Information processing system and program

Abstract

To acquire an appropriate division result of a partial area of the pancreas by performing switching between a division algorithm of a partial area of the pancreas based on a blood vessel area and a division algorithm of a partial area of the pancreas based on the shape of the pancreas and a pixel value in the area.SOLUTION: An information processing system includes: a storage unit for storing a plurality of algorithms including a first algorithm and a second algorithm, the first algorithm being the algorithm for dividing a plurality of partial areas of the pancreas in medical image data on the basis of a blood vessel area in the medical image data with regard to the medical image data on the medical image including at least part of the pancreas, and the second algorithm being the algorithm for dividing a plurality of partial areas of the pancreas in the medical image data on the basis of at least one of the shape and the pixel value of a pancreas area in the medical image data with regard to the medical image data on the medical image including at least part of the pancreas; an image data acquisition unit for acquiring medical image data as an inference object; a selection information acquisition unit; and a dividing unit.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to an information processing system. [Background technology]

[0002] When diagnosing pancreatic diseases from images, information obtained by dividing (defining) the pancreas into partial regions such as the head, body, and tail of the pancreas may be referred to. Therefore, a technique for accurately dividing these partial regions in an image is important. Non-Patent Document 1 discloses a method for dividing a pancreatic region into each partial region by aligning an average shape model in which each partial region is divided with respect to the pancreatic region in an MRI image. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Alexandre Triay Bagur, et al. “Pancreas MRI Segmentation Into Head, Body, and Tail Enables Regional Quantitative Analysis of Heterogeneous Disease”, J. Magn. Reson. Imaging, 2022. [Non-Patent Document 2] Olaf Ronneberger, et al. “U-Net:Convolutional Networks for Biomedical Image Segmentation”,MICCAI,2015. [Non-Patent Document 3] C. Hattori et al., “Centerline detection and estimation of pancreatic duct from abdominal CT images”, Proc. SPIE 12032, Medical Imaging, 2022. Summary of the Invention [Problem to be solved by the invention]

[0004] However, the boundaries of each partial region of the pancreas are defined based on the relative positional relationship between the pancreas and the surrounding blood vessels (the aorta, portal vein, superior mesenteric vein, etc.). Therefore, if the method disclosed in Non-Patent Document 1, which is not based on the blood vessel positions, is used, the partial regions are not divided as defined, and an accurate diagnosis may not be possible. However, because the appearance of blood vessel regions on an image changes significantly depending on various conditions, including the shooting conditions, it can be difficult for a computer to automatically detect their positions, and therefore the method of dividing each partial region based on the blood vessel position is not stable. [Means for solving the problem]

[0005] The present disclosure provides the following information processing system. a storage unit that stores a plurality of algorithms including a first algorithm and a second algorithm, wherein the first algorithm is an algorithm for dividing a plurality of partial regions of a pancreas in medical image data relating to a medical image including at least a part of the pancreas based on a vascular region of the medical image data, and the second algorithm is an algorithm for dividing a plurality of partial regions of a pancreas in medical image data relating to a medical image including at least a part of the pancreas based on at least one of a shape and a pixel value of the pancreas region of the medical image data. An image data acquisition unit that acquires medical image data of an inference target; A selection information acquisition unit that acquires selection information including at least one piece of information from the following (i) to (iii); (i) Image conditions of the medical image data to be inferred (ii) information on a region related to blood vessels in the medical image data to be inferred (iii) information for determining the reliability of a vascular region obtained from the medical image data of the inference target; a selection unit that selects at least one algorithm from the first algorithm and the second algorithm from the storage unit using the selection information; a partitioning unit that partitions a plurality of partial regions of the pancreas in the medical image data of the inference target by applying the selected algorithm to the medical image data of the inference target; An information processing system having the above configuration. Effect of the Invention

[0006] The information processing system of the present disclosure can more reliably divide partial regions of the pancreas by switching, depending on conditions, between an algorithm that divides multiple partial regions of the pancreas based on vascular regions and an algorithm that divides multiple partial regions of the pancreas based on at least one of the shape and pixel values ​​of the pancreatic region. [Brief description of the drawings]

[0007] [Figure 1] FIG. 2 is a diagram showing an example of the functional configuration of the information processing system according to the first embodiment. [Diagram 2] FIG. 1 is a diagram showing an example of a hardware configuration of an information processing system according to a first embodiment. [Diagram 3] FIG. 2 is a diagram showing an example of a processing procedure of the information processing system according to the first embodiment. [Figure 4] FIG. 2 is a diagram showing an example of settings of the information processing system according to the first embodiment. [Diagram 5] FIG. 2 is a view for explaining an image according to the first embodiment. [Figure 6] FIG. 11 is a diagram showing an example of the functional configuration of an information processing system according to a second embodiment. [Figure 7] FIG. 11 is a diagram showing an example of a processing procedure of an information processing system according to a second embodiment. [Figure 8] FIG. 11 is a diagram showing an example of a processing procedure of an information processing system according to a third embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] The present disclosure provides an information processing system comprising: A) a memory unit that stores a plurality of algorithms, including a first algorithm and a second algorithm, The first algorithm is an algorithm for dividing medical image data relating to a medical image including at least a part of a pancreas into a plurality of partial regions of the pancreas based on a vascular region of the medical image data; The second algorithm is an algorithm for dividing medical image data relating to a medical image including at least a part of the pancreas into a plurality of partial regions of the pancreas based on at least one of a shape and a pixel value of the pancreas region of the medical image data. B) an image data acquisition unit for acquiring medical image data to be inferred; C) a selection information acquisition unit that acquires selection information including at least one piece of information among the following (i) to (iii): (i) Image conditions of the medical image data to be inferred (ii) information on a region related to blood vessels in the medical image data to be inferred (iii) information for determining the reliability of a vascular region obtained from the medical image data of the inference target; D) a selection unit that selects at least one algorithm from the first algorithm and the second algorithm from the storage unit using the selection information; and E) A partitioning unit that partitions a plurality of partial regions of the pancreas of the medical image data to be inferred by applying the selected algorithm to the medical image data to be inferred.

[0009] The plurality of partial regions may include at least one partial region selected from the group consisting of the pancreatic head region, the pancreatic body region, and the pancreatic tail region. Only some of the partial regions may be divided, such as the pancreatic head region and the other pancreatic regions.

[0010] The plurality of algorithms may include algorithms other than the first algorithm and the second algorithm, and two or more algorithms may be used in combination to divide the subregion.

[0011] Dividing a partial region of the pancreas means dividing the pancreas region of the image data into a plurality of parts, for example, dividing the pancreas region into a pancreas head region, a pancreas body region, and a pancreas tail region. Alternatively, dividing a partial region of the pancreas means determining which partial regions are in the pancreas region of the image data, for example, determining which of the pancreas region of the image data is the pancreas head region, the pancreas body region, or the pancreas tail region. Only one partial region may be determined, or multiple partial regions may be determined.

[0012] The vascular territory can include the territory of the portal vein (superior mesenteric vein) and the territory of the aorta. The medical image may be captured by any imaging device selected from the group consisting of an X-ray computed tomography (X-ray CT) device, a nuclear magnetic resonance imaging (MRI) device, a positron emission tomography (PET) device, and an ultrasound diagnostic device.

[0013] The information processing system of the present disclosure is capable of, in the C) selection information acquisition unit, (i) acquiring image conditions of the medical image data to be inferred as selection information of an algorithm, the image conditions including at least one of a group consisting of a contrast phase, a contrast time, a reconstruction function, an X-ray dose, and an amount of noise, and in the D) selection unit, selecting the first algorithm when the image conditions are not determined to reduce the detection accuracy of the blood vessels, and selecting the second algorithm when the image conditions are determined to reduce the detection accuracy of the blood vessels.

[0014] The information processing system of the present disclosure further includes F) a vascular region acquisition unit that acquires a vascular region of the medical image data to be inferred, and the C) selection information acquisition unit can acquire the selection information based on the vascular region acquired by the F) vascular region acquisition unit.

[0015] However, the C) selection information acquisition unit may acquire selection information not based on the vascular region acquired by the F) vascular region acquisition unit. (i) Image conditions of the medical image data of the inference target, (ii) information on a region related to blood vessels of the medical image data of the inference target, and (iii) information for determining the reliability of the vascular region acquired from the medical image data of the inference target can all be acquired without using the vascular region acquisition unit. For example, image conditions of medical image data can be obtained from supplementary information, and information on blood vessel-related regions of medical image data can be obtained not from blood vessel regions but from abnormalities near blood vessels.

[0016] In the information processing system of the present disclosure, the selection information acquired in the C) selection information acquisition unit includes information on the region related to blood vessels of the medical image data of the inference target (ii), the selection information is obtained by referring to the vascular region of the medical image data of the inference target acquired in the F) vascular region acquisition unit, and includes the presence or absence of at least any abnormality of a group of abnormalities consisting of anatomical abnormalities near the vascular region, foreign bodies near the vascular region, and image abnormalities, and the D) selection unit can select the first algorithm when the selection information is information indicating the absence of at least any abnormality of the group of abnormalities, and select the second algorithm when the selection information indicates the presence of at least any abnormality of the group of abnormalities.

[0017] In the information processing system of the present disclosure, the selection information acquired by the C) selection information acquisition unit includes (iii) information for determining the reliability of the vascular region acquired from the medical image data of the inference target, and the selection information is obtained by referring to the vascular region of the medical image data of the inference target acquired by the F) vascular region acquisition unit, and includes information for determining the reliability of the vascular region including at least one of an estimated blood vessel diameter, volume, and confidence level, and the D) selection unit is capable of selecting the first algorithm when at least one of the information for determining the reliability indicates that the reliability of the vascular region is high, and selecting the second algorithm when at least one of the information for determining the reliability indicates that the reliability of the vascular region is low.

[0018] The first algorithm may be an algorithm for segmenting a plurality of partial regions of the pancreas in medical image data by a machine learning-based method using information including information on vascular regions as input. The first algorithm may be an algorithm for dividing a plurality of partial regions of the pancreas based on a pancreas region and a blood vessel region of medical image data relating to a medical image including at least a portion of the pancreas. Furthermore, the first algorithm may be an algorithm that determines boundaries between a plurality of partial regions based on the vascular region, and divides the plurality of partial regions of the pancreas based on the boundaries.

[0019] The second algorithm may be an algorithm that divides a plurality of partial regions of the pancreas in medical image data using a trained machine learning based method that uses as input medical image data relating to a medical image that includes at least a portion of the pancreas. The second algorithm may be an algorithm for dividing the pancreas into a plurality of partial regions based on a pancreas region of medical image data relating to a medical image including at least a portion of the pancreas. Moreover, the second algorithm may be an algorithm for dividing a plurality of partial regions of the pancreas based on either a statistical volume ratio of the plurality of partial regions of the pancreas or a statistical length ratio of the center lines of the pancreas. Furthermore, the second algorithm may be an algorithm for dividing the pancreas into a plurality of partial regions by fitting a shape model of the pancreas divided into a plurality of partial regions to the pancreas region.

[0020] The present disclosure also provides a program for causing a computer to function as each of the means of the information processing system, and a recording medium on which the program is stored in a computer-readable format.

[0021] Hereinafter, an embodiment of the information processing system disclosed in this specification will be described with reference to the drawings. Duplicate descriptions of identical or equivalent components, members, and processes shown in each drawing will be omitted as appropriate. In addition, in each drawing, some of the components, members, and processes will be omitted as appropriate. In the following, the present disclosure will be described using CT image data captured by an X-ray computed tomography (X-ray CT) device as an example of medical image data. Note that the embodiment of the present invention is not limited to the following embodiment, and can be applied to images captured by, for example, a magnetic resonance imaging (MRI) device, a positron emission tomography (PET) device, and an ultrasound diagnostic device.

[0022] <First embodiment> The first embodiment is an information processing system including A) to E) above, in particular C) a selection information acquisition unit that (i) acquires image conditions of the medical image data to be inferred as selection information for an algorithm.

[0023] In this embodiment, in order to divide the partial regions of the pancreas region in the CT image data, multiple algorithms are switched according to conditions and executed. First, information on the image conditions of the CT image data is obtained as information for selecting an algorithm, and based on that, a suitable algorithm is selected from multiple algorithms for dividing the multiple partial regions of the pancreas to divide the multiple partial regions of the pancreas. Here, a case where the image conditions are information on the contrast phase in particular will be described. In this embodiment, the multiple partial regions of the pancreas to be divided are the pancreatic head region, pancreatic body region, and pancreatic tail region. In this embodiment, the algorithms include a first algorithm for dividing the pancreas into a plurality of partial regions based on CT image data and a vascular region in the CT image data, and a second algorithm for dividing the pancreas into a plurality of partial regions based on a shape of the pancreas depicted in the CT image data or a pixel value of the pancreas region.

[0024] The first and second algorithms in this embodiment will be described in more detail below.

[0025] The first algorithm is an algorithm that uses an inference device that receives as input CT image data and image data (vascular region image data) representing a vascular region in the CT image data, to divide multiple partial regions of the pancreas depicted in the CT image data to be inferred. Since the first algorithm uses image data representing a vascular region as one of its inputs, it has a characteristic that the division of multiple partial regions of the pancreas is likely to be inaccurate when the reliability (detection accuracy, etc.) of the vascular region held by the vascular region image data is low.

[0026] On the other hand, the second algorithm is an algorithm that uses an inference device that receives as input CT image data to be inferred and image data representing the pancreatic region in the CT image data (pancreatic region image data) to segment multiple partial regions of the pancreas depicted in the CT image data to be inferred. According to the second algorithm, information on the vascular region is not used as an input, and even if the inference target is CT image data in which the reliability of the vascular region may be low, good segmentation results of multiple partial regions of the pancreas can be obtained.

[0027] Anatomically, the boundaries of the partial regions of the pancreas, such as the head, body, and tail of the pancreas, are defined by the surrounding blood vessels (the aorta, portal vein, superior mesenteric vein, etc.). Therefore, when the reliability of the blood vessel region is high, it is preferable to divide the multiple partial regions of the pancreas based on the blood vessel region corresponding to the CT image data, as in the first algorithm. However, since the appearance of the blood vessel region on the image changes due to various factors, it may be difficult to accurately obtain the blood vessel region depending on the image. Here, the various factors are, for example, the contrast state, and in particular, when the image data is used without using a contrast agent and has low contrast with the surrounding tissue, it becomes difficult to obtain the accurate region of the blood vessel, and the reliability of the blood vessel region may be low. Therefore, in this embodiment, the second algorithm is used when the reliability of the blood vessel region to be obtained may be low, and the first algorithm is used when the reliability of the blood vessel region to be obtained may be low, thereby obtaining a suitable division result of the multiple partial regions of the pancreas.

[0028] (Functional configuration) Hereinafter, the functional configuration of the information processing system 100 according to this embodiment will be described with reference to Fig. 1. As shown in the figure, the information processing system 100 includes a storage unit 110, an image data acquisition unit 120, a selection information acquisition unit 140, a selection unit 150, and a classification unit 160.

[0029] The storage unit 110 is an example of a computer-readable storage medium, and is a large-capacity storage device such as a hard disk drive (HDD) or a solid-state drive (SSD). The storage unit 110 holds CT image data of an inference target, supplementary information of the CT image data of an inference target, a program capable of executing a first algorithm, and a program capable of executing a second algorithm.

[0030] The image data acquisition unit 120 acquires the CT image data of the inference target and its associated information from the storage unit 110 , and transmits them to the selection information acquisition unit 140 and the division unit 160 .

[0031] The selection information acquisition unit 140 receives the supplementary information of the CT image data to be inferred from the image data acquisition unit 120, and acquires information on the contrast phase included in the supplementary information as information for selecting an algorithm. Then, the selection information acquisition unit 140 transmits the information on the contrast phase to the selection unit 150. In this embodiment, the information on the contrast phase is the time that has elapsed since the administration of the contrast agent (contrast time). Hereinafter, the information for selecting an algorithm may be referred to as selection information.

[0032] The selection unit 150 first receives information related to the contrast phase, which is selection information, from the selection information acquisition unit 140. Next, the selection unit 150 selects an algorithm to be used in the division unit 160 from among a plurality of algorithms, based on the information related to the contrast phase. Then, the selection unit 150 transmits the selection result of the algorithm to the division unit 160.

[0033] In this embodiment, the algorithms include a first algorithm for dividing the pancreas into a plurality of partial regions based on the CT image data and the vascular region corresponding to the CT image data, and a second algorithm for dividing the pancreas into a plurality of partial regions based on the shape of the pancreas depicted in the CT image data or the pixel values ​​of the pancreas region.

[0034] As described above, the first algorithm divides the pancreas into multiple partial regions based on the CT image data and the vascular region, and therefore has the characteristic that the division of the pancreas into multiple partial regions is likely to be inaccurate when the reliability (detection accuracy, etc.) of the vascular region is low. On the other hand, the second algorithm divides the pancreas into multiple partial regions based on the CT image data and the pancreas region, and therefore can obtain good division results for the multiple partial regions of the pancreas even when the inference target is CT image data in which the reliability of the vascular region may be low. Therefore, the selection unit 150 selects the second algorithm when the condition is such that the reliability of the vascular region may be low, and selects the first algorithm when that is not the case.

[0035] The division unit 160 first receives the CT image data of the inference target from the image data acquisition unit 120, and receives the selection result of the algorithm from the selection unit 150. Next, the division unit 160 reads an executable program corresponding to the selection result of the algorithm from the storage unit 110. That is, if the selection result of the algorithm is information representing a first algorithm, the division unit 160 reads a program capable of executing the first algorithm into the main memory, and if the selection result is information representing a second algorithm, the division unit 160 reads a program capable of executing the second algorithm into the main memory. Next, the division unit 160 executes the process of the selected algorithm to obtain the division result of the multiple partial regions of the pancreas corresponding to the CT image data of the inference target, and generates image data representing the multiple partial regions of the pancreas. Then, the division unit 160 stores the image data representing the multiple partial regions of the pancreas in the storage unit 110.

[0036] Here, the image data representing the plurality of partial regions of the pancreas is multi-value image data (label image data) in which the pancreatic head region, pancreatic body region, and pancreatic tail region are each expressed in a distinguishable manner. In this multi-value image data, for example, the pixel value of the pixel belonging to the pancreatic head region is 1, the pixel value of the pixel belonging to the pancreatic body region is 2, the pixel value of the pixel belonging to the pancreatic tail region is 3, and the pixel value of the pixel belonging to the other background regions is 0. Note that the image data representing the plurality of partial regions of the pancreas may be in any expression format as long as the plurality of partial regions of the pancreas can be distinguished, and may be, for example, image data in which each pixel holds the likelihood that it belongs to each partial region. In addition, the image data representing the plurality of partial regions of the pancreas may be composed of a plurality of image data, and may be, for example, image data representing the pancreatic head region, image data representing the pancreatic body region, and image data representing the pancreatic tail region. In this case, each image data may be, for example, configured to hold the pixel value of the pixel belonging to the partial region as 1 and the pixel value of the other pixels as 0, or to hold the likelihood that each pixel is a partial region as a value between 0 and 1.

[0037] Hereinafter, image data representing multiple partial regions of the pancreas may be referred to as partial region image data. In addition, the term "region image data" used hereinafter refers to image data in an expression format in which target regions can be distinguished, similar to partial region image data.

[0038] (Hardware configuration) Next, the hardware configuration of the information processing system 100 will be described with reference to Fig. 2. The information processing system 100 has the configuration of a known computer (information processing system). The information processing system 100 includes, as its hardware configuration, a CPU 201, a main memory 202, a magnetic disk 203, a display memory 204, a monitor 205, a mouse 206, and a keyboard 207.

[0039] A CPU (Central Processing Unit) 201 mainly controls the operation of each component. A main memory 202 stores a control program executed by the CPU 201 and provides a working area when the CPU 201 executes the program. A magnetic disk 203 stores programs for implementing various application software including an OS (Operating System), device drivers for peripheral devices, and programs for performing processes described below. The CPU 201 executes programs stored in the main memory 202, the magnetic disk 203, etc., thereby implementing the functions (software) of the information processing system 100 shown in FIG. 1 and processes in the flowcharts described below. The magnetic disk 203 may be the same as the storage unit 110.

[0040] The display memory 204 temporarily stores display data. The monitor 205 is, for example, a CRT monitor or a liquid crystal monitor, and displays images, text, and the like based on the data from the display memory 204. The mouse 206 and keyboard 207 allow the user to input pointing and characters, respectively. The above components are connected to each other via a common bus 208 so that they can communicate with each other.

[0041] The CPU 201 corresponds to an example of a processor or a control unit. In addition to the CPU 201, the information processing system 100 may have at least one of a GPU (Graphics Processing Unit) and an FPGA (Field-Programmable Gate Array). Also, instead of the CPU 201, the information processing system 100 may have at least one of a GPU and an FPGA. The main memory 202 and the magnetic disk 203 correspond to an example of a memory or a storage device.

[0042] (Processing Procedure) Next, a processing procedure of the information processing system 100 according to this embodiment will be described with reference to FIG.

[0043] (Step S310) In step S310, the image data acquisition unit 120 acquires the CT image data of the inference target and its associated information from the storage unit 110, and transmits them to the selection information acquisition unit 140 and the division unit 160. Here, the CT image data is a three-dimensional image obtained by imaging a region including the pancreas of the subject. In addition, the associated information may include information such as the patient ID, sex, imaging date and time, imaging device, reconstruction function, contrast phase (time elapsed since administration of contrast agent), etc.

[0044] (Step S320) In step S320, the selection information acquisition unit 140 receives the supplementary information of the CT image data to be inferred from the image data acquisition unit 120, and acquires information on the contrast phase included in the supplementary information as selection information. Then, the selection information acquisition unit 140 transmits the information on the contrast phase, which is the selection information, to the selection unit 150. In this embodiment, the information on the contrast phase is described as an example of the time elapsed since the administration of a contrast agent (contrast time). However, the information on the contrast phase is not limited to this, and may be information in which the contrast state is classified into a plurality of categories, such as an early phase, an arterial phase, a pancreatic parenchymal phase, a portal vein phase, and a delayed phase.

[0045] (Step S330) In step S330, the selection unit 150 first receives information on the contrast phase, which is selection information, from the selection information acquisition unit 140. Next, the selection unit 150 selects an algorithm to be used in the division unit 160 from among a plurality of algorithms, based on the information on the contrast phase. Then, the selection unit 150 transmits the selection result of the algorithm to the division unit 160. In this embodiment, the selection unit 150 selects an algorithm using a correspondence table between a preset contrast time and an algorithm to be selected, as shown in FIG. 4(a). The correspondence table in this embodiment is set so that the first algorithm is selected in the case of a contrast time (30 to 120 seconds) at which the contrast between blood vessels and surrounding tissues is high due to the contrast effect, and the second algorithm is selected in other cases. Here, the contrast time value at which the contrast is high, such as 30 seconds or 120 seconds, may be a preset value, or may be appropriately changed by the user. In addition, a mechanism may be provided for changing the value based on the auxiliary information acquired in step S310. For example, the above-mentioned value may be switched between the case where the reconstruction function included in the auxiliary information is a soft tissue condition and the case where the reconstruction function is a lung field condition. In this case, the reconstruction function of the lung field condition tends to have a high noise level due to high frequency emphasis. Therefore, in the case of the reconstruction function of the lung field condition, it is desirable to set the condition for selecting the first algorithm stricter than that of the soft tissue condition.

[0046] (Step S340) In step S340, the division unit 160 first receives the CT image data of the inference target from the image data acquisition unit 120, and receives the selection result of the algorithm from the selection unit 150. Next, the division unit 160 reads an executable program corresponding to the selection result of the algorithm from the storage unit 110. Next, the division unit 160 executes the process of the selected algorithm to obtain the division result of the multiple partial regions of the pancreas corresponding to the CT image data of the inference target, and generates image data representing the multiple partial regions of the pancreas. Then, the division unit 160 stores the image data representing the multiple partial regions of the pancreas in the storage unit 110.

[0047] The processing procedures of the first and second algorithms will be described in detail with reference to FIG. 5. FIG. 5(a) shows CT image data 510 to be inferred. Although the actual CT image data is three-dimensional volume data, FIG. 5 shows only one axial cross-sectional image constituting the CT image data for ease of illustration. The CT image data 510 depicts a blood vessel 511 defining the boundary between the head and body of the pancreas, a blood vessel 512 defining the boundary between the body and tail of the pancreas, and a pancreas 513. More specifically, the blood vessel 511 is the portal vein (or the superior mesenteric vein), and the blood vessel 512 is the aorta.

[0048] First, the first algorithm will be described. In the first algorithm, first, as shown in FIG. 5(b), blood vessel regions 521 and 522 are acquired from the CT image data 510 to be inferred, and blood vessel region image data 520 is generated. In this embodiment, the CT image data is used as input, and the blood vessel region 521 and the blood vessel region 522 are acquired using a machine learning model trained to segment the blood vessel region. This machine learning model is, for example, a U-Net, a type of neural network disclosed in Non-Patent Document 2, and is trained by a known method using a teacher data set consisting of CT image data and ground truth image data of the blood vessel region. In the ground truth image data of the blood vessel region, the regions corresponding to the blood vessel 511 and the blood vessel 512 are held in a distinguishable format, and the U-Net is trained to segment the regions corresponding to the blood vessel 511 and the blood vessel 512 from the CT image data to be inferred. Then, using the CT image data 510 to be inferred and the blood vessel region image data 520, as shown in FIG. 5(d), the pancreatic head region 544, the pancreatic body region 545, and the pancreatic tail region 546 are divided from the CT image data 510 to be inferred, and partial region image data 540 is generated. However, the pancreatic head region 544, the pancreatic body region 545, and the pancreatic tail region 546 are examples of the partial regions of the pancreas, and the partial regions of the pancreas are not limited to these. In this embodiment, the CT image data and the blood vessel region image data are input, and a machine learning model trained to segment each of the pancreatic head region, the pancreatic body region, and the pancreatic tail region is used to divide the multiple partial regions including the pancreatic head region 544, the pancreatic body region 545, and the pancreatic tail region 546. This machine learning model is, for example, U-Net, and is trained by a known method using a teacher dataset consisting of the CT image data, the blood vessel region image data, and the ground truth image data of each of the pancreatic head region, the pancreatic body region, and the pancreatic tail region. Here, image data obtained by combining CT image data and vascular region image data on the channel axis is used as the input to U-Net. Note that this input method is just one example, and for example, the CT image data and vascular region image data may be input from separate input layers.

[0049] According to the processing procedure of the first algorithm described above, the division unit 160 acquires a vascular region from the CT image data of the inference target, and then divides a plurality of partial regions of the pancreas from the CT image data of the inference target based on the CT image data of the inference target and the vascular region corresponding to the CT image data of the inference target, thereby generating partial region image data.

[0050] Next, the second algorithm will be described. The second algorithm divides the CT image data 510 of the inference target into a pancreatic head region 544, a pancreatic body region 545, and a pancreatic tail region 546 based on the pixel value of the pancreas 513 depicted in the CT image data 510 of the inference target, and generates partial region image data 540. In this embodiment, the pancreatic head region 544, the pancreatic body region 545, and the pancreatic tail region 546 are divided using a machine learning model that is trained to segment the pancreatic head region, the pancreatic body region, and the pancreatic tail region using the CT image data as input. This machine learning model is, for example, U-Net, and is trained by a known method using a teacher data set consisting of the CT image data and the ground truth image data of the pancreatic head region, the pancreatic body region, and the pancreatic tail region. The second algorithm is an algorithm different from the first algorithm, and is characterized in that it is not based on information on the blood vessel region corresponding to the CT image data of the inference target.

[0051] According to the processing procedure of the second algorithm described above, the division unit 160 divides the CT image data of the inference target into multiple partial regions of the pancreas based on the pixel values ​​of the pancreas depicted in the CT image data of the inference target, and generates partial region image data.

[0052] Through the above processing procedure, the information processing system 100 can obtain suitable division results for multiple partial regions of the pancreas by switching between the first algorithm and the second algorithm based on information regarding the contrast phase, which is one of the image conditions of the CT image data to be inferred.

[0053] The result of segmenting the plurality of partial regions of the pancreas obtained by the above processing procedure can be used, for example, for automatic generation of a report. As a more specific example of use, first, after the result of segmenting the plurality of partial regions of the pancreas is obtained by the above processing procedure, an external information processing device communicably connected to the information processing system 100 executes an estimation process of a region suspected of pancreatic cancer. Next, the region suspected of pancreatic cancer is acquired from the CT image data of the inference target, and the region suspected of pancreatic cancer is superimposed on the CT image data of the inference target and displayed on the monitor 205. Next, after the doctor confirms the region suspected of pancreatic cancer, the external information processing device calculates the center point of the region suspected of pancreatic cancer, identifies in which partial region the center point of the region suspected of pancreatic cancer exists by referring to the segmentation result of the partial region, and acquires the identification result. Then, the external information processing device automatically generates a report by adding information (identification result) of the partial region in which the region suspected of pancreatic cancer exists in addition to information that the region suspected of pancreatic cancer exists in the CT image data of the inference target. Note that the method of identifying the partial region in which the region suspected of pancreatic cancer exists is not limited to the above. For example, the partial region containing the most extracted suspected pancreatic cancer regions (e.g., the partial region with the largest TPR value relative to the number of pixels in the suspected pancreatic cancer region) may be determined as the partial region in which the suspected pancreatic cancer region exists. In addition, when a suspected pancreatic cancer region straddles two partial regions, information that the region exists between the two partial regions may be added and output to the report. In addition, when displaying the estimated results of the suspected pancreatic cancer region obtained by the same procedure as above on the monitor 205, information on the partial region in which the suspected pancreatic cancer region exists (identification result) may be added and displayed. Furthermore, abnormalities in the pancreas may be detected by performing image analysis for each divided partial region. As a specific example, the diameter of the pancreas may be calculated for each partial region, and a localized atrophic area of ​​the pancreas may be identified based on the difference in the diameter of the pancreas between the partial regions. Note that although several examples of use have been shown above, examples of use of the division results of multiple partial regions of the pancreas are not limited to these.

[0054] (Variations) In the above embodiment, the information processing system 100 divides the pancreas into a pancreatic head region, a pancreatic body region, and a pancreatic tail region as a plurality of partial regions of the pancreas, but is not limited to these partial regions or combinations. For example, the partial regions may be divided to include a pancreatic neck region, a pancreatic uncinate region, etc. Also, only some partial regions may be divided, such as the pancreatic head region and the other pancreatic regions.

[0055] In the above embodiment, the information processing system 100 divides the pancreas into a plurality of partial regions using an algorithm alternatively selected by the selection unit 150. However, the partial regions may be divided using two or more algorithms in combination. In this case, for example, the algorithm selection in step S330 in the above embodiment is performed by setting a condition for selecting two or more algorithms in advance, such as the correspondence table of contrast time and algorithm to be selected shown in FIG. 4(b). Then, the division unit 160 obtains the division results of the pancreas into a plurality of partial regions by executing processing using both algorithms, and obtains the final division result of the partial regions by integrating the division results of the two partial regions. The integration method is an arbitrary method using weights corresponding to the first and second algorithms set in advance. Here, a specific example of the integration method is shown. For example, when the division results of the pancreas into a plurality of partial regions obtained by each algorithm are probability maps representing the partial region-likeness, the probability maps of the same partial region obtained by each algorithm are integrated by weighted average value using weights set in advance. That is, the probability maps representing the pancreatic head region obtained by the first algorithm and the second algorithm are integrated by weighted averaging, and the pancreatic body region and the pancreatic tail region are integrated in the same manner. In the above embodiment, inference processing using U-Net is executed as the processing of the first algorithm and the second algorithm. In this case, the results of each processing (division result) are obtained as a probability map, so the results of each processing are integrated by the above method to obtain the final division result of multiple partial regions of the pancreas. As another integration method, a method may be used in which the boundary 547 between the pancreatic head region and the pancreatic body region is determined based on the division result of the second algorithm, and the boundary 548 between the pancreatic body region and the pancreatic tail region is determined based on the division result of the first algorithm. Also, the boundary positions of the division results of each algorithm may be averaged to divide them as the final boundary of the partial region.

[0056] In the above embodiment, the information on the contrast enhancement phase is obtained from the supplementary information of the CT image data to be inferred as the selection information of the algorithm, but any method may be used to obtain the information on the contrast enhancement phase. For example, the information on the contrast enhancement phase may be obtained from the CT image data to be inferred by a classifier that classifies the contrast enhancement phase (or contrast time) using the CT image data as an input. Also, the information on the contrast enhancement phase input by the user may be obtained.

[0057] In the above embodiment, information on the contrast phase is acquired as the selection information of the algorithm, but any information may be used as long as it is information on image conditions that affect the blood vessel depiction ability of the CT image data to be inferred. The image conditions of the CT image data to be inferred include, for example, a reconstruction function, an X-ray dose, a tube voltage, and an amount of noise.

[0058] If the selection information is a reconstruction function, in step S320, the selection information acquisition unit 140 acquires the reconstruction function from the auxiliary information. Then, in step S330, the selection unit 150 selects the first algorithm if the reconstruction function is a predetermined one, and selects the second algorithm if the reconstruction function is not a predetermined one. For example, in the case of a reconstruction function that is mainly used in lung image diagnosis, a kernel that emphasizes high frequency regions is used, so that the noise level tends to be higher than that of a reconstruction function used in soft tissue image diagnosis. If the division accuracy of multiple partial regions of the pancreas is greatly affected by the noise level of either the first algorithm or the second algorithm, a suitable division result of multiple partial regions of the pancreas can be obtained by switching the algorithm according to the reconstruction function as described above. In addition, when a machine learning model is used in the process of each of the first algorithm and the second algorithm, the algorithm may be selected in consideration of the reconstruction function of the CT image data included in the teacher data set used to train the machine learning model. For example, consider a case in which the teacher data set when the machine learning model that segments the vascular region used in the first algorithm is trained does not contain (or contains little) CT image data reconstructed with a reconstruction function mainly used in lung image diagnosis. In this case, it is assumed that the machine learning model for segmenting the vascular region used in the first algorithm has low extraction accuracy of the vascular region for CT image data reconstructed with a reconstruction function mainly used in image diagnosis of the lung field, and the division accuracy of the partial region is also low. Therefore, in such a case, for example, in step S330, the selection unit 150 can select the second algorithm if the CT image data to be inferred is image data reconstructed with a reconstruction function mainly used in image diagnosis of the lung field. Note that the method of selecting the algorithm based on the reconstruction function is not limited to the above.

[0059] When the selection information is the X-ray dose, for example, in step S320, the selection information acquisition unit 140 acquires information on the X-ray dose from the supplementary information. Generally, the lower the X-ray dose, the stronger the noise of the CT image data. This increases the difficulty of acquiring the blood vessel region, so in step S330, the selection unit 150 selects the first algorithm if the X-ray dose is equal to or greater than a predetermined value, and selects the second algorithm if the X-ray dose is less than the predetermined dose. As a result, the first algorithm is selected if the noise of the CT image data is expected to be low, and the second algorithm is selected otherwise. As a result, an appropriate algorithm is selected according to the noise level of the CT image data. When the selection information is the tube voltage, the noise is stronger when the tube voltage is low, so similarly to the above, the first algorithm and the second algorithm may be selected based on a predetermined value.

[0060] The selection information is not necessarily based on the additional information of the CT image data. For example, the noise amount of the CT image data obtained by analyzing the CT image data itself may be used as the selection information. In this case, for example, the noise amount contained in the CT image data can be estimated by analyzing the distribution of pixel values ​​of the CT image data to be inferred, and the estimated value can be used as the selection information to select an algorithm. That is, in step S320, the selection information acquisition unit 140 analyzes the distribution of pixel values ​​of the CT image data to be inferred to acquire information on the noise amount (for example, the variance of pixel values ​​of an area that should be homogeneous, such as an air area). Then, in step S330, the selection unit 150 selects the first algorithm if the noise amount is less than a threshold value, and selects the second algorithm if the noise amount is equal to or greater than the threshold value. As a result, if the noise of the CT image data is low, the first algorithm is selected, and if not, the second algorithm is selected. As a result, an appropriate algorithm is selected according to the noise level of the CT image data. In the above embodiment, the second algorithm divides the partial regions based on the pixel values ​​of the pancreas using a machine learning model that uses the CT image data of the inference target as an input, but the partial regions may be divided by explicitly providing information about the pancreas region on the CT image data of the inference target. For example, in step S340, the division unit 160 uses the CT image data as an input to acquire a pancreas region 533 from the CT image data 510 of the inference target using a machine learning model that has been trained to segment the pancreas region, as shown in FIG. 5(c), and generates pancreas region image data 530. Then, the division unit 160 uses the CT image data and the pancreas region image data as input to acquire multiple partial regions of the pancreas from the CT image data 510 of the inference target using a machine learning model that has been trained to segment multiple partial regions of the pancreas.

[0061] In the above embodiment and variations, the second algorithm divides the pancreas into a plurality of partial regions based on the pixel values ​​of the pancreas, but the partial regions may be divided based on the shape of the pancreas. The shape of the pancreas is, for example, the pancreas region or pancreatic duct region obtained from the CT image data to be inferred, or the center line of the pancreas, and by using these statistical shape information and shape models, it is possible to obtain results similar to those of the above embodiment. When dividing the partial regions based on statistical shape information of the pancreatic region, for example, in step S340, the division unit 160 acquires the pancreatic region from the CT image data of the inference target using a known method. Then, the division unit 160 calculates the boundaries of the partial regions based on the statistical volume and length ratio of multiple partial regions of the pancreas, and divides the acquired pancreatic region. More specifically, the division unit 160 first calculates the volumes of the pancreatic head region, pancreatic body region, and pancreatic tail region from the total volume of the acquired pancreatic region and the volume ratio of the partial regions set in advance based on statistics. Next, the division unit 160 calculates the cross-sectional area of ​​the pancreatic region for each sagittal cross section from the right shoulder side to the left shoulder side of the subject, obtains a cumulative sum, and calculates the sagittal cross section whose cumulative sum is equal to or greater than the volume of the pancreatic head region as the boundary 547 between the pancreatic head region and the pancreatic body region. Next, the cumulative sum is reset to 0, and the cross-sectional area of ​​the pancreatic region is calculated in the same manner using the sagittal cross section that is the boundary 547 between the pancreatic head region and the pancreatic body region as a reference to obtain the cumulative sum. Then, the sagittal cross section where the cumulative sum is equal to or greater than the volume of the pancreatic body region is calculated as the boundary 548 between the pancreatic body region and the pancreatic tail region. That is, the pancreatic region is divided into a plurality of partial regions by calculating the boundaries of each partial region based on a statistical volume ratio, and the divided partial regions are acquired in order from the right shoulder side to the left shoulder side as the pancreatic head region 544, the pancreatic body region 545, and the pancreatic tail region 546. As another method, the cross-sectional area of ​​the pancreatic region in a cross section perpendicular to the center line of the pancreas may be calculated along the center line of the pancreas, and the boundaries of each partial region may be calculated in the same manner as above. Note that the center line of the pancreas may be acquired by using a known method such as the technique disclosed in Non-Patent Document 3. Note that the volume ratio of the plurality of partial regions of the pancreas may be set in advance as a fixed value based on statistics as described above, or may be set by the user. Furthermore, information such as the height, weight, sex, and age of the subject may be obtained, and the set value of the volume ratio may be changed based on this information. When dividing the partial regions based on the shape model of the pancreas region, for example, in step S340, the division unit 160 fits a shape model of the pancreas in which the multiple partial regions of the pancreas are divided, to the acquired pancreas region by a known method. Then, the division unit 160 projects the partial regions of the pancreas shape model after fitting onto the acquired pancreas region, thereby acquiring a division result of the multiple partial regions of the pancreas for the CT image data to be inferred. Although the method of using the volume ratio and shape model of the pancreatic region has been described above, partial regions can also be divided using a similar method when the pancreatic duct region is used.

[0062] When the statistical length ratio of the centerline of the pancreas is used, in step S340, the sectioning unit 160 acquires the centerline of the pancreas by a known method (such as the technique disclosed in Non-Patent Document 3). Next, the sectioning unit 160 calculates the length of the centerline passing through each of the pancreatic head region, pancreatic body region, and pancreatic tail region from the ratio of the lengths of the centerlines passing through the partial regions set in advance. Next, the sectioning unit 160 tracks the centerline of the pancreas from the pancreatic head side (right shoulder side), and sets the sagittal cross section at the position where the tracked distance is equal to or greater than the length of the centerline passing through the pancreatic head region as the boundary 547 between the pancreatic head region and the pancreatic body region. Then, using the sagittal cross section corresponding to the boundary 547 as a reference, the sectioning unit 160 obtains the boundary 548 between the pancreatic body region and the pancreatic tail region in a similar manner, and divides the pancreas into a plurality of partial regions based on each boundary. The ratio of the lengths of the center lines passing through each partial region may be set by the user, or information such as the subject's height, weight, sex, and age may be obtained and the setting value of the length ratio may be changed based on that information. Alternatively, the boundaries of partial regions may be calculated based on the curvature of the running direction or the general shape (center line, etc.) of the pancreas to obtain the division results of the multiple partial regions of the pancreas. When dividing the multiple partial regions of the pancreas using the curvature of the center line of the pancreas, for example, the center line is projected onto a two-dimensional cross section in the same direction as the axial cross section, and the sagittal cross section corresponding to the position of the highest curvature on that cross section is calculated as the boundary 547 between the pancreatic head region and the pancreatic body region. Since the pancreas has a shape that bypasses the portal vein or superior mesenteric vein that defines the boundary between the pancreatic head region and the pancreatic body region, the boundary 547 between the pancreatic head region and the pancreatic body region can be calculated by the calculation method described above.

[0063] In the above description, the second algorithm for dividing the pancreas based on the shape of the pancreas depicted in the CT image data of the inference target is exemplified, but the implementation of the present disclosure is not limited thereto. For example, when there is CT image data obtained by photographing a subject at multiple contrast phases, the partial regions of the pancreas can be divided based on CT image data other than the CT image data of the inference target. For example, in step S340, the division unit 160 first searches for CT image data of a contrast phase in which a vascular region is easy to detect (the reliability is likely to be high) from among the CT image data of the multiple contrast phases, and acquires the CT image data. Next, the division unit 160 obtains a division result of the multiple partial regions of the pancreas using the first algorithm for the CT image data. Then, the division unit 160 aligns the CT image data with the CT image data of the inference target, and projects the division result of the multiple partial regions of the pancreas, thereby acquiring the division result of the multiple partial regions of the pancreas depicted in the CT image data of the inference target.

[0064] In the above embodiment and variations, in the first and second algorithms, a method of using a trained machine learning model or the like to acquire the regions of blood vessels and pancreas depicted in the CT image data to be inferred is exemplified. The method is not limited to these methods, and any method may be used as long as it is a method of acquiring each region corresponding to the CT image data to be inferred. For example, each region may be acquired by reading regional image data stored in the storage unit 110 or an external storage device, or a region annotated by user input may be acquired. In addition, each region may be acquired in a data format such as a bounding box or a landmark depending on the configuration of each algorithm, rather than regional image data.

[0065] The first algorithm is not limited to the above method, and any method, whether machine learning or image processing, can be used without compromising the essential features of the present disclosure, as long as it is a method for dividing a plurality of partial regions of the pancreas based on CT image data and a vascular region corresponding to the CT image data. Similarly, the second algorithm is also a method for dividing a plurality of partial regions of the pancreas based on the shape of the pancreas or the pixel value of the pancreas, and any method, whether machine learning or image processing, can be used without compromising the essential features of the present disclosure.

[0066] <Second embodiment> The second embodiment is an information processing system comprising A) to E), in particular, in C) a selection information acquisition unit, which acquires, as selection information for an algorithm, either (ii) information on a region related to blood vessels in the medical image data to be inferred, or (iii) information for determining the reliability of the vascular region acquired from the medical image data to be inferred.

[0067] In this embodiment, in order to divide the partial regions of the pancreas region in the CT image data, multiple algorithms are switched according to conditions. In the first embodiment, a method of switching the algorithm based on the supplementary information of the CT image data to be inferred was described, but in this embodiment, the algorithm is switched based on information about blood vessels depicted in the CT image data to be inferred. In other words, information about blood vessels is obtained as algorithm selection information, and based on that, a suitable algorithm is selected from multiple algorithms for dividing the partial regions of the pancreas to divide the partial regions of the pancreas. An example is shown in which information about blood vessels is obtained as algorithm selection information, and a suitable algorithm is selected from multiple algorithms for dividing the partial regions of the pancreas to divide the partial regions of the pancreas.

[0068] The information on blood vessels in this embodiment is the blood vessel diameter of a predetermined blood vessel calculated from the extraction result of the blood vessel region acquired by an arbitrary method from the CT image data of the inference target. The general flow of the process is as follows: first, the blood vessel region is extracted (acquired) from the CT image data of the inference target, and the blood vessel diameter is calculated from the extracted blood vessel region. Next, the reliability of the extracted blood vessel region is determined based on the calculated blood vessel diameter. Specifically, the determination of the reliability based on the blood vessel diameter in this embodiment refers to whether or not the calculated blood vessel diameter matches the standard blood vessel diameter (of the predetermined blood vessel) of the human body. Next, if the extracted blood vessel region is reliable, a first algorithm that divides partial regions based on the extracted blood vessel region is selected, and if not, a second algorithm that does not use the extracted blood vessel region is selected. Then, the selected algorithm divides multiple partial regions of the pancreas depicted in the CT image data of the inference target. That is, if the blood vessel diameter of the extracted blood vessel matches the standard blood vessel diameter of the human body (normal value), the extracted blood vessel region is deemed reliable (the region is appropriately extracted), and the first algorithm is selected. On the other hand, if it does not match the standard blood vessel diameter of the human body (abnormal value), the extracted blood vessel region is deemed unreliable (region extraction has not been performed properly), and the second algorithm is selected. In this embodiment, the multiple partial regions of the pancreas to be segmented are the pancreatic head region, pancreatic body region, and pancreatic tail region, as in the first embodiment. Hereinafter, the numerical information (blood vessel diameter in the above example) used in determining reliability may be referred to as reliability.

[0069] (Functional configuration) Hereinafter, the functional configuration of the information processing system 600 according to this embodiment will be described with reference to Fig. 6. As shown in the figure, the information processing system 600 is configured with a storage unit 610, an image data acquisition unit 620, a blood vessel region acquisition unit 630, a selection information acquisition unit 640, a selection unit 650, and a division unit 660.

[0070] The storage unit 610 is similar to the storage unit 110 in the first embodiment, and therefore a description thereof will be omitted.

[0071] The image data acquisition unit 620 acquires CT image data of the inference target from the storage unit 610 , and transmits it to the blood vessel region acquisition unit 630 and the division unit 660 .

[0072] The vascular region acquisition unit 630 first receives CT image data of the inference target from the image data acquisition unit 620. Next, the vascular region acquisition unit 630 receives the CT image data as an input, acquires a vascular region from the CT image data of the inference target using a machine learning model trained to segment the vascular region, and generates vascular region image data. Then, the vascular region acquisition unit 630 transmits the vascular region image data to the selection information acquisition unit 640 and the division unit 660.

[0073] The selection information acquisition unit 640 first receives the vascular region image data from the vascular region acquisition unit 630. Then, the selection information acquisition unit 640 acquires information for determining the reliability of the vascular region, which is information on blood vessels, from the vascular region image data, and transmits the information for determining the reliability of the vascular region to the selection unit 650. In this embodiment, the information for determining the reliability of the vascular region is a vascular diameter calculated from the vascular region held in the vascular region image data.

[0074] The selection unit 650 receives information for determining the reliability of the vascular region, which is selection information, from the selection information acquisition unit 640. Next, the selection unit 650 selects an algorithm to be used in the division unit 660 from among a plurality of algorithms, based on the information for determining the reliability of the vascular region. Then, the selection unit 650 transmits the selection result of the algorithm to the division unit 660.

[0075] In this embodiment, the algorithms include a first algorithm for dividing a plurality of partial regions of the pancreas based on CT image data and a vascular region corresponding to the CT image data, as in the first embodiment. In addition, the algorithms include at least a second algorithm for dividing a plurality of partial regions of the pancreas based on the shape of the pancreas or the pixel values ​​of the pancreatic region, as in the first embodiment. As described above, since the first algorithm is based on the vascular region, when the reliability of the extraction result of the vascular region acquired by the vascular region acquisition unit 630 is low, the division of the plurality of partial regions of the pancreas is likely to be inaccurate. Therefore, when the selection unit 650 determines that the reliability of the vascular region is low, it selects the second algorithm, and when not, it selects the first algorithm.

[0076] The sectioning unit 660 first receives the CT image data of the inference target from the image data acquisition unit 620, receives the vascular region image data from the vascular region acquisition unit 630, and receives the selection result of the algorithm from the selection unit 650. Next, the sectioning unit 660 reads an executable program corresponding to the selection result of the algorithm from the storage unit 610. Next, the sectioning unit 660 executes the process of the selected algorithm to acquire the sectioning results of multiple partial regions of the pancreas corresponding to the CT image data of the inference target, and generates partial region image data. Then, the sectioning unit 660 stores the partial region image data in the storage unit 610.

[0077] (Hardware configuration) The hardware configuration of the information processing system 600 according to this embodiment is the same as that of the first embodiment, and therefore a description thereof will be omitted.

[0078] (Processing Procedure) Next, the processing procedure of the information processing system 600 according to this embodiment will be described with reference to FIG.

[0079] (Step S710) In step S710, the image data acquisition unit 620 acquires CT image data of the inference target from the storage unit 610, and transmits it to the vascular region acquisition unit 630 and the division unit 660. The process of this step is similar to step S310 in the first embodiment, and therefore a detailed description thereof will be omitted.

[0080] (Step S720) In step S720, the vascular region acquisition unit 630 first receives CT image data of the inference target from the image data acquisition unit 620. Next, the vascular region acquisition unit 630 receives the CT image data as an input, acquires a vascular region from the CT image data of the inference target using a machine learning model trained to segment the vascular region, and generates vascular region image data. Then, the vascular region acquisition unit 630 transmits the vascular region image data to the selection information acquisition unit 640 and the division unit 660. As described above, the multiple partial regions of the pancreas divided in this embodiment are the pancreatic head region, pancreatic body region, and pancreatic tail region. The boundary between the pancreatic head region and pancreatic body region is defined by the portal vein (or superior mesenteric vein), and the boundary between the pancreatic body region and pancreatic tail region is defined by the aorta. Therefore, the vascular region acquisition unit 630 acquires the portal vein region and the aortic region using a machine learning model, and generates vascular region image data in which these regions can be distinguished. In this embodiment, U-Net, which is a kind of neural network, is used as the machine learning model. Therefore, U-Net is trained by a known method using a teacher data set consisting of CT image data, ground truth image data of the portal vein region, and ground truth image data of the aorta region. Here, the output of U-Net is a value converted into a probability by a sigmoid function or a softmax function, and the vascular region image data is a probability map representing the probability that each pixel belongs to the vascular region. Specifically, a two-channel image including an image representing the existence probability of the portal vein region and an image representing the existence probability of the aorta region is generated as the vascular region image data.

[0081] (Step S730) The selection information acquisition unit 640 first receives the vascular region image data from the vascular region acquisition unit 630. Then, the selection information acquisition unit 640 acquires information for determining the reliability of the vascular region, which is information about blood vessels, from the vascular region image data, and transmits the information for determining the reliability of the vascular region to the selection unit 650.

[0082] In this embodiment, the information for determining the reliability of the vascular region is the vascular diameter calculated from the vascular region held by the vascular region image data. Therefore, the selection information acquisition unit 640 calculates the vascular diameter for each of the portal vein region and the aorta region held by the vascular region image data. Specifically, the selection information acquisition unit 640 calculates the cross-sectional area of ​​each blood vessel for each axial cross-sectional image data constituting the vascular region image data, and calculates the blood vessel diameter for each axial cross-sectional image data by regarding the cross section of the blood vessel as a circle. At this time, since the vascular region image data is a probability map, the cross-sectional area is calculated after binarization with a predetermined threshold value. Then, the selection information acquisition unit 640 acquires the maximum value of each blood vessel diameter. That is, the selection information acquisition unit 640 acquires the maximum diameter of each blood vessel from the portal vein region and the aorta region held by the vascular region image data.

[0083] (Step S740) The selection unit 650 receives information for determining the reliability of the vascular region, which is selection information, from the selection information acquisition unit 640. Next, the selection unit 650 selects an algorithm to be used in the division unit 660 from among a plurality of algorithms, based on the information for determining the reliability of the vascular region. Then, the selection unit 650 transmits the selection result of the algorithm to the division unit 660. In this embodiment, the information for determining the reliability of the vascular region is the vascular diameter of the portal vein and the vascular diameter of the aorta acquired in step S730. If the accuracy of the blood vessels acquired by the vascular region acquisition unit 630 is low, the vascular diameter may indicate an abnormal value. If the vascular diameter indicates an abnormal value (if the reliability is low), the division of the partial region by the first algorithm may be inaccurate. Therefore, in this step, the selection unit 650 selects the first algorithm when both the vascular diameter of the portal vein and the vascular diameter of the aorta are normal values, and selects the second algorithm otherwise. Note that the normal values ​​of the vascular diameter are preset values, such as 0.7 to 12.0 mm for the portal vein and 15.0 to 35.0 mm for the aorta. The normal values ​​of the vascular diameter may be set based on statistical information. For example, information such as the height, weight, sex, and age of the subject may be acquired based on supplementary information of the CT image data, and the normal values ​​may be set based on the information.

[0084] (Step S750) The sectioning unit 660 first receives the CT image data of the inference target from the image data acquisition unit 620, receives the vascular region image data from the vascular region acquisition unit 630, and receives the selection result of the algorithm from the selection unit 650. Next, the sectioning unit 660 reads an executable program corresponding to the selection result of the algorithm from the storage unit 610. Next, the sectioning unit 660 executes the process of the selected algorithm to acquire the sectioning results of multiple partial regions of the pancreas corresponding to the CT image data of the inference target, and generates partial region image data. Then, the sectioning unit 660 stores the partial region image data in the storage unit 610.

[0085] The processing procedures of the first and second algorithms will be described in detail with reference to FIG.

[0086] First, the first algorithm will be described. In the first algorithm, first, the CT image data is input, and a machine learning model that has been trained to segment the pancreas region is used to acquire the pancreas region 533. Next, for each axial cross-sectional image data constituting the vascular region image data 520 received from the vascular region acquisition unit 630, the vascular region 521 corresponding to the portal vein or superior mesenteric vein and the left edge (the right end of the blood vessel in the figure) of the vascular region 522 corresponding to the aorta are detected. Then, the X-axis coordinate of the left edge of the vascular region 521 on the axial cross-section is detected as the boundary 547 between the pancreatic head region and the pancreatic body region, and the X-axis coordinate of the left edge of the vascular region 522 on the axial cross-section is detected as the boundary 548 between the pancreatic body region and the pancreatic tail region. By determining the boundary of the partial region for each axial cross-section in this manner, the boundary surface of the partial region becomes a curved surface. Then, based on the boundary surfaces (curved surfaces) of the detected partial regions, the pancreas region 533 is divided (determined) into a pancreas head region 544, a pancreas body region 545, and a pancreas tail region 546.

[0087] According to the processing procedure of the first algorithm described above, the sectioning unit 660 sections a plurality of partial regions of the pancreas depicted in the CT image data of the inference target based on the CT image data of the inference target and the blood vessel region corresponding to the CT image data of the inference target to generate partial region image data. Then, the sectioning unit 660 stores the partial region image data in the storage unit 610.

[0088] Next, the second algorithm will be described. As with the first algorithm, the second algorithm acquires the pancreas region 533 using a trained machine learning model. Next, based on the statistical volume ratio of the pancreas head region, pancreas body region, and pancreas tail region (for example, pancreas head:pancreas body:pancreas tail=3:1:2), the pancreas region 533 is divided into three partial regions, and the pancreas head region 544, pancreas body region 545, and pancreas tail region 546 are acquired. A specific division method has been described as a method based on the statistical volume ratio of the pancreas region in the variation of the first embodiment, and will not be described here.

[0089] According to the processing procedure of the second algorithm described above, the sectioning unit 660 divides the CT image data of the inference target into a plurality of partial regions of the pancreas based on the shape of the pancreas depicted in the CT image data of the inference target, and generates partial region image data. Then, the sectioning unit 660 stores the partial region image data in the storage unit 610.

[0090] Through the above processing procedure, the information processing system 600 can obtain suitable segmentation results for multiple partial regions of the pancreas by switching between the first algorithm and the second algorithm based on information regarding the contrast phase of the CT image data to be inferred.

[0091] (Variations) In the above embodiment, the vascular region acquisition unit 630 acquires a vascular region from the CT image data to be inferred using a trained machine learning model, but any method for acquiring a vascular region corresponding to the CT image data to be inferred may be used. For example, the vascular region may be acquired by reading vascular region image data stored in the storage unit 610 or an external storage device, or the vascular region annotated by user input may be acquired. In addition, the vascular region may be acquired in a data format such as a bounding box or a landmark, depending on the configuration of each algorithm, instead of the region image data. As an example of a configuration in which the blood vessel region is acquired as a bounding box, a case in which the bounding boxes of the blood vessels 511 and 512 regions are acquired for each axial cross section will be described. In this case, the selection information acquisition unit 640 acquires the length of the maximum side of the bounding box as the blood vessel diameter. That is, the selection information acquisition unit 640 acquires the length of the maximum side of the bounding box for all axial cross sections, and acquires the maximum value of the length of the maximum side as the maximum value of the blood vessel diameter, which is the selection information of the algorithm. Next, the selection unit 650 selects an algorithm in the same manner as in the above embodiment. Then, in the first algorithm in the division unit 660, the coordinates of the sides V1 and V2 corresponding to the left edges of the blood vessels 511 and 512 are acquired from among the sides of the bounding box of the blood vessel region, respectively, to identify the boundary 547 between the pancreatic head region and the pancreatic body region, and the boundary 548 between the pancreatic body region and the pancreatic head region. As an example of a configuration in which the vascular region is acquired as a landmark, a case in which the central coordinates of the regions of the blood vessel 511 and the blood vessel 512 are acquired for each axial cross section will be described. In this case, since it is difficult to acquire the maximum value of the vascular diameter (an example of the selection information of the algorithm) as in the above-mentioned method, for example, the selection information acquisition unit 640 acquires the average value of the certainty of the landmarks acquired for each axial cross section as the selection information. Next, if the average value of the certainty of the landmarks of each vascular region is equal to or greater than a predetermined value, the selection unit 650 determines that the reliability of the selection information of the algorithm is high and selects the first algorithm, and if not, selects the second algorithm. Then, in the first algorithm in the division unit 660, the landmarks L1 and L2 with the highest certainty are selected from the landmarks corresponding to the blood vessel 511 and the blood vessel 512, respectively, to identify the boundary 547 between the pancreatic head region and the pancreatic body region and the boundary 548 between the pancreatic body region and the pancreatic head region.

[0092] In the above embodiment, the maximum value of the blood vessel diameter is acquired as the selection information of the algorithm, but other statistics such as the average value, minimum value, and variance of the blood vessel diameter may be used. Also, any information may be used as long as it is based on the blood vessel extraction result acquired by the blood vessel region acquisition unit 630. The information based on the acquired blood vessel extraction result is, for example, the volume of the blood vessel region or the certainty of the blood vessel region. When the information based on the acquired blood vessel extraction result is volume, for example, in step S730, the selection information acquisition unit 640 calculates the volume of each of the blood vessel region 521 and the blood vessel region 522. Then, in step S740, the selection unit 650 selects the first algorithm when the volume of each of the blood vessel regions is a normal value, and selects the second algorithm otherwise. In the case where the information based on the acquired blood vessel extraction result is the certainty of the blood vessel region, for example, in step S730, the selection information acquisition unit 640 acquires the certainty of the blood vessel extraction result acquired by the blood vessel region acquisition unit 630. Since the blood vessel region image data in the above embodiment is a probability map, the selection information acquisition unit 640 acquires the maximum probability of each blood vessel region as the certainty of the blood vessel region for each axial cross-sectional image data constituting the blood vessel region image data. Since the blood vessel 511 and the blood vessel 512 are depicted in each axial cross-sectional image data constituting the CT image data 510 to be inferred, it is considered that the certainty of the extracted result of the acquired blood vessel region is low. Therefore, in step S740, the selection unit 650 selects the first algorithm when all the maximum probabilities are equal to or greater than a predetermined threshold value, and selects the second algorithm otherwise.

[0093] In the above embodiment and variations, a method is exemplified in which information based on the extraction result of blood vessels acquired by the blood vessel region acquisition unit 630 is used as the selection information of the algorithm. On the other hand, information on the blood vessel-related region of the medical image data to be inferred may be used as the selection information of the algorithm. The information on the blood vessel-related region is information obtained from the blood vessel-related region of the medical image data to be inferred, such as the presence or absence of anatomical abnormalities (diseases such as masses) near the blood vessels, the presence or absence of foreign bodies (stents, etc.), the presence or absence of image abnormalities (artifacts, etc.), etc. An example will be described in which the information on the region related to blood vessels is the presence or absence of a disease near the blood vessel. In this case, for example, in step S730, the selection information acquisition unit 640 uses a machine learning model that has been trained to segment a disease region from CT image data to be inferred to segment the disease region. If a disease region exists on the CT image data to be inferred, the distance between the blood vessel region acquired by the blood vessel region acquisition unit 630 and the disease region is calculated. If the distance between the blood vessel region and the disease region is short, it is considered that the difficulty of acquiring the blood vessel region is high, or the blood vessel has an anatomically abnormal shape. Therefore, in step S740, the selection unit 650 selects the first algorithm when the distance between the blood vessel region and the disease region is equal to or greater than a threshold value, and selects the second algorithm when the distance is less than the threshold value. The same applies to the presence or absence of a foreign body or image abnormality.

[0094] In the above embodiment, the first algorithm sets the left edge of the blood vessel region 521 as the boundary 547 between the pancreatic head region and the pancreatic body region, and the left edge of the blood vessel region 522 as the boundary 548 between the pancreatic body region and the pancreatic tail region, but the method of determining the boundary of the partial region is not limited to this. For example, instead of setting the left edge of each blood vessel region on the axial section as the boundary, the center point may be set as the boundary. Also, among the pixels in the acquired pancreatic region 533, pixels adjacent to each blood vessel region may be calculated, and each partial region may be divided based on the position of the pixel. More specifically, first, among the pixels in the pancreatic region 533, a pixel P1 adjacent to the blood vessel region 521 and a pixel P2 adjacent to the blood vessel region 522 are calculated. Next, a cross section S1 of the pancreatic region 533 at pixel P1 and a cross section S2 of the pancreatic region 533 at pixel P2 are calculated. Then, the cut surface S1 is set as a boundary 547 between the pancreatic head region and the pancreatic body region, and the cut surface S2 is set as a boundary 548 between the pancreatic body region and the pancreatic tail region, and the pancreatic region 533 is divided into a pancreatic head region 544, a pancreatic body region 545, and a pancreatic tail region 546. The cut surfaces may be determined by sagittal sections corresponding to the X coordinates of pixels P1 and P2, or by a section including pixel P1 perpendicular to the running direction of the pancreatic region and a section including pixel P2 perpendicular to the running direction of the pancreatic region. Any other method may be used as long as it is a method for dividing a plurality of partial regions of the pancreas based on the position of the blood vessel region.

[0095] <Third embodiment> The information processing system of the present disclosure according to the third embodiment includes the following. a memory unit that stores a plurality of algorithms, including a third algorithm and a fourth algorithm, The third algorithm is an algorithm that divides a plurality of partial regions of the pancreas in medical image data and is highly dependent on blood vessel regions; The fourth algorithm is an algorithm that divides a plurality of partial regions of the pancreas in medical image data and has a low dependency on blood vessel regions; an image data acquisition unit for acquiring medical image data to be inferred; A selection information acquisition unit that acquires selection information including at least one piece of information among the following (i) to (iii): (i) Image conditions of the medical image data to be inferred (ii) information on a region related to blood vessels in the medical image data to be inferred (iii) information for determining the reliability of a vascular region obtained from the medical image data of the inference target; a selection unit that selects at least one algorithm from the third algorithm and the fourth algorithm from the storage unit using the selection information; and A partitioning unit that partitions a plurality of partial regions of the pancreas in the medical image data of the inference target by applying the selected algorithm to the medical image data of the inference target.

[0096] In this embodiment, a third algorithm that is highly dependent on the vascular region and a fourth algorithm that is less dependent on the vascular region are switched based on the selection information of the algorithm. The algorithm that is highly dependent on the vascular region is an algorithm that explicitly uses the vascular region, and the algorithm that is less dependent is an algorithm that implicitly uses the vascular region. For example, the vascular region is used as part of an inference device that divides multiple partial regions of the pancreas in medical image data. In this embodiment, the third algorithm is an algorithm that detects a specific position (e.g., the left edge) of the vascular region and calculates the boundaries of each of the multiple partial regions of the pancreas to divide the partial regions. On the other hand, the fourth algorithm is an algorithm that divides the multiple partial regions of the pancreas using a machine learning model that inputs CT image data and vascular region image data. The third algorithm explicitly uses the position of the vascular region, so it is highly dependent on the vascular region, while the fourth algorithm leaves the reference of the vascular region to the machine learning model (implicitly), so it can be said that it is less dependent on the vascular region.

[0097] (Functional configuration) Based on the second embodiment, the functional configuration of an information processing system 800 according to this embodiment shown in Fig. 8 will be described below. The information processing system 800 includes a storage unit 810, an image data acquisition unit 820, a blood vessel region acquisition unit 830, a selection information acquisition unit 840, a selection unit 850, and a division unit 860.

[0098] The memory unit 810, image data acquisition unit 820, vascular region acquisition unit 830, selection information acquisition unit 840, and sectioning unit 860 are similar to the memory unit 610, image data acquisition unit 620, vascular region acquisition unit 630, selection information acquisition unit 640, and sectioning unit 860 in the second embodiment, respectively, and therefore their explanations are omitted.

[0099] The selection unit 850 receives information for determining the reliability of the vascular region, which is selection information, from the selection information acquisition unit 840. Next, the selection unit 850 selects an algorithm to be used in the division unit 860 from among a plurality of algorithms, based on the information for determining the reliability of the vascular region. Then, the selection unit 850 transmits the selection result of the algorithm to the division unit 860.

[0100] In this embodiment, the multiple algorithms include a third algorithm that is highly dependent on the vascular region and a fourth algorithm that is less dependent on the vascular region. More specifically, the third algorithm is an algorithm that detects a specific position (e.g., the left edge) of the vascular region and calculates the boundaries of each of the multiple partial regions of the pancreas to divide the partial regions. On the other hand, the fourth algorithm is an algorithm that divides the multiple partial regions of the pancreas by a machine learning model that inputs CT image data and vascular region image data. Since the third algorithm is highly dependent on the vascular region, when the reliability of the extraction result of the vascular region acquired by the vascular region acquisition unit 830 is low, the division of the multiple partial regions of the pancreas is likely to be inaccurate. Therefore, when the selection unit 850 determines that the reliability of the vascular region is low, it selects the fourth algorithm, and when not, it selects the third algorithm.

[0101] (Hardware configuration) The hardware configuration of the information processing system 800 according to this embodiment is the same as that of the second embodiment, and therefore a description thereof will be omitted.

[0102] (Processing Procedure) Next, a processing procedure of the information processing system 800 according to this embodiment will be described with reference to Fig. 7. Note that descriptions overlapping with the second embodiment will be omitted as appropriate.

[0103] (Steps S710 to S730) Steps S710 to S730 are the same as steps S710 to S730 in the second embodiment, and therefore a description thereof will be omitted.

[0104] (Step S740) The selection unit 850 receives information for determining the reliability of the vascular region, which is selection information, from the selection information acquisition unit 840. Next, the selection unit 850 selects an algorithm to be used in the division unit 860 from among a plurality of algorithms, based on the information for determining the reliability of the vascular region. Then, the selection unit 850 transmits the selection result of the algorithm to the division unit 860.

[0105] The information for determining the reliability of the vascular region in this embodiment is the vascular diameter of the portal vein and the vascular diameter of the aorta acquired in step S730, as in the second embodiment. If the vascular diameter indicates an abnormal value (if the reliability is low), the classification of the partial region by the third algorithm, which is highly dependent on the vascular region, may be inaccurate. Therefore, in this step, the selection unit 850 selects the third algorithm when both the vascular diameter of the portal vein and the vascular diameter of the aorta are normal values, and selects the fourth algorithm otherwise.

[0106] (Step S750) The sectioning unit 860 first receives CT image data of the inference target from the image data acquisition unit 820, receives vascular region image data from the vascular region acquisition unit 830, and receives the selection result of the algorithm from the selection unit 850. Next, the sectioning unit 860 reads an executable program corresponding to the selection result of the algorithm from the storage unit 810. Next, the sectioning unit 860 executes the processing of the selected algorithm to acquire the sectioning results of multiple partial regions of the pancreas corresponding to the CT image data of the inference target, and generates partial region image data. Then, the sectioning unit 860 stores the partial region image data in the storage unit 810.

[0107] The third algorithm in this embodiment, like the first algorithm in the second embodiment, divides the pancreas into a plurality of partial regions by calculating the boundary 547 and the boundary 548 based on the left edge of each of the blood vessel region 521 and the blood vessel region 522. The fourth algorithm, like the first algorithm in the first embodiment, is an algorithm that divides the pancreas into a plurality of partial regions depicted in the CT image data to be inferred by a machine learning model that inputs the CT image data and the blood vessel region image data. By the processing procedures of these algorithms, the division unit 860 divides the pancreas into a plurality of partial regions depicted in the CT image data to be inferred, and generates partial region image data. Then, the division unit 860 stores the partial region image data in the storage unit 810.

[0108] Through the above processing procedure, the information processing system 800 can obtain suitable segmentation results for multiple partial regions of the pancreas by switching between the third algorithm and the fourth algorithm based on information regarding the contrast phase of the CT image data to be inferred.

[0109] (Variations) The third algorithm may be any algorithm that explicitly uses the vascular region (highly dependent method). For example, the algorithm described in the variation of the first algorithm in the second embodiment may be used. The fourth algorithm may be any algorithm that implicitly uses the vascular region or does not use the vascular region (lowly dependent method). For example, the algorithm described in the variation of the second algorithm in the first or second embodiment may be used.

[0110] (Other Examples) The present disclosure can also be realized by a process in which a program for implementing one or more functions of the above-described embodiments and examples is supplied to a system or device via a network or a storage medium, and one or more processors in the computer of the system or device read and execute the program. It can also be realized by a circuit (e.g., ASIC) for implementing one or more functions. A computer may have one or more processors or circuits, and may include separate computers or a network of separate processors or circuits for reading and executing computer-executable instructions. The processor or circuitry may include a central processing unit (CPU), a microprocessing unit (MPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), or a field programmable gateway (FPGA), and the processor or circuitry may include a digital signal processor (DSP), a data flow processor (DFP), or a neural processing unit (NPU).

[0111] The embodiments of the present disclosure include the following configurations. (Configuration 1) a storage unit that stores a plurality of algorithms including a first algorithm and a second algorithm, wherein the first algorithm is an algorithm for dividing a plurality of partial regions of a pancreas in medical image data relating to a medical image including at least a part of the pancreas based on a vascular region of the medical image data, and the second algorithm is an algorithm for dividing a plurality of partial regions of a pancreas in medical image data relating to a medical image including at least a part of the pancreas based on at least one of a shape and a pixel value of the pancreas region of the medical image data. An image data acquisition unit that acquires medical image data of an inference target; A selection information acquisition unit that acquires selection information including at least one piece of information from the following (i) to (iii); (i) Image conditions of the medical image data to be inferred (ii) information on a region related to blood vessels in the medical image data to be inferred (iii) information for determining the reliability of a vascular region obtained from the medical image data of the inference target; a selection unit that selects at least one algorithm from the first algorithm and the second algorithm from the storage unit using the selection information; a partitioning unit that partitions a plurality of partial regions of the pancreas in the medical image data of the inference target by applying the selected algorithm to the medical image data of the inference target; An information processing system having the above configuration. (Configuration 2) 2. The information processing system according to configuration 1, wherein the plurality of partial regions includes at least one partial region selected from the group consisting of a pancreatic head region, a pancreatic body region, and a pancreatic tail region. (Configuration 3) 3. The information processing system according to claim 1 or 2, wherein the vascular region includes a portal vein region and an aortic region. (Configuration 4) The information processing system according to any one of configurations 1 to 3, wherein the medical image is captured by any imaging device selected from the group consisting of an X-ray computed tomography (X-ray CT) device, a nuclear magnetic resonance imaging (MRI) device, a positron emission tomography (PET) device, and an ultrasound diagnostic device. (Configuration 5) In the selection information acquisition unit, (i) Image conditions of the medical image data to be inferred is obtained as the algorithm selection information, The image conditions include at least one of a group consisting of a contrast phase, a contrast time, a reconstruction function, an X-ray dose, and an amount of noise; In the selection unit, selecting the first algorithm when the image condition is not determined to lower the detection accuracy of the blood vessel; The second algorithm is selected when it is determined that the image condition reduces the detection accuracy of the blood vessel. 5. An information processing system according to any one of configurations 1 to 4. (Configuration 6) moreover, a vascular region acquisition unit for acquiring a vascular region of the medical image data of the inference target; 5. The information processing system according to claim 1, wherein the selection information acquisition unit acquires the selection information based on the vascular region acquired by the vascular region acquisition unit. (Configuration 7) The selection information acquired by the selection information acquisition unit includes (ii) information on a region related to a blood vessel of the medical image data of the inference target, The selection information is The blood vessel region is obtained by referring to the blood vessel region of the medical image data of the inference target acquired by the blood vessel region acquisition unit, The presence or absence of at least one abnormality in a group of abnormalities consisting of an anatomical abnormality near the vascular region, a foreign body near the vascular region, and an image abnormality, In the selection unit, selecting the first algorithm when the selection information is information indicating that at least one of the anomalies is not present; 7. The information processing system according to claim 6, wherein the second algorithm is selected when the presence of at least one of the group of abnormalities is indicated. (Configuration 8) The selection information acquired by the selection information acquisition unit is (iii) information for determining the reliability of a blood vessel region obtained from the medical image data of the inference target; The selection information is The blood vessel region is obtained by referring to the blood vessel region of the medical image data of the inference target acquired by the blood vessel region acquisition unit, information for determining the reliability of the vascular region, including at least one of an estimated diameter, volume, and confidence level of the vascular region; The selection unit is When at least one of the pieces of information for determining the reliability of the selection information indicates that the reliability of the vascular region is high, the first algorithm is selected; The information processing system according to configuration 6, characterized in that the second algorithm is selected when at least any of the information for determining the reliability indicates that the reliability of the vascular region is low. (Configuration 9) The information processing system according to any one of configurations 1 to 8, wherein the first algorithm is an algorithm for dividing a plurality of partial regions of the pancreas in medical image data by a machine learning-based method using information including information on vascular regions as input. (Configuration 10) The information processing system according to any one of configurations 1 to 9, wherein the first algorithm is an algorithm for dividing a plurality of partial regions of the pancreas based on a pancreatic region and a vascular region of medical image data relating to a medical image including at least a portion of the pancreas. (Configuration 11) The information processing system according to any one of configurations 1 to 10, wherein the first algorithm is an algorithm for determining boundaries between a plurality of partial regions based on the vascular region, and for dividing the plurality of partial regions of the pancreas based on the boundaries. (Configuration 12) The information processing system according to any one of configurations 1 to 11, wherein the second algorithm is an algorithm for dividing a plurality of partial regions of the pancreas in the medical image data by a method based on machine learning that is learned by inputting medical image data relating to a medical image including at least a portion of the pancreas. (Configuration 13) The information processing system according to any one of configurations 1 to 12, wherein the second algorithm is an algorithm for dividing a plurality of partial regions of the pancreas based on a pancreatic region of medical image data relating to a medical image including at least a portion of the pancreas. (Configuration 14) The information processing system of configuration 13, wherein the second algorithm is an algorithm for dividing multiple sub-regions of the pancreas based on either a statistical volume ratio of multiple sub-regions of the pancreas or a statistical length ratio of centerlines of the pancreas. (Configuration 15) The information processing system according to configuration 13, wherein the second algorithm is an algorithm for dividing the pancreas into a plurality of partial regions by fitting a shape model of the pancreas divided into a plurality of partial regions to the pancreas region. (Configuration 16) a storage unit that stores a plurality of algorithms including a third algorithm and a fourth algorithm, where the third algorithm is an algorithm that divides a plurality of partial regions of the pancreas in the medical image data and has a strong dependency on the blood vessel region, and the fourth algorithm is an algorithm that divides a plurality of partial regions of the pancreas in the medical image data and has a weak dependency on the blood vessel region; An image data acquisition unit that acquires medical image data of an inference target; A selection information acquisition unit that acquires selection information including at least one piece of information from (i) to (iii) below; (i) Image conditions of the medical image data to be inferred (ii) information on a region related to blood vessels in the medical image data to be inferred (iii) information for determining the reliability of a vascular region obtained from the medical image data of the inference target; a selection unit that selects at least one algorithm from the third algorithm and the fourth algorithm from the storage unit using the selection information; a partitioning unit that partitions a plurality of partial regions of the pancreas of the medical image data of the inference target by applying the selected algorithm to the medical image data of the inference target; An information processing system having the above configuration. (Configuration 17) 17. A program for causing a computer to function as each of the means of the information processing system according to any one of configurations 1 to 16. (Configuration 18) A recording medium storing the program according to configuration 17 in a computer-readable format. [Explanation of symbols]

[0112] 110 Storage section 120 Image data acquisition unit 140 Selection information acquisition unit 150 Selection Section 160 Division 201 CPU 202 Main Memory 203 Magnetic Disk 204 Display Memory 205 Monitor 206 Mouse 207 Keyboard 208 Common Bus 510 CT image data 511 Blood vessels defining the border between the head and body of the pancreas 512 Blood vessels defining the border between the body and tail of the pancreas 513 Pancreas 520 Blood vessel area image data 521 Vascular area 522 Vascular area 530 Pancreatic Region Image Data 533 Pancreatic Region 540 Partial Area Image Data 544 Pancreatic Head Region 545 Pancreatic body region 546 Pancreatic Tail Region 547 Boundary between head and body of pancreas 548 Boundary between the body and tail of the pancreas 600 Information Processing Systems 610 Storage section 620 Image data acquisition unit 630 Blood vessel region acquisition unit 640 Selection Information Acquisition Unit 650 Selection Section 660 Division

Claims

1. a storage unit that stores a plurality of algorithms including a first algorithm and a second algorithm, wherein the first algorithm is an algorithm for dividing a plurality of partial regions of a pancreas in medical image data relating to a medical image including at least a part of the pancreas based on a vascular region of the medical image data, and the second algorithm is an algorithm for dividing a plurality of partial regions of a pancreas in medical image data relating to a medical image including at least a part of the pancreas based on at least one of a shape and a pixel value of the pancreas region of the medical image data. An image data acquisition unit that acquires medical image data of an inference target; A selection information acquisition unit that acquires selection information including at least one piece of information from the following (i) to (iii); (i) Image conditions of the medical image data to be inferred (ii) information on a blood vessel-related region of the medical image data to be inferred; (iii) information for determining the reliability of a blood vessel region obtained from the medical image data of the inference target; a selection unit that selects at least one algorithm from the first algorithm and the second algorithm from the storage unit using the selection information; a partitioning unit that partitions a plurality of partial regions of the pancreas in the medical image data of the inference target by applying the selected algorithm to the medical image data of the inference target; An information processing system having the above configuration.

2. 2. The information processing system according to claim 1, wherein the plurality of partial regions includes at least one partial region selected from the group consisting of a pancreatic head region, a pancreatic body region, and a pancreatic tail region.

3. The information processing system according to claim 1 , wherein the vascular region includes a portal vein region and an aortic region.

4. 2. The information processing system according to claim 1, wherein the medical image is captured by any imaging device selected from the group consisting of an X-ray computed tomography (X-ray CT) device, a nuclear magnetic resonance imaging (MRI) device, a positron emission tomography (PET) device, and an ultrasound diagnostic device.

5. In the selection information acquisition unit, (i) Image conditions of the medical image data to be inferred is obtained as the algorithm selection information, The image conditions include at least one of a group consisting of a contrast phase, a contrast time, a reconstruction function, an X-ray dose, and an amount of noise; In the selection unit, selecting the first algorithm when it is not determined that the image condition reduces the detection accuracy of the blood vessel; The second algorithm is selected when it is determined that the image condition reduces the detection accuracy of the blood vessel. The information processing system according to claim 1 .

6. moreover, a vascular region acquisition unit for acquiring a vascular region of the medical image data of the inference target; 5. The information processing system according to claim 1, wherein the selection information acquisition unit acquires the selection information based on the vascular region acquired by the vascular region acquisition unit.

7. The selection information acquired by the selection information acquisition unit includes (ii) information on a region related to a blood vessel of the medical image data of the inference target, The selection information is The blood vessel region is obtained by referring to the blood vessel region of the medical image data of the inference target acquired by the blood vessel region acquisition unit, The presence or absence of at least one abnormality in a group of abnormalities consisting of an anatomical abnormality near the vascular region, a foreign body near the vascular region, and an image abnormality, In the selection unit, selecting the first algorithm when the selection information is information indicating that at least any one of the anomalies is not present; 7. The information processing system according to claim 6, wherein the second algorithm is selected when the presence of at least one of the group of abnormalities is indicated.

8. The selection information acquired by the selection information acquisition unit is (iii) includes information for determining a reliability of a blood vessel region acquired from the medical image data of the inference target, The selection information is The blood vessel region is obtained by referring to the blood vessel region of the medical image data of the inference target acquired by the blood vessel region acquisition unit, information for determining the reliability of the vascular region, including at least one of an estimated diameter, volume, and confidence level of the vascular region; The selection unit is When at least one of the pieces of information for determining the reliability of the selection information indicates that the reliability of the vascular region is high, the selection information selects the first algorithm; The information processing system according to claim 6 , wherein the second algorithm is selected when at least one of the pieces of information for determining the reliability indicates that the reliability of the blood vessel region is low.

9. The information processing system according to claim 1 , wherein the first algorithm is an algorithm for dividing a plurality of partial regions of the pancreas in medical image data using a machine learning-based method that uses as input information including information regarding vascular regions.

10. The information processing system according to claim 1 , wherein the first algorithm is an algorithm for dividing a plurality of partial regions of the pancreas based on a pancreas region and a vascular region of medical image data relating to a medical image including at least a portion of the pancreas.

11. The information processing system according to claim 10, wherein the first algorithm is an algorithm for determining boundaries between a plurality of partial regions based on the vascular region, and for dividing the plurality of partial regions of the pancreas region based on the boundaries.

12. The information processing system according to any one of claims 1 to 4, wherein the second algorithm is an algorithm that divides multiple partial regions of the pancreas in the medical image data by a method based on machine learning that is learned by inputting medical image data relating to a medical image that includes at least a portion of the pancreas.

13. The information processing system according to claim 1 , wherein the second algorithm is an algorithm for dividing a pancreatic region into a plurality of partial regions based on a pancreatic region of medical image data relating to a medical image including at least a portion of the pancreas.

14. The information processing system according to claim 13 , wherein the second algorithm is an algorithm for dividing a plurality of sub-regions of the pancreas based on either a statistical volume ratio of the plurality of sub-regions of the pancreas or a statistical length ratio of the centerline of the pancreas.

15. The information processing system according to claim 13 , wherein the second algorithm is an algorithm for dividing the pancreatic region into a plurality of partial regions by fitting a shape model of the pancreas divided into a plurality of partial regions to the pancreatic region.

16. a storage unit that stores a plurality of algorithms including a third algorithm and a fourth algorithm, where the third algorithm is an algorithm that divides a plurality of partial regions of the pancreas in the medical image data and has a high dependency on the blood vessel region, and the fourth algorithm is an algorithm that divides a plurality of partial regions of the pancreas in the medical image data and has a low dependency on the blood vessel region; An image data acquisition unit that acquires medical image data of an inference target; A selection information acquisition unit that acquires selection information including at least one piece of information from the following (i) to (iii); (i) Image conditions of the medical image data to be inferred (ii) information on a blood vessel-related region of the medical image data to be inferred; (iii) information for determining the reliability of a blood vessel region obtained from the medical image data of the inference target; a selection unit that selects at least one algorithm from the third algorithm and the fourth algorithm from the storage unit using the selection information; a partitioning unit that partitions a plurality of partial regions of the pancreas of the medical image data of the inference target by applying the selected algorithm to the medical image data of the inference target; An information processing system having the above configuration.

17. 17. A program for causing a computer to function as each of the means of the information processing system according to any one of claims 1 to 4 and 16.

18. A recording medium storing the program according to claim 17 in a computer-readable format.

Images