Artificial Intelligence-Based Medical Image Conversion Method

Abstract

Regarding a method for converting medical images using artificial intelligence, specifically, it relates to a medical image conversion method using artificial intelligence that can generate MR images based on CT images or generate CT images based on MR images by utilizing GAN (Generative Adversarial Network), which is one of the deep learning technologies. According to an embodiment of the present invention, it is possible to provide a medical image conversion method that can generate MR images based on CT images and, conversely, generate CT images based on MR images.

Description

Technical Field

[0001] One aspect of the present invention relates to a method for converting medical images based on artificial intelligence. More specifically, the present invention relates to an artificial intelligence-based medical image conversion method that uses artificial intelligence to perform conversions between medical images belonging to the same domain or between medical images belonging to different domains.

Background Art

[0002] The content described in this section merely provides background information for embodiments of the present invention and does not constitute prior art.

[0003] Examinations using CT images are very inexpensive compared to treatments using MR images, can be performed in a short time, and have the advantage of having few limitations on artificial substances in the body. However, there is a problem of irradiating a large amount of radiation.

[0004] In addition, MR images have the advantage of being able to distinguish tumors more precisely because of their high resolution for soft tissues and can minimize the risk of exposure to radiation because they do not use radiation. However, MR scanners are very expensive, and it takes a lot of time and cost to take MR images. Also, in some cases, MR images cannot be taken if the patient has an artificial substance inserted in the body or has an allergy to a specific contrast agent.

[0005] Generally, the tissues observable with non-contrast-enhanced CT images and contrast-enhanced CT images are different. Also, the tissues observable with CT images and MR images are different.

[0006] Therefore, when a doctor makes a clinical judgment, it would be desirable to be able to use both CT images and MR images as the basis for diagnosis so that only the advantages of each image can be taken and accurate diagnosis can be made.

[0007] However, as described above, there is a problem that radiation exposure must be tolerated to take CT images. In addition, when an artificial substance is inserted into a patient's body or there is a contrast agent allergy, MR images cannot be taken.

[0008] Therefore, if it is possible to generate MR images based on CT images or generate CT images based on MR images, doctors can use multi-modal images to proceed to accurate diagnosis and treatment, which can improve the domestic medical technology level and contribute to the improvement of the health and quality of life of the people.

[0009] Therefore, it is urgent to develop a technology for performing conversion between medical images belonging to different domains, such as being able to generate MR images based on CT images or generate CT images based on MR images.

[0010] The background technology described above cannot necessarily be regarded as publicly known technology publicly disclosed to the general public before the filing of the embodiments of the present invention, whether it was held by the inventor for deriving the embodiments of the present invention or was technical information acquired during the derivation process.

Summary of the Invention

Problems to be Solved by the Invention

[0011] One aspect of the present invention is proposed to solve the above-mentioned problems. The object of the present invention is to provide a medical image conversion method and apparatus capable of performing conversion between medical images belonging to different domains. For example, it is possible to provide a medical image conversion method and apparatus capable of generating MR images based on CT images and, conversely, generating CT images based on MR images.

[0012] Another object of the present invention is to drive an artificial intelligence learning model that performs conversion between medical images belonging to different domains. For example, it is possible to drive an artificial intelligence learning model that can generate an MR image based on a CT image, or conversely, generate a CT image based on an MR image.

[0013] The technical problems to be solved by the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned can be clearly understood by those with ordinary knowledge in the technical field to which the present invention belongs from the following description.

Means for Solving the Problems

[0014] In order to achieve the problems raised above, one aspect of the present invention is a method for converting medical images between different domains while an artificial intelligence learning unit performs learning,

[0015] a step in which the artificial intelligence learning unit receives an input of a first medical image, a step in which a second medical image having a domain different from that of the first medical image is generated from the first medical image,

[0016] a step in which a third medical image having a domain different from that of the second medical image is generated from the second medical image, and

[0017] a step of calculating a loss value between the two images by comparing the third medical image and the first medical image, and a medical image conversion method including the steps can be provided.

[0018] In an embodiment, in the step of calculating the loss value, the artificial intelligence learning unit can be characterized by repeatedly performing the calculation of the loss value for each iteration during learning.

[0019] According to the embodiment, the iteration includes a first iteration and a second iteration that are sequentially connected. The artificial intelligence learning unit may be characterized in that it interrupts learning only when the absolute value of the change amount between the first loss value calculated in the first iteration and the second loss value calculated in the second iteration becomes smaller than a reference value.

[0020] According to the embodiment, the first medical video and the third medical video may be characterized in that they are videos belonging to the same domain.

[0021] According to the embodiment, at the stage of receiving the input of the first medical video, the first medical video may be characterized in that it includes lesion data and normal data.

[0022] According to the embodiment, when the amount of the lesion data is less than that of the normal data, the lesion data may be amplified so that the ratio of the lesion data to the normal data becomes the same.

[0023] According to the embodiment, when the first medical video has all the first slices for the target site and the second slices for the non-target site, the first medical video may be processed to remove the second slices and leave only the first slices.

[0024] According to the embodiment, when the first medical video is one of a pair of paired data videos, it may further include a step of calculating a coincidence rate by comparing the second medical video with the other one of the pair of paired data videos.

[0025] According to the embodiment, when either one of the first original video and the second original video that has a different domain from the first original video and captures the same site as the first original video is used as the first medical video for input, it may further include a step of calculating a coincidence rate by comparing the second medical video with the second original video.

[0026] According to an embodiment, the coincidence rate can be calculated using the Hausdorff distance or the Dice similarity coefficient.

[0027] According to an embodiment, when one of the second medical video and the pair of paired data videos includes a lesion site, the coincidence rate can be quantitatively calculated through the difference in the center of mass of the positions of the two lesions.

[0028] According to another aspect of the present invention, in a method for performing conversion between medical videos belonging to different domains while the artificial intelligence learning unit performs learning,

[0029] an input step in which the artificial intelligence learning unit receives an input of a first medical video, a generation step in which the artificial intelligence learning unit generates a second medical video having a domain different from that of the first medical video from the first medical video by learning, and

[0030] when the first medical video is one of a pair of paired data, a first learning step in which the artificial intelligence learning unit compares the difference between the second medical video and the other first medical video of the pair of paired data and learns in a direction to reduce the corresponding difference, can provide a medical video conversion method including one or more of them.

[0031] According to an embodiment, the paired data can include a pair of videos belonging to different domains for the same site of the same patient, or a pair of videos belonging to different domains for the same site of different patients.

[0032] After the generation stage according to the embodiment, the artificial intelligence learning unit may include a regeneration stage of regenerating, by learning, a third medical video having a domain different from that of the second medical video from the second medical video. Here, according to the embodiment, the first medical video and the third medical video may belong to the same domain.

[0033] According to the embodiment, the artificial intelligence learning unit may include a second learning stage of comparing the difference between the first medical video and the third medical video and learning in a direction of reducing the corresponding difference.

[0034] According to the embodiment, when the first medical video has all of a first slice for a target site and a second slice for a non-target site, the first medical video may be processed to remove the second slice and leave only the first slice before the input stage.

[0035] According to the embodiment, the first medical video may include lesion data and normal data.

[0036] According to the embodiment, when the amount of the lesion data is less than that of the normal data in the first medical video, the lesion data may be amplified before the first medical video is input to the artificial intelligence learning unit. Here, according to the embodiment, both the lesion data and the normal data may be defined in units of slices. For example, a slice including a lesion area may be defined as lesion data, and a slice including only normal tissue may be defined as normal data. Here, a slice may include each tomographed video obtained by continuously performing tomography in the longitudinal axis direction.

[0037] According to the embodiment, amplifying the lesion data may include replicating the lesion data so that the ratio of the lesion data to the normal data becomes the same. Here, replicating may include simply copying.

[0038] According to another aspect of the present invention, it is possible to provide a computer-readable recording medium recording a program for performing a medical video conversion method having the above-described features.

Effects of the Invention

[0039] As described above, according to one embodiment of the present invention,

[0040] it is possible to provide a medical video conversion method and an apparatus thereof that can perform conversion between medical videos belonging to different domains from each other. For example, it is possible to provide a medical video conversion method and an apparatus thereof that can generate an MR video based on a CT video, and conversely, can generate a CT video based on an MR video.

[0041] In addition, it is possible to drive an artificial intelligence learning model that performs conversion between medical videos belonging to different domains from each other. For example, it is possible to provide an artificial intelligence learning model that can generate an MR video based on a CT video, and conversely, can generate a CT video based on an MR video.

[0042] In addition, the effects of the present invention have various effects such as excellent versatility depending on the embodiment, and such effects can be clearly confirmed in the description part of the embodiments described later.

[0043] The following drawings attached to this specification illustrate one embodiment of the present invention, and serve to better understand the technical idea of the present invention together with the detailed description of the invention described above. Therefore, the present invention is not construed as being limited only to the matters described in such drawings.

Brief Description of the Drawings

[0044]

Figure 1

Figure 2

Figure 3

Figure 4

Mode for Carrying Out the Invention

[0045] The advantages, features, and the methods for achieving them of the present invention will become clear by referring to the embodiments described in detail together with the attached drawings. However, the present invention is not limited to the embodiments presented below, and can be embodied in various different forms, and it must be understood to include all conversions, equivalents, and alternatives included in the idea and technical scope of the present invention. The embodiments presented below are provided to make the disclosure of the present invention complete and to fully inform those with ordinary knowledge in the technical field to which the present invention belongs of the scope of the invention. When it is determined that a specific description of the known technology related to the description of the present invention may obscure the gist of the present invention, the detailed description thereof will be omitted. The terms used in this application are merely used to describe specific embodiments and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, terms such as "including" or "having" are intended to specify the presence of the features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, and it must be understood that the presence or possibility of addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof, etc. are not precluded in advance. Terms such as first, second, etc. can be used to describe various components, but the components should not be limited by the terms. The terms are only used for the purpose of distinguishing one component from another. Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. In the description with reference to the accompanying drawings, the same or corresponding components are given the same drawing numbers, and redundant descriptions thereof are omitted.

[0046] FIG. 1 shows a first embodiment of a medical video conversion method according to an embodiment of the present invention, and FIG. 2 shows a second embodiment of amplifying lesion data in the first embodiment of FIG. 1.

[0047] FIG. 3 shows a third embodiment of adding a step of processing the first medical video so as to remove the second slice and leave only the first slice in the first embodiment of FIG. 1, and FIG. 4 shows a fourth embodiment of adding a step of calculating a matching rate in the first embodiment of FIG. 1.

[0048] According to an embodiment of the present invention, a method for converting medical videos between different domains while the artificial intelligence learning unit performs learning can be presented.

[0049] The medical video conversion method according to this embodiment includes a step (S100) in which the artificial intelligence learning unit receives an input of a first medical video, a step (S110) of generating a second medical video whose domain is different from that of the first medical video from the first medical video,

[0050] a step (S120) of generating a third medical video whose domain is different from that of the second medical video from the second medical video, and a step (S130) of calculating a loss value between the two images by comparing the third medical video and the first medical video.

[0051] Here, the first medical image can include an image acquired through an imaging device. The imaging device can include, for example, a CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) device (not shown). Using such a CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) device, non-enhanced CT or MR images of a specific part of a patient can be taken, contrast-enhanced CT or MR images can be taken, and T1 or T2 images can also be taken.

[0052] In this specification, taking CT images or MR images can, in addition to the optical meaning, mean image reconstruction through computing. However, here, medical images do not exclude images having an optical meaning.

[0053] Here, the specific part of the patient can mean the specific part of the same patient. Also, if the parts to be imaged are equal, it can also mean the specific parts of different patients. The imaging device can take non-enhanced CT images or MR images of a specific part of the patient, contrast-enhanced CT images or MR images, and T1 or T2 images.

[0054] For example, the non-enhanced CT images or MR images taken here can be input images and can also be target images.

[0055] Here, the input video and the target video can mean a pair of paired data. For example, CT images and MR images taken of the same specific part of one patient can be a pair of paired data. Also, CT images and MR images taken of the same part of different patients can be a pair of paired data.

[0056] For example, contrast-enhanced CT images and non-contrast-enhanced CT images of the same part of the subject to be imaged can be a pair of paired data. Also, contrast-enhanced MR images and non-contrast-enhanced MR images of the same part can be a pair of paired data, and T1 images and T2 images of the same part can also be paired data.

[0057] When the non-contrast-enhanced CT image is the input image, the non-contrast-enhanced CT image as the input image and the contrast-enhanced CT image as the paired data can be the target image. When the non-contrast-enhanced MR image is the input image, the non-contrast-enhanced MR image as the input image and the contrast-enhanced CT image as the paired data can be the target image.

[0058] When the contrast-enhanced CT image is the input image, the contrast-enhanced CT image as the input image and the non-contrast-enhanced CT image as the paired data can be the target image.

[0059] When the contrast-enhanced MR image is the input image, the contrast-enhanced MR image as the input image and the non-contrast-enhanced MR image as the paired data can be the target image. These input images and target images can generally include the original images.

[0060] Explain the meaning of different Domains described in this specification. For example, CT images and MR images have different Domains as meant in this specification. Also, non-contrast enhanced images and contrast enhanced images have different Domains. T1 images and T2 images are the same MR images, but still have different Domains as meant in this specification. Ultrasonic images and CT images have different Domains. Similarly, ultrasonic images and MR images have different Domains. MR images and PET images have different Domains. That is, for example, among CT images, MR images, contrast enhanced images, non-contrast enhanced images, T1 images, T2 images, PET images, and SPECT images, any one and the remaining one have different Domains as referred to in this specification.

[0061] That is, among the images realized through Reconstruction based on the signals provided by the subject, if there are differences in the Signal, Sequence, or the image realization process (Process) etc. that form the basis of image realization, it can be said that the Domains of the images with the corresponding differences are different from each other.

[0062] To explain the medical image conversion method according to an embodiment of the present invention, take the case where an imaging device captures CT images and MR images of the same part from the same patient to obtain a pair of paired data as an example. In this case, if the CT image is determined as the input image among the pair of paired data, i.e., the CT image and the MR image, then the MR image can be determined as the target image. In this case, the CT image and the MR image can be the original images.

[0063] When the CT image is determined as the input image, the CT image determined as the input image by the medical image conversion method according to this embodiment can become the first medical image.

[0064] That is, the artificial intelligence learning unit can go through the stage of receiving the CT image as the first medical image as input. After receiving the input of the first medical image, the artificial intelligence learning unit can generate a second medical image whose domain is different from that of the first medical image from the first medical image. Here, the second medical image can be meant as an image generated based on the first medical image. For example, an MR image can be generated based on the input CT image, and in this case, the generated MR image can become the second medical image.

[0065] Here, the process in which the artificial intelligence learning unit generates a second medical image whose domain is different from that of the first medical image from the first medical image, for example, the process of generating an MR image from a CT image, can be accomplished in the following two processes according to the embodiments. However, it is not limited to the following two processes.

[0066] The first process is composed of one or more layers that perform 3D convolution operations to receive the input of two-dimensional or three-dimensional image data acquired by an imaging device, etc., and extract high-dimensional features of a size smaller than the original image, and this can be represented by an artificial neural network that extracts features. Such an artificial neural network that extracts features can receive the input of two-dimensional or three-dimensional actual image data acquired by an imaging device, perform 3D convolution operations, and extract high-dimensional image features of a size smaller than the original image.

[0067] The second process is composed of one or more layers that perform 3D deconvolution operations for the purpose of generating two-dimensional or three-dimensional image data of other modalities from the result of the first process described above, and this can be represented by an artificial neural network that generates images of other modalities. Such an artificial neural network that generates images of other modalities can perform 3D deconvolution operations from the 3D convolution operation execution result of the artificial neural network that extracts features in the first process to generate two-dimensional or three-dimensional image data of other modalities.

[0068] The medical video conversion method according to this embodiment may include a step (S130) of calculating a loss value between the generated third medical video and the first medical video after generating the third medical video. For example, in the above-described example, if the CT video is the first medical video, the MR video synthesized based on this can be the second medical video, and the re-synthesized CT video can be generated by re-synthesizing based on the synthesized MR video. Here, the re-synthesized CT video can be the third medical video.

[0069] According to the embodiment, the first medical video and the third medical video can be videos belonging to the same domain. That is, if the first medical video is a CT video, the third medical video can also be a CT video. For example, if the first medical video is an MR video, the third medical video can also be an MR video.

[0070] According to the embodiment, the first medical video and the third medical video can be videos belonging to the same domain. In this case, in the above-described example, the re-synthesized CT video, which is the third medical video, should have no difference compared to the CT video (input video), which is the first medical video. The artificial intelligence learning unit can calculate the loss value between the third medical video and the first medical video to calculate the accuracy of video conversion.

[0071] According to the embodiment, in the step (S130) of calculating the loss value, the artificial intelligence learning unit can repeatedly perform the calculation of the loss value during learning for each iteration. That is, the artificial intelligence learning unit can calculate the loss value between the regenerated video and the input video for each number of iterations while performing learning. That is, in the above-described example, the loss value can be calculated between the CT video and the re-synthesized CT video.

[0072] In the case where the artificial intelligence learning unit includes a first iteration and a second iteration in which the iterations are sequentially connected according to an embodiment, when the absolute value of the change amount between the first loss value calculated in the first iteration and the second loss value calculated in the second iteration becomes smaller than the reference value, the learning can be interrupted on the assumption that overfitting has occurred.

[0073] Artificial intelligence learning is not only adapted to the distribution of the given learning data, and learning can be completed by preventing further learning when the necessary medical images have been sufficiently learned. The artificial intelligence learning unit thus completed can already be applied to medical images having other data distributions respectively.

[0074] According to an embodiment, in the step (S100) of receiving the input of the first medical image, the first medical image can include lesion data and normal data. For example, when the CT image among a pair of paired data CT image and MR image taken by an imaging device is the first medical image, the corresponding CT image can include all of the lesion data and the normal data. Here, the lesion data can mean a part where a lesion is displayed, and the normal data can mean a normal area other than the area where the lesion is displayed. The lesion data and the normal data can exist together when the same part of the same patient is imaged, and can also exist together when the same part is imaged for different patients.

[0075] Generally, the area of the lesion data is smaller than that of the normal data. And when imaging many patients, the number of data with lesions is small. Therefore, generally, the normal data is overwhelmingly more than the lesion data.

[0076] In the case where the lesion data is less than the normal data in the step (S100) of receiving the input of the first medical image, the step (S50) of amplifying the lesion data so that the ratio of the lesion data to the normal data becomes the same can be performed prior to the step (S100) of receiving the input of the first medical image.

[0077] The step of amplifying the lesion data (S50) is, for example, when the number of normal data is 1000 and the number of lesion data is 10, in order to make the number of lesion data correspond to the number of normal data, it is replicated 100 more times to make the number of lesion data 1000, so that the number of normal data and lesion data becomes 1:1, and then this can be used as input data.

[0078] If the artificial intelligence learning part conducts much more learning based on normal data than learning based on lesion data, accurate learning cannot be carried out, and the accuracy of the result of medical image conversion can only be increased by learning the same amount of lesion data as normal data.

[0079] If the first medical image has all the first slices for the target site and the second slices for the non-target site before the step of receiving the input of the first medical image (S100) according to the embodiment, the step of processing the first medical image (S60) can be passed through to remove the second slices and only leave the first slices.

[0080] Before the first medical image is input to the artificial intelligence learning part, the step of removing the unnecessary parts for learning from the first medical image can be additionally carried out before the step of receiving the input of the first medical image (S100) so that only the part to be learned by the artificial intelligence learning part in the first medical image receives the input.

[0081] The learning target of the artificial intelligence learning part, that is, the target site may be, for example, the brain or the liver. When the target site is the brain, the imaging device that takes brain CT images performs a plurality of consecutive tomographic scans in the longitudinal axis direction while generating a plurality of slices, so the imaging range may be not only the brain but also from the head end to the knee.

[0082] In this case, in order to align the input video, i.e., the CT video, with the target video, such as the MR video, it is necessary to use only the slices corresponding to the brain region (target site) in the CT video. As a method for extracting only the target site, an intensity thresholding algorithm of the image or the like can be used according to the embodiment.

[0083] According to this embodiment, the artificial intelligence can significantly reduce the learning time by removing unnecessary input data in advance during the learning process.

[0084] When either the first original video or the second original video that is different from the first original video in domain and captures the same site as the first original video is used as the first medical video and input according to the embodiment, the step of calculating the coincidence rate by comparing the second medical video with the second original video (S70) can be further included.

[0085] A pair of paired data can include the first original video and the second original video. Here, the second original video can be a video that captures the same site as the first original video.

[0086] However, here, the first original video and the second original video do not necessarily have to be for the same patient.

[0087] For example, when the first original video is a CT video and the second original video is an MR video, when the CT video is used as the input video and input as the first medical video, the step of calculating the coincidence rate by comparing the MR video, which is the second medical video generated based on the CT video, with the MR video, which is the second original video (S70) can be added.

[0088] According to the embodiment, the coincidence rate can be calculated using the Hausdorff distance or the Dice similarity coefficient. When one of the second medical video and a pair of paired data contains a lesion site, the coincidence rate can be quantitatively calculated through the difference in the center of mass of the positions of the two lesions.

[0089] According to this embodiment, in addition to the step of calculating the loss value by comparing the regenerated medical video, that is, the third medical video with the first medical video (S130), the medical video generated other than that, that is, the second medical video, is compared with the first medical video and the paired data medical video to calculate the coincidence rate. By measuring the accuracy of the respective results when the second medical video is generated and when the third medical video is generated, the precision of medical video conversion can be improved.

[0090] Another embodiment of the present invention can present a computer-readable recording medium recording a program for executing the above-described medical video conversion method.

[0091] According to another aspect of the present invention, as an input video input to the artificial intelligence learning unit, a sub-input video including asymmetric information of the original medical video can be input together with the original medical video. That is, the artificial intelligence learning unit can receive, as an input video, paired data of the original medical video and the sub-input video.

[0092] Here, the asymmetric information can include a resultant image obtained by calculating the difference in voxel values between the original medical video and the image obtained by aligning the video reversed left and right with the original medical video using the Non-rigid Transform technique after reversing the original medical video left and right.

[0093] For example, when the artificial intelligence learning unit receives the original CT image (A) as input data for the sub-input video, after flipping the original CT image (A) horizontally, the original CT image (A) and the horizontally flipped CT image (B) are aligned, and then the difference in voxel values between the original CT image (A) and the flipped CT image (B) is calculated. The resulting image (C) can be included. Here, the resulting image (C) can become the sub-input video.

[0094] In this case, the original CT image (A) and the resulting image (C) become paired data and can be input to the artificial intelligence learning unit as a pair of input videos.

[0095] Lesions such as stroke and brain tumor show fine intensity differences on CT, but may be difficult to distinguish with the naked eye. Since such lesions are mostly distributed asymmetrically between the left and right, the difference in voxel values between the left and right can be made visible to emphasize the site with the lesion.

[0096] By using the original CT image (A) and the resulting image (C) as the input videos input to the artificial intelligence learning unit, a video with the site of the lesion emphasized can be obtained.

[0097] The embodiments according to the present invention described above can be embodied in the form of a computer program that can be executed through various components on a computer, and such a computer program can be recorded on a computer-readable medium. At this time, the medium can include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, and flash memories.

[0098] On the one hand, the computer program can be specially designed and configured for the present invention or can be known and usable to those skilled in the field of computer software. Examples of computer programs can include not only machine language codes such as those created by compilers, but also high-level language codes that can be executed by a computer using an interpreter or the like.

[0099] The use of the term "the" and similar directive terms in the specification of the present invention (especially in the claims) can apply to both singular and plural. Also, when a range is described in the present invention, the invention including the application of individual values belonging to the range (if there is no contrary description) is the same as that which describes each individual value constituting the range in the detailed description of the invention. For the steps constituting the method according to the present invention, if there is no clear description of the procedure or no contrary description, the steps can be performed in an appropriate procedure. The present invention is not necessarily limited by the described procedure of the steps. The use of all examples or exemplary terms (such as, etc.) in the present invention is simply for explaining the present invention in detail, and the scope of the present invention is not limited by the above examples or exemplary terms unless it is limited by the claims. Also, those skilled in the art will understand that various modifications, combinations, and changes can be made within the scope of the claims or their equivalents constituted by design conditions and factors. Therefore, the idea of the present invention should not be determined only by the described embodiments, and not only the claims described later, but all ranges equivalent to or equivalently changed from this claim belong to the category of the idea of the present invention.

Explanation of Reference Numerals

[0100] S100 Step of receiving the input of the first medical video S110 Step of generating the second medical video S120 Step of generating the third medical video Step S130 of calculating a loss value

Claims

1. In a method for performing conversion between medical videos belonging to different domains while an artificial intelligence learning unit performs learning, an input stage in which the artificial intelligence learning unit receives an input of a first medical video, a generation stage in which the artificial intelligence learning unit generates a second medical video whose domain is different from that of the first medical video from the first medical video by learning, and when the first medical video is one of a pair of paired data, a first learning stage in which the artificial intelligence learning unit compares the difference between the second medical video and the other first medical video of the pair of paired data and learns in a direction to reduce the corresponding difference. A medical video conversion method characterized by the above.

2. The paired data includes a pair of videos belonging to different domains for the same part of the same patient, or a pair of videos belonging to different domains for the same part of different patients. The medical video conversion method according to claim 1.

3. After the generation stage, a regeneration stage in which the artificial intelligence learning unit regenerates a third medical video whose domain is different from that of the second medical video from the second medical video by learning. The medical video conversion method according to claim 1.

4. The first medical video and the third medical video belong to the same domain. The medical video conversion method according to claim 3.

5. The artificial intelligence learning unit includes a second learning stage in which the artificial intelligence learning unit compares the difference between the first medical video and the third medical video and learns in a direction to reduce the corresponding difference. The medical video conversion method according to claim 3.

6. When the first medical video has all of a first slice for a target site and a second slice for a non-target site, a processing stage of processing the first medical video so that the second slice is removed before the input stage and only the first slice remains. The medical video conversion method according to claim 1.

7. The first medical video includes lesion data and normal data. The medical video conversion method according to claim 1.

8. When the amount of the lesion data is less than the normal data, amplify the lesion data. The medical video conversion method according to claim 7.

9. Amplifying the lesion data includes replicating the lesion data so that the ratio of the lesion data to the normal data is the same. The medical video conversion method according to claim 8.

10. A computer-readable recording medium recording a program for performing the medical video conversion method according to any one of claims 1 to 9.

Images