Image processing device, learning device, method for producing images, method for producing a learning model, and program

Abstract

Conventionally, it has not been possible to generate a high-precision Type 2 medical image of the brain, which is a different type from the Type 1 medical image, using a Type 1 medical image of the subject's brain. [Solution] An image processing device 1 comprises an image management unit 111 that stores two or more Type 1 medical images including a non-contrast image and a contrast image; a contrast effect image acquisition unit 131 that acquires a contrast effect image using two or more images including a non-contrast image and a contrast image; an input image acquisition unit 132 that acquires an input image using two or more Type 1 medical images including a contrast effect image; a prediction unit 133 that acquires a Type 2 medical image by machine learning prediction processing using a learning model and the input image; and an output unit 14 that outputs a Type 2 medical image. This allows a Type 2 medical image to be generated using a Type 1 medical image of the subject's brain.

Description

Technical Field

[0001] The present invention relates to an image processing apparatus that processes a brain image of a subject, etc.

Background Art

[0002] Conventionally, for example, there has been research on generating a methionine PET image from an MRI image (see Non-Patent Document 1). In this conventional technology, among MRI images, an attempt is made to generate a methionine PET image only from an imaging image of a contrast-enhanced T1-weighted image of glioma.

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the above conventional technology, for example, although the development of a technology for generating a methionine PET image (an example of a second type of medical image) from an MRI image (an example of a first type of medical image) has been explored and its promise has been shown, improvement in the accuracy of the generated image has been required for clinical application.

[0005] That is, in the conventional technology, it has not been possible to generate a highly accurate second type of brain medical image (for example, a methionine PET image) of a different type from the first type of medical image using the first type of medical image (for example, an MRI image) of the subject's brain. Note that the second type of medical image is an image in which a person can more easily distinguish the attention area than the first type of medical image.

[0006] Therefore, the inventors considered that, for example, MRI images often include not only contrast-enhanced T1-weighted images but also plain T1-weighted and T2-weighted images, and that this information is likely to contain useful information for generating methionine PET images. Furthermore, the inventors considered that because MRI images vary significantly depending on the facility and machine used for imaging, preprocessing is necessary before inputting them into artificial intelligence for generating methionine PET images. Moreover, the inventors considered that by appropriately preprocessing MRI images and combining multiple types of MRI images before inputting them into artificial intelligence, the accuracy of methionine PET images generated from MRI images is expected to improve. [Means for solving the problem]

[0007] The first image processing apparatus of the present invention comprises: an image management unit that stores two or more Type I medical images, including a non-contrast image which is a brain image of a subject when no contrast agent is used and a contrast image which is a brain image of a subject when a contrast agent is used; a contrast effect image acquisition unit that acquires a contrast effect image using two or more images including the non-contrast image and the contrast image; an input image acquisition unit that acquires an input image using two or more Type I medical images including the contrast effect image acquired by the contrast effect image acquisition unit; a prediction unit that performs machine learning learning processing using two or more training data having a Type II medical image of a different type from the Type I medical image as the target variable and the input image as the explanatory variable, and acquires a Type II medical image by machine learning prediction processing using the acquired learning model and the input image acquired by the input image acquisition unit; and an output unit that outputs the Type II medical image acquired by the prediction unit.

[0008] With this configuration, it is possible to generate a high-precision Type 2 medical image of the brain, which is of a different type from the Type 1 medical image, using a Type 1 medical image of the subject's brain.

[0009] Furthermore, the image processing apparatus of this second invention differs from the first invention in that the input image acquisition unit acquires an input image using a contrast-enhanced image and other first-class medical images other than the contrast-enhanced image, and the prediction unit acquires a second-class medical image by machine learning prediction processing using a learning model and the input image acquired by the input image acquisition unit.

[0010] With this configuration, it is possible to generate a high-precision Type 2 medical image of the brain, which is of a different type from the Type 1 medical image, using a Type 1 medical image of the subject's brain.

[0011] Furthermore, the image processing apparatus of this third invention, compared to the first or second invention, comprises a correction means for acquiring a contrast-enhanced image, which is a corrected non-contrast image and a corrected contrast-enhanced image, which is a corrected non-contrast image, and a correction means for acquiring a contrast-enhanced image, which is a corrected contrast-enhanced image, and a correction means for acquiring a contrast-enhanced image, which is a corrected contrast-enhanced image, using the corrected non-contrast image and the corrected contrast-enhanced image.

[0012] With this configuration, it is possible to generate a higher-precision Type 2 medical image of the brain, which is of a different type from the Type 1 medical image, using a Type 1 medical image of the subject's brain.

[0013] Furthermore, the image processing apparatus of the fourth invention, compared to the third invention, is an image processing apparatus in which the correction means corrects the pixel value among the pixel values ​​of each pixel constituting the non-contrast image so that it is between approximately 60 and approximately 90, thereby acquiring a corrected non-contrast image, and corrects the pixel value among the pixel values ​​of each pixel constituting the contrast image so that it is between approximately 60 and approximately 90, thereby acquiring a corrected contrast image.

[0014] With this configuration, it is possible to generate a higher-precision Type 2 medical image of the brain, which is of a different type from the Type 1 medical image, using a Type 1 medical image of the subject's brain.

[0015] Furthermore, the image processing apparatus of the fifth invention is an image processing apparatus comprising: an input image acquisition unit which performs R imaging, G imaging, or B imaging on each of the non-aerosol images, a erosol image, and an erosol image to a different color, and acquires a color non-aerosol image which is a colorized non-aerosol image, a color erosol image which is a colorized erosol image, and a color erosol image which is a colorized erosol image; and an input image acquisition unit which acquires an input image using the color non-aerosol image, the color erosol image, and the color erosol image.

[0016] With this configuration, it is possible to generate a high-precision, color, Type 2 medical image of the brain of a subject using a Type 1 medical image of the brain, which is of a different type from the Type 1 medical image.

[0017] Furthermore, the image processing apparatus of the sixth invention is an image processing apparatus for any one of the first to fifth inventions, wherein the first type medical image is an MRI image or a CT image, and the second type medical image is a nuclear medicine image.

[0018] This configuration allows for the generation of highly accurate nuclear medicine images of the brain using MRI or CT images of the subject's brain.

[0019] Furthermore, the image processing apparatus of the seventh invention, compared to the sixth invention, has an image management unit that stores a T1-weighted image which is a first-class medical image, a contrast-enhanced T1-weighted image which is a first-class medical image, and a T2-weighted image which is a first-class medical image; a contrast-enhanced image acquisition unit that acquires a contrast-enhanced image using the T1-weighted image and the contrast-enhanced T1-weighted image; and an input image acquisition unit that acquires an input image using the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit, the T1-weighted image, and the T2-weighted image.

[0020] With this configuration, it is possible to generate more accurate Class II medical images of the brain using T1-weighted images, contrast-enhanced T1-weighted images, and T2-weighted images of the subject's brain.

[0021] The learning device of the eighth invention also includes an original image management unit that stores two or more sets of original images, which are a set of two or more first-type medical images and second-type medical images including a non-contrast image, which is a brain image of a subject without using a contrast agent, and a contrast image, which is a brain image of the subject with a contrast agent; a contrast effect image acquisition unit that acquires a contrast effect image using two or more images including the non-contrast image and the contrast image for each of the two or more sets of original images; an input image acquisition unit that acquires an input image using two or more first-type medical images including the contrast effect image acquired by the contrast effect image acquisition unit for each of the two or more sets of original images; a teacher data acquisition unit that acquires teacher data having the input image acquired by the input image acquisition unit and the second-type medical image of the original image management unit for each of the two or more sets of original images; a learning unit that performs learning processing of machine learning using the two or more teacher data acquired by the teacher data acquisition unit and acquires a learning model; and a storage unit that stores the learning model.

[0022] With such a configuration, a learning model for generating a highly accurate second-type medical image of the brain using the first-type medical image of the brain of the subject can be created.

[0023] The learning device of the ninth invention further includes an image management unit that stores two or more first-type medical images including a non-contrast image, which is a brain image of a subject without using a contrast agent, and a contrast image, which is a brain image of the subject with a contrast agent, with respect to the eighth invention; a contrast effect image acquisition unit that acquires a contrast effect image using two or more images including the non-contrast image and the contrast image; and an input image acquisition unit that acquires an input image using two or more first-type medical images including the contrast effect image acquired by the contrast effect image acquisition unit. The teacher data in the teacher data storage unit is information having the input image acquired by the input image acquisition unit and the second-type medical image.

[0024] With such a configuration, a learning model for generating a highly accurate second-type medical image of the brain using the first-type medical image of the brain of the subject can be created.

Advantages of the Invention

[0025] According to the image processing apparatus of the present invention, using a first type of medical image of the brain of a subject, a second type of highly accurate medical image of the brain different from the first type of medical image can be generated.

Brief Description of the Drawings

[0026] [Figure 1] Block Diagram of Image Processing Apparatus 1 in Embodiment 1 [Figure 2] Flowchart for Explaining the Operation Example of the Image Processing Apparatus 1 [Figure 3] Flowchart for Explaining an Example of the Contrast Effect Image Acquisition Process [Figure 4] Flowchart for Explaining an Example of the Input Image Acquisition Process [Figure 5] Processing Schematic Diagram of the Image Processing Apparatus 1 [Figure 6] Diagram Showing an Example of the First Type of Medical Image [Figure 7] Diagram for Explaining the Tone Correction Process [Figure 8] Diagram Showing an Example of the Result of the Tone Correction Process [Figure 9] Diagram for Explaining the Process of Acquiring the Contrast Effect Zscore Image [Figure 10] Diagram for Explaining the Colorization Process [Figure 11] Diagram for Explaining the Input Image Acquisition Process [Figure 12] Diagram for Explaining the Process of Acquiring the Generated Methionine PET Image [Figure 13] Comparison Diagram between the Generated Methionine PET Image and the Actual Methionine PET Image [Figure 14] Block Diagram of Learning Apparatus 2 in Embodiment 2 [Figure 15] Flowchart for Explaining the Operation Example of the Learning Apparatus 2 [Figure 16] Block Diagram of the Computer System in the Above Embodiment

Modes for Carrying Out the Invention

[0027] The following describes embodiments of the image processing device and the like with reference to the drawings. Note that components denoted by the same reference numerals in these embodiments perform similar operations, and therefore, further explanation may be omitted.

[0028] (Embodiment 1) In this embodiment, an image processing device is described that acquires a contrast-enhanced image using two or more Type 1 medical images, which are brain images of a subject, and which include a non-contrast image and a contrast-enhanced image of the subject's brain using a contrast agent; synthesizes the two or more Type 1 medical images including the contrast-enhanced image to acquire an input image of the subject; and uses the input image and a learning model to acquire a Type 2 medical image, which is a brain image of the subject.

[0029] The image processing device can be, for example, a terminal, but it can also be a server. If the image processing device is a terminal, it can be, for example, a smartphone, a personal computer, or a tablet device, but the type is not specified. If the image processing device is a server, it can be, for example, a cloud server or an ASP server, but the type is not specified.

[0030] In this specification, information X being associated with information Y means that information Y can be obtained from information X, or information X can be obtained from information Y, and the method of association is irrelevant. Information X and information Y may be linked, may reside in the same buffer, may information X be contained in information Y, or information Y may be contained in information X, and so on.

[0031] Furthermore, in this specification, selecting or determining information Z means obtaining information Z, obtaining a pointer to information Z, obtaining the ID of information Z, setting a flag on information Z, etc., and it is sufficient to be able to access information Z.

[0032] Figure 1 is a block diagram of the image processing device 1 in this embodiment. The image processing device 1 comprises a storage unit 11, a receiving unit 12, a processing unit 13, and an output unit 14. The storage unit 11 comprises an image management unit 111 and a learning model management unit 112. The processing unit 13 comprises a contrast-enhanced image acquisition unit 131, an input image acquisition unit 132, and a prediction unit 133. The contrast-enhanced image acquisition unit 131 comprises a correction means 1311 and a contrast-enhanced image acquisition means 1312. The input image acquisition unit 132 comprises a colorization means 1321 and an input image acquisition means 1322.

[0033] The storage unit 11, which constitutes the image processing device 1, stores various types of information. These types of information include, for example, various images (described later) and a learning model (described later).

[0034] The image management unit 111 stores two or more Class I medical images. A Class I medical image is a brain image of a subject. A Class I medical image is, for example, an MRI image or a CT image. The two or more Class I medical images are brain images of a single subject.

[0035] Two or more Class I medical images include both non-contrast images and contrast-enhanced images. Non-contrast images are brain images of a subject without the use of a contrast agent. Contrast-enhanced images are brain images of a subject with the use of a contrast agent. Examples of non-contrast images include T1-weighted images and T2-weighted images. Examples of contrast-enhanced images include contrast-enhanced T1-weighted images.

[0036] The learning model management unit 112 stores the learning model. The learning model is a model obtained by performing machine learning training using two or more training data sets, each having a Type 2 medical image as the target variable and input images as explanatory variables. Preferably, the learning model is a model created by the learning device 2, which will be described later. The Type 2 medical image and input image that constitute the training data are either brain images of one subject or images based on brain images of one subject.

[0037] The Type II medical image is preferably a nuclear medicine image. A nuclear medicine image is, for example, a PET image. A nuclear medicine image may also be, for example, a SPECT image. The PET image is preferably a methionine PET image. A PET image may also be, for example, an FDG-PET image.

[0038] Here, a learning model is information constructed through the learning process of machine learning, and is used in the prediction process of machine learning. A learning model can also be called a learner, classifier, or classification model.

[0039] The machine learning algorithm used here is preferably a conditional GAN ​​that learns using pairs of input and target images. Among conditional GANs, the pix2pix algorithm is particularly preferred. However, any machine learning algorithm that can generate a Type 2 medical image (described later) from one or more Type 1 medical images is acceptable. Such algorithms may include, for example, CycleGAN, Pix2PixHD, StyleGAN, etc. The pix2pix algorithm is an image transformation model using a generative adversarial network (GAN) that aims to transform one type of image into another type of image.

[0040] The reception unit 12 receives various instructions and information. These instructions and information include, for example, start instructions, various images, and learning models.

[0041] The term "reception" encompasses the acceptance of information input from input devices such as keyboards, mice, and touch panels; the reception of information transmitted via wired or wireless communication lines; and the acceptance of information read from recording media such as optical discs, magnetic discs, and semiconductor memory.

[0042] A start command is an instruction to acquire and output a type 2 medical image using two or more images, including a non-contrast image and a contrast-enhanced image.

[0043] The processing unit 13 performs various processes. These processes include, for example, those performed by the contrast effect image acquisition unit 131, the input image acquisition unit 132, or the prediction unit 133.

[0044] The contrast-enhanced image acquisition unit 131 acquires a contrast-enhanced image using two or more images, including a non-contrast image and a contrast-enhanced image, from the image management unit 111.

[0045] The contrast-enhanced image acquisition unit 131 acquires a contrast-enhanced image, for example, using a T1-weighted image and a contrast-enhanced T1-weighted image.

[0046] The contrast-enhanced image acquisition unit 131 preferably includes a correction means 1311 and a contrast-enhanced image acquisition means 1312, as described below. However, it is not essential for the contrast-enhanced image acquisition unit 131 to perform the gradation correction processing described later.

[0047] The correction means 1311, which constitutes the contrast-enhanced image acquisition unit 131, performs grayscale correction on the non-contrast image and acquires a corrected non-contrast image, which is the corrected non-contrast image. The correction means 1311 also performs grayscale correction on the contrast-enhanced image and acquires a corrected contrast-enhanced image, which is the corrected contrast-enhanced image.

[0048] The correction means 1311 corrects the pixel values ​​of each pixel constituting the non-contrast image so that the most frequent pixel value becomes the reference pixel value, and acquires a corrected non-contrast image. The correction means 1311 also corrects the pixel values ​​of each pixel constituting the contrast image so that the most frequent pixel value becomes the reference pixel value, and acquires a corrected contrast image. This gradation correction process can absorb differences in image attribute values ​​(e.g., brightness, luminance) due to differences in equipment used to acquire Type I medical images (e.g., magnetic resonance imaging apparatus, computed tomography apparatus) and the environment in which the Type I medical images are acquired.

[0049] The reference pixel value is the pixel value with the highest number of pixels among the pixel values ​​that make up the image. Preferably, the reference pixel value is between 60 and 90. Preferably, the reference pixel value is 75. Note that the boundaries of 60 and 90 for the reference pixel value do not need to be precise; they can be considered approximately 60 and 90. Similarly, the appropriate reference pixel value of 75 does not need to be precise; it can be considered approximately 75.

[0050] The contrast-enhanced image acquisition means 1312 acquires a contrast-enhanced image using a corrected non-contrast image and a corrected contrast-enhanced image. The contrast-enhanced image acquisition means 1312 acquires a contrast-enhanced image by, for example, combining a corrected non-contrast image and a corrected contrast-enhanced image. The contrast-enhanced image acquisition means 1312 acquires a contrast-enhanced zscore image, which is a contrast-enhanced image, by, for example, combining a corrected T1-weighted image and a corrected contrast-enhanced T1 image. The contrast-enhanced image acquisition means 1312 may also acquire a contrast-enhanced image using an uncorrected non-contrast image and an uncorrected contrast-enhanced image. The contrast-enhanced image acquisition means 1312 may also acquire a contrast-enhanced image by combining an uncorrected non-contrast image and an uncorrected contrast-enhanced image. Note that combining two or more images means superimposing two or more images.

[0051] In this specification, the algorithm for combining two or more images is not limited. Examples of algorithms for combining two or more images include image synthesis using GANs, overlay, methods where the pixel value of the combined pixel is the sum of the pixel values ​​of corresponding pixels in the two or more images, methods where the pixel value of the combined pixel is the weighted average of the pixel values ​​of corresponding pixels in the two or more images, and methods using image normalization.

[0052] The input image acquisition unit 132 acquires an input image using two or more Class I medical images, including a contrast-enhanced image acquired by the contrast-enhanced image acquisition unit 131. For example, the input image acquisition unit 132 acquires an input image by synthesizing two or more Class I medical images, including a contrast-enhanced image acquired by the contrast-enhanced image acquisition unit 131. The input image acquired by the input image acquisition unit 132 is usually a single image, but it may be two or more images.

[0053] The input image acquisition unit 132 acquires an input image, for example, using a contrast-enhanced image and other Class I medical images other than the contrast-enhanced image. The input image acquisition unit 132 acquires an input image, for example, by combining a contrast-enhanced image and one or more other Class I medical images other than the contrast-enhanced image.

[0054] The input image acquisition unit 132 acquires an input image, for example, using the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit 131, the T1-weighted image, and the T2-weighted image. The input image acquisition unit 132 also acquires an input image by combining the contrast-enhanced image, the T1-weighted image, and the T2-weighted image, for example.

[0055] The input image acquisition unit 132 preferably includes a colorization means 1321 and an input image acquisition means 1322, as described below. However, it is not essential for the input image acquisition unit 132 to acquire a colorized input image.

[0056] The colorization means 1321 performs R imaging, G imaging, or B imaging on each of the non-contrast image, contrast image, and contrast effect image to a different color, thereby obtaining a color non-contrast image, a color contrast image, and a color contrast effect image.

[0057] R-imaging is the process of obtaining an image in which each pixel of a monochrome image (for example, an image with 256 grayscale levels) is colored red (R). G-imaging is the process of obtaining an image in which each pixel of a monochrome image is colored green (G). B-imaging is the process of obtaining an image in which each pixel of a monochrome image is colored blue (B).

[0058] The colorization means 1321 preferably performs R imaging on a non-contrast image to obtain a color non-contrast image, G imaging on a contrast image to obtain a color contrast image, and B imaging on a contrast effect image to obtain a color contrast effect image. However, it is not specified which color is applied to which image.

[0059] The input image acquisition means 1322 acquires an input image using a color non-contrast image, a color contrast image, and a color contrast effect image. It is preferable for the input image acquisition means 1322 to acquire an input image by combining the color non-contrast image, the color contrast image, and the color contrast effect image.

[0060] The prediction unit 133 uses the learning model from the learning model management unit 112 and the input image acquired by the input image acquisition unit 132 to acquire a Class II medical image through machine learning prediction processing. Typically, the prediction unit 133 uses the learning model and the input image to provide a module that performs machine learning prediction processing, executes the module, and acquires a Class II medical image. The machine learning prediction processing is, for example, image conversion processing using the pix2pix algorithm.

[0061] The output unit 14 outputs the Type 2 medical image acquired by the prediction unit 133. The Type 2 medical image is, for example, a nuclear medicine image such as a methionine PET image.

[0062] Here, "output" is a concept that includes displaying on a screen, projecting using a projector, printing with a printer, transmitting to an external device, storing on a recording medium, and transferring processing results to other processing devices or other programs.

[0063] The storage unit 11, the image management unit 111, and the learning model management unit 112 are preferably made of non-volatile recording media, but can also be made of volatile recording media.

[0064] The process by which information is stored in the storage unit 11, etc. is not relevant. For example, information may be stored in the storage unit 11, etc. via a recording medium, information transmitted via a communication line, etc. may be stored in the storage unit 11, etc., or information input via an input device may be stored in the storage unit 11, etc.

[0065] The reception unit 12 can be implemented using device drivers for input means such as touch panels or keyboards, or control software for menu screens. The reception unit 12 may also be implemented using wireless or wired communication means.

[0066] The processing unit 13, contrast-enhanced image acquisition unit 131, input image acquisition unit 132, prediction unit 133, correction means 1311, contrast-enhanced image acquisition means 1312, colorization means 1321, and input image acquisition means 1322 can typically be implemented using a processor, memory, etc. The processing procedures of the processing unit 13, etc., are typically implemented in software, and this software is recorded on a recording medium such as ROM. However, it may also be implemented in hardware (dedicated circuitry). The processor can be a CPU, MPU, GPU, etc., and the type is not limited.

[0067] The output unit 14 may or may not include output devices such as displays and speakers. The output unit 14 can be implemented using driver software for an output device, or using driver software for an output device and an output device. The output unit 14 may also be implemented using wireless or wired communication means.

[0068] Next, an example of the operation of the image processing device 1 will be explained using the flowchart in Figure 2. The image processing device 1 will, for example, begin the following operations upon receiving a start command.

[0069] (Step S201) The contrast-enhanced image acquisition unit 131 acquires a contrast-enhanced image using two or more Class I medical images stored in the image management unit 111. An example of such contrast-enhanced image acquisition process will be explained using the flowchart in Figure 3.

[0070] (Step S202) The input image acquisition unit 132 acquires an input image using two or more Class I medical images, including the contrast-enhanced image acquired in step S201. An example of such input image acquisition processing will be explained using the flowchart in Figure 4.

[0071] (Step S203) The prediction unit 133 acquires the learning model from the learning model management unit 112.

[0072] (Step S204) The prediction unit 133 provides the learning model acquired in step S203 and the input image acquired in step S202 to a module that performs machine learning prediction processing (for example, a module for the pix2pix algorithm), executes the module, and acquires a second-class medical image.

[0073] (Step S205) The output unit 14 outputs the Type 2 medical image acquired in step S204. The process ends.

[0074] Next, an example of the contrast-enhanced image acquisition process in step S201 will be explained using the flowchart in Figure 3.

[0075] (Step S301) The correction means 1311 assigns 1 to counter i.

[0076] (Step S302) The correction means 1311 determines whether or not there is an i-th Type 1 medical image to be used when acquiring a contrast-enhanced image. If there is an i-th Type 1 medical image, proceed to step S303; otherwise, proceed to step S305.

[0077] (Step S303) The correction means 1311 performs grayscale correction processing on the i-th first type medical image and obtains a corrected first type medical image.

[0078] (Step S304) The correction means 1311 increments the counter i by 1. Return to step S302.

[0079] (Step S305) The contrast-enhanced image acquisition means 1312 acquires two or more corrected Class I medical images to be used when acquiring the contrast-enhanced image. The corrected Class I medical images are, for example, a corrected T1-weighted image and a corrected contrast-enhanced T1-weighted image.

[0080] (Step S306) The contrast-enhanced image acquisition means 1312 synthesizes two or more corrected Class I medical images to acquire a contrast-enhanced image. It then returns to the higher-level processing. The synthesis of two or more images here is done, for example, by overlay.

[0081] Next, an example of the input image acquisition process in step S202 will be explained using the flowchart in Figure 4.

[0082] (Step S401) The input image acquisition unit 132 acquires two or more images to be used when acquiring the input image. These images include contrast-enhanced images.

[0083] (Step S402) The colorization means 1321 decides whether or not to perform colorization. If colorization is performed, the process proceeds to step S403; if colorization is not performed, the process proceeds to step S407. Note that whether or not to perform colorization is usually predetermined. Furthermore, performing colorization is preferable.

[0084] (Step S403) The colorization means 1321 assigns 1 to counter i.

[0085] (Step S404) The colorization means 1321 determines whether or not an i-th image to be colorized exists. If an i-th image exists, the process proceeds to step S405; otherwise, the process proceeds to step S407.

[0086] (Step S405) The colorization means 1321 places the color (R, G, or B) corresponding to the i-th image to be colorized onto the i-th image.

[0087] (Step S406) The colorizing means 1321 increments the counter i by 1. Return to step S404.

[0088] (Step S407) The input image acquisition means 1322 combines three or more images that have undergone colorization processing, or combines two or more images used when acquiring the input image. As a result, the input image is acquired. The process returns to the higher-level processing. Note that the combination of two or more images here is done, for example, by overlay.

[0089] The following describes specific examples of the operation of the image processing device 1 in this embodiment. Here, an example of the process by which the image processing device 1 generates a methionine PET image from an MRI image will be explained using the schematic diagram of the process in Figure 5.

[0090] The image management unit 111 of the image processing device 1 stores three Class I medical images shown in Figure 6 (501 in Figure 5). The three images are MRI brain images of a single subject, and are a T1-weighted image, a contrast-enhanced T1-weighted image, and a T2-weighted image, respectively.

[0091] Assume that the reception unit 12 of the image processing device 1 has received a processing start instruction. Then, the contrast-enhanced image acquisition unit 131 acquires a contrast-enhanced image using the three Class I medical images stored in the image management unit 111, as follows. Specifically, the correction means 1311 first performs the grayscale correction processing described above on each of the three images in Figure 6, acquires a corrected T1-enhanced image (701 in Figure 7) from the T1-enhanced image, acquires a corrected contrast-enhanced T1-enhanced image (702 in Figure 7) from the contrast-enhanced T1-enhanced image, and acquires a corrected T2-enhanced image (703 in Figure 7) from the T2-enhanced image.

[0092] Figure 8(a1) is a histogram of the number of pixels (vertical axis) against each signal intensity (horizontal axis) of the uncorrected T1-weighted MRI, (a2) is a histogram of the number of pixels (vertical axis) against each signal intensity (horizontal axis) of the uncorrected contrast-enhanced T1-weighted MRI, and (a3) ​​is a histogram of the number of pixels (vertical axis) against each signal intensity (horizontal axis) of the uncorrected T2-weighted MRI. Furthermore, Figure 8(b1) is a histogram of the number of pixels (vertical axis) against each signal intensity (horizontal axis) of the corrected T1-weighted MRI, (b2) is a histogram of the number of pixels (vertical axis) against each signal intensity (horizontal axis) of the corrected contrast-enhanced T1-weighted MRI, and (b3) is a histogram of the number of pixels (vertical axis) against each signal intensity (horizontal axis) of the corrected T2-weighted MRI. The histograms (a1), (a2), and (a3) ​​in Figure 8 show that the variability of each signal intensity in the three uncorrected images is large. When generating a methionine PET image using these three uncorrected images, it may be difficult to identify glioblastoma. On the other hand, the histograms (b1), (b2), and (b3) in Figure 8 show that in each of the three corrected images, the reference pixel value, which is the most frequent pixel value, is set to "75". When generating a methionine PET image using these three corrected images, it is possible to obtain an image in which glioblastoma can be easily identified.

[0093] Next, the contrast-enhanced image acquisition means 1312 acquires two corrected Class I medical images (corrected T1-weighted image 901 and corrected contrast-enhanced T1-weighted image 902) to be used when acquiring the contrast-enhanced image. Next, the contrast-enhanced image acquisition means 1312 synthesizes these two or more corrected Class I medical images to acquire a contrast-enhanced zscore image (903 in Figure 9). The contrast-enhanced zscore image is an example of a contrast-enhanced image.

[0094] Next, the input image acquisition unit 132 performs R imaging by overlaying red onto the T1-weighted image to obtain a color T1-weighted image 1001 (see Figure 10). Note that color T1-weighted image 1001 is an example of a color non-contrast image. The input image acquisition unit 132 also performs G imaging by overlaying green onto the T2-weighted image to obtain a color T2-weighted image 1002. Note that color T2-weighted image 1002 is an example of a color non-contrast image. The input image acquisition unit 132 also performs B imaging by overlaying blue onto the contrast-enhanced zscore image to obtain a color contrast-enhanced zscore image 1003. Note that color contrast-enhanced zscore image 1003 is an example of a color contrast-enhanced image.

[0095] Next, the input image acquisition means 1322 synthesizes three or more colorized images (1001, 1002, 1003) to acquire the input image (1101 in Figure 11). Note that the input image is 502 in Figure 5.

[0096] Next, the prediction unit 133 acquires the learning model from the learning model management unit 112. This learning model is the model described above. The learning model is, for example, the device that acquires the learning device 2, which will be described later.

[0097] Next, the prediction unit 133 provides the acquired learning model and the acquired input image (1201 in Figure 12) to the pix2pix algorithm module, executes the module, and acquires the target image, which is the generated methionine PET image (1202 in Figure 12). The generated methionine PET image is an example of a Class II medical image.

[0098] Next, the output unit 14 outputs the generated methionine PET image. In Figure 13, the actual methionine PET image of the subject is 1301. On the other hand, the generated methionine PET image 1202 output by the image processing device 1 can be said to appropriately visualize the extent of glioblastoma in the subject's brain compared to the actual methionine PET image 1301. In Figure 5, the generated methionine PET image is 503.

[0099] As described above, according to this embodiment, it is possible to generate a high-precision Type 2 medical image of the brain of a subject using a Type 1 medical image of the brain, which is of a different type from the Type 1 medical image.

[0100] Furthermore, according to this embodiment, a Type 2 medical image of the brain, which is of a different type from the Type 1 medical image, can be generated using a Type 1 medical image of the subject's brain, resulting in a high-precision color image of the brain.

[0101] Furthermore, according to this embodiment, highly accurate nuclear medicine images of the brain can be generated using MRI or CT images of the subject's brain.

[0102] Furthermore, according to this embodiment, a more accurate Class II medical image of the brain can be generated using T1-weighted images, contrast-enhanced T1-weighted images, and T2-weighted images of the subject's brain.

[0103] The processing in this embodiment may be implemented by software. This software may be distributed by software download or the like. Alternatively, this software may be recorded on a recording medium such as a CD-ROM and distributed. This also applies to other embodiments in this specification. The software that implements the image processing device 1 in this embodiment is the following program. In other words, this program provides access to a computer that can access an image management unit that stores two or more Class I medical images, including non-contrast images, which are brain images of the subject when no contrast agent is used, and contrast images, which are brain images of the subject when a contrast agent is used. A contrast-enhanced image acquisition unit acquires a contrast-enhanced image using two or more images, including the non-contrast image and the contrast-enhanced image. An input image acquisition unit acquires an input image using two or more Class I medical images, including the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit, A machine learning learning process is performed using two or more training data sets, each having a Type 2 medical image of a different type from the aforementioned Type 1 medical image as the target variable and input images as explanatory variables. A prediction unit then uses the acquired learning model and the input images acquired by the input image acquisition unit to perform a machine learning prediction process to acquire a Type 2 medical image. This is a program to cause the prediction unit to function as an output unit that outputs the Type II medical image acquired by the prediction unit.

[0104] (Embodiment 2) In this embodiment, a learning device for creating a learning model used by the image processing device 1 will be described.

[0105] Figure 14 is a block diagram of the learning device 2 in this embodiment. The learning device 2 comprises a storage unit 21, a reception unit 22, a processing unit 23, and an output unit 24. The storage unit 21 comprises a source image management unit 211 and a training data storage unit 212. The processing unit 23 comprises a contrast effect image acquisition unit 131, an input image acquisition unit 132, a training data acquisition unit 231, and a learning unit 232. The output unit 24 comprises a storage unit 241.

[0106] The storage unit 21 stores various types of information. These types of information include, for example, the original image set (described later), the training data (described later), and the learning model.

[0107] The original image management unit 211 stores two or more sets of original images. One set of original images is a collection of two or more Type 1 medical images and Type 2 medical images obtained from the brain of one subject. The set of original images is a collection of images that will be used to create the training data described later. The two or more Type 1 medical images in the set of original images are, for example, non-contrast images and contrast-enhanced images. The Type 2 medical images in the set of original images are, for example, methionine PET images.

[0108] The training data storage unit 212 stores two or more training data sets. The training data includes an input image and a second-type medical image of a different type from the first-type medical image. The input image is a so-called explanatory variable, and the second-type medical image is a so-called dependent variable. The second-type medical image is, for example, a methionine PET image. The input image and the first-type medical image in the training data are brain images of a single subject, or images based on brain images of that single subject.

[0109] The reception unit 22 receives various instructions and information. These instructions and information include, for example, instructions to start learning and training data.

[0110] The term "reception" encompasses the acceptance of information input from input devices such as keyboards, mice, and touch panels; the reception of information transmitted via wired or wireless communication lines; and the acceptance of information read from recording media such as optical discs, magnetic discs, and semiconductor memory.

[0111] The processing unit 23 performs various processes. These processes are, for example, performed by the contrast effect image acquisition unit 131, the input image acquisition unit 132, or the learning unit 232.

[0112] The training data acquisition unit 231 acquires two or more training data sets, each containing an input image acquired by the input image acquisition unit 132 and a second-class medical image from the original image management unit 211. It is preferable for the training data acquisition unit 231 to store the two or more training data sets in the training data storage unit 212.

[0113] The contrast-enhanced image acquisition unit 131 acquires a contrast-enhanced image for each of the two or more original image sets using two or more images, including a non-contrast image and the contrast-enhanced image.

[0114] The input image acquisition unit 132 acquires an input image for each of the two or more original image sets using two or more Type 1 medical images, including the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit 131.

[0115] The learning unit 232 uses two or more training data sets acquired by the training data acquisition unit 231 to perform machine learning training and acquire a training model. The training process is the process of acquiring a training model to be used by the prediction unit 133. For example, the training process is the process of acquiring a training model to be used by the pix2pix algorithm.

[0116] The output unit 24 outputs various types of information. These types of information include, for example, a learning model. Here, output is a concept that includes displaying on a screen, projecting using a projector, printing with a printer, transmitting to an external device, storing on a recording medium, and transferring processing results to other processing devices or other programs.

[0117] The storage unit 241 stores the learning model acquired by the learning unit 232. The storage unit 241 stores the learning model in, for example, the storage unit 21 or the learning model management unit 112 of the image processing device 1.

[0118] The storage unit 21, the original image management unit 211, and the training data storage unit 212 are preferably made of non-volatile recording media, but can also be made of volatile recording media.

[0119] The process by which information is stored in the storage unit 21, etc. is not relevant. For example, information may be stored in the storage unit 21, etc. via a recording medium, information transmitted via a communication line, etc. may be stored in the storage unit 21, etc., or information input via an input device may be stored in the storage unit 21, etc.

[0120] The reception unit 22 can be implemented using device drivers for input means such as touch panels or keyboards, or control software for menu screens. The reception unit 22 may also be implemented using wireless or wired communication means.

[0121] The processing unit 23, the training data acquisition unit 231, the learning unit 232, and the storage unit 241 can typically be implemented using a processor, memory, etc. The processing procedures of the processing unit 23, etc., are usually implemented in software, and this software is recorded on a recording medium such as ROM. However, it may also be implemented in hardware (dedicated circuitry). The processor can be a CPU, MPU, GPU, etc., and the type is not limited.

[0122] The output unit 24 may or may not include output devices such as displays and speakers. The output unit 24 can be implemented using driver software for the output device, or using driver software for the output device and the output device itself.

[0123] Next, an example of the operation of the learning device 2 will be explained using the flowchart in Figure 15. In the flowchart in Figure 15, the same steps as in the flowchart in Figure 2 will be omitted.

[0124] (Step S1501) The processing unit 23 assigns 1 to counter i.

[0125] (Step S1502) The processing unit 23 determines whether the i-th original image set exists in the original image management unit 211. If the i-th original image set exists, the process proceeds to step S201; otherwise, it proceeds to step S1505.

[0126] (Step S1503) The training data acquisition unit 231 constructs training data comprising the input image acquired by the input image acquisition unit 132 in step S202 and a second-class medical image from the original image management unit 211 that corresponds to the input image, and stores the training data in the training data storage unit 212.

[0127] (Step S1504) The processing unit 23 increments counter i by 1. The process returns to step S1502.

[0128] (Step S1505) The learning unit 232 uses two or more training data stored in the training data storage unit 212 to perform machine learning training and obtain a learning model.

[0129] (Step S1506) The storage unit 241 stores the learning model acquired by the learning unit 232. The process ends.

[0130] As described above, according to this embodiment, a learning model can be created for generating highly accurate Type 2 medical images of the brain using Type 1 medical images of the subject's brain.

[0131] The processing in this embodiment may be implemented in software. This software may be distributed via software download or the like. Alternatively, this software may be recorded on a recording medium such as a CD-ROM and distributed. This also applies to other embodiments in this specification. The software that implements the learning device 2 in this embodiment is the following program. In other words, this program is a program that enables a computer to access an original image management unit, which stores two or more original image sets, each set of two or more Type 1 medical images and Type 2 medical images, including non-contrast images, which are brain images of a subject when no contrast agent is used, and contrast images, which are brain images of the subject when a contrast agent is used. The program then enables a computer to function as: a contrast effect image acquisition unit that acquires a contrast effect image for each of the two or more original image sets using two or more images, including the non-contrast image and the contrast image; an input image acquisition unit that acquires an input image for each of the two or more original image sets using two or more Type 1 medical images, including the contrast effect image acquired by the contrast effect image acquisition unit; a training data acquisition unit that acquires training data for each of the two or more original image sets, which includes the input image acquired by the input image acquisition unit and the Type 2 medical images in the original image management unit; a learning unit that performs machine learning learning processing and acquires a learning model using the two or more training data acquired by the training data acquisition unit; and a storage unit that stores the learning model.

[0132] Figure 16 is a block diagram of a computer system 300 that executes the program described herein to realize the image processing apparatus 1 or learning apparatus 2 of the various embodiments described above.

[0133] In Figure 16, the computer system 300 includes a computer 301 with a CD-ROM drive, a keyboard 302, a mouse 303, and a monitor 304.

[0134] In Figure 16, the computer 301 includes, in addition to the CD-ROM drive 3012, an MPU 3013, a bus 3014 connected to the CD-ROM drive 3012, a ROM 3015 for storing programs such as boot-up programs, a RAM 3016 connected to the MPU 3013 for temporarily storing application program instructions and providing temporary storage space, and a hard disk 3017 for storing application programs, system programs, and data. Although not shown here, the computer 301 may further include a network card for providing connectivity to a LAN.

[0135] The program that causes the computer system 300 to execute the functions of the image processing device 1, etc., as described above, may be stored on the CD-ROM 3101, inserted into the CD-ROM drive 3012, and then transferred to the hard disk 3017. Alternatively, the program may be transmitted to the computer 301 via a network (not shown) and stored on the hard disk 3017. The program is loaded into the RAM 3016 during execution. The program may also be loaded directly from the CD-ROM 3101 or the network.

[0136] The program does not necessarily have to include an operating system (OS) or third-party program that causes the computer 301 to execute functions such as the image processing device 1 of the above embodiment. The program only needs to include the instruction portion that calls appropriate functions (modules) in a controlled manner and obtains the desired result. How the computer system 300 operates is well known, so a detailed explanation is omitted.

[0137] In the above program, steps such as sending information and receiving information do not include hardware-based processing, such as processing performed by a modem or interface card in the transmission step (processing that can only be performed by hardware).

[0138] Furthermore, the computer running the above program may be a single computer or multiple computers. In other words, it may perform centralized processing or distributed processing.

[0139] Furthermore, it goes without saying that in each of the above embodiments, two or more communication means present in a single device may be physically implemented in a single medium.

[0140] Furthermore, in each of the above embodiments, each process may be implemented by centralized processing by a single device, or by distributed processing by multiple devices.

[0141] It goes without saying that the present invention is not limited to the embodiments described above, and various modifications are possible, all of which are also included within the scope of the present invention. [Industrial applicability]

[0142] As described above, the image processing device 1 according to the present invention has the effect of being able to generate a high-precision Type 2 medical image of the brain of a subject using a Type 1 medical image of the brain, and is useful, for example, as an image processing device that supports the visualization of glioblastoma. [Explanation of Symbols]

[0143] 1 Image processing device 2 Learning device 11, 21 Storage Unit 12, 22 Reception Desk 13, 23 Processing Unit 14, 24 Output section 111 Image Management Department 112 Learning Model Management Department 131 Contrast-enhanced image acquisition unit 132 Input Image Acquisition Unit 133 Prediction Section 211 Former Image Management Department 212 Training data storage unit 231 Training Data Acquisition Unit 232 Learning Department 241 Storage Unit 1311 Correction means 1312 Contrast-enhanced image acquisition method 1321 Colorization method 1322 Input image acquisition means

Claims

1. An image management unit that stores two or more Class I medical images, including non-contrast images, which are brain images of the subject when no contrast agent is used, and contrast-enhanced images, which are brain images of the subject when a contrast agent is used. A contrast-enhanced image acquisition unit acquires a contrast-enhanced image using two or more images, including the non-contrast image and the contrast-enhanced image. An input image acquisition unit acquires an input image using two or more Class I medical images, including the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit, A machine learning learning process is performed using two or more training datasets, each containing a Type 2 medical image of a different type from the aforementioned Type 1 medical image as the target variable and input images as explanatory variables. A prediction unit then uses the acquired learning model and the input images acquired by the input image acquisition unit to perform a machine learning prediction process to acquire a Type 2 medical image. An image processing apparatus comprising: an output unit that outputs the second type medical image acquired by the prediction unit.

2. The aforementioned input image acquisition unit, The input image is acquired using the aforementioned contrast-enhanced image and other Class I medical images other than the said contrast-enhanced image. The prediction unit, The image processing apparatus according to claim 1, which acquires the second type medical image by machine learning prediction processing using the learning model and the input image acquired by the input image acquisition unit.

3. The aforementioned contrast-enhanced image acquisition unit is: Correction means that performs grayscale correction on the non-contrast image and the contrast-enhanced image, and acquires a corrected non-contrast image, which is the corrected non-contrast image, and a corrected contrast-enhanced image, which is the corrected contrast-enhanced image. The image processing apparatus according to claim 1 or claim 2, further comprising: a means for acquiring a contrast-enhanced image using the corrected non-contrast image and the corrected contrast-enhanced image;

4. The correction means is The image processing apparatus according to claim 3, which corrects the pixel value of each pixel constituting the non-contrast image so that the most frequent pixel value is between approximately 60 and approximately 90, and acquires the corrected non-contrast image, and corrects the pixel value of each pixel constituting the contrast image so that the most frequent pixel value is between approximately 60 and approximately 90, and acquires the corrected contrast image.

5. The aforementioned input image acquisition unit, A colorization means that performs R imaging, G imaging, or B imaging on each of the non-contrast image, the contrast image, and the contrast effect image to obtain a colorized non-contrast image (color non-contrast image), a colorized contrast image (color contrast image), and a colorized contrast effect image (color contrast effect image). An image processing apparatus according to any one of claims 1 to 4, comprising input image acquisition means for acquiring the input image using the color non-contigmented image, the color contrasted image, and the color contrasted effect image.

6. The aforementioned Type 1 medical image is an MRI image or a CT image. The image processing apparatus according to any one of claims 1 to 5, wherein the aforementioned second type medical image is a nuclear medicine image.

7. The aforementioned image management unit includes: The T1-weighted image, the contrast-enhanced T1-weighted image, and the T2-weighted image, which are all Type 1 medical images, are stored in the above-mentioned Type 1 medical image. The aforementioned contrast-enhanced image acquisition unit is: Using the T1-weighted image and the contrast-enhanced T1-weighted image, the contrast-enhanced image is acquired. The aforementioned input image acquisition unit, The image processing apparatus according to claim 6, which acquires the input image using the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit, the T1-weighted image, and the T2-weighted image.

8. A source image management unit stores two or more source image sets, which are sets of two or more Type 1 medical images and Type 2 medical images, including non-contrast images, which are brain images of the subject when no contrast agent is used, and contrast images, which are brain images of the subject when a contrast agent is used. For each of the two or more sets of source images, a contrast-enhanced image acquisition unit acquires a contrast-enhanced image using two or more images, including the non-contrast image and the contrast-enhanced image. For each of the two or more original image sets, an input image acquisition unit acquires an input image using two or more Type 1 medical images, including the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit. For each of the two or more original image sets, a training data acquisition unit acquires training data comprising the input image acquired by the input image acquisition unit and the second type medical image from the original image management unit. A learning unit performs machine learning processing using the two or more training data acquired by the training data acquisition unit and acquires a learning model. A learning device comprising a storage unit for storing the aforementioned learning model.

9. An image management unit that stores two or more Class I medical images, including non-contrast images, which are brain images of the subject when no contrast agent is used, and contrast-enhanced images, which are brain images of the subject when a contrast agent is used. A contrast-enhanced image acquisition unit acquires a contrast-enhanced image using two or more images, including the non-contrast image and the contrast-enhanced image. The system further comprises an input image acquisition unit that acquires an input image using two or more Class I medical images, including the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit, The learning device according to claim 8, wherein the training data in the training data storage unit is information comprising the input image acquired by the input image acquisition unit and the second type medical image.

10. An image production method realized by an image management unit that stores two or more Class I medical images, including a non-contrast image which is a brain image of the subject when no contrast agent is used, and a contrast image which is a brain image of the subject when a contrast agent is used, a contrast effect image acquisition unit, an input image acquisition unit, a prediction unit, and an output unit, The contrast-enhanced image acquisition unit acquires a contrast-enhanced image using two or more images, including the non-contrast image and the contrast-enhanced image, in a contrast-enhanced image acquisition step. The input image acquisition step involves the input image acquisition unit acquiring an input image using two or more Class I medical images, including the contrast-enhancing image acquired by the contrast-enhancing image acquisition unit, The prediction step involves the prediction unit performing machine learning training using two or more training datasets, each of which a Type 2 medical image of a different type from the Type 1 medical image is the target variable and the input image is the explanatory variable, and then using the acquired training model and the input image acquired by the input image acquisition unit, performing machine learning prediction to acquire a Type 2 medical image. An image production method comprising: an output step in which the output unit outputs the second type medical image acquired by the prediction unit.

11. A method for producing a learning model, comprising: a source image management unit that stores two or more source image sets, which are sets of two or more Type 1 medical images and Type 2 medical images, including non-contrast images, which are brain images of a subject when no contrast agent is used, and contrast-enhanced images, which are brain images of the subject when a contrast agent is used; a contrast-enhanced image acquisition unit; an input image acquisition unit; a training data acquisition unit; a learning unit; and a storage unit, wherein the learning model is produced by these units. The contrast-enhanced image acquisition unit acquires a contrast-enhanced image for each of the two or more original image sets using two or more images, including the non-contrast image and the contrast-enhanced image. The input image acquisition step involves the input image acquisition unit acquiring an input image for each of the two or more original image sets using two or more Type 1 medical images, including the contrast-enhancing image acquired by the contrast-enhancing image acquisition unit, The training data acquisition unit performs a training data acquisition step in which, for each of the two or more original image sets, the training data has training data consisting of the input image acquired by the input image acquisition unit and the second type medical image from the original image management unit. The learning unit performs machine learning training using the two or more training data acquired by the training data acquisition unit, and obtains a learning model in a learning step. A method for producing a learning model, comprising: an accumulation unit; and an accumulation step for accumulating the learning model.

12. A computer that can access an image management unit that stores two or more Class I medical images, including non-contrast images, which are brain images of the subject when no contrast agent is used, and contrast-enhanced images, which are brain images of the subject when a contrast agent is used. A contrast-enhanced image acquisition unit acquires a contrast-enhanced image using two or more images, including the non-contrast image and the contrast-enhanced image. An input image acquisition unit acquires an input image using two or more Class I medical images, including the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit, A machine learning learning process is performed using two or more training datasets, each containing a Type 2 medical image of a different type from the aforementioned Type 1 medical image as the target variable and input images as explanatory variables. A prediction unit then uses the acquired learning model and the input images acquired by the input image acquisition unit to perform a machine learning prediction process to acquire a Type 2 medical image. A program to cause the prediction unit to function as an output unit that outputs the second type medical image acquired by the prediction unit.

13. A computer that can access the original image management unit, which stores two or more original image sets, each set being a collection of two or more Type 1 medical images and two or more Type 2 medical images, including non-contrast images, which are brain images of the subject when no contrast agent is used, and contrast-enhanced images, which are brain images of the subject when a contrast agent is used. For each of the two or more sets of source images, a contrast-enhanced image acquisition unit acquires a contrast-enhanced image using two or more images, including the non-contrast image and the contrast-enhanced image. For each of the two or more original image sets, an input image acquisition unit acquires an input image using two or more Type 1 medical images, including the contrast-enhanced image acquired by the contrast-enhanced image acquisition unit. For each of the two or more original image sets, a training data acquisition unit acquires training data comprising the input image acquired by the input image acquisition unit and the second type medical image from the original image management unit. A learning unit performs machine learning processing using the two or more training data acquired by the training data acquisition unit and acquires a learning model. A program to function as a storage unit for accumulating the aforementioned learning model.

Images