Information processing device, information processing method and program

The information processing system accelerates lesion inference by selecting appropriate models and images based on imaging conditions, addressing the inefficiencies of existing methods.

JP2024160011AActive Publication Date: 2024-11-08CANON KK +1
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2024151370
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-11-08
Estimated Expiration
2040-05-21

AI Technical Summary

Technical Problem

Existing methods require time-consuming inference processes due to the need to select appropriate imaging conditions and inference models for each image, especially when multiple imaging conditions are involved, leading to increased inference time.

Method used

An information processing system that acquires a group of images under multiple conditions, selects an appropriate inference model based on these conditions, and performs inference using the selected model to quickly identify lesions.

Benefits of technology

Enables faster lesion inference by selecting optimal models and images, reducing the time required for processing multiple imaging condition sets.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024160011000001_ABST
    Figure 2024160011000001_ABST
Patent Text Reader

Abstract

To provide a technique that enables faster inference of a lesion from images obtained by imaging a subject under multiple imaging conditions.SOLUTION: An information processing device comprises: acquisition means which acquires an image group obtained by imaging a subject under multiple imaging conditions; selection means which selects at least one inference model from multiple inference models on the basis of at least one of the imaging conditions; and inference means which uses the selected inference model to perform inference of a lesion from the images in the acquired image group.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to an information processing device, an information processing method, and a program for a medical image. [Background technology]

[0002] In medical imaging examinations, various imaging devices (modalities) such as magnetic resonance imaging devices (MRI devices), computed tomography devices (X-ray CT devices), and ultrasound diagnostic devices are commonly used. Various imaging conditions are adopted for each modality; for example, MRI devices use T1 weighted imaging to emphasize tissue structures inside the body and T2 weighted imaging to emphasize lesions. Furthermore, images obtained with the same modality under different imaging conditions contain different information, and in clinical practice, these multiple images are used in combination to make a diagnosis.

[0003] In medical image diagnosis, various methods have been proposed for making inferences (e.g., detecting lesions) for images captured under multiple imaging conditions using an inference model created by machine learning. For example, Patent Document 1 discloses a technique for selecting an image captured under imaging conditions suitable for detecting lesions and an inference model from a group of images obtained by imaging the same subject under different imaging conditions. Patent Document 2 discloses a method for detecting a region of interest for a group of images captured under different imaging conditions using an inference model corresponding to each imaging condition, and for identifying the lesion using a map indicating the probability of the lesion in the region of interest. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2016-67573 A [Patent Document 2] U.S. Patent No. 10049451 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the above technology, if the lesion to be detected is not specified, it is not possible to select appropriate imaging conditions and inference models. Therefore, since multiple inference models that differ depending on the imaging conditions are used for each image, there is a problem that inference takes time. In addition, in the above technology, since the results obtained by using different inference models for multiple images are combined, there is a problem that the time required for inference increases as the number of imaging conditions increases.

[0006] Therefore, the technology disclosed herein provides a technology for more quickly inferring a lesion from images obtained by imaging a subject under a plurality of imaging conditions. [Means for solving the problem]

[0007] One aspect of the present disclosure is to provide an imaging apparatus comprising: an imaging unit for imaging a subject under a plurality of imaging conditions to obtain an image group; A selection means for selecting at least one inference model from a plurality of inference models based on at least one imaging condition of the plurality of imaging conditions; inference means for performing inference regarding a pathology on an image of the acquired image set using the selected inference model; The information processing device is characterized by having:

[0008] Also, one aspect of the present disclosure is a method for acquiring an image group obtained by imaging an object under a plurality of imaging conditions, comprising: A selection step of selecting at least one inference model from a plurality of inference models based on at least one imaging condition of the plurality of imaging conditions; an inference step of performing inference regarding a pathology on images of the acquired set of images using the selected inference model; The present invention relates to an information processing method.

[0009] Another aspect of the present disclosure is a program for causing a computer to function as each of the means of the information processing device described above. Effect of the Invention

[0010] According to the technology of the present disclosure, it is possible to more quickly make inferences regarding lesions from images obtained by imaging a subject under a plurality of imaging conditions. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 is a diagram showing an example of a schematic configuration of an information processing apparatus according to a first embodiment; [Diagram 2] 1 is a flowchart showing an example of processing executed in the first embodiment. [Diagram 3] FIG. 13 is a diagram showing an example of a schematic configuration of an information processing device according to a second embodiment; [Figure 4] 10 is a flowchart showing an example of processing executed in the second embodiment. [Diagram 5] FIG. 1 is a diagram showing an example of a schematic configuration of an information processing device according to a modified example; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The same or equivalent components, members, and processes shown in each drawing will be given the same reference numerals, and duplicated descriptions will be omitted as appropriate. Also, some of the components, members, and processes will be omitted in each drawing.

[0013] <First embodiment> An information processing system according to a first embodiment will be described below. In this embodiment, an appropriate inference model is selected for an image group obtained by imaging a subject under a plurality of imaging conditions, and inference is made regarding a lesion in the image. In the following description, as an example, a case is assumed in which a brain tumor is detected as a lesion from an image group including a T1 weighted image, a T2 weighted image, and a diffusion weighted image obtained by imaging the subject's head with an MRI device.

[0014] FIG. 1 is a block diagram showing an example of the configuration of an information processing system 1 in this embodiment. The information processing system 1 includes an imaging device 110, a data server 120, and an information processing device 130. The imaging device 110 is connected to the data server 120, and data captured by the imaging device 110 is stored in the data server 120. The data server 120 is connected to the information processing device 130. The information processing device 130 includes an acquisition unit 131, a selection unit 132, an inference unit 133, a generation unit 134, and an output unit 135. The functions of each unit of the information processing device 130 are realized by the CPU of the information processing device 130 executing a program stored in a memory. The information processing device 130 acquires data of a group of captured images from the data server 120, and executes the process described below using the acquired data.

[0015] In this embodiment, the imaging device 110 is an MRI device, and images the subject's head under each imaging condition of T1 weighted, T2 weighted, and diffusion weighted. The data server 120 holds a group of images of T1 weighted images, T2 weighted images, and diffusion weighted images received from the imaging device 110. The data server 120 also holds a plurality of learned inference models that have previously learned an image group including a T1 weighted image, a T2 weighted image, and a diffusion weighted image using a deep neural network or the like. Note that the image group only needs to include at least one type of image of a T1 weighted image, a T2 weighted image, and a diffusion weighted image, and may also include images captured under other imaging conditions. Also, the images used for learning are not limited to images captured by the imaging device 110, and may be appropriately selected. The data server 120 may be, for example, a system such as a PACS (Picture Archiving and Communication Systems) that stores and manages images via a network.

[0016] The acquisition unit 131 acquires an image group including a plurality of T1 weighted images, a plurality of T2 weighted images, and a plurality of diffusion weighted images from the data server 120 in response to an instruction from a user of the information processing device 130. The types of images acquired by the acquisition unit 131 are not limited to these three types, and one or more types of images may be acquired. The acquisition unit 131 corresponds to an acquisition means for acquiring an image group obtained by imaging a subject under a plurality of imaging conditions. In the following description, the plurality of T1 weighted images are referred to as T1 weighted images, or simply as T1 weighted images. The same applies to the plurality of T2 weighted images and the plurality of diffusion weighted images. The T1 weighted image is the name of an image captured by executing a pulse sequence using a first parameter, and the T2 weighted image is the name of an image captured by executing a pulse sequence using a second parameter. Typically, the parameters of the T1 weighted image are set to a longer TR than the parameters of the T2 weighted image. In this way, the imaging conditions are determined by the parameters in the sequence. In other words, the imaging conditions can be made different depending on the difference in the parameters in the sequence.

[0017] The selection unit 132 compares the image group (T1 weighted image, T2 weighted image, and diffusion weighted image) acquired by the acquisition unit 131 with the input conditions (types of images (imaging conditions) to be input to the model) of each inference model held by the data server 120. The input conditions of the inference model are associated with at least one imaging condition of the multiple imaging conditions. For example, the input conditions include a condition for inputting at least one type of image among T1 weighted image, T2 weighted image, and diffusion weighted image, such as a condition for inputting only T1 weighted image and a condition for inputting T1 and T2 weighted images. This comparison identifies an inference model that satisfies the input condition for at least one type of image among T1 weighted image, T2 weighted image, and diffusion weighted image. Then, the selection unit 132 selects an inference model with the highest predefined priority among the inference models that satisfy the input conditions of the acquired image group. Here, the criterion for the priority of the selected inference model is, for example, the high inference accuracy of detecting a lesion from an image, and the higher the inference accuracy, the higher the priority. The selection unit 132 may select not only the inference model with the highest priority, but also a predetermined number of inference models with the highest priority. In this way, the selection unit 132 selects multiple models corresponding to multiple imaging conditions. Furthermore, the criterion for the priority is not limited to the high inference accuracy, and other criteria such as the number of images for each imaging condition included in the image group may be appropriately adopted. The selection unit 132 corresponds to a selection means for selecting at least one inference model from multiple inference models based on at least one imaging condition of the multiple imaging conditions. Furthermore, the criterion for the priority corresponds to a predetermined criterion for selecting the inference model with the highest priority.

[0018] Then, the selection unit 132 selects, as a new image group, images that match the input conditions of the selected inference model from the image group acquired by the acquisition unit 131. The selection unit 132 outputs the selected inference model and the image group to the inference unit 133.

[0019] For example, it is assumed that the data server 120 holds inference models A, B, and C that perform some kind of inference on an input image. Here, the input conditions for the inference models A, B, and C are "T1-weighted images and T2-weighted images are input," "T1-weighted images, T2-weighted images, and FLAIR images are input," and "T1-weighted images are input." The priorities of the inference models A, B, and C are in the order of decreasing order: inference model B, inference model A, and inference model C. It is also assumed that the acquisition unit 131 acquires a group of images including T1-weighted images, T2-weighted images, and diffusion-weighted images.

[0020] In this case, the selection unit 132 selects an inference model in which the acquired image group satisfies the input condition. Model A and inference model C are selected. Since the image group does not include a FLAIR image, the selection unit 132 excludes inference model B from the selection targets. Then, the selection unit 132 selects inference model A, which has the highest priority. Furthermore, the selection unit 132 selects, as a new image group, a T1-weighted image and a T2-weighted image, which are input targets that match the input conditions of inference model A, from the image group acquired by the acquisition unit 131. Here, since the diffusion-weighted image included in the image group acquired by the acquisition unit 131 is not an input target for inference model A, the selection unit 132 excludes the diffusion-weighted image from the image group.

[0021] The inference unit 133 applies the selected inference model to a new image group consisting of images that match the input conditions of the inference model selected by the selection unit 132, and detects brain tumors, which are lesions in each image. When multiple inference models are selected by the selection unit 132, the inference unit 133 uses the multiple inference models to make inferences regarding lesions for each image in the image group. The inference unit 133 corresponds to an inference means that makes inferences for images in the image group.

[0022] The generating unit 134 generates a superimposed image by superimposing an index indicating the brain tumor area detected by the inference unit 133 on an image of the image group selected by the selecting unit 132. The generating unit 134 corresponds to a generating means for generating an image indicating the position of a lesion detected by inference in an image of the acquired image group.

[0023] The output unit 135 outputs the superimposed image on which the brain tumor indicators generated by the generation unit 134 are superimposed to the data server 120, a display unit of an information processing device (not shown), an external display device, or the like.

[0024] Next, a process executed by the information processing device 130 in this embodiment will be described with reference to the flowchart in Fig. 2. The CPU of the information processing device 130 starts the process of the flowchart in Fig. 2 upon receiving an instruction from the user of the information processing device 130, for example.

[0025] In step S21, the CPU of the information processing device 130 accepts designation of image capturing conditions from the user. Then, the acquisition unit 131 communicates with the data server 120 and acquires, from the data server 120, a group of images captured under imaging conditions that match the imaging conditions designated by the user. The acquisition unit 131 outputs the acquired group of images to the selection unit 132.

[0026] In step S22, first, the selection unit 132 selects an inference model that satisfies the input conditions of the image group input from the acquisition unit 131. Next, the selection unit 132 selects the inference model with the highest priority according to the priority order of the inference models stored in the memory of the information processing device 130. Then, the selection unit 132 selects images that match the input conditions defined in the selected inference model as a new image group from the image group input from the acquisition unit 131, and outputs the selected image group to the inference unit 133.

[0027] Here, when selecting an inference model that satisfies the input conditions of the image group input from the acquisition unit 131, the selection unit 132 may specify the inference model to be selected by referring to header information, etc. included in the images of the input image group. Alternatively, the selection unit 132 may use machine learning, etc. to determine under what imaging conditions the input image group was captured, and specify the inference model to be selected based on the determination result. In these cases, in step S21, the acquisition unit 131 acquires the image group from the data server 120 without performing a process of accepting designation of the imaging conditions from the user.

[0028] In step S23, the inference unit 133 detects brain tumors in each image by inputting the images selected in step S22 to the inference model selected in step S22. In this embodiment, the inference unit 133 detects brain tumors in each image by inputting the images to the inference model selected in step S22. By using a computational technique, brain tumors are detected in the images.

[0029] In step S24, the generation unit 134 generates a superimposed image by superimposing the index representing the brain tumor region detected by the inference unit 133 on each image of the image group selected in step S22. Then, the output unit 135 outputs the superimposed image generated by the generation unit 134 to the data server 120.

[0030] As described above, the information processing device according to the present embodiment can select an appropriate image group as an input for an inference model from an image group captured by an MRI device or the like, and infer a lesion in each image of the selected image group. This allows rapid inference regarding a lesion to be performed for an image group obtained by capturing images of the same subject under multiple imaging conditions.

[0031] <Second embodiment> Next, an information processing system according to a second embodiment will be described. In the following description, the same components as those in the first embodiment are denoted by the same reference numerals, and detailed description thereof will be omitted.

[0032] In images captured by an imaging device, phenomena such as artifacts that reduce the quality of inference may occur. In this embodiment, the information processing device selects an image group with a quality appropriate for inferring a lesion from an image group obtained by imaging the same subject under multiple imaging conditions. Then, the information processing device selects an inference model appropriate for the selected image group and performs inference regarding the lesion in the image. In the following description, artifacts are assumed as a phenomenon that reduces the quality of inference, but the following processing may be applied to other phenomena occurring in the image that affect the quality of inference, such as image distortion and contrast.

[0033] In this embodiment, an example of selecting an image group of appropriate quality based on an estimated value of an artifact that occurred when capturing images will be described. In the following description, a case where a motion artifact occurs when capturing images is assumed, but the type of artifact that occurs in the images is not limited to this. For example, this embodiment can be applied even when artifacts such as ring artifacts, streak artifacts, shower artifacts, shading artifacts, and beam hardening occur in the images.

[0034] 3 is a block diagram showing an example of the configuration of an information processing system 3 in this embodiment. Here, the configuration different from that in the first embodiment will be mainly described. The information processing device 330 has an acquisition unit 131, an inference unit 133, a generation unit 134, an output unit 135, a selection unit 332, and a calculation unit 336.

[0035] The selection unit 332 compares the image group acquired by the acquisition unit 131 with the input conditions of each inference model held by the data server 120, and selects an inference model for the acquired image group in accordance with the priority order. Then, the selection unit 332 selects images that match the input conditions of the selected inference model as a new image group from the image group acquired by the acquisition unit 131. The selection unit 332 outputs the selected image group to the calculation unit 336.

[0036] The calculation unit 336 divides the images into a plurality of regions for the image group selected by the selection unit 332, and calculates the occurrence probability of motion artifacts for each divided region for each imaging condition using a known method based on image processing. The calculation unit 336 corresponds to a calculation means for calculating the occurrence probability of a phenomenon that reduces the inference for each region in the image.

[0037] The probability of occurrence of a motion artifact may be calculated by applying an inference model that has been trained in advance using machine learning or the like to the image. The processor 332 outputs information indicating the probability of occurrence of the motion artifact to the selector 332.

[0038] The selection unit 332 compares the occurrence probability input from the calculation unit 336 with a predetermined threshold value of the occurrence probability of a predefined motion artifact. Then, the selection unit 332 specifies an image having an area where the occurrence probability of an artifact exceeds the threshold value. Furthermore, the selection unit 332 excludes images captured under the same imaging conditions as the images specified here from the image group output to the calculation unit 336, and performs the above process on the image group consisting of the remaining images to select an inference model again. Then, the selection unit 332 outputs the image group consisting of the images after exclusion and the reselected inference model to the inference unit 133.

[0039] Next, a process executed by the information processing device 330 in this embodiment will be described with reference to the flowchart in Fig. 4. The CPU of the information processing device 330 starts the process of the flowchart in Fig. 4 upon receiving an instruction from the user of the information processing device 330, for example.

[0040] Step S21 is the same as the process in the first embodiment. The acquisition unit 131 outputs the image group acquired from the data server 120 to the selection unit 332.

[0041] Next, in step S31, the selection unit 332 selects an inference model that satisfies the input conditions of the image group input from the acquisition unit 131. Next, the selection unit 332 selects the inference model with the highest priority according to the priority order of the inference models stored in the memory of the information processing device 130. Then, the selection unit 332 selects images that match the input conditions defined in the selected inference model as a new image group from the image group input from the acquisition unit 131. In step S31, the selection unit 332 outputs the selected image group to the calculation unit 336.

[0042] In step S32, the calculation unit 336 divides each image in the image group selected by the selection unit 332 in step S31 into a plurality of regions using a known method. Then, the calculation unit 336 calculates the occurrence probability of motion artifacts for each divided region for each image using a known method.

[0043] In step S33, the selection unit 332 determines whether the probability of occurrence of a motion artifact calculated in step S32 exceeds a predefined threshold. If there is an area where the probability of occurrence of a motion artifact exceeds the threshold (S33: Y), the process proceeds to step S34. On the other hand, if there is no area where the probability of occurrence of a motion artifact exceeds the threshold (S33: N), the process proceeds to step S23.

[0044] In step S34, the selection unit 332 excludes, from the group of images selected in step S31, images captured under the same imaging conditions as the image having an area in which the occurrence probability of motion artifacts exceeds a threshold value. Then, the process proceeds to step S35.

[0045] In step S35, the selection unit 332 performs processing similar to that in step S31 on the image group consisting of the remaining images in step S34, reselects the image group and inference model, and outputs the selected image group and inference model to the inference unit 133.

[0046] Next, the processing of steps S23 and S24 is the same as that in the first embodiment. If the processing proceeds from step S33 to step S34 (if S33:Y), then in step S23, the inference unit 133 inputs the image group selected again in step S35 into the inference model selected again in step S35, and detects a brain tumor in each image of the image group. If the processing proceeds from step S33 to step S35 without passing through step S34 (if S33:N), then in step S23, the inference unit detects a brain tumor in each image of the image group using the inference model selected in step S31 and the image group. Then, in step S In step S24, the generation unit 134 generates a superimposed image by superimposing the index representing the brain tumor region detected by the inference unit 133 on each image of the image group selected again in step S35. Then, the output unit 135 outputs the superimposed image generated by the generation unit 134 to the data server 120.

[0047] As described above, according to the information processing device of this embodiment, even if the captured image group includes images of poor quality due to artifacts, such images can be excluded and inference can be performed, thereby improving the accuracy of inference regarding lesions in images obtained by capturing images of the same subject under multiple imaging conditions.

[0048] (Other embodiments) The above-described embodiment merely shows a specific example of the present invention. The scope of the present invention is not limited to the configuration of the above-described embodiment, and various embodiments can be adopted without departing from the gist of the present invention.

[0049] Modifications of the above embodiment will be described below. The following modifications may be implemented in combination with each other, or may be implemented in appropriate combination with the above embodiment. In the following description, the same reference numerals are used for configurations similar to those of the above embodiment, and detailed description thereof will be omitted.

[0050] <Variation 1> In the above embodiment, a case where a brain tumor is detected from a T1 weighted image and a T2 weighted image obtained by imaging the head of a subject with an MRI device has been described as an example. However, as a modified example, the imaging conditions when imaging the subject with an MRI device are not limited to T1 weighted, T2 weighted, and diffusion weighted. The imaging conditions may be proton density weighted, FLAIR, susceptibility weighted, MRA (Magnetic Resonance Angiography), T2 star weighted, etc., and these imaging conditions may be combined arbitrarily.

[0051] Moreover, the imaging device 110 for imaging the subject is not limited to an MRI device, and may be an X-ray CT device, an ultrasound diagnostic device, or the like. Examples of imaging conditions for the X-ray CT device include lung field conditions and vertical section conditions. Examples of imaging conditions for the ultrasound diagnostic device include B-mode, Doppler, and elastography. Furthermore, in the above information processing system, a plurality of imaging devices may be used in combination. In this case, it is desirable to use a group of images obtained by imaging the same part of the same subject using a plurality of imaging devices. Also, a single imaging device for imaging the subject by combining a plurality of imaging devices, such as PET-CT, may be used. Therefore, in the above information processing system, the acquisition unit 131 can acquire a group of images obtained by imaging with different modalities, or an image group obtained by imaging with the same modality under a plurality of imaging conditions. Furthermore, when imaging the subject with different modalities, an image group consisting of various image types is obtained from imaging with these modalities. Here, the image type refers to the type of modality that images the subject and generates an image.

[0052] FIG. 5 shows a schematic configuration of an information processing system 5 as an example of this modified example. The same components as those in the above embodiment are denoted by the same reference numerals, and detailed description thereof will be omitted. In the information processing system 5, the data server 120 stores images generated by a first imaging device 510, a second imaging device 520, and a third imaging device 530, each of which is a different modality. For example, the first imaging device 510 can be an MRI device, the second imaging device 520 can be a CT device, and the third imaging device 530 can be an ultrasound diagnostic device. The information processing device 330 performs the above-mentioned processing on an image group consisting of a plurality of image types acquired by the data server 120, thereby making inferences regarding lesions on the images of the image group.

[0053] In this modification, the acquisition unit 131 acquires an image group consisting of a plurality of image types obtained by imaging a subject with different modalities. The selection unit 132 uses the image type as an input condition for the inference model instead of the imaging condition of the above embodiment, and selects at least one inference model from a plurality of inference models based on at least one of the plurality of image types. The selection unit 132 may select an inference model corresponding to the image type and the imaging condition. The inference unit 133 uses the selected inference model to make an inference regarding a lesion for images of the acquired image group.

[0054] The type of modality used to image the subject varies depending on the environment in which the information processing system is used. According to this modification, even if only images of a specific number of image types, such as MR images and CT images, are available, the information processing device 330 can obtain optimal inference results regarding lesions according to the image types.

[0055] In addition, the information processing device 330 may perform inference regarding a lesion for an image group including captured images obtained using a plurality of imaging conditions and captured images obtained by a plurality of image types. For example, a case is assumed in which the image group includes T1 weighted images and T2 weighted images generated by an MRI device, and CT images generated by a CT device, which is another modality. In this way, the information processing device 330 can perform inference regarding a lesion in the same manner as described above for an image group that is a combination of a plurality of images obtained by different sequences of a single modality and images obtained by another modality.

[0056] In addition, the imaging site of the subject is not limited to the head, but may be the chest or abdomen, etc., and the lesion to be inferred is not limited to brain tumor, but may be cerebral infarction, pulmonary nodule, etc. depending on the imaging site. Inference regarding the lesion is not limited to detection of the lesion, but may be benign / malignant judgment of the lesion, prognosis prediction, judgment of whether the lesion is present in the image, etc. Then, an inference model is created according to the inference target or the lesion to be detected. In addition, in the above description, the image type and the imaging condition can be regarded as the same input condition as the input condition of the inference model, that is, the imaging condition may include the type of modality used for imaging.

[0057] <Variation 2> In the above embodiment, a case has been described in which images acquired under the same imaging conditions as an image having an area where the probability of artifact occurrence exceeds a threshold are excluded from a group of images acquired from the data server 120. However, as a modified example, the selection unit 332 may exclude only images having an area where the probability of artifact occurrence exceeds a threshold, i.e., only images including artifacts, from a group of images acquired.

[0058] In this case, in step S34, the selection unit 332 removes only images having areas where the probability of artifact occurrence exceeds a threshold from the image group selected in step S31. In step S35, the selection unit 332 reselects an inference model and an image group for the image group consisting of the remaining sets of images in the same manner as described above. Then, the inference unit 133 inputs the image group reselected in step S35 to the inference model reselected in step S35, and detects brain tumors in each image of the image group.

[0059] This allows images without artifacts to be input into the inference model without being excluded, which is expected to result in more accurate inference results.

[0060] In this modified example, when an image having an area where the probability of occurrence of artifacts exceeds a threshold is excluded, the remaining images captured under the same imaging conditions as the excluded images are few, and therefore the inference accuracy using the images captured under these imaging conditions may also be low. If the ratio of images in which artifacts occur among images captured under the same imaging conditions exceeds a predetermined ratio, the selection unit 332 may exclude all images captured under the same imaging conditions as in the above embodiment. Here, the ratio of images in which artifacts occur may be the ratio of the number of images in which artifacts occur among images captured under the same imaging conditions. In addition, the number of images in which artifacts occur among images captured under the same imaging conditions exceeding a predetermined number may be regarded as the ratio of images in which artifacts occur among images captured under the same imaging conditions exceeding a predetermined ratio.

[0061] <Modification 3> In the above embodiment, a process of excluding an image having an area where the probability of artifact occurrence exceeds a threshold value regardless of the position in the image from the input target of the inference model has been described. However, as a modified example, in step S34, the selection unit 332 may determine whether or not the area where the probability of artifact occurrence exceeds a threshold value is included in a predefined region of interest. Alternatively, the selection unit 332 may determine whether or not the area where the probability of artifact occurrence exceeds a threshold value is included only in the region of interest. Here, the region of interest may be appropriately defined, for example, by being specified by a user or automatically extracted from an image acquired from the data server 120. Then, when the region of interest includes an area where the probability of artifact occurrence exceeds a threshold value, the selection unit 332 excludes an image having an area where the probability of artifact occurrence exceeds a threshold value from the input target of the inference model. As a result, even if an artifact occurs in a certain image, if the artifact does not occur in the region of interest, this image can be input to the inference model without being excluded, so that a more accurate inference result can be obtained.

[0062] <Modification 4> In the above embodiment, a case has been described in which an image having an area where the probability of occurrence of an artifact exceeds a threshold is excluded from the input target of the inference model. However, as a modified example, such an image may not be excluded from the input target of the inference model, and an inference model may be selected or inference may be performed using an image group including the image. Then, the generation unit 134 may generate information indicating that a low-quality image is used for inference, together with the inference result by the inference unit 133. Alternatively, the generation unit 134 may perform correction regarding artifacts on an image having an area where the probability of occurrence of an artifact exceeds a threshold by machine learning or the like as a correction means, and the inference unit 133 may perform inference on a group of images including the corrected images.

[0063] The present invention can also be realized by a process in which a program for implementing one or more of the functions of the above-described embodiments is supplied to a system or device via a network or a storage medium, and one or more processors in a computer of the system or device read and execute the program. The present invention can also be realized by a circuit (e.g., ASIC) that implements one or more of the functions. [Explanation of symbols]

[0064] 130, 330 Information processing device 131 Acquisition Department 132, 332 Selection section 133 Reasoning part

Claims

1. An acquisition means for acquiring a group of images obtained by imaging the subject under a plurality of imaging conditions; A selection means for selecting at least one inference model from a plurality of inference models based on at least one imaging condition of the plurality of imaging conditions; inference means for performing inference regarding a pathology on an image of the acquired image set using the selected inference model; 13. An information processing device comprising:

2. The input conditions of the plurality of inference models are respectively associated with at least one imaging condition of the plurality of imaging conditions; The selection means selects an inference model in which the acquired image group satisfies an input condition.

2. The information processing apparatus according to claim 1,

3. 3. An information processing apparatus according to claim 1, wherein the selection means selects the inference model with the highest priority according to a predetermined criterion.

4. 4. The information processing apparatus according to claim 3, wherein the predetermined criterion is the inference accuracy of an inference model.

5. 5. The information processing apparatus according to claim 3, wherein the inference means performs inference using an image captured under imaging conditions that match the input conditions of the inference model with the highest priority and the inference model with the highest priority.

6. The selection means selects a plurality of inference models corresponding to the plurality of imaging conditions, the inference means uses the selected inference models to make an inference regarding a lesion for each image in the set of images; 6. The information processing apparatus according to claim 1,

7. The information processing apparatus according to claim 1 , wherein the plurality of imaging conditions include at least T1 weighted, T2 weighted, and diffusion weighted.

8. The information processing apparatus according to claim 1 , wherein the acquisition means acquires a group of images obtained by imaging using different modalities.

9. The information processing apparatus according to claim 1 , wherein the acquisition means acquires a group of images obtained by imaging under a plurality of imaging conditions in a same modality.

10. An acquisition means for acquiring a group of images of a plurality of image types obtained by imaging a subject using different modalities; a selection means for selecting at least one inference model from a plurality of inference models based on at least one image type of the plurality of image types; inference means for performing inference regarding a pathology on an image of the acquired image set using the selected inference model; 13. An information processing device comprising:

11. The information processing apparatus according to claim 10 , wherein the selection means selects an inference model corresponding to an image type and an imaging condition.

12. For each image in the acquired set of images, a phenomenon that reduces inference in each region within the image is The method further includes a calculation means for calculating an occurrence probability, An information processing device according to any one of claims 1 to 11, characterized in that the selection means selects the inference model by excluding from the group of acquired images images any image that has an area in which the calculated probability of occurrence of the phenomenon exceeds a predetermined threshold.

13. The information processing device described in claim 12, characterized in that the selection means selects the inference model by excluding from the group of acquired images images that are acquired under the same imaging conditions as an image that has an area in which the calculated probability of occurrence of the phenomenon exceeds a predetermined threshold.

14. The method further includes a calculation means for calculating, for each image in the acquired image group, a probability of occurrence of a phenomenon that reduces inference in each region within the image, An information processing device described in any one of claims 1 to 11, characterized in that the selection means selects the inference model by excluding from the group of acquired images images whose region of interest includes an area where the calculated probability of occurrence of the phenomenon exceeds a predetermined threshold.

15. 15. The information processing apparatus according to claim 1, further comprising a generating unit that generates an image indicating a position of a lesion detected by inference by said inference unit in an image of the acquired image group.

16. A calculation means for calculating, for each image in the acquired image group, a probability of occurrence of a phenomenon that reduces inference in each region within the image; a generating means for generating information indicating that a low-quality image has been used for the inference when the inference is performed by the inference means using an image having an area in which the calculated occurrence probability of the phenomenon exceeds a predetermined threshold; 12. The information processing apparatus according to claim 1, further comprising:

17. A calculation means for calculating, for each image in the acquired image group, a probability of occurrence of a phenomenon that reduces inference in each region within the image; a correction means for performing a correction for the phenomenon on an image having an area in which the calculated occurrence probability of the phenomenon exceeds a predetermined threshold; and The inference means performs the inference on a group of images including the corrected image.

12. The information processing apparatus according to claim 1,

18. An acquisition step of acquiring a group of images obtained by imaging the subject under a plurality of imaging conditions; a selection step of selecting at least one inference model from a plurality of inference models based on at least one imaging condition of the plurality of imaging conditions; an inference step of performing inference regarding a pathology on images of the acquired set of images using the selected inference model; An information processing method comprising:

19. A program for causing a computer to function as each of the means of the information processing device according to any one of claims 1 to 17.

Citation Information

Patent Citations

  • Computer-aided detection system and method

    JP2008541889A

  • Detector generation apparatus, method and program, and image detection device

    JP2016067573A

  • Biomedical signal processing device

    JP2019111322A

  • Image Correction Using A Deep Generative Machine-Learning Model

    US20170372193A1

  • Automated lesion segmentation from MRI images

    US10049451B2