CN103838941A - Average value comparison method for nuclear magnetic resonance data processing - Google Patents

Abstract

The invention discloses an average value comparison method for nuclear magnetic resonance data processing. The average value comparison method for nuclear magnetic resonance data processing comprises the following steps that an SPM software package is used for preprocessing data so that the processing quality of following data can be ensured; fMRI data are marked through a matrix; a TCA method, an MTCA method and an IAM method are combined, and finally a CAM method is used for solving problems. The reaction time domain, reaction duration time and reaction tendency of the brain can be detected when the brain is stimulated in some extent. Particularly, the reaction time domain of the brain is displayed by counting the number of points higher than a self baseline value, and the average value comparison method for nuclear magnetic resonance data processing has the advantages of being simple in algorithm, high in calculation speed and the like.

Description

A kind of comparison mean value method of nuclear magnetic resonance data processing

Technical field

The present invention designs a kind of disposal route of nuclear magnetic resonance data, particularly a kind of statistical method for functional MRI data.

Background technology

Functional MRI (fMRI, functional magnetic resonance imaging) is a kind of emerging neuroimaging mode, and its principle is to utilize the magnetic radiography that shakes to measure the hemodynamic change that neuron activity causes.At present be mainly brain or the spinal cord that is used in research people and animal.FMRI is a kind of method of new research human brain function, there is the feature high without wound, time and spatial resolution, gradually be applied to multiple fields of Neuscience and " illustrating navigation in functional cohesion between higher nerve physiology and neuropsychological manner and cortex, art to excise to greatest extent Motor cortex pathology and to reduce postoperative complication, the differentiation degree of understanding brain tumor and prognosis judgement, disclose the aspect such as nerve and the abnormal pathophysiological change of mental illness function of cortex, all shown higher using value.Magnetic resonance imaging (functional magnetic resonance imaging, fMRI) technology can show the subtle change of the magnetic resonance signal that in brain regional internal jugular vein capillary, blood oxygenation state is risen.FMRI, as harmless and dynamic Detection Techniques, becomes observation brain activity day by day, and then discloses a kind of important method of brain and thinking relation.

Functional mri has become one of the most frequently used imaging technique of research human brain.Apply positron emission computerized tomography technology (PET scans), or be referred to as the research of PET scanning technique, give the tested radioactivity material not of the same race (but very safe) of taking, these materials are absorbed by movable brain cell in brain.Magnetic resonance imaging (magnetic resonance imaging, MRI) utilizes in magnetic field and rf wave brain and produces pulse energy, because pulse may be tuned to different frequency range, makes some atoms and magnetic field coupling.When the moment that magnetic-pulse is switched off, these atomic vibrations (resonance) also turn back to oneself initial state, and special radio frequency receiver detects these resonance and the channel information for computing machine thereof, produce accordingly the positioning image in homoatomic Nao district not.The new technology of Functional MRI, combines above-mentioned two technical advantages, and the changes of magnetic field that enters brain cell by inspection blood flow realizes cerebral function imaging, and it provides more accurate structure-function relationship.Current method is to find to activate brain district by statistical means such as t inspection and correlation analyses mostly, carry out functional localization, as use software statistics Parameter Map (Statistical Parametric Mapping, and functional neurosurgery imaging analysis (Analysis of Function NeuroImages, AFNI) etc. SPM).But how unknown to activating lasting time and general trend in brain.A kind of temporal cluster analysis method (time cluster analysis, TCA) method, can calculate easily the time zone of reaction in brain, correction time bunch analytic approach (the modified TCA proposing again subsequently, and average gray method (intensity average method MTCA), IAM) also can draw and stimulate the situation of change of the rear brain response of generation in time domain, and respectively have feature.

Summary of the invention

The problems referred to above that exist for solving prior art, the present invention will design a kind of new number that is greater than the point of self baseline value by statistics, show the time zone of brain response, thereby for functional MRI (functional magnetic resonance imaging, fMRI) statistical method of data---relatively mean value method (compare average method, CAM).

To achieve these goals, technical scheme of the present invention is as follows: a kind of CAM method being applied in nuclear magnetic resonance data processing, comprises the following steps:

A, data pre-service.

A1, be spatial alignment, each two field picture in an experiment sequence all with this sequence in the first two field picture align according to certain algorithm, correct head and move;

A2, Spatial normalization, because everyone brain is variant in anatomical structure, need to carry out Spatial normalization processing brain image tested difference, is translated into size and towards all identical standardized images.

The matrix representation of B, fMRI data.

Have m tissue points through the every width of pretreated fMRI image, suppose to gather altogether n width image, its data can represent with following matrix, and it is capable that this matrix has n, m row.

$I=\left\{\begin{array}{ccccc}{I}_{\mathrm{1,1}}& I_{\mathrm{1,2}}& I_{\mathrm{1,3}}& .........& I_{1,m}\\ I_{\mathrm{2,1}}& I_{\mathrm{2,2}}& I_{\mathrm{2,3}}& .........& I_{2,m}\\ .& .& .& & .\\ .& .& .& .........& .\\ .& .& .& & .\\ I_{n,1}& I_{n,2}& I_{n,3}& .........& I_{n,m}\end{array}\right\}---\left(1\right)$

Wherein matrix element I _i,jbe that i moment j place spatial point voxel value (is I=[I _i,j], i=1,2 ..., n, j=1,2 ..., m).

C, TCA method and MTCA method.

C1, the average baselining value of establishing j tissue points signal intensity are I _{0, j}, can derive the matrix S with same dimension, wherein S by matrix I _i,jfollowing calculating:

$S_{i,j}=\frac{|I_{i,j}-{\mathrm{<figure-callout\; id="0"\; label="I"\; filenames="HDA00002444998200011.png,HDA00002444998200022.png"\; state="\{\{state\}\}">I</figure-callout>}}_{0,j}|}{{\mathrm{<figure-callout\; id="0"\; label="I"\; filenames="HDA00002444998200011.png,HDA00002444998200022.png"\; state="\{\{state\}\}">I</figure-callout>}}_{0,j}}(i=\mathrm{1,2},...,n,j=\mathrm{1,2},...,m)---\left(2\right)$

C2, equally can be by matrix S derivational matrix W,

$W_{i,j}=\left\{\begin{array}{cc}1,& S_{i,j}=\max \{S_{1,j},S_{2,j},...,S_{n,j}\}\\ 0,& \mathrm{otherwise}\end{array}\right.(i=\mathrm{1,2},...,n,j=\mathrm{1,2},...,m);---\left(3\right)$

C3, formula (3) are that in whole change procedure, whether the j place spatial point in i time point reaches peaked judgement.It should be noted that equation (2) got absolute value, this just takes into account and promotes and suppress two kinds of cerebrations.Equation (3) can be used to select one group of tissue points, and this group tissue points is put and reached maximal value at one time.In this way, one group of tissue points selecting at each time point is Using Temporal Clustering.The size (tissue points number) of the Using Temporal Clustering element in each row vector of matrix W and that be exactly corresponding time point, Ke Yiyong

represent.Can obtain thus the vectorial K that describing one dimension time domain changes (Using Temporal Clustering size variation):

K＝(K ₁,K ₂,K ₃,…,K _n), (4)

Draw K value curve taking time point as horizontal ordinate, can be used for detecting the time zone that cerebration occurs, and the pattern of assumed response in advance.

C4, MTCA method are similar to TCA method, and just the weighted value of matrix W selection result is the gray-scale value I of original voxel _i,jinstead of 1.The matrix element W of MTCA method _i,jselection mode represent as follows:

$W_{i,j}=\left\{\begin{array}{cc}{I}_{i,j},& I_{i,j}=\max \{I_{1,j},I_{2,j},...,I_{n,j}\}\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(5\right)$

Equally, the variation of one dimension time domain (Using Temporal Clustering size variation) K still with the row of matrix W and represent.The K only here represents the overall intensity of each Using Temporal Clustering.

D, IAM method.

D1, the average baselining value of establishing j tissue points signal intensity are I _{0, j}, can derive the matrix S with same dimension by matrix I,

S _i,j=I _i,j-I _0,j （i＝1,2,…,n,j=1,2,…m）(6)

D2, do not need to calculate greatly or minimal value, i width view data average gray value is calculated as follows,

$K_i=\left(\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{S}_{i,j}\right)/m---\left(7\right)$

The average gray value that calculates n width image just can draw K value curve,

K=(K ₁,K ₂,K ₃,…,K _n) (8)

Draw the K value curve on time shaft, as the foundation that detects brain response time zone.

E, comparison averaging method (CAM method)

E1, derive the average baselining vector (I during each spatial point gathers by matrix I _0,1, I _0,2, I _0,3..., I _{0, m}), and compared with matrix I derivational matrix W, W _i,jfollowing calculating:

$W_{i,j}=\left\{\begin{array}{cc}1,& I_{i,j}\&GreaterEqual;{\mathrm{<figure-callout\; id="0"\; label="I"\; filenames="HDA00002444998200011.png,HDA00002444998200022.png"\; state="\{\{state\}\}">I</figure-callout>}}_{0,\mathrm{<figure-callout\; id="0"\; label="j"\; filenames="HDA00002444998200011.png,HDA00002444998200022.png"\; state="\{\{state\}\}">j</figure-callout>}}\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(9\right)$

E2, with TCA principle, calculate and represent the vectorial K of Using Temporal Clustering by the row of matrix W:

K＝(K ₁,K ₂,K ₃,……,K _n) (10)

Wherein

draw the K value curve of time shaft, as the foundation that detects brain response time zone.

E3, vectorial K is gone to radix processing.

If in the time that the tissue points in image is many (m is very large), at this moment the radix of vectorial K will become very large, it is a less number by comparison that the voxel that is greater than mean value is counted, the time zone of the reflection cerebration that the K value curve of time shaft just can not be clearly.Therefore, need to go radix processing to vectorial K.

K _i′＝K _i＝min(K ₁,K ₂,…,K _n) (11)

Thus, draw the vectorial K ' of a new expression Using Temporal Clustering:

K′＝(K ₁′,K ₂′,K ₃′,…,K _n′)(12)

Because K ' is through past radix processing, therefore K _i' what represent is the number higher than baseline value tissue points in each width brain image.The time shaft curve being drawn by K ' value also can better detect the variation of brain response time domain.

Compared with prior art, the present invention has following beneficial effect:

1, the present invention, by CAM method is applied in to functional MRI data processing, has simplified calculating greatly, has improved arithmetic speed.

2, the present invention is greater than the number of the point of self baseline value by statistics, shows the time zone of brain response, can detect the time zone of brain response in the time stimulating through some, and duration of the reaction and reaction tendency.

Brief description of the drawings

4, the total accompanying drawing of the present invention, wherein:

Fig. 1 is the result of calculation that adopts TCA method in the present invention;

Fig. 2 is the result of calculation that adopts MTCA method in the present invention;

Fig. 3 adopts IAM method result of calculation in the present invention.

Fig. 4 adopts CAM method result of calculation in the present invention.

Embodiment

In order to reflect the feature of this several method, we adopt the experimental data of knowing in advance brain response time zone.This test figure is the university student about 24 years old from 6, and experimental duties are tongue motion, and selected magnetic resonance scanner model is GE Echospeed 1.5T.Gather altogether 12min data (2min controls, 3min task, 2min control, 3min task, 2min control), every 2 seconds frames, totally 360 frames, after every two field picture standardization, form is 79 × 95 × 68 dot matrix, meter totally 510340 points.

In experiment, adopt respectively TCA, MTCA, IAM and CAM to process data, therefrom draw initial time and the duration of brain function reaction.Experimental result (having removed the first two field picture) is as shown in Fig. 1 (TCA method), Fig. 2 (MTCA method), Fig. 3 (IAM method), Fig. 4 (CAM).From 0 to 58 is the control stage, and 59 to 148 is task phase, and 149 to 208 is the control stage, and 209 to 298 is task phase, and 299 to 358 is the control stage.From this four width figure, can see, though 299 to 358 for the control stage, brain response is very strong, and this may be due to due to tested psychological application and brain response postpone.In a word, contrast this four width figure plots changes, be not difficult to draw CAM method observation reaction time region on effect the most obviously, the most directly perceived.

When the contained quantity of information of every two field picture few (being that m is little), and while stimulating number of times few, initial time and the duration of brain response after these four kinds of methods all can detect and stimulate, but in the time that the contained quantity of information of every two field picture is many or stimulate often, as Fig. 1, Fig. 2, Fig. 3, shown in Fig. 4, only have CAM can better reflect the time zone of brain response.This is because CAM looks for maximum point, but by each point signal intensity compared with mean value, add up all numbers that are greater than mean point in every width brain image; And TCA has only added up the number that reaches maximum of points in every width image, and the larger point of a series of signal intensity such as inferior Gao Dian is not all added up.Therefore, the information of CAM statistics is more complete, and the effect of the time zone of reflection brain work is also better.

Claims (1)

1. a comparison mean value method for nuclear magnetic resonance data processing, is characterized in that: comprise the following steps:

A, data pre-service;

A1, be spatial alignment, each two field picture in an experiment sequence all with this sequence in the first two field picture align according to certain algorithm, correct head and move;

A2, Spatial normalization, because everyone brain is variant in anatomical structure, need to carry out Spatial normalization processing brain image tested difference, is translated into size and towards all identical standardized images;

The matrix representation of B, fMRI data;

Have m tissue points through the every width of pretreated fMRI image, suppose to gather altogether n width image, its data can represent with following matrix, and it is capable that this matrix has n, m row;

$I=\left\{\begin{array}{ccccc}{I}_{\mathrm{1,1}}& I_{\mathrm{1,2}}& I_{\mathrm{1,3}}& .........& I_{1,m}\\ I_{\mathrm{2,1}}& I_{\mathrm{2,2}}& I_{\mathrm{2,3}}& .........& I_{2,m}\\ .& .& .& & .\\ .& .& .& .........& .\\ .& .& .& & .\\ I_{n,1}& I_{n,2}& I_{n,3}& .........& I_{n,m}\end{array}\right\}---\left(1\right)$

Wherein matrix element I _i,jbe that i moment j place spatial point voxel value (is I=[I _i,j], i=1,2 ..., n, j=1,2 ..., m);

C, TCA method and MTCA method;

C1, the average baselining value of establishing j tissue points signal intensity are I _{0, j}, can derive the matrix S with same dimension, wherein S by matrix I _i,jfollowing calculating:

$S_{i,j}=\frac{|I_{i,j}-I_{0,j}|}{I_{0,j}}(i=\mathrm{1,2},...,n,j=\mathrm{1,2},...,m)---\left(2\right)$

C2, equally can be by matrix S derivational matrix W,

$W_{i,j}=\left\{\begin{array}{cc}1,& S_{i,j}=\max \{S_{1,j},S_{2,j},...,S_{n,j}\}\\ 0,& \mathrm{otherwise}\end{array}\right.(i=\mathrm{1,2},...,n,j=\mathrm{1,2},...,m);---\left(3\right)$

C3, formula (3) are that in whole change procedure, whether the j place spatial point in i time point reaches peaked judgement; It should be noted that equation (2) got absolute value, this just takes into account and promotes and suppress two kinds of cerebrations; Equation (3) can be used to select one group of tissue points, and this group tissue points is put and reached maximal value at one time; In this way, one group of tissue points selecting at each time point is Using Temporal Clustering; The size of the Using Temporal Clustering element in each row vector of matrix W and that be exactly corresponding time point, Ke Yiyong

represent; Can obtain thus the vectorial K that describing one dimension time domain changes:

K＝(K ₁,K ₂,K ₃,…,K _n), (4)

Draw K value curve taking time point as horizontal ordinate, can be used for detecting the time zone that cerebration occurs, and the pattern of assumed response in advance;

C4, MTCA method are similar to TCA method, and just the weighted value of matrix W selection result is the gray-scale value I of original voxel _{i, j}instead of 1; The matrix element W of MTCA method _{i, j}selection mode represent as follows:

$W_{i,j}=\left\{\begin{array}{cc}{I}_{i,j},& I_{i,j}=\max \{I_{1,j},I_{2,j},...,I_{n,j}\}\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(5\right)$

Equally, the variation K of one dimension time domain still with the row of matrix W and represent; The K only here represents the overall intensity of each Using Temporal Clustering;

D, IAM method;

D1, the average baselining value of establishing j tissue points signal intensity are I _{0, j}, can derive the matrix S with same dimension by matrix I,

S _i,j＝I _i,j-I _0,j （i＝1,2,…,n,j=1,2,…m）(6)

D2, do not need to calculate greatly or minimal value, i width view data average gray value is calculated as follows,

$K_i=\left(\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{S}_{i,j}\right)/m---\left(7\right)$

The average gray value that calculates n width image just can draw K value curve,

K＝(K ₁,K ₂,K ₃,…,K _n) (8)

Draw the K value curve on time shaft, as the foundation that detects brain response time zone;

E, comparison averaging method

E1, derive the average baselining vector (I during each spatial point gathers by matrix I _0,1, I _0,2, I _0,3..., I _{0, m}), and compared with matrix I derivational matrix W, W _i,jfollowing calculating:

$W_{i,j}=\left\{\begin{array}{cc}1,& I_{i,j}\&GreaterEqual;I_{0,j}\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(9\right)$

E2, with TCA principle, calculate and represent the vectorial K of Using Temporal Clustering by the row of matrix W:

K＝(K ₁,K ₂,K ₃,……,K _n) (10)

Wherein

draw the K value curve of time shaft, as the foundation that detects brain response time zone;

E3, vectorial K is gone to radix processing;

If m is very large in the time that the tissue points in image is many, at this moment the radix of vectorial K will become very large, and it is a less number by comparison that the voxel that is greater than mean value is counted, the time zone of the reflection cerebration that the K value curve of time shaft just can not be clearly; Therefore, need to go radix processing to vectorial K;

K _i′=K _i-min(K ₁,K ₂,…,K _n) (11)

Thus, draw the vectorial K ' of a new expression Using Temporal Clustering:

K′=(K ₁′,K ₂′,K ₃′,…,K _n′) (12)

Because K ' is through past radix processing, therefore K _i' what represent is the number higher than baseline value tissue points in each width brain image; The time shaft curve being drawn by K ' value also can better detect the variation of brain response time domain.

Images