CN103985092B - A kind of post processing noise-eliminating method that plural number fMRI data are carried out ICA analysis - Google Patents

A kind of post processing noise-eliminating method that plural number fMRI data are carried out ICA analysis Download PDF

Info

Publication number
CN103985092B
CN103985092B CN201410191416.6A CN201410191416A CN103985092B CN 103985092 B CN103985092 B CN 103985092B CN 201410191416 A CN201410191416 A CN 201410191416A CN 103985092 B CN103985092 B CN 103985092B
Authority
CN
China
Prior art keywords
voxel
tested
noising
brain district
phase
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201410191416.6A
Other languages
Chinese (zh)
Other versions
CN103985092A (en
Inventor
林秋华
于谋川
龚晓峰
丛丰裕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dalian University of Technology
Original Assignee
Dalian University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Dalian University of Technology filed Critical Dalian University of Technology
Priority to CN201410191416.6A priority Critical patent/CN103985092B/en
Publication of CN103985092A publication Critical patent/CN103985092A/en
Application granted granted Critical
Publication of CN103985092B publication Critical patent/CN103985092B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

A kind of post processing noise-eliminating method that plural number fMRI data carry out ICA analysis, belongs to plural number fMRI data analysis field.For first obtaining its phase image phase place span for (π without the spatial activation brain district composition of phase ambiguity, π], then definition [π/4, π/4] in the range of voxel corresponding to phase value for activating voxel, the voxel that other phase values are corresponding is interference voxel, builds single tested de-noising mask and how tested group average de-noising mask accordingly, is respectively acting on single tested and group average assay, remove the voxel that amplitude is small again, just obtain final de-noising result.The present invention can ensure that complete plural fMRI data are analyzed by ICA, and then the data solving to cause because of pretreatment de-noising abandon and brain information loss problem.For the plural fMRI data gathered under Motor stimulation, pretreatment Denoising Method only estimates 26 independent elements, and the present invention can estimate 49 independent elements.

Description

A kind of post processing noise-eliminating method that plural number fMRI data are carried out ICA analysis
Technical field
The ICA that the present invention relates to plural number fMRI data analyzes, and particularly relates to a kind of plural number fMRI data are carried out ICA divide The post processing noise-eliminating method of analysis.
Background technology
Brain function research is worldwide emphasis and difficult point, and functional mri (functional Magnetic resonance imaging, fMRI) by its not damaged and high spatial resolution advantage, it has also become brain function grinds One of important means studied carefully.By using the analysis method such as GLM (general linear model) of model-driven, or data The analysis method such as independent component analysis (independent component analysis, ICA) driven, people can be from The spatial activation brain district information that fMRI extracting data is abundant.
FMRI data are plural numbers, including amplitude data and phase data.But major part fMRI analyzes method and only analyzes amplitude Data, and abandon phase data completely.Its reason is, the noise ratio amplitude data of phase data is big and characteristic is unknown, and this causes The analysis making plural number fMRI data is extremely challenging.But, containing unique brain function composition in fMRI phase data, in order to take off Show complete brain function information, it is necessary to be used.Plural number fMRI number is utilized to this end, started to explore from the beginning of, people from 2000 According to ICA method.2002, Calhoun et al. first ICA is used for plural number fMRI data analysis (Calhoun, V.D., Adali,T.,Pearlson,G.,Van Zijl,P.,Pekar,J.,2002.Independent component analysis of fMRI data in the complex domain.Magnetic Resonance in Medicine48,180-192)。 In order to eliminate phase noise, the method had abandoned the voxel of brain first half before ICA.In other words, ICA now is only Make use of the plural fMRI data of about 50%.Even so, compared with the ICA method only analyzing fMRI amplitude data, The number of voxels that activates continuously of Calhoun method extracted task Related Component the most averagely exceeds 12-23%.2009, Rodriguez et al. proposes a kind of new phase place noise-eliminating method phase masses figure (phase quality map) Denoising Method (Rodriguez,P.A.,Correa,N.M.,Eichele,T.,Calhoun,V.D.,Adali,T.,2009.Quality map thresholding for de-noising of complex-valued fMRI data and its application to ICA of fMRI.IEEE International Workshop on Machine Learning for Signal Processing(MLSP),1-6).The method is by true higher than the voxel being manually set thresholding for phase place change in fMRI phase data It is set to noise voxel, before ICA, it is removed from plural number fMRI data.Experiment shows, this part is removed data and accounted for The 30% of fMRI total amount of data, say, that when utilizing phase masses drawing method to carry out de-noising, ICA can utilize about 70% Plural number fMRI data, data user rate is improved, the brain function information more horn of plenty recovered.Therefore, phase masses figure disappears Method of making an uproar is analyzed in document at the ICA of many plural fMRI data and is applied.
But, it to be not difficult to find out, above two eliminates the method for phase noise and all eliminated part plural number fMRI before ICA Data (about at least 30%), these are removed and also contain abundant brain function information in data.To gather under a kind of Motor stimulation Plural fMRI data instance, before de-noising, the plural number independent element number that comprised of fMRI data is not less than 40, but phase place de-noising After reduce to 26, say, that about 14 brain function compositions lack because of the disappearance of carrier data.It is visible, in ICA analyzes Utilizing plural number fMRI data is to extract the key of complete brain function information.
Summary of the invention
It is an object of the invention to, cancel the phase place de-noising pretreatment before ICA, the substitute is and carry after ICA For a kind of method that spatial activation brain district composition is carried out effective de-noising, ensure that ICA can be to single tested and how tested plural number FMRI data are analyzed, and the data solving to cause because of de-noising pretreatment abandon and brain information loss problem.
The technical scheme is that, for single tested plural number fMRI data, directly carry out ICA, obtain K spatial activation Brain district compositionK=1 ..., K, K are the number of independent element.RightCarry out phasing, obtain the space without phase ambiguity Activate brain district compositionOwing to there is no pretreatment de-noising,Containing substantial amounts of amplitude noise voxel, (remove width with threshold method Value is less than the voxel of a certain threshold value) cannot eliminate.To this end, the present invention utilizes source phase image to carry out de-noising.Specifically, space Activate brain district compositionFor complex signal, including magnitude image and phase image, note phase image isFor being different from Original plural fMRI phase data (belonging to blended data), claimsFor source phase image, its phase place span be (- π,π].?In, the voxel that in the range of definition [-π/4, π/4], phase value is corresponding is activation voxel, other phase values, bag Include (-π ,-π/4) and (π/4, π] in the range of phase value, corresponding voxel is interference voxel, then be suitable to single tested spatial activation The mask of brain district composition post processing de-noising builds as follows:
BM 1 = 1 , if s ‾ ^ k , phase ( l ) ∈ [ - π / 4 , π / 4 ] 0 , otherwise - - - ( 1 )
L is the pointer of voxel in spatial activation brain district composition.BM1 is acted on noisyObtain removing much noise The spatial activation brain district composition of voxel, only reservation activation voxel:Finally, removeMiddle amplitude is micro- Little such asVoxel, obtain the list tested spatial activation brain district composition of final de-noising.
For how tested plural number fMRI data, carry out single tested ICA respectively, obtain each tested spatial activation brain district and become PointP=1 ..., P, P are tested number.RightCarry out phasing, obtain the spatial activation brain district composition without phase ambiguityFromMiddle acquisition source phase imageAnd build single tested de-noising mask respectively, it is designated as BM1p, by BM1pAct on noisy 'sThe list tested spatial activation brain district composition obtain and remove much noise voxel, only retaining activation voxel Again byThe group mean space obtaining preliminary de-noising activates brain district compositionIn order to eliminate not At the voxel of the tested middle activation of half, at single tested BM1pOn the basis of, build how tested group mean space and activate brain district composition De-noising mask is as follows:
BM 2 = 1 , if Σ p = 1 P BM 1 p ≥ P / 2 0 , otherwise - - - ( 2 )
BM2 is acted onThe group mean space obtaining further de-noising activates brain district composition RemoveMiddle amplitude is small such asVoxel, obtain final de-noising how tested group mean space activate brain district Composition.
Effect and benefit that the present invention is reached are to compare with ICA pretreatment Denoising Method, such as phase masses figure Denoising Method, Before ICA, plural number fMRI data are removed without any voxel, be guaranteed in during ICA analyzes utilizing plural number fMRI data, to obtain More brain function information.Such as, for the plural fMRI data gathered under Motor stimulation, pretreatment Denoising Method only estimates 26 Independent brain function composition, and the present invention can estimate 49 independent brain function compositions.Become for thing followed spatial activation brain district The problem that there is amplitude noise in Fen, the invention provides post processing Denoising Method based on source phase image, according to phase place model Enclose rather than amplitude size is removed noise voxel, retained and activate voxel.Therefore, the present invention can ensure that ICA is to single tested and many Tested plural number fMRI data are analyzed, and then the data solving to cause because of pretreatment de-noising abandon and ask with brain information loss Topic.
Accompanying drawing explanation
Fig. 1 is the tool that the task correlation space activation brain district composition to single tested plural number fMRI data carries out post processing de-noising Body step.
Fig. 2 is that the group average task correlation space to 16 tested plural number fMRI data activates brain district composition and carries out post processing and disappear The concrete steps made an uproar.
Detailed description of the invention
Below in conjunction with technical scheme and accompanying drawing, describe two specific embodiments of the present invention in detail.
Specific embodiment one: assume the list tested plural number fMRI data gathered under existing Motor stimulation, to tested full brain Scanning times is 165.This plural number fMRI data are carried out ICA, obtains the task correlation space containing a large amount of amplitude noises and swash Huo Nao district composition, carries out the concrete steps of post processing de-noising as shown in Figure 1 to it.
The first step, estimates the independent element number of single tested plural number fMRI data according to information theory criterion, and result is 30, adopts By PCA (principle component analysis) method, plural number fMRI data are down to 30 dimensions from 165 dimensions.
Second step, uses plural number EBM (entropy bound minimization) algorithm to the plural number after PCA dimensionality reduction FMRI data carry out ICA, obtain 30 plural ICA and estimate composition, therefrom select task correlation space to activate brain district compositionShould Containing substantial amounts of amplitude noise voxel in composition.
3rd step, selects method for correcting phase pair based on time course compositionCarry out phasing, obtain without phase place Fuzzy spatial activation brain district composition
4th step, calculates source phase imageI.e.Phase image.
5th step, uses formula (1) to build single tested de-noising mask:BM1.
6th step, acts on noisy by BM1Obtain removing much noise voxel, only reservation and activate the task of voxel Correlation space activation brain district composition:
7th step, removesInVoxel, obtain final de-noising task correlation space activate brain District's composition.
Specific embodiment two: the 16 tested plural number fMRI data gathered under existing Motor stimulation, the scanning of each tested full brain Number of times is 165.16 tested plural number fMRI data are carried out ICA respectively, obtains the group average task correlation space of final de-noising Activate the concrete steps of brain district composition as shown in Figure 2.
The first step, uses PCA that from 165 dimensions, 16 tested plural fMRI data are down to 30 dimensions respectively.
Second step, uses plural number EBM algorithm that the tested plural number fMRI data of 16 after PCA dimensionality reduction are carried out ICA respectively, selects Each tested task correlation space activates brain district compositionContaining substantial amounts of amplitude noise voxel in these compositions.
3rd step, selects method for correcting phase based on time course composition the most rightCarry out phasing, Obtain the spatial activation brain district composition without phase ambiguity
4th step, calculates source phase imageI.e.Phase image.
5th step, uses formula (1) to build each tested de-noising mask:BM11,...,BM116
6th step, by BM11,...,BM116It is respectively acting on noisyObtain the relevant sky of task of denoising Between activate brain district composition: s ‾ ^ * , bm 1 1 = s ‾ ^ * 1 · BM 1 1 , . . . , s ‾ ^ * , bm 1 16 = s ‾ ^ * 16 · BM 1 16 .
7th step, the group mean space obtaining preliminary de-noising activates brain district composition
8th step, uses formula (2) to build how tested de-noising mask:BM2.
9th step, acts on BM2The group mean space obtaining further de-noising activates brain district composition s ‾ ^ * , bm 2 group = s ‾ ^ * group · BM 2 .
Tenth step, removesVoxel, obtain the group average task correlation space of final de-noising Activate brain district composition.

Claims (2)

1. plural number fMRI data are carried out a post processing noise-eliminating method for ICA analysis, it is characterized in that, for single tested plural number FMRI data, directly carry out ICA, obtain K spatial activation brain district compositionK=1 ..., K, K are the number of independent element;RightCarry out phasing, obtain the spatial activation brain district composition without phase ambiguityAcquisition source phase imageI.e.'s Phase image, phase place span be (-π, π];?In, the voxel that in the range of definition [-π/4, π/4], phase value is corresponding For activating voxel, other phase values, including (-π ,-π/4) and (π/4, π] in the range of phase value, the voxel of correspondence is interfering bodies Element, the mask that structure is suitable to single tested spatial activation brain district composition post processing de-noising is as follows:
B M 1 = 1 , i f s ‾ ^ k , p h a s e ( l ) ∈ [ - π / 4 , π / 4 ] 0 , o t h e r w i s e
L is the pointer of voxel in spatial activation brain district composition;BM1 is acted on noisyObtain remove much noise voxel, The only spatial activation brain district composition of reservation activation voxel:Finally, removeThe body that middle amplitude is small Element, obtains the list tested spatial activation brain district composition of final de-noising;
For how tested plural number fMRI data, carry out single tested ICA respectively, obtain each tested spatial activation brain district compositionP=1 ..., P, P are tested number;RightCarry out phasing, obtain the spatial activation brain district composition without phase ambiguity FromMiddle acquisition source phase imageAnd build single tested de-noising mask respectively, it is designated as BM1p, by BM1pAct on noisyThe list tested spatial activation brain district composition obtain and remove much noise voxel, only retaining activation voxel Again byThe group mean space obtaining preliminary de-noising activates brain district compositionAt single tested BM1p On the basis of, build how tested group mean space and activate the de-noising mask of brain district composition, eliminate not at the body of the tested middle activation of half Element:
B M 2 = 1 , i f Σ p = 1 P B M 1 p ≥ P / 2 0 , o t h e r w i s e
BM2 is acted onThe group mean space obtaining further de-noising activates brain district composition RemoveThe voxel that middle amplitude is small, the how tested group mean space obtaining final de-noising activates brain district composition.
Post processing noise-eliminating method the most according to claim 1, is characterized in that, the small voxel of amplitude is defined as amplitude and is less than The voxel of 0.5.
CN201410191416.6A 2014-05-07 2014-05-07 A kind of post processing noise-eliminating method that plural number fMRI data are carried out ICA analysis Active CN103985092B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410191416.6A CN103985092B (en) 2014-05-07 2014-05-07 A kind of post processing noise-eliminating method that plural number fMRI data are carried out ICA analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410191416.6A CN103985092B (en) 2014-05-07 2014-05-07 A kind of post processing noise-eliminating method that plural number fMRI data are carried out ICA analysis

Publications (2)

Publication Number Publication Date
CN103985092A CN103985092A (en) 2014-08-13
CN103985092B true CN103985092B (en) 2017-01-04

Family

ID=51277050

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410191416.6A Active CN103985092B (en) 2014-05-07 2014-05-07 A kind of post processing noise-eliminating method that plural number fMRI data are carried out ICA analysis

Country Status (1)

Country Link
CN (1) CN103985092B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105760700B (en) * 2016-03-18 2018-06-08 大连理工大学 A kind of adaptive fixed point IVA algorithms for being suitable for more subject plural number fMRI data analyses
CN105912851B (en) * 2016-04-07 2019-04-16 大连理工大学 A method of utilizing PCA and other than ring type characteristic estimating plural number fMRI data model order
CN106875366B (en) * 2017-03-01 2019-06-21 大连理工大学 Tranquillization state plural number fMRI data are carried out with the phase exact extension detection method of ICA post-processing de-noising
CN108903942B (en) * 2018-07-09 2021-05-11 大连理工大学 Method for identifying spatial difference by using complex fMRI spatial source phase
CN114176518B (en) * 2021-12-06 2023-10-10 大连理工大学 Complex fMRI data space component phase anti-correction method for improving CNN classification performance

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103345749A (en) * 2013-06-27 2013-10-09 中国科学院自动化研究所 Method for detecting brain network function connectivity lateralization based on modality fusion

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103345749A (en) * 2013-06-27 2013-10-09 中国科学院自动化研究所 Method for detecting brain network function connectivity lateralization based on modality fusion

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
SEMI-BLIND KURTOSIS MAXIMIZATION ALGORITHM APPLIED TO COMPLEX-VALUED FMRI DATA;Qiu-Hua Lin;《2011 IEEE International Workshop on Machine Learning for Signal Processing》;20110921;1-6 *
Semiblind Spatial ICA of fMRI Using Spatial Constraints;Qiu-Hua Lin;《Human Brain Mapping》;20101231;1076-1088 *

Also Published As

Publication number Publication date
CN103985092A (en) 2014-08-13

Similar Documents

Publication Publication Date Title
CN103985092B (en) A kind of post processing noise-eliminating method that plural number fMRI data are carried out ICA analysis
CN105913393A (en) Self-adaptive wavelet threshold image de-noising algorithm and device
CN102663695B (en) DR image denoising method based on wavelet transformation and system thereof
CN103399348A (en) Denoising method for seismic signal based on Shearlet transform
CN101661616A (en) Method for enhancing images based on multi-scale edge detection in wavelet reconstruction
CN103116873B (en) Image denoising method
Rajeesh et al. Noise reduction in magnetic resonance images using wave atom shrinkage
CN101482617A (en) Synthetic aperture radar image denoising method based on non-down sampling profile wave
CN101901476A (en) SAR image de-noising method based on NSCT domain edge detection and Bishrink model
CN101980287A (en) Method for detecting image edge by nonsubsampled contourlet transform (NSCT)
CN107590785A (en) A kind of Brillouin spectrum image-recognizing method based on sobel operators
Denis et al. From patches to deep learning: Combining self-similarity and neural networks for SAR image despeckling
CN103854265A (en) Novel multi-focus image fusion technology
CN105069307A (en) Multi-subject fMRI data analysis method combining ICA and shift invariant CPD
Shuaiqi et al. Medical image fusion based on nonsubsampled direction complex wavelet transform
CN102314675B (en) Wavelet high-frequency-based Bayesian denoising method
CN101908206B (en) Morphological component analysis (MCA)-based synthetic aperture radar (SAR) image noise suppression method
CN101930605B (en) Synthetic Aperture Radar (SAR) image target extraction method and system based on two-dimensional mixing transform
Luisier et al. Fast Haar-wavelet denoising of multidimensional fluorescence microscopy data
CN103310423A (en) Mine image intensification method
CN102184530B (en) Image denoising method based on gray relation threshold value
CN105096274B (en) Infrared image noise-reduction method based on non-down sampling contourlet domain mixing statistical model
CN103455986A (en) Random noise point detection algorithm based on fractional-order differential gradient
Kalaiselvi et al. A novel wavelet thresholding technique to denoise magnetic resonance images‖
CN105069762A (en) Image denoising method based on Shearlet transform and non-linear diffusion

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant