images, and then the classification of tissues contained in each small image can be determined. Then, based on the position information of each small image and the classification of tissues contained in each small image, multiple types of tissue spatial distribution data can be spliced ​​together to obtain data. Thus, the multiple types of tissue spatial distribution data can represent: the categories to which the various tissues included in the pathological section image belong, and the distribution relationship between these tissues.

[0058] Specifically, in an optional implementation manner, step S101 includes the following sub-steps:

[0059] S1011, dividing the sample pathological slice image into multiple small block images.

[0060] In an embodiment of the present invention, the sample pathology slice image can be evenly divided into multiple small-size small-block images according to preset segmentation parameters, wherein the preset segmentation parameters can be: dividing the sample pathology slice at a 20× magnification into 150×150 pixel small-block images without overlap.

[0061] S1012: Classify the tissues included in the plurality of small-block images respectively, and mark the tissue categories of the small-block images.

[0062] In the embodiment of the present invention, the tissues included in the plurality of small images can be classified respectively using a pre-trained classification model, and the tissue categories of the small images can be labeled according to the classification results, so that the tissue categories of the small images can be used to represent the corresponding small images.

[0063] S1013, according to the spatial position information of the multiple small-block images, the tissue categories of the multiple small-block images are spliced ​​to form sample multi-category tissue spatial distribution data.

[0064] In the embodiment of the present invention, the tissue categories of each small image can be spliced ​​into the sample multi-category tissue spatial distribution data according to the position information of each small image in the original sample pathological slice image.

[0065] For ease of understanding, step S101 is explained by taking hepatocellular carcinoma as an example. It should be noted that the embodiment of the present invention is applicable to the mining of prognostic markers for any cancer.

[0066] For example, in an embodiment of the present invention, a classification model can be used to implement multi-category tissue classification for an input pathological slice image. According to relevant knowledge about hepatocellular carcinoma, common tissue types include normal, tumor, necrosis, fibrosis, inflammation, fat, and bile duct reaction. In addition, in order to ensure the classification accuracy of the model, a blank background is used as the eighth common tissue type. Therefore, the role of the classification model in an embodiment of the present invention can be specifically: classify each part of the input pathological slice image into the above eight common tissues.

[0067] Specifically, the mathematical model of the multi-category organization division process can be described as:

[0068] Using the classification model f seg The input pathological slice image Mapped to probability images of eight categories t Then transformed into eight tissue segmentation images Where m′=int(m / 150), n′=int(n / 150).

[0069] M=f seg (W)

[0070]

[0071] Specifically:

[0072] First, at a magnification of 20×, the pathological slice image W is segmented into non-overlapping small color block images I of 150×150 pixels (this value is only used as a setting value for this example. When the embodiment of the present invention is specifically applied, any adaptive value can be set according to actual needs). seg Small pieces of image Mapping The probability vector of . Among them, (i, j) is the spatial position of the small image I, p t is the probability that I belongs to the eight common tissues. Finally, all p are spliced ​​into M according to the spatial position information (i, j) of each small image block. For each position (i, j) in M, the maximum probability is taken as the tissue category represented by the position, and finally the tissue segmentation result S is generated, which is the spatial distribution data of multi-category tissues.

[0073] S102, training a preset model using the sample multi-category tissue spatial distribution data and the corresponding prognostic index labels to obtain a prognostic model.

[0074] In an embodiment of the present invention, the prognostic index corresponding to the sample pathological slice image can be determined according to the follow-up results of the patient corresponding to the sample pathological slice image, and the prognostic index is used as the prognostic index label corresponding to the sample pathological slice image. The method of determining the corresponding prognostic index according to the follow-up results of the patient can adopt any feasible method in the relevant technology, for example, the patient's life span can be mapped to the corresponding prognostic index.

[0075] In the embodiment of the present invention, the trained prognostic model can predict the corresponding prognostic index according to the input multi-category tissue spatial distribution data.

[0076] Among them, the preset model includes: a convolution layer for processing multi-channel input images and a batch normalization layer for stabilizing prediction results; the loss function of the preset model includes: prognosis accuracy loss and regularization loss.

[0077] In an optional embodiment of the present invention, the preset model uses a deep multi-channel residual network, including thirty-two parallel sub-convolutional layers and sixteen connected residual blocks, wherein the multi-channel parallel sub-convolutional layers can increase the number of features extracted by the network without changing the number of parameters, and the residual span connection transmits information losslessly to the subsequent convolutional layer. Thus, the trained prognostic model can predict the prognostic index by automatically extracting the prognostic-related features of the input multi-category tissue spatial distribution data.

[0078] In an optional implementation of the present invention, the preset model may select a ResNeXt structure to learn an abstract representation of the data through continuously connected downsampling layers, and judge the prognosis of the input data at the scale of the receptive field. This structure can improve the perception of high-frequency fine structures and is conducive to the mining of fine potential markers.

[0079] In an embodiment of the present invention, a preset model is trained using multi-category tissue spatial distribution data to obtain a prognostic model, so that the prognostic model can effectively match the attribution method and achieve accurate positioning of prognosis-related features.

[0080] In an optional embodiment of the present invention, the loss function of the preset model consists of two parts, the first part is the prognostic accuracy loss of the prognostic network; the second part is the regularization loss, so that the multi-channel prognostic network can avoid overfitting to a certain extent. Among them, the prognostic accuracy loss adopts the COX negative log-likelihood loss method; the regularization loss adopts the L1 norm based on the model parameters as the loss function.

[0081] Specifically, the loss function of the preset model can be expressed as: l = l prognosis +l reg ;

[0082] Among them, the prognostic accuracy loss can be expressed as:

[0083]

[0084] The regularization loss can be expressed as:

[0085] l reg =F(X)1;

[0086] Among them, U represents uncensored patients, R i ={j|Y j ≥Y i} is the survival time Y j The set of patient j whose disease is shorter than that of patient i. is the prognostic index given by the prognostic model, where β is the parameter of the last layer of the model, X i is the characteristic of patient i. F(X) represents the mathematical abstract expression of the prognostic model.

[0087] S103, obtaining a two-dimensional contribution heat map of the multi-category tissue spatial distribution data of the sample according to the prognostic model and the multi-category tissue spatial distribution data of the sample.

[0088] The two-dimensional contribution heat map represents: the area with high contribution to the prognostic index in the multi-category tissue spatial distribution data of the sample.

[0089] Specifically, in an embodiment of the present invention, the preset model can be trained with minimizing the loss function as the goal, multi-category tissue spatial distribution data of the samples as input, and the prognostic index corresponding to the samples as labels.

[0090] Specifically, the preset Adam optimizer can be used to optimize the model parameters, and finally the prognosis model is trained.

[0091] S104, determining potential markers according to the two-dimensional contribution heat map and the multi-category tissue spatial distribution data of the sample.

[0092] Wherein, the potential markers characterize: at least one type of tissue and the distribution relationship between each tissue.

[0093] Specifically, in an embodiment of the present invention, the attribution method in deep learning is used to visualize the area of ​​interest of the model to explore the area or feature that the prognostic model is interested in. Specifically, in an embodiment of the present invention, a gradient-based attribution method can be used, which generates a two-dimensional contribution heat map by calculating the gradient of each pixel (represented as the tissue classification of each small image) contained in the input multi-category tissue spatial distribution data relative to the prognostic index. After obtaining the two-dimensional contribution heat map, the two-dimensional contribution heat map is superimposed with the sample multi-category tissue spatial distribution data to achieve accurate and intuitive prognostic index attribution.

[0094] In an optional implementation manner, the step S104 includes the following sub-steps:

[0095] S1041, superimposing the two-dimensional contribution heat map with the sample multi-category tissue spatial distribution data to obtain a superimposed image.

[0096] S1042, determining potential markers based on the superimposed image.

[0097] The method for mining cancer prognostic markers proposed in the embodiment of the present invention builds a preset model, designs a loss function, and produces a multi-category tissue spatial distribution data set. The training set is used as a training sample, and the corresponding prognostic index is used as a label. The preset model is trained with the goal of minimizing the loss function to obtain a prognostic model, and an attribution method is used to obtain high-contribution areas for prognosis, and potential markers are determined based on the high-contribution areas.

[0098] In an optional embodiment of the present invention, another method for mining cancer prognosis markers is disclosed, referring to Figure 2 , shows a flow chart of another cancer prognosis marker mining method according to an embodiment of the present invention, which specifically includes the following steps:

[0099] S201, processing the sample pathological slice image to obtain the sample multi-category tissue spatial distribution data.

[0100] S202, training a preset model using the sample multi-category tissue spatial distribution data and the corresponding prognostic index labels to obtain a prognostic model.

[0101] S203, obtaining a two-dimensional contribution heat map of the multi-category tissue spatial distribution data of the sample according to the prognostic model and the multi-category tissue spatial distribution data of the sample.

[0102] S204, determining potential markers according to the two-dimensional contribution heat map and the multi-category tissue spatial distribution data of the sample.

[0103] Steps S201 - S204 are similar to the above-mentioned steps S101 - S104 , and are not described in detail in this embodiment of the present invention.

[0104] S205, testing the potential markers, and comparing the prognostic significance of the potential markers with existing prognostic markers.

[0105] S206, screening potential markers that are significant in the test process as new markers.

[0106] In the embodiment of the present invention, the method for testing the potential markers can adopt any cancer marker testing method in the related art.

[0107] In an optional implementation manner, the step S205 specifically includes: using the COX multivariate regression analysis method to obtain the risk ratio and significance of the potential marker.

[0108] Specifically, in the embodiment of the present invention, the Kaplan-Meier method can be used to calculate the significance of dividing patients into high-risk groups and low-risk groups using potential markers as indicators. Then, the potential markers are subjected to COX proportional risk univariate and multivariate analysis together with existing markers to compare the significance of the potential markers and the risk ratio of clinically existing cancer markers on prognosis. If the potential marker is significant, the potential marker is identified as a newly discovered cancer prognosis marker.

[0109] The cancer prognosis marker mining method proposed in the embodiment of the present invention, after determining the potential marker, can establish a mathematical model with clear pathological meaning, test the effect of the potential marker on prognosis, and compare the significance of the potential marker with the existing markers on prognosis; thereby screening the potential markers that show significant performance in the test process as new markers.

[0110] Therefore, the embodiments of the present invention can use pathological section images to mine prognostic markers that dominate the cancer, characterize the markers based on multi-category tissue spatial distribution information, and establish a mathematical model with clear pathological meaning, so as to make a high-precision prediction of the patient's prognosis based on the mathematical model of the marker, overcome the difficulty of deep learning in clinical application, and realize an interpretable and generalizable precision medicine prognosis based on deep learning.

[0111] Based on the same inventive concept, an embodiment of the present invention provides a cancer prognosis marker mining device. Figure 3 , Figure 3 Schematic diagram of a cancer prognosis marker mining device provided in an embodiment of the present invention.

[0112] like Figure 3 As shown, the device comprises:

[0113] The sample processing module 301 is used to process the sample pathological slice image to obtain the sample multi-category tissue spatial distribution data, wherein the sample multi-category tissue spatial distribution data represents: the multiple categories of tissues included in the sample pathological slice image and the distribution relationship of the multiple categories of tissues;

[0114] The training module 302 is used to train the preset model using the sample multi-category tissue spatial distribution data and the corresponding prognostic index labels to obtain a prognostic model;

[0115] An attribution module 303 is used to obtain a two-dimensional contribution heat map of the multi-category tissue spatial distribution data of the sample according to the prognostic model and the multi-category tissue spatial distribution data of the sample, wherein the two-dimensional contribution heat map represents: an area in the multi-category tissue spatial distribution data of the sample that has a high contribution to the prognostic index;

[0116] The potential marker determination module 304 is used to determine potential markers according to the two-dimensional contribution heat map and the sample multi-category tissue spatial distribution data.

[0117] Optionally, the sample processing module 301 includes:

[0118] A sub-slicing module is used to slice the sample pathological slice image into multiple small pieces of images;

[0119] A classification submodule, used to classify the tissues included in the plurality of small-block images respectively, and mark the tissue categories of the small-block images;

[0120] The splicing submodule is used to splice the tissue categories of multiple small-block images according to the spatial position information of the multiple small-block images to form sample multi-category tissue spatial distribution data.

[0121] Optionally, the device further comprises:

[0122] A testing module, used to test the potential markers and compare the prognostic significance of the potential markers with existing prognostic markers;

[0123] The screening module is used to screen potential markers that show significant performance during the test process as new markers.

[0124] Optionally, the inspection module is specifically used to:

[0125] The risk ratio and significance of the potential markers were obtained using the COX multifactor regression analysis device.

[0126] Optionally, the potential marker determination module 304 includes:

[0127] A superposition submodule, used for superimposing the two-dimensional contribution heat map with the sample multi-category tissue spatial distribution data to obtain a superimposed image;

[0128] The determination submodule is used to determine potential markers based on the superimposed image, wherein the potential markers represent: at least one type of tissue and the distribution relationship between each tissue.

[0129] Optionally, the preset model includes: a convolution layer for processing multi-channel input images and a batch normalization layer for stabilizing prediction results; the loss function of the preset model includes: prognosis accuracy loss and regularization loss.

[0130] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0131] A third aspect of an embodiment of the present invention provides a server, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps in the method for mining cancer prognosis markers described in any of the above embodiments.

[0132] A fourth aspect of the embodiments of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the cancer prognosis marker mining method described in any of the above embodiments.

[0133] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0134] It will be appreciated by those skilled in the art that the embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0135] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0136] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0137] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0138] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0139] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0140] The above is a detailed introduction to a cancer prognosis marker mining method, device, server and storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for mining cancer prognosis markers, characterized in that: The method comprises: Processing the sample pathological slice image to obtain the sample multi-category tissue spatial distribution data, wherein the sample multi-category tissue spatial distribution data represents: the multiple categories of tissues included in the sample pathological slice image and the distribution relationship of the multiple categories of tissues; The preset model is trained using the sample multi-category tissue spatial distribution data and the corresponding prognostic index labels to obtain a prognostic model, and the trained prognostic model is used to predict the corresponding prognostic index according to the input multi-category tissue spatial distribution data; According to the prognostic model and the sample multi-category tissue spatial distribution data, a two-dimensional contribution heat map of the sample multi-category tissue spatial distribution data is obtained, including: using a gradient-based attribution method, by calculating the gradient of the prognostic index relative to the tissue classification of each small block image included in the input multi-category tissue spatial distribution data to generate a two-dimensional contribution heat map; the two-dimensional contribution heat map represents: the area in the sample multi-category tissue spatial distribution data that has a high contribution to the prognostic index; A potential marker is determined according to the two-dimensional contribution heat map and the spatial distribution data of the multi-category tissues of the sample, wherein the potential marker represents: at least one category of tissues and the distribution relationship between the tissues.

2. The method according to claim 1, characterized in that Process the sample pathological slice images to obtain the sample multi-category tissue spatial distribution data, including: Slice the sample pathological slice image to obtain multiple small block images; Classifying the tissues included in the plurality of small-block images respectively, and marking the tissue categories of the small-block images; According to the spatial position information of multiple small-block images, the tissue categories of the multiple small-block images are spliced ​​to form sample multi-category tissue spatial distribution data.

3. The method according to claim 1, characterized in that The method further comprises: Testing the potential markers, and comparing the prognostic significance of the potential markers with existing prognostic markers; Potential markers that show significant results during the screening test are used as new markers.

4. The method according to claim 3, characterized in that Testing the potential markers includes: The COX multivariate regression analysis method was used to obtain the risk ratio and significance of the potential markers.

5. The method according to any one of claims 1 to 4, characterized in that: Determining potential markers according to the two-dimensional contribution heat map and the multi-category tissue spatial distribution data of the sample includes: Superimposing the two-dimensional contribution heat map with the multi-category tissue spatial distribution data of the sample to obtain a superimposed image; Potential markers are determined based on the overlaid images.

6. The method according to claim 5, characterized in that The preset model includes: a convolution layer for processing multi-channel input images and a batch normalization layer for stabilizing prediction results; the loss function of the preset model includes: prognosis accuracy loss and regularization loss.

7. A cancer prognosis marker mining device, characterized in that: The device comprises: The sample processing module is used to process the sample pathological slice image to obtain the sample multi-category tissue spatial distribution data; A training module is used to train a preset model using the sample multi-category tissue spatial distribution data and the corresponding prognostic index labels to obtain a prognostic model, and the trained prognostic model is used to predict the corresponding prognostic index according to the input multi-category tissue spatial distribution data; An attribution module is used to obtain a two-dimensional contribution heat map of the multi-category tissue spatial distribution data of the sample according to the prognostic model and the multi-category tissue spatial distribution data of the sample, comprising: using a gradient-based attribution method to generate a two-dimensional contribution heat map by calculating the gradient of the prognostic index relative to the tissue classification of each small block image included in the input multi-category tissue spatial distribution data; the two-dimensional contribution heat map represents the area in the multi-category tissue spatial distribution data of the sample that has a high contribution to the prognostic index; A potential marker determination module is used to determine potential markers based on the two-dimensional contribution heat map and the sample multi-category tissue spatial distribution data.

8. The device according to claim 7, characterized in that The sample processing module comprises: A sub-slicing module is used to slice the sample pathological slice image into multiple small pieces of images; A classification submodule, used to classify the tissues included in the plurality of small-block images respectively, and mark the tissue categories of the small-block images; The splicing submodule is used to splice the tissue categories of multiple small-block images according to the spatial position information of the multiple small-block images to form sample multi-category tissue spatial distribution data.

9. A server comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the cancer prognosis marker mining method according to any one of claims 1 to 6 are 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 steps of the cancer prognosis marker mining method according to any one of claims 1 to 6 are implemented.

Images