Artificial intelligence-based device and method for determining amyloid pet positivity

Abstract

The present invention relates to an artificial intelligence-based device and method for determining amyloid PET positivity, wherein a normal-appearing white matter dominance ratio is used and amyloid PET positivity is automatically determined based on machine learning, thereby improving the accuracy and sensitivity of amyloid PET positivity determination. The device comprises: an image acquisition unit for acquiring PET and MRI images of the brain of a diagnostic subject administered with a contrast agent, the images being acquired through tomographic scanning; a segmentation unit for segmenting normal-appearing white matter (NAWM) and white matter hyperintensity (WMH) from a fluid attenuated inversion recovery (FLAIR) image obtained from the MRI image; a standardized uptake value ratio (SUVR) calculation unit for calculating a global standardized uptake value ratio (SUVR) and a standardized uptake value ratio in normal-appearing white matter (NAWM SUVR), on the basis of the PET image and the NAWM region segmented by the segmentation unit; and an amyloid PET positivity determination unit for determining amyloid PET positivity by inputting the global SUVR and the NAWM dominance ratio (NDR) into a pre-trained prediction model, wherein the NDR is a value obtained by dividing the global SUVR by the NAWM SUVR.

Description

AI-based amyloid PET positive detection device and method

[0001] [Cross-reference to related applications]

[0002] This application claims priority to Republic of Korea Patent Application No. 10-2024-0045813, filed April 4, 2024, the entire contents of which are incorporated herein by reference.

[0003] The present invention relates to an artificial intelligence-based amyloid PET positivity determination device and method, and more particularly, to an artificial intelligence-based amyloid PET positivity determination device and method that can automatically perform amyloid PET positivity determination using a white matter dominance that appears normal.

[0004] Amyloid positron emission tomography (PET) diagnosis is a method of determining whether abnormal forms of amyloid, which are created when abnormalities occur in the metabolic process of proteins, are deposited in the brain by injecting a special contrast agent that reacts only to amyloid into the body and then performing a PET scan to determine whether the result is positive or negative depending on the presence of deposits.

[0005] These amyloid PET positivity results are typically determined by visual rating by a specialist or by using a cutoff for the Standardized Uptake Value Ratio (SUVR) based on cortical amyloid uptake through additional calculations.

[0006] However, when performing a positive amyloid PET diagnosis through a visual evaluation by a specialist, the decision takes a considerable amount of time, is dependent on subjective judgment, may have low accuracy and reproducibility, and requires an experienced specialist.

[0007] In addition, when performing amyloid PET positive judgment using the SUVR threshold, there is a problem that although amyloid PET positive and negative judgments are distinguished according to a specific threshold setting, it may not sufficiently reflect fine amyloid deposition or inter-individual variations.

[0008] The present invention has been devised to solve the above-mentioned conventional problems, and the purpose of the present invention is to provide an artificial intelligence-based amyloid PET positive judgment device and method that can increase the accuracy and sensitivity of amyloid PET positive judgment by automatically performing amyloid PET positive judgment based on machine learning and utilizing a normal-appearing white matter dominance ratio.

[0009] In order to achieve the above-described object, the artificial intelligence-based amyloid PET positive determination device according to the present invention comprises: an image acquisition unit for acquiring a PET image and an MRI image obtained by cross-sectionally scanning the brain of a diagnosis subject administered with a contrast agent; a segmentation unit for segmenting normal white matter (Normal-Appearing White Matter, NAWM) and white matter hyperintensity (White Matter Hyperintensity, WMH) from a FLAIR (Fluid Attenuated Inversion Recovery) image acquired from the MRI image; a standardized uptake value ratio calculation unit for calculating a global standardized uptake value ratio (SUVR) and a standardized uptake value ratio (NAWM SUVR) in normal white matter based on the PET image and the normal white matter (NAWM) region segmented in the segmentation unit; And an amyloid PET positivity judgment unit that inputs the global standardized uptake value ratio (SUVR) and NDR (NAWM Dominance Ratio) into the learned prediction model to determine amyloid PET positivity, wherein NDR is a value obtained by dividing the global standardized uptake value ratio (SUVR) by the standardized uptake value ratio (NAWM SUVR) in normal white matter; characterized in that it includes;

[0010] In addition, the artificial intelligence-based amyloid PET positive determination device according to the present invention is characterized by further including an MRI preprocessing unit that preprocesses the MRI image acquired by the image acquisition unit to divide the entire brain structure into a white matter brain region and a gray matter brain region.

[0011] In addition, the artificial intelligence-based amyloid PET positive judgment device according to the present invention is characterized by further including a learning unit that learns the prediction model based on the global standardized absorption value ratio (SUVR) and NDR calculated by the standardized absorption value ratio calculation unit.

[0012] In addition, in the artificial intelligence-based amyloid PET positive judgment device according to the present invention, the MRI preprocessing unit is characterized by dividing the entire brain structure into a white matter brain region and a gray matter brain region using a T1-weighted image.

[0013] In addition, in the artificial intelligence-based amyloid PET positive judgment device according to the present invention, the global standardized uptake value ratio (SUVR) is characterized in that it is calculated as a ratio of the average standardized uptake value ratio (SUVR) of multiple regions of interest to the entire cerebellum reference region.

[0014] In addition, in the artificial intelligence-based amyloid PET positive judgment device according to the present invention, the region of interest is characterized by including at least one of the frontal, parietal, lateral temporal, posterior and anterior cingulate cortex regions.

[0015] In addition, in the artificial intelligence-based amyloid PET positive judgment device according to the present invention, the prediction model is characterized in that it is implemented as a machine learning-based Gaussian mixture model.

[0016] In addition, the artificial intelligence-based amyloid PET positive determination method according to the present invention for achieving the above-described purpose comprises: an image acquisition step of acquiring a PET image and an MRI image obtained by cross-sectionally scanning the brain of a diagnosis subject administered with a contrast agent; a segmentation step of segmenting normal white matter (Normal-Appearing White Matter, NAWM) and white matter hyperintensity signal (White Matter Hyperintensity, WMH) in a FLAIR (Fluid Attenuated Inversion Recovery) image acquired from the MRI image; a standardized uptake value ratio calculation step of calculating a global standardized uptake value ratio (SUVR) and a standardized uptake value ratio (NAWM SUVR) in normal white matter based on the PET image and the normal white matter (NAWM) region segmented in the segmented section; And an amyloid PET positivity judgment step of inputting the global standardized uptake value ratio (SUVR) and NDR (NAWM Dominance Ratio) into the learned prediction model to determine amyloid PET positivity, wherein NDR is a value obtained by dividing the global standardized uptake value ratio (SUVR) by the standardized uptake value ratio in normal white matter (NAWM SUVR);

[0017] In addition, the artificial intelligence-based amyloid PET positive determination method according to the present invention is characterized by further including an MRI preprocessing step of preprocessing the MRI image acquired in the image acquisition step to divide the entire brain structure into a white matter brain region and a gray matter brain region.

[0018] In addition, the method for determining positive amyloid PET using artificial intelligence according to the present invention is characterized by further including a learning step for learning the prediction model based on the global standardized absorption value ratio (SUVR) and NDR calculated through the standardized absorption value ratio calculation step.

[0019] In addition, in the artificial intelligence-based amyloid PET positive judgment method according to the present invention, the MRI preprocessing step is characterized by being a step of dividing the entire brain structure into a white matter brain region and a gray matter brain region using a T1-weighted image.

[0020] In addition, in the artificial intelligence-based amyloid PET positive judgment method according to the present invention, the global standardized uptake value ratio (SUVR) is characterized in that it is calculated as a ratio of the average standardized uptake value ratio (SUVR) of multiple regions of interest to the entire cerebellum reference region.

[0021] In addition, in the artificial intelligence-based amyloid PET positive judgment method according to the present invention, the region of interest is characterized in that it includes at least one of the frontal, parietal, lateral temporal, posterior and anterior cingulate cortex regions.

[0022] In addition, in the artificial intelligence-based amyloid PET positive judgment method according to the present invention, the prediction model is characterized in that it is implemented as a machine learning-based Gaussian mixture model.

[0023] Specific details of other embodiments are included in the “Specific Details for Carrying Out the Invention” and the attached “Drawings.”

[0024] The advantages and / or features of the present invention and the methods for achieving them will become clear with reference to the various embodiments described in detail below together with the accompanying drawings.

[0025] However, the present invention is not limited to the configuration of each embodiment disclosed below, but may be implemented in various different forms, and each embodiment disclosed in this specification is provided only to ensure that the disclosure of the present invention is complete and to fully inform a person having ordinary skill in the art to which the present invention pertains of the scope of the present invention, and it should be understood that the present invention is defined only by the scope of each claim of the claims.

[0026] According to the present invention, by utilizing a normal-appearing white matter dominance and automatically performing amyloid PET positivity determination based on machine learning, the accuracy and sensitivity of amyloid PET positivity determination can be increased, thereby enabling early diagnosis and identification of even minute amyloid accumulation.

[0027] Additionally, since it is determined through objective data analysis rather than relying on visual assessment, subjective errors can be minimized, and clinical progression can be significantly predicted, which can help in establishing treatment and management strategies.

[0028] FIG. 1 is a schematic diagram showing the configuration of an artificial intelligence-based amyloid PET positive determination device according to one embodiment of the present invention.

[0029] FIG. 2 is a processing diagram for explaining an artificial intelligence-based amyloid PET positive determination method according to one embodiment of the present invention.

[0030] FIG. 3 is a diagram exemplarily showing the results of a multivariate regression analysis on the correlation between the amyloid PET positive determination method according to the present invention and the prior art and clinical progress.

[0031] FIG. 4 is a diagram exemplifying a confusion matrix for comparing and evaluating the amyloid PET positive detection performance of the present invention and the prior art.

[0032] Figure 5 is a drawing showing an example of a scatter plot of amyloid PET positivity judgments for the present invention and the prior art.

[0033] Figure 6 is an exemplary diagram showing a clinical progression scatter plot of the discordant group and an amyloid PET positivity judgment scatter plot for three methods of amyloid PET positivity judgment.

[0034] Before describing the present invention in detail, it should be understood that the terms or words used in this specification should not be interpreted as being unconditionally limited to their usual or dictionary meanings, and that the inventor of the present invention may appropriately define and use the concepts of various terms in order to explain his or her invention in the best possible manner, and further, that these terms or words should be interpreted as meanings and concepts that are consistent with the technical idea of ​​the present invention.

[0035] That is, it should be noted that the terms used in this specification are only used to describe preferred embodiments of the present invention, and are not intended to specifically limit the contents of the present invention, and that these terms are defined in consideration of various possibilities of the present invention.

[0036] Additionally, in this specification, it should be noted that singular expressions may include plural expressions unless the context clearly indicates a different meaning, and similarly, even if expressed in plural, may include a singular meaning.

[0037] Throughout this specification, whenever a component is described as "including" another component, it may mean that the component may further include any other component, rather than excluding any other component, unless specifically stated otherwise.

[0038] Furthermore, when it is described that a component is "located within, connected to, or installed within" another component, it should be understood that the component may be installed in direct connection with or in contact with the other component, may be installed spaced apart from the other component by a certain distance, and if it is installed spaced apart from the other component by a certain distance, there may be a third component or means for fixing or connecting the component to the other component, and the description of this third component or means may be omitted.

[0039] On the other hand, if a component is described as being "directly connected" or "directly connected" to another component, it should be understood that no third component or means exists.

[0040] Likewise, other expressions that describe the relationship between components, such as "between" and "directly between", or "adjacent to" and "directly adjacent to", should be interpreted as having the same meaning.

[0041] In addition, it should be noted that the terms “one side,” “the other side,” “one side,” “the other side,” “first,” “second,” etc. in this specification, if used, are used to clearly distinguish one component from another component, and the meaning of the component is not limited by such terms.

[0042] In addition, terms related to position, such as “upper,” “lower,” “left,” and “right,” etc., in this specification, if used, should be understood to indicate relative positions of the corresponding components in the corresponding drawings, and unless absolute positions are specified for these positions, these position-related terms should not be understood to refer to absolute positions.

[0043] Moreover, in the specification of the present invention, it should be noted that the terms “part”, “device”, “module”, “device”, etc., if used, mean a unit capable of processing one or more functions or operations, which may be implemented by hardware or software, or a combination of hardware and software.

[0044] In addition, in this specification, when specifying the drawing numbers for each component of each drawing, the same component has the same drawing number even if the component is shown in a different drawing, that is, the same reference number indicates the same component throughout the specification.

[0045] In the drawings attached to this specification, the size, position, connection relationship, etc. of each component constituting the present invention may be described with some exaggeration, reduction, or omission in order to sufficiently clearly convey the idea of ​​the present invention or for convenience of explanation, and therefore the proportions or scales may not be strict.

[0046] In addition, in the following description of the present invention, a detailed description of a configuration that is judged to unnecessarily obscure the gist of the present invention, for example, a known technology including a prior art, may be omitted.

[0047]

[0048] Hereinafter, with reference to the attached drawings, an artificial intelligence-based amyloid PET positive determination device and method according to a preferred embodiment of the present invention will be described in detail.

[0049] FIG. 1 is a schematic diagram showing the configuration of an artificial intelligence-based amyloid PET positive determination device according to one embodiment of the present invention.

[0050] As shown in Fig. 1, the artificial intelligence-based amyloid PET positive judgment device (110) according to the present invention may include an image acquisition unit (110), an MRI preprocessing unit (120), a segmentation unit (130), an SUVR calculation unit (140), a learning unit (150), an amyloid PET positive judgment unit (160), etc.

[0051] The image acquisition unit (110) can acquire PET images and MRI (Magnetic Resonance Imaging) images of the brain of a diagnostic subject to whom a contrast agent has been administered.

[0052] The image acquisition unit (110) can acquire a PET image through a PET scanner (not shown) that photographs the brain of a subject to whom a contrast agent (e.g., a radioactive label) that emits positrons is administered, and can acquire an MRI image through an MRI scanner (not shown) that obtains an MRI image by transmitting high-frequency waves to the brain of a subject to be diagnosed in a strong magnetic field and measuring the reflected electromagnetic waves.

[0053] The image acquisition unit (110) acquires PET images through a PET scanner (not shown), for example, 18 After administering F-florbetabe (FBB) to a subject for diagnosis, an image captured by a PET scanner (not shown) can be acquired after a certain absorption period.

[0054] In addition, the image acquisition unit (110) can acquire a high-resolution 3D T1-weighted image and FLAIR (Fluid-Attenuated Inversion Recovery) image through an MRI scanner (not shown).

[0055] Parameters of a T1-weighted image may include the acquired voxel size (spatial resolution of the image), echo time (time between excitation pulses), repetition time (time between two excitation pulses), field of view (FOV) (size of the spatial region covered by the MRI scanner), matrix size (size of the array of pixels or voxels that determines the resolution of the image), number of excitations (NOE) (number of times to repeat the entire scan), flip angle (angle of the RF pulse used in the MRI sequence), inversion time (time between a 180° excitation pulse and a 90° excitation pulse), etc.

[0056] These T1-weighted images can clearly visualize the white matter (WM) and gray matter (GM) of the brain.

[0057] Parameters of a FLAIR image may include echo time (the time at which the signal is read from the MRI sequence), repetition time (the interval at which images are repeatedly acquired), flip angle (the angle of the RF pulse used in the MRI sequence), axial slice thickness (AST, the thickness of each image slice), axial plane matrix size (APMS, the number of pixels in a cross-section), and acquired voxel size (the size of each voxel actually scanned).

[0058] These FLAIR images can be used to distinguish between white matter and non-white matter regions of the brain, and can be useful for detecting white matter hyperintensity (WMH) because they well visualize abnormal signals within white matter and clearly show the boundary between white matter and gray matter.

[0059] The MRI preprocessing unit (120) can preprocess MRI images among the images acquired from the image acquisition unit (110) to divide the entire brain structure into white matter brain regions and gray matter brain regions.

[0060] The MRI preprocessing unit (120) can resample MRI images and FLAIR images into square voxels (e.g., 1×1×1㎣) having the same size in all axes (horizontal, vertical, and height). This is because the voxel sizes of the original images of all diagnosis subjects are not the same.

[0061] Additionally, the MRI preprocessing unit (120) can divide the entire brain structure into white matter brain regions and gray matter brain regions using T1-weighted images.

[0062] Here, the MRI preprocessing unit (120) can segment brain regions based on regions defined in the DKT atlas (Desikan-Killiany-Tourville atlas).

[0063] The segmentation unit (130) can segment normal white matter (Normal-Appearing White Matter, NAWM) and white matter hyperintensity signal (WMH, an area with a problem in white matter) in a FLAIR image acquired from an MRI image.

[0064] Magnetic field inhomogeneity or bias that occurs during the acquisition of FLAIR images can cause non-uniform brightness and contrast in the image.

[0065] Accordingly, the segmentation unit (130) performs bias correction to correct non-uniformity occurring in the FLAIR image, thereby ensuring consistent brightness and contrast of the image and improving the accuracy of image analysis or diagnosis.

[0066] And the segmentation unit (130) can align the previously corrected individual FLAIR images to the original T1-weighted images obtained from the same MRI session as the FLAIR images by performing transformations such as translation and rotation through rigid-body registration with a cost function (used as an index for evaluating the similarity between the registered image and the original image) so that the two images are spatially aligned.

[0067] Thereafter, the segmentation unit (130) can segment the white matter hyperintensity signal (WMH) from the bias-corrected FLAIR image.

[0068] And voxels within the white matter region mask that are not marked by white matter hyperintensities (WMH) can be segmented into normal-appearing white matter, i.e., normal white matter (NAWM).

[0069] The SUVR calculation unit (140) can calculate a global standardized uptake value ratio (SUVR), a normal white matter standardized uptake value ratio (NAWM SUVR), and a WMH SUVR based on the PET image acquired by the image acquisition unit (110) and the normal white matter (NAWM) region and white matter high intensity signal (WMH) region segmented by the segmentation unit (130).

[0070] Global SUVR can be calculated as the ratio of the average SUVR of multiple regions of interest (ROI) to the entire cerebellar reference region.

[0071] Here, the region of interest may comprise at least one of the frontal, parietal, lateral temporal, posterior and anterior cingulate cortex regions.

[0072] In addition, the SUVR calculation unit (140) can calculate the SUVR value of each region by applying the NAWM region mask and WMH region mask divided by the division unit (130) to the PET image.

[0073] For example, applying 4mm FWHM smoothing (which reduces noise in an image by smoothing values ​​around pixels) to white matter regions can remove all voxels in non-white matter regions, preventing partial volume effects that occur when multiple structures are contained within a single pixel.

[0074] The learning unit (150) can learn a prediction model for determining amyloid PET positivity based on the global SUVR and NDR (NAWM Dominance Ratio) calculated by the SUVR calculation unit (140).

[0075] In an embodiment of the present invention, a prediction model for determining amyloid PET positivity may be implemented as a machine learning-based Gaussian Mixture Model (GMM), but is not limited thereto.

[0076] The prediction model applied to the present invention can be modeled through 100 GMM iterations with two mixture components, for example, using global SUVR and NDR, and a positivity threshold can be defined based on the modeled distribution.

[0077] NDR is the ratio between the global SUVR and NAWM SUVR calculated in the SUVR calculation unit (140), and is the value obtained by dividing the global SUVR by the NAWM SUVR.

[0078] The amyloid PET positive judgment unit (160) can determine amyloid PET positive by inputting the global SUVR and NDR calculated by the SUVR calculation unit (140) into the prediction model learned by the learning unit (150).

[0079] FIG. 2 is a processing diagram for explaining an artificial intelligence-based amyloid PET positive determination method according to one embodiment of the present invention.

[0080] An artificial intelligence-based amyloid PET positive determination method according to one embodiment of the present invention can be performed on a configuration substantially identical to that of the artificial intelligence-based amyloid PET positive determination device (110) illustrated in FIG. 1.

[0081] First, in step S10, PET images and MRI images of the brain of the diagnostic subject to whom contrast agent was administered can be acquired.

[0082] Specifically, a PET image can be obtained through a PET scanner (not shown) that photographs the brain of a subject to whom a contrast agent (e.g., a radioactive label) that emits positrons is administered in the above-described step S10, and an MRI image can be obtained through an MRI scanner (not shown) that obtains an MRI image by transmitting high-frequency waves to the brain of the subject to be diagnosed in a strong magnetic field and measuring the reflected electromagnetic waves.

[0083] Thereafter, in step S20, the MRI image obtained through the above-described step S10 is preprocessed to segment the entire brain structure into a white matter brain region and a gray matter brain region.

[0084] Specifically, in the above step S20, the MRI image and the FLAIR image are resampled into square voxels (e.g., 1×1×1㎣) having the same size in all axes (horizontal, vertical, and height), and then the entire brain structure can be segmented into white matter brain regions and gray matter brain regions using the T1-weighted image.

[0085] And in step S30, normal white matter (NAWM) and white matter hyperintensity signal (WMH, a problematic area in white matter) can be segmented from the FLAIR image acquired from the MRI image.

[0086] Specifically, in the above step S30, bias correction is performed to correct non-uniformity occurring in the FLAIR image, and transformations such as translation and rotation are performed on the bias-corrected individual FLAIR image through rigid body registration to align the bias-corrected individual FLAIR image and the original T1-weighted image so that they are spatially consistent, and then white matter hyperintensity signals (WMH) can be segmented from the bias-corrected FLAIR image.

[0087] And voxels within the white matter region mask that are not marked by white matter hyperintensities (WMH) can be segmented into normal-appearing white matter, i.e., normal white matter (NAWM).

[0088] In step S40, global SUVR, NAWM SUVR, and WMH SUVR can be calculated based on the PET image acquired through the above-described step S10 and the normal white matter (NAWM) region and white matter high intensity signal (WMH) region segmented in the above-described step S30.

[0089] In the above step S40, the global SUVR can be calculated as the ratio of the average SUVR of multiple regions of interest (ROIs) to the entire cerebellum reference region.

[0090] Here, the region of interest may comprise at least one of the frontal, parietal, lateral temporal, posterior and anterior cingulate cortex regions.

[0091] In addition, in the above-described step S40, the NAWM SUVR and WMH SUVR can calculate the SUVR value of each region by applying the NAWM region mask and WMH region mask segmented through the above-described step S30 to the PET image.

[0092] Thereafter, in step S50, the global SUVR and NDR calculated through step S40 can be input into the pre-learned prediction model to determine amyloid PET positivity.

[0093] The prediction model applied to the present invention can be implemented as a machine learning-based Gaussian mixture model (GMM), and can be learned in advance based on global SUVR and NDR.

[0094] FIG. 3 is a diagram exemplarily showing the results of a multivariate regression analysis on the correlation between the amyloid PET positivity determination method according to the present invention and the prior art and clinical progression. In order to determine which method among three methods of determining amyloid PET positivity, that is, a method of determining amyloid PET positivity using a prediction model according to the present invention, a method of determining amyloid PET positivity based on visual evaluation according to the prior art, and a method of determining amyloid PET positivity based on SUVR threshold, has a significant correlation with clinical progression, a multivariate regression analysis was performed while adjusting for age, gender, and education. As a result, it was confirmed that the amyloid PET positivity determination method according to the present invention has a significant correlation between the positivity determination and clinical progression compared to the prior art.

[0095] FIG. 4 is a diagram exemplarily showing a confusion matrix for comparing and evaluating the amyloid PET positivity determination performance of the present invention and the prior art. It can be confirmed that the amyloid PET positivity determination method according to the present invention has sensitivity, specificity, accuracy, and Youden's index of 0.988, 0.897, 0.936, and 0.885, respectively, and that the method for performing amyloid PET positivity determination based on the conventional SUVR threshold shows similar values ​​of 0.974, 0.916, 0.941, and 0.890 for these indices, respectively.

[0096] In addition, when the visual evaluation was used as the ground truth, the AUC (Area Under Curve) of the amyloid PET positivity determination method according to the present invention was 0.943, and the AUC of the method for performing amyloid PET positivity determination based on the conventional SUVR threshold was 0.945 (p=0.818), confirming that the two methods exhibit similar performance.

[0097] FIG. 5 is a drawing showing an example of a scatter plot of amyloid PET positivity judgments for the present invention and the prior art, and FIG. 6 is a drawing showing an example of a scatter plot of clinical progression of a mismatch group and an example of a scatter plot of amyloid PET positivity judgments for three types of amyloid PET positivity judgment methods, and it can be confirmed that the group judged positive by the amyloid PET positivity judgment method according to the present invention has significantly more clinical progression than the group judged negative.

[0098] In this way, according to the present invention, not only does it exhibit amyloid PET positivity judgment performance similar to a conventional method of performing amyloid PET positivity judgment based on an SUVR threshold, but it also enables more significant prediction of clinical progress.

[0099] Compared to conventional methods of performing amyloid PET positivity judgment based on SUVR thresholds, the accuracy and sensitivity are higher, enabling early diagnosis and identification of even minute amyloid accumulations.

[0100] Additionally, because judgment is made through objective data analysis rather than relying on visual evaluation, subjective errors can be minimized.

[0101] Additionally, as clinical progression can be significantly predicted, it can help in establishing treatment and management strategies.

[0102] Above, although some examples have been given and various preferred embodiments of the present invention have been described, the description of the various embodiments described in the “Specific Details for Carrying Out the Invention” section is merely exemplary, and those skilled in the art to which the present invention pertains will readily understand that they can carry out various modifications of the present invention or carry out equivalent implementations of the present invention based on the above description.

[0103] In addition, since the present invention can be implemented in various other forms, the present invention is not limited by the above description, and the above description is provided only to make the disclosure of the present invention complete and to fully inform a person having ordinary skill in the art to which the present invention belongs of the scope of the present invention, and it should be understood that the present invention is defined only by each claim of the claims.

[0104] [Explanation of symbols]

[0105] 100. AI-based amyloid PET positive detection device,

[0106] 110. Image acquisition unit,

[0107] 120. MRI preprocessing unit,

[0108] 130. Division,

[0109] 140. SUVR calculation section,

[0110] 150. Learning Department,

[0111] 160. Amyloid PET positive judgment section

Claims

1. Image acquisition unit that acquires PET images and MRI images of the brain of a diagnostic subject who has been administered contrast agent; A segmentation unit that segments normal white matter (Normal-Appearing White Matter, NAWM) and white matter hyperintensity signal (White Matter Hyperintensity, WMH) in a FLAIR (Fluid Attenuated Inversion Recovery) image acquired from the above MRI image; A standardized uptake value ratio calculation unit that calculates a global standardized uptake value ratio (SUVR) and a standardized uptake value ratio in normal white matter (NAWM SUVR) based on the PET image and the normal white matter (NAWM) region segmented in the segmentation unit; and An amyloid PET positivity determination unit that determines amyloid PET positivity by inputting the global standardized uptake value ratio (SUVR) and NDR (NAWM Dominance Ratio) into a pre-learned prediction model, wherein NDR is a value obtained by dividing the global standardized uptake value ratio (SUVR) by the standardized uptake value ratio (NAWM SUVR) in normal white matter; characterized in that it includes; AI-based amyloid PET positive detection device.

2. In paragraph 1, It is characterized by further including an MRI preprocessing unit that preprocesses the MRI image acquired from the image acquisition unit to divide the entire brain structure into a white matter brain region and a gray matter brain region. AI-based amyloid PET positive detection device.

3. In paragraph 1 or 2, It is characterized by further including a learning unit that learns the prediction model based on the global standardized absorption value ratio (SUVR) and NDR calculated in the above standardized absorption value ratio calculation unit. AI-based amyloid PET positive detection device.

4. In paragraph 1, The above MRI preprocessing unit is, Characterized by segmenting the entire brain structure into white matter brain regions and gray matter brain regions using a T1-weighted image. AI-based amyloid PET positive detection device.

5. In paragraph 1, The above global standardized absorption value ratio (SUVR) is Characterized by the ratio of the average standardized uptake value ratio (SUVR) of multiple regions of interest to the entire cerebellar reference region. AI-based amyloid PET positive detection device.

6. In paragraph 5, The above areas of interest are: characterized by including at least one of the frontal, parietal, lateral temporal, posterior and anterior cingulate cortex regions, AI-based amyloid PET positive detection device.

7. In paragraph 1, The above prediction model is, It is characterized by being implemented as a Gaussian mixture model based on machine learning. AI-based amyloid PET positive detection device.

8. Image acquisition step for obtaining PET images and MRI images of the brain of a diagnostic subject to whom contrast agent has been administered; A segmentation step for segmenting normal white matter (Normal-Appearing White Matter, NAWM) and white matter hyperintensity signal (White Matter Hyperintensity, WMH) in a FLAIR (Fluid Attenuated Inversion Recovery) image acquired from the above MRI image; A standardized uptake value ratio calculation step for calculating a global standardized uptake value ratio (SUVR) and a standardized uptake value ratio in normal white matter (NAWM SUVR) based on the PET image and the normal white matter (NAWM) region segmented in the segmented section; and An amyloid PET positivity determination step for determining amyloid PET positivity by inputting the global standardized uptake value ratio (SUVR) and NDR (NAWM Dominance Ratio) into a pre-learned prediction model, wherein NDR is a value obtained by dividing the global standardized uptake value ratio (SUVR) by the standardized uptake value ratio (NAWM SUVR) in normal white matter; characterized in that it includes; An artificial intelligence-based amyloid PET positive detection method.

9. In paragraph 8, A method characterized in that it further includes an MRI preprocessing step of preprocessing the MRI image acquired in the image acquisition step to divide the entire brain structure into a white matter brain region and a gray matter brain region; An artificial intelligence-based amyloid PET positive detection method.

10. In paragraph 8 or 9, It is characterized in that it further includes a learning step for learning the prediction model based on the global standardized absorption value ratio (SUVR) and NDR calculated through the standardized absorption value ratio calculation step. An artificial intelligence-based amyloid PET positive detection method.

11. In paragraph 8, The above MRI preprocessing step is, A step characterized by dividing the entire brain structure into white matter brain regions and gray matter brain regions using a T1-weighted image. An artificial intelligence-based amyloid PET positive detection method.

12. In paragraph 8, The above global standardized absorption value ratio (SUVR) is Characterized by the ratio of the average standardized uptake value ratio (SUVR) of multiple regions of interest to the entire cerebellar reference region. An artificial intelligence-based amyloid PET positive detection method.

13. In paragraph 12, The above areas of interest are: characterized by including at least one of the frontal, parietal, lateral temporal, posterior and anterior cingulate cortex regions, An artificial intelligence-based amyloid PET positive detection method.

14. In paragraph 8, The above prediction model is, It is characterized by being implemented as a Gaussian mixture model based on machine learning. An artificial intelligence-based amyloid PET positive detection method.

Images