KR102043829B1 - Method, Device and Program for Estimating Time of Lesion Occurrence - Google Patents

Abstract

In a method for estimating lesion occurrence time according to an embodiment of the present invention, a region of interest selection step of selecting a region of interest corresponding to the position of the subject from the brain image data of the subject, selecting a symmetric region symmetrical with the region of interest Selecting a symmetry region, calculating asymmetry data for each voxel of the region of interest by using the region of interest and the symmetry region, obtaining input data from the asymmetry data for each voxel, and And an input step of inputting into an estimation model for estimating the time of occurrence of the lesion, and an output step of outputting information about the time of occurrence of the lesion, based on an output value of the estimation model. In this case, the estimation model is an input / output function expressing a correlation between learning input data obtained from brain image data of a plurality of data providers and a lesion occurrence time point.

Description

Method, Device and Program for Estimating Time of Lesion Occurrence

The present invention relates to a method, apparatus and program for estimating the time of lesion development in a cerebral infarction patient.

Cerebral infarction is a disease in which the blood vessels in the brain are blocked and some cells in the brain die. Cerebral infarction occurs due to vascular occlusion due to atherosclerosis of the cerebrovascular vessels and atrial fibrillation, which results in blockage of the cerebrovascular vessels by cardioembolic clots that migrate from the heart to the brain. As the blood vessels are blocked by the blood vessels of the brain, the blood supply to the brain is blocked, and the brain tissues supplied by the blocked blood vessels are infarcted, resulting in neurological symptoms such as consciousness disorder and physical paralysis.

In the case of cerebral infarction, treatment is performed in which blood vessels are reopened early to restore brain blood flow before brain tissue is completely infarcted. At this time, how fast the blocked blood vessels are the most influential factor in the prognosis of cerebral infarction. The time to initial treatment (Golden time) is about 4.5-6 hours after the onset to minimize the sequelae caused by cerebral infarction. Diagnosis and treatment are required. However, 20-30% of patients with acute cerebral infarction have been excluded from vascular revascularization treatment because they do not know the exact time of onset of infarction.

Korea Patent Publication No. 10-0946350 (Notice date 2010.03.09)

It is an object of the present invention to provide a method, apparatus and program for estimating the time of lesion occurrence in a cerebral infarction patient in an objective and quantitative manner using brain image data. However, these problems are exemplary, and the scope of the present invention is not limited thereby.

In accordance with an aspect of the present invention, a method for estimating a lesion occurrence time point includes: selecting a region of interest from a brain image data of a subject to select a region of interest corresponding to a position of the subject; A symmetric region selection step of selecting a symmetric region symmetrical with the region of interest; Calculating asymmetry data for each voxel of the region of interest by using the region of interest and the symmetry region; An input step of acquiring input data from the asymmetry data for each voxel and inputting it into an estimation model for estimating the occurrence time of the lesion; And an output step of outputting information on a time point of occurrence of the lesion, based on an output value of the estimation model. In this case, the estimation model is an input / output function expressing a correlation between learning input data obtained from brain image data of a plurality of data providers and a lesion occurrence time point.

According to an embodiment, the estimation model may be an input / output function that outputs a probability that the time of occurrence of the lesion is before or after the predetermined time of the brain image data.

The method for estimating lesion occurrence time according to an embodiment of the present invention may further include training the estimation model before the selecting the ROI.

According to an embodiment, the ROI selection step may include: a data acquisition step of acquiring first brain image data and second brain image data of the subject; Registering the first brain image data and the second brain image data; Selecting, from the first brain image data, a lesion area in which the lesion of the subject is estimated to be located; And selecting a region matching the lesion region as the region of interest from the second brain image data.

According to an embodiment, the data acquiring step may include: acquiring the first brain image data from DWI data; And acquiring the second brain image data from liquid attenuation reverse recovery image (FLAIR) data.

The method for estimating lesion occurrence time according to an embodiment of the present invention may be performed before the step of selecting the symmetry region, and may further include calculating a line of symmetry of the brain image data.

According to an embodiment, the calculating of the asymmetry may include calculating a difference between a signal magnitude of each voxel of the ROI and an average value of all voxel signals of the symmetrical area.

According to an embodiment, the calculating of the asymmetry may include selecting an extended area having a larger size than the symmetric area when the size of the ROI is equal to or smaller than a predetermined size; And calculating a difference between a signal magnitude of each voxel of the region of interest and an average value of all voxel signals of the extended region.

The apparatus for estimating lesion occurrence time according to an embodiment of the present invention selects a region of interest corresponding to the position of the subject from the brain image data of the subject, selects a symmetric region symmetric to the region of interest, and selects the region of interest And calculating the asymmetry data for each voxel of the ROI using the symmetric region, obtaining input data from the asymmetry data for the voxel from the asymmetry data for the voxel, and then estimating the timing of occurrence of the lesion. A control unit inputting the estimation model; And an output unit configured to output information about a time point of occurrence of the lesion, based on an output value of the estimation model. In this case, the estimation model is an input / output function expressing a correlation between learning input data obtained from brain image data of a plurality of data providers and a lesion occurrence time point.

An embodiment of the present invention discloses a computer program stored in a medium for executing the above-described method for estimating lesion occurrence time.

Other aspects, features, and advantages other than those described above will become apparent from the following drawings, claims, and detailed description of the invention.

According to the method, apparatus, or program for estimating lesion occurrence time according to an embodiment of the present invention, it is possible to estimate the lesion occurrence time in an objective and quantitative manner, so that the difference between the observer and the observer when estimating the lesion occurrence time in a qualitative manner. The problem of high intra- and inter-observer variability can be solved. In addition, the onset of lesions can be quickly identified within minutes, which can help to quickly determine treatment options for the subject. Through this, it is possible to rapidly treat cerebrovascular occlusion, which can irreversibly damage the brain, such as cerebral infarction, thereby increasing the therapeutic effect.

Meanwhile, in one embodiment of the present invention, the heterogeneity information of the lesion is obtained through 'by voxel' asymmetry data, and the heterogeneity information is reflected in the learning of the estimation model, thereby improving the accuracy of the estimation of lesion occurrence time. .

Predicting / determining the incidence of cerebral infarction of a patient based on MRI automatic analysis technique as in one embodiment of the present invention provides important information for the treatment of cerebral infarction and increases the number of patients who can undergo vascular revascularization treatment. This can reduce the severity of sequelae after cerebral infarction.

1 is a diagram schematically illustrating a configuration of an apparatus for estimating lesion occurrence time according to an embodiment of the present invention.
2 is a flowchart illustrating a method for estimating lesion occurrence time according to an embodiment of the present invention.
3 is a diagram illustrating brain image data before and after image registration.
4 is a diagram illustrating brain image data before and after centering.
5 is a diagram illustrating a lesion area L of the first brain image data.
6 is a diagram illustrating a lesion area of the first brain image data, an ROI of the second brain image data, and a symmetric area.
7 is a diagram illustrating a method of calculating asymmetry according to an embodiment.
8 is a diagram visualizing the asymmetry of each voxel.
9 is a conceptual diagram schematically illustrating an input step and an output step.
10 is a graph illustrating the number of voxels per section of rFLAIR.
11 is a diagram illustrating an artificial neural network structure of the estimation model in accordance with an embodiment.
12 is a diagram comparing ROC graphs of an estimation model machine trained using signal intensity ratio (SIR) and an estimation model machine trained using rFLAIR.
13 is a block diagram of a controller according to an exemplary embodiment.
14 is a block diagram of a data learner, according to an exemplary embodiment.
15 is a block diagram of a data recognizer according to an exemplary embodiment.

As the invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail. Effects and features of the present invention, and methods of achieving them will be apparent with reference to the embodiments described below in detail together with the drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various forms.

Some embodiments of the present disclosure may be represented by functional block configurations and various processing steps. Some or all of these functional blocks may be implemented in various numbers of hardware and / or software configurations that perform particular functions. For example, the functional blocks of the present disclosure may be implemented by one or more microprocessors or by circuit configurations for a given function. In addition, for example, the functional blocks of the present disclosure may be implemented in various programming or scripting languages. The functional blocks may be implemented in algorithms running on one or more processors. In addition, the present disclosure may employ the prior art for electronic configuration, signal processing, and / or data processing. Terms such as "mechanism", "element", "means" and "configuration" may be used widely and are not limited to mechanical and physical configurations.

Throughout the specification, when a part is "connected" with another part, this includes not only a "direct connection" but also an "electrical connection" between other elements in between. In addition, when a part is said to "include" a certain component, which means that it may further include other components, except to exclude other components unless otherwise stated.

In addition, the connecting lines or connecting members between the components shown in the drawings are merely illustrative of functional connections and / or physical or circuit connections. In an actual device, the connections between components may be represented by various functional connections, physical connections, or circuit connections that are replaceable or added.

In addition, terms including ordinal numbers such as "first" or "second" as used herein may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

Where certain embodiments are otherwise implementable, certain steps may be performed out of the order described. For example, two steps described in succession may be performed substantially concurrently, or may be performed in a reverse order.

As used herein, the term "brain image" refers to an image visualizing the internal structure and / or function of the brain by a direct or indirect method, magnetic resonance imaging, computed tomography (CT) imaging, positron emission tomography (PET) , Positron Emission Tomography (POS) images, Single Photon Emission Computed Tomography (SPECT) images, and the like.

As used herein, the term “magnetic resonance image (MRI)” refers to a diagnostic technique for generating an image or a picture of an internal structure using a magnetic field and an image obtained through the magnetic resonance image.

As used herein, the term “voxel” refers to a regular grid unit in a three-dimensional space, and may mean a pixel in an image of a two-dimensional plane.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings, and the same or corresponding components will be denoted by the same reference numerals, and redundant description thereof will be omitted. .

1 is a diagram schematically illustrating a configuration of an apparatus for estimating lesion occurrence time 1000 according to an embodiment of the present invention.

The lesion occurrence time estimating apparatus 1000 shown in FIG. 1 illustrates only the components related to the present embodiment in order to prevent the features of the present embodiment from being blurred. Therefore, it will be understood by those skilled in the art that other general purpose components may be further included in addition to the components illustrated in FIG. 1.

The apparatus for estimating lesion occurrence time 1000 may correspond to at least one processor or may include at least one processor. Accordingly, the lesion occurrence time estimation apparatus 1000 may be driven in a form included in another hardware device such as a microprocessor or a general purpose computer system.

The apparatus for estimating lesion occurrence time 1000 according to an embodiment is a device for receiving brain image data of a subject, processing the same, and estimating and outputting a timing of occurrence of a lesion occurring in the subject's brain. Referring to FIG. 1, the lesion occurrence time estimating apparatus 1000 includes a controller 1100 and an output unit 1200.

The controller 1100 may include all kinds of devices capable of processing data, such as a processor. Here, the 'processor' may refer to a data processing apparatus embedded in hardware having, for example, a circuit physically structured to perform a function represented by code or instructions included in a program. Examples of such data processing devices embedded in hardware include a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, and an application-specific integrated circuit (ASIC). ), But may include a processing device such as a field programmable gate array (FPGA), but the scope of the present invention is not limited thereto.

First, the controller 1100 receives brain image data of a subject. The subject may be a patient who has a stroke due to thrombus caused by atherosclerosis of the cerebrovascular vessel and an embolism derived from the heart, but the subject is not limited thereto.

Brain image data may include, for example, magnetic resonance image (MRI) data. Magnetic resonance imaging data include diffusion weighted image (DWI) data, fluid attenuated inversion recovery (FLAIR) image data, susceptibility weighted image (SWI) data, and T1-weighted ) May include image data, T2-weighted image data, but the type of magnetic resonance image data is not limited thereto. The lesion occurrence time estimation apparatus 1000 may receive magnetic resonance image data from an external magnetic resonance imaging apparatus or a previously stored medical database (DB), but the present invention is not limited thereto.

The controller 1100 may select a region of interest (ROI) corresponding to the lesion location from the acquired brain image data. The controller 1100 may select an ROI automatically or semi-automatically through an image processing technique. For example, the controller 1100 may automatically select a region of interest (ROI) by determining that the lesion is located in a region where the signal intensity of the brain image data is lower or higher than a specific value. Alternatively, the controller 1100 may select a region of interest (ROI) semi-automatically by determining that the lesion is located in a region lower or higher than a specific value input or set by the user.

Thereafter, the controller 1100 may select a symmetric region C_ROI, contralateral ROI (see FIG. 6) that is symmetric to the ROI. In this case, 'symmetry' may be based on a sagittal plane dividing the subject from side to side. For example, when the ROI is located in the left hemisphere of the subject brain, the symmetry area C_ROI is located in the right hemisphere of the brain. The symmetry plane or the symmetry line (SL, symmetry line (see FIG. 4)), which is a reference for symmetry, may be manually set by the user, but may be automatically or semi-automatic through image processing techniques. It may also be set to). For example, after the controller 1100 calculates the plane of symmetry or the line of symmetry of the subject brain image, the user finely adjusts the angle and / or the position of the plane of symmetry or the line of symmetry line SL to make the final plane of symmetry or the line of symmetry SL. Can be set.

Thereafter, the controller 1100 may calculate asymmetry data for each voxel of the ROI by using the ROI and the symmetry region C_ROI. “Asymmetry” may refer to an indicator that quantifies the extent of brain tissue damage or the progression of a lesion. The asymmetry may be calculated for each voxel of each ROI. In this case, the asymmetry data for each voxel may include information about a relative ratio of the signal strength of the ROI to the signal strength of the symmetric region C_ROI which is a normal region. The degree of asymmetry can be quantified, for example, via an indicator 'rFLAIR', which will be described later in the description related to FIG. 6.

Thereafter, the controller 1100 may process the asymmetry data for each voxel to obtain input data, and then input the same to an estimation model for estimating the occurrence time of the lesion. In this case, the 'estimate model' may be an input / output function expressing a correlation between learning input data obtained from brain image data of a plurality of data providers and a lesion occurrence time.

The estimation model according to an embodiment may be generated by a machine learning technique. The lesion occurrence time estimation apparatus 1000 may use various techniques for training and / or learning such an estimation model. For example, the lesion occurrence estimation apparatus 1000 may classify learning input data obtained from brain image data of a plurality of data providers, such as Logistic regression, Decision tree, Nearest-neighbor classifier, Kernel discriminate analysis, Neural network, Support Vector Machine, Random Algorithms and / or methods (such as forests and boosted trees) can be used. However, this is merely exemplary and the present invention is not limited thereto.

The estimation model according to an embodiment may be a matrix data set for mapping learning input data to lesion occurrence time estimation data that may be expressed in a vector form. 'Training' or 'learning' the estimation model means error back-propagation so that the output from inputting the training input data into the estimation model is closer to the actual point of occurrence data labeled. The technique may mean adjusting the value of the matrix data set.

According to an embodiment, the estimation model may be an input / output function that outputs a probability that the lesion occurrence time of a specific subject is before or after the predetermined time point. The predetermined time point may be 4.5 hours or 6 hours before the brain image data is first obtained, for example, when the MR image is captured, but the present invention is not limited thereto. That is, the estimation model may determine whether the interval from the stroke onset time of the lesion or cerebral infarction to the point of time of acquiring the brain image is within a predetermined period. For example, the estimation model may calculate a probability that the lesion occurrence time of the subject is 6 hours before the time when the brain image was taken. In this case, the estimation model may be a model for estimating a lesion occurrence time by learning a correlation between voxel asymmetry data and a lesion occurrence time.

The output unit 1200 displays and outputs information processed by the lesion occurrence time estimating apparatus 1000. The output unit 1200 may include a display unit for displaying a lesion occurrence time estimation result and a user interface for receiving brain image data. The user may determine the treatment of the subject by referring to the result of the lesion occurrence time displayed on the output unit 1200.

2 is a flowchart illustrating a method for estimating lesion occurrence time according to an embodiment of the present invention. A method for estimating lesion occurrence time according to an embodiment includes a region of interest selection step S210, a symmetry region selection step S220, an asymmetry calculation step S230, an input step S240, and an output step S250.

Referring to FIG. 2, a region of interest selection step S210 of selecting a region of interest ROI corresponding to a lesion location of a subject from brain image data of the subject is performed. The step S210 may be performed automatically or semi-automatically by the controller 1100. The ROI selection step S210 may include a matching step, a centering step, and the like, which will be described later with reference to FIGS. 3 to 6.

Thereafter, a symmetric region selection step S220 of selecting a symmetric region C_ROI symmetrical to the ROI is performed. Before selecting the symmetry region C_ROI, centering the brain image data may be performed, which will be described later with reference to FIG. 5.

Thereafter, asymmetry calculation step S230 is performed. In the asymmetry calculation step S230, asymmetry data of each of the voxels included in the ROI is calculated by using signal strength differences between the ROI and the symmetry region C_ROI.

Subsequently, after input data is obtained from asymmetry data for each voxel, an input step S240 of inputting input data into an estimation model is performed. The input data may include information on the number distribution of voxels according to asymmetry, which will be described later in the description of FIG. 10. The input step S240 may include selecting, generating, or pre-processing input data to be input to the estimation model.

The estimation model may be learned / generated by an external device, or may be learned / generated by the controller 1100 of the apparatus for estimating lesion occurrence time 1000 according to an embodiment of the present invention. When the estimation model is trained in the control unit 1100, the estimation model should be trained before estimating a lesion occurrence time of a specific subject.

The method of estimating lesion occurrence time according to an embodiment may further include training the estimation model before the region of interest selection step S210. In this case, the estimating model training step may include acquiring training data from brain image data and lesion occurrence time data of a plurality of data providers, and machine learning the estimation model using the training data.

After the input step S240, an output step S250 of estimating and generating a lesion occurrence time based on an output value of the estimation model is performed.

That is, in the method for estimating lesion occurrence time according to an embodiment of the present invention, after obtaining input data from asymmetry data for each voxel obtained by processing brain image data, the pre-learned / generated estimation is performed to estimate the lesion occurrence time. Output the input values to the model to estimate the time of lesion occurrence in a particular subject.

Hereinafter, a method of estimating a lesion occurrence time of a specific subject using a trained estimation model will be described in detail with reference to FIGS. 3 to 9.

3 is a diagram illustrating brain image data before and after image registration.

According to an embodiment, the region of interest selection step S210 may include a data acquisition step of acquiring first brain image data and second brain image data of the subject. In this case, the data acquiring step may include acquiring first brain image data from DWI data (hereinafter referred to as DWI data) and second brain image from FLAM data (hereinafter referred to as FLAIR data). Obtaining data.

In the data acquisition step, a plurality of types of brain image data may be acquired. For example, the brain image data may include first brain image data and second brain image data. According to an embodiment, the first brain image data may be ADC (Apparent Diffusion Coefficient) image data obtained by processing DWI data, and the second brain image data may be image data obtained by processing FLAIR data.

DWI data is image data obtained by measuring the diffusion rate of water molecules in tissues. In general, DWI data is an image processing method for detecting lesions caused by cerebrovascular occlusion. At this time, the diffusion speed is determined by a b-value indicating the strength of the gradient magnetic field.

The ADC image data represents a change in signal strength of DWI images having different b values. In an embodiment of the present invention, the ADC value is calculated using the signal strength difference of the DWI image when b = 0 and b = 1000 as shown in Equation 1 below, but the present invention is not limited thereto.

<Equation 1>

On the other hand, FLAIR data is a kind of image obtained through the inversion recovery technique that suppresses the signal of the cerebrospinal fluid. FLAIR data show high sensitivity in patients with subacute cerebral infarction, and the more severe the brain tissue damage, the stronger the signal strength.

Since ADC image data and FLAIR data are obtained through different techniques, the degree of damage to tissues may vary over time after cerebral infarction. In one embodiment of the present invention, the lesion occurrence time is estimated using the difference between the ADC image data and the FLAIR data.

However, since the ADC image data and the FLAIR data are obtained by photographing images from different methods and viewpoints, the size of the voxel and the brain position on the coordinates may be different. Therefore, the step of 'matching' the two images may be performed before 'compare' the ADC image data and the FLAIR data. That is, according to an embodiment, after the data acquisition step of acquiring the first brain image data and the second brain image data of the subject, the step of matching the first brain image data and the second brain image data may be performed. In one embodiment of the present invention, by resampling the ADC image to match the size of the voxel with FLAIR, rigid registration was performed.

Referring to FIGS. 3A and 3B, before and after image processing and registration, DWI (b = 0) images, DWI (b = 1000) images, ADC (Apparent Diffusion Coefficient) images, and FLAIR images are displayed. Prior to the matching step, a step of removing a skull region through image processing may be performed.

4 is a diagram illustrating brain image data before and after centering.

According to an embodiment, the method of estimating a lesion occurrence time of a cerebral infarction may further include calculating a symmetry line SL of brain image data before selecting a symmetry region C_ROI. In an embodiment of the present invention, a region corresponding to the brain is selected using a binary mask in a DWI image having b = 0, and the right brain region is inverted to the left based on the centerline ML of the image. When inversion, the symmetry line SL is extracted by quantifying the degree of overlap with the left brain region. In this case, a DSC (Dice similarity coefficient) value, which is a measure of quantifying the degree of overlapping the inverted right brain region and the left brain region, may be determined as in Equation 2 below.

<Equation 2>

Where | X | is the number of voxels in the brain region obtained through a binary mask, | Y | is the number of voxels in the inverted region with respect to the center line ML, and | X∩Y | is the overlap region. Is the number of voxels. At this time, the center line ML or the brain region may be parallelly moved and / or rotated to calculate a line having the largest DSC value as the symmetric line SL.

Referring to (a) of FIG. 4, the midline (ML) of the image is shifted from the symmetry line (SL) passing through the sagittal plane due to the movement of the subject during brain imaging. However, when centering is performed, the center line ML and the symmetry line SL coincide with each other as shown in FIG. 4 (b).

5 is a diagram illustrating a lesion area L of the first brain image data.

Since the signal intensity in the DWI or ADC image data is different from the normal region in the region where the lesion is located, it is assumed that the lesion is located by selecting an area where the signal strength is less than a predetermined size (T) in the ADC image data. Can be selected. In this case, the predetermined size T may be preset or adjusted by the user. (A), (b), (c), and (d) of FIG. 5 are diagrams illustrating the lesion area L when the threshold values T are set to 500, 550, 600, and 650, respectively. In this case, the user may determine an appropriate threshold value T based on whether the contour of the lesion area L of the ADC image displayed on the output unit 1200 matches the bright area of the DWI image.

FIG. 6 is a diagram illustrating a lesion area L of the first brain image data, a region of interest ROI, and a symmetry region C_ROI of the second brain image data.

In an embodiment of the present invention, after selecting the lesion area L from which the lesion is estimated in the first brain image data, the region of the second brain image data matching the lesion area L is selected as the ROI. Can be selected. For example, in one embodiment, the lesion area L is selected from the ADC image data, and mapped to the matched FLAIR data to select the ROI from the FLAIR data. Through this mapping process, signal difference information between the DWI data and the FLAIR data is reflected, thereby improving the prediction accuracy of cerebral infarction.

Thereafter, a symmetric region selection step S220 is performed. The symmetry region C_ROI may be symmetrical with the ROI based on the above-described symmetry line SL. In this case, the ROI may correspond to a lesion of the subject brain, and the symmetry region C_ROI may correspond to a normal region of the subject brain. Meanwhile, when selecting the symmetry region C_ROI, a step of removing noise may be performed. For example, a value in which the signal strength of the second brain image data is greater than or equal to a predetermined size or less in the symmetry region C_ROI may be removed. In this process, for example, it may be removed from the region symmetry region (C_ROI) corresponding to the 'cerebrospinal fluid'. Accordingly, only voxels corresponding to 'tissue' may be included in the symmetry region C_ROI.

After selecting the ROI and the symmetry region C_ROI, an asymmetry calculation step S230 of obtaining asymmetry data for each voxel is performed. As described above, "asymmetry" may be an indicator for quantifying the degree of damage to the brain or the progression of the lesion. Asymmetry may be calculated for each voxel included in the region of interest (ROI). In this case, the asymmetry data for each voxel may include ratio information of signal strength of the ROI of the ROI of the symmetric region C_ROI which is a normal region.

According to an embodiment, calculating the asymmetry may include calculating a difference between a signal size of each voxel of the ROI and an average value of all voxel signals of the symmetry area C_ROI.

The degree of asymmetry can be quantified as shown in Equation 3 below, for example, through an indicator of 'rFLAIR'.

<Equation 3>

는 관심 영역(ROI)에 위치한 각 복셀에서의 FLAIR 신호의 크기를 나타내며,

는 대칭 영역(C_ROI)의 FLAIR 신호 평균값을 나타낸다. Where x and y represent the coordinates of each voxel,

Represents the magnitude of the FLAIR signal in each voxel located in the region of interest (ROI),

Denotes the average value of the FLAIR signal in the symmetry region C_ROI.

In an embodiment of the present invention, as shown in Equation 3, the asymmetry of each voxel of the ROI may be calculated based on the average value of all voxel signals of the symmetrical region C_ROI. In this case, the asymmetry of each voxel may be calculated more robustly than the signal size of one voxel of the symmetric area C_ROI corresponding to the specific voxel of the ROI.

7 is a diagram illustrating a method of calculating asymmetry according to an embodiment.

According to an embodiment, the asymmetry calculation step S230 may include selecting an extended ROI (E_ROI) having a larger size than the symmetry area C_ROI when the size of the ROI is less than or equal to the predetermined size. The method may include calculating a difference between a signal size of each voxel of the ROI and an average value of all voxel signals of the extended area E_ROI.

Referring to FIG. 7A, a lesion area L having a small size of first brain image data (illustrated as ADC image data) is displayed. Accordingly, the size of the ROI is also small in the second brain image data (exemplified as FLAIR data) matched with the first brain image data. In this case, when the asymmetry is calculated based only on the signal strength of the symmetric region C_ROI symmetric with the ROI, the signal asymmetry of each voxel becomes unsensitive to the position of the symmetric region C_ROI. become robust. Therefore, in an embodiment of the present invention, when the size of the ROI is equal to or less than a predetermined size, the 'extended area E_ROI' is selected to be larger than the symmetric area C_ROI. Thereafter, the asymmetry of each voxel of the ROI is calculated based on the signal strength of the extended region E_ROI.

For example, when the size of the ROI is small, the asymmetry of each voxel of the ROI may be quantified as shown in Equation 4 below.

<Equation 4>

는 관심 영역(ROI)에 위치한 각 복셀에서의 FLAIR 신호의 크기를 나타내며,

는 "확장 영역(E_ROI)"의 FLAIR 신호 평균값을 나타낸다. Where x and y represent the coordinates of each voxel,

Represents the magnitude of the FLAIR signal in each voxel located in the region of interest (ROI),

Denotes the average value of the FLAIR signal of the "extended area E_ROI".

The extended area E_ROI may be an area within a predetermined distance r with respect to the center of the symmetry area C_ROI. In this case, among the voxels included in the region included in the predetermined distance r, voxels having a signal strength greater than or equal to a threshold may be processed as noise and removed. According to an embodiment of the present invention, when the size of the ROI is within 100 mm ² , an area where the distance r from the center of the symmetry area C_ROI is within 13 mm is set to be selected as the extended area E_ROI. However, the present invention is not limited thereto.

8 is a diagram visualizing the asymmetry of each voxel. Voxel-specific asymmetry can be visualized in the form of a color contour map (color contour map) as shown in FIG. Through this, the user can easily determine the degree of damage of the lesion according to the location.

9 is a conceptual diagram schematically illustrating an input step S240 and an output step S250.

The voxel-specific asymmetry data contains information about how much damage progressed for each location of the lesion. In this case, the asymmetry data for each voxel of a specific subject may be processed and input data may be input to the estimation model. The previously trained estimation model may receive input data and output an output value p. In this case, the output value p may be a value representing the probability that the lesion occurrence time is before or after the specific time point. For example, when the estimation model is trained to classify whether the lesion occurrence time is 6 hours before or after brain imaging, the output value p may mean a probability that the lesion occurrence time is within 6 hours. If the output value p is equal to or greater than a specific value T, the controller may determine that the lesion occurrence time is after the specific time t. The specific value T may be set to a value having the highest prediction accuracy in the learning process of the estimation model.

When the output step S250 of outputting the lesion occurrence time is completed by the output unit 1200, the user may determine a treatment plan of the subject based on the output lesion estimation time point output.

< MATLAB  Used Example >

Hereinafter, the configuration and effects of the present invention through the embodiments will be described in more detail. The following examples are only for illustrating the present invention, but the scope of the present invention is not limited by the examples.

In an embodiment of the present invention, a program for estimating lesion incidence time was developed using MATLAB (Mathworks, MA, USA). The machine learning software, R (R Foundation for Statistical Computing, Vienna, Austria), was used.

In one embodiment of the present invention, the learning input data was obtained from 240 data providers who are cerebral infarction patients, and the estimation model was externally validated using data obtained from 49 data providers.

10 is a graph illustrating the number of voxels per section of rFLAIR. In Fig. 10, 0 to 1, 1 to 2, 2 to 3,. For example, rFLAIR is divided into 31 sections, such as 29 to 30 and 30 to 100%, but the present invention is not limited thereto.

According to an embodiment, the input data input to the estimation model may include voxel number data for each section of asymmetry.

Over time from the onset of lesion development, the damage to brain tissue becomes more severe and thus the asymmetry also increases. For example, when the number of voxels having a large asymmetry (eg, rFLAIR) is large, it may be estimated that the time of lesion occurrence is far from the time of brain imaging. On the contrary, when the number of voxels with small asymmetry is large, it may be estimated that the time of lesion occurrence is close to the time of brain imaging. That is, the time point of lesion occurrence can be inferred from the information on how the asymmetry is distributed. In an embodiment of the present invention, the correlation between the distribution of asymmetry and the time of lesion occurrence is learned through machine learning, in particular, an artificial neural network.

11 is a diagram illustrating an artificial neural network structure of the estimation model in accordance with an embodiment.

The estimation model according to an embodiment of the present invention may be a model based on an artificial neural network. The artificial neural network may be a virtual network structure including an input layer, a hidden layer, and an output layer. In FIG. 11, the number of hidden layers is one, the number of nodes IL ₁ , IL ₂ , IL ₃ ,..., And IL ₃₁ of the input layer is 31, and the nodes HL ₁ , HL ₂ , HL ₃ ,. ₆ ) and the number of nodes OL ₁ and OL ₂ of the output layer is 6, but the present invention is not limited thereto.

Input data may be input to an input layer of an artificial neural network. In this case, according to an embodiment, the number of voxels for each section of rFLAIR may be input to each node IL ₁ , IL ₂ , IL ₃ ,..., IL ₃₁ of the input layer. For example, the voxels in the rFLAIR intervals of 0-1%, 1-2%, and 2-3% are respectively provided to the first node IL ₁ , the second node IL ₂ , and the third node IL ₃ of the input layer. The number can be input. Meanwhile, the number of voxels in an rFLAIR section of 30 to 100% may be input to the thirty-first node IL ₃₁ of the input layer.

Each node HL ₁ , HL ₂ , HL ₃ ,…, HL ₆ of the hidden layer of the artificial neural network is a node IL ₁ , IL ₂ , IL ₃ ,…, IL ₃₁ of the input layer and the node OL ₁ , of the output layer. Connect OL ₂ ).

The output layer of the neural network may output information about the time point of the lesion. For example, the first node OL ₁ of the output layer may output a probability that the point of occurrence of the lesion is after a specific point in time, and the second node OL ₂ of the output layer may output a probability that the point of occurrence of the lesion is before a certain point in time. However, the present invention is not limited thereto.

That is, in one embodiment of the present invention, the estimation model may be trained so that the lesion occurrence time is classified before or after a specific time point through asymmetry data for each voxel. In this case, the specific time point may be, for example, 4.5 hours, 6 hours, and the like, but the present invention is not limited thereto.

Table 1 below shows the accuracy of the estimation model that was machine-learned to estimate the time of lesion occurrence using learning input data obtained from brain image data of 240 data providers.

4.5 hours 6 hours
Number of patients (<hour)

195

218

Number of patients (> hours)

45
22

표준편차)
10-fold cross-validation accuracy
(Average

Standard Deviation)
0.83

0.04

0.89

0.07

As shown in the above table, the cross-validation verification accuracy of each estimation model trained to estimate whether the lesion occurred before 4.5 hours or 6 hours was high, about 0.83 and 0.89, respectively.

12 is a diagram comparing ROC graphs of an estimation model machine trained using signal intensity ratio (SIR) and an estimation model machine trained using rFLAIR. Each estimation model is machine-learned to determine whether a lesion occurs within 6 hours using learning input data from brain image data of 240 data providers, and then verify performance through brain image data of 49 other data providers. It was.

12A is a graph of ROC of the estimation model using SIR. SIR is defined by Equation 5 below.

<Equation 5>

는 관심 영역(ROI)의 FLAIR 신호 평균값을 나타내고,

는 대칭 영역(C_ROI)의 FLAIR 신호 평균값을 나타낸다. 즉 SIR은 병변에 해당하는 관심 영역(ROI)의 신호 '평균값'을 이용하므로, '복셀 별'로 비대칭도를 산출하지 않고 '병변 전체'의 비대칭도만을 산출한다. here,

Represents the average value of the FLAIR signal of the region of interest (ROI),

Denotes the average value of the FLAIR signal in the symmetry region C_ROI. That is, since the SIR uses a signal 'average value' of a region of interest (ROI) corresponding to the lesion, only the asymmetry of the 'whole lesion' is calculated without calculating the asymmetry for each voxel.

12 (b) is a ROC graph of an estimation model using rFLAIR according to an embodiment of the present invention. As mentioned above, rFLAIR contains 'voxel' asymmetry information of the lesion.

Referring to FIG. 12, the AUC values of the graphs (a) and (b) are 0.644 and 0.758, respectively, and the performance of the estimated model trained through the rFLAIR method using the 'by voxel' asymmetry is better. . In addition, referring to [Table 2] below, even when the machine learning to determine whether the lesion occurred within 4.5 hours, it can be seen that the AUC value of the estimation model using the rFLAIR is higher.

AUC

4.5 hours

6 hours

SIR

0.524

0.644

rFLAIR

0.647

0.758

That is, in an embodiment of the present invention, the heterogeneity information of the lesion may be obtained through 'by voxel' asymmetry data, and the heterogeneity information may be reflected in the learning of the estimation model. That is, in one embodiment of the present invention, since the lesion occurrence time is estimated including heterogeneity information of the lesion, the accuracy of the lesion occurrence time estimation is improved.

Hereinafter, an embodiment of the controller 1100 will be described in detail.

13 is a block diagram of the controller 1100 according to an exemplary embodiment.

Referring to FIG. 13, the controller 1100 according to an embodiment may include a data learner 1110 and a data recognizer 1120.

The data learner 1110 may learn a criterion for estimating a lesion occurrence time. In an embodiment, the data learner 1110 may learn a predetermined criterion by using the learning input data input to the data learner 1110. In particular, the data learner 1110 may learn the correlation between the asymmetry data of each voxel and the time of occurrence of the lesion.

The data recognizer 1120 may estimate a lesion occurrence time based on the data. The data recognizing unit 1120 may estimate a lesion occurrence time from predetermined data by using the learned data recognition model. The data recognizing unit 1120 may acquire predetermined data according to a predetermined criterion by learning, and estimate a lesion occurrence time based on the predetermined data by using a data recognition model using the acquired data as an input value. have. In addition, the result value output by the data recognition model using the acquired data as an input value may be used to update the data recognition model.

At least one of the data learner 1110 and the data recognizer 1120 may be manufactured in the form of at least one hardware chip and mounted on the electronic device. For example, at least one of the data learner 1110 and the data recognizer 1120 may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or an existing general purpose processor (eg, a CPU). Alternatively, the electronic device may be manufactured as a part of an application processor or a graphics dedicated processor (eg, a GPU) and mounted on the aforementioned various electronic devices.

In this case, the data learner 1110 and the data recognizer 1120 may be mounted on one electronic device or may be mounted on separate electronic devices, respectively. For example, one of the data learner 1110 and the data recognizer 1120 may be included in the electronic device, and the other may be included in the server. In addition, the data learner 1110 and the data recognizer 1120 may provide model information constructed by the data learner 1110 to the data recognizer 1120 via a wired or wireless connection. The data input to 1120 may be provided to the data learner 1110 as additional learning data.

Meanwhile, at least one of the data learner 1110 and the data recognizer 1120 may be implemented as a software module. When at least one of the data learning unit 1110 and the data recognizing unit 1120 is implemented as a software module (or a program module including an instruction), the software module may be computer readable non-transitory reading. It may be stored in a non-transitory computer readable media. In this case, at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an operating system (OS), and others may be provided by a predetermined application.

14 is a block diagram of a data learner 1110 according to an embodiment.

Referring to FIG. 14, the data learner 1110 may include a data acquirer 1111, a preprocessor 1112, a training data selector 1113, a model learner 1114, and a model evaluator ( 1115).

The data acquirer 1111 may receive brain image data, for example, magnetic resonance image data, from an external brain imaging apparatus or a previously stored medical database DB.

The preprocessor 1112 may preprocess the acquired data so that the acquired data may be used for learning / training for estimating the time of lesion occurrence. The preprocessor 1112 may process the acquired data into a preset format so that the model learner 1114, which will be described later, uses the acquired data for learning to estimate a lesion occurrence time. For example, the preprocessor 1112 may perform a task of selecting / generating learning input data and input data from the input brain image.

The training data selector 1113 may select data necessary for learning from the preprocessed data. The selected data may be provided to the model learner 1114. The training data selector 1113 may select data necessary for learning from preprocessed data according to a predetermined criterion. In addition, the training data selector 1113 may select data according to a predetermined criterion by learning by the model learner 1114 to be described later.

The model learner 1114 is a component that may correspond to the above-described 'estimate model' and may learn a criterion on how to determine a lesion occurrence time based on the training data. In addition, the model learner 1114 may learn a criterion about what learning data should be used to estimate a time point of lesion occurrence.

In addition, the model learner 1114 may train the data recognition model used to estimate a lesion occurrence time using the training data. In this case, the data recognition model may be a pre-built model.

The data recognition model may be constructed in consideration of the application field of the recognition model, the purpose of learning, or the computer performance of the device. The data recognition model may be, for example, a model based on a neural network. For example, a model such as a deep neural network (DNN), a recurrent neural network (RNN), and a bidirectional recurrent deep neural network (BRDNN) may be used as the data recognition model, but the present invention is not limited thereto.

According to various embodiments of the present disclosure, when there are a plurality of pre-built data recognition models, the model learner 1114 may determine a data recognition model having a high correlation between input training data and basic training data as a data recognition model to be trained. have. In this case, the basic training data may be previously classified by the type of data, and the data recognition model may be pre-built by the type of data. For example, the basic training data is classified based on various criteria such as the region where the training data is generated, the time at which the training data is generated, the size of the training data, the genre of the training data, the creator of the training data, and the types of objects in the training data. It may be.

In addition, the model learner 1114 may train the data recognition model using, for example, a learning algorithm including an error back-propagation method or a gradient descent method.

In addition, the model learner 1114 may train the data recognition model through, for example, supervised learning using learning data as an input value. In addition, the model learner 1114 learns the type of data necessary for estimating the lesion occurrence point without guidance, for example, thereby unsupervised learning to find a criterion for estimating the lesion occurrence point. ), You can train the data recognition model. In addition, the model learner 1114 may train the data recognition model through, for example, reinforcement learning using feedback on whether the lesion occurrence time estimation result according to the learning is correct.

In addition, when the data recognition model is trained, the model learner 1114 may store the trained data recognition model. In this case, the model learner 1114 may store the learned data recognition model in a memory of the electronic device including the data recognizer 1120. Alternatively, the model learner 1114 may store the learned data recognition model in a memory of an electronic device including the data recognizer 1120, which will be described later. Alternatively, the model learner 1114 may store the learned data recognition model in a memory of a server connected to the electronic device through a wired or wireless network.

In this case, the memory in which the learned data recognition model is stored may store, for example, commands or data related to at least one other element of the electronic device. The memory may also store software and / or programs.

The model evaluator 1115 may input the evaluation data into the data recognition model, and may cause the model learner 1114 to relearn if the recognition result output from the evaluation data does not satisfy a predetermined criterion. In this case, the evaluation data may be preset data for evaluating the data recognition model.

For example, the model evaluator 1115 does not satisfy a predetermined criterion when the number or ratio of the evaluation data that is not accurate among the recognition results of the learned data recognition model for the evaluation data exceeds a preset threshold. It can be evaluated as not. For example, when a predetermined criterion is defined at a ratio of 2%, when the trained data recognition model outputs an incorrect recognition result for more than 20 evaluation data out of a total of 1000 evaluation data, the model evaluator 1115 learns. The data recognition model can be evaluated as not suitable.

On the other hand, when there are a plurality of trained data recognition models, the model evaluator 1115 evaluates whether each learned data recognition model satisfies a predetermined criterion, and uses the model satisfying the predetermined criterion as the final data recognition model. You can decide. In this case, when there are a plurality of models satisfying a predetermined criterion, the model evaluator 1115 may determine any one or a predetermined number of models which are preset in the order of the highest evaluation score as the final data recognition model.

At least one of the data acquirer 1111, the preprocessor 1112, the training data selector 1113, the model learner 1114, and the model evaluator 1115 in the data learner 1110 is at least one. May be manufactured in the form of a hardware chip and mounted on an electronic device. For example, at least one of the data acquirer 1111, the preprocessor 1112, the training data selector 1113, the model learner 1114, and the model evaluator 1115 may be artificial intelligence (AI). It may be manufactured in the form of a dedicated hardware chip, or may be manufactured as a part of an existing general purpose processor (eg, a CPU or an application processor) or a graphics dedicated processor (eg, a GPU) and mounted on the aforementioned various electronic devices.

In addition, the data acquirer 1111, the preprocessor 1112, the training data selector 1113, the model learner 1114 and the model evaluator 1115 may be mounted on one electronic device or may be separate. The electronic devices may be mounted on the electronic devices. For example, some of the data acquirer 1111, the preprocessor 1112, the training data selector 1113, the model learner 1114, and the model evaluator 1115 are included in the electronic device, and the other part is Can be included on the server.

In addition, at least one of the data acquirer 1111, the preprocessor 1112, the training data selector 1113, the model learner 1114, and the model evaluator 1115 may be implemented as a software module. At least one of the data acquirer 1111, the preprocessor 1112, the training data selector 1113, the model learner 1114, and the model evaluator 1115 includes a software module (or an instruction). If implemented as a program module, the software module may be stored in a computer readable non-transitory computer readable media. In this case, at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an operating system (OS), and others may be provided by a predetermined application.

15 is a block diagram of a data recognizer 1120 according to an exemplary embodiment.

Referring to FIG. 15, the data recognizer 1120 according to an exemplary embodiment may include a data acquirer 1121, a preprocessor 1122, a recognition data selector 1123, a recognition result provider 1124, and a model updater. 1125.

The data acquirer 1121 may acquire data necessary for estimating the time of lesion occurrence, and the preprocessor 1122 may preprocess the acquired data so that the acquired data may be used. The preprocessing unit 1122 may process the acquired data into a predetermined format so that the recognition result providing unit 1124, which will be described later, may use the acquired data to estimate a lesion occurrence time.

The recognition data selector 1123 may select data required for estimating a lesion occurrence time from the preprocessed data. The selected data may be provided to the recognition result provider 1124. The recognition data selector 1123 may select some or all of the preprocessed data according to a predetermined criterion for estimating a lesion occurrence time. In addition, the recognition data selector 1123 may select data according to a preset criterion by learning by the model learner 1114 to be described later.

The recognition result providing unit 1124 may apply the selected data to the data recognition model to predict the time of lesion occurrence. The recognition result providing unit 1124 may provide a recognition result according to the recognition purpose of the data. The recognition result provider 1124 may apply the selected data to the data recognition model by using the data selected by the recognition data selector 1123 as an input value. In addition, the recognition result may be determined by the data recognition model.

The model updater 1125 may cause the data recognition model to be updated based on the evaluation of the recognition result provided by the recognition result provider 1124. For example, the model updater 1125 may allow the model learner 1114 to update the data recognition model by providing the recognition result provided by the recognition result provider 1124 to the model learner 1114. have.

Meanwhile, at least one of the data acquisition unit 1121, the preprocessing unit 1122, the recognition data selection unit 1123, the recognition result providing unit 1124, and the model updating unit 1125 in the data recognizing unit 1120 is at least one. It may be manufactured in the form of one hardware chip and mounted on an electronic device. For example, at least one of the data acquirer 1121, the preprocessor 1122, the recognition data selector 1123, the recognition result provider 1124, and the model updater 1125 may be artificial intelligence (AI). ) May be manufactured in the form of a dedicated hardware chip, or may be manufactured as a part of an existing general purpose processor (eg, a CPU or an application processor) or a graphics dedicated processor (eg, a GPU) and mounted on the aforementioned various electronic devices.

In addition, the data obtaining unit 1121, the preprocessor 1122, the recognition data selecting unit 1123, the recognition result providing unit 1124, and the model updating unit 1125 may be mounted in one electronic device or may be separate. May be mounted on the electronic devices. For example, some of the data obtaining unit 1121, the preprocessing unit 1122, the recognition data selecting unit 1123, the recognition result providing unit 1124, and the model updating unit 1125 are included in the electronic device, and some of the remaining units are included in the electronic device. May be included in the server.

In addition, at least one of the data acquirer 1121, the preprocessor 1122, the recognition data selector 1123, the recognition result provider 1124, and the model updater 1125 may be implemented as a software module. At least one of the data obtaining unit 1121, the preprocessing unit 1122, the recognition data selecting unit 1123, the recognition result providing unit 1124, and the model updating unit 1125 includes a software module (or an instruction). When implemented as a program module, the software module may be stored in a computer-readable non-transitory computer readable media. In this case, at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an operating system (OS), and others may be provided by a predetermined application.

Meanwhile, the method for estimating lesion occurrence time according to an embodiment of the present invention may be written as a computer executable program, and may be implemented in a general-purpose digital computer operating the program using a computer readable recording medium. . The computer-readable recording medium may include a storage medium such as a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.) and an optical reading medium (eg, a CD-ROM, a DVD, etc.).

According to the method, apparatus, or program for estimating lesion occurrence time according to an embodiment of the present invention, it is possible to estimate the lesion occurrence time in an objective and quantitative manner, so that the difference between the observer and the observer when estimating the lesion occurrence time in a qualitative manner. The problem of high intra- and inter-observer variability can be solved. In addition, the onset of lesions can be quickly identified within minutes, which can help to quickly determine treatment options for the subject. Through this, it is possible to rapidly treat cerebrovascular occlusion, which can irreversibly damage the brain, such as cerebral infarction, thereby increasing the therapeutic effect.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

1000: lesion occurrence time estimation apparatus 1100: control unit
1200: output portion L: lesion area
ROI: region of interest C_ROI: region of symmetry
E_ROI: extended area

Claims (10)

A region of interest selection step of selecting a region of interest corresponding to a lesion location of the subject from brain image data of the subject;
A symmetric region selection step of selecting a symmetric region symmetrical with the region of interest;
Calculating asymmetry data for each voxel of the region of interest by using the region of interest and the symmetry region;
An input step of acquiring input data from the asymmetry data for each voxel and inputting it into an estimation model for estimating the occurrence time of the lesion; And
And outputting information on a time point of occurrence of the lesion based on an output value of the estimation model.
The estimation model,
It is an input / output function expressing the correlation between the learning input data obtained from brain image data of multiple data providers and the time of lesion occurrence,
The asymmetry calculation step,
If the size of the region of interest is less than or equal to the predetermined size,
Selecting an extended area having a larger size than the symmetric area; And
And calculating a difference between a signal magnitude of each voxel of the region of interest and an average value of all voxel signals of the extended region.
The method of claim 1,
The estimation model is
And an input / output function for outputting a probability that the point of occurrence of the lesion is before or after the predetermined point in time of the brain image data.
The method of claim 1,
Before the step of selecting the region of interest,
And training the estimated model.
The method of claim 1,
The region of interest selection step,
A data acquisition step of acquiring first brain image data and second brain image data of the subject;
Registering the first brain image data and the second brain image data;
Selecting, from the first brain image data, a lesion area in which the lesion of the subject is estimated to be located; And
And selecting a region matching the lesion region as the region of interest in the second brain image data.
The method of claim 4, wherein
The data acquisition step,
Obtaining the first brain image data from DWI data;
And acquiring the second brain image data from liquid attenuation reverse recovery image (FLAIR) data.
The method of claim 1,
Before the symmetric area selection step,
Comprising the step of calculating the symmetry line of the brain image data, lesion occurrence time estimation method.
The method of claim 1,
The asymmetry calculation step,
And calculating a difference between a signal size of each voxel of the region of interest and an average value of all voxel signals of the symmetric region.
delete From the brain image data of the subject, selecting a region of interest corresponding to the lesion location of the subject, selecting a symmetric region symmetrical to the region of interest, using the region of interest and the symmetric region asymmetry of the voxel of the region of interest Calculates data, obtains input data from the voxel-specific asymmetry data from the voxel-specific asymmetry data, and inputs it to an estimation model for estimating the time of occurrence of the lesion,
A controller configured to select an extended area having a larger size than the symmetrical area when the size of the ROI is equal to or less than a predetermined size and calculate a difference between a signal size of each voxel of the ROI and an average value of all voxel signals of the extended area; And
And an output unit configured to output information about a time point of occurrence of the lesion, based on an output value of the estimation model.
The estimation model,
A lesion occurrence time estimation apparatus, which is an input / output function representing a correlation between learning input data obtained from brain image data of a plurality of data providers and a lesion occurrence time point.
A computer program stored in a medium for carrying out the method of claim 1 using a computer.

Images