WO2024014724A1

WO2024014724A1 - Method and apparatus for predicting brain sex and brain age from brain image

Info

Publication number: WO2024014724A1
Application number: PCT/KR2023/008278
Authority: WO
Inventors: 류인균; 윤수정; 주윤지; 황재욱
Original assignee: 이화여자대학교 산학협력단; 순천향대학교 산학협력단
Priority date: 2022-07-14
Filing date: 2023-06-15
Publication date: 2024-01-18
Also published as: KR20240009795A

Abstract

Disclosed are a method and apparatus for predicting brain sex and brain age from a brain image. The method for predicting brain sex and brain age according to one embodiment of the present invention may comprise the steps of: identifying a brain image from a database; preprocessing the brain image so as to be identifiable by a brain sex prediction model and a brain age prediction model; using the brain sex prediction model to predict brain sex characteristics used to determine brain sex from the preprocessed brain image; and using the brain age prediction model to predict brain age based on the brain sex characteristics from the preprocessed brain image and the predicted brain sex characteristics.

Description

Method and device for predicting brain gender and brain age from brain imaging

The present invention relates to a method and device for predicting brain sex and brain age from brain imaging. The present invention relates to a method of learning a brain sex prediction model and a brain age prediction model.

Recently, the ‘human brain mosaic’ theory was proposed, which states that characteristics of both genders are found simultaneously in one person’s brain. Therefore, the importance of quantifying gender characteristics of the brain and nervous system is increasing. In particular, gender-specific structural characteristics in the central nervous system are not only directly related to its function, but can also lead to gender differences in the likelihood of developing mental and neurological diseases, age of onset, and symptoms. Therefore, it is important to identify brain sex characteristics.

Additionally, although people's brain aging shares the same pattern, there are individual differences in the detailed aspects. Based on this, brain health can be determined by calculating an individual's brain age by quantifying the degree to which the brain aging pattern deviates from the average brain aging pattern of the same age group. Additionally, an individual's brain age can be used as a biomarker to evaluate the potential risk of brain aging and related diseases.

Moreover, it is known that there are gender differences in biological aging patterns, including brain aging, and that gender is a major risk factor for the development of mental and neurological diseases related to aging. As there are gender differences in brain aging patterns, it is necessary to predict brain age that reflects the characteristics of each brain gender.

The present invention provides a method and device for determining a subject's brain age from structural brain images and brain gender characteristics that have undergone minimal preprocessing using a brain age prediction model, which is a hybrid deep learning model combining a convolutional neural network model and a multilayer perceptron model. to provide. Additionally, the present invention provides a method of learning a brain age prediction model that determines the subject's brain age using brain imaging.

A method for predicting brain sex and brain age according to an embodiment of the present invention, comprising: identifying a brain image from a database; Preprocessing the brain image so that a brain sex prediction model and a brain age prediction model can be identified; Predicting brain sex characteristics used to determine brain sex from a brain image for which the brain sex prediction model has been preprocessed; And the brain age prediction model may include a step of predicting brain age based on the brain gender characteristics from the preprocessed brain image and predicted brain gender characteristics.

The preprocessing step includes matching the voxel size and voxel volume of the brain image to a standard template; And it may include non-linearly registering the matched brain image to the standard template.

Predicting the brain gender feature includes extracting a first gender feature, which is a feature related to brain gender, from the preprocessed brain image; extracting a second gender characteristic, which is a feature contributing to brain gender prediction, from the first gender characteristic; analyzing the weight of the second gender characteristic; And it may include predicting a feature whose weight is greater than or equal to a threshold among the second gender features as a brain gender feature.

The step of predicting brain age based on brain sex includes inputting the preprocessed brain image into a convolutional neural network model of the brain age prediction model to extract a first age feature, which is a feature of the brain image that contributes to predicting brain age. step; extracting a second age feature, which is a feature contributing to brain age prediction, by inputting the predicted brain sex feature into a multi-layer perceptron model of the brain age prediction model; And it may include predicting brain age based on the brain gender based on the first age characteristic and the second age characteristic.

The brain image includes brain orientation information and may include the entire skull and brain region.

In a computing device that performs a method for predicting brain sex and brain age according to an embodiment of the present invention, the computing device includes a processor, the processor identifies a brain image from a database, and determines the brain sex of the brain image. The prediction model and the brain age prediction model are preprocessed so that they can be identified, the brain sex prediction model predicts brain sex characteristics used to determine brain sex from the preprocessed brain image, and the brain age prediction model predicts the brain sex characteristics of the preprocessed brain. Brain age based on the brain sex characteristics can be predicted from the image and predicted brain sex characteristics.

The processor may match the voxel size and voxel volume of the brain image to a standard template, and non-linearly register the matched brain image to the standard template.

The processor extracts a first gender feature that is a feature related to brain sex from the preprocessed brain image, extracts a second gender feature that is a feature that contributes to predicting brain sex from the first gender feature, and extracts the second gender feature from the first gender feature. The weights of the features are analyzed, and among the second gender features, the feature whose weight is greater than the threshold can be predicted as a brain gender feature.

The processor inputs the preprocessed brain image into a convolutional neural network model of the brain age prediction model to extract a first age feature, which is a feature of the brain image that contributes to brain age prediction, and applies the predicted brain sex feature to the The second age feature, which is a feature contributing to brain age prediction, is extracted by inputting it into the multi-layer perceptron model of the brain age prediction model, and the brain age based on the brain gender can be predicted based on the first age feature and the second age feature. .

A method for learning a brain age prediction model according to an embodiment of the present invention, comprising: identifying a data set including brain images, gender, and age from a database; From among the data sets, determining a learning data set to learn a brain age prediction model and a test data set to verify the brain age prediction model; Using the learning data set, a brain age prediction model is trained to predict brain age when brain images and brain sex characteristics are input; And it may include verifying the accuracy of predicting brain age of the brain age prediction model learned using the test data set.

The step of determining a learning data set for learning a brain age prediction model and a test data set for verifying the brain age prediction model among the data sets includes dividing the age distribution according to the age included in the data set and randomly selecting the learning data set. And the test data set can be determined.

It further includes the step of preprocessing the brain images included in the data set so that a learning brain age prediction model can identify them, wherein the preprocessing step includes matching the voxel size and voxel volume of the brain image to a standard template. ; And it may include non-linearly registering the matched brain image to the standard template.

According to one embodiment of the present invention, the subject's brain age can be determined using a brain age prediction model, which is a hybrid deep learning model that combines a convolutional neural network model and a multilayer perceptron model. By calculating the difference between the determined brain age and actual age (BrainAGE), you can identify each individual's brain aging state and monitor brain health by tracking changes. Additionally, a personalized brain health management protocol can be provided using BrainAGE information, which is information on the difference between the determined brain age and actual age.

Additionally, according to an embodiment of the present invention, a brain age prediction model that determines the subject's brain age can be learned using brain images.

1 is a diagram illustrating a computing device capable of predicting brain age according to an embodiment of the present invention.

Figure 2 is a flow chart for explaining a method for predicting brain age according to an embodiment of the present invention.

Figure 3 is a flow chart to explain a method of preprocessing a brain image according to an embodiment of the present invention.

Figure 4 is a flow chart for explaining a method for predicting brain characteristics according to an embodiment of the present invention.

Figure 5 is a diagram for explaining a brain sex prediction model for predicting brain sex characteristics according to an embodiment of the present invention.

Figure 6 is a flow chart to explain a method for predicting brain age according to an embodiment of the present invention.

Figure 7 is a diagram for explaining a brain age prediction model for predicting brain age according to an embodiment of the present invention.

Figure 8 is a diagram for explaining a method of learning a brain age prediction model according to an embodiment of the present invention.

Figure 9 is a diagram for explaining a method for verifying a brain age prediction model according to an embodiment of the present invention.

Hereinafter, embodiments will be described in detail with reference to the attached drawings. However, the scope of the patent application is not limited or limited by these examples. The same reference numerals in each drawing indicate the same members.

Various changes may be made to the embodiments described below. The embodiments described below are not intended to limit the embodiments, but should be understood to include all changes, equivalents, and substitutes therefor.

Terms such as first or second may be used to describe various components, but these terms should be understood only for the purpose of distinguishing one component from another component. For example, a first component may be named a second component, and similarly, the second component may also be named a first component.

Terms used in the examples are merely used to describe specific examples and are not intended to limit the examples. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by a person of ordinary skill in the technical field to which the embodiments belong. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and unless explicitly defined in the present application, should not be interpreted in an ideal or excessively formal sense. No.

In addition, when describing with reference to the accompanying drawings, identical components will be assigned the same reference numerals regardless of the reference numerals, and overlapping descriptions thereof will be omitted. In describing the embodiments, if it is determined that detailed descriptions of related known technologies may unnecessarily obscure the gist of the embodiments, the detailed descriptions are omitted.

Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

Referring to Figure 1, a database 101 and computing device 102 are shown.

The database 101 can collect human brain images from open sources. The database 101 may store human brain images collected from open sources. The database 101 can collect and store information on the gender and actual age of the subject who took the brain image that matches the human brain image, as well as the human brain image. The database 101 can purify data by removing images that are difficult to preprocess later, brain images in which part of the brain has been cut, and images in which the skull has been removed from human brain images. The database 101 can transmit the stored brain images to a computing device.

Computing device 102 may include a processor 103 and memory 104.

The computing device 102 can receive brain images from the database 101 and predict brain age. Specifically, the processor 103 included in the computing device 102 can predict the subject's brain age by identifying brain images.

The processor 103 can identify the brain image transmitted from the database 101. The processor 103 may preprocess the transmitted brain image before inputting it into the brain sex prediction model 501 and the brain age prediction model 701. Brain age can be predicted when the processor 103 inputs the preprocessed brain image into the brain sex prediction model 501 and the brain age prediction model 701. The brain sex prediction model 501 will be explained in FIGS. 4 and 5. The brain age prediction model 701 will be explained in FIGS. 7 and 8.

The memory 104 may store data generated in the process of predicting brain age. For example, the memory 104 may store the learned brain sex prediction model 501 and the brain age prediction model 701 and provide them to the processor 103 when predicting brain age.

Below, the process by which the processor 103 predicts brain age will be described.

Referring to FIG. 2, the computing device 102 can predict brain age using the processor 103, the brain sex prediction model 501, and the brain age prediction model 701. The brain sex prediction model 501 is shown in FIG. 5. Brain age prediction model 701 is shown in FIG. 7.

Specifically, the processor 103 included in the computing device 102 may predict brain age using the brain sex prediction model 501 and the brain age prediction model 701. The brain sex prediction model 501 and the brain age prediction model 701 may be models learned to predict brain sex characteristics and brain age, respectively.

In step 201, the processor 103 may receive a brain image from the database 101.

In step 202, the processor 103 may identify the brain image received by the computing device 102.

Brain images may be images collected by the database 101. Brain images may be brain images of healthy people over the age of 18 collected from open sources. Brain images are MRI images, which may be T1-weighted images.

The database 101 can purify brain images. The database 101 may delete brain images that are difficult to pre-process later, such as when the quality of the brain images is low or when the brain images do not contain direction information. The direction information of the brain image may be information about the three-dimensional x-axis, y-axis, and z-axis. Specifically, the direction information of brain images includes left/right information corresponding to the left and right brains, anterior-posterior information about the front and back of the brain, and superior/bottom information about the top and bottom of the brain. interior) information may be included. Additionally, the database 101 may purify the brain images by deleting from the database 101 brain images with the skull removed and brain images with part of the brain removed.

Accordingly, the brain image identified by the processor 103 is a T1-weighted image of a person over 18 years of age, and the brain included in the brain image is an intact brain with no part of it cut, and may be an image including the skull.

In step 203, the processor 103 may preprocess the identified brain image.

The brain image identified by the processor 103 is a brain image refined from the database 101, but since there are various MR scanners and imaging techniques, the brain image is used in the brain sex prediction model 501 and the brain age prediction model 701. It needs to be preprocessed before input.

The method for preprocessing brain images will be explained in Figure 3.

In step 204, the processor 103 may further include augmenting the brain image.

Augmentation of brain images may be necessary to reduce the possibility of bias and variance in the neural network-based deep learning process and to prevent over-fitting of deep learning models to specific brain imaging data.

The processor 103 can identify the distribution pattern of data according to gender and age group and augment neuroimaging data for a group of subjects in need of augmentation. Specifically, the processor 103 builds a brain image template for each detailed subject group based on gender and age, and applies data augmentation techniques such as shifting, rotation, and flipping to determine gender and age. Brain images can be augmented to better reflect characteristics.

In step 205, the processor 103 may input the preprocessed brain image into the brain sex prediction model 501 and the brain age prediction model 701.

In step 206, the brain gender prediction model 501 may predict brain gender characteristics from the preprocessed brain image.

The brain gender prediction model 501 may be a model learned to predict brain gender characteristics from preprocessed brain images. Specifically, the brain gender prediction model 501 may be a convolution neural network (CNN).

The method by which the brain gender prediction model 501 predicts brain gender characteristics will be further explained in FIG. 4.

In step 207, the brain sex prediction model 501 may input the predicted brain sex characteristics into the brain age prediction model 701.

In step 208, the brain age prediction model 701 may predict brain age from the brain sex characteristics input from the brain sex prediction model 501 and the preprocessed brain image input from the processor 103.

The brain age prediction model 701 may be a model learned to predict brain age based on brain sex characteristics using preprocessed brain images and brain sex characteristics. Specifically, the brain age prediction model 701 may be a learning model that combines a convolutional neural network model and a multilayer perceptron model. That is, the brain age prediction model 701 may be a hybrid deep learning-based model.

The method by which the brain age prediction model 701 predicts brain age will be further explained in FIG. 6.

In step 301, the processor 103 may match the voxel size and voxel volume of the brain image to a standard template.

Neuroimaging is a brain image collected by a database, and since the MR scanners and imaging techniques used for imaging are diverse, it is necessary to normalize them. Therefore, in order to use brain images that have undergone minimal preprocessing as input data, the processor can use the MATLAB-based SPM12 program to unify the voxel size and voxel volume of the image.

At step 302, processor 103 may non-linearly register the matched brain image to a standard template.

That is, the processor 103 not only performs linear registration that performs simple transformations of the matched brain images such as rotation, enlargement, and reduction, but also additionally deforms local parts of the matched brain images to provide details of the brain images. Nonlinear matching can be performed to accurately match the structure.

The processor 103 can normalize individual brain image data by non-linearly matching the brain image to the MNI152 standard template.

Additionally, the processor 103 can load brain image files in NifTI format with the Python NiBabel package and arrange them in a numpy array to use brain images in a convolutional neural network model. The processor 103 may perform maximum-minimum normalization on the constituent values of the NumPy array by dividing individual values by the maximum value.

Below, we will explain how the brain sex prediction model 501 predicts brain sex characteristics using normalized brain images.

The processor 103 may predict brain gender characteristics using the brain gender prediction model 501. The brain gender prediction model may be a learned CNN model.

In step 401, the brain gender prediction model 501 may extract a first gender feature, which is a feature related to brain gender, from the preprocessed brain image.

The brain sex prediction model 501 uses preprocessed T1-weighted brain images as input data and can extract features related to brain sex in 3D brain images. Additionally, the brain sex prediction model 501 can generate a brain sex feature map by applying weights to features related to brain sex.

In step 402, the brain gender prediction model 501 may extract a second gender characteristic, which is a feature contributing to brain gender prediction, from the first gender characteristic.

The second gender feature, which is a feature that contributes to brain sex prediction, may be included in the first gender feature, which is a feature related to brain sex. Structural brain imaging may include features related to brain sex. However, not all brain sex-related characteristics are helpful in predicting brain sex. Therefore, in order to predict brain sex characteristics, it is necessary to extract features that contribute to brain sex prediction among features related to brain sex. Accordingly, the brain gender prediction model 501 can extract the second gender feature, which is a main feature, from the brain gender feature map generated by applying weights to features related to brain gender.

In step 403, the brain gender prediction model 501 may analyze the weight of the second gender characteristic.

The brain gender prediction model 501 can analyze the weight of the extracted second gender feature and classify it based on a specific threshold.

In step 404, the brain gender prediction model 501 may predict brain gender characteristics based on a threshold by applying a sigmoid activation function based on the second gender characteristic.

The brain gender prediction model 501 may determine that a second gender characteristic classified based on a specific threshold is masculine if it is above a certain threshold, and can be determined to be feminine if it is less than a certain threshold. Accordingly, the brain gender prediction model 501 can predict brain gender characteristics based on the second gender characteristics determined as masculine and feminine based on a specific threshold.

Brain gender characteristics may be information that indicates the gender of the brain included in the brain image as a percentile, indicating whether it is a male brain or a female brain. For example, brain gender characteristics may be displayed as 60% masculine and 40% feminine.

Below, the structure of the brain gender prediction model 501 will be described.

Referring to FIG. 5, the brain sex prediction model 501 can predict brain sex characteristics from preprocessed brain images. The brain sex prediction model 501 includes a first block 502, a second block 503, a third block 504, a fourth block 505, a fifth block 506, a maximum pooling layer 507, It may include a planarization layer 508, a sixth block 509, and a logistic regression analysis unit 510. At this time, the brain gender prediction model 501 may repeat the first block twice.

The first block 502 and the third block 504 may include a 3D convolution layer, a ReLu function layer, a maximum pooling layer, and a batch normalization layer. The second block 503, fourth block 505, and fifth block 506 may include a 3D convolution layer and a ReLu function layer. The sixth block 509 may include a dense layer and a linear function layer.

The 3D convolution layer uses preprocessed brain images as input data and can extract features related to brain gender in the 3D brain images. In addition, the 3D convolution layer can construct a feature map for each brain sex by applying weights to features related to brain sex.

The maximum pooling layer can extract output data by subsampling key features from the brain sex feature map constructed from a 3D convolution layer.

The flattening layer 507 can convert features related to brain gender in a 3D brain image into a 1D array.

The dense layer of the sixth block 509 may be fully connected. The dense layer can extract features that contribute to the prediction of brain sex characteristics among features related to brain sex.

The batch normalization layer can stabilize the entire deep learning process by normalizing the output data in the learning process conducted on a batch basis.

The logistic regression analysis unit 510 may analyze the weights of features that contribute to the prediction of brain sex characteristics extracted from the dense layer. The logistic regression analysis unit 510 can predict brain sex characteristics by applying a sigmoid activation function.

Below, we will explain how to predict brain age using brain sex characteristics and preprocessed brain images.

The processor 103 can predict brain age using the brain age prediction model 701. The brain age prediction model 701 can predict brain age from preprocessed brain images and brain sex characteristics. The brain age prediction model 710 may be a model that combines a convolutional neural network model and a multilayer perceptron model. That is, the brain age prediction model 710 may be a hybrid deep learning model.

In step 601, the brain age prediction model 701 inputs the preprocessed brain image into the convolutional neural network model of the brain age prediction model to extract the first age feature, which is a feature of the brain image that contributes to brain age prediction. there is.

The preprocessed brain image may be a preprocessed T1-weighted image. The preprocessed brain image may be the same image as the preprocessed brain image previously input to the brain gender prediction model 501. The first age feature may be a feature that contributes to predicting brain age extracted from the preprocessed brain image.

In step 602, the brain age prediction model 701 inputs the predicted brain sex characteristics into the multi-layer perceptron model of the brain age prediction model to extract the second age feature, which is a feature contributing to brain age prediction.

The predicted brain gender features may be features output by the brain gender prediction model 501 after receiving a preprocessed brain image as input. The second age feature may be a feature that contributes to predicting brain age extracted from brain sex characteristics.

In step 603, the brain age prediction model 701 may predict brain age based on brain gender based on the first age characteristic and the second age characteristic.

Because there are marked differences between the sexes in brain development and aging processes, predicting brain age based on brain sex allows for quantified brain sex characteristics (e.g., 60% male or 60% female) rather than categorical characteristics (e.g., male or female). 40%) can be predicted to reflect brain age.

Below, the structure of the brain age prediction model 701 will be explained.

Referring to FIG. 7, the brain age prediction model 701 may include a convolutional neural network model 702 and a multilayer perceptron model 703. Specifically, the convolutional neural network model 702 can receive preprocessed brain images and extract features that contribute to predicting brain age. The multi-layer perceptron model 703 can receive brain sex characteristics predicted from the brain sex prediction model 501 and extract features that contribute to predicting brain age.

First, we will explain the structure of the convolutional neural network model 702 that extracts the first age feature.

The convolutional neural network model 702 includes a first block including a 3D convolutional layer and a ReLu function layer, a second block including a 3D convolutional layer, a batch normalization layer, and a ReLu function layer, and a third block including a max pooling layer. It may include a block 704 containing blocks. The convolutional neural network model 702 can repeat the above block 704 five times. By repeating block 704, the channels can be increased to 8, 16, 32, 64, and 128. As the channel becomes higher, the complex characteristics of brain imaging can be realized.

Afterwards, the convolutional neural network model 702 can transform the three-dimensional data into one dimension by placing a smoothing layer 705 after the block 704. The convolutional neural network model 702 may place a flattening layer 705 followed by a layer 706 including a dense layer and a ReLu function layer. In the layer 706 including the dense layer and the ReLu function layer, the dense layer may include 64 nodes. The ReLu function layer can apply an activation function.

Next, the convolutional neural network model 702 can stabilize the data by placing a batch normalization layer 707. Afterwards, the convolutional neural network model 702 can prevent over-fitting by placing a drop out layer 708.

Additionally, a layer including a dense layer and a ReLu function layer may be rearranged after the dropout layer 708. At this time, the dense layer may have 16 nodes.

In the structure of the convolutional neural network model, output data can be extracted by subsampling key features from the generated brain age feature map in a plurality of pooling layers.

The batch normalization layer can stabilize the entire deep learning process by normalizing the output data during the learning process conducted on a batch basis.

The flattening layer can convert the brain age characteristics of 3D brain images into a 1D array.

And, the dense layer can output the first age feature, which is a feature that contributes to brain age prediction.

Next, we will explain the structure of the multilayer perceptron model 703 that extracts the second age feature.

The multi-layer perceptron model 703 can repeatedly arrange layers combining a dense layer and a ReLu function layer.

The first dense layer to come can contain 512 nodes. The next dense layer may contain four nodes.

The dense layer of the multi-layer perceptron model 703 can receive the brain sex characteristics predicted in the brain sex prediction model 501 as input. Afterwards, the dense layer can output the second age feature, which is a feature that contributes to predicting brain age among brain sex features.

Lastly, we will explain the structure for the hybrid deep learning model.

The brain age prediction model 701 may arrange a combination layer so that the output values of the convolutional neural network model 702 and the multilayer perceptron model 703 can be input in parallel. The combination layer may combine the first age feature and the second age feature, which are features output through the convolutional neural network model 702 and the multilayer perceptor model 703.

Next, the brain age prediction model 701 can place a dense layer with 4 nodes. The brain age prediction model 701 can apply the ReLu activation function by placing the ReLu function layer after the dense layer.

Lastly, the brain age prediction model 701 can predict brain age by placing a dense layer with one node and then placing a linear function layer and applying a linear activation function. In dense layers, the weights of features that contribute to brain age prediction can be analyzed. Next, in the linear function layer, brain age can be predicted by applying a linear activation function.

Below, we will explain how to learn the brain age prediction model 701.

At step 801, processor 103 may identify a data set including brain images, gender, and age from a database.

Neuroimaging, gender, and age may be data collected by the database. The brain image may include direction information of the brain image. The brain included in brain imaging is an intact brain with no part of it cut, and may include the skull. Gender and age may be the actual gender and actual age of the subject who took the brain image.

In step 802, the processor 103 may determine a training data set from among the data sets to learn a brain age prediction model and a test data set to verify the brain age prediction model.

The processor 103 may randomly extract the entire data set into a training data set and a test data set according to a certain ratio. For example, you can randomly determine that the training data set is 90% and the test data set is 10% of the entire data set.

At this time, the processor 103 may divide the entire data set into specific units so that each randomly extracted set reflects the age distribution of the entire data set, and randomly extract data from the data set divided into specific units. For example, the processor 103 may divide the entire data set into 5-year age units and randomly extract data from the data set divided into 5-year age units.

In step 803, the processor 103 may train a brain age prediction model to predict brain age when brain images and brain sex characteristics are input using the learning data set.

The brain age prediction model can predict brain age according to the structural characteristics of the input brain image. Therefore, the brain age prediction model can learn to predict brain age higher when the input brain image shows a significant decrease in the thickness of the cerebral cortex compared to the average level for the same age group. In addition, the brain age prediction model is trained to predict brain age highly in cases where the input brain image shows atrophy of the brain parenchyma, large enlargement of the ventricles, strong high-intensity signals in the white matter around the ventricles, and cases where the volume of the hippocampus is greatly reduced. can do.

Additionally, the brain age prediction model can be learned to predict brain age based on the input brain sex characteristics. Therefore, the brain age prediction model can learn to predict brain age lower when the brain sex of an actual male is female, and the brain age prediction model is based on female brain sex characteristics than when it is based on male, which is actual gender information. there is. In addition, the brain age prediction model can learn to predict brain age higher when the actual female brain sex is male, and when the brain age prediction model is based on actual gender information, female, than when it is based on female brain sex characteristics. .

In other words, the brain age prediction model can predict brain age based on the brain sex predicted by the brain sex prediction model, and if the brain sex predicted by the brain sex prediction model matches the actual brain sex, the brain age prediction model can predict brain age. It can learn to predict age lower.

The training data set may contain brain images labeled with actual gender and actual age.

Accordingly, the brain gender prediction model 501 can be trained to predict brain gender using the learning data set. The brain age prediction model 701 can be trained to predict brain age using a learning data set.

In step 804, the processor 103 may verify the accuracy of predicting brain age of the learned brain age prediction model using a test data set.

The method of learning a brain age prediction model may further include a step of preprocessing data.

Accordingly, the processor 103 may further include a step of preprocessing the brain images included in the data set so that the learning brain age prediction model can identify them.

At this time, brain imaging can be in various forms depending on the MR scanner and imaging technique used. Accordingly, the processor 103 can match the voxel size and voxel volume of the brain image to the standard template and non-linearly register the matched brain image to the standard template. Here, the standard template may be the MNI152 standard template.

Additionally, the processor 103 can load a NifTI format brain image file with the Python NiBabel package and convert it into a numpy array in order to use the brain image in a convolutional neural network model. Composition normalization of a NumPy array can be done by dividing individual values by the maximum value and performing max-minimum normalization.

The processor 103 can preprocess not only brain images but also gender. The processor 103 can convert it into a vector format consisting of 0 or 1 through one-hot encoding. Therefore, in order to use gender information as input data, the processor 103 can code male as ‘M’ and female as ‘F’ and save them in csv file format. Additionally, the processor 103 can load the csv file as a dataframe and convert it into a NumPy array to use gender information in the multi-layer perceptron model. One-hot encoding of NumPy arrays can be done using LabelBinarizer from the scikit-learn library to convert the data to vector format.

Additionally, the method of learning a brain age prediction model may further include the step of preprocessing and then augmenting the data.

The brain age prediction model may further include verification with a verification data set included in the learning data set before verification with the test data set. Below, we will explain how to verify using the verification data set.

The entire data set 901 can be divided into a training data set 902 and a test data set 903. The training data set 902 can be used to learn the brain age prediction model 701, and the test data set 903 can be used to verify the trained brain age prediction model 701.

Below, we will explain k-fold cross validation to efficiently verify models while reducing the possibility of data bias and variance.

The training data set can be divided into a training data set and a validation data set. At this time, the learning data set 902 may be divided into k folds. In this specification, the brain age learning model has k of 10, but for the explanation of k-fold cross-validation, k is assumed to be 5.

The training data set 902 is divided into five folds, and a total of five splits in which the validation data sets do not overlap can be determined. Split 1 (904) to Split 5 (908) are splits with different verification data sets.

The brain age learning model can be trained using the training data set included in each split. In other words, five learned brain age learning models can be created by five splits. The learned brain age learning model can be verified using the verification data set included in each split. At this time, the average of Result 1 to Result 5, which are verification results, may be determined as the performance of the entire model.

Among the performances of all models, the model with the smallest loss function value can be determined as the model with optimal performance. Therefore, among multiple models learned using the same training data set, the model with the smallest loss function value can be determined as the model with optimal performance.

And, the determined model can be finally learned using the entire training data set.

Finally, the performance of the brain age prediction model can be verified by checking the brain age prediction accuracy of the learned model through the test data set 903.

In addition, the optimal hyperparameter can be further determined using the average and standard deviation of individual results from k-fold cross-validation.

For this purpose, in the present invention, k is 10, so the hyperparameter can be set as follows. Considering the size and learning speed of the data set, the epoch can be set to 100 and the batch size can be set to 16. The loss function can be set to mean absolute error to monitor the learning process. The learning rate can be set to 0.01 and the learning rate decay to 0.003. Additionally, the optimizer may be determined by Adam (adaptive moment estimation).

Meanwhile, the method according to the present invention is written as a program that can be executed on a computer and can be implemented in various recording media such as magnetic storage media, optical read media, and digital storage media.

Implementations of the various techniques described herein may be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or combinations thereof. Implementations may include a computer program product, i.e., an information carrier, e.g., machine-readable storage, for processing by or controlling the operation of a data processing device, e.g., a programmable processor, a computer, or multiple computers. It may be implemented as a computer program tangibly embodied in a device (computer-readable medium) or a radio signal. Computer programs, such as the computer program(s) described above, may be written in any form of programming language, including compiled or interpreted languages, and may be written as a stand-alone program or as a module, component, subroutine, or part of a computing environment. It can be deployed in any form, including as other units suitable for use. The computer program may be deployed for processing on one computer or multiple computers at one site or distributed across multiple sites and interconnected by a communications network.

Processors suitable for processing computer programs include, by way of example, both general-purpose and special-purpose microprocessors, and any one or more processors of any type of digital computer. Typically, a processor will receive instructions and data from read-only memory or random access memory, or both. Elements of a computer may include at least one processor that executes instructions and one or more memory devices that store instructions and data. Generally, a computer may include one or more mass storage devices that store data, such as magnetic, magneto-optical disks, or optical disks, receive data from, transmit data to, or both. It can also be combined to make . Information carriers suitable for embodying computer program instructions and data include, for example, semiconductor memory devices, magnetic media such as hard disks, floppy disks, and magnetic tapes, and Compact Disk Read Only Memory (CD-ROM). ), optical media such as DVD (Digital Video Disk), magneto-optical media such as Floptical Disk, ROM (Read Only Memory), RAM , Random Access Memory), flash memory, EPROM (Erasable Programmable ROM), and EEPROM (Electrically Erasable Programmable ROM). The processor and memory may be supplemented by or included in special purpose logic circuitry.

Additionally, computer-readable media can be any available media that can be accessed by a computer, and can include both computer storage media and transmission media.

Although this specification contains details of numerous specific implementations, these should not be construed as limitations on the scope of any invention or what may be claimed, but rather as descriptions of features that may be unique to particular embodiments of particular inventions. It must be understood. Certain features described herein in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination. Furthermore, although features may be described as operating in a particular combination and initially claimed as such, one or more features from a claimed combination may in some cases be excluded from that combination, and the claimed combination may be a sub-combination. It can be changed to a variant of a sub-combination.

Likewise, although operations are depicted in the drawings in a particular order, this should not be construed as requiring that those operations be performed in the specific order or sequential order shown or that all of the depicted operations must be performed to obtain desirable results. In certain cases, multitasking and parallel processing may be advantageous. Additionally, the separation of various device components in the above-described embodiments should not be construed as requiring such separation in all embodiments, and the described program components and devices may generally be integrated together into a single software product or packaged into multiple software products. You must understand that it is possible.

Meanwhile, the embodiments of the present invention disclosed in the specification and drawings are merely provided as specific examples to aid understanding, and are not intended to limit the scope of the present invention. It is obvious to those skilled in the art that in addition to the embodiments disclosed herein, other modifications based on the technical idea of the present invention can be implemented.

Claims

Identifying brain images from a database;

Preprocessing the brain image so that a brain sex prediction model and a brain age prediction model can be identified;

Predicting brain sex characteristics used to determine brain sex from a brain image for which the brain sex prediction model has been preprocessed; and

The brain age prediction model predicting brain age based on the brain sex characteristics from the preprocessed brain image and predicted brain sex characteristics.

Brain gender and brain age prediction method including.
According to paragraph 1,

The preprocessing step is,

Matching the voxel size and voxel volume of the brain image to a standard template; and

A method for predicting brain sex and brain age, comprising the step of non-linearly matching the matched brain image to the standard template.
According to paragraph 1,

The step of predicting the brain gender characteristics is,

Extracting a first gender feature, which is a feature related to brain gender, from the preprocessed brain image;

extracting a second gender characteristic, which is a feature contributing to brain gender prediction, from the first gender characteristic;

analyzing the weight of the second gender characteristic; and

A brain sex and brain age prediction method comprising predicting a feature whose weight is greater than a threshold among the second gender features as a brain sex feature.
According to paragraph 1,

The step of predicting brain age based on brain sex is,

Inputting the preprocessed brain image into a convolutional neural network model of the brain age prediction model to extract a first age feature, which is a feature of the brain image that contributes to brain age prediction;

extracting a second age feature, which is a feature contributing to brain age prediction, by inputting the predicted brain sex feature into a multi-layer perceptron model of the brain age prediction model; and

A brain sex and brain age prediction method comprising predicting brain age based on the brain sex based on a first age characteristic and a second age characteristic.
According to paragraph 1,

The brain image,

A brain sex and brain age prediction method that includes brain orientation information and brain imaging that includes the entire skull and brain region.
In a computing device that performs a brain sex and brain age prediction method,

The computing device includes a processor,

The processor,

Identify brain images from the database,

The brain image is preprocessed so that the brain sex prediction model and brain age prediction model can be identified,

The brain sex prediction model predicts brain sex characteristics used to determine brain sex from preprocessed brain images,

A computing device wherein the brain age prediction model predicts brain age based on the brain sex characteristics from the preprocessed brain image and predicted brain sex characteristics.
According to clause 6,

The processor,

Match the voxel size and voxel volume of the brain image to a standard template,

A computing device that non-linearly registers matched brain images to the standard template.
According to clause 6,

The processor,

Extracting a first gender feature, which is a feature related to brain gender, from the preprocessed brain image,

Extracting a second gender characteristic, which is a feature contributing to brain gender prediction, from the first gender characteristic,

Analyzing the weight of the second gender characteristic,

A computing device that predicts a feature whose weight is greater than a threshold among the second gender features as a brain gender feature.
According to clause 6,

The processor,

Input the preprocessed brain image into a convolutional neural network model of the brain age prediction model to extract a first age feature, which is a feature of the brain image that contributes to brain age prediction,

Input the predicted brain sex characteristics into a multi-layer perceptron model of the brain age prediction model to extract a second age feature, which is a feature contributing to brain age prediction,

A computing device for predicting brain age based on the brain gender based on a first age characteristic and a second age characteristic.
According to clause 6,

The brain image,

A computing device that contains brain orientation information and includes the entire skull and brain regions.
identifying a data set containing brain images, gender, and age from a database;

From among the data sets, determining a learning data set to learn a brain age prediction model and a test data set to verify the brain age prediction model;

Using the learning data set, a brain age prediction model is trained to predict brain age when it receives brain images and brain sex characteristics; and

Verifying the accuracy of predicting brain age of the brain age prediction model learned using the test data set

Brain age prediction model learning method including.
According to clause 11,

The step of determining a learning data set to learn the brain age prediction model and a test data set to verify the brain age prediction model from among the data sets,

A brain age prediction model learning method that randomly determines the training data set and the test data set by dividing the age distribution according to the age included in the data set.
According to clause 11,

Further comprising the step of preprocessing the brain images included in the data set so that a learning brain age prediction model can identify them,

The preprocessing step is,

Matching the voxel size and voxel volume of the brain image to a standard template; and

A brain age prediction model learning method comprising the step of non-linearly matching the matched brain image to the standard template.
According to clause 11,

The brain image,

A brain age prediction model learning method that includes brain orientation information and includes the entire skull and brain region.