CN115659144A - Sub-health personnel classification method, system and equipment based on fMRI multi-band analysis - Google Patents
Sub-health personnel classification method, system and equipment based on fMRI multi-band analysis Download PDFInfo
- Publication number
- CN115659144A CN115659144A CN202211337318.XA CN202211337318A CN115659144A CN 115659144 A CN115659144 A CN 115659144A CN 202211337318 A CN202211337318 A CN 202211337318A CN 115659144 A CN115659144 A CN 115659144A
- Authority
- CN
- China
- Prior art keywords
- sub
- frequency band
- fmri
- health
- frequency
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses a sub-health personnel classification method, a sub-health personnel classification system and sub-health personnel classification equipment based on fMRI multi-band analysis, and aims to improve classification accuracy. The method involves sub-health personnel and normal controls of data over multiple frequency bands, multiple sliding time windows, extracting features by using functional magnetic resonance imaging data of the subject, selecting features, and then classifying. The fMRI data are analyzed by using a multi-band method, and the close relation between the frequency band and the neural activity is considered, so that more implicit pathological information of sub-health personnel can be effectively mined; performing feature fusion on signals on multiple frequency bands under each sliding time window width by using a machine learning method, and performing classification prediction on sub-health personnel; and after the abnormal brain area of each frequency band is extracted, performing dynamic function connection analysis on the time sequence of the abnormal brain area under different time window widths on each frequency band, and searching for dynamic biomarkers of sub-health marine staffs to realize auxiliary diagnosis of the sub-health marine staffs.
Description
Technical Field
The invention relates to the technical field of magnetic resonance imaging data analysis, in particular to a sub-health personnel classification method, a sub-health personnel classification system and sub-health personnel classification equipment based on fMRI multi-band analysis.
Background
Functional magnetic resonance imaging based on blood oxygen dependence level is a noninvasive brain function research method widely used at present and has the characteristics of high space-time resolution and the like, wherein 0.01-0.08Hz in low-frequency components is a frequency band selected by most functional magnetic resonance researches at present, and the frequency band is considered to be related to nerve fluctuation. Some functional magnetic resonance studies do not select the 0.01-0.08Hz frequency band, but select a plurality of frequency bands with different frequency ranges. This is because it is becoming increasingly recognized that there is a close relationship between frequency bands and neural activity, with neurons not being active throughout the frequency range, but rather over some segment thereof. The human brain is a complex dynamic system that can spontaneously generate a large number of oscillatory waves. Relevant studies have shown that the range of nerve oscillation frequencies observed is wide, from less than 0.01Hz to greater than 1000Hz. Therefore, the functional magnetic resonance imaging technology based on multiple frequency bands is beginning to be applied to the research of neurological diseases, and provides favorable technical support for the research of human brain functional connection, the diagnosis of brain diseases and psychological diseases, and the like.
In the multi-band neural disease research based on functional magnetic resonance, which frequency band is selected and which characteristics are important to the performance of the classifier and the classification result. Although the conventional frequency band can ensure that most information in the original signal is retained, it is not suitable to use the frequency law range selection of the 'one-knife-cut' type for each interested region or for the interactive research of each pair of interested regions. The current methods for dividing frequency bands are few, and most methods do not accurately divide the frequency bands, so that the analysis result may be affected. In addition, in the existing multi-band research, the abnormal brain regions are mainly detected, the used analysis indexes are similar to the conventional 0.01-0.08Hz frequency band, the low-frequency amplitude, the local consistency and the like are mainly adopted, and the function connection analysis method is rarely used, and the tested dynamic function connection is even further researched to analyze the interaction between the abnormal brain regions. The methods can discover the implicit information of some nerve diseases to a certain extent, but the methods have respective defects and shortcomings in the application process, and the obtained combined features are used for classifying the nerve diseases, so that the accuracy of the method is further improved.
Therefore, the existing method for classifying sub-health personnel by using functional magnetic resonance imaging data still needs to be further developed and improved, and a more complete technical scheme needs to be provided on the basis of more in-depth research.
Disclosure of Invention
Due to the defects in the prior art, the invention provides a sub-health person classification method, system and device based on fMRI multi-band analysis, and aims to solve the problem of insufficient analysis accuracy in the prior art.
In order to achieve the above object, in one aspect, the present invention provides a sub-health person classification method based on fMRI multiband analysis, which is characterized by comprising the following steps:
step S1, extracting and preprocessing fMRI data on a single tested level: extracting and preprocessing original fMRI data to obtain a time sequence based on frequency bands, a tested object and voxels;
s2, extracting abnormal brain areas on multiple frequency bands: the data are processed by utilizing fast Fourier transform and an automatic anatomical template to obtain the low-frequency amplitude of each brain area to be tested on a plurality of frequency bands, and the abnormal brain area of sub-health personnel on each frequency band is obtained by a machine learning method;
s3, classifying the dynamic functional connection characteristics of abnormal brain areas on multiple frequency bands: and obtaining a time sequence of each tested brain area on each frequency band according to the abnormal brain areas by utilizing fast Fourier transform, fast Fourier inverse transform and an automatic anatomical template, and performing feature selection and classification on the correlation coefficient of the dynamic function connection of the sub-health personnel through machine learning.
The classification method analyzes fMRI data by using a multi-band method, considers the close relation between frequency bands and neural activity, can effectively mine more hidden pathological information of sub-health personnel, and can overcome some defects and limitations possibly brought by using conventional frequency bands; after the abnormal brain region of each frequency band is extracted, dynamic function connection analysis under different time window widths is carried out on the time sequence of the abnormal brain region on each frequency band, more hidden pathological information of sub-health personnel can be mined, and more analysis angles are provided for clinical auxiliary diagnosis.
Further, the step S1 includes the following processes:
s11, preprocessing data, wherein the preprocessing process comprises elimination of time points, time layer correction, head movement correction, space standardization and smoothing;
step S12, dividing frequency bands: firstly, obtaining the range of the whole frequency band according to the quinestet sampling theorem; according to the natural logarithm linear theory, concretizing the whole frequency band, and selecting the conventional frequency band of 0.01-0.08Hz as a comparative example;
and S13, obtaining the time sequence of each tested voxel on each frequency band through fast Fourier transform and inverse fast Fourier transform.
Further, the step S2 includes the following processes:
step S21, calculating the low-frequency amplitude of each tested and each interested region on each frequency band of the multi-band to obtain a plurality of matrixes as original characteristic matrixes, wherein the size of each matrix is a x (b + 1), a is the serial number of the tested behavior, and b is the low-frequency amplitude of the interested region; the last column is a tested label, the sub-health personnel are marked as 1, and the normal contrast is marked as 0;
s22, sorting the feature importance according to the original feature matrix on each frequency band, and sorting the feature importance in a descending order from front to back; and (3) for the rearranged features, putting the most important first N features into a classifier, wherein N is more than or equal to 1 and less than or equal to the number of the features, recording the points with the highest accuracy and the features selected at the points, and the features are abnormal brain areas needing to be extracted and further subjected to dynamic function connection analysis.
Further, the step S21 includes the following processing:
step S211, calculating a low-frequency amplitude of each voxel: fast Fourier transform is carried out on the time of each tested voxel on each frequency band to obtain a power spectrum, then evolution is carried out on the power spectrum, and the average value of the amplitude in a specific frequency range is solved to obtain the low-frequency amplitude of the voxel;
the time series after pretreatment is:
the low frequency amplitude is calculated as:
step S212, selecting the region of interest, dividing the cerebral cortex into 116 regions of interest by using an automatic anatomical template, and taking the mean value of the low-frequency amplitudes of all voxels contained in each brain area as the low-frequency amplitude of the brain area in 90 brain areas belonging to the brain.
Further, the step S22 includes the following processing:
step S221, feature importance ranking: the importance ranking is carried out on the characteristics on each frequency band by using a recursive characteristic deletion method of 10-fold cross validation, namely, the capability of distinguishing sub-healthy people from normal people in each brain region is ranked on each frequency band; the strength of the ability is expressed by a regression coefficient, and the larger the absolute value of the regression coefficient is, the stronger the importance is.
Step S222, abnormal brain region extraction: the used classifier is a support vector machine classifier which participates in 10-fold cross validation through grid search; and (3) putting the most important first N characteristics into a classifier on each frequency band, wherein N is more than or equal to 1 and less than or equal to the number of the characteristics, and recording the point with the highest accuracy and the characteristics selected at the point to obtain the abnormal brain area on each frequency band.
Further, the step S3 includes the following processes:
s31, calculating a time-varying correlation coefficient matrix of each frequency band abnormal brain region under a plurality of sliding time window widths; under the same sliding time window width, calculating the time-varying Pearson correlation coefficient of the abnormal brain region time sequence on each frequency band to obtain a plurality of Pearson correlation coefficient matrixes; the correlation coefficient of the sliding time window method is calculated by the following formula:
and S32, fusing the characteristics of multiple frequency bands under each sliding time window width, performing classification prediction, and comparing the classification prediction with the characteristics extracted from the conventional frequency bands.
Further, in step S31, the sliding time window has four widths, which are 15TRs,20TRs,25TRs, and 30TRs, respectively, and the step size is 5TRs.
Further, the step S32 includes the following processes:
step S321, feature fusion: the Pearson correlation coefficient matrix is a symmetric matrix, so that the number of an upper triangle or a lower triangle is only required to be taken; under each sliding time window width, taking out all tested dynamic function connection correlation coefficients on each frequency band, fusing the correlation coefficients together to serve as the characteristics of multiple frequency bands for subsequent characteristic selection and classification, and simultaneously carrying out the operation on the conventional frequency band of 0.01-0.08Hz as comparison;
step S322, classification effect evaluation and model evaluation: and evaluating the classification effect by using the precision, the recall ratio and the f1 score, and representing the performance of the classifier by using the area under the working characteristic curve of the subject.
On the other hand, the invention provides a sub-health personnel classification system based on fMRI multi-band analysis, which is characterized by being used for realizing the sub-health personnel classification method based on fMRI multi-band analysis, and the sub-health personnel classification system comprises a functional magnetic resonance data extraction module on a single tested level, an extraction module of abnormal brain areas on multiple bands and a dynamic functional connection feature classification module of the abnormal brain areas on the multiple bands.
In another aspect, the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor; the processor calls the computer program stored in the memory to execute the sub-health personnel classification method based on the fMRI multi-band analysis.
Compared with the prior art, the invention has the following advantages or beneficial effects:
(1) The fMRI data are analyzed by using a multi-band method, considering the close connection between the frequency band and the neural activity, neurons are not active in the whole frequency range but in some segments, the method can effectively extract more hidden pathological information of sub-health personnel, and overcomes some defects and limitations possibly brought by using the conventional frequency band;
(2) After the abnormal brain area of each frequency band is extracted, dynamic function connection analysis under different time window widths is carried out on the time sequence of the abnormal brain area on each frequency band, compared with the conventional dynamic function connection analysis, more hidden pathology information of sub-health personnel can be mined, and more analysis angles are provided for clinical auxiliary diagnosis.
Drawings
The invention and its features, aspects and advantages will become more apparent from reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings. Like reference symbols in the various drawings indicate like elements. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention.
FIG. 1 is a diagram illustrating the steps of a sub-health person classification method according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating the sub-health person classification according to an embodiment of the present invention.
Detailed Description
The structure of the present invention will be further described with reference to the accompanying drawings and specific examples, but the present invention is not limited thereto.
Example 1
Referring to fig. 1 and 2, a sub-health person classification method based on fMRI multiband analysis comprises the following steps:
step S1, extracting and preprocessing fMRI data on a single test level: and extracting and preprocessing the original fMRI data to obtain a time sequence based on frequency bands, a tested object and voxels.
S2, extracting abnormal brain areas on multiple frequency bands: the data are reprocessed by utilizing fast Fourier transform and an automatic dissection template to obtain the low-frequency amplitude of each tested brain area on a plurality of frequency bands, and the abnormal brain area of the sub-health personnel on each frequency band is obtained by a machine learning method; as a preferred embodiment, there are 6 frequency bands.
S3, classifying the dynamic functional connection characteristics of abnormal brain areas on multiple frequency bands: and obtaining a time sequence of each tested brain area on each frequency band according to the abnormal brain area by utilizing fast Fourier transform, fast Fourier inverse transform and an automatic dissection template, and performing feature selection and classification on the correlation coefficient of the dynamic function connection of the sub-health personnel through machine learning.
Referring to fig. 2, the sub-health personnel classification process based on fMRI multi-band dynamic functional connection can realize that dynamic interaction which is significantly different between sub-health personnel in multi-band and normal tested personnel is searched as a feature for classification, and a better classification effect than that of conventional frequency band research is obtained.
As a preferred embodiment, in the step S1, the method for extracting functional magnetic resonance data at a single test level specifically includes the following steps:
s11, preprocessing data, wherein the preprocessing process comprises elimination of time points, time layer correction, head movement correction, space standardization and smoothing; specifically, the related pretreatment process is as follows: (1) eliminating the previous 10 time points; (2) Time layer correction, which is to correct the acquisition time difference of the data of each tested time point; (3) Correcting the head movement, and removing the tested object with the head movement larger than 1.5mm and the rotation larger than 1.5 degrees; (4) Spatial normalization registers the functional image to the montreal neurological institute space and resamples image voxels to 3mm x 3mm; (5) Smoothing is performed using a full width half maximum gaussian kernel of 4mm x 4 mm.
Step S12, dividing frequency bands: firstly, obtaining the range of the whole frequency band according to the quinestet sampling theorem; typically TR =2, so the entire frequency band ranges from 0-0.25Hz. And according to the natural logarithm linear theory, the whole frequency band is specified, for example, the whole signal is specifically divided into the following 5 frequency bands: slow-6 (0.0040-0.0111 Hz), slow-5 (0.0111-0.0302 Hz), slow-4 (0.0302-0.0820 Hz), slow-3 (0.0820-0.2231 Hz), slow-2 (0.02231-0.2500 Hz), and a conventional frequency band of 0.01-0.08Hz is selected as a comparative example.
And S13, obtaining the time sequence of each tested voxel on each frequency band through fast Fourier transform and inverse fast Fourier transform. A typical voxel size is 3mm x 3mm.
The fast fourier transform formula used is:
the inverse fast fourier transform equation used is:
as a preferred embodiment, in step S2, the method for extracting an abnormal brain region over multiple frequency bands specifically includes the following steps:
step S21, calculating the low-frequency amplitude of each tested area and each interested area on each frequency band of 6 frequency bands to obtain 6 matrixes as original characteristic matrixes, wherein the size of each matrix is a x (b + 1), a is the serial number of the tested row and is listed as a brain area with the serial number from 1 to 90, and b is the low-frequency amplitude of the interested area; the last column of the matrix is a tested label, the sub-health personnel are marked as 1, and the normal contrast is marked as 0;
s22, sorting 1-90 rows of features according to feature importance according to the original feature matrix on each frequency band, and sorting the features in a descending order from front to back; for the rearranged features, putting the most important first N features into a classifier, wherein N is more than or equal to 1 and less than or equal to the number of the features; when the classification effect is optimal, recording the point with the highest accuracy and the characteristics selected at the point, wherein the characteristics are abnormal brain areas needing to be extracted and further subjected to dynamic function connection analysis, and colleagues record various parameters with the highest accuracy, values of various evaluation indexes and specific characteristics.
As a preferred embodiment, the step S21 specifically includes the following steps:
and S211, calculating the low-frequency amplitude of each voxel. And performing fast Fourier transform on the time of each voxel of each tested object on each frequency band to obtain a power spectrum, squaring the power spectrum, and calculating an amplitude average value in a specific frequency range to obtain the low-frequency amplitude of the voxel.
The time series after pretreatment is:
the low frequency amplitude is calculated as:
s212, selecting the region of interest, dividing the cerebral cortex into 116 regions of interest by using an automatic anatomical template, and taking the mean value of the low-frequency amplitudes of all voxels contained in each brain area as the low-frequency amplitude of the brain area in 90 brain areas belonging to the brain.
As a preferred embodiment, the step S22 specifically includes the following steps:
step S221, feature importance ranking: the importance ranking is carried out on the characteristics on each frequency band by using a recursive characteristic deletion method of 10-fold cross validation, namely, the capability of distinguishing sub-healthy people from normal people in each brain region is ranked on each frequency band; the strength of the ability is represented by a regression coefficient, and the larger the absolute value of the regression coefficient is, the stronger the importance is.
Step S222, abnormal brain region extraction: the used classifier is a support vector machine classifier which participates in 10-fold cross validation through grid search; and (3) putting the most important first N characteristics into a classifier on each frequency band, wherein N is more than or equal to 1 and less than or equal to the number of the characteristics, and recording the point with the highest accuracy and the characteristics selected at the point to obtain the abnormal brain area on each frequency band.
As a preferred embodiment, in step S3, the method for classifying the dynamic functional connectivity characteristics of the previous brain region in multiple frequency bands specifically includes the following steps:
and S31, calculating a time-varying correlation coefficient matrix of the abnormal brain region of each frequency band under the widths of four sliding time windows. Wherein, the widths of the four sliding time windows are respectively 15TRs,20TRs,25TRs and 30TRs, and the step length is 5TRs. Under the same sliding time window width, calculating the time-varying Pearson correlation coefficient of the abnormal brain region time sequence on each frequency band to obtain a plurality of Pearson correlation coefficient matrixes. The correlation coefficient of the sliding time window method is calculated as follows:
and S32, fusing the characteristics of the multiple frequency bands under each sliding time window width, performing classification prediction, and comparing the classification prediction with the characteristics extracted from the conventional frequency bands.
As a preferred embodiment, the step S32 specifically includes the following steps:
step S321, feature fusion: since the Pearson correlation coefficient matrix is a symmetric matrix, only the number of the upper triangle or the lower triangle needs to be taken. Under each sliding time window width, all tested dynamic function connection correlation coefficients are taken out on each frequency band and then fused together to serve as the characteristics of multiple frequency bands for subsequent characteristic selection and classification, and meanwhile, the operation comparison experiment is also carried out on the conventional frequency band of 0.01-0.08 Hz.
Step S322, classification effect evaluation and model evaluation: the present study evaluated the classification effect using Precision (ACC), precision (Precision/TPR), recall (Recall/TNR), and f1 score (f 1-score), and the area under the working characteristic curve of the subject was used to indicate the performance of the classifier. The calculation formula of the evaluation index is as follows:
and performing classification effect evaluation and model evaluation on the multi-band method, comparing the classification effect with the conventional frequency band, and selecting the characteristics as the biomarkers of sub-health personnel under the subsequent conditions that the classification effect is better and the model is more stable.
Example 2
A sub-health personnel classification system based on fMRI multi-band analysis is used for realizing the sub-health personnel classification method based on fMRI multi-band analysis in embodiment 1, and comprises a functional magnetic resonance data extraction module on a single tested level, an extraction module of abnormal brain areas on multiple frequency bands and a dynamic functional connection feature classification module of the abnormal brain areas on the multiple frequency bands.
Example 3
An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor; the processor invokes the computer program stored in the memory to perform the sub-health person classification method based on fMRI multiband analysis of embodiment 1.
In conclusion, the invention provides a sub-health person classification method, system and device based on fMRI multi-band analysis, so as to improve classification accuracy. The method involves sub-health personnel and normal controls of data over multiple frequency bands, multiple sliding time windows, extracting features by using functional magnetic resonance imaging data of the subject, selecting features, and then classifying. The method analyzes fMRI data by using a multi-band method, considers the close relation between the frequency band and the neural activity, and can effectively excavate more hidden pathological information of sub-health personnel; performing feature fusion on signals on multiple frequency bands under each sliding time window width by using a machine learning method, and performing classification prediction on sub-health personnel; and after the abnormal brain area of each frequency band is extracted, performing dynamic function connection analysis on the time sequence of the abnormal brain area under different time window widths on each frequency band, and searching for dynamic biomarkers of sub-health marine staffs to realize auxiliary diagnosis of the sub-health marine staffs.
The methods, theories or processes not described in detail in this embodiment are prior art and are not described herein in detail. Those skilled in the art will appreciate that variations may be implemented by those skilled in the art in combination with the prior art and the above-described embodiments, and will not be described herein in detail. Such variations do not affect the essence of the present invention, and are not described herein.
Through the description of the foregoing embodiments, it is clear to those skilled in the art that the method of the foregoing embodiments may be implemented by software plus a necessary general hardware platform, and certainly may also be implemented by hardware, but in many cases, the former is a better implementation. Based on such understanding, the technical solutions of the present invention or portions thereof contributing to the prior art may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the methods according to the embodiments of the present invention.
The above description is of the preferred embodiment of the invention. It is to be understood that the invention is not limited to the particular embodiments described above, in that devices and structures not described in detail are understood to be implemented in a manner common in the art; those skilled in the art can make many possible variations and modifications to the disclosed embodiments, or modify equivalent embodiments to equivalent variations, without departing from the spirit of the invention, using the methods and techniques disclosed above. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention are still within the scope of the protection of the technical solution of the present invention, unless the contents of the technical solution of the present invention are departed.
Claims (10)
1. The sub-health personnel classification method based on fMRI multi-band analysis is characterized by comprising the following steps of:
step S1, extracting and preprocessing fMRI data on a single test level: extracting and preprocessing original fMRI data to respectively obtain time sequences of a frequency band, a tested object and a voxel;
s2, extracting abnormal brain areas on multiple frequency bands: the data are processed by utilizing fast Fourier transform and an automatic anatomical template to obtain the low-frequency amplitude of each brain area to be tested on a plurality of frequency bands, and the abnormal brain area of sub-health personnel on each frequency band is obtained by a machine learning method;
s3, classifying the dynamic function connection characteristics of abnormal brain areas on multiple frequency bands: and obtaining a time sequence of each tested brain area on each frequency band according to the abnormal brain area by utilizing fast Fourier transform, fast Fourier inverse transform and an automatic dissection template, and performing feature selection and classification on the correlation coefficient of the dynamic function connection of the sub-health personnel through machine learning.
2. The method for classifying sub-health personnel based on fMRI multiband analysis according to claim 1, wherein said step S1 comprises the following processes:
s11, preprocessing data, wherein the preprocessing process comprises elimination of time points, time layer correction, head movement correction, space standardization and smoothing;
step S12, frequency bands are divided: firstly, obtaining the range of the whole frequency band according to the quinestet sampling theorem; according to the natural logarithm linear theory, concretizing the whole frequency band, and selecting the conventional frequency band of 0.01-0.08Hz as a comparative example;
and S13, obtaining the time sequence of each tested voxel on each frequency band through fast Fourier transform and inverse fast Fourier transform.
3. The method for classifying sub-health personnel based on fMRI multiband analysis according to claim 1, wherein said step S2 comprises the following processes:
step S21, calculating the low-frequency amplitude of each tested and each interested region on each frequency band of the multi-band to obtain a plurality of matrixes as original characteristic matrixes, wherein the size of each matrix is a x (b + 1), a is the serial number of the tested behavior, and b is the low-frequency amplitude of the interested region; the last column is a tested label, the sub-health personnel are marked as 1, and the normal contrast is marked as 0;
s22, sorting the feature importance according to the original feature matrix on each frequency band, and sorting the feature importance in a descending order from front to back; and (3) for the rearranged features, putting the most important first N features into a classifier, wherein N is more than or equal to 1 and less than or equal to the number of the features, recording the points with the highest accuracy and the features selected at the points, and the features are abnormal brain areas needing to be extracted and further subjected to dynamic function connection analysis.
4. The method for classifying sub-health persons based on fMRI multiband analysis according to claim 3, wherein said step S21 comprises the following processes:
step S211, calculating a low-frequency amplitude of each voxel: fast Fourier transform is carried out on the time of each tested voxel on each frequency band to obtain a power spectrum, then evolution is carried out on the power spectrum, and the average value of the amplitude in a specific frequency range is solved to obtain the low-frequency amplitude of the voxel;
the time series after pretreatment is:
the low frequency amplitude is calculated as:
step S212, selecting the region of interest, dividing the cerebral cortex into 116 regions of interest by using an automatic anatomical template, and taking the mean value of the low-frequency amplitudes of all voxels contained in each brain area as the low-frequency amplitude of the brain area in 90 brain areas belonging to the brain.
5. The method for classifying sub-health persons based on fMRI multiband analysis according to claim 3, wherein said step S22 comprises the following processes:
step S221, feature importance ranking: the importance ranking is carried out on the characteristics on each frequency band by using a recursive characteristic deletion method of 10-fold cross validation, namely, the capability of distinguishing sub-healthy people from normal people in each brain region is ranked on each frequency band; the strength of the ability is expressed by a regression coefficient, and the larger the absolute value of the regression coefficient is, the stronger the importance is.
Step S222, abnormal brain area extraction: the used classifier is a support vector machine classifier which participates in 10-fold cross validation through grid search; and (3) putting the most important first N characteristics into a classifier on each frequency band, wherein N is more than or equal to 1 and less than or equal to the number of the characteristics, and recording the point with the highest accuracy and the characteristics selected at the point to obtain the abnormal brain area on each frequency band.
6. The method for classifying sub-health personnel based on fMRI multiband analysis according to claim 1, wherein said step S3 comprises the following processes:
s31, calculating a time-varying correlation coefficient matrix of each frequency band abnormal brain region under a plurality of sliding time window widths; under the same sliding time window width, calculating the time-varying Pearson correlation coefficient of the abnormal brain area time sequence on each frequency band to obtain a plurality of Pearson correlation coefficient matrixes; the correlation coefficient of the sliding time window method is calculated by the following formula:
and S32, fusing the characteristics of the multiple frequency bands under each sliding time window width, performing classification prediction, and comparing the classification prediction with the characteristics extracted from the conventional frequency bands.
7. The method for classifying sub-health workers based on fMRI multiband analysis according to claim 6, wherein in said step S31, said sliding time window width is four, 15TRs,20TRs,25TRs,30TRs, and 5TRs in step size.
8. The method for classifying sub-health personnel based on fMRI multiband analysis according to claim 6, wherein said step S32 comprises the following processes:
step S321, feature fusion: the Pearson correlation coefficient matrix is a symmetric matrix, so that the number of an upper triangle or a lower triangle is only required to be taken; under each sliding time window width, taking out all tested dynamic function connection correlation coefficients on each frequency band, fusing the correlation coefficients together to serve as the characteristics of multiple frequency bands for subsequent characteristic selection and classification, and simultaneously carrying out the operation on the conventional frequency band of 0.01-0.08Hz as comparison;
step S322, classification effect evaluation and model evaluation: and evaluating the classification effect by using the precision, the recall ratio and the f1 score, and representing the performance of the classifier by using the area under the working characteristic curve of the subject.
9. The system for classifying the sub-health personnel based on the fMRI multi-band analysis is characterized by being used for realizing the method for classifying the sub-health personnel based on the fMRI multi-band analysis in any one of claims 1 to 8, and comprises a functional magnetic resonance data extraction module on a single tested level, an extraction module of abnormal brain areas on multiple frequency bands and a dynamic functional connection feature classification module of the abnormal brain areas on the multiple frequency bands.
10. An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor; the processor invokes the memory-stored computer program to perform the method for classifying sub-health personnel based on fMRI multiband analysis according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211337318.XA CN115659144A (en) | 2022-10-28 | 2022-10-28 | Sub-health personnel classification method, system and equipment based on fMRI multi-band analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211337318.XA CN115659144A (en) | 2022-10-28 | 2022-10-28 | Sub-health personnel classification method, system and equipment based on fMRI multi-band analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115659144A true CN115659144A (en) | 2023-01-31 |
Family
ID=84993740
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211337318.XA Pending CN115659144A (en) | 2022-10-28 | 2022-10-28 | Sub-health personnel classification method, system and equipment based on fMRI multi-band analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115659144A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116597994A (en) * | 2023-05-16 | 2023-08-15 | 天津大学 | Mental disease brain function activity assessment device based on brain activation clustering algorithm |
-
2022
- 2022-10-28 CN CN202211337318.XA patent/CN115659144A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116597994A (en) * | 2023-05-16 | 2023-08-15 | 天津大学 | Mental disease brain function activity assessment device based on brain activation clustering algorithm |
| CN116597994B (en) * | 2023-05-16 | 2024-05-14 | 天津大学 | Mental disease brain function activity assessment device based on brain activation clustering algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Epileptic seizure detection in EEG signals using sparse multiscale radial basis function networks and the Fisher vector approach | |
| Thirion et al. | Detection of signal synchronizations in resting-state fMRI datasets | |
| CN109522894B (en) | Method for detecting dynamic covariation of fMRI brain network | |
| Ryali et al. | Sparse logistic regression for whole-brain classification of fMRI data | |
| CN110840402A (en) | Atrial fibrillation signal identification method and system based on machine learning | |
| CN109344889B (en) | Brain disease classification apparatus, user terminal, and computer-readable storage medium | |
| CN104715261A (en) | FMRI dynamic brain function sub-network construction and parallel connection SVM weighted recognition method | |
| Jia et al. | A two-stage feature selection framework for hyperspectral image classification using few labeled samples | |
| Li et al. | MRI denoising using progressively distribution-based neural network | |
| CN112002428B (en) | Whole brain individualized brain function map construction method taking independent component network as reference | |
| Wang et al. | SACICA: a sparse approximation coefficient-based ICA model for functional magnetic resonance imaging data analysis | |
| CN112270255B (en) | Electroencephalogram signal identification method and device, electronic equipment and storage medium | |
| CN113486752B (en) | Emotion recognition method and system based on electrocardiosignal | |
| CN109009098B (en) | Electroencephalogram signal feature identification method under motor imagery state | |
| CN111540467B (en) | Schizophrenia classification identification method, operation control device and medical equipment | |
| CN115659144A (en) | Sub-health personnel classification method, system and equipment based on fMRI multi-band analysis | |
| CN111402212A (en) | Extraction method of dynamic connection activity mode of maritime brain function network | |
| Alimardani et al. | Presenting a spatial-geometric EEG feature to classify BMD and schizophrenic patients | |
| CN111414579A (en) | Method and system for acquiring brain area association information based on multi-angle correlation | |
| Li et al. | Improving medical/biological data classification performance by wavelet preprocessing | |
| Jahanian et al. | Controlling the false positive rate in fuzzy clustering using randomization: application to fMRI activation detection | |
| US20210386351A1 (en) | Brain-computer interface apparatus for minimizing signal correction process between users using clustering technology based on brain activity and method of operating the same | |
| Uçar et al. | Automatic detection of white matter hyperintensities via mask region-based convolutional neural networks using magnetic resonance images | |
| Sianipar et al. | Performance Analysis of Classifier in Detecting Epileptic Seizure Based on Discrete Wavelet Transform | |
| Rajaguru et al. | Analysis and detection of sentient state from alcoholic EEG signals using the fast Fourier transform features based multiple classifiers |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |