KR20200025853A - Method for generating predictive model based on intra-subject and inter-subject variability using structural mri - Google Patents

Abstract

Disclosed are a method for generating an intra- and inter-subject variability-based prediction model using structural magnetic resonance imaging (MRI) which is capable of generating a prediction model with increased performance and reliability and a device thereof. According to the present invention, the method uses the structural MRI to generate a prediction model. The method comprises: a step of performing structural processing which uses a plurality of pieces of scan data for each of a plurality of subjects acquired through a plurality of times of structural MRI scan for each of the plurality of subjects to acquire an anatomic measurement value for each subject; a step of using the anatomic measurement result acquired for each of the plurality of subjects to perform analysis of variance (ANOVA) of checking intra-subject variability and inter-subject variability; and a step of using a predication machine learning model for a feature region checked by the intra-subject variability and the inter-subject variability to generate a corresponding prediction model.

Description

METHODS FOR GENERATING PREDICTIVE MODEL BASED ON INTRA-SUBJECT AND INTER-SUBJECT VARIABILITY USING STRUCTURAL MRI}

The present invention relates to a method and apparatus for generating a prediction model based on intra- and inter-subject variability using structural magnetic resonance imaging.

In general, a medical imaging apparatus is a device that provides an image by acquiring biological information of a patient. Medical imaging apparatuses include an ultrasonic diagnostic apparatus, an X-ray tomography apparatus, a magnetic resonance imaging apparatus, and a medical diagnostic apparatus. Among these, magnetic resonance imaging devices are relatively free in imaging conditions, and provide an excellent contrast in soft tissue and various diagnostic information images, thus occupying an important position in the diagnosis field using medical imaging.

The magnetic resonance imaging (MRI) used in the magnetic resonance imaging apparatus is a nuclear magnetic resonance phenomenon in the hydrogen atom nucleus in the body by using a magnetic field and non-electromagnetic radiation that is harmless to the human body, the density of the atomic nucleus And physicochemical properties. Among the MRI, structural MRI is used to measure the thickness of the brain cortex, and functional MRI is used to measure brain efficiency by taking blood that passes through the brain.

Structural MRI data, such as T1 or T2 weighted images, consist of hundreds of thousands of voxels (3D pixels) per individual. These images generally yield six measurements, such as the thickness, volume, and area of the cerebral cortex in the analysis. There is a strong demand for the ability to select relevant functions from such large datasets when generating predictive models of psychological and cognitive characteristics in the field of neural imaging. Researchers and health workers now leave models with many unrelated predictors. This leads to the problem that the model is expensive to calculate, slow and difficult to understand.

Cortechs Labs focuses on predicting medical outcomes from anatomical MRI images. They calculate anatomical measurements in a similar way to ours, but leave it to the doctor to interpret the results. They do not do any kind of behavioral modeling or prediction.

On the other hand, 12 sigma generally focuses on predicting the medical outcome of medical imaging. Although their website doesn't provide much information, it's almost impossible to perform brain imaging because it uses machine vision technology.

State-of-the-art neuroimaging documents using state-of-the-art machine learning techniques suggest that the accuracy of predictions for validation datasets is approximately 70% at best, for example, to distinguish between normal and autistic patients based solely on MRI data. . These results should be improved for medical standards. Using complex technologies with large datasets makes these technologies slow and difficult to interpret.

Therefore, there is a need for a method that simplifies the predictive model of psychological and cognitive characteristics, thereby making modeling much faster and easier to interpret.

The problem to be solved by the present invention is to provide a method and apparatus for generating a predictive model that enables the selection of a very reliable measurement through a plurality of scans per subject.

In addition, since it is possible to greatly reduce the complexity of the generated prediction model and improve the statistical power, to provide a method and apparatus for generating a prediction model having the effect of increasing the performance and reliability of the prediction model.

Prediction model generation method according to an aspect of the present invention,

A method of generating a predictive model by using a magnetic resonance imaging (MRI) by the apparatus for generating a predictive model, wherein the plurality of subjects receive a plurality of pieces of scan data acquired through a plurality of subjects' structural MRI scans. Performing structural processing to obtain anatomical measurements for each subject, using intra-subject variability and inter-subject variability using anatomical measurements obtained for each of the plurality of subjects. performing an analysis of variance (ANOVA) to identify inter-subject variability, and corresponding predictive models using predictive machine learning models for feature regions identified by intra- and inter-subject variability Generating a step.

Here, the feature region is an area corresponding to a measurement value having low intra-subject variability and high inter-subject variability among the intra-subject variability and the inter-subject variability identified by the variance analysis, wherein the low intra-subject variability is set in advance. The high threshold of inter-subject variability is determined by a first preset threshold.

The performing of the structural processing may include aligning the T1 image and the T2 image of each subject by using the plurality of subject-specific scan data, and having magnetic field heterogeneity and defects and tissue boundaries of the scanning coil. A distortion correction step to correct magnetic field susceptibility, a brain extraction step to remove non-brain tissue using the distortion corrected scan data, and normalize the intensity of the image from which the non-brain tissue has been removed Intensity standardization step for separating gray matter, white matter and cerebrospinal fluid, a splitting step for dividing the white matter and gray matter into anatomical sections from the intensity normalized image, and using the segmented data A surface modeling step of generating a surface model of the boundary between the gray and white matter and the edges of the brain, and using the surface model And a measurement value calculating step for calculating a group anatomical measurements.

In addition, the step of calculating the measurement may include generating surface representations of internal and external cortical boundaries using the surface model, dividing the gray matter of the brain into different sections, and various cortical and subcortical parts ( calculating the anatomical measurement for the subcortical region.

The anatomical measurements also include thickness, area, volume, and myelin (myelin) content for the cortical and subcortical areas.

The predictive machine learning model also includes a support-vector machine and a support-vector regression model.

Predictive model generation device according to another aspect of the present invention,

Structural processing unit for obtaining anatomical measurement values for each subject using a plurality of subjects' plurality of scan data acquired through a plurality of structural MRI scans for a plurality of subjects, and anatomically obtained for each of the plurality of subjects A variance analysis unit that performs variance analysis using measurements to identify intra-subject variability and inter-subject variability, and features identified by the subject's variability and inter-subject variability It includes a machine learning modeling unit for generating a corresponding prediction model using a predictive machine learning model for the region.

The apparatus may further include a feature selector which selects the feature region to be used for generating the predictive model and transmits the feature region to the machine learning modeling unit. An area corresponding to a measurement value having intra-subject variability lower than a set first threshold value and having inter-subject variability higher than a second predetermined threshold value among the inter-subject variability is selected as the feature area.

In addition, the structural processor may be configured to align the T1 image and the T2 image of each subject by using the plurality of subject-specific scan data, and to generate magnetic field inhomogeneity and magnetic field susceptibility at defects and tissue boundaries of the scanning coil. ), A brain extractor for removing non-brain tissue using the distortion correction corrected by the distortion corrector, and the non-brain tissue by the brain extractor. The intensity standardization section for standardizing the intensity of the removed image and separating gray matter, white matter and cerebrospinal fluid, and the white matter and gray matter from the intensity normalization image by the intensity standardization section, respectively, section), the boundary between the gray and white matter and the surface edge of the brain using the data divided by the partition. A surface modeling unit for generating a Dell, and a measurement calculation step of calculating the anatomical measurement using the surface model generated by the surface modeling unit.

The anatomical measurements also include thickness, area, volume, and myelin content for the cortical and subcortical areas.

Predictive model generation apparatus according to another aspect of the present invention,

A predictive model generating device for generating a predictive model using structural magnetic resonance imaging (MRI), comprising an input unit, a memory, and a processor, wherein the input unit is obtained through a plurality of structural MRI scans for a plurality of subjects. Receiving a plurality of scan data for each subject, wherein the memory is configured to store a set of codes, and the codes are anatomically for each subject using a plurality of subject-specific scan data input through the input device. Obtaining variance analysis, variance analysis identifying intra-subject variability and inter-subject variability using anatomical measurements obtained for each of the plurality of subjects To the feature area identified by the action to be performed and the subject's variability and the subject's variability. And use the predictive machine learning model to control the processor to execute the operation of generating a corresponding predictive model.

Here, the operation of acquiring the anatomical measurement value, aligns the T1 image and the T2 image of each subject by using the plurality of scan data of each subject, the magnetic field heterogeneity and defects and tissue boundaries of the scanning coil. Correcting magnetic field susceptibility in the brain, removing non-brain tissue using the corrected scan data, normalizing the intensity of the image from which the non-brain tissue is removed, separation of gray matter, white matter and cerebrospinal fluid, division of white matter and gray matter into anatomical sections, respectively, from intensity standardized images, boundary between gray matter and white matter using segmented data And generating a surface model of the edge of the brain, and calculating the anatomical measurements using the surface model. All.

The feature region may be selected as an area corresponding to a measurement value having intra-subject variability lower than a preset first threshold value among the intra-subject variability and having a higher inter-subject variability value than the second preset threshold value among the inter-subject variability. do.

According to an embodiment of the present invention, a plurality of scans per subject allow a selection of highly reliable measurements.

In addition, since the structure of the subject can be stably measured and the brain regions that can be significantly different between subjects can be selected, the complexity of the predictive model generated can be greatly reduced and the statistical power can be improved, thereby increasing the performance and reliability of the predictive model. Has

Objective analysis of intra-subject and inter-subject variability allows the selection of relevant functions to create a simpler, faster and easier to interpret model. This allows you to explore the behavior of the model more broadly and deeply, and to obtain more meaningful models that produce better predictions.

1 is a schematic configuration diagram of an apparatus for generating a predictive model according to an embodiment of the present invention.
FIG. 2 is a detailed configuration diagram of the structural processing unit shown in FIG. 1.
3A and 3B show examples of a T1 image and a T2 image scanned for one subject.
FIG. 4 is a diagram illustrating an example in which the cortex is highlighted in white on the top of the T1 image of FIG. 3A by the dividing unit shown in FIG. 2.
FIG. 5 is a diagram illustrating a cortex in which myelin content is indicated using the image shown in FIG. 3.
FIG. 6 shows the cortical regions shown in white at the top of the myelin content of the cortex in the image of FIG. 5.
7 is a schematic flowchart of a method of generating a predictive model according to an embodiment of the present invention.
8 is a diagram schematically illustrating a concept of predictive model generation according to an embodiment of the present invention.
FIG. 9 is a diagram illustrating an example of intra-subject variability and inter-subject variability using two subjects in the method of generating a predictive model according to an embodiment of the present invention.
FIG. 10 shows a comparison of values between six scans of two different subjects for one particular volume measurement, especially for the lower parietal lobe, for a single ROI.
11 is a detailed flowchart of a process of processing a structure for each subject in a method of generating a prediction model according to an embodiment of the present invention.
12 is a schematic structural block diagram of an apparatus for generating a predictive model according to another embodiment of the present invention.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted for simplicity of explanation, and like reference numerals designate like parts throughout the specification.

Throughout the specification, when a part is said to "include" a certain component, it means that it can further include other components, except to exclude other components unless specifically stated otherwise.

In addition, the terms “… unit”, “… unit”, “module”, etc. described in the specification mean a unit that processes at least one function or operation, which may be implemented by hardware or software or a combination of hardware and software. have.

In the present invention, the use of intra subject and inter subjects variability allows the selection of anatomical treatments from MRI when generating models of psychological and cognitive characteristics. Specifically, by selecting brain regions that can be reliably measured within an individual and that can vary greatly between individuals, the complexity of such models can be greatly reduced and statistical power can be improved. In other words, the relevant features can be selected by simple and objective means, allowing a simpler model to be constructed, which makes it easier to interpret, faster to execute, more effective, more reliable, and more statistical. It can be a powerful model.

Hereinafter, an apparatus for generating a prediction model based on intra- and inter-subject variability according to an embodiment of the present invention will be described with reference to the accompanying drawings.

1 is a schematic configuration diagram of an apparatus for generating a predictive model according to an embodiment of the present invention.

As illustrated in FIG. 1, the predictive model generating apparatus 100 according to an exemplary embodiment of the present invention may include an input unit 110, a structural processor 120, a variance analyzer 130, a feature selector 140, and machine learning. The modeling unit 150 is included.

In an embodiment of the present invention, it is assumed that prediction model generation is performed using a plurality of scan data for each subject obtained by performing a plurality of T1 and T2 weighted MRI scans for many different subjects.

The input unit 110 receives a plurality of scan data for each of a plurality of subjects. For example, two scan data obtained through at least two scans are used to obtain information about intra-subject variability for each subject. Therefore, the input unit 110 receives two scan data for each subject. For example, in the case of ten subjects, a total of ten scan data are input in the case of two scan data for each of ten subjects.

The input unit 110 is directly connected to a structural MRI (not shown) or through a network to receive data scanned for each subject through the structural MRI directly from the structural MRI or a plurality of subject scan data through a storage medium or the like. Can be input.

The structural processor 120 obtains various anatomical measurement values for each subject by using a plurality of subject-specific scan data input through the input unit 110. These various anatomical measurements can be measurements such as thickness, area, volume and myelin (myelin) content for various cortical and subcortical areas. The structural processing unit 120 will be described in detail later.

The variance analyzer 130 performs an analysis of variance (ANOVA) using the measured values measured for each subject in the structural processor 120 to check variability in the subject and variability between the subjects. In this case, the analysis of variance refers to a data analysis method in which the difference between group means is compared to the variation in the group when the measured data are divided into several groups. Here, the variance analysis unit 130 compares a plurality of scans of a plurality of subjects through a variance analysis test for all the measured values measured by the structural processing unit 120. That is, the variance analyzer 130 may test, for each measurement and each region, whether the variability of the measurement through the scan of a single subject is greater than the variability of the measurement through the scan of different subjects. As such, the variance analyzer 130 may identify a region having a variable characteristic through a plurality of scans of a single subject.

The feature selector 140 selects a feature region to be used for generating a prediction model by using the analysis result of the variance analyzer 130. Specifically, the feature selector 140 selects a measurement value having low variability in the subjects and high variability between the subjects. That is, the selection of the measurement value having low intra-subject variability and high inter-subject variability by the feature selector 140 corresponds to selection of a brain region in which the structure in the subject is stably measured and can be greatly different between the subjects. Here, the low variability in the subject to be selected as the feature region is determined by the first preset threshold value, and the high variability between the subjects is determined by the second preset threshold value, which is the first threshold value and the second threshold value. The threshold can be determined through a large number of experiments and studies.

The machine learning modeling unit 150 generates a predictive model corresponding to the measured value selected by the feature selector 140 using the predictive machine learning model. The predictive machine learning model may be a support-vector machine, a support-vector regression model, or the like. Techniques for generating predictive models by performing machine learning on measured values using predictive machine learning models are well known, and thus detailed descriptions are omitted here.

As such, according to embodiments of the present invention, it is possible to select highly reliable measurements through a plurality of scans per subject, and also that the structure can be stably measured within subjects and can vary greatly between subjects. Since the complexity of the predictive model generated by selecting the brain region can be greatly reduced and the statistical power can be improved, the predictive model has the effect of increasing the performance and reliability.

FIG. 2 is a detailed configuration diagram of the structural processor 120 shown in FIG. 1.

As shown in FIG. 2, the structural processor 120 includes a distortion corrector 121, a brain extractor 122, an intensity standardizer 123, a divider 124, a surface modeler 125, and a measured value. The calculation unit 126 is included.

The distortion correcting unit 121 aligns T1 and T2 images for each subject by using scan data input from the input unit 110, and generates magnetic field heterogeneity and magnetic field susceptibility at defects and tissue boundaries of the scanning coil. To correct. 3A and 3B show examples of a T1 image and a T2 image scanned for one subject.

The brain extractor 122 removes non-brain tissue such as the neck, mouth, meninges, and skull using the scan data distortion-corrected by the distortion corrector 121.

The intensity standardizer 123 normalizes the intensity of the image from which the non-brain tissue is removed from the brain extractor 122 and separates gray matter, white matter and cerebrospinal fluid.

The divider 124 divides the white matter and the gray matter into predetermined anatomical sections, respectively, from the intensity normalized image by the intensity standardizer 123. Here, the anatomical sections used for dividing the white and gray matter are already well known and thus detailed descriptions thereof are omitted here. 4 is a diagram illustrating an example in which the cortex is highlighted in white on the top of the T1 image of FIG. 3A by the division unit 124.

The surface modeling unit 125 generates surface models of the boundary between the gray and the white matter and the edges of the brain (internal and external cortical boundaries) using the result data divided by the dividing unit 124.

The measurement calculation unit 126 uses the surface model generated by the surface modeling unit 125 to generate surface representations of the inner and outer cortical boundaries, and the gray matter (cortex and subcortex) of the brain into different sections. After partitioning, measurements such as thickness, area, volume and myelin content for various cortical and subcortical areas are calculated.

Hereinafter, with reference to the drawings will be described an example of calculating the myelin content for a particular area of the subject's scan image.

First, it is shown in FIG. 4 that the cortex is highlighted in white on top of the T1 image shown in FIG. 3.

The T1 image shown in FIG. 4 can be used to obtain a T1 image representing the cortex in which the myelin content is indicated as shown in FIG. 5. The myelin content shown in FIG. 5 is calculated by keeping only the cortex in the T1 and T2 images, partitioning the resulting T1 image of the cortex into the resulting T2 image of the cortex, and calculating the natural log of the result in each voxel. In FIG. 5, the yellow voxel is more numerous and less in the red voxel.

FIG. 6 shows the cortical regions shown in white at the top of the myelin content of the cortex in the image of FIG. 5. These images can be used to calculate the average myelin content in each region.

Hereinafter, a prediction model generating method according to an embodiment of the present invention will be described with reference to the drawings.

7 is a schematic flowchart of a method of generating a predictive model according to an embodiment of the present invention, and FIG. 8 is a view schematically illustrating a concept of generating a predictive model according to an embodiment of the present invention. The predictive model generating method in FIG. 7 may be performed by the predictive model generating apparatus described with reference to FIGS. 1 and 2.

7 and 8, first, a plurality of scan data for a plurality of subjects are input (S100). For example, as shown in Fig. 8, a plurality of subjects, namely, subjects 1,... For example, two pieces of scan data obtained through at least two scans are used to obtain information about the subject's variability per subject n.

Next, various anatomical measurement values are obtained for each test subject by performing the structure processing for each test subject using a plurality of scan data for each test subject (S110). The various anatomical measurements herein may be measurements such as thickness, area, volume and myelin content for various cortical and subcortical areas. The structural processing will be described later in detail.

Next, it is determined whether the structural processing of the step (S110) has been completed for all the subjects (S120), and if there is a subject who has not been subjected to the structural processing, the step (S110) is repeatedly performed for the subject. That is, the structural processing of step S110 should be performed for all the subjects.

Thereafter, variance analysis is performed using the measured values measured for all the subjects, and the variability in the subject and the variability between the subjects are confirmed for each subject (S130). In this step, for each measurement and each region, it may be determined whether the variability of the measurement through the scan of a single subject is greater than the variability of the measurement through the scans of different subjects. As shown in Fig. 8, a plurality of subjects, namely, subjects 1,... In other words, variance analysis is performed using all of the measured values for which the structural treatment was performed for each subject n.

Referring to FIG. 9, two scans 4-1, 4-2, 7-1 and 7-3, respectively, for two different subjects 4 and 7 for one specific volumetric measurement. Intra-subject variability and inter-subject variability, as confirmed by variance analysis as described above, after)) are shown. Each point in FIG. 9 represents values for some regions of interest (ROI). 9, it can be seen that scattering between scans of the same subject is much lower than scattering of scans of other subjects. This is an example of a measurement to be used for predictive modeling.

FIG. 10 shows a comparison of values between six scans of two different subjects (4, 7) for one particular volume measurement for a single ROI, in particular the volume of the inferior-parietal ROI). . In FIG. 10, the two distinct clusters of dots indicate the results of the variability check, which indicates low intra-subject variability and high inter-subject variability.

Next, a feature region to be used for generating a prediction model is selected using the result of the analysis of variance in step S130 (S140). Specifically, measurements with low variability in subjects and high variability between subjects are selected.

Finally, a predictive model corresponding to the measured value selected in the step S140 using the predictive machine learning model is generated (S150). As mentioned above, the predictive machine learning model may be a support vector machine, a support vector regression model, or the like.

Next, FIG. 11 is a detailed flowchart of a process of processing a structure for each subject in a prediction model generation method according to an embodiment of the present invention.

Referring to FIG. 11, first, the T1 and T2 images of each subject are aligned using a plurality of input data of each subject, and the magnetic field inhomogeneity and magnetic field sensitivity at the defect and tissue boundary of the scanning coil ( A distortion correction process for correcting susceptibility is performed (S111).

Next, a brain extraction process of removing non-brain tissue such as neck, mouth, meninges, and skull using the distortion-corrected scan data is performed (S112).

Thereafter, an intensity standardization process of standardizing the intensity of the image from which the non-brain tissue has been removed and separating gray matter, white matter and cerebrospinal fluid is performed (S113).

Next, a segmentation process of dividing the white matter and the gray matter into predetermined anatomical sections from the intensity normalized image is performed (S114).

Subsequently, a surface modeling process is performed to generate a surface model of the boundary between the gray and white matter and the edges of the brain (internal and external cortical boundaries) using the divided result data (S115).

Finally, surface models are used to create surface representations of the inner and outer cortical boundaries, divide the gray matter (cortex and subcortex) of the brain into different sections, and then the various cortical and subcortical A measurement calculation process for calculating measurement values such as thickness, area, volume, and myelin content for the area is performed (S116).

Next, a prediction model generating apparatus according to another embodiment of the present invention will be described.

12 is a schematic structural block diagram of an apparatus for generating a predictive model according to another embodiment of the present invention.

Referring to FIG. 12, the predictive model generating apparatus 200 according to another embodiment of the present invention includes an inputter 210, a memory 220, a processor 230, and a bus 240.

The input unit 210 receives a plurality of scan data for each subject obtained by performing a plurality of T1 and T2 weighted MRI scans for different subjects using the structural MRI.

Memory 220 is configured to store a set of code, which code is used to control processor 230 to perform the following operations. Such an operation may include receiving a plurality of pieces of scan data for a plurality of subjects through the input unit 210 and performing structure processing for each subject using a plurality of pieces of scan data for a plurality of subjects, thereby generating various anatomical measurement values for each subject. Performing variance analysis using the acquired operation, the measured values measured by the plurality of subjects, and confirming the variability in the subject and the variability between the subjects for each subject, and selecting the feature region to be used for generating the predictive model using the variance analysis results. And generating a corresponding prediction model by performing predictive machine learning modeling on the measured values of the selected specific region.

The operation of acquiring a variety of anatomical measurement values for each subject by performing the structure processing for each subject by using the plurality of scan data for each subject may be performed for each subject using a plurality of input data for each subject. Distortion correction operation to align T1 and T2 images, correct for magnetic field inhomogeneity and magnetic field susceptibility at defects and tissue boundaries of the scanning coil, neck, mouth, meninges, skull using distortion corrected scan data Brain extraction operation to remove non-brain tissue, such as normalization of the intensity of the non-brain tissue removed image, intensity to separate gray matter, white matter and cerebrospinal fluid Standardization operation, segmentation operation for dividing white and gray matter into predetermined anatomical sections, respectively, from the normalized image Surface modeling behavior to create a surface model of the boundary between the gray and white matter and the edges of the brain (internal and external cortical boundaries), using surface models to create surface representations of the internal and external cortical boundaries, Measurements that divide gray matter (cortex and subcortex) into different sections and then calculate measurements such as thickness, area, volume, and myelin content for various cortical and subcortical areas. Calculation operation.

Here, the feature region corresponds to a measurement value in which the variability in the subjects is lower than the first predetermined threshold and the variability between the subjects is higher than the preset second threshold.

In the present invention, selecting the relevant function through the objective analysis of the intra-subject and inter-subject variability can make the model simpler, faster and easier to interpret. This allows you to explore the behavior of the model more broadly and deeply, and to obtain more meaningful models that produce better predictions.

The present invention can be applied in many fields. Above all, it will be useful in the field of education for children and youth. Knowing what behavior trends a student will have in the future can help teachers and parents better inform the student about the best way to plan an educational program. It will also be useful in medicine, with specific applications in psychiatry and neurology, and will teach doctors and psychologists how to best diagnose and treat patients. There is potentially a much wider range of applications, such as in criminal justice systems, in government agencies and in other companies.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

Claims (13)

A method of generating a prediction model by using a structural magnetic resonance imaging (MRI), the prediction model generating device,
Performing a structural process of acquiring an anatomical measurement value for each subject using a plurality of subjects' plurality of scan data obtained through a plurality of structural MRI scans for each of a plurality of subjects,
Perform an analysis of variance (ANOVA) to identify intra-subject variability and inter-subject variability using anatomical measurements obtained for each of the plurality of subjects To do, and
Generating a corresponding predictive model using a predictive machine learning model for the feature region identified by the intra-subject variability and the inter-subject variability
Prediction model generation method comprising a. The method of claim 1,
The feature region corresponds to a measurement value having low intra-subject variability and high inter-subject variability among the intra-subject variability and the inter-subject variability identified by the analysis of variance,
The low of the subject variability is determined by a first preset threshold, the high of the inter-subject variability is determined by a second preset threshold,
How to generate a predictive model. The method of claim 2,
Performing the structural processing,
A distortion correction step of aligning the T1 image and the T2 image of each subject by using the plurality of scan data of each subject and correcting magnetic field inhomogeneity and magnetic field susceptibility at defects and tissue boundaries of the scanning coil. ,
Brain extraction step using distortion-corrected scan data to remove non-brain tissue,
An intensity normalization step of normalizing the intensity of the image from which the non-brain tissue has been removed and separating gray matter, white matter and cerebrospinal fluid,
A segmentation step of dividing the white matter and gray matter into anatomical sections, respectively, from an intensity normalized image,
A surface modeling step that uses segmented data to generate a surface model of the boundary between the gray and white matter and the edges of the brain, and
A measurement calculation step of calculating the anatomical measurement using the surface model
A prediction model generation method comprising a. The method of claim 3,
The measured value calculation step,
Generating surface representations of internal and external cortical boundaries using the surface model,
Dividing the gray matter in the brain into different sections, and
Calculating the anatomical measurements for various cortical and subcortical areas
A prediction model generation method comprising a. The method of claim 1,
The anatomical measurements include thickness, area, volume and myelin content for the cortical and subcortical areas,
How to generate a predictive model. The method of claim 1,
The predictive machine learning model includes a support-vector machine and a support-vector regression model,
How to generate a predictive model. A structural processor for acquiring an anatomical measurement value for each subject using a plurality of subjects' plurality of scan data acquired through a plurality of times of structural MRI scans for a plurality of subjects,
A variance analysis unit for performing variance analysis to determine intra-subject variability and inter-subject variability using anatomical measurements obtained for each of the plurality of subjects, and
Machine learning modeling unit for generating a corresponding prediction model using a predictive machine learning model for the feature region identified by the subject's intra-subject variability and the inter-subject variability
Predictive model generation apparatus comprising a. The method of claim 7, wherein
The apparatus may further include a feature selector which selects the feature region to be used for generating the predictive model and transmits the feature region to the machine learning modeling unit.
The feature selector may include an area corresponding to a measurement value having a lower intra-subject variability among the intra-subject variability and having a higher inter-subject variability than the second predetermined threshold value among the inter-subject variability. Selected with,
Predictive model generation device. The method of claim 8,
The structural processing unit,
Distortion correction unit to align the T1 image and T2 image for each subject using the plurality of scan data for each subject, and to correct the magnetic field inhomogeneity and magnetic field susceptibility at the defect and tissue boundary of the scanning coil ,
A brain extracting unit which removes non-brain tissue using the scan data distortion-corrected by the distortion correcting unit,
An intensity standardizing unit for standardizing the intensity of the image from which the non-brain tissue is removed by the brain extracting unit and separating gray matter, white matter and cerebrospinal fluid;
A division unit for dividing the white matter and gray matter into anatomical sections, respectively, from the intensity normalized image by the intensity standardization unit,
A surface modeling unit for generating a surface model of the boundary between the gray and the white matter and the edge of the brain using the data divided by the dividing unit, and
A measurement value calculation step of calculating the anatomical measurement value using the surface model generated by the surface modeling unit
Predictive model generation apparatus comprising a. The method of claim 7, wherein
The anatomical measurements include thickness, area, volume and myelin content for the cortical and subcortical areas,
Predictive model generation device. Predictive model generating apparatus for generating a predictive model using structural magnetic resonance imaging (MRI),
It includes an input method, memory, and processor
The input unit receives a plurality of scan data for a plurality of subjects obtained through a plurality of structural MRI scans for a plurality of subjects,
The memory is configured to store a set of codes,
The code is
Acquiring an anatomical measurement value for each subject by using a plurality of subject's scan data input through the input device;
Performing variance analysis to identify intra-subject variability and inter-subject variability using anatomical measurements obtained for each of the plurality of subjects, and
Generating a corresponding predictive model using a predictive machine learning model for the feature region identified by the intra-subject variability and the inter-subject variability
Used to control the processor to run,
Predictive model generation device. The method of claim 11,
Acquiring the anatomical measurement value,
Aligning the T1 image and the T2 image of each subject by using the plurality of subject-specific scan data, and correcting magnetic field heterogeneity and magnetic field susceptibility at defects and tissue boundaries of the scanning coil;
Removing non-brain tissue using the calibrated scan data,
Normalizing the intensity of the image from which the non-brain tissue has been removed, and separating gray matter, white matter and cerebrospinal fluid,
Splitting white and gray matter into anatomical sections, respectively, from an intensity normalized image,
Use the segmented data to generate a surface model of the boundary between the gray and white matter and the edges of the brain, and
Calculate the anatomical measurements using the surface model
Predictive model generation apparatus comprising a. The method of claim 11,
Wherein the feature region is selected as an area corresponding to a measured value having a lower intra-subject variability among the intra-subject variability and having a higher inter-subject variability than the preset second threshold among the inter-subject variability,
Predictive model generation device.

Images