Prognosis prediction device, prognosis prediction method and program

Abstract

To provide a prognosis prediction device, a method for predicting a prognosis, and a program that can increase the accuracy of predicting the prognosis of a patient with a tumor.SOLUTION: A prognosis prediction system 100 includes: a prognosis prediction information acquisition device 1 for acquiring prognosis prediction information, which is graph tumor information about an image of a tumor expressed by the amount defined by a graph theory and information on the relation between a person / an animal with the tumor and the prognosis; and a prognosis prediction device 2 for predicting the prognosis of an estimation target on the basis of the graph tumor information showing an image of an estimation target tumor.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a prognosis prediction device, a prognosis prediction method, and a program. [Background technology]

[0002] In recent years, a technology known as radiomics has been researched. Radiomics is a large-scale, comprehensive technology that deals with the relationship between radiomics features, which are thousands of types of phenotypic information extracted from large amounts of medical images, and clinical information such as pathology, prognosis, and side effects. One application of radiomics that is being studied is the prognosis prediction of patients with tumor diseases.

[0003] More specifically, research is being conducted into technologies for predicting the prognosis of patients suffering from tumor diseases using tumor phenotype information, such as tumor sphericity, the square of the total intratumoral CT value, and the variability of intratumoral CT values. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2020 / 067481 [Patent Document 2] Special Publication No. 2008-522273 [Non-patent literature]

[0005] [Non-Patent Document 1] Ahmed Hosny et.al., “Deep learning for lung cancer prognosis: A retrospective multi-cohort radiomics study”, PLoS Medicine, 15(11):e1002711, Nov 30, (2018) Summary of the Invention [Problem to be solved by the invention]

[0006] However, previously proposed tumor phenotype information, such as tumor sphericity, the square of the total intratumoral CT value, and the variability of intratumoral CT values, may not be accurate, and the accuracy of the information varies greatly between institutions and patients. Additionally, it is difficult to obtain local information within the tumor from tumor phenotype information. Therefore, prognosis predictions for patients with tumor diseases using tumor phenotype information have sometimes been inaccurate.

[0007] In view of the above circumstances, an object of the present invention is to provide a technique for improving the accuracy of prognosis prediction for patients suffering from tumor diseases. [Means for solving the problem]

[0008] One aspect of the present invention is a prognosis prediction device that includes a prognosis prediction unit that predicts the prognosis of a subject to be estimated based on graph tumor information that shows an image of a tumor that the subject has, using graph tumor information, which is information on the image of a tumor expressed using quantities defined in graph theory, and prognosis prediction information, which is information that shows the relationship between the prognosis of a person or animal that has the tumor.

[0009] One aspect of the present invention is a prognosis prediction method comprising a prognosis prediction step of predicting the prognosis of a subject based on graph tumor information showing an image of a tumor possessed by the subject, using graph tumor information, which is information on the image of a tumor expressed using quantities defined in graph theory, and prognosis prediction information, which is information showing the relationship between the prognosis of a person or animal having the tumor.

[0010] One aspect of the present invention is a program for causing the above-described prognosis prediction device to function as a computer. [Effects of the Invention]

[0011] The present invention makes it possible to improve the accuracy of prognosis prediction for patients suffering from tumor diseases. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is an explanatory diagram illustrating an overview of a prognosis prediction system 100 according to an embodiment. [Figure 2] 1 is a diagram showing an example of the hardware configuration of a prognosis prediction information acquisition device 1 according to an embodiment. [Figure 3] FIG. 2 is a diagram showing an example of the functional configuration of a control unit 11 in the embodiment. [Figure 4] 1 is a flowchart showing an example of the flow of processing executed by a prognosis prediction information acquisition device 1 in an embodiment. [Figure 5] FIG. 2 is a diagram showing an example of the hardware configuration of a prognosis prediction device 2 according to an embodiment. [Figure 6] FIG. 2 is a diagram showing an example of the functional configuration of a control unit 21 in the embodiment. [Figure 7] 1 is a flowchart showing an example of the flow of processing executed by a prognosis prediction device 2 in an embodiment. [Figure 8] FIG. 1 is a first explanatory diagram illustrating an example of the relationship between a tumor image and graph tumor information in an embodiment. [Figure 9] FIG. 2 is a second explanatory diagram illustrating an example of the relationship between a tumor image and graph tumor information in the embodiment. [Figure 10] FIG. 1 is a first explanatory diagram illustrating an example of a graph amount in an embodiment. [Figure 11] FIG. 2 is a second explanatory diagram illustrating an example of a graph amount in the embodiment. [Figure 12] FIG. 10 is a third explanatory diagram illustrating an example of a graph amount in the embodiment. [Figure 13] FIG. 1 is a first explanatory diagram illustrating an experiment according to an embodiment. [Figure 14] FIG. 2 is a second explanatory diagram illustrating an experiment according to an embodiment. [Figure 15] FIG. 3 is a third explanatory diagram illustrating an experiment according to an embodiment. [Figure 16] FIG. 4 is a fourth explanatory diagram illustrating an experiment according to an embodiment. [Figure 17] FIG. 5 is a fifth explanatory diagram illustrating an experiment in the embodiment. [Figure 18]FIG. 6 is a sixth explanatory diagram illustrating an experiment in the embodiment. [Figure 19] FIG. 7 is a seventh explanatory diagram for explaining an experiment in the embodiment. [Figure 20] FIG. 8 is an eighth explanatory diagram for explaining an experiment in the embodiment. [Figure 21] FIG. 9 is a ninth explanatory diagram for explaining an experiment in the embodiment. [Figure 22] FIG. 10 is a tenth explanatory diagram illustrating an experiment in the embodiment. [Figure 23] FIG. 11 is an eleventh explanatory diagram for explaining an experiment in the embodiment. [Figure 24] FIG. 12 is a twelfth explanatory diagram illustrating an experiment in the embodiment. [Figure 25] FIG. 10 is an explanatory diagram illustrating an example of a graph amount in a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0013] (Embodiment) 1 is an explanatory diagram illustrating an overview of a prognosis prediction system 100 according to an embodiment. The prognosis prediction system 100 includes a prognosis prediction information acquisition device 1 and a prognosis prediction device 2.

[0014] The prognosis prediction information acquisition device 1 acquires information on the image of a tumor (hereinafter referred to as "tumor image") shown in an image, which is expressed using graph quantities (hereinafter referred to as "graph tumor information"), and information showing the relationship between the prognosis of a patient with a tumor disease (hereinafter referred to as "prognosis prediction information"). A patient with a tumor disease is an example of a person or animal with a tumor. A graph quantity is a quantity defined by graph theory. A graph quantity may be any quantity defined by graph theory.

[0015] Note that the image showing the tumor (hereinafter referred to as "tumor image") may be any image obtained by a technique capable of imaging the tumor. The tumor image may be, for example, a CT (computed tomography) image. The tumor image may also be, for example, an MRI (magnetic resonance imaging) image or an ultrasound image.

[0016] A graph quantity is, for example, information indicating the topology of a graph in graph theory. Specifically, information indicating the topology in graph theory is information indicating at least the positional relationship of each vertex of the graph. A graph quantity is, for example, the weight of a vertex defined in graph theory. Therefore, a graph quantity may be a quantity indicating the positional relationship between vertices whose weights are defined and the weight of each vertex. A graph quantity may be, for example, the weight of an edge defined in graph theory. A graph quantity may be, for example, the degree of a vertex defined in graph theory.

[0017] The graph quantity may be, for example, an adjacency matrix defined in graph theory. The graph quantity may be, for example, a degree matrix defined in graph theory. The graph quantity may be, for example, a connectivity matrix defined in graph theory.

[0018] The graph quantity may be, for example, an eccentricity number as defined in graph theory. The graph quantity may be, for example, a radius of a graph as defined in graph theory. The graph quantity may be, for example, a diameter of a graph as defined in graph theory.

[0019] A graph quantity may be, for example, a quantity related to a path in the graph. A graph quantity may be, for example, a quantity related to the minimum spanning tree (see Reference 1) of the graph. A graph quantity may be, for example, a quantity related to an Eulerian cycle if the graph is an Eulerian cycle. A graph quantity may be, for example, a quantity related to a Hamiltonian cycle if the graph is a Hamiltonian cycle.

[0020] Reference 1: B. Korte and J. Fiegen (translated by Takao Asano, Yasuhito Asano, Takao Ono, and Tomio Hirata), "Combinatorial Optimization, 2nd Edition: Theory and Algorithms," Maruzen Publishing, 2012, pp. 160-163

[0021] The graph tumor information may be expressed using graph quantities. Therefore, the graph tumor information does not necessarily have to be expressed only by the graph quantities themselves. The graph tumor information may be expressed, for example, by information indicating the relationship between multiple graph quantities.

[0022] The information indicating the relationship between multiple graph quantities may be, for example, information indicating whether a graph expressing tumor information (hereinafter referred to as a "tumor expression graph") is an undirected graph. The information indicating the relationship between multiple graph quantities may be, for example, information indicating whether a tumor expression graph is a directed graph. The information indicating the relationship between multiple graph quantities may be, for example, information indicating whether a tumor expression graph is a product graph. The information indicating the relationship between multiple graph quantities may be, for example, information indicating whether a tumor expression graph is a regular graph.

[0023] The information indicating the relationship between the plurality of graph quantities is, for example, information indicating whether the tumor representation graph is an Eulerian cycle. The information indicating the relationship between the plurality of graph quantities may also be, for example, information indicating whether the tumor representation graph is a Hamiltonian cycle.

[0024] The information indicating the relationship between the plurality of graph quantities may be, for example, information indicating whether an adjacency matrix indicating tumor information has predetermined matrix properties, such as whether it is a Hermitian matrix.The information indicating the relationship between the plurality of graph quantities may be, for example, information indicating whether an order matrix indicating tumor information has predetermined matrix properties, such as whether it is a Hermitian matrix.The information indicating the relationship between the plurality of graph quantities may be, for example, information indicating whether a connection matrix indicating tumor information has predetermined matrix properties, such as whether it is a Hermitian matrix.

[0025] The information indicating the relationship between the multiple graph quantities may be, for example, information indicating the difference between the tumor representation graph and a predetermined reference graph. The information indicating the difference between the tumor representation graph and the predetermined reference graph is, for example, information indicating whether the tumor representation graph and the predetermined reference graph are isomorphic.

[0026] The prognosis prediction information is, for example, a trained learning model acquired by a machine learning method. For ease of explanation, the prognosis prediction system 100 will be described below using an example in which the prognosis prediction information is a trained learning model acquired by a machine learning method.

[0027] The prognosis prediction information acquisition device 1 uses machine learning to update a learning model (hereinafter referred to as a "prognosis prediction learning model") that predicts the prognosis of a person or animal having a tumor indicated by the graph tumor information based on the graph tumor information.

[0028] A learning model is a mathematical model that includes one or more processes whose execution conditions and order (hereinafter referred to as "execution rules") are predetermined. For simplicity of explanation, updating a learning model using machine learning methods is referred to as "learning." Updating a learning model also means appropriately adjusting the parameter values ​​in the learning model. Executing a learning model also means executing each process included in the learning model in accordance with the execution rules.

[0029] The learning model is represented by, for example, a neural network. A neural network is a circuit such as an electronic circuit, an electrical circuit, an optical circuit, or an integrated circuit that represents the learning model. Updating a learning model also means that the neural network that represents the learning model is updated through learning. Updating a neural network through learning means that the values ​​of the neural network parameters are updated. The parameters of a neural network are the parameters of the circuits that make up the neural network, and are also the parameters of the learning model represented by the circuits that make up the neural network.

[0030] The neural network that represents the prognosis prediction learning model may be any neural network that can represent the prognosis prediction learning model, such as a deep neural network.

[0031] More specifically, the prognosis prediction information acquisition device 1 updates the prognosis prediction learning model by a machine learning method using, among pairs of graph tumor information and prognosis information, the graph tumor information as an explanatory variable and the prognosis information as a target label (i.e., a target variable). The prognosis information is information indicating a prognosis. Hereinafter, the pair of graph tumor information and prognosis information is referred to as a learning dataset.

[0032] The prognosis prediction information acquisition device 1 updates the prognosis prediction learning model until a predetermined termination condition (hereinafter referred to as "learning termination condition") is satisfied. The learning termination condition is, for example, a condition that learning has been performed using a predetermined number of learning data sets. The trained prognosis prediction learning model is an example of prognosis prediction information. The trained prognosis prediction learning model is the prognosis prediction learning model at the time when the learning termination condition is satisfied.

[0033] The prognosis prediction device 2 accepts image data of a tumor image (hereinafter referred to as "tumor image data"). The prognosis prediction device 2 uses prognosis prediction information to predict the prognosis of a prediction subject having a subject tumor. The subject tumor is a tumor that appears in a tumor image shown by the tumor image data. The prediction subject is a human or an animal. The prediction subject is, for example, a patient. The prognosis prediction information used by the prognosis prediction device 2 is, for example, prognosis prediction information obtained by the prognosis prediction information acquisition device 1. Therefore, the prognosis prediction information used by the prognosis prediction device 2 is, for example, a trained prognosis prediction learning model.

[0034] Pairs of tumor image data and prognosis information (hereinafter referred to as "pre-processing training datasets") are input to the prognosis prediction information acquisition device 1. The prognosis prediction information acquisition device 1 generates one or more training datasets based on the pre-processing training datasets in accordance with predetermined rules. Hereinafter, the process of generating one or more training datasets based on the pre-processing training datasets is referred to as the training dataset generation process. The training dataset generation process is a process of generating one or more training datasets from the training dataset by generating one or more graph tumor information based on the tumor image data included in the pre-processing training dataset.

[0035] The graph tumor information generated based on tumor image data is information showing the image of the tumor shown in the tumor image represented by the tumor image data, and is information showing the value of a predefined graph quantity. The process of generating one or more pieces of graph tumor information based on tumor image data is, for example, a process of calculating the value of a predefined graph quantity based on the tumor image data. Hereinafter, the process of generating one or more pieces of graph tumor information based on tumor image data is referred to as the graph tumor information acquisition process.

[0036] 2 is a diagram showing an example of the hardware configuration of a prognosis prediction information acquisition device 1 in an embodiment. The prognosis prediction information acquisition device 1 includes a control unit 11 having a processor 91 such as a CPU (Central Processing Unit) and a memory 92 connected by a bus, and executes a program. By executing the program, the prognosis prediction information acquisition device 1 functions as a device including the control unit 11, a communication unit 12, an input unit 13, a memory unit 14, and an output unit 15.

[0037] More specifically, the processor 91 reads out a program stored in the storage unit 14 and stores the read out program in the memory 92. When the processor 91 executes the program stored in the memory 92, the prognosis prediction information acquisition device 1 functions as a device including a control unit 11, a communication unit 12, an input unit 13, a storage unit 14, and an output unit 15.

[0038] The control unit 11 controls the operations of various functional units included in the prognosis prediction information acquisition device 1. The control unit 11 executes, for example, a learning dataset generation process. The control unit 11 executes, for example, a prognosis prediction learning model.

[0039] The control unit 11 controls, for example, the operation of the output unit 15. The control unit 11 records, for example, various pieces of information generated by the execution of the prognosis prediction learning model in the storage unit 14. The control unit 11 records, for example, information input to the communication unit 12 or the input unit 13 in the storage unit 14.

[0040] The communication unit 12 is configured to include a communication interface for connecting the prognosis prediction information acquisition device 1 to an external device. The communication unit 12 communicates with the external device via wired or wireless communication. The external device is, for example, a device that transmits a pre-processing training dataset. The communication unit 12 receives a pre-processing training dataset transmitted by the device that transmits the pre-processing training dataset through communication with the device that transmits the pre-processing training dataset. The external device is, for example, a prognosis prediction device 2. The communication unit 12 transmits, for example, prognosis prediction information to the prognosis prediction device 2 through communication with the prognosis prediction device 2. The prognosis prediction information transmitted by the communication unit 12 to the prognosis prediction device 2 is, for example, a trained prognosis prediction learning model.

[0041] The input unit 13 includes input devices such as a mouse, keyboard, and touch panel. The input unit 13 may be configured as an interface that connects these input devices to the prognosis prediction information acquisition device 1. The input unit 13 accepts input of various information to the prognosis prediction information acquisition device 1. For example, an instruction to start learning is input to the input unit 13. An unprocessed learning dataset may be input to the input unit 13.

[0042] The storage unit 14 is configured using a computer-readable storage medium device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 14 stores various information related to the prognosis prediction information acquisition device 1. The storage unit 14 stores information input via, for example, the input unit 13 or the communication unit 12. The storage unit 14 stores various information generated by, for example, the execution of a prognosis prediction learning model.

[0043] It should be noted that the pre-processing training data set does not necessarily have to be input only to the communication unit 12, or only to the input unit 13. The pre-processing training data set may be input from either the communication unit 12 or the input unit 13. Furthermore, the pre-processing training data set may be stored in advance in the storage unit 14.

[0044] The output unit 15 outputs various types of information. The output unit 15 includes a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display. The output unit 15 may be configured as an interface that connects these display devices to the prognosis prediction information acquisition device 1. The output unit 15 outputs, for example, information input to the input unit 13. The output unit 15 may display, for example, the execution results of a prognosis prediction learning model. The output unit 15 may output, for example, graph tumor information included in a training dataset generated by executing a training dataset generation process.

[0045] 3 is a diagram illustrating an example of the functional configuration of the control unit 11 according to the embodiment. The control unit 11 includes a communication control unit 101, an input control unit 102, an output control unit 103, a learning dataset acquisition unit 104, a model execution unit 105, an update unit 106, a termination determination unit 107, and a memory control unit 108.

[0046] The communication control unit 101 controls the operation of the communication unit 12. Under the control of the communication control unit 101, the communication unit 12 transmits the learned prognosis prediction learning model (i.e., prognosis prediction information) to the prognosis prediction device 2. The input control unit 102 controls the operation of the input unit 13. The output control unit 103 controls the operation of the output unit 15.

[0047] The training dataset acquisition unit 104 executes a training dataset generation process on the pre-processing training dataset. The training dataset acquisition unit 104 generates one or more training datasets by executing the training dataset generation process. In this way, the training dataset acquisition unit 104 acquires a training dataset by generating a training dataset by executing the training dataset generation process.

[0048] The pre-processing training dataset on which the training dataset acquisition unit 104 executes the training dataset generation process is, for example, a pre-processing training dataset input to the communication unit 12 or the input unit 13. The pre-processing training dataset on which the training dataset acquisition unit 104 executes the training dataset generation process may be, for example, a pre-processing training dataset stored in the storage unit 14, if the pre-processing training dataset has been recorded in advance in the storage unit 14.

[0049] The model execution unit 105 executes the prognosis prediction learning model on the graph tumor information included in the training dataset acquired by the training dataset acquisition unit 104. The model execution unit 105 executes the prognosis prediction learning model to predict the prognosis of a person or animal with a tumor indicated by the graph tumor information acquired by the training dataset acquisition unit 104.

[0050] The update unit 106 updates the prognosis prediction learning model based on the prediction loss. The prediction loss is the difference between the prediction result of the model execution unit 105 and the correct label (i.e., prognosis information) included in the learning dataset acquired by the learning dataset acquisition unit 104.

[0051] Both the prediction result of the model execution unit 105 and the prognosis information are represented, for example, by a first-order tensor (i.e., a vector). Both the prediction result of the model execution unit 105 and the prognosis information may be represented, for example, by a second-order tensor (i.e., a matrix). Both the prediction result of the model execution unit 105 and the prognosis information may be represented, for example, by a third-order or higher tensor. The prediction loss is represented, for example, by the reciprocal of the dot product between a tensor representing the prediction result of the model execution unit 105 and a tensor representing the prognosis information. The larger the value of the dot product, the smaller the difference between the tensor representing the prediction result of the model execution unit 105 and the tensor representing the prognosis information, and therefore the smaller the prediction loss.

[0052] The termination determination unit 107 determines whether or not the learning termination condition is satisfied. The memory control unit 108 records various information in the memory unit 14.

[0053] 4 is a flowchart showing an example of the flow of processing executed by the prognosis prediction information acquisition device 1 according to an embodiment. A pre-processing training dataset is input to the communication unit 12 or the input unit 13, and the training dataset acquisition unit 104 acquires a training dataset by executing a training dataset generation process on the input pre-processing training dataset (step S101).

[0054] Next, the model execution unit 105 executes the prognosis prediction learning model on the graph tumor information included in the learning dataset acquired in step S101 (step S102). By executing the prognosis prediction learning model, the model execution unit 105 predicts the prognosis of a person or animal with a tumor indicated by the graph tumor information acquired in step S101.

[0055] Next, the update unit 106 updates the prognosis prediction learning model based on the difference (i.e., prediction loss) between the prediction result obtained in step S102 and the correct label included in the learning dataset obtained in step S101 to reduce the difference (step S103).

[0056] Next, the termination determination unit 107 determines whether the learning termination condition is satisfied (step S104). If the learning termination condition is satisfied (step S104: YES), the process ends. On the other hand, if the learning termination condition is not satisfied (step S104: NO), the process returns to step S101.

[0057] 5 is a diagram showing an example of the hardware configuration of a prognosis prediction device 2 in an embodiment. The prognosis prediction device 2 includes a control unit 21 having a processor 93 such as a CPU and a memory 94 connected by a bus, and executes a program. By executing the program, the prognosis prediction device 2 functions as a device including the control unit 21, a communication unit 22, an input unit 23, a storage unit 24, and an output unit 25.

[0058] More specifically, in the prognosis prediction device 2, the processor 93 reads out a program stored in the storage unit 24 and stores the read program in the memory 94. When the processor 93 executes the program stored in the memory 94, the prognosis prediction device 2 functions as a device including a control unit 21, a communication unit 22, an input unit 23, a storage unit 24, and an output unit 25.

[0059] The control unit 21 controls the operation of each functional unit included in the prognosis prediction device 2. The control unit 21 acquires, for example, tumor image data that depicts an image of a tumor of the prediction target. The control unit 21 predicts the prognosis of the prediction target using prognosis prediction information based on the acquired tumor image data. Hereinafter, the process of predicting the prognosis of the prediction target using prognosis prediction information based on the tumor image data will be referred to as prognosis prediction process.

[0060] The prognosis prediction process includes preprocessing and main processing. The preprocessing is performed before the main processing is performed. The preprocessing is a process for generating one or more pieces of graph tumor information based on tumor image data. For example, the preprocessing is a process for calculating the value of a predetermined graph quantity that was defined in the training data generation process based on the tumor image data. The preprocessing may be any process that generates one or more pieces of graph tumor information based on the tumor image data, and may be, for example, a process similar to the graph tumor information acquisition process performed in the training data set generation process.

[0061] The main processing is a process of predicting the prognosis of a prediction target using prognosis prediction information based on the graph tumor information obtained in the preprocessing. The main processing is a process of executing a trained prognosis prediction learning model on the graph tumor information obtained in the preprocessing, for example.

[0062] The control unit 21 records, for example, the results of the execution of the prognosis prediction process in the storage unit 24. The control unit 21 controls the operation of the communication unit 22, for example.

[0063] The communication unit 22 is configured to include a communication interface for connecting the prognosis prediction device 2 to an external device. The communication unit 22 communicates with the external device via wired or wireless communication. The external device that the communication unit 22 communicates with is, for example, the prognosis prediction information acquisition device 1. The communication unit 22 receives the prognosis prediction information transmitted by the prognosis prediction information acquisition device 1, for example, by communicating with the prognosis prediction information acquisition device 1.

[0064] The external device may be, for example, a device that is a source of tumor image data. In such a case, communication unit 22 receives the tumor image data through communication with the device that is a source of tumor image data. The external device may be, for example, a tumor image capturing device. The tumor image capturing device may be, for example, a CT device. The tumor image capturing device may be, for example, an MRI device.

[0065] The input unit 23 includes input devices such as a mouse, a keyboard, and a touch panel. The input unit 23 may be configured as an interface that connects these input devices to the prognosis prediction device 2. The input unit 23 accepts input of various information to the prognosis prediction device 2. For example, tumor image data may be input to the input unit 23.

[0066] The storage unit 24 is configured using a computer-readable storage medium device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 24 stores various information related to the prognosis prediction device 2. The storage unit 24 stores, for example, a program that controls the operation of each functional unit provided in the prognosis prediction device 2 in advance. The storage unit 24 stores, for example, prognosis prediction information. The prognosis prediction information stored in the storage unit 24 is prognosis prediction information obtained by the prognosis prediction information acquisition device 1. The storage unit 24 stores, for example, the results of prediction by the prognosis prediction device 2.

[0067] It should be noted that the tumor image data does not necessarily have to be input only to the communication unit 22, or only to the input unit 23. The tumor image data may be input from either the communication unit 22 or the input unit 23.

[0068] The output unit 25 outputs various types of information. The output unit 25 includes a display device such as a CRT display, a liquid crystal display, or an organic EL display. The output unit 25 may be configured as an interface that connects these display devices to the prognosis prediction device 2. The output unit 25 outputs information input to the input unit 23, for example. The output unit 25 may display a tumor image indicated by tumor image data input to the communication unit 22 or the input unit 23, for example. The output unit 25 may display the execution results of the prognosis prediction process, for example.

[0069] 6 is a diagram illustrating an example of the functional configuration of the control unit 21 according to the embodiment. The control unit 21 includes a communication control unit 210, an input control unit 220, an output control unit 230, an input data acquisition unit 240, a prognosis prediction unit 250, and a memory control unit 260.

[0070] The communication control unit 210 controls the operation of the communication unit 22. The input control unit 220 controls the operation of the input unit 23. The output control unit 230 controls the operation of the output unit 25.

[0071] The input data acquisition unit 240 acquires data input to the communication unit 22 or the input unit 23. The input data acquisition unit 240 acquires, for example, tumor image data input to the communication unit 22 or the input unit 23. For example, if the tumor image data has been input in advance to the communication unit 22 or the input unit 23 and has been recorded in advance in the storage unit 24, the input data acquisition unit 240 may read the tumor image data from the storage unit 24.

[0072] The prognosis prediction unit 250 executes a prognosis prediction process on the tumor image data acquired by the input data acquisition unit 240. By executing the prognosis prediction process, the prognosis prediction unit 250 predicts the prognosis of a person or animal having a subject tumor (i.e., a prediction target).

[0073] The prognosis prediction unit 250 includes a preprocessing execution unit 251 and a main processing execution unit 252. The preprocessing execution unit 251 executes preprocessing. The main processing execution unit 252 executes main processing. The prognosis prediction process by the prognosis prediction unit 250 is, for example, a process in which the preprocessing execution unit 251 executes preprocessing on tumor image data acquired by the input data acquisition unit 240, and the main processing execution unit 252 executes main processing on the graph tumor information acquired by the preprocessing execution unit 251. The tumor shown in the image indicated by the tumor image data acquired by the input data acquisition unit 240 is the tumor present in the estimation subject. The graph tumor information showing an image of the tumor present in the estimation subject is obtained by executing preprocessing on the tumor image data acquired by the input data acquisition unit 240. Therefore, the prognosis prediction unit 250 predicts the prognosis of the estimation subject based on the graph tumor information showing an image of the tumor present in the estimation subject.

[0074] The storage control unit 206 records various types of information in the storage unit 24 .

[0075] 7 is a flowchart showing an example of the flow of processing executed by the prognosis prediction device 2 in an embodiment. The input data acquisition unit 240 acquires tumor image data showing an image of a tumor of an estimation target (step S201). The estimation target is a person or animal whose prognosis is predicted by the prognosis prediction device 2. Next, the prognosis prediction unit 250 predicts the prognosis of the estimation target by executing a prognosis prediction process (step S202). The output control unit 230 controls the operation of the output unit 25, thereby causing the output unit 25 to display the prediction result obtained in step S202 (step S203).

[0076] <Relationship between tumor images and graph tumor information> FIG. 8 is a first explanatory diagram illustrating an example of the relationship between tumor images and graph tumor information in an embodiment. FIG. 8 shows three images, image G1, image G2, and image G3, and their relationship. Image G1 in FIG. 8 is an example of an image (i.e., a tumor image) indicated by tumor image data. Image G1 shows a cross-sectional view of the chest of the estimation target. Image G1 also shows a cross-sectional view of the tumor of the estimation target. The image of area A1 in image G1 is a cross-sectional view of the tumor of the estimation target.

[0077] Image G2 in Figure 8 is an image in which an example of a vertex in graph theory that represents a tumor in area A1 is superimposed on the image of area A1. Hereinafter, for ease of explanation, a vertex in graph theory will simply be referred to as a "vertex." Hereinafter, for ease of explanation, an edge in graph theory will simply be referred to as an "edge." Hereinafter, for ease of explanation, a graph in graph theory will simply be referred to as a "graph."

[0078] In image G2, there is one vertex for each pixel in the image of area A1. Therefore, the positional relationship of the vertices is the same as the positional relationship of the pixels. The color of each vertex shown in image G2 represents the pixel value of the corresponding pixel in the image of area A1. The image of area A1 is a grayscale image. Therefore, the pixel value represented by the vertex of image G2 is a scalar. The pixel value is, for example, a CT value.

[0079] Image G3 shows vertices that exist within area A2 in image G2. Image G3 shows an example of an edge. An edge connects one vertex to another. However, edges do not necessarily exist between all vertices. The edges shown in Figure 8 are edges where the absolute value of the difference in pixel values ​​between the vertices connected by the edge is equal to or greater than a predetermined threshold.

[0080] In the example of Figure 8, a vertex is defined for each pixel of the tumor image. However, a vertex does not necessarily have to be defined for each pixel. If a vertex is defined for each predefined unit pixel, it does not have to be defined for each pixel. A unit pixel is a group of one or more adjacent pixels. A unit pixel is, for example, a 3x3 pixel.

[0081] When one vertex corresponds to one pixel as in the example of Figure 8, the vertex represents the pixel value of the corresponding pixel. The pixel value represented by the vertex is an example of a pixel index value. The pixel index value is a predefined index value related to the pixel value of each pixel included in the unit pixel. The pixel value represented by the vertex is an example of the weight of the vertex.

[0082] The pixel index value is, for example, the pixel value itself when the image is a scalar image with pixel values ​​such as a grayscale image or a monochrome image, and the unit pixel is one pixel. The pixel index value is, for example, a statistical quantity of the distribution of pixel values ​​of each pixel included in the unit pixel when the image is a scalar image with pixel values ​​such as a grayscale image or a monochrome image, and the unit pixel is a plurality of pixels. The statistical quantity may be a representative value or a dispersion. The representative value may be, for example, the average, the median, the maximum value, or the minimum value. The dispersion may be, for example, the standard deviation.

[0083] The pixel index value is, for example, a scalar index indicating a tensor representing a pixel value when the image is a color image or other image in which pixel values ​​are expressed by tensors of first or higher order and the unit pixel is one pixel. The scalar index indicating a tensor representing a pixel value is, for example, the magnitude of the vector when the tensor is a vector (i.e., a first-order tensor). The scalar index indicating a tensor representing a pixel value is, for example, the value of the determinant when the tensor is a matrix (i.e., a second-order tensor). The scalar index indicating a tensor representing a pixel value is, for example, the sum of the squares of each element when the tensor is a third-order or higher tensor. The pixel index value represented by a vertex is an example of the vertex weight.

[0084] In this way, a vertex is a quantity defined for each unit pixel of the tumor image. Therefore, graph tumor information using vertices as graph quantities includes information indicating the position of the corresponding unit pixel within the tumor image for each vertex (hereinafter referred to as "unit pixel position information"). In the unit pixel position information, the position of each unit pixel within the tumor image is indicated, for example, by the position of the center of gravity of each unit pixel. In the unit pixel position information, the position of each unit pixel within the tumor image may be indicated, for example, by the position of a predetermined end of each unit pixel.

[0085] In Figure 8, only edges that satisfy the condition that the absolute value of the difference in pixel values ​​between the vertices connected by the edge is greater than or equal to a predetermined threshold were present. That is, in the example of Figure 8, the graph tumor information includes edges as graph quantities, and these edges satisfy the condition that the absolute value of the difference in pixel values ​​between the vertices is greater than or equal to a predetermined threshold. However, the edges included in the graph tumor information do not necessarily have to satisfy the condition that the absolute value of the difference in pixel values ​​between the vertices connected by the edge is greater than or equal to a predetermined threshold.

[0086] The edges included in the graph tumor information may be any edges as long as they satisfy a predetermined condition (hereinafter referred to as "edge condition") regarding the difference between the pixel index values ​​of the two vertices connected by the edge (hereinafter referred to as "difference condition"). Note that the edges included in the graph tumor information are edges that the graph has. The difference condition is, for example, a condition that the absolute value of the difference between the pixel index values ​​of the two vertices connected by the edge is less than or equal to a predetermined value. The difference condition may be, for example, a condition that the magnitude of the ratio of the pixel index values ​​of the two vertices connected by the edge is within a predetermined range. The difference between the pixel index values ​​of the two vertices connected by the edge is an example of the weight of the edge.

[0087] For simplicity of explanation, the prognosis prediction system 100 will be described using an example in which the unit pixel is one pixel, the image is a grayscale image, and the difference condition is that the absolute value of the difference between the pixel index values ​​of the two vertices connected by the edge is less than or equal to a predetermined value (hereinafter referred to as the "edge threshold").

[0088] FIG. 9 is a second explanatory diagram illustrating an example of the relationship between a tumor image and graph tumor information in an embodiment. More specifically, FIG. 9 is a diagram illustrating an example of the change in the graph when the edge threshold is changed. The edge threshold had a minimum of 5 HU and a maximum of 50 HU. Ten values ​​were used as the edge threshold, separated by 5 HU intervals from 5 HU to 50 HU.

[0089] Figure 9 shows images 4-1, 4-2, 4-3, 4-4, and 4-5. Images 4-1 to 4-5 are all graphs showing the same tumor image. Figure 9 shows that the graphs for images 4-1 to 4-5 have different topologies. The reason why the topologies are different despite showing the same tumor image is because the edge thresholds are different.

[0090] More specifically, Image 4-1 is an example of a graph when the edge threshold is B1. Image 4-2 is an example of a graph when the edge threshold is B2. The value B2 is greater than the value B1. Image 4-3 is an example of a graph when the edge threshold is B3. The value B3 is greater than the value B2. Image 4-4 is an example of a graph when the edge threshold is B4. The value B4 is greater than the value B3. Image 4-5 is an example of a graph when the edge threshold is B5. The value B5 is greater than the value B4.

[0091] Figure 9 shows that the larger the edge threshold, the more heterogeneous the graph represents the tumor interior. The reason for this is that the larger the edge threshold, the more heterogeneous the graph represents the tumor interior, since the larger the edge threshold, the more information on heterogeneous regions within the tumor is used to create the graph, where the CT value differences are large. More specifically, this is due to the following reason. For a graph with only edges whose weights are equal to or greater than the edge threshold, the larger the difference between the weights of one vertex and the weight of the other vertex between the two vertices connected by the edge, the larger the difference between the weights of one vertex and the weight of the other vertex between the two vertices connected by the edge. Furthermore, the larger the difference between the weights of one vertex and the weight of the other vertex between the two vertices connected by the edge, the more heterogeneous the tumor interior represented by the graph. Therefore, the larger the edge threshold, the more heterogeneous the graph represents the tumor interior.

[0092] <Graph quantity example> An example of graph quantity will be described in more detail. FIG. 10 is a first explanatory diagram illustrating an example of graph quantity in an embodiment. FIG. 10 shows a graph having a total of 25 vertices. The graph in FIG. 10 has a topology in which the 25 vertices are arranged in a 5×5 matrix. The graph in FIG. 10 is a weighted graph, and the thickness of the edges indicates the weight of the edges. The thicker the edge, the heavier the weight of the edge.

[0093] The graph quantity may be, for example, a quantity related to vertices. The quantity related to vertices is, for example, the number of vertices that the graph has. In the example of FIG. 10, the value of the graph quantity indicating the number of vertices is 25. Hereinafter, the number of vertices that the graph has is referred to as a first graph feature.

[0094] The graph quantity may be, for example, a quantity related to edges. The quantity related to edges is, for example, the number of edges the graph has. The edges of the graph satisfy the condition that the difference between the pixel index values ​​of the two vertices connected by the edge satisfies a difference condition. The edges of the graph satisfy the condition that the absolute value of the difference between the pixel values ​​of the two vertices connected by the edge is equal to or greater than an edge threshold.

[0095] For ease of explanation, the prognosis prediction system 100 will be described below using as an example a case where an edge in a graph satisfies the condition that the absolute value of the difference in pixel values ​​between the two vertices connected by the edge is equal to or greater than an edge threshold. Hereinafter, the number of edges in a graph will be referred to as a second graph feature.

[0096] The quantity related to an edge may be, for example, the sum of the weights of the edges in the graph. Hereinafter, the sum of the weights of the edges in the graph is referred to as a third graph feature. The quantity related to an edge may be, for example, the average value of the weights of the edges in the graph. Hereinafter, the average value of the weights of the edges in the graph is referred to as a fourth graph feature. The quantity related to an edge may be, for example, the maximum value of the weights of the edges in the graph. Hereinafter, the maximum value of the weights of the edges in the graph is referred to as a fifth graph feature.

[0097] The quantity related to edges may be, for example, the number of edges in the graph whose weight differs from the edge threshold by a predetermined difference or less. The condition that the difference between the weight and the edge threshold is equal to or less than the predetermined difference may be, for example, that the weight is the same as the edge threshold. Hereinafter, the number of edges in the graph whose weight satisfies the condition that the weight is the same as the edge threshold will be referred to as a sixth graph feature.

[0098] The quantity related to edges may be, for example, the number of edges in the graph whose difference between the weight and the edge threshold is greater than a predetermined difference. The condition that the difference between the weight and the edge threshold is greater than a predetermined difference may be, for example, a condition that the weight is greater than the edge threshold. Hereinafter, the number of edges in the graph whose weight is greater than the edge threshold is referred to as a seventh graph feature.

[0099] The graph quantity may be, for example, a quantity related to the degree. The quantity related to the degree is, for example, the sum of the degrees in the graph. Hereinafter, the sum of the degrees in the graph is referred to as the eighth graph feature. The quantity related to the degree may be, for example, the average value of the degrees in the graph. Hereinafter, the average value of the degrees in the graph is referred to as the ninth graph feature.

[0100] Fig. 11 is a second explanatory diagram illustrating an example of graph quantity in an embodiment. Fig. 11 shows a graph having a 5x5 topology. Fig. 11 shows that the number of edges connected to vertex C1 is 2. Since two edges are connected to vertex C1, the degree of vertex C1 is 2. Each of the two edges connected to vertex C1 is an edge that the graph has.

[0101] As described above, the graph quantity may be, for example, a quantity related to a minimum spanning tree. The quantity related to a minimum spanning tree is, for example, the sum of the weights of the edges of the minimum spanning tree. Hereinafter, the sum of the weights of the edges of the minimum spanning tree will be referred to as the 10th graph feature. The quantity related to a minimum spanning tree may be, for example, the number of edges of the minimum spanning tree. Hereinafter, the number of edges of the minimum spanning tree will be referred to as the 11th graph feature.

[0102] The quantity related to the minimum spanning tree may be the weight of the edge with the largest weight among the edges that make up the minimum spanning tree. Hereinafter, the weight of the edge with the largest weight among the edges that make up the minimum spanning tree will be referred to as the 12th graph feature. The quantity related to the minimum spanning tree may be the ratio of the number of edges that make up the minimum spanning tree to the number of edges that the graph has. Hereinafter, the ratio of the number of edges that make up the minimum spanning tree to the number of edges that the graph has will be referred to as the 13th graph feature.

[0103] The graph quantity does not necessarily have to be a quantity obtained from a single tumor image. The graph quantity may be obtained from multiple images of the same tumor taken under different conditions. For example, the graph quantity may be the sum of the values ​​of the first-type individual graph quantities. The first-type individual graph quantity is a predefined graph quantity obtained from each of multiple cross-sectional views. The multiple cross-sectional views are cross-sectional views of the same tumor, but at different positions in a predetermined direction, for example. The predetermined direction is, for example, the axial direction. Hereinafter, the sum of the values ​​of the first-type individual graph quantities will be referred to as the 14th graph feature.

[0104] The graph quantity may be, for example, the quantity obtained by dividing the sum of the values ​​of the first type individual graph quantities by the tumor volume. Hereinafter, the quantity obtained by dividing the sum of the values ​​of the first type individual graph quantities by the tumor volume will be referred to as the 15th graph feature. The graph quantity may be, for example, the quantity obtained by dividing the sum of the values ​​of the first type individual graph quantities by the number of cross-sectional views. Hereinafter, the quantity obtained by dividing the sum of the values ​​of the first type individual graph quantities by the number of cross-sectional views will be referred to as the 16th graph feature.

[0105] The graph quantity may be, for example, a quantity related to a three-dimensional graph (i.e., a spatial graph) obtained by connecting the graphs obtained for each of a plurality of cross-sectional views with edges. The multiple cross-sectional views are cross-sectional views of the same tumor, but are, for example, cross-sectional views that differ in position in a predetermined direction. The predetermined direction is, for example, the axial direction. Hereinafter, the quantity related to the three-dimensional graph obtained by connecting the graphs obtained for each of a plurality of cross-sectional views with edges will be referred to as a 17th graph feature.

[0106] The multiple cross-sectional views do not necessarily have to be multiple cross-sectional views at different spatial positions, but may be multiple cross-sectional views at different temporal positions, i.e., the multiple cross-sectional views may be the results of capturing the same cross section at different times.

[0107] The multiple cross-sectional views may be cross-sectional views at different spatial and temporal positions.

[0108] The graph quantity does not necessarily have to be a quantity obtained from a single graph. The graph quantity may be a quantity obtained based on second-type individual graph quantities. The second-type individual graph quantity is a set of predefined graph quantities obtained from multiple graphs showing the same tumor image, each of which has a different edge threshold.

[0109] A graph quantity is, for example, the slope of a function that shows the relationship between the edge threshold of a graph ordered set and the value of each element. A graph ordered set is an ordered set in which each element included in the second-type individual graph quantity is arranged in order of the magnitude of the edge threshold. Hereinafter, the slope of the function that shows the relationship between the edge threshold of a graph ordered set and the value of each element is referred to as the 18th graph feature.

[0110] FIG. 12 is a third explanatory diagram illustrating an example of a graph quantity in an embodiment. More specifically, FIG. 12 is a diagram illustrating an example of a second-type individual graph quantity. The horizontal axis of FIG. 12 indicates the edge threshold, and the vertical axis indicates the value of the graph quantity. An ordered set of values ​​on the vertical axis corresponding to each edge threshold of curve L1 in FIG. 12 is an example of a graph ordered set. In other words, curve L1 in FIG. 12 is a line expressing an example of a graph ordered set. The minimum edge threshold was 1 HU and the maximum was 50 HU. 50 values ​​ranging from 1 HU to 50 HU, separated by 1 HU intervals, were used as the edge thresholds.

[0111] The graph quantity may be, for example, the y-intercept value of a function that indicates the relationship between the edge threshold of the graph ordered set and the value of each element. Hereinafter, the y-intercept value of the function that indicates the relationship between the edge threshold of the graph ordered set and the value of each element will be referred to as the 19th graph feature.

[0112] The graph quantity may be, for example, the position of a peak of a function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element. Specifically, the position of the peak is the edge threshold that corresponds to the peak value of the function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element. Hereinafter, the position of the peak of the function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element will be referred to as the 20th graph feature.

[0113] The graph quantity may be, for example, the kurtosis of a function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element. Hereinafter, the kurtosis of the function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element will be referred to as the 21st graph feature.

[0114] The graph quantity may be, for example, the skewness of a function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element. Hereinafter, the skewness of a function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element will be referred to as the 22nd graph feature.

[0115] The graph quantity may be, for example, the interquartile range of a function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element. Hereinafter, the interquartile range of a function that indicates the relationship between the edge threshold of a graph ordered set and the value of each element will be referred to as the 23rd graph feature.

[0116] <Experimental Results> An example of the results of an experiment on prognosis prediction using the prognosis prediction system 100 will be described. The cases used in the experiment were a total of 304 cases of non-small cell lung cancer, excluding excluded cases among patients who underwent radiation therapy for lung cancer. That is, a total of 304 tumor images were used in the experiment. Each of the tumor images used in the experiment was a planning CT image or structure data. The excluded cases were cases of small cell lung cancer, cases with unknown histological type, and cases with insufficient GTV contours.

[0117] In the experiment, 70% of the 304 tumor images were used to train the learning model. In other words, 213 examples, or 70% of the 304 tumor images, were used as information contained in the training data. In the experiment, 91 examples, or 30% of the 304 tumor images, were used to verify the prediction accuracy of the trained learning model. In other words, 30% of the 304 tumor images were used as test data. However, the ratio of tumor images from deceased patients to tumor images from surviving patients among the tumor images used for training was approximately the same as the ratio of tumor images from deceased patients to tumor images from surviving patients among the tumor images used for verification.

[0118] FIG. 13 is a first explanatory diagram illustrating an experiment in an embodiment. More specifically, FIG. 13 is a diagram showing details of the cases used in the experiment. FIG. 13 shows that 304 patient cases were used in the experiment. FIG. 13 shows that the median survival time of the patients was 598 days. FIG. 13 shows that the range of patient survival time was 1 day to 3364 days.

[0119] Figure 13 shows the distribution of histology among the 304 patients. Figure 13 shows that for 135 of the 304 patients, the histology was adenocarcinoma. Figure 13 shows that for 149 of the 304 patients, the histology was squamous cell carcinoma. Figure 13 shows that for 7 of the 304 patients, the histology was large cell carcinoma. Figure 13 shows that for 13 of the 304 patients, the histology was NOS (not otherwise specified).

[0120] Figure 13 shows that patients with adenocarcinoma histology had a median survival time of 562 days, ranging from 1 to 3364 days. Figure 13 shows that patients with squamous cell carcinoma histology had a median survival time of 775 days, ranging from 8 to 3253 days. Figure 13 shows that patients with large cell carcinoma histology had a median survival time of 540 days, ranging from 279 to 2875 days. Figure 13 shows that patients with NOS histology had a median survival time of 967 days, ranging from 77 to 2562 days.

[0121] Figure 13 shows the distribution of disease stages among the 304 patients. Figure 13 shows that 83 patients had stage I disease. Figure 13 shows that 25 patients had stage II disease. Figure 13 shows that 146 patients had stage III disease. Figure 13 shows that 41 patients had stage IV disease.

[0122] Figure 13 shows that patients with stage I disease had a median survival time of 895 days, ranging from 10 to 3364 days. Figure 13 shows that patients with stage II disease had a median survival time of 418 days, ranging from 49 to 2875 days. Figure 13 shows that patients with stage III disease had a median survival time of 677.5 days, ranging from 19 to 3302 days. Figure 13 shows that patients with stage IV disease had a median survival time of 208 days, ranging from 9 to 2824 days.

[0123] FIG. 14 is a second explanatory diagram illustrating an experiment in an embodiment. More specifically, FIG. 14 is a diagram showing an example of a tumor image used in the experiment. In the experiment, a GTV (Gross Tumor Volume) region was extracted from the planning CT image used in the treatment plan, and graph tumor information was generated for the image of the extracted GTV region. That is, in the experiment, an image of the GTV region included in the planning CT image was used as the tumor image. The region surrounded by a frame W1 in FIG. 14 is an example of an image of the GTV region.

[0124] FIG. 15 is a third explanatory diagram illustrating an experiment in an embodiment. More specifically, FIG. 15 is a diagram illustrating an overview of an example of the processing flow performed in an experiment until graph tumor information is generated from a tumor image. In the experiment, multiple cross-sectional views of the patient were acquired. The cross-sectional views were axial cross-sectional views. In the experiment, an image of the GTV region was extracted for each of the multiple cross-sectional views acquired.

[0125] "Slice: 1" in FIG. 15 means the first cross-sectional view. "Slice: 2" in FIG. 15 means the second cross-sectional view. "Last Slice" in FIG. 15 means the last cross-sectional view among multiple cross-sectional views. In the case of FIG. 15, there were 20 cross-sectional views. Therefore, "Last Slice" in FIG. 15 means the 20th cross-sectional view.

[0126] In the experiment, a graph was generated for each image of each GTV region. In the experiment, a predetermined threshold was set for the edges represented by the graph. In the experiment, a process was performed in which edges below the predetermined threshold were deleted. When edges below the predetermined threshold were deleted, the graph tumor information showed the heterogeneity inside the tumor with higher accuracy than when edges were not deleted.

[0127] In the experiment, multiple thresholds were used for edge deletion. Specifically, the minimum edge threshold used in the experiment was 5 HU and the maximum was 50 HU. The edge thresholds used in the experiment were 10 values ​​separated by 5 HU intervals from 5 HU to 50 HU. In the experiment, a process to obtain the graph quantity value for each threshold was performed on the image of each GTV region. Hereinafter, the process to obtain the graph quantity value for each threshold performed on the image of each GTV region is referred to as the first-stage graph quantity acquisition process.

[0128] In the experiment, the sum of the graph quantities obtained in the first-stage graph quantity acquisition process was obtained and divided by the volume of the GTV. Hereinafter, the process of obtaining the sum of the graph quantities obtained in the first-stage graph quantity acquisition process and dividing by the volume of the GTV is referred to as the second-stage graph quantity acquisition process.

[0129] In the experiment, the graph quantities obtained in the first-stage graph quantity acquisition process and the graph quantities obtained in the second-stage graph quantity acquisition process were used. A total of 127 graph quantities were used in the experiment.

[0130] The total of 127 graph quantities were quantities classified into one of a total of 37 types of classifications, specifically the 14th graph feature calculated from the 1st graph feature, the 15th graph feature calculated from the 2nd to 12th graph features, the 18th-1st to 18th-13th graph features, the 19th-1st to 19th-0th graph features, the 20th graph feature whose elements included in the second type individual graph quantity are the 10th graph feature, and the 21st graph feature whose elements included in the second type individual graph quantity are the 10th graph feature.

[0131] The graph features 18-1 to 18-13 and 19-1 to 19-10 will be described later. In the experiment, information indicating each value of the 127 graph quantities obtained in this manner was used as graph tumor information.

[0132] Although the experiment used the 37 types of graph quantities mentioned above, the total number of graph quantities used in the experiment was greater than 23, such as 127 or 159. This is because the graph quantities used in the experiment were not necessarily one per classification. For example, for some of the 37 classifications in the experiment, graph quantities were obtained for multiple edge thresholds. Note that the number of edge thresholds in the experiment was 10. Also, for some of the 37 classifications in the experiment, graph quantities were obtained for each cross-section.

[0133] The reason will be explained in detail using an example where the number of cross sections is one. In the experiment, the number of the first graph feature was 1 regardless of the edge threshold. Therefore, the total number of graph quantities related to the vertices was 1. The second to fourth graph features were obtained for each edge threshold. The number of the fifth graph feature was 1 regardless of the edge threshold. The sixth and seventh graph features were obtained for each edge threshold. Since the number of edge thresholds was 10, the total number of graph quantities related to the edges was 51.

[0134] The eighth and ninth graph features were obtained for each edge threshold. Therefore, the total number of graph quantities related to degree was 20. The tenth to thirteenth graph features were obtained for each edge threshold. Therefore, the total number of graph quantities related to degree was 40.

[0135] The 14th graph feature was obtained for the 1st graph feature. The 15th graph feature was obtained for each edge threshold.

[0136] The 18th graph feature was obtained for each 15th graph feature. The 19th graph feature was obtained for each 15th graph feature. The 20th graph feature was obtained from the 15th graph feature calculated from the 10th graph feature. The 21st graph feature was obtained from the 15th graph feature calculated from the 10th graph feature. The 22nd graph feature was obtained from the 15th graph feature calculated from the 10th graph feature.

[0137] Before the experiment, a LASSO-cox regression model was used to verify how many types of graph quantities were desirable to use. As a result, it was confirmed that using the 127 graph quantities mentioned above out of a total of 159 graph quantities subject to verification resulted in higher prediction accuracy than when using 128 or more graph quantities. In addition to the 127 graph quantities mentioned above, the graph quantities subject to verification also included the 15th graph feature calculated from the 13th graph feature, the 18th graph feature whose elements of the second-type individual graph quantity are second graph features and whose slope is a cubic term, the 18th graph feature whose elements of the second-type individual graph quantity are second graph features and whose slope is a quadratic term, and the 18th graph feature whose elements of the second-type individual graph quantity are seventh graph features and whose slope is a cubic term. Graph feature, an 18th graph feature in which each element of the second type individual graph quantity is a 7th graph feature and the slope is a slope of a quadratic term, an 18th graph feature in which each element of the second type individual graph quantity is a 6th graph feature and the slope is a slope of a cubic term, an 18th graph feature in which each element of the second type individual graph quantity is a 6th graph feature and the slope is a slope of a quadratic term, an 18th graph feature in which each element of the second type individual graph quantity is a 6th graph feature and the slope is a slope of a linear term a feature, an 18th graph feature in which each element of the second type individual graph quantity is a fourth graph feature and the slope is a slope of a third-order term; a 18th graph feature in which each element of the second type individual graph quantity is a fourth graph feature and the slope is a slope of a second-order term; a 18th graph feature in which each element of the second type individual graph quantity is a third graph feature and the slope is a slope of a first-order term; a 18th graph feature in which each element of the second type individual graph quantity is a ninth graph feature and the slope is a slope of a third-order term an 18th graph feature in which each element of the second type individual graph quantity is a 9th graph feature and the slope is a slope of a quadratic term; an 18th graph feature in which each element of the second type individual graph quantity is an 8th graph feature and the slope is a slope of a cubic term; an 18th graph feature in which each element of the second type individual graph quantity is an 8th graph feature and the slope is a slope of a quadratic term; an 18th graph feature in which each element of the second type individual graph quantity is a 13th graph feature and the slope is a slope of a cubic term;The second type individual graph quantity included an 18th graph feature in which each element is a 13th graph feature and the slope is the slope of a quadratic term, an 18th graph feature in which each element is a 13th graph feature and the slope is the slope of a linear term, a 19th graph feature in which each element is a 13th graph feature and the slope is the slope of a linear term, an 18th graph feature in which each element is a 12th graph feature and the slope is the slope of a linear term, an 18th graph feature in which each element is a 12th graph feature and the slope is the slope of a linear term, an 18th graph feature in which each element is a 12th graph feature and the slope is the slope of a linear term, and a 22nd graph feature in which each element is a 10th graph feature.

[0138] In the experiment, learning was performed using a LASSO-Cox regression model as the learning model. In the experiment, pairs of graph tumor information and prognosis information obtained by the process shown in Figure 15 were used as learning datasets. The prognosis information in the experiment consisted of survival / death information and survival time.

[0139] In this way, in the experiment, prognosis prediction information was obtained by updating the LASSO-cox regression model through learning using the graph tumor information obtained by the process shown in Figure 15 as the explanatory variable and prognosis information as the objective variable. Therefore, the prognosis prediction information in the experiment was information indicating the relationship between the graph tumor information obtained by the process shown in Figure 15 and prognosis. Specifically, the learning in the experiment was a process of optimizing the coefficients for each graph quantity. Therefore, the LASSO-cox regression model is an example of a prognosis prediction learning model.

[0140] 16 to 19, learning in the prognosis prediction system 100 will be described in comparison with a conventional technique. The conventional technique used for comparison is a technique for predicting prognosis using 107 radiomics features.

[0141] Of the 107 radiomics features in the prior art used for comparison, 14 were features that represented tumor shape or size. Of the 107 radiomics features in the prior art used for comparison, 18 were features that represented the distribution of pixel values ​​of the tumor. Of the 107 radiomics features in the prior art used for comparison, 75 were features that represented tumor heterogeneity.

[0142] More specifically, among the features of the prior art compared to the conventional technology, the feature representing the shape or size of the tumor was generally called a "shape" feature. Among the features of the prior art compared to the conventional technology, the feature representing the distribution of pixel values ​​of the tumor was generally called a "statistical" feature. Among the features of the prior art compared to the conventional technology, the feature representing the heterogeneity of the tumor was generally called a "texture" feature.

[0143] The conventional technology for comparison is a technology that obtains a trained learning model that shows the relationship between 107 radiomics features and prognosis by updating the learning model using a machine learning method. The conventional technology for comparison is a technology that predicts prognosis using the trained learning model obtained through learning. Hereinafter, the learning model in the conventional technology for comparison will be referred to as the comparative learning model.

[0144] Fig. 16 is a fourth explanatory diagram illustrating an experiment in an embodiment. More specifically, Fig. 16 is a first diagram showing an example of the results of inputting training data into the prognosis prediction learning model of the prognosis prediction system 100 and learning the prognosis prediction learning model. The vertical axis of Fig. 16 represents the partial likelihood deviance of the conditional probability. The horizontal axis of Fig. 16 represents Log Lambda. Log Lambda refers to the parameters used to optimize the model. Specifically, the parameters used to optimize the model refer to the regularization coefficients.

[0145] FIG. 16 shows the partial likelihood of the conditional probability for each value on the horizontal axis. Line L2-1 in FIG. 16 shows the result of minimizing the error in the prognosis prediction result by the prognosis prediction system 100. Line L2-2 in FIG. 16 shows the regularization coefficient that minimizes the error in the prognosis prediction result by the prognosis prediction system 100. The error in the prognosis prediction result is the degree to which the prognosis prediction result is incorrect. Each numerical value enclosed in a box W2 in FIG. 16 represents a regularization coefficient.

[0146] FIG. 17 is a fifth explanatory diagram illustrating an experiment in an embodiment. More specifically, FIG. 17 is a second diagram showing an example of the results of inputting training data into the prognosis prediction learning model of the prognosis prediction system 100 and learning the prognosis prediction learning model. The vertical axis of FIG. 17 represents coefficients. Specifically, the coefficients are used to calculate the rad score. The rad score is an index indicating prognosis. A higher rad score indicates a worse prognosis. The horizontal axis of FIG. 17 represents Log Lambda. Log Lambda refers to a parameter used to optimize the model. Specifically, the parameter used to optimize the model refers to a regularization coefficient. That is, FIG. 17 shows the relationship between each feature and regularization. Note that the features in the experiment, which are the features in the prognosis prediction system 100, are a total of 159 graph quantities.

[0147] The feature quantities in the prognosis prediction system 100 may be any quantity that is a graph quantity that expresses graph tumor information. Therefore, the feature quantities in the prognosis prediction system 100 may be, for example, each of the first to 22nd graph features described above or below. In the example of FIG. 17, the feature quantities in the prognosis prediction system 100 are each of the 159 graph quantities described above. In this way, the feature quantities in the prognosis prediction system 100 may be any quantity that is a graph quantity that expresses graph tumor information, and the feature quantities used in the prognosis prediction system 100 may be quantities defined depending on the application scenario of the prognosis prediction system 100.

[0148] Line L3 in Figure 17 indicates the regularization coefficient that minimizes the error in the results of prognosis prediction by the prognosis prediction system 100. Each numerical value enclosed in a frame W3 in Figure 17 indicates a regularization coefficient. Note that each graph in Figure 17 indicates the coefficient of each graph quantity input to the prognosis prediction learning model. Therefore, even graphs with the same color scheme or line type (i.e., even if they exist on the same line) will show different graph quantity coefficients for each point on the horizontal axis.

[0149] Fig. 18 is a sixth explanatory diagram illustrating an experiment in an embodiment. More specifically, Fig. 18 is a first diagram illustrating an example of the results of inputting training data into a comparative learning model of a conventional technology and learning the comparative learning model. The vertical axis of Fig. 18 represents the partial likelihood deviance of the conditional probability. The horizontal axis of Fig. 18 represents Log Lambda. Log Lambda means the regularization coefficient.

[0150] Figure 18 shows the partial likelihood of the conditional probability for each value on the horizontal axis. The dotted line L4-1 in Figure 18 shows the result of minimizing the error in the prognosis prediction result using the trained comparative learning model. The line L4-2 in Figure 18 shows the regularization coefficient that minimizes the error in the prognosis prediction result using the trained comparative learning model. Each number enclosed in the box W4 in Figure 18 represents a regularization coefficient.

[0151] FIG. 19 is a seventh explanatory diagram illustrating an experiment in an embodiment. More specifically, FIG. 19 is a second diagram illustrating an example of the results of inputting training data into a comparative learning model of a conventional technology and learning the comparative learning model. The vertical axis of FIG. 19 indicates coefficients. Specifically, the coefficients are coefficients used to calculate the rad score. The horizontal axis of FIG. 19 indicates Log Lambda. Log Lambda means the regularization coefficient.

[0152] Line L5 in Figure 19 indicates the regularization coefficient that minimizes the error in the prognosis prediction results from the trained comparative learning model. Each value enclosed in box W5 in Figure 19 represents a regularization coefficient. Note that each graph in Figure 19 indicates the coefficient of each graph quantity input to the prognosis prediction learning model. Therefore, even graphs with the same color scheme or line type (i.e., even if they exist on the same line) will show different graph quantity coefficients for each point on the horizontal axis.

[0153] 16 and 17 and the results of Figures 18 and 19 show that the error in the prognosis prediction result by the prognosis prediction system 100 converges more than the error in the prognosis prediction result by the conventional technology. Because the error in the prognosis prediction result by the prognosis prediction system 100 converges more than the error in the prognosis prediction result by the conventional technology, the prognosis prediction system 100 is more useful than the conventional technology.

[0154] FIG. 20 is an eighth explanatory diagram illustrating an experiment in an embodiment. The row "Graph Theory Features" in the table of FIG. 20 indicates information about the graph quantities used in the prognosis prediction system 100 in the experiment. The row "Conventional Features" in the table of FIG. 20 indicates information about the radiomics features used in the conventional technology in the experiment. The column "Selected Features" in the table of FIG. 20 indicates the features (graph quantities) selected in the prognosis prediction system 100 in the experiment and the features selected in the comparative learning model. The column "Coefficients" in the table of FIG. 20 indicates the weights of the features selected in the trained prognosis prediction learning model used in the experiment and the weights of the features selected in the trained comparative learning model used in the experiment.

[0155] "Vertices_number of vertices" in Figure 20 means that the graph quantity is the first graph feature. "Edges_number of edges with the same value as the threshold_threshold 10" in Figure 20 means that the graph quantity is the sixth graph feature. "Histogram_number of edges with the same value as the threshold_approximate curve y-intercept" in Figure 20 means that the graph quantity is the nineteenth graph feature, each element of which is a second-type individual graph feature. The table in Figure 20 shows that the weight of the graph quantity "Vertices_number of vertices" is 5.259×10 -6 The table in Figure 20 shows that the graph quantity weight for "Edges - Number of edges with the same value as threshold - Threshold 10" is 9.591 × 10 -2 The table in FIG. 20 shows that the weight of the graph quantity of “histogram_number of edges having the same value as the threshold_y-intercept of the approximate curve” was 3.532.

[0156] The table in Figure 20 shows that the weight of Maximum2DDiameterRow / shape is 9.924 × 10 -3 The table in Figure 20 shows that the weight of ZoneVariance / glszm was 1.917 × 10 -7 The table in Figure 20 shows that the Complexity / ngtdm weight was 7.173 × 10 -5 This indicates that the results were as follows: Maximum2DDiameterRow / shape is defined as the maximum pairwise Euclidean distance between vertices of the tumor surface in the sagittal plane. ZoneVariance / glszm is defined as the variation in size of adjacent voxels with the same pixel value. Complexity / ngtdm is defined as the complexity of the matrix that quantifies the difference in the average gray level between a certain gray level and its neighbors within a certain distance.

[0157] 21 and 22 show an example of an image of a tumor with a relatively good prognosis and an example of an image of a tumor with a relatively poor prognosis.

[0158] Fig. 21 is a ninth explanatory diagram illustrating an experiment in an embodiment. More specifically, Fig. 21 is a diagram showing an example of an image of a tumor having a better prognosis than the tumor shown in Fig. 22. Fig. 21 shows images G5 and G6. Image G5 is an example of an image of a tumor having a better prognosis than the tumor shown in Fig. 22. Image G6 is an example of a graph of the tumor shown in image G5.

[0159] In the experiment, the LAD score for the tumor shown in image G5 obtained by prognosis prediction system 100 was 0.45. Meanwhile, in the experiment, the LAD score for the tumor shown in image G5 obtained by conventional techniques was 0.94.

[0160] Fig. 22 is a tenth explanatory diagram illustrating an experiment in an embodiment. More specifically, Fig. 22 is a diagram showing an example of an image of a tumor having a worse prognosis than the tumor shown in Fig. 21. Fig. 22 shows images G7 and G8. Image G7 is an example of an image of a tumor having a worse prognosis than the tumor shown in Fig. 21. Image G8 is an example of a graph of the tumor shown in image G7.

[0161] In the experiment, the LAD score for the tumor shown in image G7 obtained by prognosis prediction system 100 was 0.80. 0.80 is 1.8 times 0.45. On the other hand, in the experiment, the LAD score for the tumor shown in image G7 obtained by conventional technology was 1.08. 1.08 is 1.1 times 0.94. Thus, in the experiment, the LAD score obtained by prognosis prediction system 100 showed a clearer change in tumor status than the LAD score obtained by conventional technology.

[0162] 21 and 22 also show that tumors with a relatively poor prognosis are more heterogeneous internally than tumors with a relatively good prognosis. Figures 21 and 22 also show that tumors with a relatively poor prognosis have a higher density of edges within the graph. Thus, Figures 21 and 22 show that by displaying a tumor graph, a user of the prognosis prediction system 100 can visually estimate the prognosis.

[0163] Therefore, the output unit 25 included in the prognosis prediction device 2 may display a graph of the subject tumor. By the output unit 25 displaying the graph of the subject tumor, the prognosis prediction device 2 has the effect of enabling the user to visually estimate the prognosis.

[0164] 23 and 24, the effect of the prognosis prediction system 100 on evaluation using a Kaplan-Meier curve will be described. In the experiment, a Rad score was obtained for each test data by inputting the test data into the prognosis prediction device 2 in advance. In the experiment, the test data was classified using the median of the distribution of the obtained Rad scores as a boundary into test data with Rad scores lower than the median (hereinafter referred to as "low score data") and test data with Rad scores equal to or higher than the median (hereinafter referred to as "high score data").

[0165] In the experiment, whether there is a significant difference in survival time between a set of low-score data and a set of high-score data was evaluated using a Kaplan-Meier curve. In the experiment, evaluation of whether there is a significant difference in survival time using a Kaplan-Meier curve was performed for each of the Ladd scores obtained by the prognosis prediction system 100 and the Ladd scores obtained by the conventional technology.

[0166] FIG. 23 is an eleventh explanatory diagram illustrating an experiment in an embodiment. More specifically, FIG. 23 shows the results of evaluating whether there is a significant difference in survival time using a Kaplan-Meier curve for the Ladd scores obtained by the prognosis prediction system 100. The horizontal axis of FIG. 23 represents survival time. The vertical axis of FIG. 23 represents survival probability. The "Low Risk" line in FIG. 23 represents the survival curve for the set of low-score data. The "High Risk" line in FIG. 23 represents the survival curve for the set of high-score data. For the results of FIG. 23, the p-value in the Kaplan-Meier curve was 0.004.

[0167] FIG. 24 is a twelfth explanatory diagram illustrating an experiment in an embodiment. More specifically, FIG. 24 shows the results of evaluating whether there is a significant difference in survival time using Kaplan-Meier curves for Rad scores obtained using conventional technology. The horizontal axis of FIG. 24 represents survival time. The vertical axis of FIG. 24 represents survival probability. The "Low Risk" line in FIG. 24 represents the survival curve for the low-score data set. The "High Risk" line in FIG. 24 represents the survival curve for the high-score data set. For the results in FIG. 24, the p-value in the Kaplan-Meier curve was 0.03.

[0168] 23 and 24 show that the prognosis prediction system 100 can show significant differences in survival time using Kaplan-Meier curves in the same way as the conventional technology.

[0169] In the experiment, the accuracy of prognosis prediction was also evaluated using the C-index, which is an index of prognosis prediction accuracy, with a value closer to 1 indicating higher accuracy and a value closer to 0.5 indicating lower accuracy.

[0170] In the experiment, the C-index was calculated using test data. In the experiment, the C-index value for the accuracy of prediction by the prognosis prediction system 100 was 0.679, and the p-value in the log-rank test was 0.004. In the experiment, the C-index value for the accuracy of prediction by the conventional technology was 0.625, and the p-value in the log-rank test was 0.01. Thus, the experiment showed that the accuracy of prediction by the prognosis prediction system 100 was higher than the accuracy of prediction by the conventional technology.

[0171] The prognosis prediction information acquisition device 1 of the embodiment configured as described above acquires prognosis prediction information, which is information indicating the relationship between the graph tumor information and prognosis, using the graph tumor information. Therefore, the prognosis prediction information acquisition device 1 can improve the accuracy of prognosis prediction for patients suffering from tumor diseases.

[0172] The prognosis prediction device 2 of the embodiment configured as described above predicts a prognosis using prognosis prediction information, which is information showing the relationship between the graph tumor information and the prognosis, using the graph tumor information. Therefore, the prognosis prediction device 2 can improve the accuracy of prognosis prediction for patients suffering from tumor diseases.

[0173] The prognosis prediction system 100 of the embodiment configured in this manner uses the graph tumor information to obtain prognosis prediction information, which is information showing the relationship between the graph tumor information and prognosis. Therefore, the prognosis prediction system 100 can improve the accuracy of prognosis prediction for patients suffering from tumor diseases. Furthermore, the prognosis prediction system 100 predicts prognosis using the prognosis prediction information, which is information showing the relationship between the graph tumor information and prognosis, using the graph tumor information. Therefore, the prognosis prediction system 100 can improve the accuracy of prognosis prediction for patients suffering from tumor diseases.

[0174] (Variation) The graph quantity may be, for example, a graph quantity obtained from a graph in which the graph quantity obtained for each region as a result of dividing a tumor image into a plurality of regions is used as the weight of the vertices.

[0175] Fig. 25 is an explanatory diagram illustrating an example of graph quantities in a modified example. Fig. 25 shows that a tumor image is divided into nine regions K1 to K9. Fig. 25 shows that a graph is generated for each of the regions K1 to K9, and that a graph having vertices with the graph quantities obtained from the generated graph as weights is generated.

[0176] The graph quantity used by the prognosis prediction system 100 may be the first graph feature quantity alone. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.657. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.085, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.05.

[0177] The graph quantity used by the prognosis prediction system 100 may be the fifth graph feature only. In this case, according to the experiment using the test data of the 91 cases described above, the C-index of the prognosis prediction system 100 was 0.632. Furthermore, according to the experiment using the test data of the 91 cases described above, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.644, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.01.

[0178] The graph quantity used by the prognosis prediction system 100 may be the second graph feature quantity alone. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.557. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.030, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.328.

[0179] The graph quantity used by the prognosis prediction system 100 may be the seventh graph feature only. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.546. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.046, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.135.

[0180] The graph quantity used by the prognosis prediction system 100 may be only the sixth graph feature. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.685. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.000, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.010.

[0181] The graph quantity used by the prognosis prediction system 100 may be the fourth graph feature alone. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.641. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.118, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.044.

[0182] The graph quantity used by the prognosis prediction system 100 may be the third graph feature only. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.415. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.328, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.349.

[0183] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 10 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.490. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.293, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.517.

[0184] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 10 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.479. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.390, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.875.

[0185] If the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 10 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.665. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.001, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.006.

[0186] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 10 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.636. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.118, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.108.

[0187] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 10 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.403. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.447, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.180.

[0188] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.431. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.896, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.780.

[0189] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.424. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.990, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.780.

[0190] If the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 15 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.589. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.021, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.459.

[0191] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.631. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.112, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.118.

[0192] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.387. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.575, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.156.

[0193] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 20 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.606. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.687, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.243.

[0194] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 20 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.612. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.643, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.078.

[0195] If the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 20 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.524. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.187, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.799.

[0196] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 20 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.631. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.115, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.118.

[0197] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 20 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.375. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.660, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.156.

[0198] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.629. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.498, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.016.

[0199] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.633. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.486, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.007.

[0200] In addition, when the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.464. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.933, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.854.

[0201] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.629. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.117, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.118.

[0202] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.369. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.700, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.063.

[0203] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.641. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.445, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0204] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.640. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.467, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0205] In addition, when the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.611. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.139, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.117.

[0206] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.629. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.122, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.118.

[0207] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.363. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.696, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.021.

[0208] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 35 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.644. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.480, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.005.

[0209] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 35 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.647. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.484, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.005.

[0210] In addition, when the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 35 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.606. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.460, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.317.

[0211] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 35 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.627. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.153, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.118.

[0212] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 35 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.364. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.651, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.021.

[0213] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.645. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.530, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0214] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.646. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.553, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0215] In addition, when the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.643. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.128, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.031.

[0216] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.628. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.157, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.118.

[0217] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.363. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.598, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.022.

[0218] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 45 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.625, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0219] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 45 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.661, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.015.

[0220] If the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 45 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.663. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.054, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.029.

[0221] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 45 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.628. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.166, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.048.

[0222] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 45 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.365. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.527, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.038.

[0223] In addition, when the graph quantity used by the prognosis prediction system 100 is only the second graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.643. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.756, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.012.

[0224] In addition, when the graph quantity used by the prognosis prediction system 100 is only the seventh graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.641. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.784, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.012.

[0225] In addition, when the graph quantity used by the prognosis prediction system 100 is only the sixth graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.667. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.051, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.007.

[0226] In addition, when the graph quantity used by the prognosis prediction system 100 is only the fourth graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.626. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.177, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.048.

[0227] In addition, when the graph quantity used by the prognosis prediction system 100 is only the third graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.369. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.450, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.038.

[0228] The graph quantity used by the prognosis prediction system 100 may be the ninth graph feature only. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.632. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.047, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.077.

[0229] The graph quantity used by the prognosis prediction system 100 may be only the eighth graph feature. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.557. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.030, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.328.

[0230] If the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 10 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.041, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.040.

[0231] In addition, when the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 10 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.490. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.293, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.517.

[0232] In addition, when the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.652. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.039, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.031.

[0233] In addition, when the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.431. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.896, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.780.

[0234] If the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 20 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.659. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.039, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.058.

[0235] In addition, when the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 20 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.606. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.687, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.243.

[0236] If the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 25 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.663. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.040, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.047.

[0237] In addition, when the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.629. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.498, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.016.

[0238] In addition, when the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.671. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.042, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.049.

[0239] If the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 30 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.641. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.445, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0240] In addition, when the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 35 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.675. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.046, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.049.

[0241] If the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 35 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.644. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.480, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0242] If the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 40 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.674. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.050, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.049.

[0243] In addition, when the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.645. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.530, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0244] If the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 45 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.673. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.059, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.049.

[0245] If the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 45 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.625, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0246] In addition, when the graph quantity used by the prognosis prediction system 100 is only the ninth graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.671. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.072, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.049.

[0247] If the graph quantity used by the prognosis prediction system 100 is only the eighth graph feature, the edge threshold value may be 50 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.643. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.756, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.012.

[0248] The graph quantity used by the prognosis prediction system 100 may be the tenth graph feature only. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.408. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.551, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.244.

[0249] The graph quantity used by the prognosis prediction system 100 may be the 12th graph feature only. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.644. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.249, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.024.

[0250] The graph quantity used by the prognosis prediction system 100 may be the 11th graph feature only. In such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.619. Furthermore, in such a case, when the edge threshold value is 5 HU, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.009, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.033.

[0251] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 10 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.446. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.320, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.710.

[0252] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 10 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.640. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.247, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.017.

[0253] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 10 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.560. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.037, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.430.

[0254] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.425. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.753, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.260.

[0255] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.640. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.207, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.009.

[0256] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 15 HU. In such a case, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.496. In addition, according to the experiment using the above-mentioned 91 test data, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.405, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.697.

[0257] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 20 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.611. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.682, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.188.

[0258] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 20 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.177, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.009.

[0259] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 20 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.577. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.749, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.369.

[0260] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.633. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.356, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.030.

[0261] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.640. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.170, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.009.

[0262] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 25 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.616. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.287, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.085.

[0263] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.643. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.256, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0264] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.165, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.009.

[0265] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 30 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.636. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.165, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.006.

[0266] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 35 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.649. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.211, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.015.

[0267] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 35 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.641. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.181, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.048.

[0268] If the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 35 HU. In this case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.643. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.139, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.005.

[0269] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.650. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.204, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.015.

[0270] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.641. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.164, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.050.

[0271] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 40 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.649. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.131, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.015.

[0272] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 45 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.646. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.217, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.015.

[0273] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 45 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.640. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.165, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.039.

[0274] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 45 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.649. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.146, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.015.

[0275] In addition, when the graph quantity used by the prognosis prediction system 100 is only the tenth graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. In addition, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.260, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.017.

[0276] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 12th graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the log-rank test of the prognosis prediction system 100 was 0.156, and the p-value of the Kaplan-Meier curve of the prognosis prediction system 100 was 0.039.

[0277] In addition, when the graph quantity used by the prognosis prediction system 100 is only the 11th graph feature, the edge threshold value may be 50 HU. In such a case, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.642. Furthermore, according to the experiment using the above-mentioned 91 test data cases, the p-value of the prognosis prediction system 100 in the Log-rank test was 0.185, and the p-value of the prognosis prediction system 100 in the Kaplan-Meier curve was 0.015.

[0278] Hereinafter, an 18th graph feature in which each element of the second-type individual graph quantity is a second graph feature and the slope is the slope of a linear term will be referred to as an 18-1 graph feature. Hereinafter, a 19th graph feature in which each element of the second-type individual graph quantity is a second graph feature will be referred to as a 19-1 graph feature. Note that the slope of a linear term is the slope of the linear function term of the function obtained when a histogram created from the graph quantity is approximated by a cubic polynomial. Note that the histogram is a histogram in which the vertical axis represents the value of the graph quantity and the horizontal axis represents the edge threshold.

[0279] The graph quantity used by the prognosis prediction system 100 may be only the 18-1 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.683. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.001, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.002.

[0280] The graph quantity used by the prognosis prediction system 100 may be only the 19-1 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.617. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.005, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.095.

[0281] Hereinafter, an 18th graph feature whose elements in the second type individual graph quantity are the 7th graph feature and whose slope is the slope of a linear term will be referred to as an 18-2 graph feature. Hereinafter, a 19th graph feature whose elements in the second type individual graph quantity are the 7th graph feature will be referred to as a 19-2 graph feature. Hereinafter, a 19th graph feature whose elements in the second type individual graph quantity are the 6th graph feature will be referred to as a 19-3 graph feature.

[0282] The graph quantity used by the prognosis prediction system 100 may be only the 18-2 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.682. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.001, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.004.

[0283] The graph quantity used by the prognosis prediction system 100 may be only the 19-2 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.600. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.007, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.282.

[0284] The graph quantity used by the prognosis prediction system 100 may be only the 19-3 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.682. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.000, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.008.

[0285] Hereinafter, an 18th graph feature whose elements included in the second type individual graph quantity are the fourth graph feature and whose slope is the slope of a linear term will be referred to as an 18-3 graph feature. Hereinafter, a 19th graph feature whose elements included in the second type individual graph quantity are the fourth graph feature will be referred to as a 19-4 graph feature.

[0286] The graph quantity used by the prognosis prediction system 100 may be only the 18-3 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.623. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.269, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.093.

[0287] The graph quantity used by the prognosis prediction system 100 may be only the 19-4 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.650. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.134, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.059.

[0288] Hereinafter, an 18th graph feature whose elements included in the second type individual graph quantity are third graph features and whose slope is the slope of a cubic term will be referred to as an 18-4 graph feature. Hereinafter, an 18th graph feature whose elements included in the second type individual graph quantity are third graph features and whose slope is the slope of a quadratic term will be referred to as an 18-5 graph feature. Hereinafter, a 19th graph feature whose elements included in the second type individual graph quantity are third graph features will be referred to as a 19-5 graph feature.

[0289] The slope of the cubic term is the slope of the cubic function term of the function obtained when a histogram created from graph quantities is approximated with a cubic polynomial. The slope of the quadratic term is the slope of the quadratic function term of the function obtained when a histogram created from graph quantities is approximated with a cubic polynomial.

[0290] The graph quantity used by the prognosis prediction system 100 may be only the 18-4 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.541. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.322, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.678.

[0291] The graph quantity used by the prognosis prediction system 100 may be only the 18-5 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.612. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.018, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.137.

[0292] The graph quantity used by the prognosis prediction system 100 may be only the 19-5 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.428. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.218, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.595.

[0293] Hereinafter, an 18th graph feature whose elements included in the second type individual graph quantity are 9th graph features and whose slope is the slope of a linear term will be referred to as an 18-6 graph feature. Hereinafter, a 19th graph feature whose elements included in the second type individual graph quantity are 9th graph features will be referred to as a 19-6 graph feature.

[0294] The graph quantity used by the prognosis prediction system 100 may be only the 18-6 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.592. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.121, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.426.

[0295] The graph quantity used by the prognosis prediction system 100 may be only the 19-6 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.627. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.051, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.096.

[0296] Hereinafter, an 18th graph feature whose elements included in the second type individual graph quantity are the 8th graph feature and whose slope is the slope of a linear term will be referred to as an 18-7 graph feature. Hereinafter, a 19th graph feature whose elements included in the second type individual graph quantity are the 8th graph feature will be referred to as a 19-7 graph feature.

[0297] The graph quantity used by the prognosis prediction system 100 may be only the 18-7th graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.683. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.001, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.002.

[0298] The graph quantity used by the prognosis prediction system 100 may be only the 19-7th graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.617. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.005, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.095.

[0299] Hereinafter, an 18th graph feature whose elements in the second type individual graph quantity are the 10th graph feature and whose slope is a cubic term will be referred to as an 18-8 graph feature. Hereinafter, an 18th graph feature whose elements in the second type individual graph quantity are the 10th graph feature and whose slope is a quadratic term will be referred to as an 18-9 graph feature. Hereinafter, an 18th graph feature whose elements in the second type individual graph quantity are the 10th graph feature and whose slope is a linear term will be referred to as an 18-10 graph feature.

[0300] The graph quantity used by the prognosis prediction system 100 may be only the 18-8 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.653. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.002, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.046.

[0301] The graph quantity used by the prognosis prediction system 100 may be only the 18-9th graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.637. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.004, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.064.

[0302] The graph quantity used by the prognosis prediction system 100 may be only the 18th-10th graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.581. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.034, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.223.

[0303] Hereinafter, a 19th graph feature whose elements included in the second type individual graph quantity are the 10th graph feature will be referred to as a 19-8 graph feature. Hereinafter, a 19th graph feature whose elements included in the second type individual graph quantity are the 12th graph feature will be referred to as a 19-9 graph feature.

[0304] The graph quantity used by the prognosis prediction system 100 may be only the 19-8 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.622. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.958, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.016.

[0305] The graph quantity used by the prognosis prediction system 100 may be only the 19-9th graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.647. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.293, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.039.

[0306] Hereinafter, an 18th graph feature whose elements in the second type individual graph quantity are the 11th graph feature and whose slope is the slope of a cubic term will be referred to as an 18-11 graph feature. Hereinafter, an 18th graph feature whose elements in the second type individual graph quantity are the 11th graph feature and whose slope is the slope of a quadratic term will be referred to as an 18-12 graph feature. Hereinafter, an 18th graph feature whose elements in the second type individual graph quantity are the 11th graph feature and whose slope is the slope of a linear term will be referred to as an 18-13 graph feature.

[0307] The graph quantity used by the prognosis prediction system 100 may be only the 18th-11th graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.576. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.216, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.336.

[0308] The graph quantity used by the prognosis prediction system 100 may be only the 18th-12th graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the C-index of the prognosis prediction system 100 was 0.533. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.973, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.439.

[0309] The graph quantity used by the prognosis prediction system 100 may be only the 18th-13th graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.599. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.008, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.042.

[0310] Hereinafter, a 19th graph feature whose elements included in the second type individual graph quantity are the 11th graph feature will be referred to as a 19-10th graph feature. Hereinafter, a 20th graph feature whose elements included in the second type individual graph quantity are the 10th graph feature will be referred to as a 20-1 graph feature. Hereinafter, a 21st graph feature whose elements included in the second type individual graph quantity are the 10th graph feature will be referred to as a 21-1 graph feature.

[0311] The graph quantity used by the prognosis prediction system 100 may be only the 19-10 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.665. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.002, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.001.

[0312] The graph quantity used by the prognosis prediction system 100 may be only the 20-1 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.618. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.808, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.079.

[0313] The graph quantity used by the prognosis prediction system 100 may be only the 21-1 graph feature. In such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the C-index of the prognosis prediction system 100 was 0.577. Furthermore, in such a case, when the number of edge thresholds is 50, according to the experiment using the above-mentioned 91 test data cases, the p-value in the Log-rank test of the prognosis prediction system 100 was 0.958, and the p-value in the Kaplan-Meier curve of the prognosis prediction system 100 was 0.716.

[0314] It should be noted that the prognosis prediction information acquisition device 1 does not necessarily need to execute the learning dataset generation process. In such a case, a learning dataset is input to the prognosis prediction information acquisition device 1 instead of the pre-processing learning dataset. In such a case, the learning dataset acquisition unit 104 executes a process of acquiring a learning dataset input to the communication unit 12 or the input unit 13 instead of executing a process of generating a learning dataset based on the pre-processing learning dataset. If the learning dataset has been recorded in advance in the storage unit 14, the learning dataset acquisition unit 104 may acquire the learning dataset by reading it out from the storage unit 14.

[0315] It should be noted that the prognosis prediction device 2 does not necessarily need to perform preprocessing. In such cases, graph tumor information is input to the prognosis prediction device 2 instead of tumor image data. In such cases, the input data acquisition unit 240 acquires the graph tumor information instead of acquiring tumor image data. When the input data acquisition unit 240 acquires the graph tumor information, the prognosis prediction unit 250 executes the main processing without performing preprocessing. If the graph tumor information has been recorded in the memory unit 24 in advance, the input data acquisition unit 240 may acquire the graph tumor information by reading out the graph tumor information from the memory unit 24.

[0316] The prognosis prediction information acquisition device 1 may be implemented using a plurality of information processing devices communicably connected via a network. In this case, each functional unit of the prognosis prediction information acquisition device 1 may be distributed and implemented in the plurality of information processing devices.

[0317] The prognosis prediction device 2 may be implemented using a plurality of information processing devices communicably connected via a network. In this case, the respective functional units of the prognosis prediction device 2 may be distributed and implemented in the plurality of information processing devices.

[0318] It should be noted that the prognosis prediction information acquisition device 1 and the prognosis prediction device 2 do not necessarily need to be implemented as different devices, and the prognosis prediction information acquisition device 1 and the prognosis prediction device 2 may be implemented as a single device.

[0319] The prognosis prediction system 100 may be implemented using a plurality of information processing devices communicably connected via a network. In this case, the respective functional units of the prognosis prediction system 100 may be distributed and implemented in the plurality of information processing devices.

[0320] All or part of the functions of the prognosis prediction system 100 may be realized using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. The program may be transmitted via a telecommunications line.

[0321] Although an embodiment of the present invention has been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention. [Explanation of symbols]

[0322] 100...prognosis prediction system, 1...prognosis prediction information acquisition device, 2...prognosis prediction device, 11...control unit, 12...communication unit, 13...input unit, 14...memory unit, 15...output unit, 101...communication control unit, 102...input control unit, 103...output control unit, 104...learning dataset acquisition unit, 105...model execution unit, 106...update unit, 107...termination determination unit, 108...memory control unit, 21...control unit, 22...communication unit, 23...input unit, 24...memory unit, 25...output unit, 210...communication control unit, 220...input control unit, 230...output control unit, 240...input data acquisition unit, 250...prognosis prediction unit, 251...pre-processing execution unit, 252...main processing execution unit, 260...memory control unit, 91...processor, 92...Memory, 93...Processor, 94...Memory

Claims

1. a prognosis prediction unit that predicts the prognosis of a subject based on the graph tumor information showing an image of a tumor that the subject has, using graph tumor information, which is information on an image of a tumor expressed using quantities defined by graph theory, and prognosis prediction information, which is information showing the relationship between the prognosis of a person or animal having the tumor; Equipped with The prognosis prediction information does not include information regarding an image of a cell nucleus. Prognostic prediction device.

2. The quantities defined in the graph theory are quantities related to the vertices of the graph. The prognosis prediction device according to claim 1 .

3. The vertices of the graph are defined for each unit pixel, which is a group of one or more adjacent pixels. The prognosis prediction device according to claim 2 .

4. The quantity related to the vertices of the graph is a quantity indicating the positional relationship between vertices whose weights are defined and the weight of each vertex. The prognosis prediction device according to claim 3 .

5. The weight is a predefined index value related to the pixel value of each pixel included in the unit pixel. The prognosis prediction device according to claim 4 .

6. The quantities defined in the graph theory are quantities related to the edges of the graph. The prognosis prediction device according to claim 1 .

7. The quantity defined in the graph theory is a quantity related to the degree of the graph. The prognosis prediction device according to claim 1 .

8. The quantity defined in the graph theory is a quantity related to the minimum spanning tree of the graph. The prognosis prediction device according to claim 1 .

9. a prognosis prediction step in which a computer predicts the prognosis of a subject based on the graph tumor information showing an image of a tumor in the subject, using graph tumor information, which is information on tumor images expressed using quantities defined in graph theory, and prognosis prediction information, which is information showing the relationship between the prognosis of a person or animal having the tumor; and The prognosis prediction information does not include information regarding an image of a cell nucleus. Prognostic methods.

10. A program for causing a computer to function as the prognosis prediction device according to any one of claims 1 to 8.

Images