CN116309336B - Extraction method of key image marker of vascular cognitive impairment - Google Patents

Abstract

The present application provides a method for extracting key imaging markers of vascular cognitive impairment, the method comprising: 1) obtaining resting-state functional magnetic resonance imaging data and magnetic resonance diffusion tensor imaging data of normal people and patients with vascular cognitive impairment, analyzing and extracting resting-state functional magnetic resonance imaging data and magnetic resonance diffusion tensor imaging data to obtain imaging indicators of multimodal magnetic resonance neuroimaging data; 2) preprocessing of imaging indicators of multimodal magnetic resonance neuroimaging data; 3) selection of imaging indicators and construction of models; 4) extraction of imaging markers; 5) regression analysis of imaging markers and neurocognitive scales. The method of the present application uses unsupervised K-means clustering to develop a method for extracting multimodal neuroimaging markers, find key imaging markers among the indicators of numerous multimodal neuroimaging data; serve the early and accurate diagnosis and treatment of VCI, and provide assistance and basis for the study of clinical VCI brain mechanisms.

Description

A method for extracting key imaging markers of vascular cognitive impairment

Field of the Invention

The present invention relates to the field of neuroimaging markers. Specifically, the present application provides a method for extracting key imaging markers of vascular cognitive impairment.

Background technique

Vascular cognitive impairment (VCI) refers to a type of cognitive impairment syndrome caused by cerebrovascular risk factors and cerebrovascular diseases. With the aging of the population, the prevalence of VCI in my country is increasing, which can seriously affect the daily quality of life of patients and place a heavy mental and economic burden on the patients' families.

VCI covers all disease stages from mild cognitive impairment originating from cerebrovascular lesions to vascular dementia. It is divided into three subtypes according to clinical manifestations: vascular cognitive impairment without dementia (VCIND), vascular dementia (VaD) and mixed dementia (MD), of which VCIND is the most common subtype of VCI. According to the Chinese Cognitive and Aging Study led by his team, VCIND accounts for 42% of the total number of patients with vascular cognitive impairment in China, making it the most common subtype. The Canadian Centre for Aging Research found through a 5-year follow-up that 46% of VCIND patients will progress to VaD. This shows that early detection, early diagnosis, and early intervention of VCI patients are necessary.

At present, the diagnosis and classification of VCI are still mainly based on clinical manifestations and neuropsychological scales, which are highly subjective and not conducive to early clinical diagnosis and prevention. A large number of studies have reported that there are significant differences in brain structure and function between patients with vascular cognitive impairment and normal subjects, which is beneficial to the diagnosis of VCI. Extracting simple and objective imaging markers is of certain significance. Different modality data of magnetic resonance imaging can be used to objectively quantify changes in brain function and structure, but the characteristics of brain function and brain structure are complicated and redundant. Machine learning methods can well integrate and analyze the features of multimodal redundancy, constrain the model through the setting of machine learning objective functions, and extract contributing indicators. However, the commonly used methods are usually supervised methods, and the robustness of the model cannot be fully verified. The reasoning process of conventional machine learning methods is difficult to establish human-understandable meanings based on indicators such as VCI neuroimaging features, and its interpretability is often not strong. In addition, the research shows that there is currently no effective method for the extraction and diagnosis of imaging markers for VCI. Therefore, it is necessary to develop an unsupervised, easy-to-interpret, simple and objective method for extracting multimodal neuroimaging markers.

Summary of the invention

The image marker extraction method involved in the present invention only uses the patient's structural nuclear magnetic resonance, diffusion tensor imaging and resting functional nuclear magnetic resonance imaging data. It does not involve other basic clinical information of the patient, such as the age, gender, occupation, region and whether the subject suffers from underlying diseases. This patent is only related to the abnormal changes in brain structure, brain network and brain function in the subject's imaging data. Therefore, the application of this patent is regardless of the subject's age, gender, region, and race. It can be applied to different populations for extraction in this context and research. At the same time, the method proposed in the present invention is based on the analysis of the structure, network and function of conventional brain images, combined with machine learning algorithms to analyze the interaction between high-dimensional features, and find out the characteristics specific to VCI. It is worth noting that this image marker extraction method is not only suitable for the analysis of VCI. For other mental illnesses, their pathogenic mechanisms are usually abnormal brain function, structure and network. This image marker extraction method can also be used to mine image markers for other diseases. It has a wide range of applications and can be used to support the research of other diseases.

In order to address the deficiencies of current technologies, the present invention innovatively uses unsupervised K-means clustering to develop a method for extracting multimodal neuroimaging markers, and finds key imaging markers among the indicators of numerous multimodal neuroimaging data. The extracted markers are regressed with clinical scales to prove whether the selected imaging markers can characterize changes in brain structure and brain function, and their relevance to clinical practice. The purpose of the present invention is to provide a method for extracting imaging markers for vascular cognitive impairment. The entire imaging marker extraction method and system serve the early and accurate diagnosis and treatment of VCI, and provide assistance and basis for the study of clinical VCI brain mechanisms.

The basic idea of the method of the present invention is: using multimodal neuroimaging data, first extract and preprocess the indicators that can characterize brain structure, network and brain function in the neuroimaging data. After integrating all the indicators, input them into the Least Absolute Shrinkage and Selection Operator (LASSO) to assign weights to different imaging indicators, and then select the threshold of the number of clusters and weights according to the loss of the clustering algorithm. The number of clusters and the weight threshold are input into the clustering algorithm, and effective imaging markers are extracted through the constraints of the classification task and the final weight of the marker is given to achieve the extraction of imaging markers. The extracted imaging markers are input into the Relevant Vector Regression (RVR) model to predict the score of the neurological scale and evaluate its ability to be applied in clinical practice.

On the one hand, the present application provides a method for extracting key imaging markers of vascular cognitive impairment, the method comprising:

1) Obtain resting-state functional magnetic resonance imaging (fMRI) and magnetic resonance diffusion tensor imaging (MRI) data of patients with vascular cognitive impairment and normal experimental group, analyze and extract resting-state functional magnetic resonance imaging (fMRI) and magnetic resonance diffusion tensor imaging (MRI) data to obtain imaging indicators of multimodal MRI neuroimaging data;

2) Preprocessing of imaging indicators of multimodal MRI neuroimaging data;

3) Selection of imaging indicators and construction of models;

4) Extraction of image markers;

5) Regression analysis of imaging markers and neurocognitive scales.

Further, step 1) comprises:

1-1) Obtaining MRI data of patients with vascular cognitive impairment and normal experimental group

Eligible patients with vascular cognitive impairment and normal experimental groups were enrolled according to the inclusion and exclusion criteria. During the acquisition process, fillers were placed around the head to prevent head movement, and all subjects were told to empty their minds but not fall asleep. Siemens 3T magnetic resonance imaging equipment was used to acquire head images using a 32-channel head coil to obtain resting state magnetic resonance imaging data and magnetic resonance diffusion tensor imaging data.

1-2) Analysis and extraction of resting state functional magnetic resonance imaging data:

In the first step, the standard pipeline in the conn toolkit in Matlab was used to preprocess the resting-state functional MRI data, including functional image rearrangement and expansion, time slice correction, outlier identification, indirect segmentation and normalization, functional and structural co-registration, and smoothing based on a 6 mm full-width half-maximum Gaussian kernel.

In the second step, the data before and after smoothing were input into the RESTPlus toolkit for further processing, including: detrending, Friston 24, gray matter, white matter, and cerebrospinal fluid covariate regression and filtering;

In the third step, based on the results of the second step, the AAL spectrum was combined to obtain the indicators of the functional image: low-frequency vibration amplitude, fractional amplitude of low-frequency fluctuation, percentage amplitude of fluctuation and Kendall regional consistency coefficient;

In the fourth step, based on the results of the second step, the processed data were input into the Gretna toolkit to construct the functional connectivity of ROIs-ROIs; based on the functional connectivity and AAL atlas, graph theory imaging indicators of brain ROIs were extracted: assortativity, betweenness centrality, degree centrality, network efficiency, node clustering coefficient, node efficiency, node local efficiency, node shortest path length, rich club and small world indicators;

1-3) Analysis and extraction of magnetic resonance diffusion tensor imaging data:

In the first step, the MRI diffusion tensor imaging data were preprocessed based on the standard pipeline of the PANDA toolkit in Matlab under Linux system, including: brain mask estimation, image cropping, eddy current correction and head motion correction;

In the second step, the indices of diffusion tensor parameters were calculated and registered from individual space to MNI standard space;

In the third step, the manually segmented white matter atlas provided in the PANDA toolkit was combined to obtain the final diffusion imaging indices: fractional anisotropy, mean diffusivity, axial diffusivity, radial diffusivity, and local diffusion uniformity;

Further, step 2) comprises:

When a certain imaging index or a certain patient data is missing more than 20%, the index or patient is excluded; after excluding the data, the overall mode of the index is used to make up for other missing values; all the indexes are combined together and the standardized imaging index is calculated according to the following formula to ensure that the variables are in a unified dimension; the overall index is processed in a disordered manner;

In formula (1), _xi is the i-th value of the indicator, _ux is the overall mean of the indicator, and _stdx is the standard deviation of the indicator.

Further, step 3) includes:

In the first step, all imaging indicators were feature screened to verify whether each indicator met the variance homogeneity. If the variances were equal, the T test was used to filter out the features with significant differences, otherwise the Welch T test was used;

In the second step, after the image indicators are initially screened out, the LASSO regression algorithm is used to further screen out important features to avoid overfitting of the model; the weight of each image indicator, that is, the contribution degree of the indicator, is obtained using the following formula:

In formula (2), m is the number of samples, _yi is the actual label of the ith sample, _xi is the image index of the ith sample, w and Wi are the weight of the image index and the weight of the image index of the ith sample, respectively. is the regularization term, || _wi || ₁ is the 1-norm;

The third step is to set the number of clusters and the weight threshold, and use grid search to organize the parameters; the number of clusters is 1-11, and the weight threshold is Where min and max are the minimum and maximum functions respectively, and coef is the indicator weight value output by LASSO. The above two parameters are combined into a grid, and the parameters are defined by combining the elbow rule and the loss function of the K-means clustering algorithm. When the loss curve has the first inflection point, the horizontal coordinate value of the point is taken as the indicator threshold or the number of clusters:

In formula (3), m is the number of samples, _xi is the image index of the i-th sample, Represents the center point of the cluster to which the i-th sample belongs; μ represents the center point of all clusters; after defining the number of clusters and weight threshold, the final K-means clustering model is established;

Further, step 4) includes:

After establishing the K-means clustering model, the clustering model visualization toolkit was used to visualize the image indicators. According to the different weight distributions, the final contributing image markers and the corresponding weight values were visualized.

Further, step 5) comprises:

The first step is to extract contributing image markers from multimodal image indicators, shuffle and distribute features and labels in a ratio of 7:3, with 70% as the training set and 30% as the test set; in order to prevent possible data leakage, the training set and test set are standardized respectively;

In the second step, the standardized data were input into the RVR model, and the training set was used for training and the test set for testing. The predicted scores of the neurocognitive scale were compared with the actual scores, and the correlation between the two was verified by linear regression and Pearson.

On the other hand, the present application provides application of the above method in the study of brain mechanisms of vascular cognitive impairment.

On the other hand, the present application provides the application of the above method in macro statistics or research on the health status of a population, and the application does not include diagnostic purposes.

Examples of the above applications include, but are not limited to, research on the likelihood of developing vascular cognitive dysfunction in a specific population, research on the association between vascular cognitive dysfunction and other diseases, etc.

This application is the first to apply an unsupervised clustering model to extract neural imaging markers from multimodal images in the exploration of imaging markers for vascular cognitive impairment. It is found that the method combining fMRI and DTI has the highest sensitivity and specificity. The extracted imaging markers have a high correlation with the neural scales in clinical screening, revealing the sensitivity and sensitivity of the extracted imaging markers. The extracted imaging markers and extraction methods can serve early and accurate diagnosis and treatment, provide assistance and basis for the study of clinical VCI brain mechanisms, and have certain promotion significance and value.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 Influential imaging markers and their importance in the clustering model: A. The weight corresponding to the normal experimental group considered by the clustering; B. The weight corresponding to the VCI considered by the clustering. ;

Figure 2; Visualization of brain regions where image markers are located under resting-state functional imaging and diffusion tensor imaging modalities: A. Visualization of brain regions where image markers are located under resting-state functional imaging; B. Visualization of brain regions where image markers are located under diffusion tensor imaging.

Fig. 3 The results of predicting neurocognitive scale based on imaging markers (A-G are different prediction results);

FIG4 is a technical roadmap of the method of the present application.

Detailed ways

The image marker extraction method involved in the present invention has nothing to do with age, gender, occupation, region, and whether or not the patient suffers from underlying diseases, but is only related to abnormal changes in brain structure, network, and function in the image data. At the same time, the method proposed in this design is based on the analysis of the structure, network, and function of conventional brain images, combined with a machine learning algorithm to analyze the interaction between high-dimensional features, and find out the characteristics specific to VCI. It is worth noting that this image marker extraction method is not only suitable for the analysis of VCI. For other mental illnesses, their pathogenic mechanisms are usually abnormalities in brain function, structure, and network. This image marker extraction method can also be used to mine image markers for other diseases, has a wider range of applications, and can be used to support the research of other diseases.

Example 1 Specific process of the method for extracting neuroimaging markers of VCI:

The basic process of the VCI neuroimaging marker extraction method of this application is shown in Figure 4:

1) Obtain resting-state functional magnetic resonance imaging (fMRI) and magnetic resonance diffusion tensor imaging (MRI) data of patients with vascular cognitive impairment and normal experimental group, analyze and extract resting-state functional magnetic resonance imaging (fMRI) and magnetic resonance diffusion tensor imaging (MRI) data to obtain imaging indicators of multimodal MRI neuroimaging data;

1-1) Obtaining MRI data of patients with vascular cognitive impairment and normal experimental group

Eligible patients with vascular cognitive impairment and normal experimental groups were enrolled according to the inclusion and exclusion criteria. During the acquisition process, fillers were placed around the head to prevent head movement, and all subjects were told to empty their minds but not fall asleep. Siemens 3T magnetic resonance imaging equipment was used to acquire head images using a 32-channel head coil to obtain resting state magnetic resonance imaging data and magnetic resonance diffusion tensor imaging data.

1-2) Analysis and extraction of resting state functional magnetic resonance imaging data:

In the first step, the standard pipeline in the conn toolkit in Matlab was used to preprocess the resting-state functional magnetic resonance imaging data, including functional image rearrangement and expansion, time layer correction, outlier identification, indirect segmentation and normalization, functional and structural co-registration, and smoothing based on a Gaussian kernel with a full width at half maximum of 6 mm.

In the second step, the preprocessing of different resting MRI indicators is different, and it is necessary to select the data before and after smoothing for further processing according to the indicator. The smoothed data are input into the RESTPlus toolkit for further processing, mainly including: detrending, covariate regression (Friston 24, gray matter, white matter and cerebrospinal fluid) and filtering.

In the third step, based on the processing in the second step and combined with the AAL spectrum, the indicators of the functional image are obtained: low-frequency vibration amplitude (smoothed but not filtered before detrending), fractional amplitude of low-frequency fluctuation (smoothed but not filtered before detrending), percentage amplitude of fluctuation (smoothed and filtered before detrending) and Kendall area consistency coefficient (finally filtered and smoothed).

In the fourth step, after detrending, covariate regression and filtering operations on the smoothed data in the preprocessing, the processed data were input into the Gretna toolkit to construct the functional connectivity of ROIs-ROIs. Based on the functional connectivity and AAL atlas, graph theory imaging indicators of brain ROIs can be extracted: assortativity, betweenness centrality, degree centrality, network efficiency, node clustering coefficient, node efficiency, node local efficiency, node shortest path length, rich club and small world indicators.

1-3) Analysis and extraction of magnetic resonance diffusion tensor imaging data:

In the first step, the MRI diffusion tensor imaging data were preprocessed based on the standard pipeline of the PANDA toolkit in Matlab under Linux system, including: brain mask estimation, image cropping, eddy current correction and head motion correction.

In the second step, the indices of diffusion tensor parameters were calculated and registered from the individual space to the MNI standard space.

In the third step, the manually segmented white matter atlas provided in the PANDA toolkit was combined to obtain the final diffusion imaging indices: fractional anisotropy, mean diffusivity, axial diffusivity, radial diffusivity, and local diffusion uniformity.

2) Preprocessing of imaging indicators of multimodal MRI neuroimaging data:

It is difficult to achieve good results by directly inputting multimodal imaging indicators into the model. Preprocessing is a necessary step before modeling. When a certain imaging indicator or a patient's data is missing more than 20%, the indicator or patient will be excluded. After deleting the data, for other missing values, the overall mode of the indicator is used to make up for it. Then, all indicators are combined together and the standardized imaging indicators are calculated according to the following formula to ensure that the variables are in a unified dimension.

Among them, _xi is the i-th value in the indicator, _ux is the overall mean of the indicator, and _stdx is the standard deviation of the indicator. Finally, the overall indicators are shuffled.

3) Selection of image indicators and construction of models:

In the first step, feature screening was performed on all imaging indicators. First, each indicator was verified to be consistent with variance homogeneity. If the variances were equal, the T test was used to filter out features with significant differences, otherwise the Welch T test was used.

In the second step, after the initial screening of image indicators, the LASSO regression algorithm is used to further screen out important features to avoid overfitting of the model. The following formula is optimized to obtain the weight of each image indicator, that is, the contribution degree of the indicator.

Where m is the number of samples, _yi is the actual label of the ith sample, _xi is the image index of the ith sample, w and Wi are the weights of the image index and the weight of the image index of the ith sample, respectively. is the regularization term, and || _wi || ₁ is the 1-norm.

The third step is to adjust and establish model parameters. The parameters that need to be set are mainly the number of clusters and the weight threshold. The parameters are organized using a grid search method. The specific parameter selection range is: the number of clusters is 1-11, the weight threshold is

Among them, min and max are the minimum and maximum functions respectively, and coef is the indicator weight value output by LASSO. The above two parameters are combined into a grid, and the parameters are defined by combining the elbow rule and the loss function of the K-means clustering algorithm. When the loss curve has the first large inflection point, the horizontal coordinate value of the point is taken as the indicator threshold or the number of clusters.

Where m is the number of samples, _xi is the image index of the i-th sample, Represents the center point of the cluster to which the i-th sample belongs; μ represents the center point of all clusters; after defining the number of clusters and weight threshold, the final K-means clustering model is established.

4) Extraction of image markers:

After the K-means clustering model was established, the clustering model visualization toolkit (https://github.com/YousefGh/kmeans-feature-importance) was used to visualize the image indicators. According to the different weight distributions, the image markers that ultimately contributed and the corresponding weight values were visualized. The distribution of weight values is shown in Figure 1. The brain regions where the image markers were located were further visualized, and the visualization results are shown in Figure 2.

5) Regression analysis of imaging markers and neurocognitive scales:

In the first step, we extract contributing image markers from the multimodal image indicators, shuffle and distribute the features and labels in a ratio of 7:3, with 70% as the training set and 30% as the test set. In order to prevent possible data leakage, the training set and the test set are standardized.

In the second step, the standardized data was input into the RVR model, and the training set was used for training and the test set was used for testing. The predicted scores of the neurocognitive scale were compared with the actual scores, and the correlation between the two was verified by linear regression and Pearson, and the results are shown in Figure 3.

Claims (3)

1. A method of extracting a key image marker of a vascular cognitive disorder, the method comprising:

1) Acquiring resting state functional magnetic resonance imaging data and magnetic resonance diffusion tensor imaging data of a patient with vascular cognitive impairment and a normal experiment group, and analyzing and extracting the resting state functional magnetic resonance imaging data and the magnetic resonance diffusion tensor imaging data to obtain image indexes of multi-mode magnetic resonance neural image data;

2) Preprocessing image indexes of multi-mode magnetic resonance neural image data;

3) Selecting an image index and constructing a model;

4) Extracting an image marker;

5) Regression analysis of image markers and neurocognitive scale;

Wherein step 1) comprises:

1-1) acquiring magnetic resonance imaging data of a patient with vascular cognitive impairment and a normal experimental group

Inclusion of eligible vascular cognitive impairment patients and normal experimental groups according to inclusion exclusion criteria; filling filler around the head during the collection process to prevent head movement and inform all subjects to empty the brain but not to sleep; acquiring images of the head by using Siemens 3T magnetic resonance imaging equipment and adopting a 32-channel head coil to acquire resting-state functional magnetic resonance imaging data and magnetic resonance diffusion tensor imaging data;

1-2) analyzing and extracting resting state functional magnetic resonance imaging data:

the first step, preprocessing the resting state functional magnetic resonance imaging data by using a standard pipeline in a CONN toolkit in Matlab, and comprises the following steps: rearranging and expanding functional images, correcting a time layer, identifying outliers, indirectly segmenting and standardizing, jointly registering functions and structures and smoothing based on Gaussian kernels with the full width and half height of 6 mm;

Second, the data before and after smoothing is input into RESTPlus tool package for further processing, including: detrending, friston 24, gray matter, white matter, and cerebrospinal fluid covariates regression and filtering;

thirdly, based on the result of the second step, combining with the AAL map to obtain the index of the functional image: low frequency vibration amplitude, fractional amplitude of low frequency fluctuation, percent amplitude of fluctuation and Kendell region consistency coefficient;

Fourth, based on the result of the second step, inputting the processed data into Gretna tool kit to construct the function connection of the ROIs-ROIs; extracting graph theory image indexes of brain ROIs based on functional connection and AAL (analog to digital) atlas: homography, mesocenter, isocenter, network efficiency, node clustering coefficients, node efficiency, node local efficiency, node shortest path length, rich club and small world indexes;

1-3) analyzing and extracting magnetic resonance diffusion tensor imaging data:

The first step, preprocessing magnetic resonance diffusion tensor imaging data based on a standard pipeline of a PANDA kit in Matlab in a Linux system, wherein the preprocessing comprises the following steps: estimating brain mask, image clipping, eddy current correction and head movement correction;

Secondly, calculating indexes of diffusion tensor parameters, and registering the diffusion tensor indexes from an individual space to an MNI standard space;

thirdly, combining the manually segmented white matter atlas provided in the PANDA kit to obtain a final diffuse image index: fractional anisotropy, average diffusivity, axial diffusivity, radial diffusivity, and local diffusion uniformity;

wherein step 2) comprises:

when a certain image index or a certain patient data is lack by more than 20%, the index or the patient is excluded; after the data are removed, the overall mode of the index is used for making up for other missing values; combining all indexes together, calculating and obtaining standardized image indexes according to a formula (1), and ensuring that variables are in a unified dimension; carrying out disorder treatment on the integral index;

X _i in the formula (1) is the ith value in the index, u _x is the overall mean value of the index, and std _x is the standard deviation of the index; wherein step 3) comprises:

Firstly, screening the characteristics of all image indexes, verifying whether each index accords with variance homogeneity, if the variances are equal, filtering out the characteristics with obvious differences by using T test, otherwise, using Welch T test;

secondly, on the premise of preliminarily screening out image indexes, further screening out important features by using a LASSO regression algorithm, and avoiding the phenomenon of over-fitting of the model; using formula (2), a weight of each image index, that is, a contribution degree of the index, is obtained:

in the formula (2), m is the number of samples, y _i is the actual label of the ith sample, x _i is the image index of the ith sample, w and Wi are the weight of the image index and the weight of the image index of the ith sample respectively, Is a regular term, ||w _i||₁ is a 1-norm;

Thirdly, setting threshold values of the number and the weight of clusters, and organizing parameters by using a grid search mode; the clustering number is 1-11, and the weight threshold value Wherein min and max are minimum and maximum functions respectively, and coef is an index weight value output by LASSO; combining the number of clusters and the weight threshold value into a grid, defining parameters by combining an elbow rule and a loss function of a K-means clustering algorithm, and taking an abscissa value of a loss curve as the weight threshold value or the number of clusters when the loss curve has a first inflection point:

In the formula (3), m is the number of samples, x _i is the image index of the ith sample, Representing the center point corresponding to the cluster to which the ith sample belongs; μ represents the center point of all clusters; after defining the clustering quantity and the weight threshold, establishing a final K-means clustering model;

wherein step 4) comprises:

After a K-means clustering model is established, a clustering model visualization tool package is applied to visualize the image indexes, and finally the contributed image markers and the corresponding weight values are visualized according to different weight distributions;

wherein step 5) comprises:

Firstly, extracting contributed image markers from multi-mode image indexes, and carrying out disorder and distribution on the image markers and the labels according to the proportion of 7:3, wherein 70% is used as a training set, and 30% is used as a test set; in order to prevent possible data leakage, respectively carrying out standardized processing on the training set and the testing set;

Secondly, inputting standardized data into a related vector regression model, training by using a training set, and testing by using a testing set; the scores of the predicted neurocognitive scale are compared to the actual scores and the correlation between the two is verified by linear regression and Pearson.

2. Use of the method according to claim 1 in research of brain mechanisms of vascular cognitive impairment.

3. Use of the method according to claim 1 in macroscopic statistics or research of the health status of a population, said use not comprising diagnostic purposes.