CN113724863B - Automatic discrimination system, storage medium and equipment for autism spectrum disorder - Google Patents

Abstract

The invention belongs to the field of computer-aided diagnosis based on medical images and genetics, and provides an automatic discrimination system, a storage medium and equipment for autism spectrum disorder. The system comprises a brain connection feature extraction module, a brain connection feature extraction module and a brain white matter connection feature extraction module, wherein the brain connection feature extraction module is used for extracting a function connection map and a brain white matter connection map from a tested multi-modal magnetic resonance image to serve as brain connection features; the characteristic selection and dimension reduction module is used for carrying out sparse selection and nonlinear dimension reduction on brain connection characteristics; the genetic data characteristic extraction module is used for extracting genetic data characteristics by utilizing a gene constraint maximum likelihood method based on the tested whole genome sequencing data and the quality screening of single nucleotide polymorphism; the characteristic fusion module is used for fusing the selected brain connection characteristic and the genetic data characteristic after dimension reduction by using a typical correlation analysis method; and the classification decision module is used for inputting the fused features into the decision model and judging whether the fused features belong to the autism spectrum disorder category.

Description

Automatic discrimination system, storage medium and equipment for autism spectrum disorder

Technical Field

The invention belongs to the field of computer-aided diagnosis based on medical images and genetics, and particularly relates to an automatic discrimination system, a storage medium and equipment for autism spectrum disorder.

Background

The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.

Autism Spectrum Disorder (ASD) is caused by generalized mental development Disorder in infancy, and is mainly manifested by abnormalities in interpersonal communication and communication patterns, language and non-language communication disorders, limited interest and activity contents, stereotyped behaviors and repetition. An accurate and objective early ASD discrimination method is found, and the method has important significance for assisting the intelligent diagnosis of clinical ASD and formulating reasonable early nerve protection measures.

Some studies have initially discussed diagnostic studies of ASD based magnetic resonance imaging. However, most studies are based on single-mode brain image characteristics, such as structural Magnetic Resonance Imaging (MRI) based on which phenotypic changes such as brain cortical structural volume, surface area, etc. can be obtained; functional magnetic resonance imaging (fMRI) based functional MRI can reflect abnormalities of local brain area functional connection and metabolic degree; based on diffusion magnetic resonance imaging (dMRI), the microstructure change of the white brain fiber bundle can be reflected by quantitatively analyzing the free diffusion property of brain tissue water molecules. The inventor finds that the existing autism spectrum disorder diagnosis has the problems of single analysis characteristic and low accuracy, and cannot provide reliable auxiliary diagnosis for early ASD diagnosis.

Disclosure of Invention

In order to solve the technical problems in the background art, the present invention provides an automatic determination system for autism spectrum disorder, which can improve the classification accuracy.

In order to achieve the purpose, the invention adopts the following technical scheme:

a first aspect of the present invention provides an automatic discrimination system for autism spectrum disorder, comprising:

the brain connection feature extraction module is used for extracting a function connection map and a white matter connection map from a tested multi-modal magnetic resonance image as brain connection features;

the characteristic selection and dimension reduction module is used for carrying out sparse selection and nonlinear dimension reduction on brain connection characteristics;

the genetic data feature extraction module is used for extracting genetic data features by utilizing a gene constraint maximum likelihood method based on the tested whole genome sequencing data and the quality screening of the single nucleotide polymorphism;

a feature fusion module for fusing the selected and dimensionality reduced brain junction features and genetic data features using a canonical correlation analysis method;

and the classification decision module is used for inputting the fused features into the decision model and judging whether the features belong to the autism spectrum disorder category.

Furthermore, the decision model is formed by combining a plurality of base classifiers, each base classifier corresponds to the input feature vector to obtain the prediction probability output value of the base classifier of the test sample, and the prediction probability output value is used as the input of the logistic regression model of the next stage to train the decision model.

Further, the base classifier includes SVM, RF and BP networks.

Further, the multi-modality magnetic resonance image includes an srmri image, an fMRI image, and a dMRI image.

Further, sparse selection is carried out on brain connection features by utilizing L1 norm regularization.

Further, nonlinear dimension reduction is carried out by using a t-SNE algorithm.

Further, the extracted genetic data features include two features, the first feature is the degree of inheritance, and the second feature is the genetic correlation of white matter connection/functional connection and the behavioral score of autism.

Further, the feature fusion module is further configured to fuse the obtained different feature vectors by using canonical correlation analysis, so as to implement an optimal feature set.

A second aspect of the invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of:

extracting a functional connection and white matter connection map from a tested multi-modal magnetic resonance image to be used as a brain connection characteristic;

performing sparse selection and nonlinear dimensionality reduction on brain connection features;

based on the tested whole genome sequencing data and the quality screening of single nucleotide polymorphism, extracting the characteristics of genetic data by using a gene constraint maximum likelihood method;

fusing the selected brain connection characteristics and the genetic data characteristics after dimensionality reduction by using a typical correlation analysis method;

inputting the fused features into a decision model, and judging whether the features belong to the autism spectrum disorder category.

A third aspect of the present invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the program:

extracting a functional connection and white matter connection map from a tested multi-modal magnetic resonance image as a brain connection characteristic;

performing sparse selection and nonlinear dimensionality reduction on brain connection features;

based on the tested whole genome sequencing data and the quality screening of single nucleotide polymorphism, extracting the characteristics of genetic data by using a gene constraint maximum likelihood method;

fusing the selected brain connection characteristics and the genetic data characteristics after dimensionality reduction by using a typical correlation analysis method;

inputting the fused features into a decision model, and judging whether the features belong to the autism spectrum disorder category.

Compared with the prior art, the invention has the beneficial effects that:

aiming at the limitation that a magnetic resonance image is based on that a single-modal brain phenotype characteristic is adopted to classify autism, brain connection characteristics are extracted from the angle of a multi-modal brain network, the selected brain connection characteristics after dimension reduction are fused by a typical correlation analysis method, genetic information is fused, and the classification accuracy is improved; aiming at the high-dimensional characteristics of brain connection and the particularity of genetic characteristics, a good learning effect is realized by means of characteristic selection, dimension reduction and fusion and a model fusion strategy, and the discrimination precision of autism spectrum disorders is improved.

Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are included to illustrate an exemplary embodiment of the invention and not to limit the invention.

FIG. 1 is a block diagram of an automatic determination system for autism spectrum disorders, in accordance with an embodiment of the present invention;

fig. 2 is a schematic diagram of automatic determination of autism spectrum disorder according to an embodiment of the present invention.

Detailed Description

The invention is further described with reference to the following figures and examples.

It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention. As used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.

Example one

Referring to fig. 1, the automatic determination system for autism spectrum disorder of the present embodiment includes:

(1) And the brain connection feature extraction module is used for extracting the functional connection and the white matter connection map from the multi-modal magnetic resonance image to be tested as the brain connection features.

Wherein, the tested patients with autism spectrum disorder and normal control population from ABIDE I/II database obtain sMRI image, fMRI image and dMRI image of the tested patients.

In one or more embodiments, the system for automatically discriminating autism spectrum disorder further includes an image preprocessing module for preprocessing the multi-modal magnetic resonance image under test.

The pretreatment process of each modal characteristic is as follows:

wherein the sMRI is performed by (1) manually adjusting the origin of all sMRI to the anterior union point; (2) and (4) segmenting the MRI image to obtain gray matter, white matter and cerebrospinal fluid.

Performing (1) brain parenchyma extraction on dMRI; (2) eddy current and head motion correction to reduce the influence of blood flow and scanning head motion on the image; (3) correcting the gradient direction; (4) and (5) fitting a tensor model to obtain an anisotropy fraction diagram of the reaction water molecules and other dispersion parameters.

For fMRI: (1) format conversion and time point removal; (2) cephalic motion and temporal compartment correction; (3) spatial normalization and smoothing filtering.

Specifically, in the brain connection feature extraction module, the brain region is first defined: the left and right gray matter hemispheres are each segmented into 123 brain regions using a connectivity-based symmetric BNA246 template, the individual T1 image is then linearly aligned with the individual FA image, and the two inverse transformations are finally applied to the template in MNI space. And extracting the white matter connection characteristics, namely randomly selecting seed points from a gray matter/white matter interface by considering anatomical prior information, and performing white matter fiber tracking by using a second-order integral algorithm (iFOD 2) to further construct the white matter connection characteristics. And finally, obtaining the whole brain function connection characteristic based on the Pearson correlation of the BOLD signal of the brain area.

(2) And the feature selection and dimension reduction module is used for performing sparse selection and nonlinear dimension reduction on the brain connection features.

In the feature selection and dimension reduction module, for the brain white matter connection and brain function connection features, feature selection is firstly carried out, L1 norm regularization is mainly utilized, sparse selection is carried out on the connection features, namely, the size of a regularization term is controlled by introducing a hyper-parameter lambda, and a final loss function L is calculated as follows:

where L (W) is the loss of the unregulated term.

Wherein n is the number of features; w is a _i Representing the ith characteristic loss value; w denotes all feature matrices.

Based on the selected brain connection characteristics, carrying out nonlinear dimensionality reduction by using a t-SNE algorithm which is an algorithm for mining high-dimensional data, namely identifying observed clusters by using the similarity of data points with a plurality of characteristics to find a mode in the data so as to remove redundant characteristic information, namely searching for an optimal Gaussian standard deviation by using binary search and calculating a probability matrix; initializing a low-dimensional space by using normal distribution; and finally, gradually iterating, and calculating the probability matrix and the loss through a low-dimensional space to enable the new probability matrix of the two-dimensional space to be as close to the original high-dimensional space as possible.

(3) And the genetic data feature extraction module is used for extracting the genetic data features by utilizing a gene constraint maximum likelihood method based on the tested whole genome sequencing data and the quality screening of the single nucleotide polymorphism.

In a genetic data feature extraction module, firstly, carrying out quality screening on SNP based on the full genome data of an iPSYCH-PGC database, and mainly comprising the following steps:

knock out minimum allele frequency<0.01, genotyping Rate<90% and failed the Hudi-Winberg test (p)>1×10 ^-7 ) And filtering the interpolated information score<0.8 SNP.

Further based on the screened white matter connection and functional connection characteristics as target parameters, combining with SNP, estimating the connection characteristic variation proportion of all autosomal SNP interpretations by utilizing a genetic relationship matrix and a gene-constrained maximum likelihood method-based analysis model, namely acquiring the first genetic characteristic-heritability of brain connection information;

and finally, utilizing a bivariate gene constraint maximum likelihood model, namely utilizing the genetic covariance of the bivariate and dividing the genetic covariance by the square root of the product of the genetic variances of the bivariate and further estimating the genetic correlation between each brain white matter connection/functional connection and the behavior score of the ASD, wherein the genetic correlation is used as a second genetic characteristic of the brain connection information.

(4) And the characteristic fusion module is used for fusing the selected brain connection characteristic and the genetic data characteristic after dimension reduction by using a typical correlation analysis method.

In the feature fusion module, the obtained different feature vectors are fused by using typical correlation analysis, so that an optimal feature set is realized, and the reliability of target discrimination is improved.

Firstly, obtaining a feature vector of each mode feature, namely a brain white matter connection feature vector WC, a brain function connection feature vector FC, and a inheritance degree feature vector h based on brain white matter connection ² _WC And a inheritance degree characteristic vector h based on brain function connection ² _FC Genetic related feature vector r of brain white matter connection and autism behavior score _gWC And genetic correlation feature vector r of brain function connection and autism behavior score _gFC. 。

Based on the typical correlation analysis algorithm criterion, obtaining the optimal projection vector matrix W of each vector ₁ 、W ₂ 、……、W ₆ Further through Z1 _d ＝W ₁ WC，Z2 _d ＝W ₂ FC，......，Z6 _d ＝W ₆ r _gFC Obtaining d-dimensional vector { Z1 before each feature is fused _d ，Z2 _d ，Z3 _d ，Z4 _d ，Z5 _d ，Z6 _d }; finally using Zf ₁ ＝Z1 _d +Z2 _d +Z3 _d +Z4 _d +Z5 _d +Z6 _d And weighting and summing to obtain the optimal fusion feature vector.

(5) And the classification decision module is used for inputting the fused features into the decision model and judging whether the fused features belong to the autism spectrum disorder category.

The decision model is formed by combining a plurality of base classifiers, each base classifier corresponds to an input feature vector to obtain a prediction probability output value of the base classifier of the test sample, and the prediction probability output value is used as the input of a logistic regression model of the next stage to train the decision model.

For example: the decision model uses SVM, RF and BP network as a base classifier, inputs the characteristic vector to obtain the prediction probability output value of the base classifier of the test sample, and is used as the input of the logistic regression model of the next stage for training; and meanwhile, outputting the verification set based on the base classifier as a test set of the next stage, thereby finally obtaining a classification result.

Specifically, original data are divided into training sets and verification sets according to label categories, an SVM (support vector machine), an RF (radio frequency) network and a BP (back propagation) network are selected as a first level, 5-fold cross validation is adopted for the training sets, four-fold cross validation is firstly adopted as training data, and the other one-fold cross validation is adopted as test data, so that 5 discrimination probability values A1, a2, a3, a4 and a5 trained by each base classifier and probability values B1, B2, B3, B4 and B5 on the verification sets are obtained, wherein the probability value of five training sets is combined to be A1, and the probability value on the average verification set is B1.

After the three base classifiers are trained, a two-stage classifier logistic regression model is used for training the obtained 3 features A1, A2 and A3, and classification is carried out on the obtained features B1, B2 and B3 on the verification set, so as to obtain the final discrimination category.

Example two

The present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of:

extracting a functional connection and white matter connection map from a tested multi-modal magnetic resonance image to be used as a brain connection characteristic;

performing sparse selection and nonlinear dimension reduction on brain connection characteristics;

based on the tested whole genome sequencing data and single nucleotide polymorphism quality screening, extracting genetic data characteristics by using a gene constraint maximum likelihood method;

fusing the selected brain connection characteristics and the genetic data characteristics after dimensionality reduction by using a typical correlation analysis method;

inputting the fused features into a decision model, and judging whether the features belong to the autism spectrum disorder category.

The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.

It should be noted that, each step executed in the processor is the same as the specific implementation process of each module in the first embodiment, and will not be described again here.

EXAMPLE III

The embodiment provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the following steps:

extracting a functional connection and white matter connection map from a tested multi-modal magnetic resonance image to be used as a brain connection characteristic;

performing sparse selection and nonlinear dimension reduction on brain connection characteristics;

based on the tested whole genome sequencing data and single nucleotide polymorphism quality screening, extracting genetic data characteristics by using a gene constraint maximum likelihood method;

fusing the selected brain connection characteristics and the genetic data characteristics after dimensionality reduction by using a typical correlation analysis method;

and inputting the fused features into a decision model to judge whether the features belong to the autism spectrum disorder category.

It should be noted that, the steps executed in the processor are the same as the specific implementation process of the modules in the first embodiment, and are not described here again.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (4)

1. An automatic discrimination system for autism spectrum disorder, comprising:

the brain connection feature extraction module is used for extracting a function connection and white matter connection map from a tested multi-modal magnetic resonance image as a brain connection feature; the multi-modality magnetic resonance image comprises an sMRI image, an fMRI image and a dMRI image;

the characteristic selection and dimension reduction module is used for carrying out sparse selection and nonlinear dimension reduction on brain connection characteristics; sparse selection is carried out on brain connection characteristics by utilizing L1 norm regularization, and nonlinear dimensionality reduction is carried out by utilizing a t-SNE algorithm;

the genetic data feature extraction module is used for performing quality screening based on the tested whole genome sequencing data and single nucleotide polymorphism and extracting the genetic data features by utilizing a gene constraint maximum likelihood method; wherein the genetic data features comprise two features, the first feature being the degree of inheritance and the second feature being the genetic association of a white matter junction or functional junction with the behavioral score of autism;

a feature fusion module for fusing the selected and dimensionality reduced brain junction features and genetic data features using a canonical correlation analysis method; fusing the obtained different feature vectors by utilizing typical correlation analysis so as to realize an optimal feature set; the feature vector comprises a brain white matter connection feature vector, a brain function connection feature vector, a inheritance degree feature vector based on brain white matter connection, a inheritance degree feature vector based on brain function connection, a inheritance related feature vector of brain white matter connection and autism behavior score, and a inheritance related feature vector of brain function connection and autism behavior score;

the classification decision module is used for inputting the fused features into a decision model and judging whether the fused features belong to the autism spectrum disorder category or not; the decision model is formed by combining a plurality of base classifiers, each base classifier corresponds to an input feature vector to obtain a prediction probability output value of the base classifier of the test sample, and the prediction probability output value is used as the input of a logistic regression model of the next stage to train the decision model.

2. The system for automated discrimination of autism spectrum disorder of claim 1, wherein the base classifier comprises SVM, RF and BP networks.

3. A computer-readable storage medium on which a computer program is stored, the program, when executed by a processor, performing the steps of:

extracting a functional connection and white matter connection map from a tested multi-modal magnetic resonance image to be used as a brain connection characteristic; the multi-modality magnetic resonance image comprises an sMRI image, an fMRI image and a dMRI image;

performing sparse selection and nonlinear dimension reduction on brain connection characteristics; sparse selection is carried out on brain connection characteristics by utilizing L1 norm regularization, and nonlinear dimensionality reduction is carried out by utilizing a t-SNE algorithm;

performing quality screening based on the tested whole genome sequencing data and single nucleotide polymorphism, and extracting genetic data characteristics by using a gene constraint maximum likelihood method; wherein the genetic data characteristics comprise two characteristics, the first characteristic is degree of inheritance, and the second characteristic is genetic correlation between white matter connection or functional connection and behavioral score of autism;

fusing the selected brain connection characteristics and the genetic data characteristics after dimensionality reduction by using a typical correlation analysis method; fusing the obtained different feature vectors by utilizing typical correlation analysis so as to realize an optimal feature set; the feature vector comprises a brain white matter connection feature vector, a brain function connection feature vector, a inheritance degree feature vector based on brain white matter connection, a inheritance degree feature vector based on brain function connection, a inheritance related feature vector of brain white matter connection and autism behavior score, and a inheritance related feature vector of brain function connection and autism behavior score;

inputting the fused features into a decision model, and judging whether the features belong to the autism spectrum disorder category or not; the decision model is formed by combining a plurality of base classifiers, each base classifier corresponds to an input feature vector to obtain a prediction probability output value of the base classifier of the test sample, and the prediction probability output value is used as the input of a logistic regression model of the next stage to train the decision model.

4. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program performs the steps of:

extracting a functional connection and white matter connection map from a tested multi-modal magnetic resonance image as a brain connection characteristic; the multi-modality magnetic resonance image comprises an sMRI image, an fMRI image and a dMRI image;

performing sparse selection and nonlinear dimensionality reduction on brain connection features; sparse selection is carried out on brain connection characteristics by utilizing L1 norm regularization, and nonlinear dimensionality reduction is carried out by utilizing a t-SNE algorithm;

performing quality screening based on the tested whole genome sequencing data and single nucleotide polymorphism, and extracting genetic data characteristics by using a gene constraint maximum likelihood method; wherein the genetic data features comprise two features, the first feature being the degree of inheritance and the second feature being the genetic association of a white matter junction or functional junction with the behavioral score of autism;

fusing the selected brain connection characteristics and the genetic data characteristics after dimensionality reduction by using a typical correlation analysis method; fusing the obtained different feature vectors by utilizing typical correlation analysis so as to realize an optimal feature set; the feature vector comprises a brain white matter connection feature vector, a brain function connection feature vector, a hereditary degree feature vector based on brain white matter connection, a hereditary degree feature vector based on brain function connection, a hereditary correlation feature vector of brain white matter connection and autism behavior score, and a hereditary correlation feature vector of brain function connection and autism behavior score;

inputting the fused features into a decision model, and judging whether the features belong to the autism spectrum disorder category or not; the decision model is formed by combining a plurality of base classifiers, each base classifier corresponds to an input feature vector to obtain a prediction probability output value of the base classifier of the test sample, and the prediction probability output value is used as the input of a logistic regression model of the next stage to train the decision model.

Images