Image processing apparatus, image processing method, and program

Abstract

To provide a technology to detect a lesion area more appropriately by avoiding or suppressing erroneous detection of the lesion area.SOLUTION: An image processing device executes lesion area estimation processing using a learning tool for each of a plurality of slice images obtained by slicing a subject by cross sections having a same normal direction. The image processing device determines a first area 71a and a second area 71b as lesion areas on condition that the difference between a cross section position of the first area 71a, which is an estimated lesion area detected by the lesion area estimation processing in a first slice image L(j-1) of the plurality of slice images and a cross section position of the second area 71b, which is an estimated lesion area detected by the lesion area estimation processing in a second slice image Lj of the plurality of slice images is within a predetermined range ΔP.SELECTED DRAWING: Figure 7

Description

Technical Field

[0001] The present invention relates to a processing technique for a plurality of two-dimensional slice images and related techniques.

Background Art

[0002] There is a technique of acquiring a plurality of two-dimensional slice images (cross-sectional images) of a subject using an MRI (Magnetic Resonance Imaging) apparatus or a CT (Computed Tomography) apparatus, and identifying a lesion region using the plurality of two-dimensional slice images.

[0003] For example, there is a technique of determining a lesion region or the like in each of a plurality of two-dimensional slice images (hereinafter, also simply referred to as slice images) using a learning device learned by machine learning using a neural network (see Patent Document 1).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] By the way, in the technique of detecting a lesion region from each of a plurality of slice images using a learning device, the lesion region is not always correctly detected in all slice images. For example, in some slice images, a region that is not originally a lesion region may be erroneously detected as a lesion region. In short, the lesion region may be overdetected.

[0006] Therefore, an object of the present invention is to provide a technique capable of avoiding or suppressing misdetection of a lesion region and detecting the lesion region more accurately.

Means for Solving the Problem

[0007] In order to solve the above problems, the image processing apparatus according to the present invention includes a control unit that processes a group of slice images. The group of slice images is composed of a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction, and the subject is sliced at a plurality of different positions in the normal direction. The control unit executes a lesion region estimation process using a learning device for each of the plurality of slice images. By the lesion region estimation process, when a first region is detected as an estimated lesion region in a first slice image that is a cross-sectional image at a first position in the normal direction of the subject among the plurality of slice images, among the plurality of slice images, at a position different from the first position in the normal direction and within a predetermined distance from the first position in the normal direction, when an estimated lesion region having a cross-sectional position within a predetermined range different from the cross-sectional position of the first region is not detected in any of the slice images having such a position, the first region is determined as a non-lesion region. Among the plurality of slice images, a second slice image different from the first slice image, which is a cross-sectional image at a second position that is different from the first position in the normal direction and within a predetermined distance from the first position in the normal direction, when a second region, which is an estimated lesion region having a cross-sectional position within a predetermined range different from the cross-sectional position of the first region, is detected by the lesion region estimation process, the first region and the second region are determined as lesion regions. location It is characterized by this. and the predetermined distance is a distance equal to or greater than the distance corresponding to the thickness of two slice images ​To solve the above problems, an image processing apparatus according to the present invention includes a control unit that processes a group of slice images. The group of slice images includes a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction. The control unit executes a lesion region estimation process using a learning device for each of the plurality of slice images. When the difference between the in-plane position of a first region, which is an estimated lesion region detected by the lesion region estimation process in a first slice image among the plurality of slice images, and the in-plane position of a second region, which is an estimated lesion region detected by the lesion region estimation process in a second slice image among the plurality of slice images, is within a predetermined range, the first region and the second region are determined as lesion regions. When the first region and the second region are determined as lesion regions and there is a third slice image between the first slice image and the second slice image among the plurality of slice images, if an estimated lesion region in the third slice image, the difference in the in-plane position of which from at least one of the first region and the second region is within the predetermined range, is not detected by the lesion region estimation process, a third region complemented in the third slice image based on the first region and the second region is determined as the lesion region in the third slice image. An image processing apparatus characterized by this is provided.

[0012] The control unit executes a determination process for determining a plurality of estimated lesion regions detected in different slice images of a first group of slice images as lesion regions on the condition that the difference in the in-plane position between the plurality of estimated lesion regions is within a predetermined range. Also, for a second group of slice images different from the first group of slice images, which has a normal in a direction different from the normal of the first group of slice images, the control unit may execute the determination process. direction different from second a cross-section having a normal in a direction and is composed of a plurality of slice images obtained by slicing the subject at a plurality of different positions in the second direction The determination process may also be executed for the second group of slice images.

[0013] The lesion region may be a fracture region.

[0014] In order to solve the above problems, the image processing method according to the present invention includes: a) performing lesion region estimation processing using a learning device on each of a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction, where the subject is sliced at a plurality of different positions in the normal direction; and b) determining whether a first region detected as an estimated lesion region in a first slice image, which is a cross-sectional image at a first position in the normal direction of the subject among the plurality of slice images, is a lesion region. The step b) includes: b-1) when, in any of the plurality of slice images having a position within a fixed distance from the first position in the normal direction, no estimated lesion region having a difference in the in-plane position within a predetermined range from the in-plane position of the first region is detected by the lesion region estimation processing, determining the first region as a non-lesion region; and b-2) when, in a second slice image different from the first slice image among the plurality of slice images, which is a cross-sectional image at a second position different from the first position in the normal direction and within the predetermined distance from the first position in the normal direction, a second region, which is an estimated lesion region having a difference in the in-plane position within a predetermined range from the in-plane position of the first region, is detected by the lesion region estimation processing, determining both the first region and the second region as lesion regions. location This is characterized by the above. wherein the predetermined distance is a distance equal to or greater than the distance corresponding to the thickness of two slice images This is characterized by the above. In order to solve the above problems, an image processing method according to the present invention includes: a) performing a lesion area estimation process using a learning device on each of a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction; b) determining that a first area, which is an estimated lesion area detected by the lesion area estimation process in a first slice image among the plurality of slice images, and a second area, which is an estimated lesion area detected by the lesion area estimation process in a second slice image among the plurality of slice images, are lesion areas on the condition that a difference in in-plane positions between the two areas is within a predetermined range; and c) when the first area and the second area are determined to be lesion areas and there is a third slice image between the first slice image and the second slice image among the plurality of slice images, and an estimated lesion area in the third slice image, the in-plane position of which with respect to at least one of the first area and the second area is within the predetermined range, is not detected by the lesion area estimation process, determining a third area, which is complemented in the third slice image based on the first area and the second area, as the lesion area in the third slice image. In order to solve the above problems, a program according to the present invention causes a computer to perform: a) performing a lesion area estimation process using a learning device on each of a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction, the plurality of slice images being obtained by slicing the subject at a plurality of different positions in the normal direction; and b) determining whether a first area, which is detected as an estimated lesion area in a first slice image, which is a cross-sectional image at a first position in the normal direction of the subject, among the plurality of slice images by the lesion area estimation process, is a lesion area. The step b) includes: b-1) among the plurality of slice images, at a position different from the first position in the normal direction and at a position in the normal direction that is locationWhen, in any of the slice images having a position within a predetermined distance, no estimated lesion region having a cross-sectional position within a predetermined range of difference from the cross-sectional position of the first region in the cross-section is detected by the lesion region estimation process, determining the first region as a non-lesion region; and b-2) Among the plurality of slice images, a second slice image different from the first slice image, which is a cross-sectional image at a second position that is different from the first position in the normal direction and is within the predetermined distance from the first position in the normal direction. When a second region, which is an estimated lesion region having a cross-sectional position within a predetermined range of difference from the cross-sectional position of the first region, is detected by the lesion region estimation process in the second slice image, determining the first region and the second region as lesion regions. wherein the predetermined distance is a distance equal to or greater than the distance corresponding to the thickness of two slice images Characterized in that. To solve the above problems, the program according to the present invention causes a computer to: a) execute a lesion region estimation process using a learner for each of a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction; b) based on the condition that the difference between the cross-sectional position of a first region, which is an estimated lesion region detected by the lesion region estimation process in a first slice image among the plurality of slice images, and the cross-sectional position of a second region, which is an estimated lesion region detected by the lesion region estimation process in a second slice image among the plurality of slice images, is within a predetermined range, determining the first region and the second region as lesion regions; and c) when the first region and the second region are determined as lesion regions and there is a third slice image between the first slice image and the second slice image among the plurality of slice images, and no estimated lesion region having a cross-sectional position within the predetermined range of difference from at least one of the first region and the second region is detected by the lesion region estimation process in the third slice image, determining a third region complemented in the third slice image based on the first region and the second region as the lesion region in the third slice image. It is a program for causing the execution.

Effect of the Invention

[0015] According to the present invention, it is possible to avoid or suppress the false detection of a lesion region and detect the lesion region more accurately.

Brief Description of the Drawings

[0016]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Modes for Carrying Out the Invention

[0017] Hereinafter, embodiments of the present invention will be described based on the drawings.

[0018] <1. First Embodiment> <1-1. System Overview> FIG. 1 is a block diagram showing an image processing system 10. The image processing system 10 is a system that processes a group of slice images 210. The group of slice images 210 is composed of a plurality of two-dimensional slice images (tomographic images) 220 (see FIG. 5) obtained by slicing a subject (such as a subject) with a cross-section perpendicular to a reference axis. In other words, the group of slice images 210 has a plurality of slice images obtained by slicing the subject with cross-sections having the same normal direction.

[0019] As shown in FIG. 1, the image processing system 10 includes a slice image generation device 20 and an image processing device 30. The slice image generation device 20 and the image processing device 30 are connected by wire (or wirelessly) and can communicate with each other. The image processing device 30 receives information (such as slice images) generated (acquired) by the slice image generation device 20 from the slice image generation device 20 via a predetermined connection cable or the like.

[0020] The slice image generation device 20 is composed of an MRI (Magnetic Resonance Imaging) device or a CT (Computed Tomography) device, etc. The slice image generation device 20 generates and acquires a plurality of two-dimensional slice images (simply referred to as slice images) 220 regarding a subject (such as a subject). The plurality of slice images 220 are images obtained by slicing the subject with a cross-section perpendicular to the reference axis at a plurality of different positions on the reference axis (for example, at different positions with a pitch of 0.6 mm to 1 mm (millimeter)). The plurality of slice images 220 are acquired over a predetermined range (for example, 300 mm) in the reference axis direction and are composed of several hundred to several thousand (for example, 500) images. In the slice image generation device 20, for example, a plurality of two-dimensional slice images 220 regarding the lumbar region including the pelvis are imaged and acquired (see FIG. 5). In this way, a group of slice images 210 including a plurality of two-dimensional slice images 220 is generated.

[0021] The image processing device 30 executes an inference process for each of a plurality of slice images 220 (also referred to as 240) obtained for a certain subject using a learning model 410 (also referred to as a learned model 420) whose learning parameters have been adjusted in advance (see FIG. 2). Specifically, the image processing device 30 executes an inference process for estimating the (unknown) lesion site of each slice image 240 using the above-described learned model 420. Here, a fracture site (more specifically, a fracture site in the pelvis) is mainly exemplified as the lesion site. In this inference process, for example, the presence or absence of a fracture site and the position of the fracture site are specified (estimated). FIG. 2 is a conceptual diagram showing the processing in the inference stage in machine learning. Note that the region detected (estimated) as the lesion region by this inference process is also referred to as an estimated lesion region. Further, since this region (estimated lesion region) is a candidate region for the lesion region, it is also referred to as a lesion candidate region or the like.

[0022] Note that the learned model 420 is generated by adjusting the learning parameters of the learning model 410 (learning device) using a predetermined machine learning method. As the learning model 410, for example, a neural network model composed of a plurality of layers is used. Then, by a predetermined machine learning method (such as deep learning), the weighting coefficients and the like (learning parameters) between the plurality of layers (input layer, (one or more) intermediate layers, output layer) in the neural network model are adjusted.

[0023] The processing in this learning stage in such machine learning is, for example, executed in advance by the image processing device 30. Specifically, a learning model 410 (learned model 420) for executing an inference process for specifying the lesion site of a slice image is generated based on teacher data having a lesion site. A plurality of slice image groups having known lesion sites (more specifically, a plurality of slice image groups for a plurality of subjects) and the like are used as the teacher data.

[0024] However, without being limited thereto, the image processing apparatus 30 may build a learned model 420 in the apparatus itself (image processing apparatus 30) by acquiring learning parameters adjusted by another apparatus.

[0025] In addition, the image processing apparatus 30 executes further processing based on the inference result using the learned model 420 by machine learning. Specifically, as will be described later, the image processing apparatus 30 executes a process of determining (judging) a lesion region based on a plurality of estimated lesion regions 71 obtained as the inference result. More specifically, a process of correcting a region to be determined as a lesion region is executed based on the proximity and continuity (described later) of the lesion regions.

[0026] As shown in FIG. 1, the image processing apparatus 30 includes a controller (also referred to as a control unit) 31, a storage unit 32, and an operation unit 35.

[0027] The controller 31 is a control device built in the image processing apparatus 30 and controls the operation of the image processing apparatus 30.

[0028] The controller 31 is configured as a computer system including one or more hardware processors (e.g., a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit)). The controller 31 realizes various processes by executing a predetermined software program (hereinafter, also simply referred to as a program) stored in a storage unit (non-volatile storage unit such as a ROM and / or a hard disk) 32 in a CPU or the like. The program (specifically, a program module group) may be recorded on a portable recording medium such as a USB memory, read from the recording medium, and installed in the image processing apparatus 30. Alternatively, the program may be downloaded via a communication network or the like and installed in the image processing apparatus 30.

[0029] The controller 31 executes, for example, processing related to the inference stage in machine learning. Specifically, inference processing is executed for each of a plurality of two-dimensional slice images 220 (240) obtained for a subject using a learned model 420 (trained model 420) (see FIG. 2) with adjusted learning parameters. Further, the controller 31 also executes a determination process (judgment process) of a lesion area based on the estimated lesion area (lesion candidate area) 71. Furthermore, the controller 31 also executes a process of outputting a lesion site (such as a display process of an image including a fracture site).

[0030] The storage unit 32 is composed of a storage device such as a hard disk drive (HDD) and / or a solid state drive (SSD). The storage unit 32 stores a plurality of two-dimensional slice images 220 and the learned model 420 (including learning parameters), etc.

[0031] The operation unit 35 includes an operation input unit 35a that receives operation inputs to the image processing apparatus 30, and a display unit 35b that performs display output of various information. For example, the operation input unit 35a (also referred to as a reception unit) receives a designation operation of a two-dimensional slice image group to be processed, etc. Further, the display unit 35b displays a slice image 220 (also referred to as 250) etc. on which inference processing using the learned model 420 and various subsequent processes have been performed. As the operation input unit 35a, a mouse, a keyboard, etc. are used, and as the display unit 35b, a display (such as a liquid crystal display) is used. Also, a touch panel that functions as a part of the operation input unit 35a and also functions as a part of the display unit 35b may be provided.

[0032] Note that this image processing apparatus 30 is also referred to as a medical image processing apparatus, and the image processing system 10 is also referred to as a medical image processing system.

[0033] <1-2. Outline of Processing> FIG. 3 and FIG. 4 are flowcharts showing the processing of the image processing apparatus 30 (specifically, the controller 31 (FIG. 1)). In FIG. 3, processing including inference processing (step S11) using the learned model 420 by machine learning is shown.

[0034] First, in step S11, the controller 31 executes a lesion region estimation process (a process of estimating a lesion region) for the target slice image group. Specifically, the controller 31 executes a lesion region estimation process using the learned model 420 (a learning device) for each of the plurality of slice images constituting the slice image group. Specifically, by using the learned model 420 (a learning device), a region 71 (estimated lesion region) estimated as a lesion region is obtained in two or more of the plurality of slice images (see FIG. 5).

[0035] Next, in step S12 (described later) (see also FIG. 4, etc.), the controller 31 executes a determination process (confirmation process, etc.) of the lesion region based on the estimated lesion region 71. Specifically, a process of determining whether the estimated lesion region 71 is a lesion region or not is executed. Step S12 will be described in detail later.

[0036] Then, in step S14, as will be described later, the controller 31 displays the lesion region on the display unit 35b based on the processing result in step S12 (see FIG. 6).

[0037] <1-3. Outline of the determination process of the lesion region> <Lesion determination process using proximity> Next, the outline of the determination process (step S12) of the lesion region will be described. The determination process of the lesion region is a process of determining the lesion region based on the estimated lesion region 71 (a candidate region of the lesion region).

[0038] FIG. 5 is a diagram showing an example of a region 71 (also referred to as an estimated lesion region) estimated as a lesion region by the lesion region estimation process in step S11. In FIG. 5, in a plurality of slice images 220, each estimated lesion region 71 is shown surrounded by a bounding box 51 (a thick black frame in a rectangular shape). That is, the estimation result regarding the fracture site by the learned model 420 (learning device) is shown.

[0039] In FIG. 5, the estimated lesion regions 71 detected in some (five) of the plurality of slice images 220, i.e., slice images L(j - 1) to L(j + 3), are shown.

[0040] In FIG. 5, the estimated lesion regions 71 are detected at positions slightly to the right of the center of each of the four slice images L(j - 1), Lj, L(j + 2), and L(j + 3). Specifically, in the slice image L(j - 1) (see the lowermost row), the estimated lesion region 71a is detected, and in the slice image Lj (see the second row from the bottom), the estimated lesion region 71b is detected. Also, in the slice image L(j + 2), the estimated lesion region 71d is detected, and in the slice image L(j + 3), the estimated lesion region 71e is detected. Further, as shown in FIG. 7, in the slice image L(j + 6), the estimated lesion region 71h is detected. However, the estimated lesion region 71 is not detected at positions slightly to the right of the center in the slice images L(j + 1), L(j + 4), L(j + 5), and L(j + 7). Note that FIG. 7 is a diagram schematically showing the estimated lesion regions 71 and the like detected in a plurality of slice images L(j - 1) to L(j + 7). In the upper row of FIG. 7, the estimated lesion region 71 immediately after detection in step S11 is shown, and in the lower row of FIG. 7, the determination processing result and the like in step S12 (described later) are shown. In FIG. 7, the estimated lesion regions 71 in each slice image L are shown abstracted as elongated rectangles. Also, in FIG. 7, the in-plane position (two-dimensional position) in FIG. 5 is abstracted and shown as the position in the left-right direction (one-dimensional position) in FIG. 7.

[0041] In addition, an estimated lesion area 71p is detected at a position slightly to the left of the center of the slice image L(j - 1) (see the bottom row of FIG. 5), and an estimated lesion area 71q is detected at a position near the left end of the slice image L(j + 1) (see the third row from the bottom of FIG. 5).

[0042] Furthermore, as shown in FIG. 8, in the slice image L(j - 5), an estimated lesion area 71r is detected at a position in a cross-sectional plane different from that of the estimated lesion areas 71a, 71p, etc. (two-dimensional position within the cross-section). Note that FIG. 8 is the same as FIG. 7 and shows the estimated lesion areas 71 in slice images 220 other than the slice image 220 shown in FIG. 7 (such as the slice image L(j - 5)). As shown in FIG. 8, no estimated lesion areas 71 are detected in the slice images L(j - 4), L(j - 3), and L(j - 2).

[0043] However, the accuracy of the estimation result (step S11) by the learned model 420 is not necessarily 100%.

[0044] In the example shown in FIG. 5 and the like, actually, fracture sites exist only at positions slightly to the right of the center of each slice image. Conversely, the estimated lesion area 71p at a position slightly to the left of the center of one slice image L(j - 1) (see the bottom row of FIG. 5) is not actually a fracture area. That is, over-detection (false detection) of fracture sites has occurred. Similarly, the estimated lesion area 71q at a position near the left end of one slice image L(j + 1) (see the third row from the bottom) is not actually a fracture area, and over-detection of fracture sites has occurred. Also, the estimated lesion area 71r of the slice image L(j - 5) (see FIG. 8) is not actually a fracture area, and over-detection of fracture sites has occurred. Note that in FIG. 5, the fracture sites existing in the left half of the subject (the fracture sites at positions slightly to the right of the center of each slice image) are mainly shown. In FIG. 5 and the like, which depict the slice image (an image with the spine arranged on the lower side (an image seen from the foot side rather than the head side)), the fracture sites are shown on the right side of the figure.

[0045] Here, generally, the fracture site (fracture location) has a certain size (a size larger than a certain slice pitch (for example, 0.6 mm)) and exists over a certain range. In FIG. 8, the state where the fracture site 61 exists over a certain range from the slice image L(j - 1) to the slice image L(j + 6) is shown. The fracture site 61 exists in some slice images (several cross-sectional images) that are adjacent in its normal direction among the plurality of slice images. Also, the fracture site 61 exists at the same in-slice position (the same or a nearby in-slice position) in these several adjacent slice images (cross-sections). Note that the in-slice position is a planar position (2D position) within the slice image (cross-sectional image).

[0046] Therefore, in this embodiment, such over-detection is suppressed by utilizing such characteristics (the property that the fracture site exists not only in a single slice image but also in a plurality of mutually adjacent slice images (and at relatively close in-slice positions), which is also simply referred to as proximity).

[0047] Specifically, first, if the estimated lesion region 71 is not detected in any of the plurality of slice images 220, it is determined that no lesion region (fracture location) exists. Further, even if the estimated lesion region 71 is detected only in a single slice image among the plurality of slice images 220, it is determined that no lesion region exists.

[0048] In addition, when an estimated lesion region 71 is detected in two or more slice images (including the first slice image and the second slice image) among a plurality of slice images 220, it is determined whether each estimated lesion region 71 is a lesion region (a region including a fracture site) as follows. Specifically, on the condition that the distance between the in-plane position of the estimated lesion region 71 (also referred to as the first region) in the first slice image and the in-plane position of the estimated lesion region 71 (also referred to as the second region) in the second slice image is within a predetermined range ΔP, the estimated lesion region 71 (the first region) in the first slice image is determined to be a lesion region. When the distance between the in-plane position of the first region and the in-plane position of the second region is within the predetermined range ΔP, it is determined that both the first region and the second region are fracture sites. As the in-plane position of each estimated lesion region (the first region and the second region) 71, for example, a representative position such as the central position (or the centroid position) of each estimated lesion region 71 may be used.

[0049] In this way, by utilizing the property (proximity) that the fracture site exists not only in a single slice image but also in a plurality of slice images adjacent to each other, over-detection of the fracture site can be suppressed. Specifically, even when two estimated lesion regions 71 (the first region and the second region) are detected in different slice images 220, when the difference between the in-plane position of the first region and the in-plane position of the second region is not within the predetermined range ΔP, the first region and the second region are not determined to be (basically) lesion regions. Therefore, it is possible to avoid or suppress false detection of the lesion region and detect the lesion region more accurately.

[0050] Furthermore, in this embodiment, on the condition that the second slice image exists within a predetermined separation distance D from the first slice image in the normal direction of the first slice image, the estimated lesion region 71 of the first slice image (and the estimated lesion region 71 of the second slice image) is determined to be a fracture site. The predetermined separation distance D is, for example, a distance (several millimeters to several tens of millimeters) corresponding to the thickness of several to several tens of slice images (tomographic images).

[0051] Specifically, if no other estimated lesion region 71 that satisfies a predetermined condition is detected in any other slice image in which the separation distance d in the normal direction from the first slice image is smaller than a predetermined value D, the first region is determined as a non-lesion region. The predetermined condition is that the difference regarding the in-plane position with respect to the estimated lesion region 71 of the first slice image is within a predetermined range ΔP (having an in-plane position proximity relationship).

[0052] In other words, several slice images near the first slice image are set as a search range. And if no estimated lesion region 71 in which the difference from the estimated lesion region 71a regarding the in-plane position is within the predetermined range ΔP is detected by the lesion region estimation process in any of the slice images in the search range, the estimated lesion region of the first slice image is determined as a non-lesion region.

[0053] For example, when a predetermined separation distance D is set to a distance corresponding to 3.1 slice images (3 or more and less than 4), regarding the slice image L(j - 1), six neighboring slice images L(j - 4), L(j - 3), L(j - 2), Lj, (j + 1), L(j + 2) in the cross-sectional normal direction are set as the search range (see FIGS. 7 and 8, etc.).

[0054] Then, the in-plane position Pa of the estimated lesion region (lesion candidate region) 71a detected in the slice image L(j - 1) is compared with the in-plane position Pb of the estimated lesion region (lesion candidate region) 71b detected in another slice image Lj (the slice image adjacent in the normal direction). Based on the fact that the difference between the in-plane position Pa of the estimated lesion region 71a and the in-plane position Pb of the estimated lesion region 71b is within the predetermined range ΔP (that is, both positions Pa and Pb exist within a predetermined vicinity range), these estimated lesion regions 71a and 71b are determined (decided) as lesion regions. As the in-plane positions (Pa, Pb, etc.) of each estimated lesion region 71, for example, representative positions such as the central position (or the centroid position) of each estimated lesion region 71 may be used as described above.

[0055] Whether the difference between the in-section position Pa and the in-section position Pb is within a predetermined range ΔP may be determined by whether the absolute value of the difference (discrepancy) between the in-section position Pa and the in-section position Pb is within a predetermined range (specifically, whether the in-section position Pb exists within a range of a predetermined radius ΔR from the in-section position Pa, etc.). Alternatively, based on whether the degree of overlap between the estimated lesion region 71a and the estimated lesion region 71b is equal to or greater than a predetermined level (for example, 50% or more), it may be determined whether the difference between the in-section position Pa and the in-section position Pb is within a predetermined range ΔP (see Fig. 9). The upper, middle, and lower parts of Fig. 9 show the states where the degree of overlap between the estimated lesion region 71a and the estimated lesion region 71b is 50% or more, approximately 50%, and less than 50%, respectively. For example, when having a positional relationship (in-section positional relationship) as shown in the upper or middle part of Fig. 9, it may be determined that the difference between the in-section position Pa of the estimated lesion region 71a and the in-section position Pb of the estimated lesion region 71b is within a predetermined range ΔP.

[0056] Similarly, based on the fact that the difference between the in-section position Pb of the estimated lesion region 71b and the in-section position Pd of the estimated lesion region 71d is within a predetermined range ΔP (that is, both positions Pb and Pd exist within a predetermined vicinity range), these estimated lesion regions 71b and 71d are determined as lesion regions. The same applies to the estimated lesion regions 71d and 71e, and the estimated lesion regions 71e and 71h.

[0057] Also, the in-plane position Pp of the estimated lesion region (lesion candidate region) 71p detected in the slice image L(j - 1) is compared with the in-plane position Pb of the estimated lesion region (lesion candidate region) 71b detected in another slice image Lj (a slice image adjacent in the normal direction). However, the difference between the in-plane position Pp of the estimated lesion region 71p and the in-plane position Pb of the estimated lesion region 71b exceeds a predetermined range ΔP (that is, both positions Pp and Pb do not exist within a predetermined vicinity range). Therefore, the estimated lesion region 71p is not confirmed as a lesion region. Further, a process of searching for other slice images 220 existing within a predetermined separation distance D in the normal direction from the slice image L(j - 1) is repeatedly executed. For example, the in-plane position Pp of the estimated lesion region 71p is also compared with the in-plane position Pd of the estimated lesion region 71d detected in another slice image L(j + 2). Since the difference between the in-plane position Pp of the estimated lesion region 71p and the in-plane position Pd of the estimated lesion region 71d also exceeds the predetermined range ΔP, the estimated lesion region 71p is not confirmed as a lesion region.

[0058] From the slice image L(j - 1), in any slice image within a predetermined separation distance D in the normal direction (for example, within a distance range corresponding to the thickness of three slice images (tomographic images)), there is no estimated lesion region 71 having an in-plane position with a difference within the predetermined range ΔP from the in-plane position Pp. In other words, an estimated lesion region 71 having a difference within the predetermined range from the estimated lesion region 71p regarding the in-plane position is not detected by the lesion region estimation process. In this case, the estimated lesion region 71p of the slice image L(j - 1) is determined as a non-lesion region 73(73p).

[0059] Similarly, the estimated lesion region 71q of the slice image L(j + 1) and the estimated lesion region 71r of the slice image L(j - 5) are also determined as non-lesion regions 73(73p).

[0060] According to such processing, the estimated lesion regions 71r, 71p, and 71q that are over-detected in the lesion region estimation process (step S11) are determined as non-lesion regions 73. That is, it is possible to eliminate the lesions over-detected by the learned model 420 (learning device).

[0061] Here, although six slice images 220 (three in the up-down (front-back) direction, for a total of six) are set as the search range, it is not limited to this. Specifically, two or fewer, or four or more (for example, about several to several tens) slice images 220 may be set as the search range.

[0062] <Lesion determination process using continuity> In addition, it is possible that the first region (estimated lesion region 71) of the first slice image and the second region (estimated lesion region 71) of the second slice image are determined (confirmed) as lesion regions, and there may be another third slice image between the first slice image and the second slice image. For example, when the estimated lesion region 71b of the slice image Lj and the estimated lesion region 71d of the slice image L(j + 2) are confirmed as lesion regions, there may be another slice image L(j + 1) between the slice image Lj and the slice image L(j + 2).

[0063] Considering the continuity of fractures (the property that fracture sites exist in a plurality of consecutive slice images), in such an intermediate third slice image (for example, slice image L(j + 1)), an estimated lesion region 71 (specifically, within the proximity range) should originally be detected. However, in reality, in the third slice image, the estimated lesion region 71 may not be detected by the lesion region estimation process (step S11). Specifically, in the third slice image, there are cases where the estimated lesion region 71 with a difference in cross-sectional position within a predetermined range ΔP from at least one of the first region (for example, 71b) of the first slice image and the second region (for example, 71d) of the second slice image is not detected by the lesion region estimation process (step S11). In short, there may be a detection omission of the estimated lesion region 71 in the third slice image.

[0064] In this embodiment, even in such a case, it is determined that a lesion region exists in the third slice image. For example, a third region 72c that is complemented within the third slice image L(j + 1) based on the first region 71b and the second region 71d is determined (assumed determination) as the lesion region in the third slice image. The position of the third region 72c may be calculated by being complemented (for example, averaged (weighted average, etc.)) based on the position of the first region 71b and the position of the second region 71d. The size of the third region 72c may be calculated by being complemented (for example, averaged) based on the size of the first region 71b and the size of the second region 71d. Alternatively, the size of the third region 72c may be obtained as the larger (or smaller) size of the first region 71b and the second region 71d, or may be a fixed size. Although the third region 72c is not a region detected as a presumed lesion region by the lesion region estimation process (step S11), it is a region that has been additionally determined (assumed determination) (step S32) as a presumed lesion region (also referred to as an additional determination region or a complemented lesion region 72).

[0065] In this way, by utilizing the property that a fracture site exists in a plurality of consecutive slice images (also simply referred to as continuity), it is possible to particularly well suppress or avoid overlooking the fracture site.

[0066] <1-4. Details of the determination process of the lesion region> Next, with reference to FIG. 3, the details of the process of step S12 will be described.

[0067] First, in step S21, the controller 31 sets a certain presumed lesion region 71 among the plurality of slice images 220 as the first determination target region. In other words, a presumed lesion region 71 that has not yet been set as the first determination target region is set as the first determination target region. Then, the controller 31 determines whether there is another presumed lesion region 71 (also referred to as the second determination target region) within the slice images 220 at a predetermined separation distance D in the normal direction from the slice image 220 to which the first determination target region belongs.

[0068] When it is determined that the other estimated lesion region 71 (second determination target region) does not exist within the separation distance D from the first determination target region, the first determination target region is determined as a non-lesion region 73 (step S26). For example, when there is no other estimated lesion region 71 (second determination target region) within a predetermined separation distance D in the normal direction from the estimated lesion region 71r (first determination target region), the estimated lesion region 71r is determined to be a non-lesion region (not including a fracture site (fracture region)) (see FIG. 8). That is, it is possible to eliminate the over-detected estimated lesion region 71 (noise suppression). Thereafter, the process proceeds to step S39. Although not explicitly shown in the flowchart of FIG. 5, when no estimated lesion region 71 is detected at all, the process of FIG. 4 immediately ends. In this case, at step S14, it is displayed that no estimated lesion region 71 is detected at all (no lesion exists).

[0069] On the other hand, when it is determined that the first determination target region exists and the other estimated lesion region 71 (second determination target region) exists within the predetermined separation distance D, the process proceeds to step S23.

[0070] At step S23, it is determined whether the difference between the in-plane positions of the first determination target region and the second determination target region is within a predetermined range ΔP (whether the in-plane positions of the first determination target region and the second determination target region exist within a predetermined vicinity range).

[0071] When it is determined at step S23 that the difference between the in-plane positions of the two regions is not within the predetermined range ΔP, the process proceeds to step S27. For example, when the estimated lesion region 71p is the first determination target region and the estimated lesion region 71b is the second determination target region, it is determined that the difference between the in-plane positions of the two regions 71p and 71b is not within the predetermined range ΔP, and the process proceeds to step S27. At step S27, it is determined whether the search within the separation distance D has been completed.

[0072] If the search is not completed, the process returns to step S21.

[0073] When the exploration is completed, the process proceeds to step S28. In step S28, the first determination target region is determined as a non-lesion region. In other words, among a plurality of slice images, if there is no estimated lesion region 71 in any other slice image where the separation distance d in the normal direction from the first slice image is smaller than the predetermined value D and the difference from the first region regarding the in-plane position is within the predetermined range ΔP, the first region is determined as a non-lesion region. For example, among a plurality of slice images, if there is no estimated lesion region 71 in any other slice images L(j - 1) to L(j + 2) where the separation distance d in the normal direction from the slice image L(j - 1) is smaller than the predetermined value D and the difference from the estimated lesion region 71p regarding the in-plane position is within the predetermined range ΔP, the estimated lesion region 71p is determined as a non-lesion region 73 (73p). Similarly, the estimated lesion regions 71q and 71r are also determined as non-lesion regions 73 (73q and 73r). In this way, it is possible to eliminate the over-detected estimated lesion region 71 (noise suppression). After that, the process proceeds to step S39.

[0074] If it is determined in step S23 that the difference between the in-plane positions of the two regions is within the predetermined range ΔP, the process proceeds to step S25. In step S25, the first determination target region is determined (confirmed) as a lesion region (step S25). Also, the second determination target region is determined (confirmed) as a lesion region. For example, as described above, when the first determination target region is the estimated lesion region 71a and the second determination target region is the estimated lesion region 71b, both regions 71a and 71b are determined as lesion regions. Also, when the first determination target region is the estimated lesion region 71b and the second determination target region is the estimated lesion region 71d, both regions 71b and 71d are determined as lesion regions.

[0075] In step S31 following step S25, it is determined whether there is an intermediate slice image between the first slice image and the second slice image. For example, when the first determination target region is the estimated lesion region 71b and the second determination target region is the estimated lesion region 71d, it is determined that there is an intermediate slice image L(j + 1) between the slice image Lj and the slice image L(j + 2).

[0076] If there is no intermediate slice image, proceed to step S33.

[0077] If there is an intermediate slice image, it is determined that a detection omission has occurred in the estimated lesion region 71 in the intermediate slice image (for example, slice image L(j + 1)), and the process proceeds to step S32. In step S32, the lesion region is complemented in the intermediate slice image.

[0078] For example, based on the first determination target region 71b of the slice image Lj and the second determination target region 71d of the slice image L(j + 2), the estimated lesion region 71 (also referred to as the complemented lesion region 72c) is complemented within the intermediate slice image L(j + 1) (see the lower part of FIG. 7). In other words, the complemented lesion region 72c is generated. More specifically, based on the position and size of the bounding box 51 surrounding the first determination target region 71b and the position and size of the bounding box 51 surrounding the second determination target region 71d, the position and size of the bounding box 51c in the estimated lesion region 71c are determined. Then, the region within the bounding box 51c in the estimated lesion region 71c is determined as the complemented lesion region 72c.

[0079] Similarly, based on the first determination target region 71e of the slice image L(j + 3) and the second determination target region 71h of the slice image L(j + 6), the estimated lesion region 71 (also referred to as the complemented lesion region 72f) is complemented within the intermediate slice image L(j + 4) (see the lower part of FIG. 7). Further, based on both regions 71e and 71h, the estimated lesion region 71 (also referred to as the complemented lesion region 72g) is complemented within the intermediate slice image L(j + 5) (see the lower part of FIG. 7).

[0080] Thereafter, in step S33, a second determination target area is set as a new first determination target area, and in step S34, it is determined whether there is a new second determination target area that satisfies a predetermined condition C2. The condition C2 is that there is an undetermined estimated lesion area 71 (new second determination target area) within a predetermined separation distance D from the new first determination target area and the in-plane positions of both areas (the new first determination target area and the new second determination target area) are within a predetermined vicinity range.

[0081] When there is a new second determination target area that satisfies condition C2, the process returns from step S34 to step S25. For example, when there is an undetermined estimated lesion area 71d (new second determination target area) within a predetermined separation distance D from the new first determination target area 71b and the in-plane positions of both areas 71b, 71d are within a predetermined vicinity range, the process returns to step S25. Then, both areas 71b, 71d are determined as lesion areas.

[0082] On the other hand, when there is no new second determination target area that satisfies condition C2, the process proceeds from step S34 to step S39. For example, when there is no undetermined estimated lesion area 71 (new second determination target area) within a predetermined separation distance D from the new first determination target area 71h, the process proceeds to step S39.

[0083] Thereafter, in step S39, it is determined whether the above-described processing has been completed for all the estimated lesion areas 71.

[0084] If there remains an unprocessed estimated lesion area 71 (an estimated lesion area 71 that has not yet been set as the first determination target area), the process returns to step S21.

[0085] On the other hand, if there remains no unprocessed estimated lesion area 71, the process of FIG. 4 (step S12) ends. Then, the process proceeds to step S14 (FIG. 3).

[0086] According to the above processing (especially the loop of steps S25, S31 to S34), the estimated lesion regions 71a, 71b, 71d, 71e, 71h and the complementary lesion regions 72c, 72f, 72g are detected as regions whose in-section positions are close in two or more different slice images (see FIG. 7). These regions 71a, 71b, 72c, 71d, 71e, 72f, 72g, 71h are an aggregate (accumulation) of regions determined to be actually fractured, forming one lesion space (fracture space). In particular, by complementing the complementary lesion regions 72c, 72f, 72g, detection omission of the lesion regions is avoided or suppressed.

[0087] Conversely, the estimated lesion regions 71p, 71q, 71r are determined not to be actual fracture regions (lesion regions). That is, over-detection is suppressed.

[0088] <1-4. Display processing of lesion region (step S14)> Next, the processing of step S14 will be described.

[0089] FIG. 6 is a diagram showing a display example (display example of each slice image 220) in step S14.

[0090] As shown in FIG. 6, a predetermined number (five in FIG. 6) of consecutive slice images 220 are arranged and displayed along a predetermined direction (here, the vertical direction) (simultaneously). Also, according to a screen scroll operation using a mouse or the like, images not visible at the current time (slice images 220 after the sixth one, etc.) can also be displayed. That is, the plurality of slice images 220 can be sequentially switched and displayed by a predetermined number (sequentially). Here, a predetermined number of slice images 220 are arranged and displayed within the screen (simultaneously), but it is not limited to this. For example, the plurality of slice images 220 may be sequentially switched and displayed one by one (sequentially).

[0091] As shown in FIG. 6, in each slice image 220, the three types of regions classified (formed) as described above are displayed in different modes (different display colors, etc.).

[0092] For example, the estimated lesion regions 71a, 71b, 71d, 73e, etc. determined in step S25 are surrounded by a red bounding box 51 and displayed.

[0093] In addition, the complementary lesion regions (also referred to as recognized lesion regions recognized (regarded as determination) as lesion regions) 72 (72c, etc.) complemented in step S32 are surrounded by a green bounding box 51 and displayed.

[0094] Furthermore, the estimated lesion regions 71p, 71q changed to non-lesion regions in steps S26 and S28 are surrounded by a purple bounding box 51 (dashed line in FIG. 6) and displayed.

[0095] In this way, the estimated lesion regions 71a, etc. determined (judged) as lesion regions (fracture regions) and the recognized lesion regions (complementary lesion regions) 72c, etc. recognized (regarded as determination) as lesion regions are displayed in different manners from each other. Therefore, the user can easily distinguish between the estimated lesion region 71 and the recognized lesion region 72.

[0096] In addition, since the lesion regions (71a and / or 72c, etc.) and the non-lesion regions (71p, etc.) are displayed in different manners from each other, the user can easily distinguish between the lesion region and the non-lesion region.

[0097] Here, although these three types of regions are displayed in different manners from each other, it is not limited thereto. For example, the estimated lesion regions 71a, etc. determined in step S25 and the complementary lesion regions 72c, etc. complemented in step S32 may be surrounded by a bounding box 51 of the same color (for example, red) and displayed. Also, the bounding box surrounding the estimated lesion regions 71p, etc. changed to non-lesion regions in steps S26 and S28 may not be displayed (may be made non-displayed).

[0098] <2. Second Embodiment> The second embodiment is a modification of the first embodiment. Hereinafter, the description will focus on the differences from the first embodiment.

[0099] In the first embodiment described above, the processes of steps S11 and S12 (see FIGS. 3 and 4) are executed only for a single group of slice images with respect to a single reference axis direction. In particular, for a single group of slice images, a determination process (FIG. 4) is executed to determine that a plurality of estimated lesion regions detected in different slice images among them are lesion regions on the condition that the difference in the in-plane positions between the plurality of estimated lesion regions falls within a predetermined range ΔP.

[0100] In this second embodiment, the same determination process (FIG. 4) is executed not only for one group of slice images (the first group of slice images) but also for a group of slice images different from the one group of slice images (the second group of slice images). The same determination process includes a process (steps S21, S23, S25, etc.) of determining that a plurality of estimated lesion regions detected in different slice images among them are lesion regions on the condition that the difference in the in-plane positions between the plurality of estimated lesion regions falls within a predetermined range ΔP with respect to the second group of slice images. The same determination process also includes a process (steps S31, S32, etc.) of determining a complementary lesion region complemented in an intermediate slice image as a lesion region in the intermediate slice image.

[0101] In the second embodiment, such an aspect will be described.

[0102] The plurality of groups of slice images 210 are obtained for each of a plurality of different reference axes (the normal directions of the cross-sections) with respect to the same subject. In other words, the groups of slice images 210 (a plurality of two-dimensional slice images) are further photographed after changing the direction of the cross-section. As a result, a plurality of groups of slice images 210 are imaged (generated).

[0103] Each group of slice images 210 has a plurality of slice images obtained by slicing a subject in a cross-section having a unique normal direction for each group of images. In other words, each group of slice images 210 has a plurality of slice images obtained by slicing a subject in a cross-section having a normal line in a direction different from the normal direction with respect to other groups of slice images. Note that each group of slice images 210 is also referred to as a (the i-th) group of slice images Gi (where i = 1, ..., N; N is the number of groups of slice images and is a natural number of 2 or more).

[0104] Each group of slice images 210 can be obtained by slicing and imaging a subject in a cross-section in a direction different from the cross-sections of other groups of slice images (for example, each cross-section such as an axial cross-section, a coronal cross-section, a sagittal cross-section, etc.) (see FIG. 12).

[0105] FIG. 12 is a diagram showing various types of cross-sections of an axial cross-section, a coronal cross-section, and a sagittal cross-section. An axial cross-section is a cross-section orthogonal to the body axis (also referred to as an axial section or a transverse section). A coronal cross-section is a cross-section that cuts across (a cross-section that divides the body, etc., into a front side and a rear side) (also referred to as a coronal section or a coronal plane). A sagittal cross-section is a longitudinal cross-section (a cross-section that divides the body, etc., into a left side and a right side) (also referred to as a sagittal section or a sagittal plane). Note that as types of cross-sections, there are also oblique cross-sections (tilted cross-sections) (oblique sections), etc.

[0106] For example, one group of slice images 210 is composed of a plurality of two-dimensional slice images 220 sliced in an axial cross-section (a plurality of two-dimensional slice images 220 having the normal direction of the axial cross-section as the reference axis direction) (see FIG. 5, etc.).

[0107] Also, another group of slice images 210 is composed of a plurality of two-dimensional slice images 220 sliced in a sagittal cross-section (a plurality of two-dimensional slice images 220 having the normal direction of the sagittal cross-section as the reference axis direction). Alternatively, the another group of slice images 210 may be composed of a plurality of two-dimensional slice images 220 sliced in a coronal cross-section (a plurality of two-dimensional slice images 220 having the normal direction of the coronal cross-section as the reference axis direction).

[0108] FIG. 11 is a flowchart showing the processing of the image processing apparatus 30 (specifically, the controller 31) according to the second embodiment.

[0109] As shown in FIG. 11, in the second embodiment as well, the same processing as in the first embodiment (see FIG. 3) is executed. Specifically, first, for one group of slice images, the processing of steps S11 and S12 is executed. Then, for the other groups of slices as well, the processing of steps S11 and S12 (see also FIG. 4) is executed respectively.

[0110] In this way, for each of a plurality (for example, two) of groups of slice images, the processing of steps S11 and S12 is executed.

[0111] And when it is determined in step S13 that the processing for a predetermined number (for example, two) of two-dimensional slice image groups 210 is completed, the process proceeds to step S14.

[0112] In step S14, for each two-dimensional slice image group 210, the same display processing as in step S14 of the first embodiment is executed.

[0113] According to this, it is possible to obtain the same effect as in the first embodiment. In short, based on the property (proximity) that a fracture site exists not only in a single slice image but also in a plurality of slice images close to each other, over-detection of the fracture site can be suppressed. Also, based on the property (continuity) that a fracture site exists in a plurality of consecutive slice images, overlooking of the fracture site can be suppressed.

[0114] Moreover, in each of a plurality of groups of slice images in different directions from each other, an estimated lesion region 71 or the like within the slice image is detected, so it is possible to suppress overlooking of the fracture site depending on the direction of the fracture site (see FIG. 13).

[0115] FIG. 13 is a diagram schematically showing a situation where two fracture sites 61 and 62 exist. In FIG. 13, each slice image 220 (220a) sliced in the axial cross-section is shown as an elongated rectangular region that is long in the horizontal direction. That is, each slice image 220a in FIG. 13 is schematically shown as having a minute thickness (about 0.6 mm to 1 mm).

[0116] The right fracture site 61 extends non-parallel (substantially perpendicular direction) to the slice cross-section and exists across a plurality of slice images 220 (220a) sliced in the axial cross-section. Therefore, by the above-described processing, the fracture location can be well detected in a plurality of slice images regarding the axial cross-section.

[0117] On the other hand, the left fracture site 62 extends in a direction parallel to the slice cross-section and exists only in a single slice image 220 sliced in the axial cross-section. In such a case, even if the above-described processing (S11, S12) is performed on a single slice image group 210a (a plurality of slice images 220a) regarding the axial cross-section as in the first embodiment, there is a possibility that the fracture site 62 may be overlooked. For example, even if the fracture site 61 is detected as the estimated lesion region 71 in step S11, there is a possibility that the estimated lesion region 71 may be determined as the non-lesion region 73 in step S26.

[0118] On the contrary, if, as in the second embodiment, the above-described processing (S11, S12) is also performed on another slice image group 210b (a plurality of slice images 220b) regarding a cross-section in another direction (for example, the sagittal cross-section), it is possible to avoid or suppress overlooking the fracture site 62.

[0119] Thus, according to the second embodiment, it is possible to suppress or avoid overlooking (such as fractures extending in a direction parallel to the slice image cross-section and / or fractures existing in the gaps between adjacent slice images) that may occur when determining the lesion area using only a single slice image group. In other words, it is possible to suppress or avoid overlooking the lesion area more than when determining the lesion area using only a single slice image group (in a single direction). To put it simply, it is possible to suppress or avoid overlooking that may occur in exchange for noise suppression (excluding the over-detected estimated lesion area 71).

[0120] Here, although the mode in which the processes of steps S11 and S12 are executed for each of the two slice image groups has been exemplified, it is not limited to this. For example, the processes of steps S11 and S12 may be executed for each of three or more slice image groups regarding different reference axes. Also, in step S14, for each of the three or more slice image groups, similar to the first embodiment and the like, the fracture site or the like may be displayed. According to such a mode, since the lesion area is determined using three slice image groups in different directions, it is possible to suppress or avoid overlooking the lesion area more than when determining the lesion area using only two slice image groups.

[0121] Also, each of the plurality of slice image groups 210 is not limited to being composed of a plurality of slice images 220 sliced in an axial cross-section, a coronal cross-section, or a sagittal cross-section. Each slice image group 210 may be composed of, for example, a plurality of two-dimensional slice images 220 sliced in an oblique cross-section (a cross-section inclined at an arbitrary angle). Also, each slice image group 210 may be an image group having a normal line orthogonal to the normal line of another image group, or may be an image group having a normal line inclined at an arbitrary angle with respect to the normal line of another image group.

[0122] Further, the plurality of slice image groups 210 may be obtained by imaging slice image groups in respective directions with respect to mutually different reference axes as described above, but are not limited thereto. For example, while actually imaging one slice image group in one direction, image conversion processing may be performed on the one slice image group to generate slice image groups in other directions (for example, two slice image groups each having other two directions as the reference axis directions), whereby the plurality of slice image groups may be obtained.

[0123] <3. Modifications, etc.> The embodiments of the present invention have been described above, but the present invention is not limited to the above-described content.

[0124] For example, in each of the above embodiments, etc., the presence or absence of the estimated lesion region 71 within the vicinity range with respect to the estimated lesion region 71 in a certain slice image is determined by searching different slice images within a predetermined separation distance D in the normal direction of the certain slice image. Specifically, among the plurality of slice images constituting the slice image group, a slice image in which the separation distance d in the normal direction from the first slice image (the slice image to which the first region belongs) is smaller than the predetermined value D is set as the search range (search target). Then, within the search range, if there is no estimated lesion region 71 detected by the lesion region estimation process (step S11) and having a difference from the first region regarding the in-plane position within a predetermined range, the first region is determined to be a non-lesion region. Briefly speaking, it is premised that the two estimated lesion regions 71 to be determined are detected in different slice images within a predetermined separation distance D in the reference axis direction of the slice image group.

[0125] However, the present invention is not limited thereto, and the condition regarding the predetermined separation distance D may not be considered.

[0126] For example, the two estimated lesion regions 71 to be determined are not limited to the two estimated lesion regions 71 within a predetermined separation distance D, and may be two estimated lesion regions 71 detected in any two of all the slice images constituting the slice image group. That is, all of the plurality of slice images 220 constituting the slice image group may be set as the search range. Then, on the condition that the difference in the cross-sectional positions of the two estimated lesion regions 71 is within a predetermined range ΔP, the two estimated lesion regions 71 may be determined as lesion regions.

[0127] In other words, if, in any of the slice images other than the first slice image among all the slice images, an estimated lesion region 71 within the vicinity range of the first region (the "estimated lesion region 71 whose difference from the first region regarding the in-plane position is within the predetermined range ΔP") is not detected by the lesion region estimation process, the first region may be determined as a non-lesion region.

[0128] Also, in the above-described embodiment and the like, on the condition that the difference in the cross-sectional positions of the two estimated lesion regions 71 detected in different slice images within a predetermined separation distance D in the reference axis direction of the slice image group is within the predetermined range ΔP, the two estimated lesion regions 71 are determined as lesion regions. This predetermined range ΔP may be a fixed value, but is not limited thereto.

[0129] For example, the predetermined range ΔP may be a value determined according to the separation distance d from a certain slice image in the reference axis direction (a value that increases according to the separation distance d (such as B×d), etc.). Specifically, as shown in FIG. 10, the estimated lesion region 71b may be determined as a lesion region on the condition that the difference from the in-plane position of the estimated lesion region 71a is within "B×Δd". Also, the estimated lesion region 71d may be determined as a lesion region on the condition that the difference from the in-plane position of the estimated lesion region 71a is within "B×3×Δd". Here, the value Δd is the slice pitch, and the value B is a constant (proportional constant). Note that FIG. 10 is a schematic diagram showing a detection example of the estimated lesion region 71 according to a modified example.

[0130] In addition, in each of the above embodiments, a fracture region is exemplified as the lesion region, but the present invention is not limited thereto, and the lesion region may be a region of a lesion related to blood vessels (such as an aneurysm). Alternatively, the lesion region may be a region of a lesion such as cancer.

Explanation of Signs

[0131] 10 Image processing system 20 Slice image generation device 30 Image processing device 51 Bounding box 61, 62 Fracture site 71a~71h, 71p~71r Estimated lesion region 72 Complemented lesion region (certified lesion region) 73 Non-lesion region 210 Slice image group 220 Slice image ΔP Predetermined range d Separation distance D Predetermined separation distance Gi Slice image group Lj Slice image

Claims

1. A control unit that processes a group of slice images, comprising: The group of slice images is a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction, and is composed of a plurality of slice images obtained by slicing the subject at a plurality of different positions in the normal direction, The control unit: Performs lesion region estimation processing using a learning device on each of the plurality of slice images, When, by the lesion region estimation processing, a first region is detected as an estimated lesion region in a first slice image that is a cross-sectional image at a first position in the normal direction of the subject among the plurality of slice images, Among the plurality of slice images, when an estimated lesion region having an in-plane position within a predetermined range different from the in-plane position of the first region is not detected by the lesion region estimation processing in any slice image having a position different from the first position in the normal direction and within a predetermined distance from the first position in the normal direction, the first region is determined as a non-lesion region, Among the plurality of slice images, in a second slice image different from the first slice image, which is a cross-sectional image at a second position that is different from the first position in the normal direction and within the predetermined distance from the first position in the normal direction, when a second region, which is an estimated lesion region having an in-plane position within a predetermined range different from the in-plane position of the first region, is detected by the lesion region estimation processing, the first region and the second region are determined as lesion regions, The image processing apparatus, wherein the predetermined distance is a distance equal to or greater than a distance corresponding to the thickness of two slice images.

2. A control unit that processes a group of slice images, comprising: The group of slice images has a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction, The control unit: Performs lesion region estimation processing using a learning device on each of the plurality of slice images, On the condition that the difference between the in-plane position of a first region, which is a putative lesion region detected by the lesion region estimation process in a first slice image among the plurality of slice images, and the in-plane position of a second region, which is a putative lesion region detected by the lesion region estimation process in a second slice image among the plurality of slice images, is within a predetermined range, the first region and the second region are determined as lesion regions. When the first region and the second region are determined as lesion regions and there is a third slice image between the first slice image and the second slice image among the plurality of slice images, and a putative lesion region in the third slice image, the difference in in-plane position of which from at least one of the first region and the second region is within the predetermined range, is not detected by the lesion region estimation process, a third region complemented in the third slice image based on the first region and the second region is determined as the lesion region in the third slice image. An image processing apparatus characterized by the above.

3. The control unit Regarding a first slice image group that is the slice image group, executes a determination process of determining a plurality of putative lesion regions as lesion regions on the condition that the differences in in-plane position among the plurality of putative lesion regions detected in different slice images thereof are within a predetermined range. Regarding a second slice image group that is different from the first slice image group and is composed of a plurality of slice images obtained by slicing the subject at a plurality of different positions in a second direction and having a normal line in a second direction different from the normal line direction of the first slice image group, the determination process is also executed. The image processing apparatus according to claim 1 or claim 2, characterized by the above.

4. The image processing apparatus according to any one of claims 1 to 3, characterized in that the lesion region is a fracture region.

5. An image processing method, comprising: a) Executing a lesion region estimation process using a learning device for each of a plurality of slice images obtained by slicing a subject with a cross-section having the same normal line direction and at a plurality of different positions in the normal line direction. b) determining whether a first region detected as a presumed lesion region in a first slice image, which is a cross-sectional image at a first position in the normal direction of the subject in the plurality of slice images by the lesion region estimation process, is a lesion region; comprising; the step b) includes: b-1) when, in any of the plurality of slice images having a position different from the first position in the normal direction and within a predetermined distance from the first position in the normal direction, a presumed lesion region having an in-plane position within a predetermined range different from the in-plane position of the first region is not detected by the lesion region estimation process, determining the first region as a non-lesion region; b-2) when, in a second slice image different from the first slice image among the plurality of slice images, which is a cross-sectional image at a second position different from the first position in the normal direction and within the predetermined distance from the first position in the normal direction, a second region, which is a presumed lesion region having an in-plane position within a predetermined range different from the in-plane position of the first region, is detected by the lesion region estimation process, determining the first region and the second region as lesion regions; comprising; wherein the predetermined distance is a distance equal to or greater than the distance corresponding to the thickness of two slice images. An image processing method characterized by this. **Claim 6** An image processing method, comprising: a) performing a lesion region estimation process using a learning device for each of a plurality of slice images obtained by slicing a subject with cross-sections having the same normal direction; b) determining the first region and the second region as lesion regions on the condition that the difference between the in-plane position of the first region, which is a presumed lesion region detected by the lesion region estimation process in the first slice image among the plurality of slice images, and the in-plane position of the second region, which is a presumed lesion region detected by the lesion region estimation process in the second slice image among the plurality of slice images, is within a predetermined range; c) When the first region and the second region are determined as lesion regions and a third slice image exists between the first slice image and the second slice image among the plurality of slice images, if a lesion region estimated in the third slice image, where the difference in the in-plane position between at least one of the first region and the second region is within the predetermined range, is not detected by the lesion region estimation process, determining, as a lesion region in the third slice image, a third region that is complemented in the third slice image based on the first region and the second region; An image processing method characterized by comprising the above. **Claim 7** A program for causing a computer to a) performing a lesion region estimation process using a learning device for each of a plurality of slice images obtained by slicing a subject in cross-sections having the same normal direction, where the subject is sliced at a plurality of different positions in the normal direction; b) determining whether a first region detected as an estimated lesion region in a first slice image, which is a cross-sectional image at a first position in the normal direction of the subject among the plurality of slice images, is a lesion region; wherein step b) includes b-1) when, among the plurality of slice images, in any slice image having a position within a predetermined distance from the first position in the normal direction and different from the first position in the normal direction, an estimated lesion region having an in-plane position difference within the predetermined range from the in-plane position of the first region is not detected by the lesion region estimation process, determining the first region as a non-lesion region; b-2) when, among the plurality of slice images, in a second slice image different from the first slice image, which is a cross-sectional image at a second position that is different from the first position in the normal direction and within the predetermined distance from the first position in the normal direction, a second region, which is an estimated lesion region having an in-plane position difference within the predetermined range from the in-plane position of the first region, is detected by the lesion region estimation process, determining the first region and the second region as lesion regions; and comprising A program characterized in that the predetermined distance is a distance equal to or greater than the distance corresponding to the thickness of two slice images.

8. On a computer, a) Executing a lesion region estimation process using a learning device for each of a plurality of slice images obtained by slicing a subject in a cross-section having the same normal direction; b) When the difference between the in-section position of a first region, which is an estimated lesion region detected by the lesion region estimation process in a first slice image among the plurality of slice images, and the in-section position of a second region, which is an estimated lesion region detected by the lesion region estimation process in a second slice image among the plurality of slice images, is within a predetermined range, determining the first region and the second region as lesion regions; c) When the first region and the second region are determined as lesion regions and there is a third slice image between the first slice image and the second slice image among the plurality of slice images, if an estimated lesion region in the third slice image, the difference in the in-section position of which from at least one of the first region and the second region is within the predetermined range, is not detected by the lesion region estimation process, determining a third region, which is complemented in the third slice image based on the first region and the second region, as the lesion region in the third slice image. A program for causing the above to be executed.

Images