Cerebrovascular branch normalization method in cerebrovascular image, standardized cerebrovascular structure generating method and analysis apparatus

Abstract

A method for normalizing cerebral vascular branches extracted from a cerebral vascular image comprises the steps of: an analysis device receiving a subject’s cerebral vascular image; the analysis device extracting a plurality of vascular unit structures from the cerebral vascular image; the analysis device extracting feature values ​​for each of the plurality of vascular unit structures; the analysis device inputting the feature values ​​of each of the plurality of vascular unit structures into a first learning model trained in advance to classify the chunk to which each of the plurality of vascular unit structures belongs; the analysis device inputting the feature values ​​of each of the vascular unit structures belonging to the same chunk into a second learning model trained in advance to classify a plurality of vascular branches constituting the vascular unit structures belonging to the same chunk; and the analysis device dividing at least one vascular branch among the plurality of vascular branches into a predetermined number of zones and setting an index for all of the zones or for zones located at a certain interval among the zones.

Description

Technology Field

[0001] The technology described below relates to a technique for normalizing and analyzing vascular branches in cerebral vascular imaging. Background Technology

[0002] Intracranial vascular structures or structural variations must be identified for the diagnosis and treatment of cerebrovascular diseases such as stroke. Magnetic resonance angiography (MRA) is a widely used tool for evaluating cerebral artery disease. Prior art literature

[0003] Michelle Livne et al., A U-Net Deep Learning Framework for High Performance Vessel Segmentation in Patients With Cerebrovascular Disease, Front. Neurosci., 28 February 2019. The problem to be solved

[0004] There have been previous studies aimed at analyzing and quantitatively assessing cerebral vascular structures based on imaging. However, due to the complexity of cerebral vascular structures and individual differences, these studies had limitations in classifying structures or identifying variations solely through imaging.

[0005] The technology described below aims to provide a technique for evaluating cerebral vascular structures based on imaging. The technology described below aims to provide a technique for classifying vascular structures at the level of cerebral arterial branches and evaluating subjects. means of solving the problem

[0006] A method for normalizing cerebral vascular branches extracted from a cerebral vascular image comprises the steps of: an analysis device receiving a subject’s cerebral vascular image; the analysis device extracting a plurality of vascular unit structures from the cerebral vascular image; the analysis device extracting feature values ​​for each of the plurality of vascular unit structures; the analysis device inputting the feature values ​​of each of the plurality of vascular unit structures into a first learning model trained in advance to classify the chunk to which each of the plurality of vascular unit structures belongs; the analysis device inputting the feature values ​​of each of the vascular unit structures belonging to the same chunk into a second learning model trained in advance to classify a plurality of vascular branches constituting the vascular unit structures belonging to the same chunk; and the analysis device dividing at least one vascular branch among the plurality of vascular branches into a predetermined number of zones and setting an index for all of the zones or for zones located at a certain interval among the zones.

[0007] A method for generating standardized cerebrovascular structure information includes the steps of: an analysis device receiving cerebrovascular images of subjects belonging to a population; the analysis device setting indices for at least one vascular branch for each of the subjects using the cerebrovascular images of the subjects; and the analysis device generating cerebrovascular structure information for the population by averaging the positions of the same index in the at least one vascular branch of the subjects.

[0008] An analysis device for evaluating a subject using standardized cerebrovascular structure information includes an input device for receiving a cerebrovascular image of the subject, a first learning model for classifying a chunk to which a vascular unit structure belongs, a second learning model for classifying cerebrovascular branches of a vascular unit structure belonging to the same chunk, a storage device for storing standardized cerebrovascular structure information of a population, and a computation device for extracting a plurality of vascular unit structures based on geometric features of a three-dimensional model from the cerebrovascular image, inputting feature values ​​for each of the plurality of vascular unit structures into the first learning model to classify the chunk to which each of the plurality of vascular unit structures belongs, inputting feature values ​​for each of the vascular unit structures belonging to the same chunk into the second learning model to classify a plurality of vascular branches formed by the vascular unit structures belonging to the same chunk, assigning an index that separates the vascular unit belonging to at least one of the plurality of vascular branches at equal intervals, and comparing the position of the index of the at least one vascular branch of the subject with the position of the index of the standardized cerebrovascular structure information. Effects of the invention

[0009] The technology described below classifies structures by cerebral artery region and can be used for the quantification, evaluation, or prediction of cerebrovascular diseases. The technology described below can evaluate cerebrovascular structures with individual variability by quantifying cerebral vascular branches. The technology described below can standardize the cerebrovascular structure of a specific population. Brief explanation of the drawing

[0010] Figure 1 is an example of a system for extracting structural features of a subject's cerebral artery. Figure 2 is an example of the process of classifying cerebral artery branches. Figure 3 is an example of the training and validation process for a model used for classifying cerebral artery branches. Figure 4 shows the results of predicting chunks for a healthy control group. Figure 5 shows the results of predicting chunks for the ICAS group. Figure 6 shows the result of predicting chunks for the Stroke group. Figure 7 is an ROC curve showing the performance for the Stroke group. Figure 8 is an ROC curve showing the performance of the ICAS group. Figure 9 shows the results of cerebral artery branch classification for a normal control group. Figure 10 shows the results of the cerebral artery branch classification of the ICAS group. Figure 11 shows the results of the cerebral artery branch classification of the Stroke group. Figure 12 is an example of the process of quantifying cerebrovascular structure. Figure 13 is an example of indexing performed on a single blood vessel branch. Figure 14 is an example of the process of standardizing information on cerebral artery branches. Figure 15 is another example of the process of standardizing information on cerebral artery branches. Figure 16 is an example of the process of evaluating a subject's cerebrovascular structure using a standard cerebrovascular template. Figure 17 is an example of an analysis device for analyzing the cerebral artery branches of a subject. Specific details for implementing the invention

[0011] The technology described below is subject to various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit the technology described below to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the technology described below.

[0012] Terms such as first, second, A, B, etc., may be used to describe various components, but such components are not limited by the said terms and are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of rights of the technology described below, the first component may be named the second component, and similarly, the second component may be named the first component. The term "and / or" includes a combination of multiple related described items or any of the multiple related described items.

[0013] In terms used in this specification, singular expressions should be understood to include plural expressions unless the context clearly indicates otherwise, and terms such as “includes” should be understood to mean that the described features, number, steps, actions, components, parts, or combinations thereof exist, and not to exclude the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0014] Before providing a detailed description of the drawings, it is to clarify that the classification of components in this specification is merely based on the primary function each component is responsible for. That is, two or more components described below may be combined into a single component, or a single component may be divided into two or more components based on more subdivided functions. Furthermore, each component described below may additionally perform some or all of the functions of other components in addition to its own primary function, and it goes without saying that some of the primary functions of each component may be exclusively performed by other components.

[0015] Furthermore, in performing the method or operation method, each process constituting the method may occur differently from the specified order unless a specific order is clearly indicated in the context. That is, each process may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

[0017] The technique described below is a method for quantifying the cerebral artery structure of a subject using cerebral artery MRA images. The technique described below is a method for characterizing the cerebral artery structure based on cerebral artery branches.

[0018] The technology described below analyzes cerebral artery MRA images to classify vascular structures at the level of artery branches. The technology described below extracts artery branch-based structural features for a subject. Cerebral artery structural features are defined by artery branch information extracted from a specific subject's cerebral artery MRA image. Cerebral artery structural features are unique to specific brain disease-related phenotype groups (disease group or normal group, etc.). The rationale for this is discussed later.

[0019] The researcher analyzed the images using TOF (time-of-flight) MRA. Therefore, the following description is based on TOF MRA. However, the techniques described below can be applied to other types of medical images.

[0020] The following explains that the analysis device classifies and standardizes cerebral artery branches by analyzing TOF MRA. The analysis device can standardize the cerebral vascular structure by consistently quantifying the cerebral vascular structure of each individual in the population and generalizing the quantified cerebral vascular structure. The analysis device can be implemented as various devices capable of consistent data processing. For example, the analysis device can be implemented as a PC, a server on a network, a smart device, or a chipset with a dedicated program embedded therein.

[0021] Standardized vascular information is general vascular structure information for a population. For example, standardized vascular information may be the average vascular structure information for a specific population. Standardized vascular information can be referred to as a cerebrovascular standard template. Here, the population can be defined by phenotypes such as gender, age group, and specific diseases. For example, the population may be any one of healthy people, healthy men, healthy women, people with cerebrovascular disease, men with cerebrovascular disease, women with cerebrovascular disease, people with specific diseases, men with specific diseases, and women with specific diseases.

[0022] In addition, the analysis device can evaluate the cerebrovascular structure of a specific subject using a standard cerebrovascular template.

[0024] FIG. 1 is an example of a system (100) for evaluating the cerebral artery structure of a subject. FIG. 1 illustrates an example in which the analysis device is a computer terminal (130) and a server (140).

[0025] The MRA equipment (110) generates an MRA image of a subject. The MRA image or TOF MRA image generated by the MRA equipment (110) may be stored in an EMR (Electronic Medical Record, 120) or a separate database. A subject refers to a person who is evaluated for abnormalities in the cerebrovascular structure.

[0026] The computer terminal (130) can receive MRA images from the MRA equipment (110) or EMR (120) via a wired or wireless network. In some cases, the computer terminal (130) may be a device physically connected to the MRA equipment (110). The computer terminal (130) can extract certain cerebral blood vessel structures from the MRA images and input the features of the extracted cerebral blood vessel structures into a pre-established learning model to derive cerebral artery branch classification results. The computer terminal (130) can perform quantification on the classified cerebral artery branches. In this case, quantification corresponds to the process of extracting features in specific intervals so that the structure of each cerebral artery branch can be identified. The structure of the quantified cerebral artery branches can be defined based on the locations of consecutive indices. The specific quantification process will be described later. The computer terminal (130) can evaluate the subject's cerebral blood vessel structure by comparing the subject's quantified cerebral blood vessel structure with a standard cerebral blood vessel template for the population to which the subject belongs. The standard template database (150) stores standard cerebrovascular templates for one or more populations. The process of generating standard cerebrovascular templates is described later. User A can check the analysis results on the computer terminal (130). The analysis results may include whether the subject's cerebrovascular structure is normal or abnormal.

[0027] The server (140) can receive MRA images from the MRA equipment (110) or EMR (120). The server (140) can extract certain cerebral blood vessel structures from the MRA images and input the features of the extracted cerebral blood vessel structures into a pre-established learning model to derive cerebral artery branch classification results. The server (140) can perform quantification on the classified cerebral artery branches. The server (140) can evaluate the cerebral blood vessel structure of a subject by comparing the subject's quantified cerebral blood vessel structure with a standard cerebral blood vessel template for the population to which the subject belongs. The server (140) can receive a standard cerebral blood vessel template for the population from the standard template DB (150). The server (140) can transmit the analysis results to the terminal of user A. User A can check the analysis results through the user terminal.

[0028] The computer terminal (130) and / or server (140) may also store the analysis results in the EMR (120).

[0029] First, the process of classifying the subjects' cerebral vascular structures is explained. The process by which the researcher constructed a model for classifying cerebral artery branches and the results of its verification are described.

[0030] Figure 2 is an example of a process (200) for classifying cerebral artery branches. Focusing on Figure 2, the researcher explains the process of extracting vascular structures from TOF MRA images and classifying vascular branches based on the characteristics of the vascular structures. In the explanation of Figure 2, the image processing and model building processes performed by the researcher are also to be explained.

[0031] The analysis device receives a TOF MRA image (210). The analysis device extracts structures for image feature extraction from the TOF MRA image (220). The analysis device can detect desired vascular structures using geometric processing to reconstruct a 3D model.

[0032] Vascular structures refer to spots, segments, chunks, and branches. The analysis device can extract vascular structures from an input image using a specific image processing program and extract the features of the corresponding structures.

[0033] First, the vascular structural system used by the researcher is described. The researcher organized a vascular unit structure into four hierarchical levels, which differs from the methods conventionally used in the field of clinical neurology.

[0034] A spot is the basic unit of a 3D cerebral artery tree cubic cell with a constant spacing (size) from the artery centerline. The researcher defined the spot as a cubic cell with a spacing of 0.2801 mm from the artery centerline.

[0035] A segment consists of multiple spots distinguished based on branching points in the vascular structure. In other words, a segment corresponds to a set of multiple consecutive spots located between the branching points of a blood vessel.

[0036] A vessel branch consists of multiple segments. A vessel branch can be identified as one of specific types based on the segments according to the geometry of the vessel branching point. The researcher classified vessel branches into 62 branch types based on the geometry of the vessel branching point. In other words, vessel branches can be consistently classified according to the vascular unit structure proposed by the researcher. Meanwhile, vessel branches can be identified using nomenclature traditionally used in the clinical field.

[0037] Vascular branches can be reorganized and classified according to clinical criteria into (i) symmetry, (ii) anterior or posterior, (iii) basal or pituitary, (iv) middle cerebral arteries (MCA), anterior cerebral arteries (ACA), or posterior cerebral arteries (PCA). Based on this, the researcher defined chunks as structures superior to the branches. Chunks can be classified into types using at least one criterion from a group including the criteria of (i) symmetry, (ii) anterior or posterior, (iii) basal or pituitary, and (iv) middle cerebral arteries (MCA), anterior cerebral arteries (ACA), and posterior cerebral arteries (PCA). In other words, a chunk is a classification of vascular branches into higher-level groups based on clinical criteria. The reason for defining chunks is that it is difficult to classify vascular branches into detailed structural units all at once.

[0038] The researcher defined 20 types of vascular chunks as shown in Table 1 below. Each chunk can be further subdivided into one or more branches.

[0040] 순번 청크 (약어) 청크 코드 분지(약어) 분지 코드 1 Anterior communicating artery(ACOA) A0 anterior communicating artery(ACoA) A0.01 2 Right internal carotid artery (RtICA) A1 Right internal carotid artery (Rt ICA) A1.01 Right ophthalmic artery(Rt Ophthalmic) A1.02 Right anterior choroidal artery(Rt AChA) A1.03 3 Left internal carotid artery(LtICA) A2 Left internal carotid artery(Lt ICA) A2.01 Left ophthalmic artery(Lt Ophthalmic) A2.02 Left anterior choroidal artery(Lt AChA) A2.03 4 Right anterior cerebral basal arteries - middle cerebral artery(RtBasalMCA) A3 Right MCA M1(Rt M1) A3.01 Right M2 superior(Rt MCA Superior) A3.02 Right M2 inferior(Rt MCA Inferior) A3.03 5 Left anterior cerebral basal arteries - middle cerebral artery(LtBasalMCA) A4 Left anterior basal MCA(Lt M1) A4.01 Left_MCA_Superior(Lt MCA Superior) A4.02 Left_MCA_Inferior(Lt MCA Inferior) A4.03 6 Right anterior cerebral basal arteries - anterior cerebral artery(RtBasalACA) A5 Right ACA A1(Rt A1) A5.01 Right ACA A2(Rt A2) A5.02 Right ACA A1+A2(Rt A1+A2) A5.03 7 Left anterior cerebral basal arteries - anterior cerebral artery(LtBasalACA) A6 Left ACA A1(Lt A1) A6.01 Left ACA A2(Lt A2) A6.02 Left ACA A1+A2(Lt A1+A2) A6.03 8 Right anterior cerebral pial arteries - middle cerebral artery(RtPialMCA) A7 Right orbitofrontal artery(Rt MCA lat OFA) A7.01 Right MCA PreRolandic artery(Rt MCA PreRolandic) A7.02 Right MCA Rolandic artery(Rt MCA Rolandic) A7.03 Right MCA Anterior Parietal artery(Rt MCA AntPerietal) A7.04 Right MCA Postior Parietal artery(Rt MCA PostParietal) A7.05 Right MCA Angular artery(Rt MCA Angular) A7.06 Right MCA Posterior Temporal artery(Rt MCA PostTemporal) A7.07 Right MCA Midtemporal artery(Rt MCA MidTemporal) A7.08 Right MCA Antior temporal artery(Rt MCA AntTemporal) A7.09 9 Left anterior cerebral pial arteries - middle cerebral artery(LtPialMCA) A8 Left MCA orbitofrontal artery(Lt MCA lat OFA) A8.01 Left MCA PreRolandic artery(Lt MCA PreRolandic) A8.02 Left MCA Rolandic artery(Lt MCA Rolandic) A8.03 Left MCA Anterior Parietal artery(Lt MCA AntPerietal) A8.04 Left MCA Postior Parietal artery(Lt MCA PostParietal) A8.05 Left MCA Angular artery(Lt MCA Angular) A8.06 Left MCA Posterior Temporal artery(Lt MCA PostTemporal) A8.07 Left MCA Midtemporal artery(Lt MCA MidTemporal) A8.08 Left MCA Antior temporal artery(Lt MCA AntTemporal) A8.09 10 Right anterior cerebral pial arteries - anterior cerebral artery(RtPialACA) A9 Right ACA orbitofrontal artery(Rt ACA med OFA) A9.01 Right ACA Frontopolar artery(Rt A2 Frontopolar) A9.02 Right ACA Callosamarginal artery(Rt ACA Callosamarginal) A9.03 Right ACA Pericallosal artery(Rt ACA Pericallosal) A9.04 11 Left anterior cerebral pial arteries - anterior cerebral artery(LtPialACA) A10 Left ACA orbitofrontal artery(Lt ACA med OFA) A10.01 Left ACA Frontopolar artery(Lt A2 Frontopolar) A10.02 Left ACA Callosamarginal artery(Lt ACA Callosamarginal) A10.03 Left ACA Pericallosal artery(Lt ACA Pericallosal) A10.04 12 Right posterior - vertebral artery(RtVA) P1 Right ve Rightebral artery(Rt VA) P1.01 13 Left posterior - vertebral artery(LtVA) P2 Left ve Rightebral artery(Lt VA) P2.01 14 Right posterior basal - posterior cerebral artery(RtBasalPCA) P3 Right PCA P1(Rt P1) P3.01 Right PCA P2(Rt P2) P3.02 Right PCA P1, P2(Rt P1+P2) P3.03 Right PCA P3,P4(Rt P3,P4) P3.04 15 Left posterior basal - posterior cerebral artery(LtBasalPCA) P4 Left PCA P1(Lt P1) P4.01 Left PCA P2(Lt P2) P4.02 Left PCA P1+P2(Lt P1+P2) P4.03 16 Right posterior pial - posterior cerebral artery(RtPialPCA) P5 Right posterior communcating artery(Rt PCoA) P5.01 Right Hippocampal artery(Rt Hippocampal artery) P5.02 Right PCA Anterior Temporal artery(Rt PCA AnteriorTemporal) P5.03 Right PCA Posterior Temporal artery(Rt PCA PosteriorTemporal) P5.04 Right parieto-occipital artery(Rt parieto-occipital) P5.05 Right calcarine artery(Rt Calcarine) P5.06 17 Left posterior pial - posterior cerebral artery(LtPialPCA) P6 Left posterior communcating artery(Lt PCoA) P6.01 Left_Hippocampal artery(Lt Hippocampal artery) P6.02 Left_PCA_Anterio Rightemporal(Lt PCA AnteriorTemporal) P6.03 Left_PCA_Posterio Rightemporal(Lt PCA PosteriorTemporal) P6.04 Left parieto-occipital artery (Lt parieto-occipital) P6.05 Left calcarine artery(Lt Calcarine) P6.06 18 Right posterior - superior cerebral artery, anterior inferior cerebral artery, posterior inferiorcerebellar artery(RtCbll) P7 Right posterior inferior cerebellar artery(Rt PICA) P7.01 Right anterior inferior cerebellar artery(Rt AICA) P7.02 Right internal auditory artery(Rt IAA) P7.03 Right superior cerebellar artery (Rt SCA) P7.04 19 Left posterior - superior cerebral artery, anterior inferior cerebral artery, posterior inferior cerebellar artery (LtCbll) P8 Left posterior inferior cerebellar artery (Lt PICA) P8.01 Left anterior inferior cerebellar artery (Lt AICA) P8.02 Left internal auditory artery (Lt IAA) P8.03 Left superior cerebellar artery (Lt SCA) P8.04 20 Basilar artery (BA) P0 Basilar artery(BA) P0.01

[0041] The researcher acquired images of the intracranial arteries on a 3.0 T Philips Achieva MRI scanner (Philips Medical Systems). The researcher used whole-brain 3D MRA images with the TOF protocol collected from each participant. Isotropic 0.284 × 0.284 mm 3The parameters for the voxel size were as follows: an echo time of 4.59 ms, a repeat time of 22 ms, a flip angle of 23°, an RBW of 130 Hz / pixel, a GRAPPA factor of 3, and a reference line of 32.

[0042] The data format used by the researcher is DICOM in TOF format. The researcher anonymized the raw data using DICOM Anonymizer Pro and performed region growing using an angiography analysis program that generates segmented cerebral angiography masks.

[0043] This paper describes the process of extracting vascular surfaces from cerebral vascular MRA. The researcher performed isosurface dissection to generate vascular surface models using the VMTK (vascular modeling toolkit) library. The researcher removed artifacts using bicubic interpolation and resampled the coarse images on a regular planar grid. Through this process, the z-axis voxels were aligned to the isovoxel image scale. The continuous 3D space can be subdivided into multiple cells based on each vertex of the isosurfaces. Through this process, the researcher can extract major artery centerlines from the boundary surfaces of each cell in cerebral vascular MRA.

[0044] The analysis device can divide the surface of blood vessels into cells of a certain size in a brain blood vessel MRA image and extract the starting point and framework of the centerline of the brain artery based on the surface of the blood vessels.

[0045] The analysis device can make the endpoint of the centerline more distinct by performing vascular skeletonization. The analysis device can (i) skeletonize the cerebral vascular region and surface, (ii) prune branches under a predetermined threshold, (iii) generate a linked list of tree structures based on the refined skeleton structure, and (iv) determine the endpoint by designating leaf nodes from the linked list. The analysis device can extract the centerline of the blood vessel by tracing the boundary of the cell connecting the determined start point and the endpoint.

[0046] Subsequently, the analysis device characterizes the vascular feature vectors for the segmented groups based on the branching points of the centerline (230). Vascular features include cerebral vascular cross-sectional area, maximally inscribed sphere radius, minimum diameter, maximum diameter, maximum-minimum radius ratio, surface circumference, torsion, curvature, and luminal circularity.

[0047] The researcher built a model that performs automated segmentation of brain vascular systems at the standard naming level in brain MRA images and performs classification (labeling) of cerebral artery branches. The analysis device classifies vascular structures at the chunk level using a first learning model built in advance and classifies each chunk into a specific branch (cerebral artery branch) using a second learning model (240).

[0048] The researcher used a multi-layer perceptron (MLP) as a supervised learning model.

[0049] Of course, other models may be used for the learning model that classifies cerebral artery branches. Here, machine learning models include decision trees, RF (random forest), KNN (K-nearest neighbor), Naive Bayes, SVM (support vector machine), and ANN (artificial neural network). Meanwhile, ANN is a statistical learning algorithm that mimics biological neural networks. Various neural network models are being researched. Like general artificial neural networks, DNN (deep learning network) can model complex non-linear relationships. Various types of DNN models have been studied. For example, there are CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), RBM (Restricted Boltzmann Machine), DBN (Deep Belief Network), GAN (Generative Adversarial Network), and RL (Relation Networks).

[0050] The analysis device first performs classification at the chunk level (Step 1. Chunk level modeling).

[0051] The analysis device first inputs the feature vector of each spot, extracted from the TOF MRA image in spot units, into the first learning model. Here, the feature vector of the spot may be a value for at least one of the aforementioned vascular features. Furthermore, the feature vector of the spot may further include the brightness value of the spot region.

[0052] The researcher used a DNN as the first learning model. The first learning model classifies the input spots into chunks. This process is repeated for all spots. As a result, the analysis device obtains chunk classification results (first classification) for all spots.

[0053] Furthermore, after the analysis device completes primary chunk classification for all spots, it can perform voting-based classification (secondary classification) to improve accuracy. To this end, the analysis device distinguishes segments based on the branching points of blood vessels. Through this process, the analysis device can distinguish the segments of the entire blood vessel. The analysis device can perform majority voting-based chunk assignment on a segment-by-segment basis. For each spot belonging to the same segment, the analysis device checks the classification results of the first learning model and determines the classification result with the most occurrences (chunk type) as the final classification result for the spots constituting that segment. Through this, the analysis device can classify the chunk to which the corresponding segment (= spots belonging to that segment) belongs on a segment-by-segment basis. This process can be repeated for all segments.

[0054] Now, the analysis device performs sub-classification in units of classified chunks (Step 2. Sub-chunk modeling).

[0055] The analysis device inputs each spot belonging to the same chunk into the second learning model in chunk units. The second learning model classifies the input spots into specific cerebral artery branches. This process is repeated for all chunks. Consequently, the analysis device obtains cerebral artery branch classification results (primary classification) for all spots.

[0056] Meanwhile, the second learning model may be a model of the same type as the first learning model (DNN). Also, the second learning model may be a machine learning model of a different type from the second learning model. Furthermore, the second learning model may be an ensemble model.

[0057] Furthermore, after the analysis device completes the cerebral artery branch classification results (primary classification) for all spots, it can perform a voting-based classification (secondary classification) to improve accuracy. The analysis device can perform chunk assignment based on majority voting on a chunk-by-chunk basis. For each spot belonging to the same chunk, the analysis device can verify the classification results provided by the second learning model and determine the classification result with the most occurrences (cerebral artery branch type) as the final classification result for the spots constituting that chunk. Through this process, the analysis device can classify the cerebral artery branches to which a corresponding chunk (= the spots belonging to that chunk) belongs on a chunk-by-chunk basis. This process can be repeated for all chunks.

[0058] Meanwhile, the analysis device may verify or correct the cerebral artery branch classification results (primary classification) for the spots in a different way. For example, the analysis device can verify whether the direction is opposite based on 3D coordinates for segments (branches belonging to the corresponding segment) in which branches are clearly separated left-right or up-down in the image data, and if the direction is opposite, it can correct the classification information.

[0059] The analysis device produces information in which classification is performed in units of cerebral artery branches in the brain MRA image of the subject (patient) through the process illustrated in FIG. 2 (250). Based on the final classification result, medical staff can diagnose the disease and take treatment measures for the subject.

[0061] Table 2 below shows examples of cerebrovascular features used in chunk classification and cerebral artery branch classification.

[0063] Features explanation X,Y,Z Vascular centerline coordinates Area Cross-sectional area of ​​a blood vessel at a specific point Max Inscribed Sphere Radius Maximum inscribed sphere radius Min-Diameter Minimum diameter of the cross-section at a specific point Max-Diameter Maximum diameter of the cross-section at a specific point Hyduralic Luminal Diameter Hydrodynamic vessel diameter Perimemter perimeter of a cross-section at a specific point Min-Max Diameter Ratio Max-minimum radius ratio Luminal Circularity Internal cavity epicenter Curvature Curvature of the centerline at a specific point Torsion Distortion of the centerline at a specific point

[0065] Figure 3 is an example of a learning and validation process (300) for a model used for classifying cerebral artery branches.

[0066] The researcher conducted the study on clinically confirmed stroke patients with consent from Samsung Medical Center. The control cohort consisted of healthy individuals who underwent MRA imaging at the Samsung Medical Center Health Screening Center from January 1, 2013, to December 31, 2016.

[0067] The researcher studied healthy control subjects, stroke patients with intracranial atherosclerosis (ICAS) (Stroke with ICAS, ICAS group), and general stroke patients (Stroke group). The researcher included 157 participants aged 20 to 94 years. The 157 participants were divided into 42 control subjects, 46 ICAS group participants, and 69 Stroke group participants. The researcher used 70% of the collected cohort data as training data and 30% as validation data.

[0068] The researcher identified the structure of cerebral arteries in advance by analyzing the participants' TOF MRAs (image identification using experts). In other words, the researcher classified the cerebral artery branches in the training and validation data beforehand. The classification information regarding the cerebral artery branches is referred to simply as "classification information" below.

[0069] The researcher builds a data pool for learning and verification (310). The data pool includes brain MRA images of the participants.

[0070] The researcher extracts training data and validation data (320). The researcher prepares training data and internal validation data based on MRA images obtained through their affiliated institution. The training data and validation data each include the brain MRA images of a specific subject and classification information regarding those images. Additionally, the researcher performed external validation using images provided by their affiliated institution. Of course, external validation data could also be prepared by collecting MRA images and classification information from an external public database (DB).

[0071] The researcher trains the aforementioned learning model using training data (330). The training of the learning model is performed on an analysis device or a separate computer device. Hereinafter, the device for constructing the learning model is referred to as the learning device. The learning device trains the first learning model and the second learning model using training data. The learning device extracts the aforementioned vascular unit structure from the MRA image. The process of extracting the vascular unit structure is as described above. The vascular unit structure includes the aforementioned spots, segments, etc. The learning device inputs features at the spot level from the extracted vascular unit structure into the first learning model to classify the chunk to which the corresponding spot belongs. In this process, the learning device inputs the features of the spot into the first learning model and updates the parameters of the first learning model by comparing the probability value of the chunk classification output by the first learning model with the classification information for the corresponding spot. Figure 3 illustrates the process of training using MRA image i, and the training process is performed iteratively using various training data. By repeating this process, the first learning model is trained to output the chunk classification of the corresponding spot based on the features of the input spot.

[0072] Meanwhile, during the training process of the learning model, the learning device may also determine the final classification results of the spots belonging to a segment by voting based on the classification results of the spots belonging to that segment.

[0073] The learning unit classifies the cerebral artery branch to which a spot belongs by inputting the features of each spot within the same chunk into a second learning model on a chunk-by-chunk basis. The input spots are those belonging to the same chunk. The learning unit inputs the features of the spots within the same chunk into the second learning model and updates the parameters of the second learning model by comparing the probability value of the cerebral artery branch classification output by the second learning model with the classification information for the corresponding spot. By repeating this process, the second learning model is trained to output the cerebral artery branch classification of the spot based on the input features of the spots within the same chunk.

[0074] Meanwhile, during the training process of the learning model, the learning device may also determine the final classification results of the spots belonging to a chunk by voting based on the classification results of the spots belonging to that chunk.

[0075] Subsequently, the learning device validates the learned model using internal validation data and / or external validation data (340). The learning device extracts vascular unit structures from the input MRA images and inputs the features of each spot into the first learning model to classify the chunk to which the spot belongs. Subsequently, based on the result of the chunk classification, the learning device inputs the features of each spot belonging to the same chunk into the second learning model to classify the cerebral artery branch to which the spot belongs. The researcher verified the performance of the classification results using the validation data.

[0077] Figures 4 through 6 show the results of verifying the predictive performance of chunks within the cerebral vascular structure. Figure 4 shows the results of predicting chunks for a healthy control group. Figure 5 shows the results of predicting chunks for the ICAS group. Figure 6 shows the results of predicting chunks for the Stroke group. In the graphs of Figures 4 through 6, the vertical axis represents the true label, and the horizontal axis represents the predicted label. The model constructed by the researcher demonstrated a prediction accuracy of 82% only for the left anterior basal artery (LtBasalACA, A6) chunk among the 20 chunks, while the remaining chunks showed prediction accuracy ranging from 87% to 99%. Furthermore, the model constructed by the researcher demonstrated similar prediction accuracy for each chunk in the healthy control group, the Stroke group, and the ICAS group. However, anterior communicating artery (ACOA, A0) showed some predictive deviation depending on the target group. This appears to be due to the limited sample size for ACOA and high anatomical variability. The AUC (area under the curve)-ROC (receiver operating characteristic) for chunk classification showed high performance overall, ranging from 0.99 to 1.00, while the PRC (precision-recall curve) was 0.992.

[0079] The researchers also verified the classification accuracy of the model for cerebral artery branches. Table 3 below shows the prediction accuracy of major cerebral artery branches belonging to 20 chunks from the experiment. Table 3 below aggregates the results for the entire cohort (control group, ICAS group, and Stroke group). If there was only one cerebral artery branch belonging to a chunk, the prediction accuracy for that branch was not indicated separately.

[0080] Chunk code Chunks Accuracy (%) Major cerebral artery branches Accuracy (%) A1 Right ICA 98-99 Right ICA 100 Right OA 94-99 Right ACHA 85 A2 Left ICA 96-99 Left ICA 100 Left OA 91-95 Left ACHA 91 A3 Right anterior basal MCA 90-95 Right M1 96 Right MCAS 87-91 Right MCAI 90-96 A4 Left anterior basal MCA 92 Left M1 95-97 Left MCAS 85-92 Left MCAI 92-94 A7 Right anterior pial MCA 96-99 Right MCALO 95 right MCAPR 87-98 right MCAR 95-97 right MCAAP 87-95 right MCAPP 93-95 right MCAA 94-97 right MCAPT 90-94 right MCAMT 93 right MCAAT 90-94 right MCAPF 82-85 A8 Left anterior pial MCA 97-99 Left MCALO 81 left MCAPR 93-96 left MCAR 94-96 left MCAAP 89-94 left MCAPP 90-94 left MCAA 94-97 left MCAPT 92-94 left MCAMT 91 left MCAAT 90-92 left MCAPF 88 A5 Right anterior basal ACA 87-92 Right A1 93-95 right A2 97-99 right A1A2 97 A6 Left anterior basal ACA 82-89 Left A1 94-95 left A2 96-97 left A1A2 76-100 A9 Right anterior pial ACA 88-95 Right ACAMO 89 Right A2F 90-96 Right ACAC 94-97 Right ACAP 97-98 A10 Left anterior pial ACA 88-94 Left ACAMO 88 Left A2F 93-96 Left ACAC 93-96 Left ACAP 97-98 P1 Right posterior VA 95-97 Right VA P2 Left posterior VA 94-97 Left VA P3 Right posterior basal PCA 91-97 Right P1 86-89 Right P2 90-92 Right P1P2 87-94 Right P3P4 98 P4 Left posterior basal PCA 94-97 Left P1 86-90 Left P2 93-95 Left P1P2 94-97 Left P3P4 98-99 P5 Right posterior pial PCA 87-93 Right PPA 88 right HA NA right PCAAT 86-93 right PCAPT 96-100 right PCALP NA right PCOA 93-98 P6 Left posterior pial PCA 89-94 Left PPA 87 left HA NA left PCAAT 81-94 left PCAPT 97-98 left PCALP 100 left PCOA 78-95 P7 Right SCA, AICA, and PICA 94-97 Right PICA 95-98 Right AICA 94-97 Right IAA 79 Right SCA 99 P8 Left SCA, AICA, and PICA 90-97 Left PICA 96-98 Left AICA 89-96 Left IAA NA Left SCA 98-99 P0 BA 93-97 BA (G190) A0 ACOA 37-91 ACOA (G200)

[0081] 상기 표 3에서 약어는 다음과 같은 의미이다. ICA(internal carotid arteries), OA(ophthalmic arteries), ACHA(anterior choroidal arteries), VA(vertebral arteries), PICA(posterior inferior cerebellar arteries), AICA(anterior inferior cerebellar arteries), IAA(internal auditory arteries), SCA(superior cerebellar arteries), PCOA(posterior communi-cating arteries),PCA(posterior cerebral arteries), P1(pre-communicating PCA), P2(post-communicating PCA), P1P2(the coalescence among P1 and P2), P3P4(the concoction of quadrigeminal and calcarine PCA), PPA(direct peduncular perforating arteries), HA(hippocampal arteries), PCAAT(ante-rior temporal PCA), PCAPT(posterior temporal PCA), PCALP(lateral posterior choroidal arteries), M1(sphenoidal middle cerebral artery), MCA(middle cerebral arteries), MCAS(superior division of MCA), MCAI(inferior division of MCA), MCALO(lateral orbitofrontal arteries), MCAPR(pre-Rolandic MCA), MCAR(Rolandic MCA), MCAAP(anterior parietal MCA), MCAPP(posterior parietal MCA), MCAA(angular MCA),The arteries are MCAPT (posterior temporal MCA), MCAMT (middle temporal MCA), MCAAT (anterior temporal MCA), MCAPF (pre-frontal MCA), ACA (anterior cerebral arteries), A1 (horizontal pre-communicating ACA), A2 (vertical post-communicating pre-callosal ACA), A1A2 (the combination of A1 and A2), ACAMO (medial orbitofrontal ACA), A2F (frontopolar vertical post-communicating pre-callosal ACA), ACAC (callosomarginal ACA), ACAP (peri-callosal ACA), BA (basilar artery), and ACOA (anterior communicating artery). We examine the classification prediction accuracy by cohort. In the control group, the prediction accuracy was 90–99%, except for right MCAPF, left A1 / A2, and left PCAAT. In the ICAS group, the prediction accuracy was 91–100% for ICA, 85–98% for MCA, 88–100% for ACA, 87–100% for PCA, and 96–99% for SCA-AICA-PICA. In the Stroke group, the prediction accuracy was 94–100% for ICA, 90–96% for MCA, 94–98% for ACA, 90–99% for PCA, and 96–99% for SCA-AICA-PICA. The AUC-ROC for cerebral artery branch classification showed high performance overall at 0.99, and the PRC was 0.992.

[0083] The researcher also performed external verification. External verification was conducted to verify the performance in identifying stroke patients. The researcher built a training model using only data from the control group. In other words, the training model used for external verification is a model that outputs a probability value regarding whether a patient is a stroke patient, without classifying cerebral artery branches. The researcher evaluated whether the ICAS group (46 people) and the Stroke group (69 people) could distinguish stroke patients.

[0084] Figures 7 and 8 show the results of verifying the performance of identifying stroke patients. Figure 7 is the ROC curve representing the performance for the Stroke group. For the Stroke group, the micro-average AUC was 0.97, and the macro-average AUC was 0.96. Figure 8 is the ROC curve representing the performance for the ICAS group. For the ICAS group, the micro-average AUC was 0.95, and the macro-average AUC was 0.92. The model for identifying stroke patients demonstrated significantly high classification accuracy overall.

[0086] Cerebral artery structural characteristics based on cerebral artery branches can be information specific to a particular subject. Figures 9 to 11 visualize the cerebral artery branch classification results for the control group, the ICAS group, and the Stroke group. The average number of cerebral artery branches extracted from subjects in each group was 29. Figure 9 shows the cerebral artery branch classification results for the normal control group. Figure 10 shows the cerebral artery branch classification results for the ICAS group. Figure 11 shows the cerebral artery branch classification results for the Stroke group. Figures 9 to 11 show the results of visually distinguishing and displaying cerebral artery branches in the same coordinate space, and it can be seen that there are regions in the three groups where the location or shape of the cerebral artery branch(s) differs from one another.

[0087] Furthermore, the morphology of cerebral blood vessels varies significantly from person to person. In other words, even within the same group (such as normal individuals or patients), it may not be easy to compare the location of a specific blood vessel in one subject with the corresponding location in another. Accordingly, the researcher proposes a process to consistently quantify the aforementioned structural characteristics of cerebral arteries.

[0088] The following describes the process by which the analysis device quantifies the structural characteristics of the cerebral arteries of the subjects. Quantification corresponds to the process of defining the cerebral artery structure in a standardized space. Furthermore, the analysis device can generate a standard cerebral vascular template for the population by averaging the quantified cerebral artery structures for specific subjects within the population.

[0090] FIG. 12 is an example of a process (400) for quantifying cerebrovascular structure. FIG. 12 corresponds to an example of quantifying cerebrovascular structure for one subject.

[0091] The analysis device receives the subject's TOF MRA image and extracts the vascular structure (410). The process of extracting the vascular structure is as described above.

[0092] The analysis device can extract features of the extracted blood vessel structure (420). Blood vessel features include cerebral blood vessel cross-sectional area, maximum inscribed sphere radius, minimum diameter, maximum diameter, maximum-minimum radius ratio, surface circumference (circumference), distortion, curvature, and lumen roundness.

[0093] The analysis device can label (classify) blood vessels by cerebral artery branch unit based on blood vessel characteristics as described in FIG. 2 (430).

[0094] Subsequently, the analysis device performs indexing on the corresponding blood vessel using the classified blood vessel structure (e.g., cerebral artery branch) as a unit (440). This process corresponds to the quantification process of the blood vessel structure. The analysis device can divide a specific section for the classified specific cerebral artery branch and assign an index according to the section or according to sections at regular intervals. Meanwhile, a specific cerebral artery branch may have different structures and lengths depending on the individual. FIG. 12 illustrates three blood vessel structures having different shapes and lengths for the same cerebral artery branch. This is intended to illustrate that even the same blood vessel structure unit may have different shapes and lengths. FIG. 12 illustrates an example in which indexing is performed from 1 to 5 for the same cerebral artery branch. The structure of the quantified cerebral artery branch can be evaluated based on the position of the index.

[0096] The indexing process for specific cerebral artery branches is described below. The analysis device can classify specific cerebral artery branches in the subject's cerebral blood vessels and perform indexing on the classified specific cerebral artery branches (e.g., A4.01 in Table 2).

[0097] Indexing for specific vascular branches can be performed in various ways. The analysis device can divide a specific cerebral artery branch into a fixed number (N) of segments. In this case, each segment has the same length. The analysis device can assign an index based on a specific point (e.g., the center) within the divided segments. The point where the index is assigned can vary. For instance, the point where the index is assigned may be the center of the segment, the beginning of the segment, or the end of the segment. For the following explanation, it is assumed that the point where the index is assigned is the center of the segment. The analysis device can assign an index to each of the N segments. Alternatively, the analysis device can assign an index to the N segments at regular intervals.

[0098] Figure 13 is an example of indexing performed on a single blood vessel branch. Figure 13 is an example illustrating a blood vessel branch having N branches. Figure 13 is an example of assigning indices to divided sections. For convenience of explanation, Figure 13 assumes a straight blood vessel. In Figure 13, a circle may be a spot. In this case, each section of the cerebral artery branch may be an area containing one or a certain number of spots.

[0099] Figure 13 is an example of assigning an index to a single identical cerebral artery branch (e.g., A4.01).

[0100] FIG. 13(A) is an example of assigning an index to the cerebral artery branches of Subject A. The analysis device extracts the cerebral blood vessels of Subject A and performs classification at the level of cerebral artery branches. Subsequently, the analysis device can perform indexing by cerebral artery branch. The analysis device can divide Subject A's cerebral artery branches (A4.01) into N segments and assign an index to each segment. FIG. 13(A) is an example in which Subject A's cerebral artery branches are divided into N segments and then indexed. In this case, each segment has a length of D.

[0101] FIG. 13(B) is an example in which an index is assigned to the cerebral artery branch of Subject B. The cerebral artery branch (A4.01) of Subject B has a longer length than the cerebral artery branch of Subject A. The analysis device can divide the cerebral artery branch (A4.01) of Subject B into N segments and assign an index to each segment. FIG. 13(B) is an example in which the cerebral artery branch of Subject B is divided into N segments and an index is assigned. In this case, each segment has a length of D'.

[0103] Furthermore, the analysis device can perform quantification (indexing) on ​​all cerebrovascular structures belonging to a specific population and prepare a standard cerebrovascular template for that population. The standard cerebrovascular template can be prepared for each vascular region (chunk or cerebral artery branch).

[0104] For example, the analysis device can prepare a standard cerebrovascular template for a specific group (e.g., a stroke group). In this case, the stroke group consists of multiple subjects, and the subjects may have slightly different cerebrovascular morphologies. The analysis device can standardize the cerebral artery branches for the stroke group to a consistent degree.

[0105] Figure 14 is an example of the process of standardizing information on cerebral artery branches. Figure 14 is an example of generating a standard cerebral blood vessel template for the same specific cerebral artery branch. That is, the multiple cerebral blood vessel branches illustrated in Figure 14 are cerebral blood vessel branches that have the same branch code. In Figure 14, a circle within a blood vessel may be a single spot.

[0106] The analysis device normalizes the same cerebral artery branches for subjects A, B, and C, who belong to the same population. Figure 14 shows an example where indices are assigned at regular intervals. The cerebral artery branches of subjects A, B, and C differ slightly in shape from one another. The analysis device can average the positions of the normalized cerebral artery branches of subjects A, B, and C based on the same index. For example, the analysis device can align the cerebral artery branches of subjects A, B, and C in the same direction, place them in the same coordinate space, and then extract the index positions of each branch. The analysis device can average the positions of the same index in the cerebral artery branches of subjects A, B, and C. The analysis device can perform the same operation for all cerebral artery branches. Through this process, the analysis device can construct a standard cerebrovascular template for a specific population (e.g., healthy men or men with cerebrovascular disease).

[0108] Figure 15 is another example of the process of standardizing information on cerebral artery branches. Figure 15 is an example of generating a standard cerebral blood vessel template for the same specific cerebral artery branch. In Figure 15, a circle within a blood vessel may be a single spot.

[0109] The analysis device normalizes the same cerebral artery branches for subjects D, E, and F, who belong to the same population. Figure 15 shows an example where indices are assigned at regular intervals. The cerebral artery branches of subjects D, E, and F differ slightly in shape from one another. Subject D's cerebral artery branch is longer than that of the other subjects. The analysis device can average the positions of the normalized cerebral artery branches of subjects D, E, and F based on the same index. For example, the analysis device can align the cerebral artery branches of subjects D, E, and F in the same direction, place them in the same coordinate space, and then extract the index positions of each branch. The analysis device can average the positions of the same index in the cerebral artery branches of subjects D, E, and F. The analysis device can perform the same operation on all cerebral artery branches. Through this process, the analysis device can construct a standard cerebrovascular template for a specific population (e.g., healthy men or men with cerebrovascular disease).

[0111] FIG. 16 is an example of a process (500) for evaluating the cerebrovascular structure of a subject using a standard cerebrovascular template.

[0112] The analysis device first constructs a cerebrovascular standard template database (DB) for a specific population (510). The process of constructing the cerebrovascular standard template is as described in FIGS. 12 to 15. The analysis device can generate a cerebrovascular standard template for the corresponding population by classifying and normalizing the cerebral artery branches of subjects belonging to the population by cerebral artery branch, and then averaging the index positions. At this time, the specific population can be defined by gender and health status (presence or absence of a specific brain disease). It is assumed that the cerebrovascular standard template DB stores a cerebrovascular standard template for a healthy male group, a cerebrovascular standard template for a healthy female group, a cerebrovascular standard template for a male group with a specific brain disease, and a cerebrovascular standard template for a female group with a specific brain disease.

[0113] The analysis device acquires a brain MRA image of the subject being evaluated (520). The analysis device can extract brain blood vessels from the subject's brain MRA image and classify brain artery branches based on the characteristics of the extracted brain blood vessels (530).

[0114] The analysis device performs indexing (normalization) on at least one cerebral artery branch among the subject's cerebral artery branches that is the subject of analysis (540). At this time, the analysis device may perform indexing on all of the subject's cerebral artery branches. Furthermore, the analysis device may selectively index specific cerebral artery branch(s) that are significant for the subject. The latter is a case where the association between a specific disease and the structure of specific cerebral artery branch(s) has been identified in advance.

[0115] The analysis device can compare the subject's entire cerebral artery branches or specific cerebral artery branch(s) with the population's standard cerebral blood vessel template (550). The analysis device can compare each of the subject's entire cerebral artery branches with the standard cerebral blood vessel templates for those cerebral artery branches. Alternatively, the analysis device can compare specific cerebral artery branch(s) with the standard cerebral blood vessel templates for those cerebral artery branches.

[0116] For example, if the subject is male, the analysis device can compare the subject's cerebral artery branches with a standard cerebrovascular template of a normal male group. The analysis device can determine the difference between the structure of the subject's cerebral artery branches and the structure of the standard cerebrovascular template by comparing the indices of the subject's cerebral artery branches with the locations of the corresponding indices in the standard cerebrovascular template. If there are cerebral artery branch(s) that have a difference exceeding a preset threshold, the analysis device can output the corresponding cerebral artery branch and the difference value. Alternatively, if there are cerebral artery branch(s) that have a difference exceeding a preset threshold, the analysis device can determine that the subject does not fall within the normal range (high probability of a specific disease).

[0117] Alternatively, if the subject is male, the analysis device can compare the subject's cerebral artery branches with a standard cerebrovascular template for a group of men with specific brain diseases. The analysis device can determine the similarity between the structure of the subject's cerebral artery branches and the structure of the standard cerebrovascular template by comparing the indices of the subject's cerebral artery branches with the locations of the corresponding indices in the standard cerebrovascular template. If there are cerebral artery branch(s) with a difference smaller than a preset threshold (i.e., structural similarity), the analysis device can output the corresponding cerebral artery branch(s) and the difference value. Alternatively, if there are cerebral artery branch(s) with a difference smaller than a preset threshold, the analysis device can determine that the subject does not fall within the normal range (high probability of a specific disease).

[0119] FIG. 17 is an example of an analysis device (600) for analyzing the cerebral artery branches of a subject. The analysis device (600) indexes the cerebral artery branches of the subject. The analysis device (600) can evaluate the subject based on the cerebral artery branches and the index. For example, the analysis device (600) can classify the cerebral artery branches of the subject and evaluate the subject's phenotype based on this. Alternatively, the analysis device (600) can classify the cerebral artery branches of the subject and evaluate the effect of medication based on this. Alternatively, the analysis device (600) can standardize the characteristics of the cerebral blood vessels based on the cerebral artery branches for a specific population.

[0120] The analysis device (600) can be physically implemented in various forms. For example, the analysis device (600) can take the form of a computer device such as a PC, a network server, a chipset dedicated to data processing, etc.

[0121] The analysis device (600) may include a storage device (610), a memory (620), a computation device (630), an interface device (640), a communication device (650), and an output device (660).

[0122] The storage device (610) can store MRA images generated from the MRA equipment.

[0123] The storage device (610) can store code or a program for extracting the aforementioned cerebral vascular structures (spots, segments, etc.) from an MRA image.

[0124] The storage device (610) may store code or a program for extracting feature values ​​(feature vectors) for a spot of a cerebral blood vessel structure. (i) The feature values ​​may include at least one of the following elements: blood vessel centerline coordinates, cerebral blood vessel cross-sectional area, maximum inscribed sphere radius, minimum diameter, maximum diameter, maximum-minimum radius ratio, hydrodynamic blood vessel inner diameter, perimeter of the blood vessel cross-section, surface circumference, distortion, curvature, and lumen roundness. (ii) Furthermore, the feature values ​​may further include brightness values ​​for the corresponding spot area.

[0125] The storage device (610) can store a first learning model that classifies belonging chunks by spot unit as described in FIGS. 2 and 3, and a second learning model that classifies cerebral artery branches for each of the spots belonging to the same chunk.

[0126] The storage device (610) can store reference information for subject evaluation. The reference information is a standard cerebrovascular template for a specific population. The population can be diverse, such as normal subjects, subjects with a specific phenotype, subjects who have received medication, etc. Furthermore, the population can be further distinguished by information such as gender, and the storage device (610) can hold standard cerebrovascular templates for each subgroup (normal males, females with a specific cerebrovascular disease, males who have received treatment medication for a specific disease, etc.).

[0127] The memory (620) can store data and information generated during the process in which the analysis device (600) classifies cerebral artery branches in an MRA image, indexes cerebral artery branches, and evaluates indexed cerebral artery branches.

[0128] The interface device (640) is a device that receives certain commands and data from the outside.

[0129] The interface device (640) can receive an MRA image from a physically connected input device or an external storage device.

[0130] The interface device (640) can receive a standard template of the population of cerebrovascular diseases from a physically connected input device or an external storage device.

[0131] The interface device (640) can receive a selection command for a specific cerebral artery branch to be analyzed among the cerebral artery branches.

[0132] The interface device (640) may also transmit the results of analyzing the subject's MRA image to an external object.

[0133] The communication device (650) refers to a configuration that receives and transmits certain information through a wired or wireless network.

[0134] The communication device (650) can receive MRA images from an external object.

[0135] The communication device (650) can receive a standard template of the population of cerebrovascular diseases from an external object.

[0136] The communication device (650) can receive a selection command for a specific cerebral artery branch to be analyzed among the cerebral artery branches.

[0137] The communication device (650) may also transmit the results of analyzing the subject's MRA image to an external object, such as a user terminal.

[0138] The interface device (640) may be a device that receives data from the communication device (650) and transmits it into the analysis device (600).

[0139] The output device (660) is a device that outputs certain information. The output device (660) can output an interface required for the data processing process, an MRA image, a cerebral blood vessel structure extracted from the MRA image, a chunk classification, a cerebral artery branch classification, a cerebral artery branch index, and an analysis result based on the indexed cerebral artery branch.

[0140] The computing device (630) can classify cerebral artery branches in an MRA image.

[0141] The computing device (630) can reconstruct the structure in the MRA image through geometric processing. At this time, the computing device (630) can identify spots, segments, etc., which are the aforementioned cerebral blood vessel structures.

[0142] The computing device (630) can distinguish multiple cells based on each vertex of the dorsal surfaces in a continuous 3D space in an MRA image. In this process, the computing device (630) can perform preprocessing steps such as noise removal and normalization of the image. The computing device (630) can distinguish the structure into multiple cells constituting the surface of the blood vessel and extract the major artery centerline from the boundary surface of each cell in the cerebral blood vessel MRA.

[0143] The computing device (630) can divide the surface of the blood vessel into cells of a certain size in a brain blood vessel MRA image and extract the starting point and skeleton of the centerline of the brain artery based on the surface of the blood vessel. The computing device (630) can perform blood vessel skeleton refinement to make the end point of the centerline more distinct. The computing device (630) can (i) skeletonize the brain blood vessel region and surface, (ii) cut off branches under a predetermined threshold, (iii) generate a linked list of tree structures based on the refined skeleton structure, and (iv) determine the end point by specifying leaf nodes from the linked list. The analysis device can extract the centerline of the blood vessel by tracking the boundary surface of the cell connecting the determined starting point and the end point.

[0144] The computing device (630) can identify spots, which are the basic units of 3D cerebral artery tree cubic cells that are spaced at regular intervals from the artery centerline. Additionally, the computing device (630) can identify segments, which are specific regions in the vascular structure where multiple spots are separated based on branching points.

[0145] The computing device (630) can classify the chunk to which the input spots belong by inputting the spots extracted from the MRA image into the first learning model as a single spot unit. The computing device (630) can extract feature values ​​of the spots and input them into the first learning model. The operation and learning process of the first learning model are as described above. The first learning model may be a DNN-based model. Through this process, the belonging chunk is determined for each individual spot.

[0146] Additionally, the computing device (630) can determine the final classification result on a segment basis. As described above, the computing device (630) can set the classification result value with the most classification results based on the chunk classification results of spots belonging to the same segment as the chunk classification result of spots belonging to that segment.

[0147] The computing device (630) can classify the cerebral artery branch to which the input spot belongs by inputting spots belonging to the same chunk into the second learning model as a single spot unit. The computing device (630) can extract feature values ​​of the spots and input them into the second learning model. The operation and learning process of the second learning model are as described above. The second learning model may be a DNN-based model. The second learning model may also be an ensemble model using heterogeneous learning models.

[0148] Additionally, the computing device (630) can determine the final classification result in chunk units. As described above, the computing device (630) can set the classification result value with the most classification results as the classification result of the spots belonging to the corresponding chunk based on the classification results of the cerebral artery branches of the spots belonging to the same chunk.

[0149] Furthermore, the computing device (630) may verify or correct the cerebral artery branch classification results (primary classification) for the spots in a different way. For example, the analysis device can verify whether the direction is opposite based on 3D coordinates for segments (branches belonging to the corresponding segment) in which the branches are clearly separated left-right or up-down in the image data, and if the direction is opposite, it can correct the classification information.

[0150] Through this process, the computing device (630) can classify the structure into units of cerebral artery branches extracted from the MRA image.

[0151] Meanwhile, the computing device (630) may classify whether a patient is a patient with a specific brain disease based on the MRA image. In this case, the computing device (630) must use a pre-trained model for brain disease classification. This was mentioned in the aforementioned external verification process. The computing device (630) may also classify whether the subject of the image is a patient with a specific brain disease by extracting spots from the MRA image and inputting feature values ​​for the spots into a brain disease classification model.

[0152] A learning model for classifying brain diseases may be a two-stage model. That is, the first learning model classifies chunks of spots extracted from MRA images, and the second learning model may receive feature values ​​of the spots in units of the same chunks to ultimately classify whether or not a brain disease exists. In this case, the second learning model may receive feature values ​​in units of a single spot and classify the presence of brain disease based on the evaluation results for all spots. Alternatively, the second learning model may receive feature values ​​of all spots within the same chunk at once to classify the presence of brain disease. The second learning model may also be a model that extracts vascular branching characteristics for the spots or for each individual spot and classifies the presence of brain disease based on the extracted characteristics. This is a matter to be determined by the training process of the second learning model and the training data.

[0153] The computing device (630) can perform normalization on at least some of the classified cerebral artery branches. The computing device (630) may normalize all of the subject's cerebral artery branches, or may normalize only specific cerebral artery branch(s) that are the subject of analysis.

[0154] Furthermore, the computing device (630) may generate a cerebrovascular standard template for a specific population using the index of the cerebral artery branches of subjects belonging to that population. The process of generating the cerebrovascular standard template is as described in FIGS. 12 to 15.

[0155] The computing device (630) can evaluate the subject's cerebral vascular structure by comparing the subject's indexed cerebral artery branch(s) with a population cerebral vascular standard template. This process is the same as the process described in FIG. 16.

[0156] The computing device (630) may be a device such as a processor, AP, or a chip with a program embedded in it that processes data and performs certain operations.

[0158] In addition, the image processing method, MRA-based cerebral artery blood vessel analysis method, cerebral artery branch classification method, cerebral blood vessel standard template generation method, and brain disease classification method described above may be implemented as a program (or application) comprising an executable algorithm that can be executed on a computer. The program may be provided by storing it on a transitory or non-transitory computer-readable medium.

[0159] A non-transient readable medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, the various applications or programs described above may be stored and provided on a non-transient readable medium such as a CD, DVD, hard disk, Blu-ray disc, USB, memory card, ROM (read-only memory), PROM (programmable read-only memory), EPROM (Erasable PROM, EPROM), EEPROM (Electrically EPROM), or flash memory.

[0160] Transient readable media refers to various types of RAM such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synclink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).

[0161] The embodiments and drawings attached to this specification merely clearly illustrate a part of the technical ideas included in the aforementioned technology, and it is self-evident that all variations and specific embodiments that can be easily inferred by a person skilled in the art within the scope of the technical ideas included in the specification and drawings of the aforementioned technology are included within the scope of the rights of the aforementioned technology.

Claims

Claim 1 A step in which an analysis device receives a brain blood vessel image of a subject; a step in which the analysis device extracts a plurality of vascular unit structures from the brain blood vessel image; a step in which the analysis device extracts feature values ​​for each of the plurality of vascular unit structures; a step in which the analysis device inputs the feature values ​​of each of the plurality of vascular unit structures into a first learning model that has been pre-trained to classify the chunk to which each of the plurality of vascular unit structures belongs; a step in which the analysis device inputs the feature values ​​of each of the vascular unit structures belonging to the same chunk into a second learning model that has been pre-trained to classify a plurality of vascular branches constituting the vascular unit structures belonging to the same chunk; A method for normalizing a cerebral blood vessel branch extracted from a cerebral blood vessel image, wherein the analysis device divides at least one of the plurality of blood vessel branches into a predetermined number of zones, and sets an index for all of the zones or for zones located at a certain interval among the zones to generate a normalized blood vessel branch, wherein the analysis device divides a region of a certain size based on the centerline of the cerebral blood vessel in the cerebral blood vessel image to extract the plurality of blood vessel unit structures, and the feature value is geometric structure information calculated through an image processing process for each of the plurality of blood vessel unit structures. Claim 2 A method for normalizing cerebral blood vessel branches extracted from a cerebral blood vessel image, wherein the vascular unit structure is a spot, and the spot is a cubic cell having a constant interval from the arterial centerline extracted from the cerebral blood vessel image. Claim 3 In claim 1, the method for normalizing cerebral vascular branches extracted from cerebral vascular images, wherein the feature values ​​include cerebral vascular cross-sectional area, maximum inscribed sphere radius, minimum diameter, maximum diameter, maximum-minimum radius ratio, surface circumference, distortion, curvature, and lumen roundness. Claim 4 A method for normalizing cerebral blood vessel branches extracted from a cerebral blood vessel image, wherein the step of classifying the chunks comprises: a step in which the analysis device performs a first chunk classification for each of the plurality of vascular unit structures using the first learning model; and a step in which the analysis device performs a second chunk classification for the vascular unit structures belonging to the same segment using a majority voting method based on the first chunk classification results of the vascular unit structures belonging to the same segment among the plurality of vascular unit structures, wherein the segment consists of vascular unit structures belonging to a region distinguished by a branching point in the vascular structure. Claim 5 In claim 1, the step of classifying the blood vessel branches comprises: a step in which the analysis device performs a primary blood vessel branch classification for each of the plurality of blood vessel unit structures belonging to the same chunk using the second learning model; and a step in which the analysis device performs a secondary blood vessel branch classification for the blood vessel unit structures belonging to the same chunk using a majority voting method based on the primary blood vessel branch classification results of the blood vessel unit structures belonging to the same chunk among the plurality of blood vessel unit structures, wherein the method for normalizing blood vessel branches extracted from a brain blood vessel image. Claim 6 The method comprises the steps of: an analysis device receiving brain vascular images of subjects belonging to a population; the analysis device setting indices for at least one vascular branch for each of the subjects using the brain vascular images of the subjects; and the analysis device generating brain vascular structure information for the population by averaging the positions of the same index in the at least one vascular branch of the subjects, wherein the step of setting indices for at least one vascular branch includes: a step in which the analysis device inputs the feature values ​​of each of a plurality of vascular unit structures extracted from the brain vascular images into a first learning model that has been pre-learned to classify the chunk to which each of the plurality of vascular unit structures belongs; and a step in which the analysis device inputs the feature values ​​of each of the vascular unit structures belonging to the same chunk into a second learning model that has been pre-learned to classify the at least one vascular branch formed by the vascular unit structures belonging to the same chunk. A method for generating standardized cerebrovascular structure information, wherein the analysis device divides the at least one blood vessel branch into a predetermined number of zones and sets an index for all of the zones or for zones located at a certain interval among the zones, wherein the feature value is geometric structure information calculated through an image processing process for each of the plurality of blood vessel unit structures. Claim 7 An input device for receiving a subject's cerebral vascular image; a first learning model for classifying the chunk to which a vascular unit structure belongs, a second learning model for classifying the cerebral vascular branches of the vascular unit structure belonging to the same chunk, and a storage device for storing standardized cerebral vascular structure information of the population; The analysis device includes a computing device that evaluates a subject by comparing the position of the index of the at least one blood vessel branch of the subject with the position of the index of the standardized cerebrovascular structure information, wherein the computing device extracts the plurality of blood vessel unit structures by dividing an area of ​​a certain size based on the centerline of the cerebrovascular vessel in the cerebrovascular image, classifies the chunk to which each of the plurality of blood vessel unit structures belongs by inputting the feature value of each of the blood vessel unit structures belonging to the same chunk to the second learning model, classifies the plurality of blood vessel branches formed by the blood vessel unit structures belonging to the same chunk, assigns an index that divides the blood vessel units belonging to the blood vessel branch among the plurality of blood vessel branches at equal intervals, and evaluates the subject by comparing the position of the index of the at least one blood vessel branch of the subject with the position of the index of the standardized cerebrovascular structure information. Claim 8 An analysis device for evaluating a subject using standardized cerebrovascular structure information, wherein the vascular unit structure is a spot, and the spot is a cubic cell having a constant interval from the arterial centerline extracted from the cerebrovascular image. Claim 9 In claim 7, the analysis device evaluates a subject using standardized cerebrovascular structural information, wherein the feature values ​​include cerebral blood vessel cross-sectional area, maximum inscribed sphere radius, minimum diameter, maximum diameter, maximum-minimum radius ratio, surface circumference, distortion, curvature, and lumen roundness. Claim 10 In claim 7, the computing device performs a primary chunk classification for each of the plurality of vascular unit structures using the learning model, and performs a secondary chunk classification for the vascular unit structures belonging to the same segment using a majority voting method based on the primary chunk classification results of the vascular unit structures belonging to the same segment among the plurality of vascular unit structures, wherein the segment is an analysis device that evaluates a subject using standardized cerebrovascular structure information consisting of vascular unit structures belonging to a region distinguished by a branching point in the vascular structure.

Images