CN110473202A - A kind of timing invariant feature extraction method of high-order dynamic function connects network - Google Patents
A kind of timing invariant feature extraction method of high-order dynamic function connects network Download PDFInfo
- Publication number
- CN110473202A CN110473202A CN201910528161.0A CN201910528161A CN110473202A CN 110473202 A CN110473202 A CN 110473202A CN 201910528161 A CN201910528161 A CN 201910528161A CN 110473202 A CN110473202 A CN 110473202A
- Authority
- CN
- China
- Prior art keywords
- correlation
- timing
- sequence
- brain
- order
- 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.)
- Granted
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Abstract
The invention discloses a kind of timing invariant feature extraction method of high-order dynamic function connects network, key steps are as follows: (1) for given sliding window window width and step-length, entire magnetic resonance sequences are divided into multiple subsequences；(2) correlation between each brain area in each subsequence is calculated, obtains dynamic function connection network, and then calculate the Central Moment Feature of the dynamic function catenation sequence of any two brain area；(3) for each subsequence of dynamic function connection network, the catenation sequence of each brain area and other brain areas is considered as an one-dimensional random sequence, the correlation of any pair of brain area is calculated, and then constructs a high-order dynamic function connects network and obtains its Central Moment Feature.The present invention connects the timing invariant features of network and high-order dynamic function connects network using Central Moment Feature as dynamic function, can capture the profound incidence relation of brain function connection, provide invariant feature for subsequent Medical Image Processing.
Description
◆ technical field
The present invention relates to the technical fields of neuroimaging and machine learning, and in particular to a kind of timing of high-order dynamic functional network
Invariant feature extraction method.
◆ background technique
Tranquillization state functional MRI (rs-FMRI, resting-state functional magnetic resonance
It imaging is changed based on blood oxygen caused by brain neurological motion and is produced in the case where subject avoids the systematic appraisal environment of an activation
Raw magnetic vibration shadowgraph technique, is a kind of highly effective neuroimaging mode that the mankind find out brain activity rule.Due to rs-
FMRI has non-invasive and radiation damage free and higher spatial and temporal resolution, at present in cognitive science, Neuscience, medicine
The fields such as of science and mental disease using very extensive.
For the information effectively extracted and included using rs-FMRI, need using certain strategy and method to rs-
FMRI data are handled, the important feature contained to capture rs-FMRI.Dynamic function wherein based on sliding window strategy
Connection network (D-FCN, dynamic functional connectivity network) is the capture subtle dynamic of rs-FMRI
The important way of variation characteristic.For an individual, enableIndicate theiA brain
The average rs-FMRI time series of functional areas, whereinMIndicate the number of samples in entire sweep time section,NIndicate cerebral cortex
Functional areas number.Then the basic step of D-FNC building can be stated are as follows:
1) rs-FMRI in entire sweep time section is divided into multiple rs-FMRI subsequences.
Using sliding window setting technique, by rs-FMRI time seriesIt is divided intoKA subsequence, it may be assumed that
(1)
WhereinIndicate the subsequence number of rs-FMRI,TIndicate sliding window width,SIndicate sliding step
It is long.
2) D-FCN is constructed.
ToThe rs-FMRI sequence of a child window, correlation between the sequence of calculation.I.e.
(2)
ObviouslyIt is a correlation matrix, describes any pair of brain areaIn a short time
Interior correlation.Based on formula (1), we are availableA correlation sequence, i.e.,
(3)
Wherein,Describe any pair of variation that brain area occurs with sweep time.
Show currently, having numerous evidences: based on D-FCN constructed by sliding window strategy, can preferably react brain area it
Between the dynamic change that connects, and this variation can reveal that the difference between Different Individual between brain tissue and brain function
Property, therefore there are huge application prospects in fields such as cognitive science, Neuscience, pharmacology and mental diseases.
However, there are following several by D-FCN although D-FCN, which understands brain activity for us, provides a new channel
A obvious disadvantage: (1) D-FCN is especially sensitive to time sequencing.I.e. under tranquillization state environment, Different Individual is in entire magnetic resonance
The brain neurological motion rule in sweep time section, does not have temporal consistency.That is, Different Individual is same
The correlation between brain area in the magnetic resonance imaging period is not consistent.Therefore, for Different Individual, along sweep time
Dynamic function connection subsequence (i.e. D-FCN) there are temporal wrong matchings.(2) D-FCN only reflects any two brain
Dynamic link library characteristic between area, can not reflect the incidence relation between multiple brain areas.So which kind of method to obtain D- using
Incidence relation between FCN timing invariant features and multiple brain areas is still to have problem to be solved now.
◆ summary of the invention
To solve the above-mentioned problems, the present invention provides a kind of extraction side for obtaining D-FCN and high-order D-FCN timing invariant features
Method.This feature reflects the order of information of function connects, the function connects relationship between multiple brain areas can be captured, after can be used for
Continuous brain injury detection, Observation on rehabilitation effect etc..
◆ the specific steps of the present invention are as follows:
Step 1) Initial parameter sets.Setting sliding window widthT, sliding stepS。
Step 2) acquisition of functional MRI subsequence.It will be whole along sweep time axis using given sliding window and step-length
A magnetic resonance sequences are divided into multiple subsequences.
Step 3) obtain D-FCN.Correlation between each brain area is calculated in each child window, and then obtains D-FCN.
Step 4) obtain D-FCN Central Moment Feature.Calculate the center of the dynamic function catenation sequence of any two brain area
Moment characteristics.Since the central moment of a random sequence has timing invariance, the Central Moment Feature of D-FCN can be used as D-
The timing invariant features of FCN.
Step 5) obtain high-order dynamic function connects network (Ho-D-FCN, high-order dynamic
Functional connectivity network).I.e. for each subsequence of D-FCN, by each brain area and other brains
The catenation sequence in area is considered as an one-dimensional random sequence, calculates the correlation of any pair of brain area, i.e., " correlation of correlation ",
It reflects the incidence relation between multiple brain areas, and then constructs a high-order dynamic function connects network, i.e. Ho-D-FCN.
Step 6) obtain Ho-D-FCN Central Moment Feature.Similar step 4, the High order central moment for obtaining Ho-D-FCN are special
Sign, the high-order timing invariant features as D-FCN.
◆ beneficial effects of the present invention:
1) Central Moment Feature of D-FCN has timing invariance, can effectively overcome the timing sensitive question of D-FCN.
2) while the Central Moment Feature of Ho-D-FCN keeps timing invariance, the association for being able to reflect multiple brain areas is closed
System explores brain neurological motion rule for depth and provides richer information.
◆ Detailed description of the invention
Fig. 1 is the flow diagram of invention
◆ specific embodiment
With reference to the accompanying drawing (Fig. 1) the invention will be further described with embodiment.As shown in Figure 1, comprising the following steps:
Step 1) Initial parameter sets.Set sliding window window widthT, sliding stepS。
Step 2) acquisition of functional MRI subsequence.It enablesIndicate some
The of individualiThe average rs-FMRI time series of a brain domain, whereinMIndicate the number of samples in entire sweep time section,NIndicate corticocerebral functional areas number.Using sliding window setting technique, by the rs-FMRI time series in entire sweep time sectionIt is divided intoKA subsequence, i.e.,
(1)
WhereinIndicate the subsequence number of rs-FMRI.
Step 3) D-FCN(dynamic function connect network) building.ToThe rs- of a child window
FMRI sequence, using correlation between the Pearson correlation coefficient sequence of calculation.I.e.
(2)
ObviouslyIt is a correlation matrix, describes any pair of brain areaIn a short time
Interior correlation.Based on formula (1), we are availableA correlation sequence, i.e.,
(3)
Wherein,Describe the variation that correlation occurs with sweep time between any pair of brain area.
Step 4) obtain D-FCN(dynamic function connect network) Central Moment Feature.By formula (3), we are available
One function connects sequence, i.e.,
= [] (), (4)
WhereinReflectiA brain area and thej Dynamic link library relationship between a brain area along the time axis.In fact, not
Dynamic link library relationship with individual between corresponding brain areaThere is no the synchronization of time consistency, i.e. Different Individual is corresponding
The Dynamic link library relationship of brain area there are great differences property.For this purpose, in order to obtain the timing invariant features of Dynamic link library sequence, we
The Central Moment Feature of each dynamic sequence is sought, i.e.,
(), (8)
WhereinIndicate central moment order,It indicates= [] mean value.Obviously,
Since there is the central moment of one-dimensional random signal sequence timing not cease to be faithful, so() can
Using as dynamic function catenation sequence= [] () timing invariant features,
For subsequent image classification, auxiliary diagnosis etc..
Step 5) Ho-D-FCN(high-order dynamic function connects network) building.It considers(see formula (4)) are only anti-
The Dynamic link library relationship between two brain areas has been reflected, the Dynamic link library relationship between multiple brain areas can not be captured.In order to obtain
Dynamic link library relationship between multiple brain areas, we define a higher order dynamic network Ho-D-FCN as follows.
,), (9)
Wherein, it illustratesiA brain area iskIn a period with other brains
Correlation between area, so,IllustrateiA brain area and thejIt is more between a brain area " correlation of correlation "
A kind of embodiment of correlation between a brain area, we term it " higher order dependencies ".Based on formula (9), we are available
A correlation sequence, i.e.,
(10)
Obviously,Describe the variation that correlation occurs with sweep time between multiple brain areas.
Step 6) obtain Ho-D-FCN Central Moment Feature.According to formula (9), our available high-order function
Catenation sequence, i.e.,
= [] (), (11)
WhereinReflectiThe correlation of a brain area (with other brain areas) and thej A brain area (to other brain areas) it is related
Dynamic link library relationship between property along the time axis, the i.e. Relationship Between Dynamic Change of " correlation of correlation ".Similar to step 4, it is
The timing invariant features of this Dynamic link library sequence of acquisition, we seek the Central Moment Feature of each dynamic sequence, i.e.,
(), (12)
WhereinIndicate central moment order,It indicates= [] mean value.It is aobvious
So, since there is the central moment of one-dimensional random signal sequence timing not cease to be faithful, so()
It can be used as high-order dynamic function connects sequence= [] timing invariant features, use
In subsequent image classification, auxiliary diagnosis etc..
By formula (8) and formula (12), we obtain D-FCN(dynamic function connection network respectively) and Ho-D-FCN
The timing invariant features of (high-order dynamic function connects network)() and(), and it is used for subsequent brain image processing, obtain preferable result.
Claims (5)
1. a kind of timing invariant feature extraction method of high-order dynamic function connects network, characterized in that mainly comprising following several
A step: step 1, Initial parameter sets;Set sliding window window width size, sliding step length;Step 2, function magnetic is obtained
The subsequence that resonates is enabledIndicate the of some individualiThe average rs- of a brain domain
FMRI time series, wherein M indicates the number of samples in entire sweep time section, and N indicates corticocerebral functional areas number;
Using sliding window setting technique, by the rs-FMRI time series in entire sweep time sectionIt is divided into K subsequence, i.e.,
(1)
WhereinIndicate the subsequence number of rs-FMRI, T indicates that sliding window width, S indicate sliding step;
Step 3.D-FCN(dynamic function connect network) building;Step 4. obtain D-FCN(dynamic function connect network) central moment
Feature;Step 5.Ho-D-FCN(high-order dynamic function connects network) building;Step 6. obtains Ho-D-FCN(high-order dynamic function
Network can be connected) Central Moment Feature.
2. a kind of timing invariant feature extraction method of high-order dynamic function connects network as described in claim 1, feature
It is that the step 3 includes: correlation sequenceDefinition: i.e.
(3)
Here,Describe the variation that correlation occurs with sweep time between any pair of brain area；WhereinIt is a correlation matrix, describes any pair of brain areaCorrelation within a short time;
Its is defined as: is toThe rs-FMRI sequence of a child window, using the Pearson correlation coefficient sequence of calculation it
Between correlation, i.e.,
(2)。
3. a kind of timing invariant feature extraction method of high-order dynamic function connects network as described in claim 1, feature
It is that the step 4 includes:
To any pair of brain areaDefine a function connects sequence:
= [] (), (4)
WhereinReflect Dynamic link library relationship along the time axis between i-th brain area and jth brain area；In fact, not
Dynamic link library relationship with individual between corresponding brain areaThere is no temporal consistency, i.e. the synchronization phase of Different Individual
Answer Dynamic link library relationship there are great differences the property of brain area；For this purpose, in order to obtain the timing invariant features of Dynamic link library sequence, I
Seek the Central Moment Feature of each dynamic sequence, it may be assumed that
(), (8)
WhereinIndicate central moment order,It indicates= [] mean value；Clearly as
There is the central moment of one-dimensional random signal sequence timing not cease to be faithful, so() can be used as
Dynamic function catenation sequence= [] () timing invariant features, be used for subsequent shadow
As classification, auxiliary diagnosis etc..
4. a kind of timing invariant feature extraction method of high-order dynamic function connects network as described in claim 1, feature
It is that the step 5 includes:
Define a higher order dependencies sequence:
(10)
Here,Describe the variation that correlation occurs with sweep time between multiple brain areas；WhereinDetermine
Justice is as follows:
,) (9)
Wherein, it illustrate i-th of brain area within k-th of period with other brain areas it
Between correlation, so,It illustrates between i-th of brain area and j-th of brain area " correlation of correlation ", is multiple brains
A kind of embodiment of correlation between area, we term it " higher order dependencies ".
5. a kind of timing invariant feature extraction method of high-order dynamic function connects network as described in claim 1, feature
It is that the step 6 includes:
To any pair of brain areaDefine a high-order function connects sequence:
= [] (), (11)
WhereinThe correlation for reflecting i-th brain area (to other brain areas) is related with jth brain area (with other brain areas)
Dynamic link library relationship between property along the time axis, the i.e. Relationship Between Dynamic Change of " correlation of correlation "；It is this in order to obtain
The timing invariant features of Dynamic link library sequence, we seek the Central Moment Feature of each dynamic sequence, i.e.,
(), (12)
WhereinIndicate central moment order,It indicates= [] mean value；Clearly as
There is the central moment of one-dimensional random signal sequence timing not cease to be faithful, so() can be used as
High-order dynamic function connects sequence= [] timing invariant features, be used for subsequent image
Classification, auxiliary diagnosis etc..
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910528161.0A CN110473202B (en) | 2019-06-18 | 2019-06-18 | Time sequence invariant feature extraction method for high-order dynamic function connection network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910528161.0A CN110473202B (en) | 2019-06-18 | 2019-06-18 | Time sequence invariant feature extraction method for high-order dynamic function connection network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110473202A true CN110473202A (en) | 2019-11-19 |
| CN110473202B CN110473202B (en) | 2023-03-31 |
Family
ID=68506926
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910528161.0A Active CN110473202B (en) | 2019-06-18 | 2019-06-18 | Time sequence invariant feature extraction method for high-order dynamic function connection network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110473202B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111325288A (en) * | 2020-03-17 | 2020-06-23 | 山东工商学院 | Clustering idea-based multi-view dynamic brain network characteristic dimension reduction method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100220906A1 (en) * | 2007-05-29 | 2010-09-02 | Michael Abramoff | Methods and systems for determining optimal features for classifying patterns or objects in images |
| CN104700120A (en) * | 2015-03-23 | 2015-06-10 | 南京工业大学 | Method for extracting and classifying fMRI features based on adaptive entropy algorithm for projection clustering (APEC) |
| US9633430B2 (en) * | 2011-12-28 | 2017-04-25 | Institute Of Automation, Chinese Academy Of Sciences | Method for analyzing functional MRI brain images |
| CN106618575A (en) * | 2017-01-09 | 2017-05-10 | 天津大学 | Method for collecting cerebral function connection spontaneous fluctuation variability |
| US20170340261A1 (en) * | 2014-12-08 | 2017-11-30 | Rutgers, The State University Of New Jersey | System and method for measuring physiologically relevant motion |
| US10034645B1 (en) * | 2017-04-13 | 2018-07-31 | The Board Of Trustees Of The Leland Stanford Junior University | Systems and methods for detecting complex networks in MRI image data |
| US20180256912A1 (en) * | 2017-03-07 | 2018-09-13 | The Regents Of The University Of California | Method for monitoring treatment of neuropsychiatric disorders |
| US20190114773A1 (en) * | 2017-10-13 | 2019-04-18 | Beijing Curacloud Technology Co., Ltd. | Systems and methods for cross-modality image segmentation |
-
2019
- 2019-06-18 CN CN201910528161.0A patent/CN110473202B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100220906A1 (en) * | 2007-05-29 | 2010-09-02 | Michael Abramoff | Methods and systems for determining optimal features for classifying patterns or objects in images |
| US9633430B2 (en) * | 2011-12-28 | 2017-04-25 | Institute Of Automation, Chinese Academy Of Sciences | Method for analyzing functional MRI brain images |
| US20170340261A1 (en) * | 2014-12-08 | 2017-11-30 | Rutgers, The State University Of New Jersey | System and method for measuring physiologically relevant motion |
| CN104700120A (en) * | 2015-03-23 | 2015-06-10 | 南京工业大学 | Method for extracting and classifying fMRI features based on adaptive entropy algorithm for projection clustering (APEC) |
| CN106618575A (en) * | 2017-01-09 | 2017-05-10 | 天津大学 | Method for collecting cerebral function connection spontaneous fluctuation variability |
| US20180256912A1 (en) * | 2017-03-07 | 2018-09-13 | The Regents Of The University Of California | Method for monitoring treatment of neuropsychiatric disorders |
| US10034645B1 (en) * | 2017-04-13 | 2018-07-31 | The Board Of Trustees Of The Leland Stanford Junior University | Systems and methods for detecting complex networks in MRI image data |
| US20190114773A1 (en) * | 2017-10-13 | 2019-04-18 | Beijing Curacloud Technology Co., Ltd. | Systems and methods for cross-modality image segmentation |
Non-Patent Citations (4)
| Title |
|---|
| ZHAO F 等: "Diagnosis of Autism Spectrum Disorder Using Central-Moment", 《FRONTIERS IN NEUROENCE》 * |
| 张泽中等: "基于深度学习的胃癌病理图像分类方法", 《计算机科学》 * |
| 袁悦铭等: "基于静息态功能磁共振成像的动态功能连接分析及临床应用研究进展", 《磁共振成像》 * |
| 龙雨涵等: "非线性脑区相关性分析及动态脑网络构建方法", 《信号处理》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111325288A (en) * | 2020-03-17 | 2020-06-23 | 山东工商学院 | Clustering idea-based multi-view dynamic brain network characteristic dimension reduction method |
| CN111325288B (en) * | 2020-03-17 | 2022-02-25 | 山东工商学院 | Clustering idea-based multi-view dynamic brain network characteristic dimension reduction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110473202B (en) | 2023-03-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Aquino et al. | Identifying and removing widespread signal deflections from fMRI data: Rethinking the global signal regression problem | |
| Haak et al. | Connectopic mapping with resting-state fMRI | |
| Lange et al. | Non-linear Fourier time series analysis for human brain mapping by functional magnetic resonance imaging | |
| Ding et al. | Classification and quantification of neuronal fiber pathways using diffusion tensor MRI | |
| McIntosh et al. | Partial least squares analysis of neuroimaging data: applications and advances | |
| Bell et al. | Object representations in the temporal cortex of monkeys and humans as revealed by functional magnetic resonance imaging | |
| LaConte et al. | The evaluation of preprocessing choices in single-subject BOLD fMRI using NPAIRS performance metrics | |
| Li et al. | Surface-based single-subject morphological brain networks: effects of morphological index, brain parcellation and similarity measure, sample size-varying stability and test-retest reliability | |
| CN110232332A (en) | Extraction and brain state classification method and system for dynamic function connection local linear embedded features | |
| US9633430B2 (en) | Method for analyzing functional MRI brain images | |
| Iyer et al. | Inferring functional connectivity in MRI using Bayesian network structure learning with a modified PC algorithm | |
| Kundu et al. | Estimating dynamic brain functional networks using multi-subject fMRI data | |
| Caruyer et al. | On facilitating the use of HARDI in population studies by creating rotation-invariant markers | |
| Chatzichristos et al. | Coupled tensor decompositions for data fusion | |
| CN110473202A (en) | A kind of timing invariant feature extraction method of high-order dynamic function connects network | |
| Li et al. | Diffusion tensor-based fast marching for modeling human brain connectivity network | |
| Huang et al. | Learning shared neural manifolds from multi-subject FMRI data | |
| CN111369637B (en) | DWI fiber optimization reconstruction method and system for fusing white matter functional signals | |
| CN113786185A (en) | Static brain network feature extraction method and system based on convolutional neural network | |
| Liang et al. | Shape modeling and clustering of white matter fiber tracts using fourier descriptors | |
| Achard et al. | Inter-regional correlation estimators for functional magnetic resonance imaging | |
| Frank et al. | JEDI: Joint Estimation Diffusion Imaging of macroscopic and microscopic tissue properties | |
| Demir et al. | Online agglomerative hierarchical clustering of neural fiber tracts | |
| Parmar et al. | Visualizing temporal brain-state changes for fMRI using t-distributed stochastic neighbor embedding | |
| Goebel | Revealing brain activity and white matter structure using functional and diffusion-weighted magnetic resonance imaging |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |