CN113591582A - Character recognition device and method based on resting state functional magnetic resonance data - Google Patents
Character recognition device and method based on resting state functional magnetic resonance data Download PDFInfo
- Publication number
- CN113591582A CN113591582A CN202110743959.4A CN202110743959A CN113591582A CN 113591582 A CN113591582 A CN 113591582A CN 202110743959 A CN202110743959 A CN 202110743959A CN 113591582 A CN113591582 A CN 113591582A
- Authority
- CN
- China
- Prior art keywords
- brain
- factor
- identified
- data
- person
- 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 65
- 230000000284 resting effect Effects 0.000 title claims abstract description 31
- 210000004556 brain Anatomy 0.000 claims abstract description 134
- 239000011159 matrix material Substances 0.000 claims abstract description 51
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims abstract description 42
- 239000008280 blood Substances 0.000 claims abstract description 42
- 210000004369 blood Anatomy 0.000 claims abstract description 42
- 230000001419 dependent effect Effects 0.000 claims abstract description 42
- 229910052760 oxygen Inorganic materials 0.000 claims abstract description 42
- 239000001301 oxygen Substances 0.000 claims abstract description 42
- 238000012880 independent component analysis Methods 0.000 claims abstract description 23
- 238000007781 pre-processing Methods 0.000 claims abstract description 15
- 238000010220 Pearson correlation analysis Methods 0.000 claims abstract description 13
- 230000009467 reduction Effects 0.000 claims description 37
- 238000012549 training Methods 0.000 claims description 32
- 238000012360 testing method Methods 0.000 claims description 18
- 238000012937 correction Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 9
- 230000004886 head movement Effects 0.000 claims description 8
- 239000000178 monomer Substances 0.000 claims description 7
- 238000012847 principal component analysis method Methods 0.000 claims description 7
- 238000012795 verification Methods 0.000 claims description 6
- BTCSSZJGUNDROE-UHFFFAOYSA-N gamma-aminobutyric acid Chemical compound NCCCC(O)=O BTCSSZJGUNDROE-UHFFFAOYSA-N 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000012417 linear regression Methods 0.000 abstract description 9
- 238000004590 computer program Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 5
- 230000003925 brain function Effects 0.000 description 4
- 230000001272 neurogenic effect Effects 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 2
- 238000005481 NMR spectroscopy Methods 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000000205 computational method Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000002996 emotional effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000002599 functional magnetic resonance imaging Methods 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000003997 social interaction Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2134—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on separation criteria, e.g. independent component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses a character recognition method and a system, wherein the method comprises the following steps: acquiring resting state functional magnetic resonance data of a person to be identified; preprocessing the resting state functional magnetic resonance data; determining brain areas of the preprocessed magnetic resonance data by using a group independent component analysis method, and extracting blood oxygen level dependent signals of all the brain areas; carrying out Pearson correlation analysis on the blood oxygen level dependent signals of all the brain areas of each person to be identified to obtain a brain network of each person to be identified; inputting the brain network of each person to be identified as a characteristic into a multi-factor identification model, and outputting a character identification result; the multi-factor identification model comprises a multi-factor similarity matrix; the multiple factors are different kinds of personality labels. The method improves the accuracy of the linear regression model in the existing computer program for identifying the five-factor score, and further improves the accuracy of character identification.
Description
Technical Field
The invention relates to the technical field of personality analysis artificial intelligence, in particular to a personality identification device and method based on resting state functional magnetic resonance data.
Background
The character is the sum of the more stable psychological activities of people, which is the basis for effectively finishing work by only one person who can do or not do the exhibition and developing social interaction. The character to a certain extent measures the influence of a person on the society and the attitude of the surrounding objects for a long time. Most of the traits of a personality are closely related to five factors, including: neurogenic (emotional stability), extroversion, openness, humanity and conscientiousness, therefore, current methods of measuring character are mainly measured using five-factor models (FFM). The current five-factor model (FFM) mainly utilizes brain function network to identify the L of the scoring condition of five factors1A regularized linear regression model.
L1The expression of the regularized linear regression model is:
wherein X ═ X1，x2，...，xn]T∈Rn×dIs a data matrix, n is the number of subjects, d is the number of brain function networks (features) of each person, a weight vector Wk∈Rd×1Model parameters for identifying the kth factor, Yk∈Rn×1Is the output vector whose elements represent the fraction of the kth factor being tested.
However, in the prior art, the linear regression model only identifies scores of five factors according to the brain function network, and does not consider the relationship among the factors, so that a large error exists between the identified factor score and the actual score.
Disclosure of Invention
The invention aims to provide a character recognition device and method based on resting state function magnetic resonance data, so as to improve the accuracy of recognizing five-factor scores by a linear regression model in the existing computer program and further improve the accuracy of character recognition.
In order to achieve the above object, the present invention provides a character recognition method based on resting state functional magnetic resonance data, the method comprising:
acquiring resting state functional magnetic resonance data of a person to be identified;
preprocessing the resting state functional magnetic resonance data;
determining brain areas of the preprocessed magnetic resonance data by using a group independent component analysis method, and extracting blood oxygen level dependent signals of all the brain areas;
carrying out Pearson correlation analysis on the blood oxygen level dependent signals of all the brain areas of each person to be identified to obtain a brain network of each person to be identified;
inputting the brain network of each person to be identified as a characteristic into a multi-factor identification model, and outputting a character identification result; the multi-factor identification model comprises a multi-factor similarity matrix; the multiple factors are different kinds of personality labels.
Optionally, the multi-factor recognition model is:
wherein X ═ X1，x2，...，xn]T∈Rn×dThe method comprises the steps that an input feature matrix of a person to be identified, namely a brain network matrix of the person to be identified; n is the number of people to be identified; d is the number of brain networks of each person to be identified; rn×dData corresponding to all brain networks of n persons to be identified; weight vector Wk∈Rd×1Model parameters for identifying the kth factor; y isk∈Rn×1Is an output vector whose elements represent the fraction of the kth factor of the person to be identified; sijA similarity matrix being a factor i and a factor j; alpha and beta are multi-factor identification model parameters.
Optionally, the method for determining the similarity matrix in the multi-factor recognition model includes:
determining the multi-factor recognition model matrix form as:
wherein, a product of Hadamard, D is a degree matrix of S [ ] S, D-S [ ] S is an S [ ] S Laplacian matrix, tr [ ] S is a trace of a calculation matrix, i.e., the sum of diagonal elements is calculated;
jinjin tea
Its gradient expression is
And gradient descending is carried out by a gradient descending step length gamma, and the gradient descending formula is as follows:
alpha | W | ceiling ray according to near-end optimization1，1The operator of (2) is defined as:
sgn(Wij) Is a sign function, outputs WijPositive and negative of (A), abs (W)ij) D is the number of brain network features of each person to be identified, and l is the number of factors;
keeping W in a feasible domain through an operation operator, fixing W, and converting the multi-factor recognition model into an unconstrained equation by using a Laplacian operator:
solving equation about Si，jThe partial derivatives of and λ are expressed as:
order to
Solving two simultaneous equations to obtain a similarity matrix as:
optionally, the preprocessing the resting state functional magnetic resonance data specifically includes:
selecting reference header data;
registering the resting state functional magnetic resonance data with the reference head data in space according to a set unit volume to obtain head movement correction data;
distinguishing and deleting noise components in the head movement correction data by using an independent component analysis method to obtain de-noising data;
and registering the denoising data of each person to be identified to a standard space to obtain preprocessed magnetic resonance data.
Optionally, the determining, by using a group independent component analysis method, brain regions of the preprocessed magnetic resonance data, and extracting blood oxygen level dependent signals of each brain region specifically include:
performing monomer dimensionality reduction on the preprocessed magnetic resonance data of a single person to be identified by adopting a principal component analysis method to obtain monomer dimensionality reduction data;
splicing the monomer dimensionality reduction data of all the persons to be identified in time according to corresponding voxel positions, and performing dimensionality reduction on the time dimension by using the principal component analysis method of the group again to obtain total dimensionality reduction data;
decomposing the overall dimension reduction data by using an independent component analysis method to obtain each brain region with independent space;
and taking each brain area as a template, and extracting the blood oxygen level dependent signals of different brain areas of all the persons to be identified.
Optionally, the extracting the blood oxygen level dependent signals of different brain areas of all the persons to be identified by using each brain area as a template specifically includes:
extracting blood oxygen level dependent signals of all voxels in each brain region of each person to be identified;
and adding and re-averaging the blood oxygen level dependent signals of all the voxels in each brain area to obtain the blood oxygen level dependent signal of each brain area.
Optionally, the obtaining of the brain network of each to-be-identified person by performing pearson correlation analysis on the blood oxygen level dependent signals of each brain area of each to-be-identified person specifically includes:
substituting the blood oxygen level dependent signals of all brain areas of each person to be identified into a Pearson correlation coefficient formula to obtain a Pearson correlation coefficient matrix;
and performing Fisher z transformation on the Pearson correlation coefficient matrix to obtain the brain network of each person to be identified.
Optionally, the training method of the multi-factor recognition model specifically includes:
acquiring resting state functional magnetic resonance data of a testee;
preprocessing the resting state functional magnetic resonance data;
determining brain areas of the preprocessed magnetic resonance data by using a group independent component analysis method, and extracting blood oxygen level dependent signals of all the brain areas;
carrying out Pearson correlation analysis on the blood oxygen level dependent signals of the brain areas of each testee to obtain a brain network of each testee;
and inputting the brain network of each testee as a characteristic and the five-factor score of the testee as a label into the multi-factor recognition model to be trained for training and testing to obtain the multi-factor recognition model.
Optionally, the training and testing are performed by inputting the brain network of each subject as a feature and the five-factor score of the subject as a label into the multi-factor recognition model to be trained to obtain the multi-factor recognition model, and specifically includes:
dividing experimental data into three groups of a boy group, a girl group and all testees; randomly dividing the three groups of data sets into five folds, taking four folds as a training set, and taking one fold as a testing set;
inputting the brain network of each testee in the training set as a characteristic and the five-factor score of the testee as a label into a multi-factor recognition model for training to obtain a weight W and parameters alpha and beta; then, the reliability of the multi-factor recognition model with the weight W and the parameters alpha and beta is tested by using the test set, and a first-time training multi-factor recognition model is obtained;
randomly dividing the training set into five folds, taking four folds as retraining data, and taking one fold as re-verification data; retraining the first training multi-factor recognition model by utilizing retraining data to obtain a weight W; selecting parameters alpha and beta by adopting a grid search method, and verifying the parameters alpha and beta in the first-time training multi-factor recognition model by utilizing the re-verification data, wherein the value ranges of the alpha and the beta are [2 ]0，21，...，211](ii) a The cycle is five times; obtaining an optimal weight W and optimal parameters alpha and beta;
and evaluating the optimal weight W and the optimal parameters alpha and beta by using the test set, circulating for five times, and taking the five times to obtain the average of errors to obtain the multi-factor recognition model.
The invention also provides a character recognition device based on the resting state functional magnetic resonance data, and the system comprises:
the data acquisition unit is used for acquiring resting state functional magnetic resonance data of the person to be identified;
the preprocessing unit is used for preprocessing the resting state functional magnetic resonance data;
a brain region determining unit, configured to determine a brain region by using a group independent component analysis method for the preprocessed magnetic resonance data, and extract a blood oxygen level dependent signal of each brain region;
the feature extraction unit is used for carrying out Pearson correlation analysis on the blood oxygen level dependent signals of all the brain areas of each person to be identified to obtain a brain network of each person to be identified;
the recognition unit is used for inputting the brain network of each person to be recognized as a feature into the multi-factor recognition model and outputting a character recognition result; the multi-factor identification model comprises a multi-factor similarity matrix; the multiple factors are different kinds of personality labels.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the multi-factor recognition model established in the character recognition method and the system provided by the invention comprises the five-factor similarity matrix, the similarity matrix can automatically update the recognition factor score according to the brain network data, the relation among all factors is considered, the accuracy of recognizing the five-factor score by the linear regression model in the existing computer program is improved, and the accuracy of character recognition is further improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
FIG. 1 is a representation of the correlation of Pearson for a five factor correlation;
FIG. 2 is a flowchart of a multifactor recognition model training process in an embodiment of the present invention;
fig. 3 is a flowchart of a character recognition method according to an embodiment of the present invention;
fig. 4 is a system block diagram of a character recognition system according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a character recognition method and a character recognition system, which are used for improving the accuracy of recognizing five-factor scores by a linear regression model in the conventional computer program and further improving the accuracy of character recognition.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Pearson's correlation analysis was performed on the scores of the five major factors, and the correlation analysis is shown in FIG. 1, where A: and C, humanization is facilitated: accountability, O: openness, N: neurogenic, E: outward property. As shown in FIG. 1, the correlation between the five factors is strong, so that the correlation information is added into the multi-factor recognition model to improve the final character recognition performance.
In performing linear regression, if the factor i and the factor j have a strong relationship, their weights should be close, and thus it is expressed as
In order to prevent a situation of deterioration,adding into
Constraint, here SijThe similarity of factor i and factor j is measured. Therefore, the multi-factor identification model determined by the invention is as follows:
wherein X ═ X1，x2，...，xn]T∈Rn×dInputting a characteristic matrix, namely a brain network matrix of the testee, by the testee; n is the number of subjects; d is the number of brain networks per subject; rn×dExperimental data corresponding to all brain networks of n testees; weight vector Wk∈Rd×1Model parameters for identifying the kth factor; y isk∈Rn×1Is an output vector whose elements represent the scores of the kth factor of the subject; sijA similarity matrix being a factor i and a factor j; alpha and beta are multi-factor identification model parameters; l represents the total number of factors.
The method for determining the similarity matrix in the multi-factor recognition model comprises the following steps:
a1, determining the matrix form of the multi-factor recognition model as:
wherein, a hadamard product represents two matrices, D is a degree matrix of S [ < i > S </i >, D-S [ < i > S </i >, which are traces of the matrices, i.e., sums of diagonal elements of the matrices are calculated;
the parameters W and S in the multi-factor recognition model need to be optimized, and the embodiment is implemented by using alternate optimization. In addition, since equation (2) is convex, but since L1The existence of the norm results in the undifferentiation of the multifactor recognition model. Therefore, it is desirable to fix S to solve for W using near-end optimization. The method comprises the following specific steps:
a2, jin
Its gradient expression is
And gradient descending is carried out by a gradient descending step length gamma, and the gradient descending formula is as follows:
a3, optimizing alpha | W | ceiling according to the near end1，1The operator of (2) is defined as:
sgn(Wij) Is a sign function, outputs WijPositive and negative of (A), abs (W)ij) D is the number of brain network features of each subject, and l is the number of factors, as a function of absolute value.
A4, keeping W in a feasible domain through an operator, fixing W, and converting the multi-factor recognition model into an unconstrained equation by using a Laplacian:
a5, find equation about Si，jThe partial derivatives of and λ are expressed as:
a6, order
Solving two simultaneous equations (6) and (7) yields the similarity matrix as:
as shown in fig. 2, based on the multi-factor recognition model, the present embodiment provides a training method of the multi-factor recognition model, where the training method includes:
step 201: acquiring resting state functional magnetic resonance data of a testee and five-factor scores of the testee;
and scanning by adopting a nuclear magnetic resonance scanner to obtain resting state functional magnetic resonance data (rs-fMRI) of the testee, and then obtaining the five-factor score of the testee in a mode of a cause-free questionnaire. The score is used for training of the subsequent model and comparison of the recognition result. Of course, the five-factor score can also be obtained by other existing means, and is not limited herein. It should be noted that the present embodiment is to identify five factors as categories, and certainly, the present embodiment can be divided into more categories of characters, for example, there are 10 factors and 7 factors in the prior art, and the changes to the number of factors are all within the protection scope of the present invention.
Step 202: and preprocessing the resting state functional magnetic resonance data.
The step 202 specifically includes: head motion correction, denoising, and image registration.
Head movement correction: the human subject inevitably generates a small amount of head movement in the process of scanning data, so that head data obtained for the first time in scanning or obtained in the middle time is selected as reference head data, and then the resting state functional magnetic resonance data is respectively registered with the reference head data in space according to a set unit volume to obtain head movement correction data, so that the correct overlapping of the head movement correction data and the reference head data is ensured, and the total movement visible in the original bold data is eliminated.
Denoising: independent Component Analysis (ICA) is a computational method for separating a multivariate signal into additive subcomponents, and by dividing head motion correction data into good components and noise components, structural noise is removed, and the ICA automatically deletes artifacts or noise caused by a scanning machine or heartbeat, so as to obtain denoised data.
Image registration: in order to extract signals of the same brain region of different testees, the de-noising data of each tester are registered to a standard space, and preprocessed magnetic resonance data are obtained.
Step 203: determining brain regions of the preprocessed magnetic resonance data by using a group independent component analysis method, and extracting Blood Oxygen level dependent signals (Bold signals) of all the brain regions;
the brain is divided into different regions, and it is therefore necessary to determine from the preprocessed magnetic resonance data which voxels, i.e. which specific locations of the brain, each region contains. The function of the same brain region, shortly brain region, is determined to be the same from the preprocessed magnetic resonance data. For example: the signals generated by the vision and motion zones are simultaneous, and when one zone is active, the signals generated by the other zone are similar.
Since each brain region has more signals (due to more voxels), the present embodiment represents the final signal of this brain region according to the average of all voxel signals.
In the embodiment, a brain network is taken as a characteristic, the brain network is the mutual relation among different brain areas, and the brain network is represented by a Pearson correlation coefficient; the Bold signal of each brain region is used as an input to measure the correlation between brain regions.
Obtaining a scanning image of the brain (i.e. a four-dimensional image) for each test; and representing the gray value in the brain scanning image as the Bold signal at a certain determined moment, extracting the Bold signals of all voxels of the image according to the corresponding position of each brain region in the template, and further averaging to obtain the final Bold signal.
Based on the above idea, the step 203 may specifically include:
firstly, for all tested blood oxygen level dependent signals, due to the fact that the number of voxels is excessive, and the time series of all tested subjects is too long, the calculation amount is too large, therefore, the single dimension reduction is performed on the preprocessed magnetic resonance data of a single subject by adopting a principal component analysis method, and the single dimension reduction data is obtained: data H obtained by preprocessing a certain tested jj(Hj∈RV*KV is the number of voxels, K is the number of time points sampled), dimensionality reduction by PCA:
Xj＝HjDj
wherein, Xj(Xj∈RV*LL is the number of time points after dimension reduction) is a matrix after dimension reduction, DjIs a dimension reduction matrix.
After the single tested dimensionality reduction is finished, connecting all tested data together and putting the data into a matrix for second dimensionality reduction, namely splicing the single dimensionality reduction data of all tested persons according to voxel positions in time, and reducing dimensionality in a time dimension by using the principal component analysis method of the group again to obtain total dimensionality reduction data F: the overall system is tested to perform the dimensionality reduction after the splicing of the corresponding voxel positions in time, and the following results are obtained:
FT＝(Y1D1，Y2D2，L，YnDn)*G
F(F∈RY*Vy represents the number of time points after dimensionality reduction) represents a matrix after overall dimensionality reduction of the data of the human subject, G is a dimensionality reduction matrix of Ln x Y, G is also determined by PCA decomposition, L is the number of time points after dimensionality reduction of the data corresponding to n test subjects, and n represents the number of test subjects.
According to the matrix form of the obtained overall dimension reduction data F, the overall dimension reduction data F is used as an observation data set, V voxels are provided, and Y is carried out on the voxelsAnd (3) decomposing the overall dimension reduction data by utilizing an independent component analysis method to obtain each brain region with independent space: the spatially independent component map obtained from ICA is Cc＝(Cc1,Cc2,L,CcV) And c is 1,2, L, Y, and the mixing matrix is M, the decomposition formula of ICA can be expressed as:
F＝M*C(F∈RY*V,M∈RY*Z,C∈RZ*V)
each column of M is a time series of corresponding signals, and each row of C is a spatially independent component, i.e., an independent brain region. Thus, brain partitioning is achieved.
The blood oxygen level dependent signals of different brain areas of all the subjects are extracted by taking each brain area as a template. Such as: a trial j extracts the Bold signals of all voxels of a brain region c, adds all the Bold signals and averages the summed signals to obtain the Bold signal xcIs the representative signal of the tested brain region c.
Step 204: carrying out Pearson correlation analysis on the blood oxygen level dependent signals of the brain areas of each testee to obtain a brain network of each testee;
performing Pearson correlation (Pearson correlation) on the obtained blood oxygen level dependent signals to obtain a brain network corresponding to each tested object, and further performing Fisher z transformation on the brain network so as to perform approximate variation stabilization processing on the data. The brain network is pulled into vectors (connections) as features. Specifically, substituting the blood oxygen level dependent signals of each brain area of each subject into a pearson correlation coefficient formula to obtain a pearson correlation coefficient matrix;
pearson correlation coefficient:
here x in the Pearson correlation coefficienti，xjThe Bold signals representing brain region i and brain region j,
each represents xi，xjAverage value of (1), xi，xj∈Rt×1Thus, the resulting matrix of pearson correlation coefficients is:
wherein N represents the number of brain regions.
Fisher z-transformed to obtain:
step 205: taking the brain network of each testee as a characteristic, taking the five-factor score of the testee as a label, inputting the label into the multi-factor recognition model for training and testing to obtain a trained multi-factor recognition model;
the step 205 specifically includes:
dividing experimental data into three groups of a boy group, a girl group and all testees; randomly dividing the three groups of data sets into five folds, taking four folds as a training set, and taking one fold as a testing set;
inputting the brain network of each testee in the training set as a characteristic and the five-factor score of the testee as a label into a multi-factor recognition model for training to obtain a weight W and parameters alpha and beta; then, the reliability of the multi-factor recognition model with the weight W and the parameters alpha and beta is tested by using the test set, and a first-time training multi-factor recognition model is obtained;
randomly dividing the training set into five folds, taking four folds as retraining data, and taking one fold as re-verification data; retraining the first training multi-factor recognition model by utilizing retraining data to obtain a weight W; selecting parameters alpha and beta by adopting a grid search method, and verifying the first-time training multi-factor recognition by utilizing re-verification dataThe values of the selected parameters alpha and beta in the model are in a value range of [2 ]0，21，...，211L, |; the cycle is five times; obtaining an optimal weight W and optimal parameters alpha and beta;
and evaluating the optimal weight W and the optimal parameters alpha and beta by using the test set, circulating for five times, and taking the five times to obtain the average of errors to obtain the trained multi-factor recognition model.
The identification data obtained in this example is compared with the average absolute error of the identification data obtained by the existing method.
By performing experiments, the following results were obtained, including Mean Absolute Error (MAE), percentage of improvement, and run time comparisons.
The average absolute error is expressed as follows:
here, the
Respectively representing the identification scores and the real scores of all the factors k of the testees, and t represents the number of the testees in the test set. The comparative results are shown in Table 1.
TABLE 1
Meanwhile, the invention also obtains a similarity matrix, Sf,SmThe similarity matrix represents the group of all, women and men, respectively. The method comprises the following specific steps:
compared with the prior art, the invention has the advantages that: the recognition of the five factors is simultaneous, no matter the linear recognition model or the elastic network is used for independently recognizing the five factors, the relation among the factors is not considered, and the similarity matrix is introduced to automatically update the recognition factor score according to the data of brain function connection; the similarity matrix obtained at the same time can be used to determine which factors have the most similarity.
The generated effect is as follows: compared with the prior art, the factor score and the real score identified by the method are smaller, the average absolute error is 5.3671, and the average absolute error of the linear regression model is as follows: 8.7241, the error of the elastic network is 7.5747.
Compared with the two existing models, the multi-factor recognition model has great progress in four factors of humanity (A), conscientiousness (C), openness (O) and outward type (E). The reason is that each of the four factors has a good positive correlation with other factors, and the improvement in identifying the neurogenic (N) factor is small. The reason is that the neurogenic property has only a weak or negative correlation with other factors.
In addition, due to the complexity of the proposed model, the time for the model to solve the running code of the process is about 40 times that of the univariate linear model and about 20 times that of the elastic net.
As shown in fig. 3, a character recognition method is provided based on the obtained trained multi-factor recognition model, and the method includes:
step 301: acquiring resting state functional magnetic resonance data of a person to be identified;
step 302: preprocessing the resting state functional magnetic resonance data;
step 303: determining brain areas of the preprocessed magnetic resonance data by using a group independent component analysis method, and extracting blood oxygen level dependent signals of all the brain areas;
step 304: carrying out Pearson correlation analysis on the blood oxygen level dependent signals of all brain areas of each person to be identified to obtain a brain network of each person to be identified;
step 305: and inputting the brain network of each person to be identified as a feature into the trained multi-factor identification model for identification to obtain the score of each factor, and determining the character identification result according to the score of each factor.
As shown in fig. 4, the present invention also provides a character recognition system corresponding to the character recognition method, the system including:
a data acquiring unit 401, configured to acquire resting-state functional magnetic resonance data of a person to be identified;
a preprocessing unit 402, configured to preprocess the resting-state functional magnetic resonance data;
a brain region determining unit 403, configured to determine a brain region by using a group-independent component analysis method on the preprocessed magnetic resonance data, and extract a blood oxygen level dependent signal of each brain region;
the brain region determining unit 403 specifically includes:
the single dimension reduction module is used for performing single dimension reduction on the preprocessed magnetic resonance data of a single testee by adopting a principal component analysis method to obtain single dimension reduction data;
the overall dimension reduction module is used for splicing the monomer dimension reduction data of all the persons to be identified according to voxel positions in time to obtain overall dimension reduction data;
the brain area decomposition module is used for decomposing the overall dimension reduction data by using an independent component analysis method to obtain each brain area which is independent in space;
and the blood oxygen level dependent signal extraction module is used for taking each brain area as a template and extracting the blood oxygen level dependent signals of different brain areas of all the persons to be identified.
A feature extraction unit 404, configured to perform pearson correlation analysis on the blood oxygen level dependent signals of each brain region of each subject to obtain a brain network of each person to be identified;
the recognition unit 405 is configured to input the brain network of each person to be recognized as a feature into the multi-factor recognition model, and output a character recognition result; the multi-factor identification model comprises a multi-factor similarity matrix; the multiple factors are different kinds of personality labels.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (10)
1. A character recognition method based on resting state functional magnetic resonance data is characterized in that the method comprises the following steps:
acquiring resting state functional magnetic resonance data of a person to be identified;
preprocessing the resting state functional magnetic resonance data;
determining brain areas of the preprocessed magnetic resonance data by using a group independent component analysis method, and extracting blood oxygen level dependent signals of all the brain areas;
carrying out Pearson correlation analysis on the blood oxygen level dependent signals of all the brain areas of each person to be identified to obtain a brain network of each person to be identified;
inputting the brain network of each person to be identified as a characteristic into a multi-factor identification model, and outputting a character identification result; the multi-factor identification model comprises a multi-factor similarity matrix; the multiple factors are different kinds of personality labels.
2. The personality recognition method of claim 1, wherein the multi-factor recognition model is:
wherein X ═ X1,x2,…,xn]T∈Rn×dThe method comprises the steps that an input feature matrix of a person to be identified, namely a brain network matrix of the person to be identified; n is the number of people to be identified; d is the number of brain networks of each person to be identified; rn×dData corresponding to all brain networks of n persons to be identified; weight vector Wk∈Rd×1Model parameters for identifying the kth factor; y isk∈Rn×1Is an output vector whose elements represent the fraction of the kth factor of the person to be identified; sijA similarity matrix being a factor i and a factor j; alpha and beta are multi-factor identification model parameters.
3. The character recognition method according to claim 2, wherein the method for determining the similarity matrix in the multi-factor recognition model comprises:
determining the multi-factor recognition model matrix form as:
wherein, a product of Hadamard, D is a degree matrix of S [ ] S, D-S [ ] S is an S [ ] S Laplacian matrix, tr [ ] S is a trace of a calculation matrix, i.e., the sum of diagonal elements is calculated;
order to
Its gradient expression is
And gradient descending is carried out by a gradient descending step length gamma, and the gradient descending formula is as follows:
alpha | W | ceiling ray according to near-end optimization1,1The operator of (2) is defined as:
sgn(Wij) Is a sign function, outputs WijPositive and negative of (A), abs (W)ij) D is the number of brain network features of each person to be identified, and l is the number of factors;
keeping W in a feasible domain through an operation operator, fixing W, and converting the multi-factor recognition model into an unconstrained equation by using a Laplacian operator:
solving equation about Si,jThe partial derivatives of and λ are expressed as:
order to
Solving two simultaneous equations to obtain a similarity matrix as:
4. the personality identification method according to claim 1, wherein the preprocessing of the resting-state functional magnetic resonance data specifically comprises:
selecting reference header data;
registering the resting state functional magnetic resonance data with the reference head data in space according to a set unit volume to obtain head movement correction data;
distinguishing and deleting noise components in the head movement correction data by using an independent component analysis method to obtain de-noising data;
and registering the denoising data of each person to be identified to a standard space to obtain preprocessed magnetic resonance data.
5. The character recognition method of claim 1, wherein the determining the brain regions of the preprocessed magnetic resonance data by group-independent component analysis and extracting the blood oxygen level dependent signals of the respective brain regions comprises:
performing monomer dimensionality reduction on the preprocessed magnetic resonance data of a single person to be identified by adopting a principal component analysis method to obtain monomer dimensionality reduction data;
splicing the monomer dimensionality reduction data of all the persons to be identified in time according to corresponding voxel positions, and performing dimensionality reduction on the time dimension by using the principal component analysis method of the group again to obtain total dimensionality reduction data;
decomposing the overall dimension reduction data by using an independent component analysis method to obtain each brain region with independent space;
and taking each brain area as a template, and extracting the blood oxygen level dependent signals of different brain areas of all the persons to be identified.
6. The character recognition method according to claim 5, wherein the extracting of the blood oxygen level dependent signals of different brain areas of all the persons to be recognized by using each brain area as a template specifically comprises:
extracting blood oxygen level dependent signals of all voxels in each brain region of each person to be identified;
and adding and re-averaging the blood oxygen level dependent signals of all the voxels in each brain area to obtain the blood oxygen level dependent signal of each brain area.
7. The character recognition method according to claim 1, wherein the performing the Pearson correlation analysis on the blood oxygen level dependent signals of the brain areas of each person to be recognized to obtain the brain network of each person to be recognized specifically comprises:
substituting the blood oxygen level dependent signals of all brain areas of each person to be identified into a Pearson correlation coefficient formula to obtain a Pearson correlation coefficient matrix;
and performing Fisher z transformation on the Pearson correlation coefficient matrix to obtain the brain network of each person to be identified.
8. The personality recognition method of claim 1, wherein the training method for the multi-factor recognition model specifically comprises:
acquiring resting state functional magnetic resonance data of a testee and a multi-factor score of the testee;
preprocessing the resting state functional magnetic resonance data;
determining brain areas of the preprocessed magnetic resonance data by using a group independent component analysis method, and extracting blood oxygen level dependent signals of all the brain areas;
carrying out Pearson correlation analysis on the blood oxygen level dependent signals of the brain areas of each testee to obtain a brain network of each testee;
and inputting the brain network of each testee as a characteristic and the multifactor score of the testee as a label into the multifactor recognition model to be trained for training and testing to obtain the multifactor recognition model.
9. The character recognition method according to claim 1, wherein the training and testing are performed by inputting the brain network of each subject as a feature and the five-factor score of the subject as a label into the multi-factor recognition model to be trained, and the obtaining of the multi-factor recognition model specifically comprises:
dividing experimental data into three groups of a boy group, a girl group and all testees; randomly dividing the three groups of data sets into five folds, taking four folds as a training set, and taking one fold as a testing set;
inputting the brain network of each testee in the training set as a characteristic and the five-factor score of the testee as a label into a multi-factor recognition model for training to obtain a weight W and parameters alpha and beta; then, the reliability of the multi-factor recognition model with the weight W and the parameters alpha and beta is tested by using the test set, and a first-time training multi-factor recognition model is obtained;
randomly dividing the training set into five folds, and taking four folds as retraining dataAnd folding as re-verification data; retraining the first training multi-factor recognition model by utilizing retraining data to obtain a weight W; selecting parameters alpha and beta by adopting a grid search method, and verifying the parameters alpha and beta in the first-time training multi-factor recognition model by utilizing the re-verification data, wherein the value ranges of the alpha and the beta are [2 ]0，21，...，211](ii) a The cycle is five times; obtaining an optimal weight W and optimal parameters alpha and beta;
and evaluating the optimal weight W and the optimal parameters alpha and beta by using the test set, circulating for five times, and taking the five times to obtain the average of errors to obtain the multi-factor recognition model.
10. A device for personality identification based on resting functional magnetic resonance data, the system comprising:
the data acquisition unit is used for acquiring resting state functional magnetic resonance data of the person to be identified;
the preprocessing unit is used for preprocessing the resting state functional magnetic resonance data;
a brain region determining unit, configured to determine a brain region by using a group independent component analysis method for the preprocessed magnetic resonance data, and extract a blood oxygen level dependent signal of each brain region;
the feature extraction unit is used for carrying out Pearson correlation analysis on the blood oxygen level dependent signals of all the brain areas of each person to be identified to obtain a brain network of each person to be identified;
the recognition unit is used for inputting the brain network of each person to be recognized as a feature into the multi-factor recognition model and outputting a character recognition result; the multi-factor identification model comprises a multi-factor similarity matrix; the multiple factors are different kinds of personality labels.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110743959.4A CN113591582B (en) | 2021-07-01 | 2021-07-01 | Character recognition device and method based on resting state functional magnetic resonance data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110743959.4A CN113591582B (en) | 2021-07-01 | 2021-07-01 | Character recognition device and method based on resting state functional magnetic resonance data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113591582A true CN113591582A (en) | 2021-11-02 |
| CN113591582B CN113591582B (en) | 2024-05-31 |
Family
ID=78245693
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110743959.4A Active CN113591582B (en) | 2021-07-01 | 2021-07-01 | Character recognition device and method based on resting state functional magnetic resonance data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113591582B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102293656A (en) * | 2011-05-25 | 2011-12-28 | 四川大学华西医院 | Emotional stability evaluation system based on magnetic resonance imaging and evaluation method thereof |
| CN107392907A (en) * | 2017-09-01 | 2017-11-24 | 上海理工大学 | Parahippocampal gyrus function division method based on tranquillization state FMRI |
| US20190188458A1 (en) * | 2017-12-15 | 2019-06-20 | Industrial Technology Research Institute | Method and device for recognizing facial expressions |
| US20200069237A1 (en) * | 2017-02-28 | 2020-03-05 | 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 |
| CN112837819A (en) * | 2021-01-20 | 2021-05-25 | 尹丽君 | Method for establishing acute kidney injury prediction model after coronary artery bypass grafting |
-
2021
- 2021-07-01 CN CN202110743959.4A patent/CN113591582B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102293656A (en) * | 2011-05-25 | 2011-12-28 | 四川大学华西医院 | Emotional stability evaluation system based on magnetic resonance imaging and evaluation method thereof |
| US20200069237A1 (en) * | 2017-02-28 | 2020-03-05 | 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 |
| CN107392907A (en) * | 2017-09-01 | 2017-11-24 | 上海理工大学 | Parahippocampal gyrus function division method based on tranquillization state FMRI |
| US20190188458A1 (en) * | 2017-12-15 | 2019-06-20 | Industrial Technology Research Institute | Method and device for recognizing facial expressions |
| CN112837819A (en) * | 2021-01-20 | 2021-05-25 | 尹丽君 | Method for establishing acute kidney injury prediction model after coronary artery bypass grafting |
Non-Patent Citations (1)
| Title |
|---|
| 李雨明: ""结合表型信息的阿尔兹海默症图卷积神经网络分类方法研究"", 《中国生物医学工程学报》, vol. 2, no. 40, pages 177 - 187 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113591582B (en) | 2024-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109409416B (en) | Feature vector dimension reduction method, medical image identification method, device and storage medium | |
| US9020250B2 (en) | Methods and systems for building a universal dress style learner | |
| CN108229330A (en) | Face fusion recognition methods and device, electronic equipment and storage medium | |
| CN111461176A (en) | Multi-mode fusion method, device, medium and equipment based on normalized mutual information | |
| CN105117731A (en) | Community partition method of brain functional network | |
| CN109255289A (en) | A kind of across aging face identification method generating model based on unified formula | |
| CN112270255B (en) | Electroencephalogram signal identification method and device, electronic equipment and storage medium | |
| CN111568412A (en) | Method and device for reconstructing visual image by utilizing electroencephalogram signal | |
| CN113486752A (en) | Emotion identification method and system based on electrocardiosignals | |
| CN113569805A (en) | Action recognition method and device, electronic equipment and storage medium | |
| CN114170478A (en) | Defect detection and positioning method and system based on cross-image local feature alignment | |
| CN112418166A (en) | Emotion distribution learning method based on multi-mode information | |
| CN111488934A (en) | Brain image data processing method, storage medium, computer device and apparatus | |
| CN108805181B (en) | Image classification device and method based on multi-classification model | |
| CN110192860A (en) | A kind of the Brian Imaging intelligent test analyzing method and system of network-oriented information cognition | |
| CN111310711B (en) | Face image recognition method and system based on two-dimensional singular spectrum analysis and EMD fusion | |
| CN113591582B (en) | Character recognition device and method based on resting state functional magnetic resonance data | |
| CN111414579A (en) | Method and system for acquiring brain area association information based on multi-angle correlation | |
| Li et al. | Improving medical/biological data classification performance by wavelet preprocessing | |
| CN115019396A (en) | Learning state monitoring method, device, equipment and medium | |
| Yavuz et al. | Automatic lipreading with principle component analysis | |
| CN115116117A (en) | Learning input data acquisition method based on multi-mode fusion network | |
| CN110458058B (en) | Expression recognition method and device | |
| CN113255789A (en) | Video quality evaluation method based on confrontation network and multi-tested electroencephalogram signals | |
| Kuzilek et al. | Comparison of jade and canonical correlation analysis for ecg de-noising |
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 |