阅读说明：本技术 一种阿尔茨海默症早期认知功能下降预测方法 (Method for predicting early cognitive function decline of Alzheimer's disease ) 是由 张翼飞 于 2020-07-08 设计创作，主要内容包括：一种阿尔茨海默症早期认知功能下降预测方法,获取被测对象的静息态功能磁共振影像组,对静息态功能磁共振影像组进行预处理,对预处理后的静息态功能磁共振影像组进行脑网络划分,基于静息态全脑灰质功能磁功能共振影像和脑网络划分后的功能磁共振影像组的时间序列,分别构建连接矩阵,基于构建好的连接矩阵进行节点度计算与度分布统计,并进行脑网络度分布的模型拟合,基于训练数据构建预测模型,依据估计出的各个脑网络度分布的模型参数,获得预测认知功能得分并输出。本发明过程简单,易操作,检验结果更加准确,更加客观定量,实用性更高,更加有助于辅助医生临床决策。(A prediction method for early cognitive function decline of Alzheimer's disease comprises the steps of obtaining a resting state functional magnetic resonance image group of a measured object, preprocessing the resting state functional magnetic resonance image group, carrying out brain network division on the preprocessed resting state functional magnetic resonance image group, respectively constructing connection matrixes based on time sequences of resting state full-gray-matter functional magnetic resonance images and the brain network divided functional magnetic resonance image group, carrying out node degree calculation and degree distribution statistics based on the constructed connection matrixes, carrying out model fitting of brain network degree distribution, constructing a prediction model based on training data, and obtaining and outputting predicted cognitive function scores according to estimated model parameters of the brain network degree distribution. The method has the advantages of simple process, easy operation, more accurate test result, more objective quantification and higher practicability, and is more favorable for assisting the clinical decision of doctors.)

1. A method for predicting early cognitive function decline in Alzheimer's disease, comprising the steps of:

step S1, obtaining a resting state functional magnetic resonance image group of the tested object;

step S2, preprocessing the resting state functional magnetic resonance image group;

step S3, brain network division is carried out on the preprocessed resting state functional magnetic resonance image group;

step S4, respectively constructing connection matrixes based on the resting-state full-grey-brain-matter functional magnetic resonance images and the time sequences of the functional magnetic resonance image groups divided by the brain network;

step S5, calculating node degrees and counting degree distribution based on the constructed connection matrix, and performing model fitting of brain network degree distribution;

and step S6, constructing a prediction model based on the training data, and obtaining and outputting a predicted cognitive function score according to the estimated model parameters of each brain network degree distribution.

2. The method of predicting early cognitive decline in alzheimer's disease according to claim 1, wherein said brain network partitioning comprises: the brain template or the independent component analysis method or the prior region information-based method is adopted to divide the brain region into different large-scale brain function network regions.

3. The method according to claim 1, wherein the functional connectivity FC is used to construct a functional connectivity matrix, the constructed functional connectivity matrix is a two-dimensional n × n matrix, where n is the total number of nodes in each network, and the elements in the functional connectivity matrix are the functional connectivity strengths of the corresponding nodes in the matrix rows and columns.

4. The method of predicting early cognitive decline in alzheimer's disease according to claim 1 wherein said node degree calculation is divided into three dimensions:

global degree: degree distribution for the whole brain network, i.e. the whole brain; for a sub-functional network, i.e. the degree distribution of connections between nodes in a large-scale functional sub-network and other nodes around the whole brain;

network internal degree: degree distribution between the nodes in the divided brain function sub-network and other nodes in the network;

network externality: according to the degree distribution between the nodes in the divided brain function sub-networks and the nodes outside the network in the whole brain, namely the degree of the nodes in the network is the global degree of the network minus the degree of the nodes in the network;

and fitting a brain network degree distribution model by adopting Weibull distribution.

5. The method of predicting early cognitive decline in alzheimer's disease according to claim 1, wherein the calculation of the node degree is to calculate the number of edges connecting each node, and with each voxel as a node, the calculation formula of each voxel degree is as follows:

wherein r is_ijRepresenting the Pearson correlation coefficient, N, between voxel i and voxel j_voxelsRepresents the number of voxels within the calculated brain network range, and T is a threshold value.

6. The method of predicting early cognitive decline in alzheimer's disease according to claim 1, wherein said prediction model is:

cognitive score β₀+β₁× Beta parameter + β for Weibull distribution₂× age + β₃× sex + β₄× degree of education;

wherein, β₀Is a constant term, β₁，β₂，β₃，β₄Is the regression coefficient, beta is a parameter of the weibull distribution.

7. A storage device, wherein a plurality of program codes are stored in the storage device, and the program codes are suitable for being loaded and executed by a processor to realize the method for predicting early cognitive decline in alzheimer's disease according to any one of claims 1-6.

8. A processing device comprising a processor and a memory device;

the processor is adapted to execute various program codes;

the storage device stores a plurality of program codes, and the program codes are suitable for being loaded and executed by a processor to realize the early cognitive function decline prediction method of Alzheimer's disease according to any one of claims 1-6.

Technical Field

The invention relates to the field of biomedicine, relates to a graph theory method of a large-scale brain function network, and particularly relates to a prediction method of cognitive function decline of the resting-state large-scale brain function network in the early stage of neuropsychiatric diseases and/or diseases such as Alzheimer's disease from the aspect of network topological organization structure.

Background

Mild Cognitive Impairment (MCI) is a precursor stage between healthy aging and (eventually) alzheimer's disease. Alzheimer's Disease (AD), commonly known as senile dementia, is the most common and important degenerative disease of the brain, mainly manifested as a decline in situational memory, thinking, behavior and daily activities. According to the global Alzheimer's disease Association, the rate of this disease in the elderly aged 65 and over 85 years is as high as about 15% and 37%, and is on a gradual increase and trend towards younger age. The disease is incurable, and is classified as four killers of human health together with cardiovascular diseases, cerebrovascular diseases and cancers. Many clinical drug trials are directed to intervention from early stages of the disease to prevent or slow the disease. Because of the occult occurrence of AD, the current missed diagnosis rate is about 73.1%, and most confirmed patients have reached the middle and later period of the disease. In fact, as early as about 20 years before clinical dementia symptoms, pathogenic Amyloid-beta (Amyloid-beta) has begun to deposit in the patient's brain. Therefore, detection of early stages of AD is a prerequisite for early intervention and is of great importance.

Disclosure of Invention

The invention provides a method for predicting the early cognitive function decline of Alzheimer's disease, which has the advantages of simple process, easy operation, more accurate inspection result, more objective quantification and higher practicability, and is more favorable for assisting the clinical decision of doctors.

In order to achieve the above object, the present invention provides a method for predicting cognitive function decline in an early stage of alzheimer's disease, comprising the steps of:

step S1, obtaining a resting state functional magnetic resonance image group of the tested object;

step S2, preprocessing the resting state functional magnetic resonance image group;

step S3, brain network division is carried out on the preprocessed resting state functional magnetic resonance image group;

step S4, respectively constructing connection matrixes based on the resting-state full-grey-brain-matter functional magnetic resonance images and the time sequences of the functional magnetic resonance image groups divided by the brain network;

step S5, calculating node degrees and counting degree distribution based on the constructed connection matrix, and performing model fitting of brain network degree distribution;

and step S6, constructing a prediction model based on the training data, and obtaining and outputting a predicted cognitive function score according to the estimated model parameters of each brain network degree distribution.

The brain network partition comprises: the brain template or the independent component analysis method or the prior region information-based method is adopted to divide the brain region into different large-scale brain function network regions.

And constructing a functional connection matrix by using the functional connection FC, wherein the constructed functional connection matrix is a two-dimensional matrix of n multiplied by n, n is the total number of nodes of each network, and the elements in the functional connection matrix are the functional connection strength of the nodes corresponding to the rows and the columns of the matrix.

The node degree calculation is divided into three dimensions:

global degree: degree distribution for the whole brain network, i.e. the whole brain; for a sub-functional network, i.e. the degree distribution of connections between nodes in a large-scale functional sub-network and other nodes around the whole brain;

network internal degree: degree distribution between the nodes in the divided brain function sub-network and other nodes in the network;

network externality: according to the degree distribution between the nodes in the divided brain function sub-networks and the nodes outside the network in the whole brain, namely the degree of the nodes in the network is the global degree of the network minus the degree of the nodes in the network;

and fitting a brain network degree distribution model by adopting Weibull distribution.

Calculating the node degree, namely calculating the number of connected edges of each node, and taking each voxel as a node, wherein the calculation formula of each voxel degree is as follows:

wherein r is_ijRepresenting the Pearson correlation coefficient, N, between voxel i and voxel j_voxelsRepresenting the number of voxels in the computed brain networkAnd T is a threshold value.

The prediction model is as follows:

cognitive score β₀+β₁× Beta parameter + β for Weibull distribution₂× age + β₃× sex + β₄× degree of education;

wherein, β₀Is a constant term, β₁，β₂，β₃，β₄Is the regression coefficient, beta is a parameter of the weibull distribution.

The invention also provides a storage device, wherein a plurality of program codes are stored in the storage device, and the program codes are suitable for being loaded and executed by a processor to realize the early cognitive function decline prediction method for the Alzheimer's disease.

The invention also provides a processing device, which comprises a processor and a storage device;

the processor is adapted to execute various program codes;

the storage device stores a plurality of program codes, and the program codes are suitable for being loaded and executed by a processor to realize the early cognitive function decline prediction method for the Alzheimer's disease.

The invention has the beneficial effects that:

1. according to the method, through fitting the network degree distribution of the whole brain network and different large-scale functional sub-networks, the early cognitive function decline prediction of the Alzheimer's disease by resting-state functional magnetic resonance imaging is quickly and conveniently realized by adopting a computer. The method comprises the steps of extracting time sequence signals from preprocessed brain function magnetic resonance data by using a predefined brain template, generating a connection matrix by using methods such as functional connection and the like, calculating network node degree distribution by using the connection matrix, performing model fitting on the degree distribution, and performing effective cognitive function prediction on a group of input function magnetic resonance data of a measured object by using a prediction model. Compared with the conventional biomarker method for predicting the cognitive function (the conventional method is mainly characterized in that the conventional method is connected from a certain function(s), the neural activity of a certain specific brain region(s) is taken as a biomarker and can only reflect local brain change), the method predicts the degree distribution of a whole brain network and various large-scale functional sub-networks, namely the angle of change of the overall organization topological structure of the network, and predicts different dimensions of the global degree, the network internal degree and the network external degree respectively, so that the detection result is more accurate and comprehensive, and the practicability is higher.

2. The invention carries out degree distribution calculation and fitting on the whole brain and large-scale brain function network based on resting state functional magnetic resonance imaging. Compared with other biomarkers or prediction methods, the method for acquiring data through resting state functional magnetic resonance has the advantages that for the aged patients, particularly the patients with cognitive impairment, complex tasks do not need to be completed in a matching mode, the operation is easier, and the clinical practice is realized. Compared with other network construction and degree distribution fitting processes (such as a conditional probability method like a Bayesian network), the method has the advantages of simple process, easy operation and clear meaning of each step. The method can discover the cognitive decline of the patient as early as possible, has more objective and quantitative advantages compared with a questionnaire survey method for the old patient to perceive the cognitive decline by self, and is helpful for assisting the clinical decision of doctors.

Drawings

Fig. 1 is a schematic flow chart of the method for predicting early cognitive decline in alzheimer's disease based on resting-state functional magnetic resonance imaging according to the present invention.

Fig. 2 is a cumulative distribution equation fitting curve of three alternative model distributions, which is an example of the global degree of a measured object of a mild cognitive impairment patient according to an embodiment of the method for predicting early cognitive decline in alzheimer's disease based on a resting-state functional magnetic resonance image.

Fig. 3 is a fitting graph of the cognitive performance score predicted by the global degree fitting weibull distribution beta parameter for the whole brain network and the 7 sub-functional networks according to the method for predicting early cognitive performance decline of alzheimer's disease based on the resting state functional magnetic resonance image.

Detailed Description

The preferred embodiment of the present invention will be described in detail below with reference to fig. 1 to 3. The examples described herein are intended to better illustrate the invention and are not intended to limit the invention.

It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other. The scope of the invention is not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions also falls into the protection scope of the invention.

The resting-state functional Magnetic resonance imaging (rsfMRI) is a non-invasive imaging technology which is simple and easy to operate and does not need to be tested to participate in tasks, and has the advantage of easy clinical practice for old testees with cognitive impairment. The imaging method measures the change of the ratio of oxyhemoglobin to deoxyhemoglobin in local blood of a brain activity area through Blood Oxygen Level Dependence (BOLD) to determine the activity degree of cranial nerves. Functional Connectivity (FC) is the synchronization of spontaneous low-frequency fluctuating brain activity between different brain regions calculated by blood oxygen level dependence. The human brain is a spatially integrated but functionally distributed network with a highly efficient topological organization of many brain regions and neural connections (interactions) between brain regions. By utilizing a mathematical graph theory method, the interaction relationship between cranial nerves can be effectively described: each brain region is a node in the network and the functional connections between brain regions are edges in the network. Studying the topology of brain networks can yield valuable information about the junction area, robustness level and information transfer capabilities of the network. One of the most important attributes describing the network topology is the degree distribution (degree distribution). A degree distribution is a graph-theoretic metric that describes a probability distribution of the number of connections between nodes in a network. In the brain network of rsfMRI, there is currently no consensus as to which model can better fit the fitness distribution.

It is considered that neuropsychiatric diseases including AD may be considered as "dysarthric syndrome". In the precursor stage of AD, various brain functions including memory, attention, executive functions, speech, vision, etc. are affected. In brain networks based on resting state functional magnetic resonance, alterations in neural-coherence connectivity between certain brain regions of MCI patients are found, as well as alterations in brain neural activity in large-scale functional networks. These results indicate that the degree distribution change in the resting brain network is an independent, early brain change, which may have occurred decades before the clinical onset of AD. Since human advanced cognitive function does not involve a single brain function network, the whole brain and several large-scale brain function networks may change during the precursor stage of AD. The existing method is mostly to connect certain (several) cranial nerves, and the nerve activity change of certain (several) specific areas is used as a biomarker. The whole brain network and the sub-function brain network are established by a set of method with clear flow and universal applicability, the degree distribution fitting of the networks is carried out, the network change of the disease in the early stage is measured from the angle of degree distribution-the overall topological organization structure of the network, and the cognitive function decline of the early patient is predicted, so that an effective biomarker of AD early brain network topology can be provided, and technical support is provided for diagnosis and cognitive function prediction.

Aiming at the concealment of the Alzheimer disease for decades, the invention enters the middle and later stages of the disease when the Alzheimer disease has clinical symptoms, and because no cure method exists in the disease at present and only early discovery and early intervention can be relied on, the development of the disease is predicted by a non-invasive neuroimaging means in the early stage of the disease, and related biomarkers are found, which is very important. Conventional biomarkers and imaging procedures via resting-state functional magnetic resonance imaging are limited, and usually only certain pivotal brain region(s) or certain functional connection(s) are used. The invention predicts the cognitive activity decline of patients in early stage of diseases from the change of network topological organization structure by using the graph theory moderate distribution method from the perspective of a whole brain network and a large-scale brain function sub-network from the global brain and the sub-system respectively. Provides support for early diagnosis, intervention and prevention of relevant neuropsychiatric diseases such as Alzheimer's disease, mild cognitive impairment and the like in clinic.

The invention provides a method for predicting early cognitive function decline of Alzheimer's disease based on a resting state functional magnetic resonance image, which is used for constructing a brain network on the basis of the resting state functional magnetic resonance image scanned by mild cognitive impairment crowds at the precursor stage of the Alzheimer's disease according to a connection matrix generated by acquired rsfMRI of a tested object in a whole brain network and different large-scale functional networks. And performing model fitting of degree distribution by calculating the degree distribution of the generated brain network, and outputting the predicted cognitive ability score of the measured object based on the prediction model provided by the invention.

The invention provides a method for predicting early cognitive function decline of Alzheimer's disease based on resting state functional magnetic resonance images, which comprises the following steps:

step S1, obtaining a resting state functional magnetic resonance image of the object to be measured;

the functional magnetic resonance image group comprises functional magnetic resonance images of a plurality of time points.

And step S2, preprocessing the functional magnetic resonance image.

And step S3, performing brain network division on the preprocessed functional magnetic resonance image.

And S4, respectively constructing connection matrixes based on the full gray matter functional magnetic resonance image and the time sequence of the divided functional magnetic resonance image.

And step S5, calculating node degrees and counting degree distribution based on the constructed connection matrix, and performing model fitting of brain network degree distribution.

Step S6, constructing a prediction model based on the training data, and obtaining and outputting a prediction cognitive function score according to the estimated model parameters of each brain network degree distribution;

the prediction model can be a model which is constructed based on a General Linear Model (GLM) and the like and used for predicting the cognitive function score of the measured object corresponding to the brain network degree distribution model parameters.

In order to more clearly describe the method for predicting early cognitive decline of alzheimer's disease based on resting-state functional magnetic resonance images, the following will describe each step in the embodiment of the method of the present invention in detail with reference to fig. 1.

The invention provides a method for predicting early cognitive function decline of Alzheimer' S disease based on a resting state functional magnetic resonance image, which comprises the following steps of S1-S6:

step S1, acquiring a functional magnetic resonance image group of the measured object; the functional magnetic resonance image group comprises functional magnetic resonance images of a plurality of time points.

The resting state functional magnetic resonance imaging is to scan a measured object to obtain a magnetic resonance image without performing any specific task or action while keeping the measured object still, and in the process, the measured object is generally required to be awake and cannot sleep as far as possible without any thing.

In this embodiment, the acquisition of the functional magnetic resonance image is to acquire the brain functional magnetic resonance image based on the blood oxygen level dependence within several minutes or ten and several minutes at certain scanning intervals for the measured object within a certain time period.

And step S2, preprocessing the functional magnetic resonance image.

The acquired original resting-state functional magnetic resonance image (rsfMRI) contains much noise, and in order to make the image have a better signal-to-noise ratio for subsequent calculation, the image is usually preprocessed. The functional magnetic resonance image may be preprocessed using one or more of the following preprocessing methods:

1. the first 10 time points were removed to eliminate machine instability.

2. And (5) correcting the head movement. In a preferred embodiment of the invention, the head movement is greater than 3mm and 3 degrees, and if the head movement is greater than the standard, the image obtained is not used due to poor quality.

3. And (5) correcting the time layer.

4. And (3) carrying out unified segmentation by using T1 or carrying out user-defined segmentation by using a structural magnetic resonance image of the detected object to obtain the segmented gray matter, white matter and cerebrospinal fluid template.

5. Registration to the standard space. A preferred embodiment of the invention registers to the montreal normalized space (MNI space) and resamples the image to voxels (voxels) 3mm x 3 mm.

6. And removing the linear drift.

7. And (6) regression covariates. In a preferred embodiment of the invention, regression is performed on 24-directional cranial parameters as well as global, white matter and cerebrospinal fluid signals.

8. And (6) filtering. In a preferred embodiment of the invention, frequencies other than 0.01-0.1Hz in the image are filtered.

Step S3: and carrying out brain network division on the preprocessed functional magnetic resonance image.

The brain regions can be divided into large-scale brain network regions with different functions by adopting a brain template or other methods such as independent component analysis, prior region information-based methods and the like.

In a preferred embodiment of the present invention, the brain network is divided by using an existing brain template, and the whole brain is divided into 7 sub-function networks, including a Visual network (V), a body motion network (SM), a Dorsal Attention network (DA), a Ventral Attention network (VA), a marginal network (Limbic, Lim), a frontier network (FP), and a Default network (Default Mode, DM), where the brain template of the 7 networks is obtained by a clustering method based on 1000 data-driven tests.

Step S4: and respectively constructing connection matrixes based on the full-gray-matter functional magnetic resonance image and the time sequence of the divided functional magnetic resonance image.

Obtaining the whole brain gray matter functional magnetic resonance image by using the intersection of the segmented gray matter template obtained in the preprocessing process and the whole brain functional magnetic resonance image; the constructed connection matrix is a two-dimensional matrix of n × n, and a Functional Connection (FC) is used to construct the functional connection matrix in a preferred embodiment of the invention;

where n is the total number of nodes per network. The selection of the whole brain and the divided sub-function network nodes can select brain areas with different scales as the nodes.

In a preferred embodiment of the present invention, each voxel is used as a node, and the maximum resolution (i.e., voxel) obtained by the functional magnetic resonance image scan is used as a node unit, so that the result is more accurate. The voxels may also be reduced in dimension, and the values in a region may be averaged by resampling to obtain nodes of larger scale, which may reduce computation time but may reduce accuracy.

In a preferred embodiment of the invention, the elements in the connection matrix are the functional connection strengths of the nodes corresponding to the rows and columns of the matrix. The functional connection is calculated by a pearson correlation coefficient metric, the formula of which is as follows:

where x, y are the time series of two nodes, s_x，s_yAs standard deviation, cov_xyIs the covariance of the two time series. Other methods such as active connections may also be used to construct the connection matrix.

Step S5: and calculating node degrees and counting degree distribution based on the constructed connection matrix, and fitting a correlation model of the degree distribution.

Based on the connection matrix (i.e., the information of the corresponding edge in the network) in step S4, it is determined whether there is an edge between the node pairs of the constructed network, and the number of edges of each node in the network is counted.

The element cards in the connection matrix can be thresholded by adopting a threshold value method, the statistics above the threshold value are edges, and the statistics below the threshold value are not carried out. In a preferred embodiment of the invention, the network is constructed using a binary method, i.e. 1 edge is considered to exist by the threshold value and no edge is considered to exist by the threshold value. The threshold value may be selected empirically based on data analysis, typically ranging from 0.3 to 0.7, or may be selected by other methods. In a preferred embodiment of the invention, the threshold value is empirically selected to be 0.4 or 0.5 for the connectivity and the small world property index of the constructed network, and only the positive value elements in the functional connection matrix are clipped, and the negative value elements are discarded.

In a preferred embodiment of the present invention, each voxel is used as a node, and a calculation formula of each voxel degree is as follows:

wherein r is_ijRepresenting the Pearson correlation coefficient, N, between voxel i and voxel j_voxelsRepresents the number of voxels within the calculated brain network range, and T is a threshold value.

The calculation of the node degree is divided into three dimensions. (1) Global degree: degree distribution for the whole brain network, i.e. the whole brain; for a sub-functional network, i.e. the degree distribution of connections between nodes in a large-scale functional sub-network and other nodes around the whole brain; (2) network internal degree: degree distribution between the nodes in the divided brain function sub-network and other nodes in the network; (3) network externality: and according to the degree distribution between the nodes in the divided brain function sub-network and the nodes outside the network (but inside the whole brain), namely the degree of the nodes in the network is the global degree of the network minus the degree of the nodes in the network.

In a preferred embodiment of the present invention, in the process of calculating the node degree, since the voxels are used as the node scale, erroneous counting of the shared signal between adjacent voxels is avoided, and some adjacent edges are excluded by calculating the spatial euclidean distance between node pairs. If the euclidean distance between the node pair center coordinates is less than a certain value (20 mm is used in this embodiment), the degree between them is not calculated. In this embodiment, voxels are used as a node scale, and since there are many nodes, in order to save calculation time and storage space, the correlation matrix constructed in step S4 is not saved, and the correlation result is directly used for calculating and counting the node degree distribution in step S5.

And (3) counting the network degree distribution, namely sequencing all the calculated node degrees in the network, counting by a probability distribution method in probability statistics, and performing subsequent model fitting.

According to different construction methods adopted for nodes and edges during network construction, different models can be used for parameter fitting, and screening and comparing processes for the models need to be added in the process. The model parameter estimation method may be selected as desired. In a preferred embodiment of the present invention, a Maximum Likelihood Estimation (MLE) method is used to fit the three most likely candidate models; comparing the fitting degree between different alternative models by adopting the standardized log-likelihood ratio and the related p value thereof; and fitting and comparing the sample Distribution and a theoretical Distribution curve by adopting a complementary cumulative Distribution equation (CCDF) to obtain a Weibull Distribution (Weibull Distribution) as an optimal model.

Step S6: constructing a prediction model based on training data, and obtaining and outputting a predicted cognitive function score according to the estimated parameters of each brain network degree distribution model; the prediction model can be a model which is constructed based on methods such as a General Linear Model (GLM) and the like and is used for predicting the cognitive function scores of the tested object corresponding to the degree distribution model parameters.

In a preferred embodiment of the present invention, the general linear model method used in step S6 to construct the prediction model from the training data of 30 subjects is as follows:

cognitive score β₀+β₁× Beta parameter + β for Weibull distribution₂× age + β₃× sex + β₄× educational level.

The construction and validation of the model for predicting early cognitive decline in alzheimer's disease according to the present invention will be described with reference to examples. In this example, the functional magnetic resonance image of the sample can be processed through steps 1-6.

Step 1, acquiring a resting state functional magnetic resonance image group of 30 patients with Mild Cognitive Impairment (MCI) and basic information of the resting state functional magnetic resonance image group, the gender, the age, the education level, the cognitive function and the like.

The scanning process of the embodiment is as follows: each object is scanned for 480 seconds by 960 pieces of functional magnetic resonance images. Using a multiband scan sequence, the layer acceleration factor is 4, the repetition Time (TR)/echo Time (TE) is 500/30ms, and the Flip Angle (FA) is 47Angle, field of view (FOV) 231 × 231mm2, matrix 66 × 66, slice 36, thickness 3.5mm, voxel size 3.5 × 3.5.5 3.5 × 3.5.5 mm³The echo spacing is 0.4ms, and the bandwidth is 3444 Hz/pixel. The tested object is informed of eye closing scanning, does not want anything and does not fall asleep as much as possible, and the actions of the head and limbs are reduced as much as possible.

And 2, preprocessing the functional magnetic resonance image of each measured object by a Matlab-based brain image processing software SPM (statistical Parameter mapping) kit to remove noise and increase the signal-to-noise ratio. The pretreatment process comprises the following steps: removing functional MRI images of the first 10 time points; aligning the rest images to the first image; head movement correction, wherein the head movement is eliminated by 3 degrees beyond 3 millimeters; and (4) dividing the functional magnetic resonance image by T1 unified division to obtain the divided gray matter, white matter and cerebrospinal fluid template. Segmented gray, white and cerebrospinal fluid images; registration to montreal normalized space (MNI space) and resampling the image to voxels 3mm x 3 mm; removing linear drift; regression covariates, which carry out regression on the head movement parameters in 24 directions, the global signals, the white matter signals and the cerebrospinal fluid signals; and filtering, namely filtering frequencies except 0.01-0.1Hz in the image. This embodiment has no preprocessing step for temporal layer correction because it employs a multiband (multiband) scan sequence.

And 3, respectively carrying out brain network division on the obtained preprocessed functional magnetic resonance images to obtain brain areas divided into large-scale functional sub-networks.

The brain template can be used for partitioning, and the brain template can be divided into large-scale brain network areas with different functions. In this embodiment, a seven-network resting brain function template of Yeo is adopted to divide a brain network, and a whole brain is divided into 7 sub-function networks, including a visual network (V), a body movement network (SM), a dorsal attention network (DA), a ventral attention network (VA), a border network (Lim), a forehead network (FP) and a default network (DM).

And 4, step 4: and respectively constructing connection matrixes based on the full-gray-matter functional magnetic resonance image and the time sequence of the divided functional magnetic resonance image.

In the embodiment, a voxel is used as a node scale, a constructed connection matrix is a functional connection matrix, and the size of the connection matrix is a two-dimensional matrix of n multiplied by n; in a whole brain network, n is 47294. The whole brain network and the 7 sub-functional networks of this embodiment both intersect with the gray matter template, i.e. only voxels in gray matter are considered.

The elements in the connection matrix in this embodiment are the functional connection strengths of the corresponding nodes of the rows and columns of the matrix, and the functional connection is calculated by the pearson correlation coefficient measurement.

And 5: and carrying out node degree calculation and degree distribution statistics and carrying out relevant model fitting of degree distribution.

The degrees (i.e., the number of edges) of each node in the network are counted based on step 4. And adopting a threshold value method to threshold the element cards in the connection matrix. In this embodiment, the threshold is selected to be 0.4, the threshold is selected for the positive value element card in the functional connection matrix, and the negative value element is discarded.

And calculating degree distribution of three dimensions of the whole brain network and the 7 sub-function networks, namely, the global degree, the network internal degree and the network external degree. And (4) calculating the space Euclidean distance between the node pairs, and if the Euclidean distance between the node pairs and the central coordinates is less than 20mm, not calculating the degree between the node pairs and the central coordinates. In this embodiment, three most commonly used models in a brain network are selected for comparison and screening, namely power law distribution (power law), power law distribution with exponential truncation (power law with exponential cutoff) and Weibull distribution (Weibull distribution), and the fitting degree of the normalized log-likelihood ratio and the related p value thereof are compared among different candidate distributions, and the comparison result is shown in table 1. The probability density functions of the three alternative models and the parameters to be estimated are shown in table 2.

TABLE 1

Table 1 shows the results of comparing and screening the models by selecting the three most commonly used models, i.e., power law distribution, exponential truncated power law distribution, and weibull distribution, and fitting the three candidate distributions by using the normalized log-likelihood ratios and their associated p values. The elements in the table represent the values of the normalized log-likelihood ratios of pairwise comparison of the alternative distributions, the positive value represents that the Weibull distribution has better fitting effect, and the negative value represents that the alternative model has better fitting effect. It can be seen that the values are all positive values, i.e. the weibull distribution fitting effect is better.

TABLE 2

Table 2 shows the probability density functions and their corresponding parameters to be estimated for the three most common models.

Fitting and comparing the sample distribution and the theoretical distribution curve by adopting a complementary cumulative distribution equation, taking the global degree of a measured object as an example, the fitting result is shown in figure 2, and the fitting degree of the Weibull distribution in all networks can be seen to be better; and performing parameter estimation on the Weibull distribution by adopting maximum likelihood estimation.

The calculation result shows that the fitting effect of the Weibull distribution on the degree distribution (the global degree, the network internal degree and the network external degree) of the three dimensions of the whole brain and each sub-function network is better than that of other two alternative distributions in the embodiment.

Step 6: and (5) constructing a prediction model according to the Weibull distribution parameters estimated in the step (5), and checking the significance of the hypothesis model.

The model for the cognitive function score of the tested object corresponding to the predictive degree distribution model parameters, which is constructed based on the general linear model method, is as follows (wherein the age and the educational degree are both accurate to the number of years):

cognitive score β₀+β₁× Weibull distribution parameter + β₂× age + β₃× sex + β₄× educational level.

Taking the global degree as an example, the model result shows that, after multiple comparison and correction of False Discovery Rate (FDR), for the parameter beta (β) of weibull distribution, the results of the whole brain network and the rest 6 networks are significant (the corrected p value is less than 0.05) except for the ventral attention network, the corrected p value of the ventral attention network is 0.1 (also has a certain prediction trend), that is, the cognitive ability score can be predicted by the parameter beta (β) of weibull distribution, and the prediction results of the whole brain network and each sub-function network are shown in fig. 3.

And constructing a prediction model through training data, and obtaining and outputting a predicted cognitive function score according to the parameters of each brain network degree distribution model estimated by the measured object.

The invention can be used for predicting cognitive function decline in other mental or neurological diseases including Alzheimer's disease and in early stage of the diseases.

It should be noted that, for brevity and convenience of description, for those skilled in the art to understand, the above-mentioned specific working processes and related descriptions of the data storage device and the processing device may refer to corresponding processes in the foregoing method embodiments, and therefore, the descriptions are omitted here.

Those of skill in the art will appreciate that the various illustrative logical blocks, methods, steps, and modules described in connection with the embodiments disclosed herein may be implemented as computer software, electronic hardware, or combinations of both. Program code corresponding to the associated software modules, methods, and steps may be stored in a memory including read only memory ROM, random access memory RAM, electrically programmable ROM, electrically erasable programmable ROM, hard disk, removable hard disk, optical disk, registers, floppy disk, magnetic tape, or any other form of storage medium known in the art. In the foregoing description, the steps and components of the various examples have been described generally in terms of functionality to clearly illustrate the interchangeability of software and electronic hardware. The aforementioned functions may be implemented or realized in software or electronic hardware, depending on the respective technical solution and the specific application scenario and design requirement constraints. Those skilled in the art will be able to select different methods for implementing the described functionality according to the specific requirements of the application scenario, but such implementation should not be considered as exceeding the scope of the present invention.

The terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a method, process, article, apparatus, or device that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such method, process, article, apparatus, or device.

The invention has the beneficial effects that:

1. according to the method, through fitting the network degree distribution of the whole brain network and different large-scale functional sub-networks, the early cognitive function decline prediction of the Alzheimer's disease by resting-state functional magnetic resonance imaging is quickly and conveniently realized by adopting a computer. The method comprises the steps of extracting time sequence signals from preprocessed brain functional magnetic resonance data by using a predefined brain template, generating a connection matrix by using methods such as a Pearson correlation coefficient and the like, calculating node degree distribution in a network by using the connection matrix, performing model fitting on the degree distribution, and performing effective cognitive function prediction on a group of input functional magnetic resonance data of a tested object by using a prediction model. Compared with the prior biomarker method for predicting cognitive functions (the prior method is mostly connected from a certain number of lines, the nerve activity of a certain specific brain area(s) is taken as a biomarker and only reflects local change of the brain), the method predicts the overall organization topological structure of the network through the degree distribution of the whole brain network and different large-scale functional sub-networks. And through predicting three different dimensions of the global degree, the network internal degree and the network external degree, the detection result is more accurate and comprehensive, and the practicability is higher.

2. The invention carries out degree distribution calculation and fitting on the whole brain network and different large-scale brain function networks based on the resting state functional magnetic resonance imaging. Compared with other biomarkers or prediction methods, the method for acquiring data through resting state functional magnetic resonance has the advantages that for the aged patients, particularly the patients with cognitive impairment, complex tasks do not need to be completed in a matching mode, the operation is easier, and the clinical practice is realized. Compared with other network construction and degree distribution fitting processes (such as a conditional probability method like a Bayesian network), the method has the advantages of simple process, easy operation and clear meaning of each step. The method can discover the cognitive decline of the patient as early as possible, has more objective and quantitative advantages compared with a questionnaire survey method for the old patient to perceive the cognitive decline by self, and is helpful for assisting the clinical decision of doctors.

While the present invention has been described in detail with reference to the preferred embodiments, it should be understood that the above description should not be taken as limiting the invention. Various modifications and alterations to this invention will become apparent to those skilled in the art upon reading the foregoing description. Accordingly, the scope of the invention should be determined from the following claims.