CN101292871A - Method for specification extraction of magnetic resonance imaging brain active region based on pattern recognition - Google Patents
Method for specification extraction of magnetic resonance imaging brain active region based on pattern recognition Download PDFInfo
- Publication number
- CN101292871A CN101292871A CNA2007100986913A CN200710098691A CN101292871A CN 101292871 A CN101292871 A CN 101292871A CN A2007100986913 A CNA2007100986913 A CN A2007100986913A CN 200710098691 A CN200710098691 A CN 200710098691A CN 101292871 A CN101292871 A CN 101292871A
- Authority
- CN
- China
- Prior art keywords
- msub
- mrow
- region
- brain
- local
- 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
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000002595 magnetic resonance imaging Methods 0.000 title claims abstract description 7
- 210000004556 brain Anatomy 0.000 title claims description 96
- 238000003909 pattern recognition Methods 0.000 title claims description 3
- 238000000605 extraction Methods 0.000 title description 6
- 230000000638 stimulation Effects 0.000 claims abstract description 53
- 238000002599 functional magnetic resonance imaging Methods 0.000 claims abstract description 40
- 230000000694 effects Effects 0.000 claims abstract description 39
- 238000004458 analytical method Methods 0.000 claims description 88
- 230000004913 activation Effects 0.000 claims description 78
- 238000012360 testing method Methods 0.000 claims description 44
- 238000004422 calculation algorithm Methods 0.000 claims description 34
- 230000007177 brain activity Effects 0.000 claims description 29
- 230000006870 function Effects 0.000 claims description 27
- 238000012937 correction Methods 0.000 claims description 11
- 238000012567 pattern recognition method Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000007689 inspection Methods 0.000 claims description 6
- 238000012952 Resampling Methods 0.000 claims description 4
- 230000008707 rearrangement Effects 0.000 claims description 4
- 238000007635 classification algorithm Methods 0.000 claims description 3
- 238000010219 correlation analysis Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000013401 experimental design Methods 0.000 claims description 2
- 230000002490 cerebral effect Effects 0.000 abstract description 14
- 238000005516 engineering process Methods 0.000 abstract description 7
- 230000033001 locomotion Effects 0.000 abstract 3
- 238000001514 detection method Methods 0.000 description 22
- 238000009499 grossing Methods 0.000 description 18
- 238000004088 simulation Methods 0.000 description 17
- 238000007405 data analysis Methods 0.000 description 9
- 230000003925 brain function Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000000004 hemodynamic effect Effects 0.000 description 4
- 230000035945 sensitivity Effects 0.000 description 4
- 230000001149 cognitive effect Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000010354 integration Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- YBJHBAHKTGYVGT-ZKWXMUAHSA-N (+)-Biotin Chemical group N1C(=O)N[C@@H]2[C@H](CCCCC(=O)O)SC[C@@H]21 YBJHBAHKTGYVGT-ZKWXMUAHSA-N 0.000 description 2
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 2
- 239000008280 blood Substances 0.000 description 2
- 210000004369 blood Anatomy 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 229910052760 oxygen Inorganic materials 0.000 description 2
- 239000001301 oxygen Substances 0.000 description 2
- 101100009524 Anas platyrhynchos HSD17B12 gene Proteins 0.000 description 1
- 238000012935 Averaging Methods 0.000 description 1
- 235000002566 Capsicum Nutrition 0.000 description 1
- 239000006002 Pepper Substances 0.000 description 1
- 235000016761 Piper aduncum Nutrition 0.000 description 1
- 235000017804 Piper guineense Nutrition 0.000 description 1
- 244000203593 Piper nigrum Species 0.000 description 1
- 235000008184 Piper nigrum Nutrition 0.000 description 1
- 101100256975 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) SIP2 gene Proteins 0.000 description 1
- 101100203850 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) SPT2 gene Proteins 0.000 description 1
- 101100291461 Schizosaccharomyces pombe (strain 972 / ATCC 24843) spm2 gene Proteins 0.000 description 1
- 230000001746 atrial effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229960002685 biotin Drugs 0.000 description 1
- 235000020958 biotin Nutrition 0.000 description 1
- 239000011616 biotin Substances 0.000 description 1
- 230000003727 cerebral blood flow Effects 0.000 description 1
- 230000036992 cognitive tasks Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 230000009916 joint effect Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000002610 neuroimaging Methods 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000007473 univariate analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Landscapes
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The present invention discloses an arithmetic of picking up magnetic resonance imaging cerebral active regions by sorting based on mode identification, which comprises the steps that cerebral active regions are extracted based on the multi-element mode distance between fine activity modes in partial cerebral regions for the pretreatment of an fMRI image; a partial consistent cerebral region is obtained by clustering; the combined activities of a plurality of tissues inside the partial consistent cerebral region are used for constructing the multi-element mode; the multi-element distance function is constructed by a mode sorting method to measure the separable characters of the partial cerebral region motion under different stimulation conditions, so as to judge whether the cerebral region is activated or not. The present invention indicates the cerebral motions under different stimulation conditions by multi-element mode information formed by multiple tissues inside the partial cerebral region directly, the multi-element mode can reflect the partial cerebral motion state over all, and multi-element statistical distance can be effectively integrated with the information in the partial cerebral region to measure the difference between different cerebral activate states, so the multi-element mode and the multi-element statistical distance ensure that the arithmetic of the present invention can detect the fine cerebral action mode more completely than the traditional fMRI analyzing technology.
Description
Technical Field
The invention belongs to the technical field of neuroimaging data analysis, particularly relates to an extraction algorithm of a magnetic resonance imaging fMRI activation region, and particularly relates to an fMRI brain activation region by using a multivariate pattern recognition method.
Background
With functional Magnetic Resonance Imaging (fMRI), it is widely used in human brain function research due to its high spatial and temporal resolution, non-invasive, and other features. fMRI generally refers to Blood Oxygen Level Dependent (BOLD) based fMRI imaging that reflects brain activity by measuring changes in magnetic resonance signals caused by changes in cerebral blood flow and oxygen content of the cerebral blood due to neural activity. With the rapid growth of experimental data in recent years, reasonably effective fMRI data analysis techniques have gained increasing importance. The core problem in functional data analysis is to look for brain regions whose activity can significantly distinguish between different experimental conditions (e.g., stimulation conditions and baseline conditions) based on measured fMRI data. Reasonable fMRI analysis methods need to take into account both the general organization principle of human brain function and some characteristics of the fMRI data itself.
Brain function is generally believed to follow two basic organizational principles: function integration and function specialization. On a large spatial scale (scale), a complex brain function may be performed by many function-specific brain regions through interaction (integration); conversely, a specific brain region may represent or process many different cognitive tasks, with different distributed brain activities on a fine spatial scale representing different external stimuli. On the other hand, with the advancement of MRI technology, the spatial resolution of fMRI images is constantly increasing. Conventional fMRI data can achieve a high signal-to-noise ratio at voxel widths of about 4 mm in each dimension. Currently, voxels with a width of 2 mm in each dimension are already reliably available on standard 3T clinical Magnetic Resonance (MRI) machines. With the use of ultra-high field (≧ 4T) magnetic resonance techniques, the spatial resolution of fMRI images is approaching on a sub-millimeter scale. The improvement of the spatial resolution of fMRI images provides the possibility for us to study brain function on a finer spatial scale: however, most conventional fMRI data analysis is based on a Generalized Linear Model (GLM), which is essentially a univariate statistical technique that determines whether each voxel is activated by analyzing the time series of the voxel in isolation, completely ignoring the interrelations between different voxels (especially the pattern information implicit in the local brain region). Thus, conventional unary analysis techniques cannot detect patterns of brain activity on a fine scale caused by experimental stimuli, and voxels constituting these patterns that are weakly responsive to the stimuli. In addition, to improve signal-to-noise ratio and statistical effectiveness, unary techniques typically employ gaussian kernels to spatially smooth fMRI data. Data smoothing essentially corresponds to low-pass filtering, which blurs the fine brain activity signals that contain information relevant to neuroscience. Thus, a large amount of fine information will be filtered out and the high spatial resolution information provided by fMRI is still far from being utilized.
Disclosure of Invention
In order to be able to distinguish brain activity patterns caused by different stimuli on a finer scale and to make full use of the information contained in fMRI data, the invention describes a new class of algorithms (LMDM) for detecting brain activation regions based on pattern recognition methods. The invention directly uses the multi-mode information composed of multiple voxels in the local brain area to represent the brain activity under different stimulation conditions, and further measures the separability of the local brain area activity under different stimulation conditions by taking the multi-element distance between different modes as statistic. The multivariate mode can comprehensively reflect the local brain activity state, and the multivariate statistical distance as the statistic can more effectively integrate the multi-voxel information in the local brain area to distinguish different brain activation states. Therefore, compared with a unitary analysis technology, the method can more accurately extract the brain activation region and judge the brain activity state under different cognitive conditions.
The GLM method assumes that the individual voxels are independent of each other, so that the mode information contained in the local brain region is completely ignored when performing the localization of the brain activation region. From a cognitive neuroscience perspective, it ignores the possibility that there is no one-to-one correspondence between specific functions of the human brain and individual voxels in fMRI data. The invention aims to provide a brain activation region detection algorithm based on a pattern recognition method, which can more accurately extract a brain activation region and judge the brain activity state under different cognitive conditions, in order to solve the defects in the existing fMRI data analysis technology, distinguish brain activity patterns caused by different stimuli on finer scales and fully utilize information contained in fMRI data.
In order to realize the purpose of the invention, the invention provides a brain activation region detection algorithm based on a pattern recognition method, which comprises the following steps:
a pretreatment step A: preprocessing the acquired magnetic resonance image, removing confusion factors and obtaining standardized image data;
and B, constructing a local consistent brain region: for each voxel v of the image data0Using region growing algorithm with v0As seed points, local consensus brain region N (v) containing K-idiosin was obtained0)；
A time sequence dividing step C: dividing local consistent brain region N (v) according to stimulation condition category corresponding to experimental design0) A time sequence of a plurality of voxels in the time sequence, each type of stimulation condition corresponds to a group of different multi-metadata samples;
constructing a multivariate mode statistical distance step D: under different stimulation conditions, the local consistent brain region N (v)0) The spatial mode formed by the K voxels in the array corresponds to different K-dimensional random distribution; constructing a multivariate statistical distance function using pattern recognition methods to measure local consensus brain region N (v)0) Statistical distances between data samples corresponding to different conditions; calculating a multivariate statistical distance by iterating the above process for each voxel to obtain a multivariate statistical distance parameter map;
a multivariate distance statistical parameter graph hypothesis testing step E: suppose that in each locally consistent brain region N (v)0) The activity pattern is the same under different stimulation conditions, using non-parametersThe sequence test is used for testing the statistical distance parameter corresponding to each voxel to obtain a significance nonparametric sequence test value image;
hypothesis testing multiple comparison correction step F: the method of false discovery rate FDR is used for eliminating multiple comparisons to obtain local brain regions, namely corresponding activation regions, which can obviously distinguish different test conditions.
According to an embodiment of the invention, the preprocessing step of the normalized image data further comprises:
step A.1: standardizing different samples of image data to enable the mean value to be 0 and the standard deviation to be 1;
step A.2: different features of the image data were normalized to have a mean of 0 and a standard deviation of 1.
According to an embodiment of the present invention, the step of constructing a local consistent brain region further comprises:
step B.1: measuring the activity similarity property among different voxels by using the Pearson correlation coefficient of the voxel activity time sequence as a criterion;
step B.2: using voxels v0As the seed points, the region growing algorithm selects the domain voxels with the most similar activity to the seed points each time as the local consistent brain region; the region growing is carried out iteratively, and the operation is stopped after the specified size of the local region is reached; so that the resulting voxel activity within the local region is all substantially uniform.
According to an embodiment of the present invention, the step of constructing a multivariate pattern statistical distance further includes:
step D.1: voxel v0Local consensus brain region N (v)0) Under different stimulation conditions X and Y, the measured combined activity of K voxels in the local area is obtained from two different multivariate random variables X ═ X (X1，X2，…，Xi，…，XK)T，Y＝(Y1，Y2，…，Yi，…，YK)TI-1, 2, …, K samples the resulting pattern samples;
step D.2: respectively corresponding to the sample set S by the stimulation conditions X and YXAnd SY(ii) a Constructing different multivariate distance functions to measure the sample set S according to the pattern recognition methodXAnd SYThe multivariate distance is a Fisher distance 0 distance function derived according to Fisher linear discriminant analysis, a maximum boundary distance function derived according to a support vector machine, and a multivariate distance function derived by other mode classification algorithms;
step D.3: here, FLDF is taken as an example and set S according to Fisher' S Linear discriminant analysisXAnd SYThe FLDF between is calculated by:
wherein, 
are respectively a sample SXAnd SYThe average vector of (d) is:
Σ is the sample mixture covariance:
According to the embodiment of the invention, the multivariate distance statistical parameter graph hypothesis test step comprises the following steps:
in each local brain region N (v)0) The activity pattern is the same under different stimulation conditions, namely: under the null assumption that no significant separation is possible, use is made ofThe parameter sequence test is used for testing the statistical distance parameter corresponding to each voxel to obtain a significance nonparametric sequence test value image;
non-parametric sequence verification is performed by magnetic resonance image data resampling as follows: the data space mode is unchanged, and the time sequence corresponding to the magnetic resonance image data is randomly rearranged; the random rearrangement destroys the correlation between the magnetic resonance image signals and the test conditions, and reserves the complete space structure, thereby calculating the non-parameter sequence test values corresponding to the multivariate distance statistics among different local brain activity modes.
According to an embodiment of the present invention, the hypothesis testing multiple comparison correction step further comprises:
step F.1: selecting a boundary value q (0 < q < 1) of the false discovery rate FDR as the expected maximum false discovery rate FDR;
step F.2: sorting the values of the non-parameter sequence inspection p-value graph obtained by the non-parameter arrangement inspection from small to large: p (1) ≦ p (2) ≦ … ≦ p (m), where p (i) is the corresponding voxel statistic v (i), and m is the number of voxels common to the fMRI data examined;
step F.3: let r satisfy the inequality
<math>
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>≤</mo>
<mfrac>
<mi>i</mi>
<mi>V</mi>
</mfrac>
<mfrac>
<mi>q</mi>
<mrow>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
</math>
Where c (m) is a predetermined constant, and its choice is related to the distribution of voxels, and there are 2 choices under different conditions: c (m) is equal to 1,
<math>
<mrow>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mi>Σ</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>V</mi>
</munderover>
<mfrac>
<mn>1</mn>
<mi>i</mi>
</mfrac>
<mo>;</mo>
</mrow>
</math>
step F.4: the results were obtained: rejecting null hypothesis condition, i.e. true activated voxels are v (1), v (2), …, v (r), i.e. voxels whose statistic is greater than v (r) are activated voxels, called v (r) as threshold value determined for multiple tests;
step F.5: and (3) carrying out threshold segmentation on the statistical parameter graph according to a multi-test threshold v (r) obtained by FDR correction to obtain brain activation regions which can obviously distinguish different test conditions.
The invention has the beneficial effects that:
fMRI data contains rich brain activity information, both at the large spatial scale level and at the fine scale level. However, until recently, there has been relatively little research into how to extract directly, using the fine information hidden in fMRI data. The traditional unary analysis technology only depends on the time series information of a single voxel to extract a brain activation region, and completely ignores the spatial mode information contained in a local brain region. In addition, spatial smoothing pre-processing, which is typically employed by unary analysis techniques, can also unreasonably filter out much of the high frequency information that is implicit on the fine scale in the data. The method firstly uses a space mode vector formed by a plurality of voxels to express the brain activity state in a local brain area, further uses multivariate distance statistics to integrate local brain area information, and judges the brain activity difference under different stimulation conditions. The multivariate mode can comprehensively reflect the activity state of the local brain, and the multivariate statistical distance can effectively integrate information in the local brain area to measure the difference between different brain activation states, so that the multivariate statistical distance and the information in the local brain area can ensure that the LMDM method can more completely detect the fine brain activation mode than the traditional univariate analysis technology. Both simulation test and real data test results prove that the algorithm can completely detect the brain activation region compared with the traditional unitary analysis method, and a new way is provided for brain function data analysis.
Drawings
FIG. 1 is a schematic diagram of the algorithm calculation flow of the present invention.
Fig. 2a and 2b are schematic diagrams of extracting local consistent brain regions in the present invention.
FIG. 3 is a diagram illustrating a concept of pattern classification for measuring statistical distances between different activation patterns according to the present invention.
Fig. 4 is a comparison of the inventive algorithm and a conventional active area detection algorithm in an example.
Fig. 5 is a comparison of the inventive algorithm and a conventional active area detection algorithm in an example.
Fig. 6 is a comparison of the inventive algorithm and a conventional active area detection algorithm in an example.
Fig. 7 is a comparison of the inventive algorithm and a conventional active area detection algorithm in an example.
Fig. 8 is a comparison of the inventive algorithm and a conventional active area detection algorithm in an example.
Detailed Description
The invention will be further described with reference to fig. 1 and the examples, which are intended to illustrate the invention without limiting it.
The following is a detailed description of the steps involved in the process of the invention:
first, the form of the specific embodiment is as follows:
extracting a brain activation region based on a multi-mode distance between fine activity modes contained in a local brain region, and preprocessing an fMRI image; clustering to obtain a local consistent brain area; constructing a multivariate pattern by using the joint activities of a plurality of voxels in a locally consistent brain region; and constructing a multivariate distance function by using a mode classification method to measure separable properties of local brain region activities under different stimulation conditions, and judging whether the brain region is activated. The invention directly uses the multi-element mode information composed of multiple voxels in the local brain area to represent the brain activity under different stimulation conditions, the multi-element mode can comprehensively reflect the local brain activity state, the multi-element statistical distance can effectively integrate the information in the local brain area to measure the difference between different brain activation states, and the two together ensure that the invention can more completely detect the fine brain activation mode compared with the traditional fMRI analysis technology.
The step of preprocessing the normalized image data comprises:
step A.1, different samples of image data are standardized, so that the mean value is 0 and the standard deviation is 1;
step A.2, different characteristics of the image data are standardized, so that the mean value is 0 and the standard deviation is 1.
Secondly, constructing a local consistent brain region, further comprising:
step B.1, measuring the activity similarity between different voxels by using the Pearson correlation coefficient of the voxel activity time sequence as a criterion;
step B.2 Using voxel v0As the seed points, the region growing algorithm selects the domain voxels with the most similar activity to the seed points each time as the local consistent brain region; the region growing is carried out iteratively, and the operation is stopped after the specified size of the local region is reached; so that the resulting voxel activity within the local region is all substantially uniform.
Thirdly, constructing a multivariate mode statistical distance step, further comprising:
step D.1: voxel v0Local consensus brain region N (v)0) Under different stimulation conditions X and Y, the measured combined activity of K voxels in the local area is from two different multivariate randomThe machine variable X ═ X1，X2，…，Xi，…，XK)T，Y＝(Y1，Y2，…，Yi，…，YK)TI-1, 2, …, K samples the resulting pattern samples;
step D.2: respectively corresponding to the sample set S by the stimulation conditions X and YXAnd SY(ii) a Constructing different multivariate distance functions to measure the sample set S according to the pattern recognition methodXAnd SYThe multiple distances are Fisher distance functions derived according to Fisher linear discriminant analysis, maximum boundary distance functions derived according to a support vector machine, and multiple distance functions derived by other mode classification algorithms;
step D.3: here, FLDF is taken as an example and set S according to Fisher' S Linear discriminant analysisXAnd SYThe FLDF between is calculated by:
wherein,
are respectively a sample SXAnd SYThe average vector of (d) is:
Σ is the sample mixture covariance:
Fourthly, a step of hypothesis testing of the multivariate distance statistical parameter graph:
in each local brain region N (v)0) The activity pattern is the same under different stimulation conditions, namely: under the assumption of zero which cannot be obviously separated, a non-parameter sequence test is used for testing the statistical distance parameter corresponding to each voxel to obtain a significance non-parameter sequence test value graph;
non-parametric sequence verification is performed by magnetic resonance image data resampling as follows: the data space mode is unchanged, and the time sequence corresponding to the magnetic resonance image data is randomly rearranged; the random rearrangement destroys the correlation between the magnetic resonance image signals and the test conditions, and reserves the complete space structure, thereby calculating the non-parameter sequence test values corresponding to the multivariate distance statistics among different local brain activity modes.
Fifthly, the step of hypothesis test multiple comparison and correction further comprises the following steps:
step F.1, selecting a boundary value q (0 < q < 1) of the error discovery rate FDR, wherein the boundary value q is the expected maximum error discovery rate FDR;
step F.2, sorting the values of the non-parameter sequence inspection p-value graph obtained by the non-parameter arrangement inspection from small to large: p (1) ≦ p (2) ≦ … ≦ p (m), where p (i) is the corresponding voxel statistic v (i), and m is the number of voxels common to the fMRI data examined;
step F.3, assuming that r satisfies the inequality
<math>
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>≤</mo>
<mfrac>
<mi>i</mi>
<mi>V</mi>
</mfrac>
<mfrac>
<mi>q</mi>
<mrow>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
</math>
Where c (m) is a predetermined constant, and its choice is related to the distribution of voxels, and there are 2 choices under different conditions: c (m) is equal to 1,
<math>
<mrow>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mi>Σ</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>V</mi>
</munderover>
<mfrac>
<mn>1</mn>
<mi>i</mi>
</mfrac>
<mo>;</mo>
</mrow>
</math>
step F.4, the result is obtained: rejecting null hypothesis condition, i.e. true activated voxels are v (1), v (2), …, v (r), i.e. voxels whose statistic is greater than v (r) are activated voxels, called v (r) as threshold value determined for multiple tests;
and F.5, performing threshold segmentation on the statistical parameter graph according to the multi-test threshold v (r) obtained by FDR correction to obtain brain activation regions capable of distinguishing different test conditions obviously.
Second, simulation data experiment A is introduced
A.1 simulation data design
Since the position and shape of the activation region of the real mri data cannot be accurately known, the simulation data (the position, shape and voxel activity of the activation region are preset) is designed firstly, and the simulation data is used for comparing the performance of different activation region detection methods. The simulation data comprises two types of stimulation conditions, and the slow presentation event related design is adopted. Each stimulus is presented for 500 milliseconds, with different event initiation intervals ranging from 16 to 20 seconds. Each type of stimulation condition contained 30 events, presented in a random order. The fMRI time series corresponding to brain activation by each type of stimulus was simulated by convolving the standard hemodynamic Function (cHRF) with a sequence of stimulus events. The data are specified as follows: the number of layers (number of slices) is 5, the voxel size (voxel sizes) is 3 × 3 × 3 cubic millimeters, the repetition Time (TR) is 2000 milliseconds, and the matrix (matrix) is 64 × 64.
Five activation areas with different shapes and sizes were simulated (fig. 6a), containing 10, 30, 90, 180, 270 voxels. Within the activation zone, the spatial effects caused by each type of stimulus obey a gaussian white noise distribution. Outside the activation zone, the stimulus causes no effect. The activation mode caused by various stimuli is simulated by Gaussian white noise, so that the energy of an effect signal can be uniformly dispersed in different spatial frequency bands, and the local average signal (low-frequency component) and the fine spatial structure (high-frequency component) of the simulated activation brain area both contain information related to stimulation conditions, thereby being closer to real fMRI data. The brain activity signal caused by the stimulus is then added to a spatiotemporal noise background. To simulate the local spatial correlation properties of noise in real fMRI data, the noise background is generated as follows: firstly, space-time white Gaussian noise is obtained through simulation, and then the generated white Gaussian noise is smoothed by using a Gaussian kernel with the Full Width at half maximum (FWHM) of 3.5 mm, so that a noise background with local space correlation characteristics is obtained. To examine the performance of the activation region extraction algorithm at different Contrast to noise ratios (CNR), five sets of data with different Contrast to noise ratios were generated, with the Contrast to noise ratios respectively taking on values of 0.2, 0.4, 0.6, 0.8, and 1.0. The contrast-to-noise ratio is defined as the ratio of the maximum absolute value of the spatial mean signal amplitude in each brain activation region to the background noise standard deviation.
A.2 simulation data analysis
The simulation data is analyzed in three ways:
(1) raw data-GLM analysis;
(2) gaussian kernel smoothing data-GLM analysis;
(3) and (4) LMDM analysis.
Gaussian kernel smoothing data-in GLM analysis, gaussian kernel filters with 6 mm and 9 mm full width at half maximum are used to smooth the data, respectively. These two gaussian kernel filters are employed because they are commonly employed in fMRI data analysis and literature reports. The local region size K adopted by LMDM analysis is 10 voxels and 30 voxels respectively, which ensures that the number of local voxels used by LMDM analysis is approximately equal to the number of local voxels covered by two Gaussian kernels in Gaussian kernel smooth data-GLM analysis, and enables the local spatial signal integration efficiency of LMDM analysis and Gaussian kernel smooth data-GLM analysis to be objectively compared.
The Receiver Operating Characteristics (ROC) curve is used to quantitatively compare the performance of different analytical methods. And setting different thresholds for segmenting the statistical parameter graph obtained by calculation of a specific analysis method so as to obtain different activation graphs. The ROC curve plots the true activation rate (true activation rate) and false activation rate (false activation rate) of the activation map detected by the analysis method as the threshold value varies from the minimum value to the maximum value in the parameter map. The true activation rate is defined as the number of true activation voxels on the ratio of the number of voxels correctly identified as activation voxels; and the number of voxels erroneously identified as activated voxels compared to the number of true non-activated voxels is defined as a false activation rate. According to the signal detection theory, the ROC curve represents the restriction relationship between sensitivity (sensitivity) and specificity (specificity) of a signal detection method, and can be used for quantitative comparison of different signal detection methods. Here, the true activation rate corresponds to the sensitivity of the active region extraction algorithm, and the false activation rate corresponds to the specificity of the active region extraction algorithm. The area enclosed by the ROC curve (ROC area) can be used as a comprehensive index to measure the degree of high sensitivity and specificity of the signal detection method. That is, this metric measures how correctly the statistical parameters computed by a particular method can distinguish a simulated activated voxel from background noise.
Fig. 2 is a schematic diagram of extracting a local consistent brain region in the present invention, fig. 2a illustrates seed voxels used in a region growing process, and fig. 2b illustrates a local consistent region obtained by the region growing process.
FIG. 3 is a schematic diagram of the concept of pattern classification for measuring the statistical distance between different activation patterns according to the present invention (here, linear discriminant analysis is taken as an example). In fig. 3, open circles represent sample points measured under the stimulation condition X, and filled circles represent sample points measured under the stimulation condition Y. For the sake of illustration on a two-dimensional map, it is assumed here that the local spatial mode consists of two voxels (X-axis and Y-axis). The linear discriminant analysis enables the two types of samples to have the best separability on an optimal discriminant axis by projecting the two types of sample points on the axis, and the separability of the two types of samples on the axis can be measured through a Fisher linear discriminant function.
In the comparison between the algorithm of the present invention and the conventional active region detection algorithm in the example of fig. 4, the curves are obtained by averaging 30 simulation experiments, and the situations that the area enclosed by ROC curves corresponding to the three analysis methods varies with the contrast-to-noise ratio are given. It can be clearly observed that at any contrast-to-noise ratio, the data smoothing degrades the detection performance of the GLM analysis. The more severe the data smoothing (the larger the smoothing window), the more the performance degrades because many of the high frequency information on the fine scale is filtered out when the gaussian kernel smoothes the data. The performance of LMDM analysis is superior to GLM analysis in each case because LMDM completely describes and exploits the brain activity information contained in the local region by directly using the spatial pattern of multiple voxels.
Fig. 5 is a comparison of the algorithm of the present invention and a conventional active area detection algorithm in an example. There are shown ROC curves for the three analysis methods under an example of simulation data with a contrast to noise ratio of 0.6. It can be seen that the ROC curve for the LMDM analysis is in all cases at the top left, higher than the ROC curve for all GLM analyses. The ROC curve obtained by the original data-GLM analysis is positioned between the ROC curves obtained by the LMDM analysis and the Gaussian kernel smoothing data-GLM analysis, and the ROC curve shows that the performance of the original data-GLM analysis is worse than that of the LMDM analysis but better than that of the Gaussian kernel smoothing data-GLM analysis. The simulation data have the same energy level in all frequency bands lower than the Nyquist frequency (determined by the voxel size), and the Gaussian kernel smoothing data-GLM analysis filters out high-frequency information and reduces signal energy, so that the high-frequency information cannot be utilized, the detection performance is reduced, and the performance of the three is the worst; original data-GLM analysis is not filtered, the high-frequency information is reserved, but the performance is lower than that of LMDM analysis because only information of a single voxel is utilized and information in a local brain area is not utilized; the LMDM analysis effectively utilizes the information of all frequency bands distributed in a local space mode to judge different brain activity states, so that the performance of the LMDM analysis is optimal.
For visual understanding of fig. 4, the quantitative results given in fig. 5, fig. 6(a), (b), (c), (d) compare the algorithm of the present invention with the conventional activation region detection algorithm in the example. Three methods are presented for the non-thresholded parameter plots (all taken from the middle layer of simulation data) obtained in fig. 5 on a sample of the simulation data used with a contrast noise of 0.6. Fig. 6(a) simulates an active area profile. FIG. 6(b) raw data, GLM analysis results. FIG. 6(c) data was filtered through a Gaussian kernel with a half-height width of 9 mm and analyzed by GLM. Fig. 6(d) shows the LMDM analysis result with the local region size set to 30 voxels. In the figure, the real brain activation zone boundaries have been marked by contour lines. It can be seen that the absolute t-value map obtained by the raw data-GLM analysis does not well detect some weak moving voxels in the simulated activation region, and the overall signal-to-noise ratio of the parameter map is also low. After Gaussian kernel smoothing data with the half-height width of 9 mm are adopted, the signal-to-noise ratio of a parameter map extracted by GLM analysis is improved, but some obvious random pseudo-activation voxels appear outside a real simulation activation region. Compared with two types of GLM analysis, the signal-to-noise ratio of the multivariate distance statistical parameter map obtained by the LMDM analysis is higher, a real brain activation region is accurately detected, and meanwhile, no pseudo activation voxel appears outside the real activation region. These undivided parametric maps qualitatively illustrate that LMDM analysis can more accurately detect local brain regions that can distinguish between different brain activity states by using multivariate statistical distances to integrate complete information within the local brain regions.
Thirdly, real data experiment B is introduced
B.1 true fMRI data acquisition
6 subjects were tested in each half of the study. The experiment uses a block design with two types of stimulation conditions, one type of stimulation condition is a face picture, and the other type of stimulation condition is a house picture. Presenting a stimulation (human face or house) picture with a number of 'ten' in the center in the stimulation condition group block; within the baseline chunk (baseline), the presentation content is simply a picture of "ten". The subject is asked to place the point of regard on the "ten" number while viewing the picture. Each stimulation block lasted 30 seconds during which a total of 20 homogeneous pictures were presented, each picture was presented for 500 milliseconds, with pictures "ten" at stimulation picture intervals. The baseline block length is about 10 seconds, allowing the baseline block length to be variable to more fully capture the hemodynamic response of the human brain to the stimulus, thereby facilitating detection of brain activation regions. Bold mri data were acquired using a T2-weighted gradient Echo Planar Imaging (EPI) sequence with detailed parameters as follows: the layer thickness (slice thickness) is 4 mm, the window (FOV) is 240 × 240 mm square, the repetition Time (TR) is 2000 ms, and the matrix (matrix) is 64 × 64.
The data were preprocessed using SPM2(http:// www.fil.ion.ucl.ac.uk/SPM /), which included: time alignment, spatial rearrangement, de-baseline drift, removal of non-brain voxels. Additionally, for LMDM analysis, the fMRI time sequence was moved forward for 4 seconds to offset hemodynamic response delays; the data corresponding to the block transitions were discarded, and only the data measured at hemodynamic stability were retained.
B.2 true fMRI data analysis
The real data is analyzed in the same way as the simulation data: (1) raw data-GLM analysis, (2) gaussian kernel smoothing data-GLM analysis, and (3) LMDM analysis. As can be seen from simulation experiments, when the size of the local region is set to be 30 voxels, the LMDM analysis detection performance is the best, and the number of the used local voxels is approximately equal to the number of the local voxels covered by a Gaussian kernel with the half-height width of 9 mm. The data is analyzed using only these two parameters in the following. As a result, as shown in fig. 7, the main brain regions (FFA and PPA) corresponding to the difference between the facial image stimulation and the atrial image stimulation can be detected by all of the three analysis methods. In the figure, all significance p-value maps were obtained by 1000 independent row resampling. Except that the significance p-value plots for both gaussian kernel smoothing data-GLM analysis and LMDM analysis have been corrected using FDR, setting the average FDR to not exceed q 0.05; while the raw data-GLM analysis gives a significant p-value map without FDR correction, no voxel will achieve significant activation if the same correction as the other two is used. The activation graph obtained by GLM analysis shows that the voxel activity caused by the stimulation of the face picture at the position is obviously stronger than the voxel activity caused by the stimulation of the house picture, or the voxel activity caused by the stimulation of the house picture at the position is obviously stronger than the activity caused by the stimulation of the face picture. The activation graph obtained by the LMDM analysis shows that the spatial mode formed by combining the corresponding voxels and the voxel activities in the surrounding local area has good separability under the two types of picture stimulation, namely the modes of local brain area activity presentation caused by the two types of stimulation conditions are obviously different. As shown by comparing the algorithm of the present invention with the conventional activation region detection algorithm in the examples of fig. 7a, 7b, and 7c, the raw data-GLM analysis can approximately obtain the location of the brain activation region, but there is severe salt and pepper noise. The contrast noise of a single voxel signal in fMRI data is low, and the raw data-GLM analysis result of fig. 7a only uses single voxel activity information to determine the state of a voxel, but does not use information contained in a local brain region, so that the signal-to-noise ratio of the obtained parameter map is low. After data was filtered using a gaussian kernel with a half-width height of 9 mm, the activation map obtained using the GLM analysis results was cleaner and more concentrated than the activation map obtained from the original data-GLM analysis, but only those core activation voxels with stronger differences in response to the two types of stimuli were detected (fig. 7 b). However, the LMDM analysis using local region size of 30 voxels detected more activated voxels than the gaussian kernel smoothed data-GLM analysis, which were distributed around the central activation region obtained by the gaussian kernel smoothed data-GLM analysis. This indicates that besides the core activation region detected by the gaussian kernel smoothing data-GLM analysis, there are still many voxels around it that contain fine information and respond weakly to experimental stimuli (fig. 7c), and when these voxels are jointly analyzed, it can also distinguish the different experimental stimulus conditions well. The inability of GLM analysis to detect these marginal, weakly informative voxels arises mainly from two reasons: firstly, GLM analysis is essentially unitary analysis, and when activated region detection is carried out, each voxel is processed separately, so that spatial mode information contained in a local region cannot be comprehensively utilized; the other is that while on one hand, gaussian kernel filtering enhances the statistical effectiveness of GLM analysis and improves the signal-to-noise ratio of the biotin signal in the core activation region, it unreasonably filters out most of the fine brain activity information contained in the local region. By analyzing all the tested data, we further found that the activation region detected by the LMDM analysis is more extended and distributed than that detected by the gaussian kernel smoothing data-GLM analysis, which is consistent with the previous research results.
In fig. 7, L indicates the left side of the side corresponding to the axial position of the human brain, and R indicates the right side of the side corresponding to the axial position of the human brain. Face represents a Face picture stimulation condition, and house picture stimulation condition. In the figure, voxels with activation difference to the face picture and the house picture are both checked by the permatation. b and c use FDR correction to ensure that the average FDR of all activated voxels is less than 0.05.
To quantitatively compare the repeatability (similarity or consistency) of the parameter maps given by different analysis methods among the tested samples, the tested parameter maps are firstly registered on a standard MNI template, and then the Pearson correlation coefficient (r) of each pair of tested parameter maps is calculated to measure the similarity between the tested parameter maps. In the example of fig. 8, the algorithm of the present invention is compared with the conventional active area detection algorithm, and the consistency of parameter maps corresponding to the three analysis methods among different subjects is shown in the figure, and GLM, sm-GLM, and LMDM on the horizontal axis represent the raw data-GLM analysis, gaussian kernel smoothing data-GLM analysis, and LMDM analysis, respectively; the vertical axis scale represents the pearson correlation coefficient. 15 pairs (C) for each analytical method in the form of mean. + -. standard mean error (r. + -. SEM)6 2) The pearson correlation coefficient of the parameter map was counted. Of the three analysis results, the statistical parameter map obtained by the raw data-GLM analysis has the worst similarity (r ═ 0.25 ± 0.04) among the tested results, which is caused by the low signal-to-noise ratio of the parameter map. The similarity of the parameter map of the LMDM analysis between the subjects (r 0.49 ± 0.03) is worse than that of the gaussian kernel smoothing data-GLM analysis (r 0.57 ± 0.02). This is because on a fine spatial scale, the brain activation patterns themselves vary greatly from subject to subject; it may also be an LMDThe M analysis does not adopt space smoothing, and is sensitive to errors introduced in a space standardization process, so that the consistency of a tested space parameter map is reduced. Therefore, the method is particularly suitable for the fMRI brain activation area of a single tested object, and can extract the brain activation information of the single tested object more finely compared with the traditional fMRI activation area extraction method.
The above description is of embodiments for carrying out the invention, and it will be understood by those skilled in the art that any modification or partial replacement without departing from the scope of the invention is intended to be included within the scope of the appended claims.
Claims (6)
1. A method for extracting a magnetic resonance imaging brain activation region based on pattern recognition classification is characterized by comprising the following steps: comprises the following steps:
a pretreatment step A: preprocessing the acquired magnetic resonance image, removing confusion factors and obtaining standardized image data;
and B, constructing a local consistent brain region: for each voxel v of the image data0Using region growing algorithm with v0As seed points, local consensus brain region N (v) containing K-idiosin was obtained0)；
When dividingStep C: dividing local consistent brain region N (v) according to stimulation condition category corresponding to experimental design0) A time sequence of a plurality of voxels in the time sequence, each type of stimulation condition corresponds to a group of different multi-metadata samples;
constructing a multivariate mode statistical distance step D: under different stimulation conditions, the local consistent brain region N (v)0) The spatial mode formed by the K voxels in the array corresponds to different K-dimensional random distribution; constructing a multivariate statistical distance function using pattern recognition methods to measure local consensus brain region N (v)0) Statistical distances between data samples corresponding to different conditions; calculating a multivariate statistical distance by iterating the above process for each voxel to obtain a multivariate statistical distance parameter map;
a multivariate distance statistical parameter graph hypothesis testing step E: suppose that in each locally consistent brain region N (v)0) The activity of the activity mode under different stimulation conditions is the same, and a non-parameter sequence test is used for testing the statistical distance parameter corresponding to each voxel to obtain a significance non-parameter sequence test value graph;
hypothesis testing multiple comparison correction step F: the method of false discovery rate FDR is used for eliminating multiple comparisons to obtain local brain regions, namely corresponding activation regions, which can obviously distinguish different test conditions.
2. The method for extracting mri brain activation region of claim 1 wherein the preprocessing step normalizes image data further comprising:
step A.1, different samples of image data are standardized, so that the mean value is 0 and the standard deviation is 1;
step A.2, different characteristics of the image data are standardized, so that the mean value is 0 and the standard deviation is 1.
3. The method for extracting an mri brain activation region as set forth in claim 1, wherein said step of constructing a locally consistent brain region further includes:
step B.1, measuring the activity similarity between different voxels by using the Pearson correlation coefficient of the voxel activity time sequence as a criterion;
step B.2 Using voxel v0As the seed points, the region growing algorithm selects the domain voxels with the most similar activity to the seed points each time as the local consistent brain region; the region growing is carried out iteratively, and the operation is stopped after the specified size of the local region is reached; so that the resulting voxel activity within the local region is all substantially uniform.
4. The method for extracting mri brain activation region of claim 1, wherein said constructing a multivariate pattern statistical distance step further comprises:
step D.1: voxel v0Local consensus brain region N (v)0) Under different stimulation conditions X and Y, the measured combined activity of K voxels in the local area is obtained from two different multivariate random variables X ═ X (X1，X2，…，Xi，…，XK)T，Y＝(Y1，Y2，…，Yi，…，YK)TI-1, 2, …, K samples the resulting pattern samples;
step D.2: respectively corresponding to the sample set S by the stimulation conditions X and YXAnd SY(ii) a Constructing different multivariate distance functions to measure the sample set S according to the pattern recognition methodXAnd SYThe multiple distances are Fisher distance functions derived according to Fisher linear discriminant analysis, maximum boundary distance functions derived according to a support vector machine, and multiple distance functions derived by other mode classification algorithms;
step D.3: here, FLDF is taken as an example and set S according to Fisher' S Linear discriminant analysisXAnd SYThe FLDF between is calculated by:
wherein,
are respectively a sample SXAnd SYThe average vector of (d) is:
Σ is the sample mixture covariance:
5. The method for extracting an mri brain activation region as set forth in claim 1, wherein the multivariate distance statistical parameter map hypothesis test step:
in each local brain region N (v)0) The activity pattern is the same under different stimulation conditions, namely: under the assumption of zero which cannot be obviously separated, a non-parameter sequence test is used for testing the statistical distance parameter corresponding to each voxel to obtain a significance non-parameter sequence test value graph;
non-parametric sequence verification is performed by magnetic resonance image data resampling as follows: the data space mode is unchanged, and the time sequence corresponding to the magnetic resonance image data is randomly rearranged; the random rearrangement destroys the correlation between the magnetic resonance image signals and the test conditions, and reserves the complete space structure, thereby calculating the non-parameter sequence test values corresponding to the multivariate distance statistics among different local brain activity modes.
6. The method for extracting mri brain activation regions of claim 1 wherein the hypothesis testing multiple comparison correction step further comprises:
step F.1, selecting a boundary value q (0 < q < 1) of the error discovery rate FDR, wherein the boundary value q is the expected maximum error discovery rate FDR;
step F.2, sorting the values of the non-parameter sequence inspection p-value graph obtained by the non-parameter arrangement inspection from small to large: p (1) ≦ p (2) ≦ … ≦ p (m), where p (i) is the corresponding voxel statistic v (i), and m is the number of voxels common to the fMRI data examined;
step F.3, assuming that r satisfies the inequality
<math>
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>≤</mo>
<mfrac>
<mi>i</mi>
<mi>V</mi>
</mfrac>
<mfrac>
<mi>q</mi>
<mrow>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
</math>
Where c (m) is a predetermined constant, and its choice is related to the distribution of voxels, and there are 2 choices under different conditions: c (m) is equal to 1,
<math>
<mrow>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mi>Σ</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>V</mi>
</munderover>
<mfrac>
<mn>1</mn>
<mi>i</mi>
</mfrac>
<mo>;</mo>
</mrow>
</math>
step F.4, the result is obtained: rejecting null hypothesis condition, i.e. true activated voxels are v (1), v (2), …, v (r), i.e. voxels whose statistic is greater than v (r) are activated voxels, called v (r) as threshold value determined for multiple tests;
and F.5, performing threshold segmentation on the statistical parameter graph according to the multi-test threshold v (r) obtained by FDR correction to obtain brain activation regions capable of distinguishing different test conditions obviously.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200710098691A CN101292871B (en) | 2007-04-25 | 2007-04-25 | Method for specification extraction of magnetic resonance imaging brain active region based on pattern recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200710098691A CN101292871B (en) | 2007-04-25 | 2007-04-25 | Method for specification extraction of magnetic resonance imaging brain active region based on pattern recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101292871A true CN101292871A (en) | 2008-10-29 |
| CN101292871B CN101292871B (en) | 2010-05-26 |
Family
ID=40063606
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200710098691A Expired - Fee Related CN101292871B (en) | 2007-04-25 | 2007-04-25 | Method for specification extraction of magnetic resonance imaging brain active region based on pattern recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101292871B (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101847261A (en) * | 2010-05-10 | 2010-09-29 | 南京航空航天大学 | Obtaining method of rat head magnetic resonance image Monte Carlo simulation model |
| CN102508184A (en) * | 2011-10-26 | 2012-06-20 | 中国科学院自动化研究所 | Brain function active region detection method based on moving average time series models |
| CN101706561B (en) * | 2009-11-20 | 2012-07-04 | 电子科技大学 | Clustering method for functional magnetic resonance images |
| CN101666865B (en) * | 2009-09-09 | 2012-12-19 | 西北工业大学 | Method for registering diffusion tensor nuclear magnetic resonance image in local quick traveling mode |
| CN103020653A (en) * | 2012-12-07 | 2013-04-03 | 中国科学院自动化研究所 | Structure and function magnetic resonance image united classification method based on network analysis |
| WO2013097118A1 (en) * | 2011-12-28 | 2013-07-04 | 中国科学院自动化研究所 | Method for processing brain function magnetic resonance data |
| CN101744612B (en) * | 2008-11-28 | 2014-07-23 | Ge医疗系统环球技术有限公司 | Blood flow dynamic analysis apparatus and magnetic resonance imaging system |
| CN105513058A (en) * | 2015-12-01 | 2016-04-20 | 中国科学院苏州生物医学工程技术研究所 | Brain active region detection method and device |
| CN105653861A (en) * | 2015-12-31 | 2016-06-08 | 首都医科大学宣武医院 | Similarity calculation method between category concepts based on fMRI |
| CN106548206A (en) * | 2016-10-27 | 2017-03-29 | 太原理工大学 | Multi-modal nuclear magnetic resonance image data classification method based on minimum spanning tree |
| CN108197661A (en) * | 2018-01-19 | 2018-06-22 | 上海海事大学 | Cognitive activities state classification system and method based on voxel level brain function information |
| CN108596228A (en) * | 2018-04-13 | 2018-09-28 | 江南大学 | A kind of brain functional magnetic resonance image classification based on unsupervised fuzzy system |
| CN109830286A (en) * | 2019-02-13 | 2019-05-31 | 四川大学 | Brain function magnetic resonance code energy imaging method based on nonparametric statistics |
| CN110087546A (en) * | 2016-12-21 | 2019-08-02 | 弗门尼舍有限公司 | For determining the fMRI method of the brain activation pattern of the feeling excited in response to smell |
| CN111329446A (en) * | 2020-02-26 | 2020-06-26 | 四川大学华西医院 | Visual stimulation system and method for processing spatial frequency of facial pores through brain visual pathway |
| CN112529918A (en) * | 2020-12-23 | 2021-03-19 | 沈阳东软智能医疗科技研究院有限公司 | Method, device and equipment for ventricular region segmentation in brain CT image |
| CN112686902A (en) * | 2019-10-17 | 2021-04-20 | 西安邮电大学 | Two-stage calculation method for brain glioma identification and segmentation in nuclear magnetic resonance image |
| CN116583221A (en) * | 2021-09-17 | 2023-08-11 | 皇家飞利浦有限公司 | Determination of subject-specific hemodynamic response functions |
| CN118470118A (en) * | 2024-07-12 | 2024-08-09 | 广州市第一人民医院（广州消化疾病中心、广州医科大学附属市一人民医院、华南理工大学附属第二医院） | RTMS target positioning method and system based on multi-mode image |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11826133B2 (en) * | 2017-02-28 | 2023-11-28 | Board Of Trustees Of Michigan State University | Method and system for determining brain-state dependent functional areas of unitary pooled activity and associated dynamic networks with functional magnetic resonance imaging |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6292683B1 (en) * | 1999-05-18 | 2001-09-18 | General Electric Company | Method and apparatus for tracking motion in MR images |
| EP1219243A1 (en) * | 2000-12-28 | 2002-07-03 | Matsushita Electric Works, Ltd. | Non-invasive brain function examination |
| DE10162927A1 (en) * | 2001-12-20 | 2003-07-17 | Siemens Ag | Evaluation of images of the brain obtained using functional magnetic resonance tomography |
| CN1294875C (en) * | 2003-12-12 | 2007-01-17 | 中国科学院自动化研究所 | Time series analysis method of nuclear magnetic resonance for brain functions based on constrained optimization |
| CN1626031A (en) * | 2003-12-12 | 2005-06-15 | 中国科学院自动化研究所 | Method for detecting functional connection between brain regions based on graph theory |
| CN1939211A (en) * | 2005-09-30 | 2007-04-04 | 中国科学院自动化研究所 | Analysis of sea-horse quantitative shape in magnetic resonance image |
-
2007
- 2007-04-25 CN CN200710098691A patent/CN101292871B/en not_active Expired - Fee Related
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101744612B (en) * | 2008-11-28 | 2014-07-23 | Ge医疗系统环球技术有限公司 | Blood flow dynamic analysis apparatus and magnetic resonance imaging system |
| CN101666865B (en) * | 2009-09-09 | 2012-12-19 | 西北工业大学 | Method for registering diffusion tensor nuclear magnetic resonance image in local quick traveling mode |
| CN101706561B (en) * | 2009-11-20 | 2012-07-04 | 电子科技大学 | Clustering method for functional magnetic resonance images |
| CN101847261B (en) * | 2010-05-10 | 2012-02-15 | 南京航空航天大学 | Obtaining method of rat head magnetic resonance image Monte Carlo simulation model |
| CN101847261A (en) * | 2010-05-10 | 2010-09-29 | 南京航空航天大学 | Obtaining method of rat head magnetic resonance image Monte Carlo simulation model |
| CN102508184B (en) * | 2011-10-26 | 2015-04-08 | 中国科学院自动化研究所 | Brain function active region detection method based on moving average time series models |
| CN102508184A (en) * | 2011-10-26 | 2012-06-20 | 中国科学院自动化研究所 | Brain function active region detection method based on moving average time series models |
| WO2013097118A1 (en) * | 2011-12-28 | 2013-07-04 | 中国科学院自动化研究所 | Method for processing brain function magnetic resonance data |
| US9633430B2 (en) | 2011-12-28 | 2017-04-25 | Institute Of Automation, Chinese Academy Of Sciences | Method for analyzing functional MRI brain images |
| CN103020653A (en) * | 2012-12-07 | 2013-04-03 | 中国科学院自动化研究所 | Structure and function magnetic resonance image united classification method based on network analysis |
| CN103020653B (en) * | 2012-12-07 | 2015-07-22 | 中国科学院自动化研究所 | Structure and function magnetic resonance image united classification method based on network analysis |
| CN105513058A (en) * | 2015-12-01 | 2016-04-20 | 中国科学院苏州生物医学工程技术研究所 | Brain active region detection method and device |
| CN105513058B (en) * | 2015-12-01 | 2019-04-23 | 中国科学院苏州生物医学工程技术研究所 | A kind of brain active region detection method and device |
| CN105653861A (en) * | 2015-12-31 | 2016-06-08 | 首都医科大学宣武医院 | Similarity calculation method between category concepts based on fMRI |
| CN105653861B (en) * | 2015-12-31 | 2018-03-23 | 首都医科大学宣武医院 | Similarity calculation method between category concepts based on fMRI |
| CN106548206A (en) * | 2016-10-27 | 2017-03-29 | 太原理工大学 | Multi-modal nuclear magnetic resonance image data classification method based on minimum spanning tree |
| CN106548206B (en) * | 2016-10-27 | 2019-08-02 | 太原理工大学 | Multi-modal nuclear magnetic resonance image data classification method based on minimum spanning tree |
| CN110087546A (en) * | 2016-12-21 | 2019-08-02 | 弗门尼舍有限公司 | For determining the fMRI method of the brain activation pattern of the feeling excited in response to smell |
| CN108197661B (en) * | 2018-01-19 | 2022-03-01 | 上海海事大学 | Cognitive activity state classification system and method based on voxel horizontal brain function information |
| CN108197661A (en) * | 2018-01-19 | 2018-06-22 | 上海海事大学 | Cognitive activities state classification system and method based on voxel level brain function information |
| CN108596228B (en) * | 2018-04-13 | 2022-08-23 | 江南大学 | Brain function magnetic resonance image classification method based on unsupervised fuzzy system |
| CN108596228A (en) * | 2018-04-13 | 2018-09-28 | 江南大学 | A kind of brain functional magnetic resonance image classification based on unsupervised fuzzy system |
| CN109830286B (en) * | 2019-02-13 | 2022-09-30 | 四川大学 | Brain function magnetic resonance encoding energy imaging method based on nonparametric statistics |
| CN109830286A (en) * | 2019-02-13 | 2019-05-31 | 四川大学 | Brain function magnetic resonance code energy imaging method based on nonparametric statistics |
| CN112686902B (en) * | 2019-10-17 | 2023-02-03 | 西安邮电大学 | Two-stage calculation method for brain glioma identification and segmentation in nuclear magnetic resonance image |
| CN112686902A (en) * | 2019-10-17 | 2021-04-20 | 西安邮电大学 | Two-stage calculation method for brain glioma identification and segmentation in nuclear magnetic resonance image |
| CN111329446B (en) * | 2020-02-26 | 2022-06-10 | 四川大学华西医院 | Visual stimulation system and method for processing spatial frequency of facial pores through brain visual pathway |
| CN111329446A (en) * | 2020-02-26 | 2020-06-26 | 四川大学华西医院 | Visual stimulation system and method for processing spatial frequency of facial pores through brain visual pathway |
| CN112529918A (en) * | 2020-12-23 | 2021-03-19 | 沈阳东软智能医疗科技研究院有限公司 | Method, device and equipment for ventricular region segmentation in brain CT image |
| CN112529918B (en) * | 2020-12-23 | 2024-02-27 | 沈阳东软智能医疗科技研究院有限公司 | Method, device and equipment for segmenting brain room area in brain CT image |
| CN116583221A (en) * | 2021-09-17 | 2023-08-11 | 皇家飞利浦有限公司 | Determination of subject-specific hemodynamic response functions |
| CN118470118A (en) * | 2024-07-12 | 2024-08-09 | 广州市第一人民医院（广州消化疾病中心、广州医科大学附属市一人民医院、华南理工大学附属第二医院） | RTMS target positioning method and system based on multi-mode image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101292871B (en) | 2010-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101292871B (en) | Method for specification extraction of magnetic resonance imaging brain active region based on pattern recognition | |
| Perlbarg et al. | CORSICA: correction of structured noise in fMRI by automatic identification of ICA components | |
| Kundu et al. | Differentiating BOLD and non-BOLD signals in fMRI time series using multi-echo EPI | |
| LaConte et al. | The evaluation of preprocessing choices in single-subject BOLD fMRI using NPAIRS performance metrics | |
| Sui et al. | An ICA-based method for the identification of optimal FMRI features and components using combined group-discriminative techniques | |
| US7995825B2 (en) | Histogram segmentation of FLAIR images | |
| CN102509123B (en) | Brain function magnetic resonance image classification method based on complex network | |
| WO2011115956A1 (en) | Systems and methods for detection and prediction of brain disorders based on neural network interaction | |
| US7916909B2 (en) | Method and means for image processing | |
| Ambrosen et al. | Validation of structural brain connectivity networks: The impact of scanning parameters | |
| Dean III et al. | Multivariate characterization of white matter heterogeneity in autism spectrum disorder | |
| Zhong et al. | Automated white matter hyperintensity detection in multiple sclerosis using 3D T2 FLAIR | |
| Duncan et al. | Identifying preseizure state in intracranial EEG data using diffusion kernels | |
| Sole et al. | Anisotropic 2-D and 3-D averaging of fMRI signals | |
| Katsaloulis et al. | Fractal dimension and lacunarity of tractography images of the human brain | |
| Meyer-Bäse et al. | Unsupervised clustering of fMRI and MRI time series | |
| Auffermann et al. | Cluster significance testing using the bootstrap | |
| Laudadio et al. | Fast nosological imaging using canonical correlation analysis of brain data obtained by two‐dimensional turbo spectroscopic imaging | |
| Ellingson et al. | Gray and white matter delineation in the human spinal cord using diffusion tensor imaging and fuzzy logic | |
| Khalid et al. | Adaptive 2DCCA based approach for improving spatial specificity of activation detection in functional MRI | |
| Parmar et al. | Effectiveness Analysis of Fuzzy Unsupervised Clustering Algorithms for Brain Tissue Segmentation in Single Channel MR Image | |
| Auffermann et al. | Nonadditive two-way ANOVA for event-related fMRI data analysis | |
| WO2012040219A2 (en) | Context differential autocorrelation mapping of functional mri | |
| Khademi | Medical image processing techniques for the objective quantification of pathology in magnetic resonance images of the brain | |
| Simonetti | Investigation of brain tumor classification and its reliability using chemometrics on MR spectroscopy and MR imaging data |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20100526 Termination date: 20210425 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |