Program, method for generating neural network model, information processing method, and information processing apparatus

Abstract

To provide a program that can output information on lymph node metastasis of stomach cancer by using a neural network model to reduce the burden on a patient's body.SOLUTION: A program according to one aspect causes a computer to execute processing of: acquiring, for a stomach cancer specimen classified to endoscopic curability C-2 through endoscopic submucosal dissection (ESD), the lesion size, the lesion invasive depth, the presence or absence of lymph vessel invasion, the presence or absence of vein invasion, the main tissue type, and the presence or absence of a tissue mixing type; and inputting the acquired lesion size, lesion invasive depth, presence or absence of lymph vessel invasion, presence or absence of vein invasion, main tissue type, and presence or absence of a tissue mixing type as explanatory variables, to a neural network model 171 trained to output information on lymph node metastasis when the lesion size, the lesion invasive depth, the presence or absence of lymph vessel invasion, the presence or absence of vein invasion, the main tissue type, and the presence or absence of a tissue mixing type are input thereto, thereby outputting information on the lymph node metastasis.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a program, a neural network model generating method, an information processing method, and an information processing device. [Background technology]

[0002] In the medical field, there is a technique for determining lymph node metastasis of gastric cancer. Patent Document 1 discloses a method for determining lymph node metastasis of gastric cancer, which quantifies the mRNA of cytokeratin 19 in a detection sample prepared using lymph node tissue suspected of metastasis of gastric cancer, and determines the metastasis of gastric cancer to lymph nodes based on the quantitative value of the mRNA. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2008-194028 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the invention of Patent Document 1 requires collection of a sample containing lymph node tissue from a patient with stomach cancer, which imposes a large physical burden on the patient.

[0005] One aspect of the present invention is to provide a program or the like capable of outputting information regarding lymph node metastasis of gastric cancer using a neural network model in order to reduce the physical burden on the patient. [Means for solving the problem]

[0006] A program relating to one aspect acquires the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type of a gastric cancer specimen classified as endoscopic cure level C-2 (eCuraC2) through endoscopic submucosal dissection (ESD), and causes a computer to execute a process of outputting information regarding lymph node metastasis by inputting the acquired lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type as explanatory variables into a neural network model trained to output information regarding lymph node metastasis when the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type are input. Effect of the Invention

[0007] In one aspect, the neural network model is capable of outputting information regarding lymph node metastasis of gastric cancer. [Brief description of the drawings]

[0008] [Figure 1] FIG. 2 is a block diagram showing an example of the configuration of a computer. [Diagram 2] FIG. 11 is an explanatory diagram showing an example of the record layout of a training data DB and a prediction result DB. [Diagram 3] FIG. 13 is an explanatory diagram for explaining a lymph node metastasis information output model. [Figure 4] 13 is a flowchart showing the steps of a process for generating a lymph node metastasis information output model. [Diagram 5] FIG. 13 is an explanatory diagram showing an example of a lymph node metastasis prediction screen. [Figure 6] FIG. 13 is an explanatory diagram showing an example of a lymph node metastasis prediction result screen. [Figure 7] 13 is a flowchart showing a processing procedure for outputting information regarding lymph node metastasis. [Figure 8] FIG. 11 is an explanatory diagram showing an example of the record layout of a training data DB and a prediction result DB in embodiment 2. [Figure 9]FIG. 11 is an explanatory diagram for explaining a lymph node metastasis information output model in the second embodiment. [Figure 10] FIG. 11 is an explanatory diagram showing an example of a lymph node metastasis prediction screen in the second embodiment. [Figure 11] FIG. 11 is an explanatory diagram showing an example of a lymph node metastasis prediction result screen in embodiment 2. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, the present invention will be described in detail with reference to the drawings showing embodiments thereof.

[0010] (Embodiment 1) The first embodiment relates to an embodiment in which information regarding lymph node metastasis is output by a neural network model.

[0011] The system of this embodiment includes an information processing device 1. The information processing device 1 is an information processing device that processes, stores, and transmits / receives various types of information. The information processing device 1 is, for example, a server device, a personal computer, a general-purpose tablet PC (personal computer), a smartphone, a mobile phone, a wearable device such as an Apple Watch (registered trademark), or an information processing device such as a tablet. For the sake of simplicity, the information processing device 1 will be read as a computer 1 below.

[0012] The computer 1 according to this embodiment acquires the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type of a gastric cancer specimen classified as endoscopic cure level C-2 through endoscopic submucosal dissection. The computer 1 outputs information related to lymph node metastasis by inputting the acquired lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type as explanatory variables into a lymph node metastasis information output model (neural network model) 171 described later.

[0013] 1 is a block diagram showing an example of the configuration of a computer 1. The computer 1 includes a control unit 11, a storage unit 12, a communication unit 13, an input unit 14, a display unit 15, a reading unit 16, and a large-capacity storage unit 17. Each component is connected by a bus B.

[0014] The control unit 11 includes an arithmetic processing device such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), or a quantum processor, and performs various information processing, control processing, etc. related to the computer 1 by reading and executing a control program 1P (program product) stored in the memory unit 12.

[0015] The control program 1P can be deployed to be executed on a single computer, or on multiple computers located at one site, or distributed across multiple sites and interconnected by a communications network. Although the control unit 11 is described in FIG. 1 as being a single processor, it may be a multiprocessor.

[0016] The storage unit 12 includes memory elements such as a RAM (Random Access Memory) and a ROM (Read Only Memory), and stores the control program 1P or data required for the control unit 11 to execute processing. The storage unit 12 also temporarily stores data required for the control unit 11 to execute arithmetic processing. The communication unit 13 is a communication module for performing processing related to communication.

[0017] The input unit 14 may be a keyboard, a mouse, or a touch panel integrated with the display unit 15. The display unit 15 is a liquid crystal display, an organic EL (electroluminescence) display, or the like, and displays various information according to instructions from the control unit 11.

[0018] The reading unit 16 reads the portable storage medium 1a including a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc)-ROM. The control unit 11 may read the control program 1P from the portable storage medium 1a via the reading unit 16 and store it in the mass storage unit 17. The control unit 11 may also download the control program 1P from another computer via a network or the like and store it in the mass storage unit 17. Furthermore, the control unit 11 may read the control program 1P from the semiconductor memory 1b.

[0019] The mass storage unit 17 includes a recording medium such as a hard disk drive (HDD), a solid state drive (SSD), etc. The mass storage unit 17 includes a lymph node metastasis information output model 171, a training data DB 172, and a prediction result DB 173.

[0020] The lymph node metastasis information output model 171 is an output device (predictor) that outputs information related to lymph node metastasis based on a plurality of explanatory variables, and is a trained model generated by deep learning. The explanatory variables will be described later. The training data DB 172 stores training data for constructing (generating) the lymph node metastasis information output model 171. The prediction result DB 173 stores the results predicted by the lymph node metastasis information output model 171.

[0021] In this embodiment, the storage unit 12 and the large-capacity storage unit 17 may be configured as an integrated storage device. The large-capacity storage unit 17 may be configured with a plurality of storage devices. Furthermore, the large-capacity storage unit 17 may be an external storage device connected to the computer 1.

[0022] The computer 1 may execute various information processing and control processing, etc., as a single computer, or may execute the processing in a distributed manner among multiple computers. Furthermore, the computer 1 may be realized by multiple virtual machines provided in one computer, or may be realized by using a cloud server.

[0023] FIG. 2 is an explanatory diagram showing an example of the record layout of the training data DB 172 and the prediction result DB 173.

[0024] The training data DB 172 includes a training ID string, an input data string, and a transition presence / absence string. The training ID string stores a unique ID of each piece of training data in order to identify each piece of training data.

[0025] The input data column stores a plurality of explanatory variables and the values ​​of each explanatory variable. The explanatory variables include the lesion size, the lesion depth, the presence or absence of lymphatic invasion, the presence or absence of venous invasion, the main tissue type, and the presence or absence of mixed tissue type of a gastric cancer specimen classified as having endoscopic cure degree C-2 through endoscopic submucosal dissection. The presence or absence of metastasis column stores information on the presence or absence of lymph node metastasis, including the presence of metastasis indicating positive, or the absence of metastasis indicating negative.

[0026] The prediction result DB 173 includes a number sequence, a lesion size sequence, a lesion depth sequence, a lymphatic invasion presence / absence sequence, a venous invasion presence / absence sequence, a main tissue type sequence, a tissue mixture type presence / absence sequence, and a probability value sequence. The number sequence stores a uniquely specified number of prediction result data in order to identify each prediction result data.

[0027] The lesion size column stores the size of the lesion in the gastric cancer specimen. The lesion invasion depth column stores a value corresponding to deep submucosal invasive cancer, mild submucosal invasive cancer, or mucosal invasive cancer as the lesion invasion depth. For example, the value corresponding to deep submucosal invasive cancer is 1, the value corresponding to mild submucosal invasive cancer is 0.5, and the value corresponding to mucosal invasive cancer is 0.

[0028] The lymphatic invasion presence / absence column stores information on the presence or absence of lymphatic invasion. For example, the presence of lymphatic invasion is set to 1, and the absence of lymphatic invasion is set to 0. The venous invasion presence / absence column stores information on the presence or absence of venous invasion. For example, the presence of venous invasion is set to 1, and the absence of venous invasion is set to 0. The main tissue type column stores values ​​corresponding to moderately differentiated adenocarcinoma (tub2) or other tissue types. For example, a value corresponding to moderately differentiated adenocarcinoma is set to 1, and a value corresponding to other tissue types is set to 0.

[0029] The tissue mixing type presence / absence column stores information on the presence / absence of a tissue mixing type. For example, the presence of a tissue mixing type is set to 1, and the absence of a tissue mixing type (single tissue type) is set to 0. The probability value column stores the probability value of lymph node metastasis predicted by the lymph node metastasis information output model 171.

[0030] The above-mentioned storage formats of each DB are merely examples, and other storage formats may be used as long as the relationships between the data are maintained.

[0031] FIG. 3 is an explanatory diagram for explaining the lymph node metastasis information output model 171. The lymph node metastasis information output model 171 is used as a program module that is a part of artificial intelligence software. The lymph node metastasis information output model 171 is a learning model that outputs information about lymph node metastasis when multiple explanatory variables are input. The explanatory variables include the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of tissue mixed type of a gastric cancer specimen classified as endoscopic cure level C-2 through endoscopic submucosal dissection.

[0032] The lymph node metastasis information output model 171 of this embodiment performs output processing of information on lymph node metastasis using a DNN (Deep Neural Network(s)). The lymph node metastasis information output model 171 has an input layer that accepts input of multiple explanatory variables, an output layer that outputs information on lymph node metastasis, and multiple intermediate layers (hidden layers) that extract features of the multiple explanatory variables.

[0033] The input layer has a plurality of neurons (nodes; units) that accept input of a plurality of explanatory variables, and passes the input explanatory variables to the intermediate layer. Each intermediate layer has a plurality of neurons, and extracts features of the plurality of explanatory variables. For example, the lymph node metastasis information output model 171 has a first intermediate layer and a second intermediate layer. The first intermediate layer may have 6 neurons, and the second intermediate layer may have 18 neurons.

[0034] The intermediate layer acquires effective features by passing the explanatory variables input from the input layer through multiple intermediate layers (e.g., fully connected layers) and using, for example, a sigmoid function, a tanh function, or a ReLu (Rectified Linear Unit) function as an activation function connecting the neurons in each layer.

[0035] The output layer includes, for example, a sigmoid function or a softmax function, and outputs information about predicted lymph node metastasis based on the feature values ​​output from the intermediate layer. The information about lymph node metastasis includes, for example, a probability value of lymph node metastasis of early gastric cancer, a risk level classified according to the probability value of lymph node metastasis, or information about the presence or absence of lymph node metastasis.

[0036] The probability value of lymph node metastasis is, for example, a continuous probability value (for example, a value ranging from "0" to "1") indicating the rate of lymph node metastasis. The risk levels include, for example, low risk where the probability value of lymph node metastasis is less than 0.25, medium risk where the probability value of lymph node metastasis is 0.25% or more and less than 0.65, and high risk where the probability value of lymph node metastasis is 0.65 or more. The information on the presence or absence of lymph node metastasis includes the presence of metastasis (positive) or the absence of metastasis (negative), and for example, the presence of metastasis is represented as 1 and the absence of metastasis is represented as 0.

[0037] In this embodiment, an example is described in which the information regarding lymph node metastasis output from the lymph node metastasis information output model 171 is a probability value of lymph node metastasis, but the same can be applied to information regarding other types of lymph node metastasis.

[0038] The computer 1 performs learning using the training data stored in the training data DB 172. The training data is generated based on a large amount of data set collected from past cases at medical institutions. The training data may be data that is separately created manually. The training data is a combination of multiple explanatory variables and information on lymph node metastasis associated with each other.

[0039] The information on lymph node metastasis includes, for example, a probability value of lymph node metastasis, a risk level classified according to the probability value of lymph node metastasis, or information on the presence or absence of lymph node metastasis, etc. Note that in this embodiment, data in which a plurality of explanatory variables are associated with information on the presence or absence of lymph node metastasis (with or without metastasis) is used as training data and will be described as an example, but the same can be applied to other types of information on lymph node metastasis.

[0040] Each record in the training data DB 172 is training data. Each explanatory variable in the input data column and the value of each explanatory variable are input data. The values ​​in the transition column are correct answer data (transition or no transition) for each explanatory variable and the value of each explanatory variable.

[0041] The computer 1 inputs multiple explanatory variables to the input layer, and obtains a predicted probability value of lymph node metastasis from the output layer after arithmetic processing in the intermediate layer. The computer 1 compares the probability value of lymph node metastasis output from the output layer with the correct value in the training data, and optimizes the parameters used in the arithmetic processing in the intermediate layer so that the output value from the output layer approaches the correct value.

[0042] The parameters are, for example, weights (coupling coefficients) between neurons, etc. Although there are no particular limitations on the method of optimizing the parameters, for example, the computer 1 optimizes various parameters using the error backpropagation method.

[0043] The computer 1 performs the above process for each record stored in the training data DB 172, and performs learning of the lymph node metastasis information output model 171. This makes it possible to construct a model capable of outputting a probability value of lymph node metastasis. Note that the above learning process may be performed by another computer (not shown), and the lymph node metastasis information output model 171 may be deployed.

[0044] When the computer 1 acquires multiple explanatory variables, it inputs the acquired multiple explanatory variables to the lymph node metastasis information output model 171. The computer 1 performs a calculation process to extract features of the multiple explanatory variables in the intermediate layer of the lymph node metastasis information output model 171. The computer 1 inputs the extracted features to the output layer of the lymph node metastasis information output model 171, and outputs a probability value of lymph node metastasis.

[0045] As shown in the figure, a probability value of lymph node metastasis of "0.85" is output for input data which are explanatory variables including the lesion size of the gastric cancer specimen, the depth of lesion, the presence or absence of lymphatic invasion, the presence or absence of venous invasion, the main tissue type, and the presence or absence of mixed tissue type.

[0046] The classification result for lymph node metastasis (e.g., metastasis and no metastasis) may be output from the lymph node metastasis information output model 171. For example, for input data that is a plurality of explanatory variables, a prediction result in which the probability values ​​of "metastasis" and "no metastasis" are "0.85" and "0.15", respectively, are output.

[0047] Also, a predetermined threshold value may be used to output the prediction result. For example, when the computer 1 determines that the probability value of "metastasis" (0.85) is equal to or greater than a predetermined threshold value (e.g., 0.80), the computer 1 outputs the probability value of "metastasis" as the prediction result. Note that, without using the above-mentioned threshold value, the classification result corresponding to the highest probability value may be output as the prediction result from among the probability values ​​of each classification result predicted by the lymph node metastasis information output model 171.

[0048] In addition, the lymph node metastasis information output model 171 is not limited to DNN, and may be realized by other models such as a multi-layer perceptron (MLP), a recurrent neural network (RNN), a decision tree model, a random forest model, a k-nearest neighbor method, a binary tree search, a Bayesian network, a regression tree, a logistic regression, a support vector machine (SVM), a k-Nearest Neighbor algorithm (k-NN), a naive Bayes classifier, or a Transformer network.

[0049] 4 is a flowchart showing the procedure of the generation process of the lymph node metastasis information output model 171. The control unit 11 of the computer 1 acquires multiple training data sets of combinations in which multiple explanatory variables (lesion size of a gastric cancer specimen, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type, etc.) are associated with information on lymph node metastasis (e.g., presence or absence of metastasis) from the training data DB 172 of the mass storage unit 17 (step S101).

[0050] Using the acquired training data, the control unit 11 generates a lymph node metastasis information output model 171 that receives multiple explanatory variables as input and outputs information related to lymph node metastasis (step S102). The control unit 11 stores the generated lymph node metastasis information output model 171 in the storage unit 12 or the mass storage unit 17 (step S103), and ends the series of processes.

[0051] 5 is an explanatory diagram showing an example of a lymph node metastasis prediction screen. The screen includes a lesion size input field 11a, a lesion invasion depth selection field 11b, a main tissue type selection field 11c, a tissue mixture type selection field 11d, a lymphatic invasion selection field 11e, a venous invasion selection field 11f, and a prediction button 11g.

[0052] The lesion size input field 11a is an input field for accepting input of the lesion size of the gastric cancer specimen. The lesion depth selection field 11b is an icon for selecting the lesion depth as deep submucosal invasive cancer (SM2 or deeper), mild submucosal invasive cancer (SM1), or mucosal invasive cancer (M).

[0053] The main tissue type selection field 11c is an icon for selecting moderately differentiated adenocarcinoma (tub2) or other tissue type (Other). The mixed tissue type selection field 11d is an icon for selecting mixed tissue type (Mixed type) or pure tissue type (Pure type). The lymphatic vessel invasion selection field 11e is an icon for selecting lymphatic vessel invasion (Positive) or no lymphatic vessel invasion (Negative).

[0054] The venous invasion selection field 11f is an icon for selecting "there is venous invasion (Positive)" or "there is no venous invasion (Negative)." The prediction button 11g is a button for outputting information on lymph node metastasis using the lymph node metastasis information output model 171.

[0055] When the computer 1 receives an input operation in the lesion size input field 11a, the computer 1 acquires the input lesion size.

[0056] When the computer 1 receives a selection operation in the lesion depth selection field 11b, it acquires a value corresponding to the selected deep submucosal invasive cancer, mild submucosal invasive cancer, or mucosal invasive cancer. For example, the value corresponding to deep submucosal invasive cancer is set to 1, the value corresponding to mild submucosal invasive cancer is set to 0.5, and the value corresponding to mucosal invasive cancer is set to 0.

[0057] When the computer 1 receives a selection operation in the main tissue type selection field 11c, it acquires a value corresponding to the selected moderately differentiated adenocarcinoma or other tissue type. For example, the presence of mixed tissue type is set to 1, and the absence of mixed tissue type (single tissue type) is set to 0.

[0058] When the computer 1 receives a selection operation in the tissue mixture type selection field 11d, it acquires a value corresponding to the selected tissue mixture type or single tissue type. For example, the tissue mixture type is set to 1, and the single tissue type is set to 0.

[0059] When the computer 1 receives a selection operation in the lymphatic vessel invasion selection field 11e, the computer 1 obtains a value corresponding to whether lymphatic vessel invasion is present or not. For example, the computer 1 may set the value to 1 for lymphatic vessel invasion and the value to 0 for no lymphatic vessel invasion.

[0060] When the computer 1 receives a selection operation in the venous invasion selection field 11f, the computer 1 acquires a value corresponding to whether or not there is venous invasion. For example, the computer 1 may set the value to 1 for venous invasion and 0 for no venous invasion.

[0061] When the computer 1 receives a touch operation of the prediction button 11g, it inputs the lesion size input in the lesion size input field 11a, the lesion depth selected in the lesion depth invasion selection field 11b, the main tissue type selected in the main tissue type selection field 11c, the mixed tissue type selected in the mixed tissue type selection field 11d, the presence or absence of lymphatic invasion selected in the lymphatic invasion selection field 11e, and the presence or absence of venous invasion selected in the venous invasion selection field 11f as explanatory variables into the lymph node metastasis information output model 171.

[0062] The computer 1 uses the lymph node metastasis information output model 171 to output information on lymph node metastasis based on a plurality of input explanatory variables. The computer 1 stores the output information on lymph node metastasis in the prediction result DB 173. Specifically, the computer 1 assigns a number to the prediction result. The computer 1 stores information on the lesion size, the lesion depth, the presence or absence of lymph node invasion, the presence or absence of venous invasion, the main tissue type, the presence or absence of tissue mixed type, and lymph node metastasis (e.g., the probability value of lymph node metastasis) in the prediction result DB 173 in association with the assigned number.

[0063] The computer 1 displays the information on lymph node metastasis output from the lymph node metastasis information output model 171 on a screen (FIG. 6, which will be described later).

[0064] 6 is an explanatory diagram showing an example of a lymph node metastasis prediction result screen. The screen includes a metastasis risk display field 12a and a contribution degree display field 12b. The metastasis risk display field 12a is a display field that displays the metastasis risk as information on lymph node metastasis (e.g., the probability value of lymph node metastasis). The contribution degree display field 12b is a display field that displays the explanatory variables (items) that contributed to the metastasis risk result and the contribution degree of each explanatory variable.

[0065] The computer 1 displays, as a lymph node metastasis risk, the information on lymph node metastasis output from the lymph node metastasis information output model 171 in the metastasis risk display field 12a. As shown in the figure, the lymph node metastasis risk is 0.85.

[0066] For example, the computer 1 calculates the contribution (degree of influence; degree of contribution) to the lymph node metastasis risk, which is the output data, using the Shapley Value, based on a plurality of explanatory variables included in the input data.

[0067] SHAP (SHapley Additive exPlanations) is a method for evaluating the validity of prediction grounds, and is a method for visualizing the degree of influence that each input value (feature value) to the lymph node metastasis information output model 171, which is a predictor, has on the prediction result. In this embodiment, the input values ​​are multiple explanatory variables (names and values ​​of explanatory variables), and the degree of influence that has been given to the prediction result of metastasis risk is the contribution degree of the explanatory variables that contributed to the result of metastasis risk.

[0068] For application to the lymph node metastasis information output model 171, for example, the contribution of feature quantity X=(X1, X2, X3, X4, X5, X6) to the predicted value is calculated using the Shapley value. X1, X2, X3, X4, X5, and X6 indicate explanatory variables.

[0069] For example, X1 is the lesion size, X2 is the lesion depth, X3 is the main tissue type, X4 is the presence or absence of mixed tissue type, X5 is the presence or absence of lymph node invasion, and X6 is the presence or absence of venous invasion. Here, an example of six explanatory variables has been described, but the number of explanatory variables is arbitrary and is not limited to six. For example, the treatment method in the second embodiment described later may be included in the explanatory variables.

[0070] Let the model be f(X) and the average predicted value be E[f(X)]. Let each instance have features (x1, x2, x3, x4, x5, x6) = x, and let the predicted value at this time be f(x). It is possible to find how much each feature affects the deviation between the average predicted value E[f(X)] and the predicted value f(x) of each instance.

[0071] Specifically, the predicted value f(x) of each instance is E[f(X|X1=x1,X2=x2,X3=x3,X4=x4,X5=x5,X6=x6)]=f(x1,x2,x3,x4,x5,x6)=f(x), so by conditioning the average predicted value from E[f(X)] with X1,X2,X3,X4,X5,X6, we can determine how knowing the features affects the prediction of each instance. Note that the order of conditioning is X1,X2,X3,X4,X5,X6, but this is not the only possible order.

[0072] X1, X2, X3, X4, X5, and X6 are added in that order, and then conditioning is performed on the combinations of X1, X2, X3, X4, X5, and X6, and on all of the permutations of X1, X2, X3, X4, X5, and X6. By calculating the average of the marginal effect (contribution) that each of the features obtained in each case has on the predicted value, the contribution of the explanatory variables that contributed to the metastasis risk results can be calculated.

[0073] For example, a linear regression method such as LIME (Local Interpretable Model-agnostic Explanations) may be applied to the lymph node metastasis information output model 171 to calculate the contribution of the explanatory variables that contributed to the result of the metastasis risk.

[0074] In this way, the computer 1 inputs the multiple explanatory variables acquired to the lymph node metastasis information output model 171 and outputs information on lymph node metastasis as metastasis risk. Based on the multiple explanatory variables that are input data, the computer 1 uses a Shapley library or the like to identify explanatory variables that contributed to the metastasis risk result obtained by the lymph node metastasis information output model 171. The computer 1 displays the identified explanatory variables and the contribution of the explanatory variables in the contribution display field 12b.

[0075] 7 is a flowchart showing a processing procedure for outputting information on lymph node metastasis. The control unit 11 of the computer 1 displays an object for receiving a plurality of explanatory variables on the display unit 15 (step S111). The explanatory variables include lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of tissue mixture type.

[0076] For example, the object includes an input field for accepting input of lesion size (lesion size input field 11a), an icon for selecting the lesion depth (deep submucosal invasive carcinoma, mildly submucosal invasive carcinoma, or mucosal invasive carcinoma) (lesion depth invasion selection field 11b), an icon for selecting the main tissue type (moderately differentiated adenocarcinoma or other tissue type) (main tissue type selection field 11c), an icon for selecting the presence or absence of mixed tissue type (mixed tissue type or single tissue type) (mixed tissue type selection field 11d), an icon for selecting the presence or absence of lymphatic invasion (with lymphatic invasion or without lymphatic invasion) (lymphatic invasion selection field 11e), and an icon for selecting the presence or absence of venous invasion (with venous invasion or without venous invasion) (venous invasion selection field 11f).

[0077] The control unit 11 receives input or selection of multiple explanatory variables from the input unit 14 through the object (step S112). The control unit 11 inputs the received multiple explanatory variables to the lymph node metastasis information output model 171 (step S113). The control unit 11 uses the lymph node metastasis information output model 171 to output information on lymph node metastasis as a metastasis risk based on the input multiple explanatory variables (step S114).

[0078] The control unit 11 uses the Shapley library or the like based on the multiple explanatory variables that are input data to identify explanatory variables that contributed to the metastasis risk result from the lymph node metastasis information output model 171 (step S115). The control unit 11 displays the metastasis risk and the explanatory variables that contributed to the identified metastasis risk result on the display unit 15 (step S116).

[0079] The control unit 11 stores the information on lymph node metastasis output from the lymph node metastasis information output model 171 in the prediction result DB 173 of the mass storage unit 17 (step S117). Specifically, the control unit 11 assigns a number to the prediction result. The computer 1 stores the lesion size, the lesion depth, the presence or absence of lymph node invasion, the presence or absence of venous invasion, the main tissue type, the presence or absence of tissue mixed type, and information on lymph node metastasis (for example, the probability value of lymph node metastasis) in the prediction result DB 173 in association with the assigned number. The control unit 11 ends the process.

[0080] According to this embodiment, it is possible to generate a lymph node metastasis information output model 171 based on training data including information on gastric cancer lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type, as well as information on lymph node metastasis.

[0081] According to this embodiment, using the lymph node metastasis information output model 171, it is possible to output information regarding lymph node metastasis.

[0082] According to this embodiment, it is possible to output the metastasis risk as information relating to lymph node metastasis, and the explanatory variables that contributed to the result of the metastasis risk. (Embodiment 2) The second embodiment relates to a form in which information regarding lymph node metastasis is output when a treatment method is further input as an explanatory variable to the lymph node metastasis information output model 171. Note that a description of the contents that overlap with the first embodiment will be omitted.

[0083] The lymph node metastasis information output model 171 in this embodiment is trained to output information on lymph node metastasis when a treatment method is further input as an explanatory variable. The treatment method includes surgery, ESD, and both ESD and additional surgery after ESD (ESD+surgery).

[0084] Fig. 8 is an explanatory diagram showing an example of the record layout of the training data DB 172 and the prediction result DB 173 in the embodiment 2. Note that a description of the contents that overlap with Fig. 2 will be omitted.

[0085] The training data DB 172 of the present embodiment is created based on a large amount of data sets collected from surgical cases, ESD cases, and ESD+ surgical cases. For example, the training data DB 172 may be created based on data sets collected from 2970 surgical cases, 414 ESD cases, and 122 ESD+ surgical cases.

[0086] Surgical cases are cases where endoscopic treatment was not possible or where surgery was required due to the depth of the gastric cancer, and surgical procedures such as gastrectomy or combined resection were performed. ESD cases are cases where endoscopic treatment was possible but only endoscopic submucosal dissection was performed. ESD + surgical cases are cases where gastric cancer could not be completely removed, or where the pathological diagnosis after ESD showed a pathologically non-curative resection, and therefore additional surgery with lymph node dissection was recommended, and therefore surgical procedures such as lymph node dissection were performed after ESD.

[0087] The input data string of the training data DB 172 stores a plurality of explanatory variables and the values ​​of each explanatory variable. The explanatory variables include the size of the lesion in the gastric cancer specimen, the depth of the lesion, the presence or absence of lymphatic invasion, the presence or absence of venous invasion, the main tissue type, the presence or absence of tissue mixture type, and the treatment method.

[0088] As shown in the figure, as an example, training data with a training ID of "00001" is training data for a surgical case, training data with a training ID of "00002" is training data for an ESD case, and training data with a training ID of "00003" is training data for an ESD+surgical case.

[0089] The prediction result DB 173 includes a treatment method column. The treatment method column stores values ​​corresponding to surgery, ESD, or ESD+surgery. For ESD or ESD+surgery, a common value corresponding to ESD or ESD+surgery is stored in the treatment method column. For example, the value corresponding to surgery is set to 1, and the common value corresponding to ESD or ESD+surgery is set to 0.

[0090] Fig. 9 is an explanatory diagram for explaining the lymph node metastasis information output model 171 in the embodiment 2. Note that the same reference numerals are used for the contents that overlap with Fig. 3, and the explanation thereof will be omitted.

[0091] The computer 1 performs learning using the training data stored in the training data DB 172. The training data is a combination of multiple explanatory variables and information on lymph node metastasis in association with each other. The explanatory variables in this embodiment include the lesion size of the gastric cancer specimen, the lesion depth, the presence or absence of lymphatic invasion, the presence or absence of venous invasion, the main tissue type, the presence or absence of mixed tissue type, and the treatment method. Note that the information on lymph node metastasis is the same as in the first embodiment, and therefore will not be described.

[0092] The computer 1 acquires a plurality of training data from the training data DB 172. The computer 1 determines the type of treatment method (surgery, ESD, or ESD+surgery) included in each acquired training data. If the type of treatment method is surgery, the computer 1 sets the value of the explanatory variable representing the treatment method to 1. If the type of treatment method is ESD or ESD+surgery, the computer 1 sets the value of the explanatory variable representing the treatment method to 0.

[0093] When the explanatory variable value, which is the treatment method, is stored in the training data DB 172 in a numerical format of "0" or "1", the above-mentioned judgment process can be omitted. The computer 1 performs a learning process of the lymph node metastasis information output model 171 using the acquired training data. The learning process of the lymph node metastasis information output model 171 is the same as that of the first embodiment, and therefore a description thereof will be omitted.

[0094] When the computer 1 acquires multiple explanatory variables, it inputs the acquired multiple explanatory variables to the lymph node metastasis information output model 171. The computer 1 performs a calculation process to extract features of the multiple explanatory variables in the intermediate layer of the lymph node metastasis information output model 171. The computer 1 inputs the extracted features to the output layer of the lymph node metastasis information output model 171, and outputs a probability value of lymph node metastasis.

[0095] As shown in the figure, the probability value of lymph node metastasis, "0.89", is output for explanatory variables including the size of the lesion in the gastric cancer specimen, the depth of the lesion, the presence or absence of lymphatic invasion, the presence or absence of venous invasion, the main tissue type, the presence or absence of mixed tissue type, and the treatment method.

[0096] The flowchart showing the procedure of the generation process of the lymph node metastasis information output model 171 based on training data including a treatment method is similar to the flowchart shown in FIG. 4, and therefore the description thereof will be omitted.

[0097] Fig. 10 is an explanatory diagram showing an example of a lymph node metastasis prediction screen in embodiment 2. Note that the same reference numerals are used for the contents that overlap with Fig. 5, and the description thereof will be omitted. This screen includes a treatment method selection field 11h. The treatment method selection field 11h is an icon for selecting either surgery, ESD, or ESD + surgery.

[0098] When the computer 1 receives a selection operation in the treatment method selection field 11h, it acquires a value corresponding to the selected surgery, ESD, or ESD+surgery. For example, the value corresponding to surgery is set to 1, the value corresponding to ESD is set to 0, and the value corresponding to ESD+surgery is set to 0.

[0099] When the computer 1 receives a touch operation of the prediction button 11g, it inputs the lesion size input in the lesion size input field 11a, the lesion depth selected in the lesion depth invasion selection field 11b, the main tissue type selected in the main tissue type selection field 11c, the mixed tissue type selected in the mixed tissue type selection field 11d, the presence or absence of lymphatic invasion selected in the lymphatic invasion selection field 11e, the presence or absence of venous invasion selected in the venous invasion selection field 11f, and the treatment method selected in the treatment method selection field 11h as explanatory variables into the lymph node metastasis information output model 171.

[0100] The computer 1 outputs information on lymph node metastasis based on a plurality of input explanatory variables using the lymph node metastasis information output model 171. The computer 1 displays the information on lymph node metastasis output from the lymph node metastasis information output model 171 on a screen (FIG. 11 described later).

[0101] Fig. 11 is an explanatory diagram showing an example of a lymph node metastasis prediction result screen in embodiment 2. Note that the same reference numerals are used for the contents that overlap with Fig. 6, and the description thereof will be omitted.

[0102] The computer 1 displays, as a lymph node metastasis risk, the information on lymph node metastasis output from the lymph node metastasis information output model 171 in the metastasis risk display field 12a. As shown in the figure, the lymph node metastasis risk is 0.89.

[0103] Based on a plurality of explanatory variables that are input data, the computer 1 uses the Shapley library or the like to identify explanatory variables that contributed to the metastasis risk result obtained by the lymph node metastasis information output model 171. The computer 1 displays the identified explanatory variables and the contributions of the explanatory variables in the contribution display field 12b.

[0104] According to this embodiment, it is possible to output information on lymph node metastasis by using a lymph node metastasis information output model 171 that is trained based on explanatory variables including a treatment method.

[0105] The embodiments disclosed herein are illustrative in all respects and should not be considered as limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the scope and meaning equivalent to the claims.

[0106] The matters described in each embodiment can be combined with each other. In addition, the independent claims and dependent claims described in the claims can be combined with each other in any and all combinations regardless of the citation format. Furthermore, the claims use a format in which a claim cites two or more other claims (multiple claim format), but this is not limited to this. The claims may also be written in a format in which a multiple claim cites at least one other multiple claim (multi-multi claim). [Explanation of symbols]

[0107] 1. Information processing equipment (computer) 1a Removable storage media 1b Semiconductor memory 1P control program 11 Control section 12 Storage section 13. Communications Department 14 Input section 15 Display section 16 Reading unit 17 Mass storage 171 Lymph node metastasis information output model (neural network model) 172 Training Data DB 173 Prediction result DB

Claims

1. Obtain the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main histological type, and presence or absence of mixed histological type for gastric cancer specimens classified as endoscopic cure grade C-2 (eCuraC2) through endoscopic submucosal dissection (ESD); The neural network model is trained to output information regarding lymph node metastasis when the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type are input, and information regarding lymph node metastasis is output by inputting the acquired lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type as explanatory variables. A program that causes a computer to carry out processing.

2. The neural network model is further trained to output information about lymph node metastasis when a treatment method including both surgery, ESD, and ESD and additional surgery after ESD is input as an explanatory variable. The program according to claim 1.

3. When inputting the treatment method into the neural network model, a common value corresponding to both surgeries, ESD or ESD and additional surgery after ESD, is input into the neural network model. The program according to claim 2.

4. Outputting icons for selecting one of deep submucosal invasive cancer, mild submucosal invasive cancer, and mucosal invasive cancer as the depth of invasion of the lesion; Accepting a selection of the icon; Inputting information corresponding to the selected icon into the neural network model.

3. The program according to claim 1 or 2.

5. Outputting the metastasis risk as information on the lymph node metastasis and explanatory variables that contributed to the result of the metastasis risk 3. The program according to claim 1 or 2.

6. Obtain training data including information on the size of the lesion, the depth of invasion of the lesion, the presence or absence of lymphatic invasion, the presence or absence of venous invasion, the main tissue type, the presence or absence of mixed tissue type, and treatment methods and lymph node metastasis for gastric cancer classified as having an endoscopic cure level of C-2 through endoscopic submucosal dissection; Based on the acquired training data, a neural network model is generated that outputs information regarding lymph node metastasis when the lesion size, the lesion depth, the presence or absence of lymphatic invasion, the presence or absence of venous invasion, the main tissue type, the presence or absence of tissue mixture type, and the treatment method are input. How to generate neural network models.

7. For gastric cancer specimens classified as C-2 in endoscopic cure level through endoscopic submucosal dissection, the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main histological type, and presence or absence of mixed histological type are obtained; The neural network model has been trained to output information regarding lymph node metastasis when the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type are input, and information regarding lymph node metastasis is output by inputting the acquired lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type. An information processing method in which processing is performed by a computer.

8. An information processing device including a control unit, The control unit is For gastric cancer specimens classified as C-2 in endoscopic cure level through endoscopic submucosal dissection, the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main histological type, and presence or absence of mixed histological type are obtained; The neural network model has been trained to output information regarding lymph node metastasis when the lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type are input, and information regarding lymph node metastasis is output by inputting the acquired lesion size, lesion depth, presence or absence of lymphatic invasion, presence or absence of venous invasion, main tissue type, and presence or absence of mixed tissue type. Information processing device.

Images