A method and apparatus for analyzing human body composition using medical images

Abstract

The present invention discloses a method and device for analyzing human body composition using medical images. The human body composition analysis device trains a first artificial intelligence model by using learning data of human tissue areas marked with muscle or fat in a learning medical image, and then uses the first artificial intelligence model to distinguish human tissue from the medical image to be detected, and outputs human body composition information identified according to the area, volume or weight of the human tissue in the medical image to be detected. The present invention is a technology developed through the 2020 Artificial Intelligence Technology Commercialization Support Project (CY20053) of the Seoul Industrial Development Agency of Seoul Special City, "Artificial Intelligence Verification and Commercialization of Opportunistic Screening for Metabolic Syndrome, Osteoporosis and Sarcopenia Based on Computed Tomography (CT) Images".

Description

Technical Field

[0001] Embodiments of the present invention relate to human body composition analysis, and more particularly, to a method and apparatus for analyzing human body composition using medical images. Background Art

[0002] Generally, body composition analysis is performed by measuring bioimpedance to calculate body fat, muscle mass, etc. Bioimpedance measurement is a method of estimating muscle mass and fat mass by connecting electrodes to the subject's body and applying a small current to measure resistance. However, the problem with the bioimpedance method is that the measurement accuracy varies depending on the subject's movement or muscle tension during measurement, the measured skin area, the composition of individual internal organs or bones, etc.

[0003] The present invention is a technology developed through the "AI Validation and Commercialization of Opportunistic Screening for Metabolic Syndrome, Osteoporosis, and Sarcopenia Based on Computed Tomography (CT) Images" of the 2020 AI Technology Commercialization Support Project (CY20053) of the Seoul Industrial Development Agency of Seoul. Summary of the invention

[0004] Technical issues

[0005] The object of the present invention is to provide a method and a device for accurately measuring human body composition using medical images.

[0006] Technical Solution

[0007] In order to achieve the technical problem, an example of a method for analyzing human body composition using medical images according to an embodiment of the present invention includes: training a first artificial intelligence model using learning data for marking human tissue areas in learning medical images, wherein the human tissue areas are at least one of skin, bones, internal organs, blood vessels, muscles or fat; receiving an input of a medical image to be detected; using the first artificial intelligence model to distinguish human tissue from the medical image to be detected; and outputting human body composition information identified according to the area, volume or weight of the human tissue in the medical image to be detected.

[0008] In order to achieve the technical problem, an example of a human body composition analysis device according to an embodiment of the present invention includes: a first artificial intelligence model, which is trained by using learning data of human tissue areas marked in medical images, wherein the human tissue areas are at least one of skin, bones, internal organs, blood vessels, muscles or fat; a first classification unit, which inputs the medical image to be detected into the first artificial intelligence model to distinguish human tissue; and a human body composition recognition unit, which recognizes and outputs human body composition information based on the human tissue distinguished by the first classification unit.

[0009] Beneficial Effects

[0010] According to an embodiment of the present invention, by analyzing medical images, human body composition for the whole or part of a human body can be accurately identified. Since human body composition is identified from medical images, when there are past medical images of the subject, past human body composition can be identified. In addition, medical images accumulated over a period of time can be used to identify changes in human body composition over a period of time. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 FIG. 4 is a schematic diagram of an example of a human body composition analysis device according to an embodiment of the present invention.

[0012] Figure 2 FIG. 4 is a flowchart of an example of a human body composition analysis method according to an embodiment of the present invention.

[0013] Figure 3 and Figure 4 A schematic diagram of an example of a first artificial intelligence model according to an embodiment of the present invention.

[0014] Figure 5 It is a schematic diagram of an example of a method for identifying body composition of a part of a human body according to an embodiment of the present invention.

[0015] Figure 6 A schematic diagram of an example of learning data for a second artificial intelligence model according to an embodiment of the present invention.

[0016] Figure 7 A schematic diagram of an example of learning data for a third artificial intelligence model according to an embodiment of the present invention.

[0017] Figure 8 A schematic diagram of an example of a method for generating learning data for an artificial intelligence model according to an embodiment of the present invention.

[0018] Fig. 9 FIG. 1 is a schematic diagram of an example of biological information according to an embodiment of the present invention.

[0019] Fig.10 FIG. 4 is a structural diagram of a first embodiment of a human body composition analysis device according to an embodiment of the present invention.

[0020] Fig.11 FIG. 4 is a structural diagram of a second embodiment of a human body composition analysis device according to an embodiment of the present invention.

[0021] Fig.12 FIG. 4 is a structural diagram of a third embodiment of a human body composition analysis device according to an embodiment of the present invention.

[0022] Fig.13FIG. 4 is a structural diagram of a fourth embodiment of a human body composition analysis device according to an embodiment of the present invention.

[0023] Fig.14 FIG. 4 is a schematic diagram of an example of a human tissue segmentation method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] A method and apparatus for analyzing human body composition using medical images according to an embodiment of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Figure 1 FIG. 4 is a schematic diagram of an example of a human body composition analysis device according to an embodiment of the present invention.

[0026] See also Figure 1 When receiving the input of the medical image 110, the human body composition analysis device 100 analyzes and outputs the human body composition therein. The human body composition analysis device 100 can be implemented as various types of computing devices including a processor, a memory, an input / output device, etc., or can be implemented as a cloud system, etc. As another example, the human body composition analysis device 100 can be implemented as a part of a medical device, such as a computed tomography (CT) instrument. In addition, the human body composition analysis device 100 can be implemented in various forms, such as in various storage devices (e.g., USB memory, etc.) including software that implements the method of this embodiment.

[0027] The medical image 110 is a three-dimensional medical image obtained by photographing a cross section of a human body. As an example of a three-dimensional medical image, there are CT images, magnetic resonance imaging (MRI) images, etc. In addition, the medical image can be various types of images obtained by photographing a cross section of a human body. The medical image used for learning the artificial intelligence model is referred to as a learning medical image, and the medical image used to identify the components of the human body is referred to as a medical image to be detected.

[0028] The human body composition information 120 refers to information on various human body tissues, such as the area, weight, volume of human body tissues or the amount of fat, muscle mass, skeletal muscle mass, amount of fat in internal organs, volume of specific organs, skin area, etc. The human body composition analysis device 100 can output the human body composition information 120 in various forms, such as text or graphics, etc., an example of which is as follows: Fig. 9 shown.

[0029] Figure 2 FIG. 4 is a flowchart of an example of a human body composition analysis method according to an embodiment of the present invention.

[0030] See also Figure 2The human body composition analysis device 100 trains a first artificial intelligence model, which distinguishes at least one human tissue among skin, bones, internal organs, blood vessels, muscles or fat from a medical image (S200). An example of the first artificial intelligence model is as follows Figure 3 and Figure 4 shown.

[0031] The human body composition analysis device 100 inputs the medical image to be detected into the learned first artificial intelligence model to distinguish human tissues such as fat or muscle (S210). For example, the medical image to be detected is composed of a plurality of CT images for the whole or partial area of ​​the human body. The human body composition analysis device 100 can input the plurality of CT images into the first artificial intelligence model and obtain the result of distinguishing human tissues such as skin, internal organs, fat or muscle from each CT image.

[0032] The human body composition analysis device 100 uses the first artificial intelligence model to identify the area, volume or weight of human tissues distinguished from the medical image to be detected, or uses them to identify human body composition information (S230). For example, when the medical image to be detected is composed of a plurality of CT images of cross-sections of the human body, the human body composition analysis device can use the area of ​​human tissues such as visceral organs, fat or muscle obtained from each CT image and the interval between each CT image to calculate the volume, can use the density of each human tissue (i.e., the density of a specific visceral organ, fat or muscle, etc.) to calculate the weight, and can also use the brightness value to obtain the mass.

[0033] Figure 3 and Figure 4 A schematic diagram of an example of a first artificial intelligence model according to an embodiment of the present invention.

[0034] See also Figure 3 and Figure 4 , the first artificial intelligence model 300 is a model generated by a learning process such as deep learning or machine learning. The first artificial intelligence model 300 can be implemented in various architectures, such as a convolutional neural network (CNN), a dense convolutional neural network (DenseNet), U-Net, GoogLeNet, and a generative adversarial network. For example, when the artificial intelligence model 200 is implemented as a CNN, the artificial intelligence model 300 uses the learning data 400 to perform a learning process for adjusting the connection weights of the artificial neural network.

[0035] The first artificial intelligence model 300 can be trained using the learning data 400 for marking at least one human tissue region of skin, bone, internal organ, blood vessel, muscle or fat. For example, when it is necessary to generate a first artificial intelligence model for distinguishing fat regions in a medical image, the first artificial intelligence model 300 can be trained using the learning data 400 for marking fat regions in a learning medical image. For another example, when it is necessary to generate a first artificial intelligence model 300 for simultaneously distinguishing multiple human tissues such as skin, fat and muscle, bone or nerve, the first artificial intelligence model can be trained using the learning data 400 for marking each human tissue to be distinguished in a learning medical image.

[0036] When receiving the input of the medical image to be detected 310, the learned first artificial intelligence model 300 outputs a result 320 of distinguishing various human tissues from the medical image to be detected 310. In another embodiment, the result of distinguishing human tissues from the medical image to be detected 310 can be used as learning data for further training the first artificial intelligence model.

[0037] Figure 5 A schematic diagram of an example of a method for identifying human body composition for a specific part of a human body according to an embodiment of the present invention.

[0038] See also Figure 5 , the human body composition analysis device 100 may include a first artificial intelligence model 500 and a second artificial intelligence model 510 or a third artificial intelligence model 520. For example, the human body composition analysis device 100 may include a first artificial intelligence model 500 and a second artificial intelligence model 510 ( Fig.11 ), or may include a first artificial intelligence model 500 and a third artificial intelligence model 520 ( Fig.12 ), or may include all of the first artificial intelligence model 500, the second artificial intelligence model 510, and the third artificial intelligence model 520. However, in this embodiment, for ease of explanation, a case where all of the first artificial intelligence model 500, the second artificial intelligence model 510, and the third artificial intelligence model 520 are included will be described.

[0039] The first artificial intelligence model 500 may be a model for distinguishing human tissues, and the second artificial intelligence model 510 and the third artificial intelligence model 520 may be models for identifying specific positions of the human body. That is, the first artificial intelligence model 500 may be a segmentation model of human tissues, and the second artificial intelligence model 510 and the third artificial intelligence model 520 may be an anatomical localization model.

[0040] The second artificial intelligence model 510 is a model trained by using learning data generated based on specific locations of spine markers. Figure 6 As shown, the second artificial intelligence model 510 can be trained using learning data that marks a specific position 610 (e.g., L3 position) of the spine based on the sagittal plane of the spine. At this time, the learned second artificial intelligence model 510 predicts and outputs the L3 position 550 when receiving input of information about the spine (e.g., a spine image or voxel data about a spine region, etc.).

[0041] The third artificial intelligence model 520 is a model trained by using learning data generated based on the spine marking predetermined area. Figure 7 As shown, the third artificial intelligence model can be trained using learning data of a segment region 710 (e.g., a lumbar region) of the spine labeled based on the coronal plane of the spine. At this time, the learned third artificial intelligence model 520 predicts and outputs the lumbar region 560 between the ribs and the pelvis when receiving input of information about the skeleton (e.g., a skeleton image or voxel data about the skeleton region).

[0042] The human body composition analysis device 100 generates information for a bone region (e.g., a spine region) to be input to the second artificial intelligence model 510 or the third artificial intelligence model using the first artificial intelligence model 500. For example, the human body composition analysis device 100 can use the first artificial intelligence model 500 to distinguish a bone region 540 including a spine from a medical image to be detected 530, and the first artificial intelligence model 500 is trained to distinguish between human tissues such as skin, internal organs, fat or muscle, and bone regions.

[0043] For example, the human body composition analysis device 100 sets the voxel value of the bone area 540 distinguished in the medical image to be detected 530 to "1", generates information about the bone area with the voxel value of the remaining area of ​​the medical image to be detected 530 set to "0", and then inputs it into the second artificial intelligence model 510 or the third artificial intelligence model 520, so that a specific position or a specific area of ​​the medical image to be detected 530 can be identified. Among them, the voxel values ​​"1" and "0" are only used to distinguish the values ​​of the bone area and the remaining area, and are not limited to this. In order to distinguish the bone area from the remaining area in the medical image to be detected 530, the voxel value can be set in various forms. In another embodiment, the human body composition analysis device 100 keeps the voxel value of the bone area of ​​the medical image to be detected 530 unchanged, and can generate bone area information with the voxel value of the remaining area of ​​the medical image to be detected set to "0".

[0044] The human body composition analysis device 100 can identify and output the human body composition of a predetermined position or a predetermined area of ​​the human body according to the predetermined position or predetermined area of ​​the medical image to be detected identified by the second artificial intelligence model 510 or the third artificial intelligence model 520. For example, when the medical image 530 to be detected is composed of a plurality of CT images for the whole body or the abdominal area, the human body composition analysis device 100 can identify the CT image corresponding to the predetermined position in the plurality of CT images constituting the medical image to be detected or the plurality of CT images corresponding to the predetermined area through the second artificial intelligence model 510 or the third artificial intelligence model 520. The human body composition analysis device 100 can identify the human body composition based on the human tissue distinguished in the plurality of CT images corresponding to the predetermined position or the predetermined area. As an example, the predetermined position 550 can be an umbilical position defined based on the spine (e.g., the L3 position of the spine), and the predetermined area 560 can be an abdominal area defined based on the spine (e.g., between the ribs and the pelvis).

[0045] Figure 6 A schematic diagram of an example of learning data for a second artificial intelligence model according to an embodiment of the present invention.

[0046] See also Figure 6 Since the second artificial intelligence model is a model for identifying anatomical positions, learning data for the second artificial intelligence model can be generated by marking a predetermined position (e.g., L3) 610 based on the spine. For example, the human body composition analysis device 100 may display an image 600 including the spine on a screen and receive an input of a specific position 610 from a user to generate learning data.

[0047] Although the present embodiment shows the learning data showing the sagittal plane of the spine, the learning data does not necessarily have to be a sagittal plane image of the spine, as long as it is learning data based on specific locations marked in the bone region. Figure 7 As shown, learning data based on coronal plane marking predetermined positions can be used. In addition, for a better understanding of the present invention, although the present embodiment shows an example of learning data including an image of the entire spine, learning data can be generated by marking specific positions based on images of partial areas of the bone region rather than the entire bone region. In another embodiment, the results of distinguishing human tissues by the first artificial intelligence model can be used to generate learning data for the second artificial intelligence model. This will refer to Figure 8 Describe again.

[0048] Figure 7 A schematic diagram of an example of learning data for a third artificial intelligence model according to an embodiment of the present invention.

[0049] See also Figure 7Since the third artificial intelligence model is a model for identifying anatomical positions, learning data for the second artificial intelligence model can be generated by marking a predetermined position (e.g., an area between the ribs and the pelvis, etc.) based on a bone region. For example, the human body composition analysis device 100 may display a spine image 700 on a screen, and may generate learning data by receiving a user's selection of a predetermined area 710.

[0050] This embodiment shows the learning data of the coronal plane showing the bone region, but is not limited thereto. Figure 6 As shown, learning data can be generated by marking a predetermined area based on a sagittal plane image of the bone area. In addition, learning data can be generated by marking a predetermined area based on a partial image of the bone area rather than an entire image of the bone area. In another embodiment, the human tissue differentiation result of the first artificial intelligence model can be used to generate learning data for the third artificial intelligence model. This will refer to Figure 8 Describe again.

[0051] Figure 8 A schematic diagram of an example of a method for generating learning data for an artificial intelligence model according to an embodiment of the present invention.

[0052] See also Figure 8 , the first learning data 810 for training the first artificial intelligence model 820 can be generated by marking 800 human tissue regions such as skin, internal organs, fat or muscle, and bones on the learning medical image. For example, the user can directly distinguish and mark human tissue regions such as skin, internal organs, fat or muscle on the learning medical image. Alternatively, various conventional segmentation algorithms for distinguishing human tissues such as bones, internal organs, fat or muscle are used to distinguish each human tissue in the medical image, and the first learning data can be automatically generated by the information of marking fat or muscle in the distinguished human tissues.

[0053] The first artificial intelligence model 820 can be used to generate the second learning data 850 for training the second artificial intelligence model and the third learning data 870 for training the third artificial intelligence model. For example, the first artificial intelligence model 820 can be a model for distinguishing the bone region 830 including the spine in the medical image. At this time, the human body composition analysis device 100 displays the bone (i.e., spine) region distinguished from the learning medical image on the screen through the first artificial intelligence model 820, and receives the input of the predetermined position (e.g., L3 position) 840 or the predetermined area (e.g., the waist area between the ribs and the pelvis) 860 from the user, so that the second learning data 850 or the third learning data 870 can be generated.

[0054] Fig. 9FIG. 1 is a schematic diagram of an example of biological information according to an embodiment of the present invention.

[0055] See also Fig. 9 The human body composition analysis device 100 identifies human body composition information such as skin area, internal organ area, skeletal muscle area, visceral fat area, etc. based on various human tissues distinguished by the first artificial intelligence model, and can display it in various forms, such as numbers or charts.

[0056] Since the human body composition analysis device 100 distinguishes human body tissues such as fat and muscle in each medical image of a human body cross section, the human body composition information 900 can be identified according to the cross section of the human body and displayed according to the position of the cross section. In another embodiment, the human body composition analysis device can output the human body composition information in the form of an image file or an Excel file of "Microsoft Corporation (Microsoft)".

[0057] Fig.10 FIG. 4 is a structural diagram of a first embodiment of a human body composition analysis device according to an embodiment of the present invention.

[0058] See also Fig.10 , the human body composition analysis device 100 includes a first artificial intelligence model 1000, a first classification unit 1010, a human body composition recognition unit 1020, a first learning data generation unit 1030, and a first learning unit 1040. When the human body composition analysis device 100 uses the learned first artificial intelligence model 1000, the first learning data generation unit 1030 and the first learning unit 1040 may be omitted.

[0059] The first artificial intelligence model 1000 is a model for distinguishing human tissues such as skin, internal organs, bones, fat or muscles when receiving an input of a medical image to be detected. An example of the first artificial intelligence model 1000 is as follows: Figure 3 and Figure 4 shown.

[0060] The first classification unit 1010 distinguishes at least one human tissue such as skin, internal organs, bones, fat or muscle by inputting the medical image to be detected into the learned first artificial intelligence model 1000. For example, when the medical image to be detected is composed of a plurality of CT images of a cross section of a human body, the first classification unit 1010 inputs each CT image constituting the medical image to be detected into the first artificial intelligence model 1000, so as to obtain the distinction result of each human tissue such as fat or muscle in each CT image.

[0061] The human body composition recognition unit 1020 recognizes and outputs human body composition information including the area, volume, weight, mass, etc. of skin, internal organs, bones, fat or muscles, etc. according to the human body tissue distinction result of the medical image to be detected recognized by the first classification unit 1010.

[0062] The first learning data generating unit 1030 generates first learning data for marking human tissue regions such as skin, internal organs, bones, fat or muscle in the learning medical image. For example, the first learning data generating unit 1030 may display the learning medical image on the screen and receive input of each human tissue region from the user to generate learning data. In addition, the first learning data generating unit 1030 may automatically distinguish and mark each region in the learning medical image by applying various conventional region segmentation algorithms for distinguishing each human tissue region to generate the first learning data.

[0063] The first learning unit 1040 trains the first artificial intelligence model 1000 by using the first learning data.

[0064] Fig.11 FIG. 4 is a structural diagram of a second embodiment of a human body composition analysis device according to an embodiment of the present invention.

[0065] See also Fig.11 , the human body composition analysis device 100 includes a first artificial intelligence model 1100, a first classification unit 1110, a second artificial intelligence model 1120, a second classification unit 1130, a human body composition recognition unit 1140, a second learning data generation unit 1150 and a second learning unit 1160. When the human body composition analysis device 100 uses the learned second artificial intelligence model 1120, the second learning data generation unit 1150 and the second learning unit 1160 can be omitted. In another embodiment, the human body composition analysis device 100 may also include Fig.10 The first learning data generating unit 1030 and the first learning unit 1040 in.

[0066] The first artificial intelligence model 1100 and the first classification unit 1110 have Fig.10 The same configuration is not repeated here.

[0067] The second artificial intelligence model 1120 is a model that recognizes and outputs a specific position of a bone region when the bone region information is input. An example of the second artificial intelligence model 1120 is as follows: Figure 5 and Figure 6 shown.

[0068] The second classification unit 1130 inputs the bone region information into the second artificial intelligence model 1120 to identify the predetermined position. For example, the second classification unit 1130 can identify the L3 position by inputting the spine information into the second artificial intelligence model that outputs the L3 position of the spine. The spine information input to the second artificial intelligence model 1120 is information about the bone region of the medical image to be detected that is distinguished by the first classification unit 1110 using the first artificial intelligence model 1100 (for example, voxel information including the bone image or bone region of the spine). For example, the first artificial intelligence model 1100 is a model that not only distinguishes human tissues such as fat or muscle but also distinguishes bone regions from the medical image to be detected. The second classification unit 1130 can input the bone region information output by the first classification unit 1110 into the second artificial intelligence model 1120 to identify the predetermined position of the medical image to be detected.

[0069] The human body composition identification unit 1140 can identify and output human body composition information corresponding to a predetermined position of the human body based on the medical image to be detected corresponding to the predetermined position identified by the second classification unit 1130 and the human tissue area information of the medical image to be detected identified by the first classification unit 1110.

[0070] The second learning data generating unit 1150 generates second learning data for the second artificial intelligence model 1120. The second learning data can be generated by marking a specific location on an image of the entire spine or an image of a portion of the spine. As an example, the second learning data generating unit 1150 can be generated using a bone region distinguished from a learning medical image by the first artificial intelligence model 1100. Figure 8 An example of a method of generating second learning data for a second artificial intelligence model 1120 using the first artificial intelligence model 1100 is shown.

[0071] The second learning unit 1160 trains the second artificial intelligence model 1120 by using the second learning data.

[0072] Fig.12 FIG. 4 is a structural diagram of a third embodiment of a human body composition analysis device according to an embodiment of the present invention.

[0073] See also Fig.12 , the human body composition analysis device 100 includes a first artificial intelligence model 1200, a first classification unit 1210, a third artificial intelligence model 1220, a third classification unit 1230, a human body composition recognition unit 1240, a third learning data generation unit 1250, and a third learning unit 1260. When the human body composition analysis device 100 uses the learned third artificial intelligence model 1220, the third learning data generation unit 1250 and the third learning unit 1260 can be omitted. In another embodiment, the human body composition analysis device 100 may also include Fig.10 The first learning data generating unit 1030 and the first learning unit 1040 in.

[0074] The first artificial intelligence model 1200 and the first classification unit 1210 have Fig.10 The same configuration is not repeated here.

[0075] The third artificial intelligence model 1220 is a model that recognizes and outputs a predetermined area of ​​the spine when the spine information is input. An example of the third artificial intelligence model 1220 is as follows: Figure 5 and Figure 7 shown.

[0076] The third classification unit 1230 inputs the spine information into the third artificial intelligence model 1220 to identify the predetermined area. For example, the third classification unit 1230 can input the spine information into the third artificial intelligence model based on the L3 output of the spine and the upper and lower predetermined (i.e., abdominal area) areas to identify the abdominal area of ​​the medical image to be detected. The spine information input to the third artificial intelligence model 1220 is information about the spine area distinguished by the first classification unit 1210. For example, the first artificial intelligence model 1200 is a model that not only distinguishes human tissues such as fat or muscle from the medical image to be detected, but also distinguishes bone areas. The third classification unit 1230 can input the bone area information output by the first classification unit 1210 into the third artificial intelligence model 1220 to identify the predetermined position of the medical image to be detected.

[0077] The human body composition identification unit 1240 can identify human body composition information corresponding to a predetermined area of ​​the human body based on the medical image to be detected corresponding to the predetermined area identified by the third classification unit 1230 and the human tissue area information of the medical image to be detected identified by the first classification unit 1210.

[0078] The third learning data generating unit 1250 generates third learning data for the third artificial intelligence model 1220. The third learning data can be generated by marking a predetermined area on an image of the entire spine or an image of a portion of the spine. As an example, the third learning data generating unit 1250 can be generated using a bone area distinguished from a learning medical image by the first artificial intelligence model 1200. Figure 8 An example of a method of generating third learning data using the first artificial intelligence model 1200 is shown.

[0079] The third learning unit 1260 trains the third artificial intelligence model 1220 by using the third learning data.

[0080] Fig.13 FIG. 4 is a structural diagram of a fourth embodiment of a human body composition analysis device according to an embodiment of the present invention.

[0081] See also Fig.13 , the human body composition analysis device 100 includes a first artificial intelligence model 1300, a first classification unit 1310, a second artificial intelligence model 1320, a second classification unit 1330, a third artificial intelligence model 1340, a third classification unit 1350 and a human body composition recognition unit 1360. In another embodiment, the human body composition analysis device 100 may further include a learning unit for training each artificial intelligence model 1300, 1320 and 1340.

[0082] The first artificial intelligence model 1300 and the first classification unit 1310 are as follows Fig.10 As shown, the second artificial intelligence model 1320 and the second classification unit 1330 are as shown in Fig.11 In addition, the third artificial intelligence model 1340 and the third classification unit 1350 are as shown in Fig.12 shown.

[0083] The human body component identification unit 1360 can identify human body component information about a predetermined position or a predetermined area of ​​the human body based on the human body tissue area information of the medical image to be detected identified by the first classification unit 1310, the predetermined position information of the medical image to be detected identified by the second classification unit 1330, and the predetermined area information of the medical image to be detected identified by the third classification unit 1350.

[0084] Fig.14 FIG. 4 is a schematic diagram of an example of a human tissue segmentation method according to an embodiment of the present invention.

[0085] See also Fig.14 The human body composition analysis device 100 includes a first artificial intelligence model 1400 and a cascade model 1410. The first artificial intelligence model 1400 can be used with Figure 3 The cascade model 1410 has a structure that sequentially connects at least one artificial intelligence model 1420 and 1422. The output value of the second artificial intelligence model 1420 becomes the input value of the third artificial intelligence model, and thus, the output value of the N-1th artificial intelligence model becomes the input value of the Nth artificial intelligence model 1422. The number of artificial intelligence models 1420 and 1422 sequentially connected to the cascade model 1410 can be variously changed according to the embodiment. Of course, the cascade model 1410 can also be composed of one artificial intelligence model 1420.

[0086] Figure 5The second artificial intelligence model 510 and the third artificial intelligence model 520 shown in the figure are models for identifying anatomical positions of the human body, but the multiple artificial intelligence models 1420 and 1422 sequentially connected to the cascade model 1410 of this embodiment are models for subdividing the human tissues distinguished by the first artificial intelligence model. In other words, the first artificial intelligence model 1400 is a model for segmenting at least one human tissue region of skin, bone, internal organs, blood vessels, muscle or fat from a medical image, and the cascade model 1410 is a model for subdividing the human tissues distinguished by the first artificial intelligence model 1400.

[0087] In one embodiment, the first artificial intelligence model 1400 is a model for distinguishing muscles from medical images, and the cascade model 1410 may be a model that subdivides the muscles distinguished by the first artificial intelligence model 1400 into leg muscles, trunk muscles, etc. The subdivision step may vary depending on the number of artificial intelligence models 1420 and 1422 present in the cascade model 1410. For example, when the cascade model is composed of two artificial intelligence models, the second artificial intelligence model 1420 is a model that distinguishes muscles into trunk, legs, arms, etc., and the third artificial intelligence model may be a model that further subdivides leg muscles into calves and thighs.

[0088] In another embodiment, the first artificial intelligence model 1400 may be a model for distinguishing various internal organs from a medical image, and the cascade model 1410 may be a model for subdividing the distinguished internal organs. For example, when the first artificial intelligence model 1400 distinguishes lung tissue from a medical image, the second artificial intelligence model 1422 of the cascade model 1410 distinguishes the lung tissue into the right lung and the left lung, and the third artificial intelligence model may distinguish the right lung into the upper lobe, the lower lobe, and the middle lobe.

[0089] In another embodiment, the first artificial intelligence model 1400 is a model for distinguishing bone regions from medical images, the second artificial intelligence model 1422 of the cascade model 1410 distinguishes spinal bone regions from bone regions, and the third artificial intelligence model can be a model for dividing the spinal bones into cervical, thoracic and lumbar vertebrae.

[0090] Figures 10 to 13 The first artificial intelligence models 1000, 1100, 1200 and 1300 can be replaced by the first artificial intelligence model 1400 and the cascade model 1410 of this embodiment. At this time, Figures 10 to 13The first classification units 1010, 1110, 1210 and 1310 use the first artificial intelligence model 1400 to distinguish the first human tissue from the medical image to be detected, and the second to N-th human tissues that are subdivided by the artificial intelligence models 1420 and 1422 of the cascade model 1410 can be obtained. The human body component recognition units 1020, 1140, 1240 and 1360 can recognize the human body components by using at least one human tissue from the first to N-th human tissues. For example, for the model for recognizing muscles, the human body component recognition units 1020, 1140, 1240 and 1360 can use the first human tissue area obtained by the first artificial intelligence model to recognize the total muscle mass of the human body, use the second human tissue area obtained by the second artificial intelligence model to recognize the muscle mass of the arm or leg, etc., and use the third human tissue area obtained by the third artificial intelligence model to respectively recognize the muscle mass of the calf and thigh.

[0091] In another embodiment, the first artificial intelligence model is connected to a plurality of cascade models, and the second artificial intelligence model of each cascade model can be connected to a plurality of third artificial intelligence models, so that a tree structure can be formed. For example, when the first artificial intelligence model is a model for distinguishing bones, fat, and muscles, respectively, there may be a first cascade model, a second cascade model, and a third cascade model for subdividing bones, fat, muscles, etc., respectively. When the first cascade model is a model for distinguishing bones, and the second artificial intelligence model of the first cascade model is a model for distinguishing bones into vertebrae and ribs, the second artificial intelligence model of the first cascade model can be connected to the 3-1st artificial intelligence model for subdividing vertebrae and the 3-2nd artificial intelligence model for subdividing ribs. In this way, according to the number and degree of subdivision of the human tissues subdivided, a connection relationship of artificial intelligence models of various shapes in a tree shape can be generated.

[0092] The present invention can also be implemented as a computer-readable code on a computer-readable recording medium. Computer-readable recording media include all types of recording devices that store computer system-readable data. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage devices. In addition, the computer-readable recording medium is dispersed in a networked computer system to store and execute the computer-readable code in a decentralized manner.

[0093] The above description is mainly based on the preferred embodiments of the present invention. It will be appreciated by those skilled in the art that the present invention can be implemented in a converted form without departing from the basic features of the present invention. Therefore, the disclosed embodiments should be considered from an illustrative perspective rather than a limiting perspective. The scope of the present invention is within the scope of the claims rather than the foregoing description, and all differences within the same scope should be interpreted as being included in the invention.

Claims

1. A method for analyzing human body composition using medical images, characterized in that: include: Training a first artificial intelligence model using learning data of marking at least one human tissue region in a learning medical image, wherein the human tissue region includes skin, bone, internal organs, blood vessels, muscle or fat; receiving a medical image to be detected; Using the first artificial intelligence model to distinguish human tissue from the medical image to be detected; Inputting the skeleton area of ​​the medical image to be detected distinguished by the first artificial intelligence model into the second artificial intelligence model or the third artificial intelligence model to identify a predetermined position or area, wherein the second artificial intelligence model is trained using learning data marking predetermined locations of the spine, wherein the third artificial intelligence model is trained using learning data that labels a predetermined region of the spine; and Output human body composition information of the area, volume or weight of human tissue corresponding to the predetermined position or region identified according to the medical image to be detected.

2. The method for analyzing human body composition using medical images according to claim 1, characterized in that: Differentiating the human tissue includes: inputting the first human tissue differentiated by the first artificial intelligence model into a cascade model in which at least one artificial intelligence model is sequentially connected and subdividing it into a second human tissue, and Outputting the human body composition information includes: identifying the human body composition information for at least one of the first human tissue and the second human tissue.

3. The method for analyzing human body composition using medical images according to claim 2, characterized in that: The cascade model includes at least two artificial intelligence models connected in sequence.

4. The method for analyzing human body composition using medical images according to claim 1, characterized in that: The medical image for learning or the medical image to be detected is a computer tomography (CT) image taken of the entire human body or a predetermined area.

5. The method for analyzing human body composition using medical images according to claim 1, characterized in that: Outputting the human body composition information includes: The human body composition information is displayed and output by marking the human body composition information for the human body cross section according to each cross-sectional position.

6. The method for analyzing human body composition using medical images according to claim 1, characterized in that: The training of the first artificial intelligence model includes: training the first artificial intelligence model using learning data that labels at least one area of ​​skin, bones, internal organs, blood vessels, muscles or fat.

7. The method for analyzing human body composition using medical images according to claim 1, characterized in that: Also includes: Learning data for the second artificial intelligence model is generated by marking predetermined positions on the bone area distinguished from the medical image used for learning by the first artificial intelligence model.

8. The method for analyzing human body composition using medical images according to claim 1, characterized in that: Also includes: The learning data of the third artificial intelligence model is generated by marking predetermined areas on the bone areas distinguished from the medical image used for learning by the first artificial intelligence model.

9. The method for analyzing human body composition using medical images according to claim 1, characterized in that: The predetermined area is an abdominal area defined based on the spine.

10. A human body composition analysis device, characterized in that: include: A first artificial intelligence model is trained by using learning data of at least one human tissue region labeled in a medical image, wherein the human tissue region is skin, bone, internal organ, blood vessel, muscle or fat; A first classification unit, which distinguishes human tissues by inputting the medical image to be detected into the first artificial intelligence model, wherein the human tissues include muscle, fat or bone; A human body composition recognition unit, which recognizes and outputs human body composition information based on the differentiated human tissues; a second artificial intelligence model that is trained using learning data that labels predetermined locations of the spine; as well as a second classification unit, which inputs the skeleton region of the to-be-detected medical image distinguished by the first artificial intelligence model into the second artificial intelligence model to identify a predetermined position, The human body composition recognition unit recognizes human body composition information for a predetermined position recognized by the second classification unit.

11. The human body composition analysis device according to claim 10, characterized in that: Also includes: Cascade model, connect at least one artificial intelligence model in sequence, The first classification unit subdivides the first human tissue into at least one second human tissue by inputting the first human tissue distinguished by the first artificial intelligence model into the cascade model, and The human body composition recognition unit recognizes human body composition information for at least one of the first human tissue and the second human tissue.

12. The human body composition analysis device according to claim 10, characterized in that: Also includes: A first learning data generating unit, which generates learning data for the first artificial intelligence model by marking a human tissue region in a medical image, wherein the human tissue region is at least one of skin, bone, internal organ, blood vessel, muscle or fat; A second learning data generating unit generates learning data of the second artificial intelligence model by marking a predetermined position on the bone region distinguished by using the first artificial intelligence model.

13. The human body composition analysis device according to claim 10, characterized in that: Also includes: a third artificial intelligence model, which is trained using learning data based on marked predetermined areas of the spine; a third classification unit, which inputs the skeleton region of the medical image to be detected into the third artificial intelligence model to identify a predetermined region, and The human body composition recognition unit recognizes human body composition information for the predetermined area recognized by the third classification unit.

14. The human body composition analysis device according to claim 13, characterized in that: Also includes: A first learning data generating unit, which generates learning data for the first artificial intelligence model by marking at least one human tissue region in the medical image, wherein the human tissue region includes skin, bones, internal organs, blood vessels, muscles or fat; A third learning data generating unit generates learning data of the third artificial intelligence model by marking a predetermined area on the skeleton area distinguished by using the first artificial intelligence model.

15. A computer-readable recording medium recording a computer program for executing the method according to claim 1.

Images