阅读说明：本技术 基于傅里叶同步压缩变换可靠划分脑低频波动子区域方法 (Method for reliably dividing brain low-frequency fluctuation sub-region based on Fourier synchronous compression transformation ) 是由 王倪传 俞钦 颜虹杰 仲兆满 于 2020-11-18 设计创作，主要内容包括：本发明公开了基于傅里叶同步压缩变换可靠划分脑低频波动子区域方法,首先对脑功能磁共振数据进行预处理及去漂移处理,采用傅里叶同步压缩变换,将大脑的时间信号转换到时频信号,并作为脑体素的表征信息,从而可从时频重构领域解析脑区之间的动态关联。由于时频信号的特殊性及大脑信号体素众多、数据集庞大,引入自动目标生成过程算法适配FSST数据下的Kmeans算法,以空间投影法找寻类的初始标签,再进行距离计算,重新进行类心计算,直到迭代完成,找到最优的类心。距离计算选择相关系数,区域选择数据集中低频区域的并集。最后将数据还原到空间,再从脑空间图谱观测不同分类之间的空间联系。(The invention discloses a method for reliably dividing brain low-frequency fluctuation sub-regions based on Fourier synchronous compression transformation. Due to the particularity of time-frequency signals, numerous voxels of brain signals and huge data sets, an automatic target generation process algorithm is introduced to adapt to a Kmeans algorithm under FSST data, an initial label of a class is found by a space projection method, distance calculation is carried out again, and the heart-like calculation is carried out again until iteration is completed to find the optimal heart-like. The distance calculation selects the correlation coefficient and the region selects the union of the low frequency regions in the data set. And finally, restoring the data to a space, and observing the spatial relation among different classifications from the brain space atlas.)

1. The method for reliably dividing the brain low-frequency fluctuation subarea based on Fourier synchronous compression transformation is characterized by comprising the following steps of:

step 1: dividing a low-frequency area, searching areas with low-frequency amplitude extracted from the same tested state at different moments, selecting the areas by adopting a traditional ALFF method, removing some noise areas and selecting the areas with strict verification;

step 2: performing time-frequency reconstruction on the preprocessed functional nuclear magnetic resonance signals, converting the preprocessed functional nuclear magnetic resonance signals into time-frequency domains, and acquiring related time-frequency-power spectrograms to form a time-frequency data set of the nuclear magnetic resonance signals;

and step 3: based on the time-frequency dynamic correlation angle, the ATGP-Kmeans is adopted to adapt to FSST data by utilizing the dynamic synchronicity of different voxels in time-frequency, correlation coefficient calculation is carried out, clustering is completed, the clustering is mapped into a spatial brain map, and the correlation of different brain areas in the space is observed.

2. The method for reliably dividing the brain low-frequency fluctuation subarea based on the Fourier synchronous compressive transformation as claimed in claim 1, wherein: and selecting a classification region in the step 1, wherein a low-frequency region is used as a research region.

3. The method for reliably dividing the brain low-frequency fluctuation subarea based on the Fourier synchronous compressive transformation as claimed in claim 1, wherein: in the step 2 and the step 3, through time-frequency reconstruction of data, an adaptive classification algorithm is introduced, the data are clustered and then mapped into a space, and time-frequency relation of different brain areas embodied on the space is researched.

4. The method for reliably dividing the brain low-frequency fluctuation subarea based on the Fourier synchronous compressive transformation as claimed in claim 1, wherein: the selection of the region in the step 1 adopts a traditional ALFF method formula as follows:a_k(f_k)，b_k(f_k) Are respectively asReal and imaginary parts.

5. The method for reliably dividing the brain low-frequency fluctuation subarea based on the Fourier synchronous compressive transformation as claimed in claim 1, wherein: removing some noise regions in the step 1, and selectingChecking a stricter area;

is the average number of samples and is the average number of samples,for the standard deviation of the samples, n is the number of samples, and the statistic t is given in the zero hypothesis: the condition that μ ═ μ 0 is true obeys a t distribution with a degree of freedom of n.

6. The method for reliably dividing the brain low-frequency fluctuation subarea based on the Fourier synchronous compressive transformation as claimed in claim 1, wherein: the formula for performing time-frequency reconstruction on the preprocessed functional nuclear magnetic resonance signals in the step 2 is as follows:converting the time spectrum into a time-frequency domain to obtain a related time-frequency-power spectrogram; the signal f (t) is a plurality of f_k(t) composition, with the STFT predominant ridge at (t, φ'_k(t)), can be approximated bySubstitute phi'_k(t)。

7. The method for reliably dividing the brain low-frequency fluctuation subarea based on the Fourier synchronous compressive transformation as claimed in claim 1, wherein: in the step 3, different voxels are utilized in time based on the time-frequency dynamic correlation angleInter-frequency dynamic synchronism and brain local correlation principle, the low-frequency voxel signals are processed by adopting a clustering means, and due to the redundancy and complexity of the signals, an ATGP algorithm is introduced: t is t₁＝arg{max_r[r^Tr]R are all voxels to be observed, and U is^TThe pseudo-inverse of U is set to,as an initial centroid selection algorithm of Kmeans, ATGP-Kmeans is adopted to adapt FSST data, and a loss function is minimized:

8. the method for reliably dividing the brain low-frequency fluctuation subarea based on the Fourier synchronous compressive transformation as claimed in claim 1, wherein: and 3, performing correlation coefficient calculation to complete clustering, wherein the correlation coefficient calculation is selected:reselecting a class center:and finishing clustering until the maximum distance is reached, mapping the clustering to a spatial brain map, and observing the correlation of different brain areas in the space.

Technical Field

The invention relates to the field of brain low-frequency fluctuation region subdivision, in particular to a method for reliably dividing a brain low-frequency fluctuation sub-region based on Fourier synchronous compression transformation.

Background

Blood oxygen level dependent effects of the brain can be used to characterize metabolic conditions of neurons in the brain, thereby indirectly reflecting neuronal activity. Researches show that the human brain in a resting state has a spontaneous low-frequency fluctuation brain function network, the frequency band of the network is usually between 0.01HZ and 0.08HZ, and the oscillation wave of the frequency band reflects the excitation degree of cortical local activity and information exchange among brain areas. Low frequency amplitude is an important indicator of decoded brain activity. The analysis of the low-frequency amplitude sub-regions can find out the correlation between the regions, and can study the functional connectivity of the brain region from another angle, thereby providing a new starting point for the brain connectivity study. Based on the above research background, the present invention combines a latest research result: the FSST algorithm is more suitable for analyzing signals with faster frequency conversion, and partial signals in the brain wave signals have more frequent oscillation phenomena, so that the FSST algorithm is more suitable for exploring areas with obvious oscillation of the brain wave signals. In addition, because FSST algorithm data has a large amount of redundancy and large voxels, the invention also provides a classification algorithm for adapting data: ATGP-Kmeans is used for dividing the time frequency data more reliably and improving the value of the brain time frequency signal clustering result.

Disclosure of Invention

The invention aims to study whether dynamic frequency correlation exists between sub-areas in a low-frequency area under time-frequency fluctuation or not from the time-frequency space angle and whether the space connectivity of a brain can be analyzed from the time-frequency angle or not. And then, by remapping the brain image, the correlation of brain voxels in a spatial region is observed so as to verify the idea proposed by the background technology.

In order to achieve the purpose, the invention provides the following technical scheme: the method for reliably dividing the brain low-frequency fluctuation subarea based on Fourier synchronous compression transformation comprises the following steps:

step 1: dividing a low-frequency area, searching areas with low-frequency amplitude extracted from the same tested state at different moments, selecting the areas by adopting a traditional ALFF method, removing some noise areas and selecting the areas with strict verification;

step 2: performing time-frequency reconstruction on the preprocessed functional nuclear magnetic resonance signals, converting the preprocessed functional nuclear magnetic resonance signals into time-frequency domains, and acquiring related time-frequency-power spectrograms to form a time-frequency data set of the nuclear magnetic resonance signals;

and step 3: based on the time-frequency dynamic correlation angle, the ATGP-Kmeans is adopted to adapt to FSST data by utilizing the dynamic synchronicity of different voxels in time-frequency, correlation coefficient calculation is carried out, clustering is completed, the clustering is mapped into a spatial brain map, and the correlation of different brain areas in the space is observed.

As a preferred technical solution of the present invention, in the step 1, a low frequency region is used as a research region for selecting the classification region;

as a preferred technical scheme of the invention, in the step 2 and the step 3, through time-frequency reconstruction of data, an adaptive classification algorithm is introduced, the data are clustered and then mapped into a space, and time-frequency relation of different brain areas embodied on the space is researched.

The selection of the region in the step 1 adopts a traditional ALFF method formula as follows:a_k(f_k)，b_k(f_k) Real and imaginary parts, respectively.

As a preferred technical scheme of the invention, in the step 1, some noise regions are removed, and selectedChecking a stricter area;

is the average number of samples and is the average number of samples,for the standard deviation of the samples, n is the number of samples, and the statistic t is given in the zero hypothesis: the condition that μ ═ μ 0 is true obeys a t distribution with a degree of freedom of n.

In a preferred embodiment of the present invention, the step 2 is performed on the preprocessed functional nuclear magnetic resonance signalThe time-frequency reconstruction formula is as follows:converting the time spectrum into a time-frequency domain to obtain a related time-frequency-power spectrogram; the signal f (t) is a plurality of f_k(t) composition, with the STFT predominant ridge at (t, φ'_k(t)), can be approximated bySubstitute phi'_k(t)。

As a preferred technical solution of the present invention, in the step 3, based on the time-frequency dynamic correlation angle, the dynamic synchronicity of different voxels in time-frequency and the local correlation principle of the brain are utilized, and the low-frequency voxel signals are processed by means of clustering, and due to the redundancy and the complex nature of the signals, an ATGP algorithm is introduced: t is t₁＝arg{max_r[r^Tr]R are all voxels to be observed, and U is^TThe pseudo-inverse of U is set to,as an initial centroid selection algorithm of Kmeans, ATGP-Kmeans is adopted to adapt FSST data, and a loss function is minimized:

as a preferred technical solution of the present invention, in the step 3, the correlation coefficient calculation is performed to complete the clustering, wherein the correlation coefficient calculation is selected:reselecting a class center:and finishing clustering until the maximum distance is reached, mapping the clustering to a spatial brain map, and observing the correlation of different brain areas in the space.

The invention has the beneficial effects that: the method starts with the selection of a low-frequency region, selects the low-frequency region which is subjected to strict T test correction, performs time-frequency reconstruction and clustering algorithm and remaps a spatial map; specifically, time-frequency analysis is carried out on a low-frequency region through an FSST algorithm, a time-frequency signal is obtained and used as representation information of a voxel, an ATGP algorithm is introduced to adapt to a Kmeans algorithm for clustering, and finally a brain map is remapped to obtain a spatial correlation diagram.

Drawings

FIG. 1 is a schematic flow diagram of the present invention;

FIG. 2 is a flow chart of the present invention.

Detailed Description

The following detailed description of the preferred embodiments of the present invention, taken in conjunction with the accompanying drawings, will make the advantages and features of the invention more readily understood by those skilled in the art, and thus will more clearly and distinctly define the scope of the invention.

Example (b): referring to fig. 1-2, the present invention provides a technical solution: the method comprises the following steps:

step 1: dividing a low-frequency area, searching areas with low-frequency amplitude extracted from the same tested state at different moments, selecting the areas by adopting a traditional ALFF method, removing some noise areas and selecting the areas with strict verification;

step 2: performing time-frequency reconstruction on the preprocessed functional nuclear magnetic resonance signals, converting the preprocessed functional nuclear magnetic resonance signals into time-frequency domains, and acquiring related time-frequency-power spectrograms to form a time-frequency data set of the nuclear magnetic resonance signals;

and step 3: based on the time-frequency dynamic correlation angle, the ATGP-Kmeans is adopted to adapt to FSST data by utilizing the dynamic synchronicity of different voxels in time-frequency, correlation coefficient calculation is carried out, clustering is completed, the clustering is mapped into a spatial brain map, and the correlation of different brain areas in the space is observed.

The invention aims to study whether dynamic frequency correlation exists between sub-areas in a low-frequency area under time-frequency fluctuation or not from the time-frequency space angle and whether the space connectivity of the brain can be analyzed from the time-frequency angle or not. Based on data redundancy under the FSST algorithm and a huge voxel group of a brain region, firstly, an ATGP algorithm is introduced to replace a traditional random sample selection algorithm to serve as an initial class center of Kmeans, and therefore the classification efficiency and reliability are improved. And then the brain voxels are observed to be associated in the space region through remapping of the brain map.

For convenience of description, terms specific to the present invention are first defined as follows:

brain low frequency fluctuation subregion:

the brain hypo-frequency fluctuation sub-region is referred to in the present invention as: based on the low-frequency oscillation attribute of human brain activity, when decoding a low-frequency oscillation signal, a main body region of the low-frequency signal is taken as an analysis key point to study whether dynamic association on time frequency exists in sub-regions in the region.

Secondly, the method comprises the following specific steps:

step 1: selection of low-frequency region of brain function nuclear magnetic resonance signal: and dividing a low-frequency area, and searching areas of the low-frequency amplitude of the union set extracted from the same tested state at different times. The traditional ALFF method is adopted for selecting the area, meanwhile, some noise areas are removed, and the area with stricter verification is selected.

Step 1.1, dividing a low-frequency area, and searching areas of union low-frequency amplitude extracted from the same tested state at different times; the selection of the regions uses the conventional ALFF method:a_k(f_k)，b_k(f_k) Real and imaginary parts, respectively.

Step 1.2, removing some noise areas and selectingChecking a stricter area;

is the average number of samples and is the average number of samples,is the standard deviation of the samples and n is the number of samples. The statistic t is at zero hypothesis: the condition that μ ═ μ 0 is true obeys a t distribution with a degree of freedom of n.

Step 2: time-frequency reconstruction and voxel characterization signal formation: performing time-frequency reconstruction on the preprocessed functional nuclear magnetic resonance signals, converting the preprocessed functional nuclear magnetic resonance signals into time-frequency domains, and acquiring related time-frequency-power spectrograms to form a time-frequency data set of the nuclear magnetic resonance signals;

step 2.1, performing time-frequency reconstruction on the preprocessed functional nuclear magnetic resonance signals: the signal f (t) is a plurality of f_k(t) composition, with the STFT predominant ridge at (t, φ'_k(t)), can be approximated bySubstitute phi'_k(t), the time-frequency reconstruction formula is as follows:converting the time spectrum into a time-frequency domain to obtain a related time-frequency-power spectrogram;

step 2.2, changing the time amplitude signal of each voxel into a time-frequency-power spectrum signal to form a time-frequency data set of nuclear magnetic resonance signals;

and step 3: remapping of voxel clustering and space: based on the time-frequency dynamic correlation angle, the ATGP-Kmeans is adopted to adapt to FSST data by utilizing the dynamic synchronicity of different voxels in time-frequency, correlation coefficient calculation is carried out, clustering is completed, the clustering is mapped into a spatial brain map, and the correlation of different brain areas in the space is observed.

And 3.1, processing the low-frequency voxel signals by adopting a clustering means based on a time-frequency dynamic correlation angle and by utilizing the dynamic synchronism of different voxels in time-frequency and the local correlation principle of the brain. Due to the redundant and cumbersome nature of these signals, the ATGP algorithm was introduced: t is t₁＝arg{max_r[r^Tr]R are all voxels to be observed, and U is^TThe pseudo-inverse of U is set to,as an initial centroid selection algorithm for Kmeans, ATGP-Kmeans is adopted to adapt FSST data. Minimization of the loss function:

and 3.2, calculating the distance by adopting a correlation coefficient:reselecting a class center:clustering is completed until the maximum distance is reached. And then mapped into a spatial brain map to observe the association of different brain regions in the space.

The method starts with the selection of a low-frequency region, selects the low-frequency region which is subjected to strict T test correction, performs time-frequency reconstruction and clustering algorithm and remaps a spatial map; specifically, time-frequency analysis is carried out on a low-frequency region through an FSST algorithm, a time-frequency signal is obtained and used as representation information of a voxel, an ATGP algorithm is introduced to adapt to a Kmeans algorithm for clustering, and finally a brain graph is remapped to obtain a spatial correlation graph. The invention is beneficial to the research of the relevance between the low-frequency brain activity areas of the human brain and is applied to the research of mental and mental diseases, brain diseases, occupational plasticity and the like.

The above examples only show some embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention.