Learning medical image data generation device, learning medical image data generation method, and recording medium

Abstract

The learning-use medical image data generation apparatus acquires a first medical image (S1), acquires a second medical image (S2) that is different from the first medical image and is obtained by imaging substantially the same part as the first medical image, generates lesion region information about a lesion portion in the second medical image (S5), and generates learning-use medical image data in which the first medical image and the lesion region information are associated (S6).

Description

TECHNICAL FIELD

[0001] The present application relates to a learning medical image data generation apparatus, a learning medical image data generation method, and a recording medium that generate learning medical image data. BACKGROUND

[0002] In recent years, a computer-aided diagnosis (CAD) that shows a candidate position of a lesion portion with respect to a medical image and displays discrimination information has been developed. For example, in Japanese Patent Application Laid-Open No. 2005-185560, a medical image processing system that causes an apparatus to learn a medical image as teacher data to detect or discriminate a lesion portion is proposed. In Japanese Patent Application Laid-Open No. 2005-185560, a medical image processing system capable of updating teacher data to improve the expertise of CAD is disclosed.

[0003] However, in generating teacher data, it is sometimes difficult to identify a lesion portion from a case. If it is a doctor who is an expert in the field of the case, the doctor can identify the lesion portion, but for a doctor who is not an expert in the field of the case, it is sometimes difficult to determine the presence or absence of a lesion portion and a lesion region thereof. For example, in an endoscope image of a general observation using white light, for a doctor who is not an expert in the field of the case, it is sometimes difficult to identify a lesion portion from a case.

[0004] Therefore, in a case where a medical image in which it is difficult to determine the presence or absence of such a lesion portion and a lesion region thereof is used as teacher data, an expert must actually observe the medical image and perform a job of specifying a lesion region of a lesion portion. Therefore, there is a problem that a large amount of time is taken for generation of teacher data in which image data of a lesion portion is provided to a medical image processing system to learn.

[0005] Therefore, an object of the present application is to provide a learning medical image data generation apparatus, a learning medical image data generation method, and a program that can simply generate teacher data. SUMMARY

[0006] Means for solving the problem

[0007] A learning medical image data generation apparatus of one embodiment of the present application includes a first image acquisition unit that acquires a first medical image, a second image acquisition unit that acquires a second medical image different from the first medical image, the second medical image being an image obtained by imaging substantially the same portion as the first medical image, a lesion region information generation unit that generates lesion region information about a lesion portion in the second medical image, and a data generation unit that generates learning medical image data in which the first medical image and the lesion region information are associated with each other.

[0008] In the learning medical image data generation method of one embodiment of the present application, a first medical image is acquired, a second medical image different from the first medical image is acquired, the second medical image is an image obtained by imaging substantially the same site as the first medical image, lesion region information about a lesion portion in the second medical image is generated, and learning medical image data in which the first medical image is associated with the lesion region information is generated.

[0009] The program of one embodiment of the present application causes a computer to perform processing of acquiring a first medical image, acquiring a second medical image different from the first medical image, the second medical image being an image obtained by imaging substantially the same site as the first medical image, generating lesion region information about a lesion portion in the second medical image, and generating learning medical image data in which the first medical image is associated with the lesion region information. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 is a block diagram showing the structure of a server of the present embodiment.

[0011] Figure 2 is a block diagram showing the structure of a server of the present embodiment.

[0012] Figure 3 is a flowchart showing an example of a processing flow of image data according to the present embodiment.

[0013] Figure 4 is a diagram for explaining a process of generating teacher data according to the present embodiment.

[0014] Figure 5 is a diagram showing the structure of a teacher data table according to the present embodiment. DETAILED DESCRIPTION

[0015] Hereinafter, the embodiments will be described with use of the drawings.

[0016] (System structure)

[0017] Figure 1 is a block diagram showing the structure of a medical system according to the present embodiment. The medical system 1 is an endoscope system including an endoscope device 2 and a computer-aided diagnosis device (hereinafter referred to as a CAD device) 3. The endoscope device 2 has an endoscope 4 having an insertion portion 4a, a video processor 6 connected to a general-purpose cable 5 extending from the endoscope 4, and a display device 7 connected to the video processor 6. The endoscope device 2 is connected to the CAD device 3 through a signal line 8.

[0018] When the physician inserts the insertion section 4a into the subject, an image of a site in the subject obtained through the observation window of the front end section of the insertion section 4a is displayed on the display device 7 as an endoscope image. The physician observes the endoscope image displayed on the display device 7 and performs finding and identification of a lesion. In addition, illumination light from the light source device 6d provided in the video processor 6, which illuminates the observation site, is emitted from the illumination window of the front end section of the insertion section 4a through a light guide that is inserted into the insertion section 4a.

[0019] The video processor 6 has a control section 6a, a storage device 6b, an operation panel 6c, and a light source device 6d. The physician or the like can operate the operation panel 6c and give various instructions to the endoscope device 2. The endoscope device 2 has a first observation mode and a second observation mode different from the first observation mode. The endoscope device 2 has two observation modes, a white light observation mode as a so-called normal light observation mode and a narrow-band light observation mode as a so-called special light observation mode, as observation modes. Switching of the observation modes can be performed by the physician or the like operating the operation panel 6c. Thus, the physician can select a desired one of the two observation modes to observe the inside of the subject. In addition, the physician can also switch the observation modes during observation.

[0020] The light source device 6d emits white light when the observation mode is the white light observation mode and emits a prescribed narrow-band light when the observation mode is the narrow-band light observation mode. The white light is wide-band light that includes the wavelengths of RGB. The narrow-band light is, for example, two narrow-band lights having center wavelengths of 415 nm and 540 nm.

[0021] Reflected light from the observation site is photoelectrically converted by an unillustrated imaging element to generate an endoscope image. The generated endoscope image is displayed on the display device 7, and the physician can store it as a still image or a moving image in the storage device 6b by pressing a release switch 4c provided in the operation section 4b of the endoscope 4. Thus, image data of the endoscope image obtained in each observation mode is recorded in the storage device 6b.

[0022] The control section 6a includes a central processing device (CPU), a ROM, and a RAM. The storage device 6b is a rewritable large-capacity nonvolatile memory such as a hard disk device. The control section 6a realizes various functions of the endoscope device 2 by reading out software programs for the various functions stored in the ROM and the storage device 6b and expanding and executing them in the RAM.

[0023] The endoscope apparatus 2 and the CAD apparatus 3 are capable of communicating through a signal line 8. Image data of an endoscope image from the endoscope apparatus 2 is input to the CAD apparatus 3 in real time. The CAD apparatus 3 detects a lesion portion in the input image data of the endoscope image, and outputs its detection result information to the endoscope apparatus 2. The endoscope apparatus 2 displays the detection result information together with the endoscope image on a display apparatus 7.

[0024] The CAD apparatus 3 has a lesion portion detection program LDP. Here, a lesion portion detection algorithm of the lesion portion detection program LDP utilizes a model obtained by machine learning using image data containing a lesion portion as teacher data. Therefore, the CAD apparatus 3 uses this lesion portion detection program LDP to perform detection of a lesion portion in an endoscope image received from the endoscope apparatus 2 in real time. When the CAD apparatus 3 finds a lesion portion in the endoscope image, it extracts region information of the found lesion portion, and transmits a detection message of the lesion portion or the like to the endoscope apparatus 2. The detection message from the CAD apparatus 3 is displayed on the display apparatus 7 of the endoscope apparatus 2. As a result, a doctor is able to recognize the lesion portion detection result from the CAD apparatus 3 while observing the endoscope image displayed on the display apparatus 7.

[0025] The endoscope apparatus 2 is connected to a server 11 through a network 12. The network 12 can be the Internet, or can be a LAN. The server 11 has a processor 13 and a storage apparatus 14. As described later, the server 11 has a teacher data generation program TDC that generates teacher data from image data of an endoscope image obtained in the endoscope apparatus 2.

[0026] Figure 2 is a block diagram showing the structure of the server 11. The server 11 has a processor 13, a storage apparatus 14, a communication circuit 15, a display apparatus 16, and an input apparatus 17 including a keyboard and a mouse. The processor 13, the storage apparatus 14, the communication circuit 15, the display apparatus 16, and the input apparatus 17 are connected to each other via a bus 18. Further, the display apparatus 16 and the input apparatus 17 are connected to the bus 18 via interfaces (I / F) 16a, 17a, respectively.

[0027] The processor 13 includes a central processing device (CPU), a ROM, and a RAM. The processor 13 reads and executes programs stored in the ROM and the storage apparatus 14. Further, a part of the processor 13 can also be constituted by a semiconductor device such as an FPGA (Field Programmable Gate Array), an electronic circuit, or the like.

[0028] The storage device 14 includes a program storage area 14a that stores various programs, a first image storage area 14b that stores the first image Gl, a second image storage area 14c that stores the second image G2, and a teacher data storage area 14d that stores teacher data. The first image Gl is a white light image obtained in the white light observation mode in the endoscope device 2, and the second image G2 is a narrow band light image obtained in the narrow band light observation mode in the endoscope device 2.

[0029] The program storage area 14a contains a teacher data generation program TDC and a lesion portion detection algorithm generation program LDC. The teacher data generation program TDC is a software program that generates teacher data for a lesion portion detection algorithm using endoscope images obtained in the endoscope device 2. The processing regarding the teacher data generation program TDC is described later.

[0030] The lesion portion detection algorithm of the lesion portion detection algorithm generation program LDC is an algorithm that is learned using teacher data generated by the teacher data generation program TDC. The lesion portion detection algorithm generation program LDC generates a partial program of a detection algorithm of a lesion portion detection program LDP stored in the CAD device 3.

[0031] Further, the teacher data generation program TDC is executed in the server 11 here, but can also be executed in a computer such as a personal computer.

[0032] (Generation processing of teacher data)

[0033] The generation processing of teacher data is described. Figure 3 is a flowchart that shows an example of the processing flow of image data. Figure 3 The processing of the teacher data generation program TDC is shown. The teacher data generation program TDC is read out from the storage device 14 by the processor 13 and executed.

[0034] Before the generation processing of teacher data is performed, endoscope images of each observation mode taken in the endoscope device 2 are transmitted to the server 11 via the network 12. A white light image taken and obtained in the white light observation mode is stored as the first image Gl in the first image storage area 14b. A narrow band light image taken and obtained in the narrow band light observation mode is stored as the second image G2 in the second image storage area 14c. Thus, image data of a plurality of endoscope images in the two observation modes taken at each examination in the endoscope device 2 is stored and accumulated in the first image storage area 14b and the second image storage area 14c.

[0035] Also, here, the image data of the first image G1 and the second image G2 are transmitted from the endoscope device 2 to the server 11 via the network 12, but the image data of the first image G1 and the second image G2 can also be transmitted from the storage device 6b to a storage medium such as a USB (Universal Serial Bus: Universal Serial Bus) memory and recorded, and the storage medium can be attached to the server 11 to store the image data of the first image G1 and the second image G2 to the first image storage area 14b and the second image storage area 14c.

[0036] The processor 13 executes the teacher data generation program TDC to generate the teacher data using the first image and the second image. First, the processor 13 acquires the first image G1 from the first image storage area 14b (step (hereinafter, abbreviated as S) 1). The acquisition of the first image G1 is performed by causing a data generator who generates the teacher data to select one from among a plurality of endoscopic images stored in the first image storage area 14b. For example, the display device 16 is caused to display a plurality of white light images stored in the first image storage area 14b, and the data generator selects one from among the displayed plurality of white light images through the input device 17.

[0037] Next, the processor 13 acquires the second image G2 from the second image storage area 14c (S2). The acquisition of the second image G2 is also performed by causing the data generator who generates the teacher data to select one from among a plurality of endoscopic images stored in the second image storage area 14c. For example, the display device 16 is caused to display a plurality of narrow band light images stored in the second image storage area 14c, and the data generator selects one from among the displayed plurality of narrow band light images.

[0038] Therefore, the process of S1 constitutes a first image acquisition section that acquires a first medical image, and the process of S2 constitutes a second image acquisition section that acquires a second medical image that is taken of substantially the same site as the first medical image, which is different from the first medical image.

[0039] Also, when the white light images stored in the first image storage area 14b are dynamic images, the data generator selects and acquires one still image that is determined by temporarily stopping while viewing the dynamic image being reproduced as the first image G1. Similarly, when the narrow band light images stored in the second image storage area 14c are dynamic images, the data generator selects and acquires one still image that is determined by temporarily stopping while viewing the dynamic image being reproduced as the second image G2. That is, the first medical image and the second medical image can each be a still image selected from a dynamic image.

[0040] The processor 13 extracts a lesion region of the lesion portion in the acquired second image G2 (S3). The extraction of the lesion region of the lesion portion can be performed in accordance with a difference in hue, the presence or absence of a prescribed feature quantity, or the like in the second image G2. For example, a region having a pixel value of a prescribed color and a luminance value of a prescribed threshold value or more is extracted as the lesion region.

[0041] In addition, here, the processor 13 extracts the lesion region by image processing, but the data generator can also set the lesion region by drawing the boundary of the region of the lesion portion using the input device 17 such as a mouse on the second image G2 displayed on the display device 16.

[0042] The processor 13 performs the alignment of the first image G1 and the second image G2 (S4). In S3, two or more feature points in each of the first image G1 and the second image G2 are extracted, and the offset amount of the first image G1 and the second image G2 is detected and adjusted on the basis of the two or more extracted feature points. In addition, the position of the blood vessel pattern in the first image G1 and the position of the blood vessel pattern in the second image G2, or the like can be compared, and the offset amount of the first image G1 and the second image G2 can be detected. The processing of S4 constitutes an alignment section that performs the alignment of the first medical image and the second medical image.

[0043] Figure 4 is a diagram for explaining the process of generating the teacher data. In S1, S2, the first image G1 that is a white light image and the second image G2 that is a narrow band light image are selected by the data generator. The first image G1 is a white light image, and the lesion portion is displayed in a manner that is difficult to visually recognize. On the other hand, the second image G2 is a narrow band light image using the above-described two narrow band lights, and the lesion region LR is displayed in the second image G2 in a recognizable manner. Since the second image is acquired at a timing different from the timing at which the first image G1 is acquired, the first image G1 and the second image G2 are images obtained by photographing substantially the same portion in the subject, but differ in the viewpoint position and the like.

[0044] Therefore, in S4, two or more feature points in each of the first image G1 and the second image G2 are detected, the offset amount of the second image G2 with respect to the first image G1 is calculated on the basis of the positions of the two or more feature points, and the alignment of the first image G1 and the second image G2 is performed on the basis of the calculated offset amount. In Figure 4 In, the center position of the second image G2 indicated by the double-dotted line is not only offset in the XY direction with respect to the center position of the first image G1, but also the second image G2 is rotated by an angle θ with respect to the first image G1 about the line-of-sight direction. Therefore, in S4, the positional offset amount of the second image G2 with respect to the first image G1 in the XY direction and the rotation angle θ about the line-of-sight direction are detected, and the alignment of the subject image in the first image and the subject image in the second image is performed.

[0045] Additionally, alignment may include scaling up or down the size of the second image G2 used to adjust the size of the subject. This is because there may be cases where the distance from the front end of the insertion part 4a to the subject when acquiring the first image G1 is different from the distance from the front end of the insertion part 4a to the subject when acquiring the second image G2.

[0046] In addition, Figure 4 In the example, the positional offset of the second image G2 relative to the first image G1 in the XY direction and the rotation angle θ around the viewing direction are detected. However, the angle between the viewing direction of the first image G1 and the viewing direction of the second image G2 can also be detected and aligned. That is, when the plane perpendicular to the viewing direction of the first image G1 and the plane perpendicular to the viewing direction of the second image G2 are not parallel, correction can be performed to deform the second image G2 so that the plane perpendicular to the viewing direction of the second image G2 and the plane perpendicular to the viewing direction of the first image G1 are parallel.

[0047] Following S4, processor 13 sets the mask region MR (S5). The mask region MR is the region that masks the area outside the lesion region LR in the first image G1. Therefore, in Figure 4 In the second image G2 after alignment, the area outside the lesion region LR in the first image G1 (represented by double-dotted lines) is the masking region MR (represented by diagonal lines) of the masking image MG. In S5, the masking image MG with the specified masking region MR is generated.

[0048] Additionally, while a masked region MR is defined here, lesion region information representing the lesion region LR can also be generated in S5. Specifically, lesion region information can also be generated in S5 by specifying the region MR. Figure 4 Information about the region (i.e., the lesion region LR) indicated by the double-dotted line in the first image G1.

[0049] Therefore, the processing in S5 constitutes a lesion region information generation unit, which generates lesion region information (masking region information or lesion region information) about the lesion in the second medical image (narrowband light image). In S5, lesion region information is generated based on the lesion region of the lesion in the second medical image after alignment with the first medical image.

[0050] Following S5, processor 13 generates teacher data and stores it in teacher data storage area 14d in teacher data table TBL (S6). The teacher data generated here includes image data of the first image G1 and masking region information MRI of the masking image MG. That is, the teacher data constitutes medical image data for learning that associates the first image G1 with lesion region information.

[0051] Figure 5 is a diagram showing the structure of the teacher data table TBL. The teacher data table TBL stores a plurality of sets of data, each set of data being constituted by image data of a first image G1 and mask region information MRI representing a lesion region in the first image G1. As shown, for example, the first image data "G100001" and the mask region information "MRI 00001" constitute one set of data. The data of each set constitutes one teacher data. By this, the processing of S6 constitutes a data generation section that generates learning medical image data in which a first medical image (white light image) is associated with lesion region information. Figure 5

[0052] By executing the above processing of S1 to S6, one teacher data is generated. Therefore, the data generator can simply generate a larger number of teacher data as learning medical image data by repeating the processing of S1 to S6.

[0053] The generated teacher data is used as learning medical image data for a lesion portion detection model of a lesion portion detection algorithm of the lesion portion detection program LDP. By increasing the teacher data, the lesion portion detection model obtained by machine learning is updated, and improvement of the lesion portion detection accuracy of the lesion portion detection algorithm is expected. The lesion portion detection program LDP using the updated lesion portion detection model is transmitted from the server 11 to the CAD apparatus 3 via the network 12 as shown by the dotted line, and the software program of the lesion portion detection algorithm is updated. As a result, in the endoscope apparatus 2, the detection result information in which the lesion portion detection accuracy is improved is displayed together with the endoscope image on the display device 7.

[0054] Therefore, according to the above-described embodiment, a learning medical image data generation apparatus that can simply generate teacher data can be provided.

[0055] In addition, in the above-described embodiment, the alignment processing (S4) of the first image G1 and the second image is performed, but in the case of the first image G1 and the second image G2 obtained at the switching timing of the observation mode, the alignment processing can not be performed.

[0056] For example, sometimes the white light image is taken in the white light observation mode, and immediately after that, the observation mode is switched from the white light observation mode to the narrow band light observation mode by operating the operation panel 6c, and immediately after the switching, the narrow band light image is taken. In such a case, sometimes the position and the orientation of the insertion portion 4a of the endoscope 4 hardly change.

[0057] ​Alternatively, according to the observation mode switching timing, when a white light image (or a narrow-band light image) immediately before the switching and a narrow-band light image (or a white light image) immediately after the switching are automatically acquired as the first image G1 and the second image G2, respectively, the position and the orientation of the insertion portion 4a of the endoscope 4 at the time of acquisition of the first image G1 and the second image G2 hardly change.

[0058] Therefore, in a case where the teacher data is generated from the first image G1 and the second image G2 thus acquired, the processing of S4 of Figure 3 can be omitted.

[0059] In addition, in the above-described embodiment, the endoscope device 2 has two observation modes of the white light observation mode and the narrow-band light observation mode, the first image is a white light image, and the second image is a narrow-band light image of two narrow-band lights having center wavelengths of 415 nm and 540 nm, but as long as a wavelength capable of detecting a desired lesion is used, the second image can be a narrow-band light image of a wavelength band other than 415 nm and 540 nm.

[0060] Also, the endoscope device 2 can have an observation mode other than the white light observation mode and the narrow-band light observation mode, the first image can be a white light image, and the second image can be a fluorescence observation image.

[0061] Further, the first image can be a white light image, and the second image can be a staining image obtained by staining an observation site with Lugol's solution containing iodine or the like, a coloring image obtained by scattering indigo carmine or the like on the observation site.

[0062] Also, the first image can be a white light image, and the second image can be an image obtained by another modality (an X-ray image, an ultrasonic wave image, or the like), an image including a region in which a biopsy is performed by forceps later, or the like.

[0063] According to the above-described embodiment and each modification example, for example, in a case where a physician other than an expert physician cannot or hardly recognize a lesion portion, teacher data can be quickly generated. As a result, the learning medical image data generation apparatus of the above-described embodiment and each modification example contributes to quickly providing a CAD apparatus capable of performing a prompt of a biopsy site with a high degree of accuracy in which a missed detection rate of a lesion portion is reduced.

[0064] Further, a program for executing the above-described operations is recorded or stored as a whole or a part thereof in a removable medium such as a floppy disk, a CD-ROM, and the like, a storage medium such as a hard disk, and the like. The program is read by a computer, and all or a part of the operations are executed. Alternatively, the whole or a part of the program can be distributed or provided via a communication network. A user can easily implement the learning medical image data generation apparatus of the present application by downloading the program via the communication network and installing it in the computer, or installing it in the computer from the recording medium.

[0065] The present application is not limited to the above-described embodiments, and various modifications, changes, and the like can be made within a scope that does not change the gist of the present application.

Claims

1. A learning-use medical image data generation apparatus, wherein, The learning medical image data generation apparatus has: a first image acquisition section that acquires a first medical image; a second image acquisition section that acquires a second medical image different from the first medical image, the second medical image being an image obtained by imaging substantially the same site as the first medical image; an alignment section that performs alignment between the first medical image and the second medical image; a lesion region information generation section that generates lesion region information about a lesion portion within the second medical image after the alignment is performed; and a data generation section that generates a plurality of sets of learning medical image data by machine learning training models, the learning medical image data being generated by associating the lesion region information about the lesion portion within the second medical image after the alignment is performed with the first medical image.

2. The learning medical image data generation apparatus according to claim 1, wherein the first medical image and the second medical image are each a still image selected from a dynamic image.

3. The learning medical image data generation apparatus according to claim 1, wherein the first medical image is an image imaged in a first observation mode, and the second medical image is an image imaged in a second observation mode different from the first observation mode.

4. The learning medical image data generation apparatus according to claim 3, wherein the first observation mode is a white light observation mode, the second observation mode is a narrow band light observation mode.

5. The learning medical image data generation apparatus according to claim 1, wherein the alignment section detects an angle formed by a line of sight direction of the first medical image and a line of sight direction of the second medical image, and performs the alignment based on the angle.

6. A learning-use medical image data generation apparatus, wherein The learning medical image data generation apparatus has: an alignment section that performs alignment between a first medical image and a second medical image, the second medical image being an image obtained by imaging substantially the same site as the first medical image, different from the first medical image; a lesion region information generation section that generates lesion region information about a lesion portion within the second medical image after the alignment is performed; and a data generation section that generates a plurality of sets of learning medical image data by machine learning training models, the learning medical image data being generated by associating the lesion region information about the lesion portion within the second medical image after the alignment is performed with the first medical image.

7. A lesion region information generating apparatus, wherein, The lesion region information generation apparatus has: a first image acquisition section that acquires a first medical image, a second image acquisition section that acquires a second medical image different from the first medical image, the second medical image being an image obtained by imaging substantially the same site as the first medical image, an alignment section that performs alignment between the first medical image and the second medical image, a lesion region information generation section that generates lesion region information about a lesion portion within the second medical image after the alignment is performed, a data generation section that generates learning medical image data in which the lesion region information about the lesion portion in the second medical image after the alignment is associated with the first medical image.

8. A lesion region information generating apparatus, wherein, The lesion region information generation apparatus has: an alignment section that performs alignment between a first medical image and a second medical image that is an image obtained by imaging substantially the same portion as the first medical image, a lesion region information generation section that generates lesion region information about a lesion portion in the second medical image after the alignment, a data generation section that generates learning medical image data in which the lesion region information about the lesion portion in the second medical image after the alignment is associated with the first medical image.

9. A recording medium having recorded thereon a program, wherein The program causes a computer to execute the following processing: acquiring a first medical image; acquiring a second medical image that is an image obtained by imaging substantially the same portion as the first medical image, performing alignment between the first medical image and the second medical image, generating lesion region information about a lesion portion in the second medical image after the alignment, and generating a plurality of sets of learning medical image data in which the lesion region information about the lesion portion in the second medical image after the alignment is associated with the first medical image, by training a model by machine learning.

10. A recording medium having recorded thereon a program, wherein The program causes a computer to execute the following processing: performing alignment between a first medical image and a second medical image that is an image obtained by imaging substantially the same portion as the first medical image, generating lesion region information about a lesion portion in the second medical image after the alignment, and generating a plurality of sets of learning medical image data in which the lesion region information about the lesion portion in the second medical image after the alignment is associated with the first medical image, by training a model by machine learning.

Images