CN105046709A - Nuclear magnetic resonance imaging based brain age analysis method - Google Patents

Abstract

The invention discloses a nuclear magnetic resonance imaging based brain age analysis method. Firstly, by utilizing information in a brain structure image, constructing a brain structure network of a functional brain region and voxel level, and according to a functional image signal, establishing a brain function network of the functional brain region and the voxel level; secondly, with a statistical method, characteristics and rules that the brain structure and the brain function are changed along with age growing, and change of a brain network connection mode are analyzed so as to obtain differential characteristics that the brain structure, the brain function and the brain network are changed along with age growing, and with a recursive characteristic growth method, selecting required effective characteristics and reducing characteristic dimensions; and finally, by applying a linear support vector machine, establishing a brain age regression model and analyzing the brain age. The method can effectively reveal an internal relation and difference between the physiological age and the brain age of people; and scientific and objective theoretical basis and method are provided for research on human brain development.

Description

A kind of brain analytical approach in age based on nuclear magnetic resonance image

Technical field

The present invention relates to the technical field of digital image analysis and process, refer in particular to a kind of brain analytical approach in age based on nuclear magnetic resonance image.

Background technology

The all one's life of people goes through from fetus to infant, arrive multiple stages such as children and adolescents, middle age, old age again, in the process, and other histoorgan of health is similar, brain also exists and increases growth, growth, maturation, slowly atrophy and decline stage rapidly fast, and follows certain rule.

Research finds, the grey matter of human brain from birth to childhood increase rapidly, grey matter volume reaches maximum, and pubarche slowly declines, in inverse u shape; White matter is then from birth to juvenile period, and sharp increase, after puberty, increasess slowly, until the manhood starts again slow decline.As can be seen here, in all ages and classes stage, the growth course of human brain shows diverse rule and feature, if estimate brain development situation or degree age with brain, then the existing substantial connection of physiological age of it and people, has again its exclusive characteristic.Disclose the internal relation (as consistance and otherness) of brain age and physiological age, to growth course and the Evolution of understanding National People's Congress's brain, the change etc. of understanding human brain cognitive ability, psychological behavior, has important scientific research value.

Magnetic resonance imaging is that human brain research provides a kind of safety without wound, accurately and reliably means, and it can obtain relevant human brain tissue structure image clearly, with the dynamic realtime functional image of reflection cerebration situation.By brain 26S Proteasome Structure and Function image, the change of human brain structure and fuction and the transmission situation of nerve information can be analyzed.

Recently, along with the rise of brain Connecting groups, about the research of brain network increases gradually, researcher utilizes graph theory and information theory, builds brain network, pays close attention to the connection between brain district, truck and information transmission.Brain network analysis method is also applied to the research of brain, as core node, network local/global efficiency, minimum path length, worldlet attribute, anti-attack ability etc.

Utilize brain structure, brain function image, and the Neurobiology feature such as brain network, build brain analytical model in age, and be used for model analyzing brain brain age, namely build a kind of comprehensive effective brain analytical model in age and method, there is not yet open report.

Summary of the invention

The present invention aims to provide a kind of science reliably based on the brain analytical approach in age of nuclear magnetic resonance image, the development condition assessing brain can be effectively applied to, for exploring growth course and the rule of human brain, disclose the variation characteristic of cognitive ability, being familiar with its psychological behavior characteristic provides important evidence.

For achieving the above object, technical scheme provided by the present invention is: a kind of brain analytical approach in age based on nuclear magnetic resonance image, first, utilizes brain structure image information, the brain structural network of constructing function brain district and voxel level; According to cerebral function signal of video signal, set up the brain function network of function brain district and voxel level; Then, analyze the structure of brain, brain function increased and the characteristic sum rule of change with the age, and the change of brain network connection mode, thus obtain brain structure, brain function and brain network and increase and the otherness feature of change with the age; Utilize recursive feature growth method, the validity feature needed for selection thereupon, reduce intrinsic dimensionality; Finally, application linear SVM, sets up brain regression model in age, analyzes age brain.

Brain analytical approach in age based on nuclear magnetic resonance image of the present invention, comprises the following steps:

1) regression model is set up

1.1) MR image data is gathered

Obtain brain structure and brain function MR image;

1.2) image data process

Comprise a normal moveout correction, registration, Spatial normalization, smoothing processing, wherein, to brain structure image carry out ectocinerea and white matter of brain analyze time, needing tissue segmentation is grey matter, white matter and cerebrospinal fluid, brain function image first also takes interbed and corrects, then carries out subsequent treatment;

Also comprise brain structure and brain function network struction, namely utilize through pretreated image, application drawing opinion and information theory, in function brain district or voxel level, set up brain 26S Proteasome Structure and Function network;

1.3) analysis and extraction of features

In function brain district or voxel level, application single-sample t-test method, analyzes brain structure, brain function image and the brain network characterization situation of change with the age, namely obtains the feature with age generation marked change;

1.4) Feature Dimension Reduction

Using iterative feature growth method, carries out dimensionality reduction to brain structure, brain function and brain network characterization respectively, selects validity feature, records the index of effective proper vector;

1.5) regretional analysis

By brain function, brain structure and the brain network characterization selected after fusion dimensionality reduction, using it as independent variable, using physiological age as dependent variable, set up linear regression model (LRM) therebetween, carry out regretional analysis;

2) brain is analyzed age

2.1) image acquisitions

Obtain the brain MR 26S Proteasome Structure and Function image of object to be analyzed;

2.2) image data process

Comprise a normal moveout correction, registration, Spatial normalization, smoothing processing, wherein, structure image must be split, and functional image first also takes interbed and corrects, then carries out subsequent treatment;

Utilize through pretreated image, application drawing opinion and information theory, in brain district or voxel level, set up brain 26S Proteasome Structure and Function network;

2.3) feature extraction

According to step 1.4) the validity feature index vector that obtains, extracts corresponding brain structure, brain function and brain network characterization;

2.4) brain is analyzed age

Utilize step 1.5) regression model set up, using the feature obtained as independent variable, brain predicts the brain age of object to be assessed as dependent variable age.

In step 1.3) in, brain structure, brain function image and brain network are with the variation characteristic at age, comprise brain structure image central gray, white matter, average diffusivity in DTI image and fractional anisotropy, brain blood flow in arterial spin labeling image, the Blood oxygen level dependence of fMRI, based on locally coherence, low frequency amplitude, the mark low frequency amplitude of tranquillization state fMRI, and the limit in brain network characterization connects, the betweenness centrad of node, degree centrad.And all features all utilize z-score method to carry out returning generalized.

In step 1.4) in, adopt following iterative characteristic growth method to carry out dimensionality reduction:

1.4.1) initialization: a certain feature vector, X of getting n training sample ₀=[X ₁, X ₂..., X _i... X _n] ^tand corresponding age vector y=[y ₁, y ₂..., y ₃..., y _n] ^t; The given validity feature quantity k at every turn chosen, or ratio r ∈ [01], then k=n × r; The total characteristic quantity expected is m, wherein m=c × k, c be greater than 1 integer;

1.4.2) initialization feature subset index vector s=[1,2 ..., n], treat sequencing feature vector X _candidate=X ₀, validity feature vector X _selected=[];

1.4.3) training sample X=X is selected according to index vector ₀(:, s), training regression model α=regression (X, y), and calculate weight vector w=Σ _iα _iy _ix _i, utilize score function c (i)=(w _i) ², sort from high to low by score, fetch bit arranges the feature X of front k _sorted=[1 ... k], added in effective sequencing feature vector, and from treating to shift out sequencing feature vector, upgraded validity feature vector X _selected, treat sequencing feature vector X _candidatewith index vector s;

1.4.4) step 1.4.3 is repeated) until the dimension of validity feature vector equals m;

1.4.5) step 1.4.1 is repeated) until all features all complete dimension-reduction treatment.

Compared with prior art, tool has the following advantages and beneficial effect in the present invention:

1) make full use of the multi-modal feature advantage separately of neuroimaging, its fusion application is measured in the structure of brain analytical model in age and brain age;

2) statistical analysis technique is combined with iterative characteristic dimension reduction method, in numerous feature, select and the closely-related key character of physiological age, or with the most significant feature of change of age;

3) disclose internal relation and the otherness in physiological age and brain age, can be applicable to brain development situation and the field such as cognitive ability assess, behaviouristics and psychology, provide science, objective theoretical foundation and method for human brain development studies.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the analytical approach in age of brain described in the embodiment of the present invention.

Fig. 2 is embodiment of the present invention population of adolescent training set physiological age and estimates brain corresponding diagram in age.

Fig. 3 is embodiment of the present invention population of adolescent test set physiological age and estimates brain corresponding diagram in age.

Fig. 4 is embodiment of the present invention elderly population training set physiological age and estimates brain corresponding diagram in age.

Fig. 5 is embodiment of the present invention elderly population test set physiological age and estimates brain corresponding diagram in age.

Embodiment

Below in conjunction with specific embodiment, the invention will be further described.

Brain analytical approach in age based on nuclear magnetic resonance image of the present invention, first, utilizes the grey matter in brain structure image, white matter, cortex information, the brain structural network of constructing function brain district and voxel level; According to cerebral function signal of video signal, as Blood oxygen level dependence (BOLD) and the CBF image of arterial spin labeling imaging (ASL), set up the brain function network of function brain district and voxel level; Then, analyze the structure of brain, brain function increased and the characteristic sum rule of change with the age, and the change of brain network connection mode, thus obtain brain structure, brain function and brain network and increase and the significance difference opposite sex feature of change with the age; Utilize recursive feature growth method thereupon, select more efficiently feature, reduce intrinsic dimensionality; Finally, application linear SVM, sets up brain regression model in age, analyzes age brain.It comprises the following steps:

1) regression model is set up

1.1) MR image data is gathered

Obtain brain structure and brain function MR image.

1.2) image data process

Comprise a normal moveout correction, registration, Spatial normalization, the process such as level and smooth, wherein, to structure image carry out ectocinerea and white matter of brain analyze time, needing tissue segmentation is grey matter, white matter and cerebrospinal fluid, functional image first also takes interbed and corrects, then carries out subsequent treatment;

Also comprise brain structure and brain function network struction, namely utilize through pretreated image, application drawing opinion and information theory, in function brain district or voxel level, set up brain 26S Proteasome Structure and Function network.

1.3) analysis and extraction of features

Statistics Application method, as t inspection, analyzes brain structure, brain function image and brain network, with the variation characteristic at age.As brain structure image central gray, white matter, average diffusivity (meandiffusion in DTI image, and fractional anisotropy (fractionalanisotropy MD), FA), arterial spin labeling image (arterialspinlabelingimaging, ASL) the brain blood flow cerebralbloodflow in, CBF), Blood oxygen level dependence (the bloodoxygenationleveldependent of fMRI, BOLD), based on the local one value property of tranquillization state fMRI, low frequency amplitude, mark low frequency amplitude etc., and the feature of brain network such as limit connects, the betweenness centrad of node, degree centrad etc.Obtain the feature with change of age by single-sample t-test method, the growth with the age is described, it changes highly significant, also namely closely related with the age.

1.4) Feature Dimension Reduction

The three-dimensional MR image of human brain, data volume is very large, as one 512 (length) × 512 (wide) × 120 (number of plies) image, has more than 3,000 ten thousand voxels, even if through statistical test, voxel or the voxel feature with marked change often have between 10 ten thousand to hundred ten thousand.So dote on large data volume, need to reduce dimension further, its reason has: one, can increase a large amount of operation time and memory cost; Its two, the statistical analysis technique of employing, all voxels or voxel feature are separate, do not consider the relation between them or relevance, therefore there is bulk redundancy; Its three, because of statistical error or mistake, part voxel or voxel feature may be there is, impact return with classification accuracy.

Following iterative characteristic growth method is adopted to carry out dimensionality reduction:

1.4.1) initialization: a certain feature vector, X of getting n training sample ₀=[X ₁, X ₂..., X _i... X _n] ^tand corresponding age vector y=[y ₁, y ₂..., y ₃..., y _n] ^t; The given validity feature quantity k at every turn chosen, or ratio r ∈ [01], then k=n × r; The total characteristic quantity expected is m, wherein m=c × k, c be greater than 1 integer;

1.4.2) initialization feature subset index vector s=[1,2 ..., n], treat sequencing feature vector X _candidate=X ₀, validity feature vector X _selected=[];

1.4.3) training sample X=X is selected according to index vector ₀(:, s), training regression model α=regression (X, y), and calculate weight vector w=Σ _iα _iy _ix _i, utilize score function c (i)=(w _i) ², sort from high to low by score, fetch bit arranges the feature X of front k _sorted=[1 ... k], added in effective sequencing feature vector, and from treating to shift out sequencing feature vector, upgraded validity feature vector X _selected, treat sequencing feature vector X _candidatewith index vector s;

1.4.4) step 1.4.3 is repeated) until the dimension of validity feature vector equals m;

1.4.5) step 1.4.1 is repeated) until all features all complete dimension-reduction treatment.

1.5) regretional analysis

Merge brain function, brain structure and brain network characterization as independent variable, using physiological age as dependent variable, set up linear regression model (LRM) therebetween, carry out regretional analysis.

2) brain is analyzed age

2.1) image acquisitions

Obtain the brain MR 26S Proteasome Structure and Function image of object to be analyzed.

2.2) image data process

Comprise a normal moveout correction, registration, Spatial normalization, smoothing processing, wherein, structure image must be split, and functional image first also takes interbed and corrects, then carries out subsequent treatment;

Utilize through pretreated image, application drawing opinion and information theory, in brain district or voxel level, set up brain 26S Proteasome Structure and Function network.

2.3) feature extraction

According to step 1.4) the validity feature index vector that obtains, extracts corresponding brain structure, brain function and brain network characterization.

2.4) brain is analyzed age

Utilize step 1.5) regression model set up, using the feature obtained as independent variable, brain predicts the brain age of object to be assessed as dependent variable age, the difference of analyses and prediction brain age and its actual physiological age.

Select teenager 106 example and the elderly 123 example below, composition graphs 1 to Fig. 5 is specifically described said method of the present invention.Wherein, age of two groups of crowds, sex are as shown in table 1 below.From teenager, Stochastic choice 80% sample is as training set (85 example), and 20% as test set (21 example).From middle-aged and old Stochastic choice 80% sample as training set (98 example), 20% as test set (25 example).MRT1W1 brain structure image and tranquillization state brain function image are the collection of 1.5T nuclear magnetic resonane scanne.

Table 1

	Quantity	Male/female ratio	The range of age	Age average/standard deviation
					Teenager's group	106	57/49	14-25	18.8±3.1
Person in middle and old age's group	122	65/57	50.1～84.5	66.5±10.6

As shown in Figure 1, the brain analytical approach in age described in the present embodiment, its concrete condition is as follows:

One, regression model is set up

1) data processing

For tranquillization state brain function image, first carry out time horizon correction, a normal moveout correction, registration, Spatial normalization carried out to all MR images, and remove locus deviation that head moves scope and template more than 5cm or angular deviation the data more than 20 °; Brain structure image is split, is divided into white matter of brain, ectocinerea and cerebrospinal fluid; Finally, Gaussian smoothing (halfwidth is 8mm) is carried out to image;

To through pretreated image, based on AAL template, application drawing opinion and information theory, set up brain 26S Proteasome Structure and Function network respectively.

2) signature analysis and extraction

The feature adopted comprises:

A) brain architectural feature: grey matter volume (graymattervolume, GMV);

B) brain function feature: locally coherence

Locally coherence (regionalhomogeneity, ReHo) is a kind of parameter investigating local cerebral functional activity situation, it reflects the similarity in brain function time-series image between set point and neighborhood point or synchronism.Obtain by calculating Kendall's coefficient (Kendall ' scoefficientofconcordanceKCC), it is defined as follows:

$W=\frac{\mathrm{\Σ}{\left(R_i\right)}^2-n{\left(\stackrel{\‾}{R}\right)}^2}{\frac{1}{12}{K}^2(n^3-n)}$

Wherein, W is Dare coefficient or ReHo value, R _ibe the total number of grades at i-th time point a certain count strong point and Neighborhood Number strong point, n functional imaging sequence number, the quantity of K set point and investigation neighborhood point.

C) brain network characterization: degree centrad

Degree centrad (degreecentrality, DC) centrality that brain network node connects is investigated, it reflects effect or the status of nodes, its value is larger, illustrate that connected node is more, connect maximum nodes and be considered to core node, they are in very important status in the information interaction of network.

After calculating above-listed three category features, carry out t inspection, analyze these features and increase the most significant brain district of change with the age.Wherein, the p value of grey matter volume is 0.02, and the p value of ReHo and DC of structure and fuction network is all set to 0.05, and adopts FalseDiscoveryRate (FDR) to correct.

3) Feature Dimension Reduction

Adopt iterative characteristic growing method to carrying out dimensionality reduction:

The given validity feature quantity k=100 at every turn chosen, GMV, ReHo, the dimension threshold value of the validity feature vector of structural network DC and functional network DC is respectively 6000,4500,4000,4000.

4) regression model is set up

After dimensionality reduction, obtain effective brain structure, brain function and brain network characterization, using it as independent variable, the age is dependent variable, and application linear SVM regression model, carries out regretional analysis.

Fig. 2 is the physiological age of population of adolescent training set and the corresponding relation in brain age that obtains through regretional analysis.

As calculated, this training set estimates that the difference average of brain age and actual physiological age is 1.03, and standard deviation is 0.51, and related coefficient is 0.93.Illustrate to there is close inner link between the brain age of this group crowd and actual physiological age.

Wherein solid line is linear regression Trendline, and the slope of dotted line is 1.The upper left point of dotted line, illustrates that its brain is greater than actual physiological age age, otherwise, illustrate that brain is less than physiological age age.

Fig. 4 is corresponding relation between the physiological age of mid-aged population test training set and the brain age obtained through regretional analysis.

As calculated, this training set estimates that the difference average of brain age and actual physiological age is 1.32, and standard deviation is 0.97, and related coefficient is 0.92.

By the brain analysis result in age of Different age group crowd, demonstrate this regression model and can set up effective internal association between estimation brain age and actual physiological age, but there are differences both also illustrating, i.e. inconsistency.

Two, brain is analyzed age

1) data processing

Select 21 examples of teenager's test set and 25 number of cases certificates of middle-aged and old test set, adopt the method for first step Data processing above respectively, brain image is processed, and set up brain network.

2) feature extraction

With reference to the validity feature index vector of the first step above, extract corresponding brain knot, brain function and brain network characterization respectively.

3) brain is analyzed age

Utilize the regression model that the first step is above set up, the brain age of assessment tested object.

Fig. 3 and Fig. 5 is respectively the physiological age of teenager's test set and middle-aged and old test set two groups of crowds and estimates brain corresponding diagram in age.

Teenager's test set and middle-aged and old test set, it estimates that the difference average of brain age and actual physiological age and standard deviation are respectively 1.20 ± 0.61,1.81 ± 0.96, and related coefficient is respectively 0.87,0.90.

Above result, demonstrate brain analytical model in age and the method for the present invention's proposition, effectively can assess teenager and middle-aged and old brain ages, error and standard deviation are all less, and brain age and physiological age exist close relationship.

The examples of implementation of the above are only the preferred embodiment of the present invention, not limit practical range of the present invention with this, therefore the change that all shapes according to the present invention, principle are done, all should be encompassed in protection scope of the present invention.

Claims (4)

1. based on a brain analytical approach in age for nuclear magnetic resonance image, it is characterized in that: first, utilize brain structure image information, the brain structural network of constructing function brain district and voxel level; According to cerebral function signal of video signal, set up the brain function network of function brain district and voxel level; Then, analyze the structure of brain, brain function increased and the characteristic sum rule of change with the age, and the change of brain network connection mode, thus obtain brain structure, brain function and brain network and increase and the otherness feature of change with the age; Utilize recursive feature growth method, the validity feature needed for selection thereupon, reduce intrinsic dimensionality; Finally, application linear SVM, sets up brain regression model in age, analyzes age brain.

2. a kind of brain analytical approach in age based on nuclear magnetic resonance image according to claim 1, is characterized in that, comprise the following steps:

1) regression model is set up

1.1) MR image data is gathered

Obtain brain structure and brain function MR image;

1.2) image data process

Comprise a normal moveout correction, registration, Spatial normalization, smoothing processing, wherein, to brain structure image carry out ectocinerea and white matter of brain analyze time, needing tissue segmentation is grey matter, white matter and cerebrospinal fluid, brain function image first also takes interbed and corrects, then carries out subsequent treatment;

Also comprise brain structure and brain function network struction, namely utilize through pretreated image, application drawing opinion and information theory, in function brain district or voxel level, set up brain 26S Proteasome Structure and Function network;

1.3) analysis and extraction of features

In function brain district or voxel level, application single-sample t-test method, analyzes brain structure, brain function image and the brain network characterization situation of change with the age, namely obtains the feature with age generation marked change;

1.4) Feature Dimension Reduction

Using iterative feature growth method, carries out dimensionality reduction to brain structure, brain function and brain network characterization respectively, selects validity feature, records the index of effective proper vector;

1.5) regretional analysis

By brain function, brain structure and the brain network characterization selected after fusion dimensionality reduction, using it as independent variable, using physiological age as dependent variable, set up linear regression model (LRM) therebetween, carry out regretional analysis;

2) brain is analyzed age

2.1) image acquisitions

Obtain the brain MR 26S Proteasome Structure and Function image of object to be analyzed;

2.2) image data process

Comprise a normal moveout correction, registration, Spatial normalization, smoothing processing, wherein, structure image must be split, and functional image first also takes interbed and corrects, then carries out subsequent treatment;

Utilize through pretreated image, application drawing opinion and information theory, in brain district or voxel level, set up brain 26S Proteasome Structure and Function network;

2.3) feature extraction

According to step 1.4) the validity feature index vector that obtains, extracts corresponding brain structure, brain function and brain network characterization;

2.4) brain is analyzed age

Utilize step 1.5) regression model set up, using the feature obtained as independent variable, brain predicts the brain age of object to be assessed as dependent variable age.

3. a kind of brain analytical approach in age based on nuclear magnetic resonance image according to claim 2, it is characterized in that: in step 1.3) in, brain structure, brain function image and brain network are with the variation characteristic at age, comprise brain structure image central gray, white matter, average diffusivity in DTI image and fractional anisotropy, brain blood flow in arterial spin labeling image, the Blood oxygen level dependence of fMRI, based on the local one value property of tranquillization state fMRI, low frequency amplitude, mark low frequency amplitude, and the limit in brain network characterization connects, the betweenness centrad of node, degree centrad, and all features all utilize z-score method to carry out returning generalized.

4. a kind of brain analytical approach in age based on nuclear magnetic resonance image according to claim 2, is characterized in that: in step 1.4) in, adopt following iterative characteristic growth method to carry out dimensionality reduction:

1.4.1) initialization: a certain feature vector, X of getting n training sample ₀=[X ₁, X ₂..., X _i... X _n] ^tand corresponding age vector y=[y ₁, y ₂..., y ₃..., y _n] ^t; The given validity feature quantity k at every turn chosen, or ratio r ∈ [01], then k=n × r; The total characteristic quantity expected is m, wherein m=c × k, c be greater than 1 integer;

1.4.2) initialization feature subset index vector s=[1,2 ..., n], treat sequencing feature vector X _candidate=X ₀, validity feature vector X _selected=[];

1.4.3) training sample X=X is selected according to index vector ₀(:, s), training regression model α=regression (X, y), and calculate weight vector W=Σ _iα _iy _ix _i, utilize score function c (i)=(w _i) ², sort from high to low by score, fetch bit arranges the feature X of front k _sorted=[1 ... k], added in effective sequencing feature vector, and from treating to shift out sequencing feature vector, upgraded validity feature vector X _selected, treat sequencing feature vector X _candidatewith index vector s;

1.4.4) step 1.4.3 is repeated) until the dimension of validity feature vector equals m;

1.4.5) step 1.4.1 is repeated) until all features all complete dimension-reduction treatment.