CN106023194A - Amygdaloid nucleus spectral clustering segmentation method based on resting state function connection - Google Patents

Abstract

The invention discloses an amygdaloid nucleus spectral clustering segmentation method based on resting state function connection, and the method is used for carrying out the automatic high-efficiency segmentation of an encephalic region based on a spectral clustering algorithm according to the similarity of internal voxel functions of an amygdaloid nucleus. The method comprises the steps: firstly carrying out the preprocessing of resting state magnetic resonance data; secondly carrying out the extraction of the encephalic region of the amygdaloid nucleus; thirdly carrying out the connection calculation of internal voxel whole-brain functions of the amygdaloid nucleus; and finally carrying out the spectral clustering segmentation of a function connection matrix. The automatic segmentation algorithm proposed by the invention and an amygdaloid nucleus clinic dissection result are enabled to be greatly consistent with each other, and the stability and noise interference resistance are enabled to be more satisfying. Compared with a conventional manual segmentation method, the method is simpler and more convenient and efficient, and is high in repeatability.

Description

The corpus amygdaloideum spectral clustering dividing method connected based on tranquillization state function

Technical field

The invention belongs to image processing field, be specifically related to a kind of based on the connection of tranquillization state function Corpus amygdaloideum spectral clustering dividing method, particularly relates to use spectral clustering based on each voxel of corpus amygdaloideum The full brain connection mode similarity method that carries out Semen Armeniacae Amarum nucleon brain district automatic segmentation.

Background technology

FMRI is one of main non-invasive methods of research cerebration, brain function, has millimeter The spatial resolution of level.The proposition of BOLD-fMRI method and the development research to brain cognitive function Having breakthrough progress, it has become neuroscience and has explored the important of human brain neuromechanism Instrument.Refer to as fMRI based on Blood oxygen level dependence (blood oxygen level-dependent, BOLD) nuclear magnetic resonance, it is by measuring the cerebral blood flow and brain blood oxygen caused by neural activity Cerebration is reacted in the magnetic resonance signal change caused Deng composition transfer.

Corpus amygdaloideum is the brain district of core the most in human emotion's memory network, and it is in the generation of emotion With express, society judges and plays important role in the cognitive activities such as facial recognition.Semen Armeniacae Amarum The exception of nuclear structure and function can cause the mental sickness such as autism, depression, Parkinson's disease, It has become as the focus of Neuropathological Study, but domestic and international at present, corpus amygdaloideum internal structure Relatively fewer with the Study on thinning of function, the most quickly corpus amygdaloideum brain district is carried out by exact automatic Divide and be particularly important.Human brain is the five-star part of central nervous system, it be by What 14000000000 neurocytes were constituted has the most systematical organ, and has identity function Cell group activity has great consistency, equally, inside corpus amygdaloideum, has identical merit The nucleus of energy is rolled into a ball its function and also can be embodied concordance in a way.Research shows tranquillization During state, the low-frequency fluctuation BOLD signal (0.01～0.1Hz) in brain Different brain region is relevant, Motor region, vision, language district, the dependency of low frequency amplitude fluctuation of auditory area are proved in succession, It is quiet that these relevant low-frequency fluctuations constitute that tranquillization state function connects between network, and different crowd It is consistent that breath state function connects network.In corpus amygdaloideum, there is the nucleus group of identical function, Its tranquillization state function also has the similarity of height.Pass through based on such theoretical basis present invention Tranquillization state functional imaging is similar according to the function connection mode of full brain of voxel each in specific brain regions district Property with spectral clustering, corpus amygdaloideum brain district is split, attempt further investigation merit within corpus amygdaloideum Can, fully understanding the mechanism of action of corpus amygdaloideum to come for the preventing and treating of relevant disease clinically provides plan Slightly.

At present, corpus amygdaloideum carries out segmentation mainly by there being the clinician on certain medical anatomy basis Corpus amygdaloideum brain district is manually divided by the feature according to corpus amygdaloideum brain district nucleus unity structure. On the one hand such dividing method has the highest medical anatomy context request, the opposing party to researcher Face inefficiency, the longest, it is unfavorable for promoting.

Summary of the invention

Based on the efficiency manually carrying out brain Division according to corpus amygdaloideum nucleus group architectural feature Low and time-consuming long, affected bigger present situation by subjective factors, it is an object of the invention to propose A kind of corpus amygdaloideum spectral clustering dividing method connected based on tranquillization state function, it is possible to stable, efficient, Exactly human brain corpus amygdaloideum region is carried out automatic segmentation.

In order to achieve the above object, the technical scheme is that

The corpus amygdaloideum spectral clustering dividing method connected based on tranquillization state function, specifically comprises the following steps that

(1), carry out tranquillization state magnetic resonance data acquisition to tested, and the data of collection are carried out Pretreatment, pretreatment refers to improve the signal to noise ratio of Brain mapping picture, and thin retaining view data While joint, tested image and standard form is made to carry out affine registration conversion；

(2), tested corpus amygdaloideum brain district is extracted, by pretreated tranquillization state functional MRI figure As registration to MNI normed space, use AAL (automated anatomical labeling) Cortex and infracortical grey matter are divided into 90 Ge Nao districts by template, left and right corpus amygdaloideum correspondence respectively It is No. 41 brain districts and No. 42 brain districts, equally corpus amygdaloideum ROI is registrated to MNI space, thus Extract tested corpus amygdaloideum brain district；

(3), by calculating the full brain connection mode of each voxel in corpus amygdaloideum, and then obtain every Full brain connection mode similarity matrix between two voxels, extract each voxel of corpus amygdaloideum time Between sequence signal；To all voxels letter in the tranquillization state signal employing brain district in other 88 Ge Nao districts Number average mode is extracted；Secondly, by calculate the voxel signal within each corpus amygdaloideum and its Correlation coefficient between the signal in his 88 Ge Nao districts, the full brain obtaining corpus amygdaloideum voxel connects square Battle array M, wherein (i j) represents that i-th corpus amygdaloideum voxel connects to the function in jth brain district to element Connect intensity, one corpus amygdaloideum voxel of each line description of this matrix under tranquillization state and full brain its Connection mode between his brain district, value is closer to 1, and dependency is the biggest；Then every two row are calculated Correlation coefficient between connection mode just can measure two voxels at full brain function connection mode On similarity, be designated as similarity matrix N, the most each element representation corpus amygdaloideum i-th body The plain similarity to jth voxel on full brain function connection mode；

(4), the spectral clustering full brain connection mode similarity matrix N to obtaining in step 3 is used Carry out cluster segmentation, it is achieved the subregion segmentation of corpus amygdaloideum.

Step (4) particularly as follows:

(1) sample point x is made₁,x₂,…,x_nRepresent that the function that the N being clustered arbitrarily often goes connects Vector, makes s_ijRepresent the connection weights between two summits of structural map, with Gauss similarity function Defining the weights between summit, wherein parameter σ is referred to as scale parameter, wherein s_ijIt is defined as

s_ij=exp (-| | x_i-x_j||²/2σ²)

Then the similarity matrix of sample is S=(s_ij) (i, j=1,2 ..., n)；

(2) input: similarity matrix S ∈ R^n×n, cluster classification number k；Set up the similar of sample Property connect figure, making W is its weight matrix；

(3) calculate non-standardization Laplacian Matrix L, each column element of W is added up Obtain N number of number, they are put on the diagonal, be the most all zero, form a N*N Matrix, be designated as D, and make L=D-W；Front k the minimum eigenvalue institute calculating L is right Characteristic vector v answered₁,v₂,…v_k；

(4) V ∈ R is made^n×kFor v₁,v₂,…v_kBy the matrix formed by row arrangement；

(5) for i=1,2 ..., n, make y_i∈R^kThe i-th row for matrix V；

(6) k mean algorithm is utilized will to belong to space R^kData y_i(i=1,2 ..., n) it is polymerized to k Class, is C₁,…C_k, just complete the cluster of each pixel of corpus amygdaloideum, it is achieved the sub-district of corpus amygdaloideum Regional partition.

The innovative point of the present invention is: with the contrast of corpus amygdaloideum anatomical atlas it appeared that the present invention The corpus amygdaloideum spectral clustering automatic segmentation algorithm connected based on tranquillization state function proposed and corpus amygdaloideum Clinical anatomic result obtains concordance significantly, on the other hand quantitative assessment segmentation result table Bright, anatomical atlas of the present invention and physiology collection of illustrative plates have the similarity of more than 70%, and in stability Also relatively satisfactory result is obtained with anti-noise jamming aspect, relative to traditional manual dividing method, Simpler, convenient, high many advantages efficient, repeatable.

Accompanying drawing explanation

Fig. 1 is the flow chart of the present invention.

Fig. 2 is tranquillization state function image pretreatment process figure.

Fig. 3 is comparison diagram before and after head segmentation.

Fig. 4 is a dynamic(al) correction result.

Fig. 5 is contrast before and after function picture registration, and left side is function picture before registration, and centre is MNI152 standard form, right side is function picture after registration, between registration image and template Difference reduce.

Fig. 6 is that corpus amygdaloideum is split by the method for corpus amygdaloideum physiology anatomical atlas and present invention proposition Comparative result figure.

Fig. 7 is spectral clustering segmentation corpus amygdaloideum result and juiche collection of illustrative plates Duplication cartogram.

Fig. 8 is that 30 tested left side corpus amygdaloideum segmentation results contrast with brain map spatial simlanty.

Fig. 9 is that 30 tested right side corpus amygdaloideum segmentation results contrast with brain map spatial simlanty.

Figure 10 is corpus amygdaloideum segmentation result stability contrast under different signal to noise ratio.

Detailed description of the invention

Below in conjunction with the accompanying drawings the present invention is done narration in detail.

The present invention carries out corpus amygdaloideum brain based on the connection of tranquillization state function and distinguishes segmentation method principle such as Fig. 1 Shown in.

(1), first the original tranquillization state MR data gathered is carried out pretreatment, due to magnetic Various effect of noise during resonance scan, individuality self exists on yardstick and position Difference, be highly desirable to before analytical data data are done certain pretreatment.Whole Experiment data acquisition in, main noise information source includes: (1) physical header moves；(2) Interlayer difference sweep time in image；(3) inhomogeneities of exterior magnetic field.Brain mapping picture is pre- Process is while retaining brain function image detail, uses Brain mapping picture to enter with standard form The pretreatment of row affine registration mapping mode, and improve the signal to noise ratio of Brain mapping picture.

The original tranquillization state MR data gathered is carried out pretreatment, uses fMRI data to locate in advance Reason, based on AFNI and FSL software programming Batch literary composition under linux system Ubuntu14.04 Part (data batch processing) completes.Pretreatment process is shown in Fig. 2, mainly comprises the following aspects:

1) structure is as head segmentation

The structure picture collected generally comprises head information, needs to carry out head segmentation, thus The artifact that elimination skull position the introduces impact on subsequent data analysis.There is provided based on FSL software 3drefit, 3dresample and fast segment realize head skull and brain interior tissue The segmentation of structure, segmentation result is as shown in Figure 3.

2) time unifying

Hematodinamics function shows that the response stimulated is had the regular hour to postpone by blood, due to Collecting full brain image during one TR, this causes the image of each layer not at synchronization Gathering, and be to occur in section whole sweep time, time adjustment is through being similar to interpolation Method each tomographic image is carried out a layer time-triggered protocol so that each tomographic image in the TR cycle Approximate and obtain at synchronization.

3) head dynamic(al) correction

Carry out tested brain typically being regarded a rigid body, therefore at fMRI as during dynamic(al) correction In experiment the motion of tested head can be approximated to a kind of rigid motion, the most only translation transformation with The combination of rotation transformation.Select the first two field picture of single-subject as reference picture, pass through The 3dvolreg function of AFNI makes remaining all image registrate with reference picture, and result is such as Shown in Fig. 4.If head is dynamic more than a voxel, remove this tested.

4) space smoothing

Space smoothing uses Gaussian function to carry out Gaussian smoothing, can effectively slacken random noise pair The impact of fMRI signal, improves the signal to noise ratio of data.Three-dimensional Gaussian function is the more commonly used Spatial smoothing method, its full width at half maximum determines the dynamics of space smoothing, selects half Gao Quan herein Wide (Full Width at Half Maximum, FWHM) is that the gaussian kernel function of 6mm is carried out Data smoothing.

5) time domain bandpass filtering

Tranquillization state fMRI signal is a kind of low-frequency fluctuation, and frequency is concentrated mainly on 0.01～0.1Hz, This low-frequency fluctuation reflects spontaneous neural activity.Therefore employing frequency range is The physiological noise relevant with breathing, heart beating etc. removed by the band filter of 0.01～0.1Hz.

6) go linearly

Owing to the long-term work of machine causes temperature to raise or tested being not suitable with, over time Accumulation can there is linear drift, need to carry out linear.

7) image segmentation

In order to remove cerebrospinal fluid (Cerebro-Spinal Fluid, CSF), white matter (White Matter, The redundant signals such as WM), needs to split structure picture, utilizes the information system that segmentation obtains Make cerebrospinal fluid and white matter template；

8) redundancy is removed

Remove white matter, cerebrospinal fluid, full brain signal and head and move the redundant signals such as artefact.

(2), in standard MNI space, tested corpus amygdaloideum brain district is extracted.In order to be accurately positioned Fructus Pruni Ren Henao district, first registrates tranquillization state magnetic resonance image (MRI) to MNI normed space；For improving Registration accuracy, uses two step registrations: first function picture is registrated to structure picture, secondly by structure As being registrated to normed space, thus function picture is registrated to standard by the transformation matrix obtained by utilizing Space be 3mm × 3mm × 3mm by voxel resampling；Registration template uses MNI template, should Template is by Montreal, CAN Neuroscience Research institute (Montreal Neurological Institute, MNI) research and development form；Whole registration process uses the linear registration that FMRIB provides Instrument realizes, and before and after registration, image comparison is as shown in Figure 5；AAL(automated anatomical Labeling) cortex and infracortical grey matter are divided into 90 Ge Nao districts, left and right Semen Armeniacae Amarum by template Core corresponds to No. 41 brain districts and No. 42 brain districts respectively, by same to No. 41 brain districts and No. 42 brain districts Registration, to standard MNI space, just can carry out as us as template splitting further Target brain district.

3, by step 2 at the position base having obtained tranquillization state functional MRI data corpus amygdaloideum On plinth, it is possible to obtain the time series of the single voxel of corpus amygdaloideum, on the other hand, AAL is calculated Other 88 Ge Nao district sequence average time in addition to left and right sides corpus amygdaloideum in template, then ask Obtain the correlation coefficient of each voxel of corpus amygdaloideum and this 88 Ge Nao district sequence average time, obtain Fructus Pruni The full brain connection matrix M of core nucleome element, wherein (i j) represents that i-th corpus amygdaloideum voxel is to jth to element The function bonding strength in Ge Nao district；One corpus amygdaloideum voxel of each line description of this matrix is quiet Connection mode under breath state and between complete other brain districts of brain, value is closer to 1, and dependency is the biggest； The reading of whole tranquillization state magnetic resonance image data, seasonal effect in time series are extracted and correlation calculations is equal Realizing based on python 2.7 language, this algorithm scientific algorithm part mainly uses Numpy (Numeric Python) module realizes, and NumPy is the calculating bag of a basic science, it Provide many senior numerical value programming tools, such as: the process of matrix data type, vector, with And the computing storehouse of precision, aim at and carry out strict digital processing and produce；To MR data Reading and write then relies on NiBabel module, NiBabel provides common medical image data The interface that form reads.

4, use spectral clustering that the full brain connection mode matrix N obtained in step 3 is clustered Segmentation, particularly as follows:

(1) sample point x is made₁,x₂,…,x_nRepresent that the function that the N being clustered arbitrarily often goes connects Vector, makes s_ijRepresent the connection weights between two summits of structural map, with Gauss similarity function Defining the weights between summit, wherein parameter σ is referred to as scale parameter, wherein s_ijIt is defined as

s_ij=exp (-| | x_i-x_j||²/2σ²)

Then the similarity matrix of sample is S=(s_ij) (i, j=1,2 ..., n)；

(2) input: similarity matrix S ∈ R^n×n, cluster classification number k；Set up the similar of sample Property connect figure, making W is its weight matrix；

(3) calculate non-standardization Laplacian Matrix L, each column element of W is added up Obtain N number of number, they are put on the diagonal, be the most all zero, form a N*N Matrix, be designated as D, and make L=D-W；Front k the minimum eigenvalue institute calculating L is right Characteristic vector v answered₁,v₂,…v_k；

(4) V ∈ R is made^n×kFor v₁,v₂,…v_kBy the matrix formed by row arrangement；

(5) for i=1,2 ..., n, make y_i∈R^kThe i-th row for matrix V；

(6) k mean algorithm is utilized will to belong to space R^kData y_i(i=1,2 ..., n) it is polymerized to k Class, is C₁,…C_k, just complete the cluster of each pixel of corpus amygdaloideum, it is achieved the sub-district of corpus amygdaloideum Regional partition.

The corpus amygdaloideum dividing method based on the connection of tranquillization state function proposed with regard to the present invention below takes Experimental result compare with corpus amygdaloideum clinical physiological anatomical atlas, and steady to this method The aspects such as qualitative, anti-noise jamming do certain discussion.

Calculate based on cytoarchitectural research, julich collection of illustrative plates in 05 year according to Amunts et al. And construct the probability template of these corpus amygdaloideum subprovinces.Corpus amygdaloideum be divide into three sons by this collection of illustrative plates Region, respectively Basolateral Nucleus (the laterobasal, LB), shallow table core (the Superficial, SF) and central medial nucleus (the centromedial, CM), corresponding diagram 6 respectively Blue, green and red brain district in Zuo Ce.LB includes secondary stratiform core, core, substrate at the end Nucleus lateralis and nucleus lateralis, CM is made up of inner side core and central nucleus, and SF includes corpus amygdaloideum pears shape skin Matter transitional areas, anterior amygdaloid area, corpus amygdaloideum-hippocampus and veutro and rear side cortex core.Fig. 6 Middle julich collection of illustrative plates demonstrates the spatial relation of corpus amygdaloideum subregion, is then this on the right side of Fig. 6 Invent to be connected with other each brain zone function based on corpus amygdaloideum voxel of object and carry out spectral clustering to Semen Armeniacae Amarum The result that core divides, the two carries out contrast can find that the cluster result of the present invention and collection of illustrative plates have height Degree concordance, shows the feasibility of this method.

On the other hand, for further quantitative assessment segmentation result, the present invention repeats 30 Tested tranquillization state functional image data carries out corpus amygdaloideum brain and distinguishes and cut, from the accuracy of algorithm and It is inquired into by two angles of stability.

Karr Pearson came algorithm based on Mark Lewis-Francis Galton proposes the linear phase of Pearson came Close coefficient, generally this correlation coefficient also referred to as " Pearson correlation coefficients ".Correlation coefficient ρ is examined Examining the degree of correlation of two variablees, span is between-1～1, and wherein, 1 represents that variable is complete Full positive correlation, 0 represents unrelated, and-1 represents perfect negative correlation.

${\mathrm{\ρ}}_{X,Y}=\frac{\mathrm{cov}(X,Y)}{{\mathrm{\σ}}_X{\mathrm{\σ}}_Y}=\frac{E\[(X-{\mathrm{\μ}}_X)(Y-{\mathrm{\μ}}_Y)\]}{{\mathrm{\σ}}_X{\mathrm{\σ}}_Y}$

In formula: ρ (X, Y) variable X and the correlation coefficient of Y；Cov (X, Y) variable X and The covariance of Y；σ_XThe standard variance of variable X；σ_YThe standard side of variable Y Difference；μ_XThe average of variable X；μ_YThe average of variable Y.

We see 3-D view to be measured and the reference picture B of size m × n × o as stochastic variable, Then the correlation coefficient between two images is

$CC(A,B)=\frac{\underset{i=1}{\overset{m}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}\underset{k=1}{\overset{o}{\Σ}}(A_{i,j,k}-\stackrel{\‾}{A})(B_{i,j,k}-\stackrel{\‾}{B})}{\sqrt{\left(\underset{i=1}{\overset{m}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}\underset{k=1}{\overset{o}{\Σ}}{\left(A_{i,j,k}-\stackrel{\‾}{A}\right)}^2\right)\left(\underset{i=1}{\overset{m}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}\underset{k=1}{\overset{o}{\Σ}}{\left(B_{i,j,k}-\stackrel{\‾}{B}\right)}^2\right)}}$

In formula, A represents 3-D view to be evaluated, and B represents reference picture, A_i,j,kRepresent in testing image Pixel, B_i,j,kRepresent the pixel in reference picture,Represent the average of image A,Table Diagram is as the average of B, and m represents that picturedeep, n represent the columns of image, and o represents image Highly.The correlation coefficient of image A and image B is the least (closer to 0), explanatory diagram as A and The similarity of image B is the least；Correlation coefficient the biggest (closer to 1), explanatory diagram is as A and B Similarity the biggest.The phylogenetic group analysis result calculating the present invention is corresponding with julich collection of illustrative plates respectively Corpus amygdaloideum subregion Pearson correlation coefficients as it is shown in fig. 7, result closer to 1 represent two Person's spatial distribution is the most similar, and CM subprovince, right side segmentation result is closest with collection of illustrative plates as seen from the figure (CC=0.78), right side SF subprovince relatively low relative to spectrogram similarity (CC=0.65).

The present invention segmentation result by each tested corpus amygdaloideum subprovince and the group analysis of this data set Result carries out space correlation, acute with the change of the space overlap rate of group analysis result with all subregion Strong degree reflects the stability of partitioning algorithm herein, Fig. 8 and Fig. 9 describes left side and right side Semen Armeniacae Amarum nucleon district single-subject and the overlapping cases of group analysis result, empty by CM on the right side of calculating Between the variance of correlation coefficient be 0.0303 maximum, show the result stability ratio split this subprovince More weak, and the variance of LB brain district, right side space correlation coefficient is 0.0017 minimum, so this The segmentation result repeatability of brain district clustering algorithm is the highest, and stability is preferably also (other brain district Variance be respectively var (CML)=0.0041, var (LBL)=0.0185, var (SFR)= 0.0098, var (SFL)=0.0194)

On the basis of the accuracy discussing partitioning algorithm and stability, superposition varying strength Noise carries out brain Division again to investigate inventive algorithm in the time series of corpus amygdaloideum voxel Vigorousness.First the time series of each voxel in brain district, corpus amygdaloideum left and right is extracted, pass through Calculating and signal to noise ratio is respectively 10db, the white Gaussian noise of 30db, 50db, 70db, 90db is added to Each time series carries out dividing based on the corpus amygdaloideum subprovince on the basis of spectral clustering shown in Fig. 1 again Cutting algorithm, whole experimental simulation process realizes at python 2.7 platform.

After superimposed noise, the cluster segmentation of 6 subprovinces of corpus amygdaloideum is as shown in Figure 10, wherein (A) Figure reflects under different noise based on ROI voxel time series and remaining 88 brain of ALL template The function connection matrix that district's mean sequence calculates carries out segmentation result and does not add the feelings of noise Kuang Xiagezi district spatial simlanty situation of change, the view data of different signal to noise ratios is equal as seen from the figure Achieve satisfied segmentation result, in the case of signal to noise ratio is more than 70db, Ge Zi district All maintain the similarity of the result 100% being close to and do not add noise, at the low letter of 10db Zao Bixiage district Pearson correlation coefficients also can keep about 0.7.

In sum, present invention aim at proposing one based on tranquillization state functional mri Can stablize, efficiently, automatization's human brain corpus amygdaloideum region segmentation method accurately, for clinic The research further of the emotion related psychiatric conditions such as upper depression, autism and preventing and treating provide plan Slightly.

Claims (2)

1. the corpus amygdaloideum spectral clustering dividing method connected based on tranquillization state function, it is characterised in that Specifically comprise the following steps that

(1), carry out tranquillization state magnetic resonance image data collection to tested, and carry out magnetic resonance image (MRI) Data prediction, magnetic resonance image data pretreatment is while retaining image detail, uses Brain mapping picture and standard form carry out the pretreatment of affine registration mapping mode, and improve brain merit The signal to noise ratio of energy image；

(2), extract tested corpus amygdaloideum brain district, use AAL (automated anatomical Labeling) cortex and infracortical grey matter are divided into 90 Ge Nao districts, left and right Semen Armeniacae Amarum by template Core corresponds to No. 41 brain districts and No. 42 brain districts respectively, equally corpus amygdaloideum ROI is registrated to MNI Space, thus extract tested corpus amygdaloideum brain district；

(3), by calculating the full brain connection mode of each voxel in corpus amygdaloideum, and then every two are obtained Full brain connection mode similarity matrix between individual voxel, extracts the time of each voxel of corpus amygdaloideum Sequence signal；To all voxel signals in the tranquillization state signal employing brain district in other 88 Ge Nao districts Average mode is extracted；Secondly, by calculate the voxel signal within each corpus amygdaloideum and other Correlation coefficient between the signal in 88 Ge Nao districts, obtains the full brain connection matrix of corpus amygdaloideum voxel M, wherein (i j) represents that i-th corpus amygdaloideum voxel connects to the function in jth brain district to element Intensity.One corpus amygdaloideum voxel of each line description of this matrix under tranquillization state and full brain other Connection mode between brain district, value is closer to 1, and dependency is the biggest；Then every two row are calculated even Connect the correlation coefficient between pattern and just can measure two voxels on full brain function connection mode Similarity, be designated as similarity matrix N, the most each element representation corpus amygdaloideum i-th voxel To jth voxel similarity on full brain function connection mode；

(4), use spectral clustering that the full brain connection mode similarity matrix N obtained in step 3 is entered Row cluster segmentation, it is achieved the subregion segmentation of corpus amygdaloideum.

The corpus amygdaloideum spectral clustering connected based on tranquillization state function the most according to claim 1 Dividing method, it is characterised in that step (4) particularly as follows:

(1) sample point x is made₁,x₂,…,x_nRepresent that the function that the N being clustered arbitrarily often goes connects Vector, makes s_ijRepresent the connection weights between two summits of structural map, with Gauss similarity function Defining the weights between summit, wherein parameter σ is referred to as scale parameter, wherein s_ijIt is defined as

s_ij=exp (-| | x_i-x_j||²/2σ²)

Then the similarity matrix of sample is S=(s_ij) (i, j=1,2 ..., n),

(2) input: similarity matrix S ∈ R^n×n, cluster classification number k；Set up the phase of sample Connecting figure like property, making W is its weight matrix；

(3) calculate non-standardization Laplacian Matrix L, each column element of W is added up Obtain N number of number, they are put on the diagonal, be the most all zero, form a N*N Matrix, be designated as D, and make L=D-W；Front k the minimum eigenvalue institute calculating L is right Characteristic vector v answered₁,v₂,…v_k；

(4) V ∈ R is madeⁿ×^kFor v₁,v₂,…v_kBy the matrix formed by row arrangement；

(5) for i=1,2 ..., n, make y_i∈R^kThe i-th row for matrix V；

(6) k mean algorithm is utilized will to belong to space R^kData y_i(i=1,2 ..., n) it is polymerized to k Class, is C₁,…C_k, just complete the cluster of each pixel of corpus amygdaloideum, it is achieved the sub-district of corpus amygdaloideum Regional partition.