JP7706883B2 - Information processing device, information processing method, information processing system, and program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, an information processing system, and a program that select an inference model based on a first inference result.

There is a medical system known that applies inference models based on machine learning technology to medical data acquired from medical imaging devices and medical data acquired from medical information systems, and makes inferences about specific diseases (disease detection, benign/malignant discrimination, prognosis prediction, risk prediction, etc.).

Patent document 1 discloses a technology in which, when a user specifies a type of a given disease, an appropriate inference model for the given disease is selected according to the specified type.

JP 2016-193295 A

However, with the technology in Patent Document 1, if a user wanted to make inferences about multiple diseases from medical data, the user had to specify the disease type for each of the multiple diseases, which was time-consuming.

The present invention aims to provide an information processing device, an information processing method, and a program that can select an inference model based on a first inference result for the medical data to be inferred.

To solve the above problem, an information processing device according to one aspect of the present invention has a selection unit that acquires a first inference result based on a first inference for medical data and selects an inference model to be applied to the medical data based on the first inference result, and an output unit that outputs information on the selected inference model.

According to the present invention, an inference model can be selected based on the results of a first inference on the medical data to be inferred.

FIG. 1 is a diagram showing the arrangement of an information processing apparatus according to a first embodiment. FIG. 1 is a diagram showing a hardware configuration of an information processing apparatus according to a first embodiment. FIG. 4 is a flowchart showing processing of the information processing apparatus according to the first embodiment. FIG. 11 is a flowchart showing a process of an information processing apparatus according to a second embodiment. FIG. 11 is a tree diagram illustrating the relationship between diseases according to the second embodiment. FIG. 11 is a flowchart showing a process of an information processing apparatus according to a third embodiment.

A preferred embodiment of the present invention will now be described with reference to the accompanying drawings.

[First embodiment]
The present invention is preferably applicable to medical data such as raw data captured by a medical imaging device and diagnostic image data generated by image reconstruction from the raw data. Examples of medical imaging devices that acquire medical data include X-ray CT devices, MRI devices, SPECT devices, and PET devices. The medical data may be composed of not only image data, but also waveform data such as electrocardiograms, voice data such as medical interviews, text data such as electronic medical records, blood test results, and other data.

The inference model described below is, for example, a pre-trained model that has been machine-trained to infer a specific disease from medical data that is the subject of inference. Disease inference refers to performing inference tasks such as classifying diseases, detecting disease areas, and extracting disease areas for medical image data. The class of the inference model corresponds to the subject that the inference model infers, and can be set appropriately during learning. A pre-trained model is an inference model that has undergone training in advance, and additional training can also be performed as appropriate. The medical data input to the inference model may be raw data captured by a medical imaging device, three-dimensional volume data, two-dimensional data, or one-dimensional data.

Figure 1 shows the configuration of an information processing device according to the first embodiment. Here, as an example, chest CT image data captured by an X-ray CT device will be described as medical data to be inferred. Also, as an example, a case will be given in which the presence or absence of cancer is inferred by a first inference, and the presence or absence of bleeding, which is a complication of cancer, is inferred by a second inference.

The information processing device 1 includes an acquisition unit 10 that acquires medical data to be inferred, an inference unit 11 that performs inference on the medical data, a selection unit 12 that selects an inference model to be applied to the medical data based on the inference result for the medical data, and an output unit 13 that outputs the selected inference model and the inference result by the selected inference model. The information processing device 1 is connected to a data server 2 via a network. The data server 2 is a computer system (e.g., PACS) that accumulates and shares image data for image diagnosis. The information processing device 1 is also connected to a display device 3 that displays output information by the output unit 13 and an input device 4 for inputting various operations by a user to the information processing device 1. The display device 3 and the input device 4 may each be composed of an independent device. Alternatively, the display device 3 and the input device 4 may be composed of a device integrated with the information processing device 1. The configuration of each part of the information processing device 1 will be described below. Here, the display device 3 is, for example, a display attached to the information processing device 1 or a mobile terminal of a hospital staff member via an external server, and the input device 4 is a mouse, a keyboard, or the like. Each component of the information processing device 1 may function as an information processing system made up of multiple devices.

(Acquisition unit 10)
The acquisition unit 10 acquires medical data of the inference target and transmits the acquired medical data to the inference unit 11. The medical data of the inference target acquired by the acquisition unit 10 may be medical data selected by the user via the input device 4, or medical data transmitted from the data server 2 or a medical imaging device. Furthermore, the acquisition unit 10 acquires selection information of a first inference model for performing a first inference from the user and transmits it to the inference unit 11. In this embodiment, a process will be described in which a first inference model including cancer as a class as the first disease is selected by the user via the input device 4. Here, the selection of the first inference model is not limited to the selection by the user, and an inference model may be selected with reference to a selection condition based on the acquired medical data of the inference target, or an inference model including a disease with a high probability of existence as a class may be selected by applying image processing or the like to the medical data of the inference target.

(Inference part 11)
The inference unit 11 performs inference by applying the selected first inference model to the medical data transmitted from the acquisition unit 10. The inference unit 11 performs a first inference on the medical data to be inferred, using the first inference model transmitted from the acquisition unit 10. Here, the first inference is performed by an inference model that infers cancer. The result of the first inference in the inference unit 11 is transmitted to the selection unit 12.

The inference unit 11 also performs a second inference on the medical data to be inferred using a second inference model selected by the selection unit 12 described later, and transmits the inference result to the output unit 13. The inference result transmitted by the inference unit 11 to the output unit 13 is, for example, information including information indicating the presence of a disease, the location or area of the disease, and the like. Here, the second inference is an inference using a model selected by the selection unit 12 to infer, for example, the presence or absence of bleeding. Bleeding, which is a second disease, is a complication of cancer, which is a first disease. When cancer is detected as the first disease in the inference unit 11, inference is performed using the inference model selected by the selection unit 12 with the complication of cancer as the second disease. In addition, the relationship between the first disease and the second disease is not limited to this, and for example, when cancer is detected as the first disease, inference may be performed using an inference model that determines the benignity or malignancy of cancer as the second disease. In addition, inference is performed using a model that infers the type of the first disease detected as the second disease in response to the detection of the first disease. A configuration in which the selection unit 12 selects the second inference model based on the relationship between the first disease and the second disease will be described later.

Note that the inference unit 11 is not a required component of the information processing device 1, and may be replaced by, for example, obtaining an inference result from another device.

The inference model that the inference unit 11 applies to the medical data to be inferred is, for example, an artificial intelligence model such as a deep neural network. The inference model used by the inference unit 11 is an artificial intelligence model that has been trained in advance to infer a specific disease from the medical data to be inferred, using a machine learning method such as deep learning.

The inference unit 11 may use an inference model that infers a disease from raw data captured by a medical imaging device as an inference model to be applied to the medical data to be inferred. The inference unit 11 may also use an inference model that infers a disease from two-dimensional data such as tomographic image data, or a model that identifies a disease from volume data such as three-dimensional image data. Alternatively, the inference unit 11 may use an inference model that identifies a disease from time-series data (moving image data or image data of multiple time phases).

The inference task performed by the inference model in the inference unit 11 may be to classify medical data, to detect a classification target from within medical data, or to extract a region of a classification target. Depending on the inference task performed by the inference model in the inference unit 11, the inference unit 11 may output the presence or absence of a disease for the medical data to be inferred, or may output an ROI (Region of Interest) or VOI (Volume of Interest) including the disease region, or may output a bounding box in the form of a rectangle or a cuboid that surrounds the disease region.

In this embodiment, the inference unit 11 uses an inference model configured to infer the presence or absence of a disease from three-dimensional medical image data (volume data).

The learning data used to train the inference model is set according to the inference task. For example, in the present embodiment, when training an inference model that infers the presence or absence of a disease from three-dimensional medical image data, learning data consisting of medical image data and a label (teacher signal) indicating a specific disease area contained in the medical image data is used. The label here corresponds to the class (disease) to be detected from the medical data. The class corresponding to the label can be inferred by the inference model trained by pairing the label and the medical data.

When the inference task is detection, the label is information that indicates the ROI or VOI that contains the disease to be detected from the medical image data.

In this embodiment, chest CT image data collected from a large number of subjects (e.g., subjects with cancer or subjects with bleeding) is used as learning data. Note that, although a trained model that has been trained in advance is implemented in the information processing device 1, this is not limiting, and the information processing device 1 may have a function of performing additional learning using additional training data.

(Selection Unit 12)
The selection unit 12 selects a second inference model to be applied to the medical data to be inferred based on the first inference result by the first inference. Note that the selection unit 12 of the information processing device 1 can select the second inference model to be applied to the medical data if the inference result by the first inference can be acquired. Therefore, the selection of the first inference model described above and the inference by the first inference model may be performed by another device.

The selection unit 12 acquires the first inference result inferred by the inference unit 11 using the first inference model including the first disease (cancer) as a class, selected by the user and acquired by the acquisition unit 10. When the first inference result transmitted from the inference unit 11 is information indicating that the first disease (cancer) has been detected in the medical data to be inferred, the selection unit 12 selects a second inference model including a second disease different from the first disease as a class. Here, the first disease and the second disease are diseases having a clinical relationship with each other. The clinical relationship refers to, for example, a causal relationship between the first disease and the second disease, or a relationship in which there is a second disease that should be confirmed when the first disease is detected. The selection of the second inference model by the selection unit 12 can be automatically performed by an information processing device such as a computer, and the effort required for the user to make a manual selection can be reduced.

The selection unit 12 refers to relationship information that defines the relationship between the first disease and the second disease, and selects a second inference model that includes the second disease as a class. The relationship information is defined by a list, a directed graph, a tree diagram, or the like, and is held by the selection unit 12. The selection unit 12 holds the relationship between the first disease and the second disease in a list or the like, refers to the list based on the inference result obtained by the first inference, and if there is a matching disease, selects a second inference model that includes the second disease as a class. The selection unit 12 can select the second inference model by referring to relationship information that defines the clinical relationship between diseases.

The first disease and the second disease are, for example, cancer and bleeding, which is a complication of cancer. The selection unit 12 can select the second inference model by defining the first disease and the second disease suspected of being a complication as relationship information in advance. The first disease and the second disease selected by the selection unit 12 are, for example, cerebral infarction and aortic dissection. In order to administer t-PA, a thrombolytic drug, to cerebral infarction, it is necessary to confirm that aortic dissection has not occurred. Therefore, when cerebral infarction is detected as the first disease, the selection unit 12 selects an inference model that includes aortic dissection as a class as the second disease as the second inference model. Note that the relationship between the first disease and the second disease is not limited to this, and for example, when cancer is detected as the first disease, the selection unit 12 may select a second inference model that determines whether the cancer is benign or malignant as the second disease. In addition, in response to the detection of the first disease, the selection unit 12 may select a second inference model that infers the type of the detected first disease as the second disease. Here, in the information processing device 1, the selection unit 12 acquires a first inference result from a first inference on the medical data, and selects a second inference model to be applied to the medical data based on the first inference result. Furthermore, the output unit 13 described below outputs information on the inference model selected by the selection unit 12. With this configuration, an inference model can be selected based on the inference result on the medical data to be inferred, and the user's effort can be reduced even when multiple diseases are to be inferred from medical data.

(Output unit 13)
When the output unit 13 acquires information on the second inference model selected from the selection unit 12, it outputs information on the selected model to the display device 3. When the output unit 13 acquires an inference result for the medical data to be inferred from the inference unit 11, it outputs the inference result. The output unit 13 may also output, together with the inference result, information to identify the subject (subject information such as subject ID and name), the medical data to be inferred, and whether or not a next action should be taken based on the inference result (for example, in the case where cancer is present and bleeding occurs, presenting the bleeding area to encourage hemostasis).

(Hardware configuration of information processing device)
The information processing device 1 may be configured by a computer including a processor, a memory, a storage, and the like. In this case, the functions and processing of the acquisition unit 10, the inference unit 11, the selection unit 12, the output unit 13, and the like are realized by loading a program stored in the storage into the memory and executing the program by the processor. However, this configuration is not limited, and for example, all or part of the functions of the acquisition unit 10, the inference unit 11, the selection unit 12, and the output unit 13 may be realized by a dedicated processor (such as an ASIC) or an FPGA. Alternatively, part of the arithmetic processing may be executed by a processor such as a GPU or a DSP. In addition, the information processing device 1 may be configured by a single piece of hardware, or may be configured by multiple pieces of hardware. For example, the functions and processing of the information processing device 1 may be realized by multiple computers working together using cloud computing or distributed computing.

Figure 2 shows an example of a specific configuration of the information processing device 1. In this example, the information processing device 1 has a CPU 20, a GPU 21, a RAM 22, a ROM 23, and a storage device 24, which are connected by a system bus 25.

(Operation of information processing device)
Next, the process executed by the information processing device 1 will be described with reference to the flowchart in Fig. 3. Note that the steps constituting the flowchart may be performed by different devices, and the process result may be obtained without performing the process of each step, and the step may be regarded as having been processed.

In step S30, the acquisition unit 10 acquires medical data to be inferred from the data server 2. Here, for example, the acquisition unit 10 acquires three-dimensional X-ray CT image data (volume data) of the chest. The acquisition unit 10 also acquires information on a first inference model to be applied to the first inference for the medical data to be inferred from the user. For example, the user selects an inference model that infers cancer as the first disease as the first inference model.

Here, the first inference model may be selected not by user selection, but by comparison with pre-stored conditions, the type of medical data to be inferred, or by applying image processing to the medical data to be inferred. The selection of the first inference model may also be performed by a different device. When the acquisition unit 10 acquires the medical data to be inferred and information on the first inference model to perform the first inference, it transmits the acquired information to the inference unit 11, and the process proceeds to step S31.

In step S31, the inference unit 11 performs inference by applying the first inference model (a model for inferring cancer) to the medical data of the inference target transmitted from the acquisition unit 10, based on the medical data of the inference target and information on the first inference model. The inference unit 11 acquires the presence or absence of cancer as first disease information from the first inference model. After transmitting the first inference result based on the first inference model to the selection unit 12, the inference unit 11 proceeds to the process of step S32.

In addition, the information processing device 1 may obtain the processing results in steps S30 and S31 and perform processing from step S32 onwards.

In step S32, the selection unit 12 obtains a first inference result based on the first inference for the medical data. If a first disease is detected in the medical data, the process proceeds to step S33, and if a first disease is not detected in the medical data, the process ends.

In step S33, the selection unit 12 selects a second inference model to be applied to the medical data based on the first inference result. Here, the selection unit 12 identifies a second disease that is clinically related to the first disease detected by the first inference using one of the methods described above, and selects an inference model that includes the identified second disease as a class as the second inference model. The selection unit 12 can automatically select the second inference model using an information processing device. When the selection unit 12 selects the second inference model to be applied to the medical data, it transmits information about the selected second inference model to the inference unit 11, and proceeds to step S34. Here, the selection unit 12 refers to the list and selects an inference model that detects the presence or absence of bleeding, which is a second disease that is clinically related to cancer, as the second inference model.

When the selection unit 12 selects an inference model to be applied to the medical data, the selection unit 12 may output information about the selected inference model via the output unit 13. In addition, a different device may receive the output and perform the steps described below.

In step S34, the inference unit 11 acquires information about a second inference model that performs a second inference on the medical data that is the inference target selected by the selection unit 12. Based on the information about the second inference model, the inference unit 11 applies the second inference model to the medical data and performs a second inference to detect the presence or absence of bleeding. After performing the second inference, the inference unit 11 transmits the second inference result by the second inference to the output unit 13, and the process proceeds to step S35.

In step S35, the output unit 13 outputs the second inference result obtained from the inference unit 11. The output by the output unit 13 is presented to the user via the display device 3 connected to the information processing device 1. When the inference unit 11 detects bleeding in the medical data, the output unit 13 outputs the result. The output unit 13 may also output the inference result based on the first inference.

According to the information processing device 1 of this embodiment, a second inference model that includes a second disease different from the first disease in its class can be selected based on the result of the first inference regarding the first disease, thereby reducing the user's effort in selecting a model. This makes it possible to significantly reduce the time until the inference result for the second disease is displayed, compared to the case where the user selects multiple inference models corresponding to multiple diseases.

(Modification 1 of the first embodiment)
In the above embodiment, the case where cancer is taken as an example of the first disease and bleeding is inferred as the second disease has been described.

The inference subject of the present invention is not limited to this, and an inference model targeting another disease may be selected as the second inference model. For example, when cerebral hemorrhage is detected as the first disease, cerebral infarction is detected as the second disease. Furthermore, the inference subject is not limited to a disease.

The present invention can be applied when it is possible to select an inference model to be applied to medical data based on a first inference result for the medical data, and when the first inference target and the second inference target have a clinical relationship.

(Modification 2 of the First Embodiment)
In the above-described embodiment, a configuration has been described in which a single disease is treated as a first disease, and a first inference is performed using an inference model including the first disease as a class. The present invention is not limited to this, and a plurality of diseases may be targeted as the first disease. The inference unit 11 may perform the first inference using an inference model including each of a plurality of first diseases as a class, and the selection unit 12 may select a second inference model based on the first inference result. The selection unit 12 may also infer a plurality of first diseases using a plurality of inference models, and select an inference model for performing the second inference from each of the inference results by the plurality of inference models. The first inference using a plurality of inference models and the second inference selected by the selection unit 12 based on a plurality of first inference results may be performed in parallel or sequentially.

In addition, when multiple diseases are targeted as the first disease, the result of combining the first inference results performed by the inference unit 11 for the multiple diseases may be transmitted to the selection unit 12 as the first inference result, and the selection unit 12 may select an inference model for performing the second inference.

For example, if multiple cancer-related diseases such as bleeding, pleural effusion, and cardiac tamponade are set as the first disease, the inference unit 11 performs inference using an inference model that detects each of the multiple diseases. Next, the inference unit 11 combines the inference results for each of the multiple diseases and transmits them to the selection unit 12 as a first inference result.

The selection unit 12 selects an inference model for detecting the presence or absence of cancer as a second disease that is clinically related to bleeding, pleural effusion, and cardiac tannade. The inference unit 11 performs a second inference using the inference model for detecting the presence or absence of cancer selected by the selection unit 12.

In addition, the result regarding the detected disease among the multiple first diseases may be used as an input to the second inference by the inference unit 11. Also, the selection unit 12 may select an inference model that includes, as a class, a second disease that has a clinical relationship with the detected disease among the multiple first diseases.

(Modification 3 of the first embodiment)
In the above-described embodiment, the selection unit 12 refers to a list that defines a clinical relationship between a first disease and a second disease set in advance, and selects an inference model that includes, as a class, a second disease that is in a clinical relationship with the first disease detected by the first inference. The present invention is not limited to this, and the clinical relationship between the first disease and the second disease may be defined by a tree diagram, a directed graph, or the like.

In addition, the inference model selected by the selection unit 12 does not necessarily have to be selected according to a predefined clinical relationship. For example, the selection unit 12 may hold structured data corresponding to an inference model, and the inference model may be selected according to the structured data. For example, the selection unit 12 holds structured data of chest, lung, and cancer in association with a first inference model for detecting lung cancer. The selection unit 12 selects an inference model for performing a second inference based on structured data corresponding to the inference model that performed the first inference. Here, the selection unit 12 selects an inference model that includes at least one of chest, lung, and cancer in the structured data or class as the inference model for performing the second inference.

(Fourth Modification of the First Embodiment)
When the first inference model for inferring a first disease is changed, the selection unit 12 identifies a second disease that has a clinical relationship with the first disease corresponding to the changed first inference model, and selects a second inference model that includes the identified second disease as a class. Here, since the inference model is defined by the relationship between diseases such as the first disease and the second disease, the second inference model selected by the selection unit 12 before the change of the first inference model and the second inference model selected by the selection unit 12 after the change of the first inference model are different inference models. Note that even when the first inference model is changed, if the first disease before the change and the first disease after the change have a clinical relationship with a common second disease, the same inference model may be selected by the selection unit 12 as the second inference model.

Second Embodiment
In the second embodiment, a case will be described in which, when a first disease is present, the selection unit 12 selects a plurality of diseases as second diseases, and the inference unit 11 executes a second inference.

In this embodiment, the inference unit 11 uses an example of inferring the presence or absence of myocardial infarction from whole-body CT image data captured by an X-ray CT device, which is medical data to be inferred, using an inference model that detects myocardial infarction as a first disease. When myocardial infarction is detected, the selection unit 12 selects an inference model that detects pericarditis, ventricular aneurysm, and myocardial rupture, which are complications that have a clinical relationship with myocardial infarction, as a second inference model for performing a second inference.

Here, the inference model that performs the first inference and the second inference model that performs the second inference perform a segmentation task to extract the detected disease area, and the output unit 13 outputs to the display device 3, as the inference result, medical image data in which the disease area is highlighted for the medical image data that is the inference target.

Hereinafter, regarding the details of each component of the information processing device according to this embodiment, explanations of overlapping parts will be omitted as appropriate, and only parts different from the first embodiment will be explained. In this embodiment, the selection unit 12 selects multiple diseases as the second disease to be inferred.

(Selection Unit 12)
The selection unit 12 selects a second inference model for inferring a second disease based on the first inference result by the inference unit 11. The selection unit 11 holds in advance a tree diagram that defines the clinical relationship between the first disease and the second disease, and, when the first disease is present, identifies a plurality of second diseases that are clinically related to the first disease. The selection unit 12 also selects a plurality of inference models that include at least one of the identified plurality of second diseases as a class, and transmits information of the selected second inference model to the inference unit 11. The inference unit 11 uses the selected second inference model to perform a second inference on the medical data to be inferred.

(Operation of information processing device)
Next, the process executed by the information processing device 1 will be described with reference to the flowchart of Fig. 4 and the tree diagram of Fig. 5. The tree diagram of Fig. 5 is a diagram showing an example of defining clinical relationships between diseases. For example, for myocardial infarction, which is a first disease, the selection unit 12 holds pericarditis, ventricular aneurysm, and myocardial rupture, which are possible complications, as second diseases that are clinically related to myocardial infarction. Furthermore, pericarditis, which is the second disease, has a clinical relationship with cardiac tannade, which is a third disease.

In step S30, the acquisition unit 10 acquires medical data to be inferred. Here, the acquisition unit 10 acquires three-dimensional X-ray CT image data (volume data) of the chest and transmits it to the inference unit 11. The acquisition unit 10 also acquires information on a first inference model to be applied to the first inference on the medical data to be inferred from the user. Here, it is assumed that the user has selected an inference model that detects myocardial infarction as the first inference model. The acquisition unit 10 transmits the acquired information to the inference unit 11 and proceeds to step S31.

In step S31, the inference unit 11 performs a first inference to infer the presence or absence of myocardial infarction using an inference model for detecting myocardial infarction from medical data (medical image data) that is the subject of inference. After transmitting the first inference result to the selection unit, the inference unit 11 proceeds to step S42.

In step S42, the selection unit 12 obtains a first inference result based on a first inference model that includes the first disease as a class. If myocardial infarction is detected by the first inference, the process proceeds to step S43. If myocardial infarction is not detected by the first inference, the process ends.

In step S43, the selection unit 12 refers to the tree diagram and identifies a plurality of second diseases. Referring to FIG. 5, pericarditis, ventricular aneurysm, and myocardial rupture are defined in the tree diagram as diseases having a clinical relationship with myocardial infarction. Therefore, the selection unit 12 selects an inference model that infers pericarditis, ventricular aneurysm, and myocardial rupture as the second disease as the second inference model. Here, the selection unit 12 selects an inference model that includes at least one of the plurality of second diseases as a class. For example, the selection unit 12 may select an inference model that includes all of pericarditis, ventricular aneurysm, and myocardial rupture as classes as the second inference model. Alternatively, the selection unit 12 may select a plurality of inference models that detect each of pericarditis, ventricular aneurysm, and myocardial rupture as the second inference model. Here, the selection unit 12 selects three inference models that detect each of pericarditis, ventricular aneurysm, and myocardial rupture, transmits information on the selected inference model to the inference unit 11, and proceeds to step S44.

In step S44, the inference unit 11 applies the multiple inference models selected by the selection unit 12 to the medical data to be inferred, and performs a second inference. Inferences using multiple inference models may be processed in parallel or sequentially. The inference unit 11 transmits the results of the second inference to the output unit 13, and proceeds to step S45.

In step S45, the output unit 13 outputs the first inference result and the second inference result in a comparable manner. For example, the output unit 13 outputs medical image data in which the detected disease area is superimposed on the medical data (medical image data) to be inferred, to the display device 3. The medical data to be output may be three-dimensional volume data or two-dimensional tomographic image data.

According to the information processing device of this embodiment, the selection unit 12 can select multiple inference models corresponding to multiple second diseases based on the result of inference regarding the first disease. Furthermore, the selection by the selection unit 12 reduces the effort required for the user to select a model, while allowing the inference unit 11 to perform inference efficiently.

This makes it possible to significantly reduce the time it takes to display the inference results for the second disease, compared to when the user first infers myocardial infarction from the medical data to be inferred, and then selects inference models for pericarditis, ventricular aneurysm, and myocardial rupture and performs the inference process.

(Modification 1 of the second embodiment)
In the above embodiment, the first inference regarding the first disease and the second inference regarding the second disease have been described. The present invention is not limited to this embodiment, and in the case where there is a third disease having a clinical relationship with the second disease as defined in the tree diagram of Fig. 5, or in the case where there is another tree diagram that defines the clinical relationship between the third disease and another disease for the third disease having a clinical relationship with the second disease, the inference process defined in the tree diagram may be performed by the inference unit 11.

(Modification 2 of the second embodiment)
In the above embodiment, the selection unit 12 selects multiple inference models based on the first inference result, and each of the selected multiple inference models performs inference on the medical data to be inferred. The present invention is not limited to this embodiment, and the inference unit 11 may input the first inference result as additional information to a second inference model that performs a second inference, and perform a second inference regarding a second disease. This allows the information processing device 1 to perform inference with higher accuracy.

(Modification 3 of the second embodiment)
In this modification, the inference unit 11 may input the second inference result from the second inference as additional information to the inference model that performed the first inference, and the inference model that performed the first inference may perform the third inference. This improves the accuracy of the inference result obtained from the inference model that performed the first inference. In addition, in combination with the modification 2 of the second embodiment described above, it is possible to perform inference with even higher accuracy by performing inferences multiple times.

(Variation 4 of the Second Embodiment)
In the above-described embodiment, the selection unit 12 selects multiple inference models based on the first inference result, and the inference unit 11 performs the second inference using the selected multiple inference models. In this modified example, the selection unit 12 identifies multiple second diseases that are clinically related to the first disease detected by the first inference, and transmits the identified multiple second diseases to the output unit 13. The output unit 13 outputs information on the multiple second diseases, and the user may select a disease to be the target of the second inference from the information on the multiple second diseases via the input device 4.

[Third embodiment]
In the information processing device 1 according to the third embodiment, when the first disease is not present, the selection unit 12 selects a second disease from among a plurality of disease candidates as the second disease.

In this embodiment, head CT image data captured by an X-ray CT scanner is used as medical data to be inferred, and an example in which cerebral infarction is not detected is described as the first inference result in the inference unit 11. Below, explanations of configurations and flows similar to those of the above-described embodiments will be omitted as appropriate, and differences will be described.

(Selection Unit 12)
The selection unit 12 selects a second inference model for performing a second inference based on the first inference result in the inference unit 11. The selection unit 12 holds in advance a list that defines the clinical relationship between the first disease and the second disease, and selects the second inference model based on the first inference result. The list defines the clinical relationship for the presence or absence of the first disease. The selection unit 11 holds, as information on the list, information including the second disease to be selected when the first disease is not detected, for example.

(Operation of information processing device)
Next, the process executed by the information processing device 1 in this embodiment will be described with reference to the flowchart of FIG.

In step S62, the selection unit 12 acquires a first inference result from the inference unit 11. If the first inference result indicates that a first disease is detected, the selection unit 12 advances the process to step S32 or step S42 described above. If the first disease is not detected, the selection unit 12 advances the process to step S63.

Steps S63 to S65 correspond to steps S43 to S45 in FIG. 4, so their explanations are omitted here.

According to the information processing device of this embodiment, it is possible to efficiently select an inference model for performing a second inference based on the result of a first inference regarding a first disease. This makes it possible to shorten the time until the inference result for the second disease is displayed.

(Variation 1)
In the above embodiment, the case where X-ray CT image data is used as medical data to be inferred has been described. However, the application of the present invention is not limited to this, and raw data, MR images, SPECT images, and PET images from various medical imaging devices may be used as medical data to be inferred.

(Variation 2)
In the first embodiment, the inference model performs an inference task of inferring the presence or absence of a disease in three-dimensional image data. In the second embodiment, the inference model performs an inference task of segmenting a three-dimensional disease area in three-dimensional image data. In the third embodiment, the inference model performs an inference task of inferring the presence or absence of a disease in three-dimensional image data.

However, the present invention is not limited to these aspects, and the inference task may be to obtain a bounding box that encompasses a disease area in an image, classify a disease, infer the risk of developing a disease, or perform any other inference related to a disease.

Other Embodiments
The present invention can also be realized by executing the following process: That is, software (programs) that realize the functions of the above-described embodiments are supplied to a system or device via a network or various storage media, and the computer (or CPU, MPU, etc.) of the system or device reads and executes the program.

Claims (19)

a selection unit that acquires a first inference result by applying a first inference model to medical data, and selects a second inference model to be applied to a second inference on the medical data based on the first inference result;
an output unit that outputs information about the selected second inference model;
an information processing device comprising: a selection unit that refers to relationship information defining a relationship between a first inference target and a second inference target based on the first inference result; and when the second inference target corresponding to the first inference result is included in the relationship information, selects an inference model that includes the second inference target corresponding to the first inference result as a class as the second inference model. An acquisition unit that acquires the medical data;
The information processing apparatus according to claim 1 , further comprising an inference unit that performs inference on the medical data. the inference unit performs a second inference on the medical data using the selected second inference model;
The information processing apparatus according to claim 2 , wherein the output unit outputs an inference result by the inference unit. The information processing device according to any one of claims 1 to 3, characterized in that the selection unit selects, as the second inference model, an inference model that includes, as a class, a second disease that is different from the first disease detected by the first inference. The information processing device according to claim 4, characterized in that the first disease and the second disease are diseases that have a clinical relationship with each other. The information processing device according to claim 5, characterized in that the clinical relationship is a causal relationship between the first disease and the second disease, or a relationship between the first disease and the second disease that should be confirmed for the first disease. The information processing device according to claim 5 or 6, characterized in that the selection unit selects, as the second inference model, an inference model that includes, as a class, a plurality of second diseases that have a clinical relationship with the first disease. The information processing device according to claim 3, characterized in that the inference unit performs the first inference using the first inference model that includes a first disease as a class, and the second inference using the second inference model that includes a second disease different from the first disease as a class. The information processing device according to any one of claims 1 to 8, characterized in that the selection unit generates a plurality of the first inference results by performing the first inference using an inference model that includes each of the plurality of first diseases as a class, or by inferring the plurality of first diseases using a plurality of inference models, and selects the inference model that corresponds to the plurality of the first inference results as the second inference model. The information processing device according to claim 3 or 8, characterized in that the inference unit performs the second inference by further using the first inference result from the first inference as an input to the second inference model. The information processing device according to claim 8 or 10, characterized in that the inference unit performs a third inference by further inputting the second inference result from the second inference into the first inference model. The information processing device according to any one of claims 1 to 11, characterized in that the selection unit selects the second inference model based on a selection instruction input by a user. An information processing device as described in any one of claims 1 to 12, characterized in that when the first inference model used to perform the first inference is changed, the selection unit selects as a second inference model an inference model that includes a first disease corresponding to the changed first inference model and a second disease that is clinically related to the first disease as a class. The information processing device according to any one of claims 4 to 7, characterized in that the selection unit refers to relationship information that defines the relationship between the first disease and the second disease, and when there is a first disease that matches the first inference result, selects a second inference model that includes, as a class, a second disease that corresponds to the matching first disease. The information processing device according to claim 14, characterized in that the selection unit holds the relationship information in at least one of a list, a directed graph, and a tree diagram. 1. A computer-implemented information processing method, comprising:
a selection step of obtaining a first inference result by applying a first inference model to medical data, and selecting a second inference model to be applied to a second inference on the medical data based on the first inference result;
an output step of outputting information of the selected second inference model;
an information processing method comprising: a step of: referring to relationship information defining a relationship between a first inference target and a second inference target based on the first inference result; and selecting, when the second inference target corresponding to the first inference result is included in the relationship information, an inference model that includes the second inference target corresponding to the first inference result as a class as the second inference model. A program for executing the information processing method according to claim 16 on a computer. An acquisition unit for acquiring medical data;
a selection unit that acquires a first inference result by applying a first inference model to the medical data, and selects a second inference model to be applied to a second inference on the medical data based on the first inference result;
an inference unit that performs the second inference on the medical data using the selected second inference model;
an output unit that outputs an inference result based on the second inference;
an information processing system comprising: a selection unit that refers to relationship information defining a relationship between a first inference target and a second inference target based on the first inference result; and when the second inference target corresponding to the first inference result is included in the relationship information, selects an inference model that includes the second inference target corresponding to the first inference result as a class as the second inference model. a selection unit that acquires a first inference result by applying a first inference model to medical data, and selects a second inference model to be applied to a second inference on the medical data based on the first inference result;
an output unit that outputs information about the selected second inference model;
and when the first inference result indicates that a first disease has not been detected, the selection unit refers to relationship information that defines a relationship with a second disease corresponding to the case where the first disease has not been detected, and selects an inference model that includes the second disease as a class as the second inference model.

Images