Method and apparatus for automatically analyzing metabolic health imaging indices using artificial intelligence-based positron emission tomography and computed tomography images

Abstract

A method for automatically analyzing metabolic health imaging indices, performed by an apparatus for automatically analyzing metabolic health imaging indices, comprises the steps of: receiving a positron emission tomography (PET) image and a computed tomography (CT) image; identifying a region of interest in the CT image by using a fixed artificial intelligence model; adjusting the identified region of interest by using an interactive artificial intelligence model; and analyzing imaging indices by analyzing the PET image and the CT image for the adjusted region of interest.

Description

Method and device for automatically analyzing metabolic health imaging indicators using artificial intelligence-based positron emission tomography and computed tomography images

[0001] The present invention relates to image analysis technology, and more particularly, to a method and device for automatically analyzing metabolic health image indices that automatically segment and quantify target health image indices in positron emission tomography and computed tomography (PET / CT) images.

[0002] A major risk factor for cardiovascular diseases such as stroke or myocardial infarction is a deterioration in complex metabolic health, including obesity, insulin resistance, and diabetes. This deterioration in complex metabolic health is caused by abnormal glucose metabolism in multiple organs, such as visceral fat, subcutaneous fat, liver, and muscles, and the resulting inflammatory response. Multi-organ analysis of positron emission tomography (PET) images and morphological analysis of computed tomography (CT) images can be used to assess metabolic health risk.

[0003] Positron emission tomography (PET) and computed tomography (CT) are acquired using the same scanner (i.e., a PET / CT scanner). PET is a nuclear medicine molecular imaging technique that visualizes whole-body glucose metabolism. It is useful for identifying and quantifying metabolic diseases and allows for functional segmentation of regions of interest (e.g., visceral fat, subcutaneous fat, liver, or muscle), whereas CT is better suited for morphological segmentation of regions of interest. However, currently, regions of interest (ROIs) are manually drawn and quantified on the acquired images, such as visceral fat, subcutaneous fat, liver, or muscle. This is time-consuming and presents problems in image analysis, including reduced assessment accuracy. Therefore, it is difficult to expect active clinical use. In other words, although current technology can observe glucose metabolic diseases in regions of interest through PET and CT images of visceral fat, subcutaneous fat, liver, or muscle, manual segmentation is required, limiting quantitative analysis. This reliance on expert judgment can lead to issues of consistency and accuracy.

[0004] [Prior Art Literature]

[0005] [Patent Document]

[0006] (Patent Document 1) U.S. Patent No. 7,412,280

[0007] (Patent Document 2) Korean Patent Publication No. 10-2024-0008838

[0008] (Patent Document 3) Korean Patent No. 10-2409464

[0009] (Patent Document 4) Korean Patent Publication No. 10-2020-0056927

[0010] The purpose of the present invention is to provide a method and device for automatically analyzing metabolic health imaging indicators, which automates and systematizes the analysis process of existing PET and CT images by semi-automating the segmentation process through the introduction of artificial intelligence technology, and enables them to be utilized as new automated diagnostic indicators.

[0011] The purpose of the present invention is to provide a method and device for automatically analyzing metabolic health imaging indicators that enable quantitative analysis of metabolism in various parts of the human body by enabling users to quickly and universally segment and easily designate and modify areas semi-automatically by supporting easy segmentation and quantification of various areas of interest through fixed artificial intelligence and conversational artificial intelligence.

[0012] The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0013] In order to achieve the above object, a first aspect of the present invention is a method for automatically analyzing metabolic health imaging indices performed in an automatic metabolic health imaging indices analysis device, comprising the steps of: receiving positron emission tomography (PET) images and computed tomography (CT) images; identifying a region of interest in the CT images through a fixed artificial intelligence model; adjusting the region of interest for the identified region of interest through an interactive artificial intelligence model; and analyzing the image indices through analysis of PET images and CT images for the adjusted region of interest.

[0014] Preferably, the fixed artificial intelligence model and the conversational artificial intelligence model may correspond to a deep learning model based on a convolutional neural network (CNN) or a vision transformer network (ViT).

[0015] Preferably, the step of identifying the region of interest may include the step of extracting a region of interest for fat, muscle tissue, or tissue related to metabolic health within the CT image, and the step of visualizing the extracted region of interest.

[0016] Preferably, the step of extracting the region of interest may include a step of segmenting the entire target region based on a 3D U-Net deep learning model in the CT image, a step of extracting a CT slice corresponding to the region of interest for the target region, and a step of automatically segmenting muscle, visceral fat, or subcutaneous fat in the CT slice based on a 2D U-Net deep learning model to extract the region of interest.

[0017] Preferably, the step of extracting the CT slice may include the step of instance-segmenting a region of interest for the target region, and the step of automatically extracting a CT slice corresponding to a region designated by a user among the instance-segmented regions.

[0018] Preferably, the step of adjusting the region of interest may include a step of receiving a region adjustment command from a user through a user interface for the segmented region of interest, a step of distinguishing a positive region and a negative region in the segmented region of interest according to the region adjustment command, and a step of adjusting and editing the segmented region of interest in real time based on the positive region and the negative region.

[0019] Preferably, the step of receiving the region control command may include a step of receiving a selection for a positive region or a selection for a negative region from the user through the user interface.

[0020] Preferably, the step of analyzing the image indicator may include the step of combining the adjusted region of interest with the PET image, the step of analyzing metabolic activity based on the PET image and CT image, and the step of implementing the analyzed metabolic activity as a 3D model.

[0021] Preferably, the step of analyzing the metabolic activity may include a step of calculating quantitative values ​​for the region of interest for each level of the PET image and CT image, and a step of diagnosing a metabolic disease based on the quantitative values ​​and previously stored health checkup values.

[0022] Preferably, the step of calculating the quantitative value may include a step of calculating the number of voxels, volume, or surface area for a region of interest for each level of the CT image, a step of calculating an average value, minimum value, maximum value, or peak value for a standardized uptake value for a region of interest for each level of the PET image, a step of calculating the total amount of glucose target activity based on the average value for the standardized uptake value and the volume of the region of interest, and a step of calculating the uptake non-uniformity of glucose through analysis of the variance for the standardized uptake value or texture pixels for the region of interest.

[0023] In order to achieve the above object, the second aspect of the present invention is a metabolic health image index automatic analysis device, comprising: an image receiving unit that receives positron emission tomography (PET) images and computed tomography (CT) images; a region of interest identification unit that identifies a region of interest in the CT images through a fixed artificial intelligence model; a region of interest adjustment unit that adjusts the region of interest through an interactive artificial intelligence model for the identified region of interest; and an image index analysis unit that analyzes image indexes through analysis of PET images and CT images for the adjusted region of interest.

[0024] Preferably, the fixed artificial intelligence model and the conversational artificial intelligence model may correspond to a deep learning model based on a convolutional neural network (CNN) or a vision transformer network (ViT).

[0025] Preferably, the region of interest identification unit can extract a region of interest for fat, muscle tissue, or tissue related to metabolic health within the CT image, and visualize the extracted region of interest.

[0026] Preferably, the region of interest adjustment unit can receive a region adjustment command from a user through a user interface for the segmented region of interest, distinguish a positive region and a negative region in the segmented region of interest according to the region adjustment command, and adjust and edit the segmented region of interest in real time based on the positive region and the negative region.

[0027] Preferably, the image index analysis unit can combine the adjusted region of interest with the PET image, analyze metabolic activity based on the PET image and CT image, and implement the analyzed metabolic activity as a 3D model.

[0028] A third aspect of the present invention for achieving the above purpose is characterized in that, in a computer program stored in a computer-readable medium, when a command of the computer program is executed, a method for automatically analyzing metabolic health image indicators is performed.

[0029] As described above, according to the present invention, the conversational artificial intelligence has the effect of increasing flexibility by being able to subdivide various areas and easily adjust and update the subdivided areas according to the user's needs.

[0030] In addition, it is possible to derive metabolic health imaging indices that reduce evaluation time and have high accuracy at the same time, and the derived metabolic health imaging indices can be used as next-generation metabolic health evaluation indices that directly reflect etiology to supplement inaccurate indirect metabolic health evaluation indices such as existing body mass index (BMI) or waist circumference.

[0031] In addition, it can be used to evaluate the efficacy and predict the prognosis of blockbuster drugs that improve obesity and metabolic health with high market penetration, such as Saxenda or Wigovi, and it has the effect of being used to evaluate the efficacy of similar new drug candidates.

[0032] FIG. 1 is a block diagram of an automatic metabolic health imaging index analysis device according to a preferred embodiment of the present invention.

[0033] Figure 2 is a flowchart of a method for automatically analyzing metabolic health imaging indicators according to one embodiment.

[0034] FIG. 3 is an exemplary diagram illustrating a method for automatically analyzing metabolic health imaging indicators according to one embodiment.

[0035] Figures 4 to 7 are exemplary diagrams illustrating a method for identifying a region of interest according to one embodiment.

[0036] Figures 5 to 10 are exemplary diagrams illustrating a method for adjusting a region of interest according to one embodiment.

[0037] Figure 11 is an exemplary diagram illustrating a method for analyzing image indicators according to one embodiment.

[0038] Hereinafter, the advantages and features of the present invention, and the methods for achieving them, will be clarified by referring to the embodiments described in detail below together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms. These embodiments are provided only to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Like reference numerals refer to like elements throughout the specification. "And / or" includes each and every combination of one or more of the mentioned items.

[0039] Although the terms first, second, etc. are used to describe various elements, components, and / or sections, these elements, components, and / or sections are not limited by these terms. These terms are only used to distinguish one element, component, or section from other elements, components, or sections. Accordingly, it should be understood that a first element, a first component, or a first section referred to below may also be a second element, a second component, or a second section within the technical spirit of the present invention.

[0040] Additionally, the identifiers (e.g., a, b, c, etc.) for each step are used for convenience of explanation and do not describe the order of each step. Each step may occur in a different order than stated unless the context clearly indicates a specific order. That is, each step may occur in the same order as stated, may be performed substantially simultaneously, or may be performed in the opposite order.

[0041] The terminology used herein is for the purpose of describing embodiments only and is not intended to be limiting of the present invention. In this specification, the singular also includes the plural unless the context clearly dictates otherwise. As used herein, the terms “comprises” and / or “comprising” do not exclude the presence or addition of one or more other components, steps, operations, and / or elements.

[0042] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in their common sense to those of ordinary skill in the art to which the present invention pertains. Furthermore, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0043] Furthermore, when describing embodiments of the present invention, detailed descriptions of known functions or configurations will be omitted if they are deemed to unnecessarily obscure the gist of the present invention. Furthermore, the terms described below are defined based on their functions in the embodiments of the present invention and may vary depending on the intent or custom of the user or operator. Therefore, their definitions should be based on the overall content of this specification.

[0044] Figure 1 is a block diagram showing an automatic metabolic health imaging index analysis device according to a preferred embodiment of the present invention.

[0045] The automatic metabolic health imaging index analysis device (100) (hereinafter referred to as the “analysis device”) is a device for performing the automatic metabolic health imaging index analysis method according to the present invention, and automatically segments and identifies regions of interest using a deep learning algorithm based on positron emission tomography (PET) images and computed tomography (CT) images captured by a PET / CT scanner, thereby quantitatively analyzing metabolism of various parts of the human body. Preferably, the deep learning model used in the analysis device (100) is composed of a fixed artificial intelligence model and an interactive artificial intelligence model, and the deep learning model may be based on a convolutional neural network (CNN) and a vision transformer network (ViT), which are optimized for extracting and classifying features from images. The analysis device (100) first trains the deep learning model using training data, and applies the trained model to new PET and CT images to automatically identify regions of interest and quantitatively analyze metabolic activity.

[0046] Preferably, the automatic metabolic health imaging index analysis device (100) is a computer that can install and execute an application or program for performing an automatic metabolic health imaging index analysis method, and has a user interface so that data input and output can be controlled. Here, a computer refers to all types of hardware devices including at least one processor, and may be understood to encompass software configurations operating on the hardware device according to an embodiment. For example, a computer may be understood to encompass, but is not limited to, a smartphone, a tablet PC, a desktop, a laptop, and all user clients and applications running on each device.

[0047] Referring to FIG. 1, the automatic metabolic health image index analysis device (100) includes an image receiving unit (110), an area of ​​interest identification unit (120), an area of ​​interest adjustment unit (130), an image index analysis unit (140), and a control unit (150). Here, the control unit (150) controls the operations and data flow of the image receiving unit (110), the area of ​​interest identification unit (120), the area of ​​interest adjustment unit (130), and the image index analysis unit (140).

[0048] The image receiving unit (110) receives PET and CT images captured from a PET / CT scanner. Preferably, the image receiving unit (110) is connected to the PET / CT scanner and can receive image data directly from the PET / CT scanner or receive image data from a separate storage device that stores the image data.

[0049] The region of interest identification unit (120) uniformly segments the region desired by the user (e.g., subcutaneous fat, visceral fat, or muscle, etc.) based on CT images among image data through a fixed artificial intelligence model to identify the region of interest. Preferably, the fixed artificial intelligence model has a UNETR structure and may include a feature extraction unit (or encoder) composed of a ViT (Vision Transformer) and a decoder composed of a convolutional neural network structure.

[0050] The region of interest adjustment unit (130) automatically adjusts the identified region of interest based on a region adjustment command input by the user through a user interface based on an interactive artificial intelligence model. Preferably, the user can adjust the region of interest using a positive point, a negative point, a region of interest box (ROI box), or a region adjustment command.

[0051] The image index analysis unit (140) combines the adjusted area of ​​interest with the PET image, derives quantitative indices through PET image analysis, and diagnoses detailed metabolic diseases of an organ by utilizing the derived quantitative indices.

[0052] The operations performed through each component of the automatic metabolic health imaging indicator analysis device (100) illustrated in FIG. 1 will be described in detail below with reference to FIG. 2. While each step to be described with reference to FIG. 2 is described as being performed by different components, this is not limited thereto, and depending on the embodiment, at least some of the steps may be performed in the same or different components.

[0053] Figure 2 is a flowchart illustrating a method for automatically analyzing metabolic health imaging indicators according to one embodiment.

[0054] Referring to Fig. 2, the image receiving unit (110) receives PET images and CT images (step S210). Preferably, the user can upload medical image files corresponding to PET and CT images to the analysis device (100) using a user interface connected to the analysis device (100) and select a desired image. For example, the PET images and CT images received through the image receiving unit (110) are as shown in Fig. 3 (a) and (b). Referring to Fig. 3 (a), as a PET image, an area with high metabolic activity in the patient's body is displayed brighter, thereby enabling identification of tissues or lesions with active glucose metabolism, and referring to Fig. 3 (b), as a CT image, it is taken under the same circumstances as PET imaging, clearly represents the morphological structure of the body, and by using this, it is possible to automatically identify and distinguish areas of interest such as specific tissues or areas (e.g., visceral fat and muscles, etc.).

[0055] The region of interest identification unit (120) identifies the region of interest within the CT image using a fixed artificial intelligence model (step S220). More specifically, the region of interest identification unit (120) segments the entire target region from the CT image based on the 3D U-Net deep learning model. Referring to FIG. 4, when a CT image as shown in (a) of FIG. 4 is extracted, a target region (e.g., spine region) to be analyzed is determined by the user through the user interface, and as indicated in yellow in (b) of FIG. 4, the region of interest identification unit (120) segments the entire target region based on the 3D U-Net deep learning model. Preferably, the user can input the target region through the user interface, for example, input L4 corresponding to the spine number as text.

[0056] Next, the region of interest identification unit (120) extracts a CT slice corresponding to the region of interest for the target region. More specifically, the region of interest identification unit (120) can segment the region of interest for the target region into instances and automatically extract a CT slice corresponding to a region specified by the user among the instance segmented regions. Preferably, the region of interest identification unit (120) can set the region of interest for the target region by considering an error that occurs during instance segmentation or can set the region of interest for the target region through interactive artificial intelligence. For example, if the target region is a vertebra, since an error in instance segmentation occurs at both ends of the vertebra, only the middle part of the vertebra can be set as the region of interest and instance segmentation can proceed. Referring to FIG. 5, the region of interest identification unit (120) divides the region of interest (ROI) for the spine vertebra into T12, L1, L2, L3, L4, and L5 regions as indicated by the colors in (a), and when the user designates a desired region through a user interface in the instance-divided region as in (b), the region of interest identification unit (120) can automatically extract a CT slice corresponding to the region designated by the user.

[0057] Preferably, the region of interest identification unit (120) can automatically segment muscle, visceral fat, or subcutaneous fat from a CT slice based on a 2D U-Net deep learning model and extract the region of interest. For example, referring to FIG. 6, the region of interest identification unit (120) can automatically segment and extract muscle, visceral fat, and subcutaneous fat as indicated by color in (b) by applying a 2D U-Net deep learning model to a CT slice designated by a user as shown in (a).

[0058] Preferably, the region of interest identification unit (120) can visualize the extracted region of interest. The region of interest identification unit (120) applies the region of interest extracted based on the CT slice specified by the user to each instance-segmented CT slice, thereby extracting and visualizing the region of interest from each CT slice. For example, referring to FIG. 7, the region of interest extracted from each CT slice corresponding to T12, L1, L2, L3, L4, and L5 instance-segmented vertebrae can be visualized and provided. That is, based on a fixed artificial intelligence model, fat, muscle tissue, or tissues related to metabolic health in the CT image are automatically identified, and each identified region is finally color-coded and displayed to the user.

[0059] The region of interest adjustment unit (130) adjusts the region of interest using an interactive artificial intelligence model for the identified region of interest (step S230). That is, the region of interest identified in step S220 is distinguished from the positive region (region of interest) and the negative region (region of non-interest) through an interactive interface that allows the user to adjust the region of interest, thereby enabling adjustment and detailed editing.

[0060] More specifically, the region of interest adjustment unit (130) receives a region adjustment command from the user through a user interface for the region of interest segmented by the region of interest identification unit (120). Preferably, the region of interest adjustment unit (130) can receive a selection of a positive region or a negative region from the user through the user interface. For example, referring to FIG. 8, the user can input a segmentation region adjustment command by using a mouse to click a desired point in the segmented region of interest, a negative region, an unwanted point, click or mark a more necessary area, or set the segmented region of interest to be relatively narrow or wide. That is, in FIG. 8, the region clicked by the user with a red dot is a region to be excluded from the region of interest, and the region clicked by the user with a green dot is a region to be included in the region of interest.

[0061] Next, the region of interest adjustment unit (130) can distinguish positive and negative regions from the segmented region of interest according to the region adjustment command, and can adjust and edit the segmented region of interest in real time based on the positive and negative regions. More specifically, referring to FIG. 9, the region of interest adjustment unit (130) removes noise from the positive and negative regions entered by the user through Gaussian blur through a user interface, encodes them through a click encoder, and connects them with the encoded segmented region of interest, and then applies an interactive artificial intelligence model to obtain the final region of interest. Here, the click encoder encodes the location where the user clicked and passes the information to the network, and after applying a Gaussian blur to the location where the user clicked, encodes it through a stacked network of Convolution-Relu-Convolution-Relu. The region of interest adjusted and edited in real time through the region of interest adjustment unit (130) is as shown in FIG. 10. (a) of Fig. 10 is a region of interest segmented by the region of interest identification unit (120), and (b) is a region of interest adjusted through the region of interest adjustment unit (130). Referring to the part indicated by the red square box, it can be seen that the region of interest has been adjusted through the interactive artificial intelligence model. Specifically, the part indicated by the green arrow is a part that the user selected as a positive region, and it can be seen that it is not included in the region of interest in (a) but is included in the region of interest in (b), and the part indicated by the red arrow is a part that the user selected as a negative region, and it can be seen that it is included in the region of interest in (a) but is not included in the region of interest in (b).

[0062] As a result, as shown in (c) of FIG. 3, a region of interest, which is a specific tissue or area to be analyzed, is acquired through the region of interest identification unit (120) and the region of interest adjustment unit (130). As an example, subcutaneous fat, visceral fat, and the psoas muscles on the left and right are indicated as designated regions of interest, but other regions can also be used universally. In addition, the data on the region of interest finally acquired can be added to the training data of the deep learning algorithm applied to the present invention and used for region of interest extraction training. If a fixed artificial intelligence model is used, it is difficult to correct an inaccurate segmented image, and a lot of time is consumed in the process of manually correcting all segmented images, which may reduce the evaluation accuracy. However, in the present invention, an interactive artificial intelligence model allows an incorrectly segmented part or a part that should be included to be input through a user interface, thereby easily correcting the segmented image, and through this process, the indices in the PET image, which will be described below, can be automatically quantified.

[0063] The image index analysis unit (140) analyzes image indexes through PET image analysis of the adjusted region of interest (step S240). That is, the image index analysis unit (140) analyzes metabolic activity by combining regions of interest identified through fixed and interactive artificial intelligence models with PET images, implements them into a 3D model, and presents them, thereby deriving quantitative indexes.

[0064] Preferably, the image index analysis unit (140) obtains quantitative values ​​corresponding to the number of voxels, volume, and surface area through fixed artificial intelligence based on the CT image, and combines the region of interest adjusted by the region of interest adjustment unit (130) with the PET image, and then obtains quantitative values ​​corresponding to the mean value (SUV_mean), minimum value (SUV_min), maximum value (SUV_max), peak value (SUV_peak), total amount of glucose metabolic activity, and glucose absorption non-uniformity for the standard uptake value (SUV) in the region of interest designated by the user based on the PET image. The quantitative values ​​obtained through the PET image are very sensitive to the selected region of interest and have a misalignment problem between the PET and CT, so they are edited in detail, but are performed through simple interactive artificial intelligence. More specifically, referring to FIG. 11, the image index analysis unit (140) combines a region of interest, such as subcutaneous fat, visceral fat, or muscle, of each CT slice with a PET image. Then, the image index analysis unit (140) analyzes metabolic activity based on the PET image with the combined region of interest. Referring to FIGS. 3 and 11, quantitative indices that can be derived from PET image analysis in the region of interest and quantitative indices that can be derived from CT image analysis are listed, and the corresponding indices can be utilized as quantitative indices related to specific organ metabolic diseases. The image index analysis unit (140) analyzes the PET image and CT image for the region of interest to calculate quantitative values ​​and diagnoses metabolic diseases based on the quantitative values ​​and previously stored health checkup values. Here, the health checkup values ​​correspond to age, height, weight, waist circumference, neutral fat, HDL, or blood pressure. Each of the indices obtained by the image index analysis unit (140) and the calculation method are as follows.

[0065] 1) Number of voxels: The number of voxels in the region of interest. A voxel is a unit particle in a 3D image, corresponding to small cubes that make up the volume of the region.

[0066] 2) Volume: Total volume of the region of interest, usually calculated by multiplying the number of voxels by the size of each voxel.

[0067] 3) Surface area: The surface area of ​​the area of ​​interest, calculated by quantitatively estimating the boundary area of ​​the tissue where metabolic activity occurs.

[0068] 4) SUV_mean (Standard Uptake Value - Mean): The average uptake of radiolabeled glucose within the region of interest, corresponding to the average intensity of metabolic activity.

[0069] 5) SUV_min (Standard Uptake Value - Minimum): Corresponds to the minimum uptake of radiolabeled glucose observed within the region of interest.

[0070] 6) SUV_max (Standard Uptake Value - Maximum): The maximum uptake of radiolabeled glucose observed within the region of interest, indicating the point of highest metabolic activity.

[0071] 7) SUV_peak (Standard Uptake Value - Peak): Corresponds to the maximum SUV value measured in a small area within the region of interest.

[0072] 8) Total Lesion Glycolysis (TLG): The total amount of glucose metabolic activity across the entire region of interest, calculated by multiplying SUV_mean by the volume of the region and reflecting the overall activity of metabolic activity.

[0073] 9) Heterogeneity: The degree of non-uniformity in the absorption of radiolabeled glucose observed within the region of interest, which can be calculated through the coefficient of variation of SUV or advanced texture pixel analysis.

[0074] Preferably, the image index analysis unit (140) can obtain analysis results in the form of a table as shown in FIG. 11 or in the form of [Table 1] below, and can implement the analyzed metabolic activity as a 3D model.

[0075] #VoxelVolume(cm 3 )SurfaceArea(cm 2 )SUV mean SUV min SUV max SUV peak TLGSUV cov SpinelevelPsoas(L)150001202502.51.04.53.830015.5L4Psoas(R)145001182452.41.14.33.7283.228.7L4Visceralfat3500030060 01.50.82.82.045045.9L5Spleen200001502803.01.21.54.245025.4L1BoneMarrow250002004002.00.93.53.04000.3L5Subcutaneous fat300002805500.90.41.81.22520.2L4

[0076] The present invention may reside on any form of computer-readable recording medium well known in the art.

[0077] The components of the present invention may be implemented as programs (or applications) and stored on a medium to be executed in conjunction with a computer, which is hardware. The components of the present invention may be implemented as software programs or software elements. Similarly, the embodiments may be implemented in a programming or scripting language such as C, C++, Java, or an assembler, including various algorithms implemented as a combination of data structures, processes, routines, or other programming components. Functional aspects may be implemented as algorithms that are executed on one or more processors.

[0078] Although the preferred embodiments of the automatic analysis method and device for metabolic health imaging indicators according to the present invention have been described above, the present invention is not limited thereto, and various modifications can be made within the scope of the patent claims, the detailed description of the invention, and the attached drawings, and these also belong to the present invention.

[0079] [Explanation of symbols]

[0080] 100: Automatic analysis device for metabolic health imaging indicators

[0081] 110: Video receiver

[0082] 120: Area of ​​Interest Identification Section

[0083] 130: Area of ​​Interest Adjustment Section

[0084] 140: Video Indicator Analysis Department

[0085] 150: Control unit

Claims

1. In a method for automatically analyzing metabolic health image indicators performed in an automatic metabolic health image indicator analysis device, A step of receiving positron emission tomography (PET) images and computed tomography (CT) images; A step of identifying an area of ​​interest using a fixed artificial intelligence model within the above CT image; A step of adjusting the area of ​​interest through an interactive artificial intelligence model for the above-mentioned identified area of ​​interest; and A method for automatically analyzing metabolic health imaging indices, comprising a step of analyzing imaging indices through PET image and CT image analysis for the above-described adjusted region of interest.

2. In paragraph 1, the fixed artificial intelligence model and the conversational artificial intelligence model, A method for automatically analyzing metabolic health imaging indicators, characterized by a deep learning model based on a convolutional neural network (CNN) or a vision transformer network (ViT).

3. In the first paragraph, the step of identifying the area of ​​interest comprises: A step of extracting a region of interest for fat, muscle tissue, or tissue related to metabolic health within the CT image; and A method for automatically analyzing metabolic health imaging indicators, characterized by including a step of visualizing the extracted region of interest.

4. In the third paragraph, the step of extracting the region of interest comprises: A step of segmenting the entire target area based on a 3D U-Net deep learning model in the above CT image; A step of extracting a CT slice corresponding to the region of interest for the above target area; and A method for automatically analyzing metabolic health imaging indicators, characterized by including a step of automatically segmenting muscles, visceral fat, or subcutaneous fat based on a 2D U-Net deep learning model in the CT slice and extracting a region of interest.

5. In the fourth paragraph, the step of extracting the CT slice is: a step of instance segmenting a region of interest for the target region; and An automatic analysis method for metabolic health imaging indicators, characterized by including a step of automatically extracting a CT slice corresponding to an area designated by a user among the above instance-divided areas.

6. In the first paragraph, the step of adjusting the area of ​​interest comprises: A step of receiving an area control command from a user through a user interface for the above-detailed area of ​​interest; A step of distinguishing a positive area and a negative area in the segmented area of ​​interest according to the above area control command; and A method for automatically analyzing metabolic health imaging indicators, characterized by including a step of adjusting and editing the segmented region of interest in real time based on the positive and negative regions.

7. In the 6th paragraph, the step of receiving the area control command is: A method for automatically analyzing metabolic health imaging indicators, characterized in that it includes a step of receiving a selection for a positive area or a selection for a negative area from a user through the user interface.

8. In the first paragraph, the step of analyzing the image indicator comprises: A step of combining the adjusted region of interest with the PET image; A step of analyzing metabolic activity based on the PET image and CT image; and A method for automatically analyzing metabolic health imaging indicators, characterized by including a step of implementing the analyzed metabolic activity as a 3D model.

9. In paragraph 8, the step of analyzing the metabolic activity comprises: A step of calculating quantitative values ​​for the region of interest for each level of the PET image and CT image; and A method for automatically analyzing metabolic health imaging indicators, characterized by including a step of diagnosing metabolic diseases based on the above quantitative values ​​and previously stored health checkup values.

10. In paragraph 9, the step of calculating the quantitative value is as follows: A step of calculating the number of voxels, volume, or surface area for a region of interest for each level of the CT image; A step of calculating an average, minimum, maximum, or peak value for standard uptake values ​​for a region of interest for each level of the PET image; A step of calculating the total amount of glucose target activity based on the average value for the above standard intake value and the volume of the above region of interest; and A method for automatically analyzing metabolic health imaging indicators, characterized in that it comprises a step of calculating the absorption non-uniformity of glucose through analysis of the dispersion for the above standard intake value or texture pixels for the above region of interest.

11. Image receiving unit for receiving positron emission tomography (PET) images and computed tomography (CT) images; A region of interest identification unit that identifies a region of interest through a fixed artificial intelligence model within the CT image; An area of ​​interest adjustment unit that adjusts the area of ​​interest using an interactive artificial intelligence model for the above-mentioned identified area of ​​interest; and An automatic metabolic health imaging index analysis device including an imaging index analysis unit that analyzes imaging indexes through PET image and CT image analysis for the above-described adjusted region of interest.

12. In paragraph 11, the fixed artificial intelligence model and the conversational artificial intelligence model, An automatic metabolic health imaging index analysis device characterized by a deep learning model based on a convolutional neural network (CNN) or a vision transformer network (ViT).

13. In paragraph 11, the area of ​​interest identification unit, An automatic metabolic health imaging index analysis device characterized by extracting a region of interest for fat, muscle tissue, or tissue related to metabolic health from the CT image and visualizing the extracted region of interest.

14. In paragraph 11, the area of ​​interest adjustment unit, An automatic analysis device for metabolic health imaging indicators, characterized in that it receives a region adjustment command from a user through a user interface for the above-described segmented region of interest, distinguishes a positive region and a negative region in the above-described segmented region of interest according to the region adjustment command, and adjusts and edits the above-described segmented region of interest in real time based on the positive region and the negative region.

15. In paragraph 11, the image indicator analysis unit, An automatic metabolic health image index analysis device characterized by combining the adjusted region of interest with the PET image, analyzing metabolic activity based on the PET image and CT image, and implementing the analyzed metabolic activity as a 3D model.

16. In a computer program stored in a computer-readable medium, A computer program stored in a computer-readable medium, characterized in that when the command of the above computer program is executed, a method according to any one of claims 1 to 10 is performed.

Images