Method for dividing target points of brain regions for autism neuromodulation

Abstract

The application discloses a brain region target point division method for autism neural regulation, which comprises the following steps: acquiring resting-state functional magnetic resonance imaging data of an autism patient, and pre-processing the functional magnetic resonance imaging data; extracting back-lateral prefrontal cortex ba9 and ba46 in the pre-processed data as target brain regions by using a Brodmann template; unfolding the pre-processed image data into a two-dimensional time sequence, and performing sliding window processing on the time sequence to obtain a plurality of sliding window time sequences; calculating a functional connection matrix of voxel points in the target brain region and all voxel points in the whole brain gray matter in each sliding window time sequence; performing dimension reduction processing on all the functional connection matrices based on local similarity to obtain a reduced dimension functional connection matrix; acquiring a reference clustering template of a normal person which has been constructed, performing Kmeans clustering on the functional connection matrix of the autism patient according to the clustering cluster number, and obtaining sub-region division of the target brain region.

Description

Technical Field

[0001] This invention relates to a method for dividing brain regions, specifically to a method for dividing brain regions into target points for neuromodulation in autism. Background Technology

[0002] The brain is the most intricate system in the human body, with numerous neurons forming a neural network. Autism, as a typical disorder of this neural network, often involves abnormal functional (neural) connections between multiple brain regions. Current research on neuromodulation of autism primarily focuses on the bilateral dorsolateral prefrontal cortex (DLPFC) as effective therapeutic targets. However, the bilateral DLPFC covers a vast cortical area, and due to variations in brain development, lesions differ between individuals. This makes it difficult for researchers to precisely pinpoint stimulation targets for different subjects in their neuromodulation studies of autism, resulting in inaccurate targeting references. Summary of the Invention

[0003] In view of the above-mentioned shortcomings in the prior art, the brain region target segmentation method for autism neuromodulation provided by the present invention solves the problem that the prior art cannot accurately segment the brain regions of autistic patients.

[0004] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0005] A method for identifying brain region targets for neuromodulation in autism is provided, comprising the following steps:

[0006] S1. Obtain resting-state functional magnetic resonance imaging (fMRI) data of autistic patients and preprocess the fMRI data.

[0007] S2. The Brodmann template was used to extract ba9 and ba46 from the dorsolateral prefrontal cortex of the preprocessed data as target brain regions.

[0008] S3. Expand the preprocessed image data into a two-dimensional time series, and perform sliding window processing on the time series to obtain several sliding window time series;

[0009] S4. Calculate the functional connectivity matrix between the voxel points of the target brain region and all voxel points in the whole brain gray matter in each sliding window time series.

[0010] S5. Based on local similarity, all functional connectivity matrices are dimensionality-reduced to obtain dimensionality-reduced functional connectivity matrices.

[0011] S6. Obtain the reference clustering template of normal individuals that has been constructed. Based on the number of clusters, perform K-means clustering on the functional connectivity matrix of autistic patients to obtain the sub-region division of the target brain region.

[0012] The beneficial effects of this invention are as follows: This scheme combines the already constructed reference clustering template and performs fine division of brain regions based on the functional connectivity matrix of voxel points, finds subregions in which the neural networks of autistic patients show significant differences, and uses these subregions as key research areas for researchers. This can effectively avoid errors caused by artificially locating stimulation targets, thereby ensuring the accuracy of research on the causes of autism.

[0013] Furthermore, methods for constructing reference clustering templates for normal individuals include:

[0014] A1. Acquire resting-state functional magnetic resonance imaging data from normal individuals at different sites and perform preprocessing;

[0015] A2. Obtain the dimensionality reduction functional connection matrix for each normal person using steps S1 to S5;

[0016] A3. Based on the weighted multi-source data fusion, the dimensionality-reduced functional connection matrix of all normal personnel is fused to obtain the final functional connection matrix;

[0017] A4. Calculate the Euclidean distance between voxel points in the final functional connectivity matrix as the similarity matrix, and perform group-level AP clustering on the similarity matrix to obtain the initial group-level clusters.

[0018] A5. Based on the silhouette coefficient and Calinski-Harabaz Index, find the largest cluster silhouette of the initial group-level cluster and obtain the brain regions of the normal group-level sub-regions as reference clustering templates.

[0019] The beneficial effects of the above technical solution are: the results obtained from large sample data from multiple sites are more reliable than those obtained from small samples, taking into account both the time characteristics of the data and reducing computational complexity, and the voxel-based clustering results are more refined.

[0020] Furthermore, step S5 further includes:

[0021] S51. Calculate the Euclidean distance between each row in each functional connectivity matrix, and arrange all the Euclidean distances into a square matrix to form a functional connectivity similarity matrix.

[0022] S52. Merge all functional connectivity similarity matrices of the same normal person / autistic patient to obtain a dynamic functional connectivity similarity matrix;

[0023] S53. Subtract the adjacent two layers in the dynamic function connection similarity matrix to obtain the inter-layer difference, and define a mask matrix. Use the mask matrix to traverse each layer of the inter-layer difference with a preset step size.

[0024] S54. Calculate all positive values ​​to form a positive gradient matrix in each mask region, and calculate all negative values ​​to form a negative gradient matrix.

[0025] S55. Extract the maximum and minimum values ​​from all positive and negative gradient matrices of the inter-layer difference, square them, and rearrange them into local gradient increase matrices Sp. e and the local gradient reduction matrix Sn e ;

[0026] S56. Calculate the variance of all positive gradient matrices and the variance of all negative gradient matrices for the inter-layer differences, constructing positive and negative variance matrices respectively, and normalize them to obtain the normalized positive variance matrix Varp. e and the normalized negative variance matrix Varn e ;

[0027] S57. Increase the matrix Sp for each local gradient. e and the local gradient reduction matrix Sn e A first-order linear fit is performed on the window time series of each element to obtain the dynamic trend matrix KSn. e and KSp e ;

[0028] S58, respectively for Varp e Varn e KSn e and KSp e Reconstruction is performed, and the reconstruction matrix reVarp is obtained respectively. e reVarn e 、reKSn e and reKSp e ;

[0029] S59. Assign each interlayer difference to its corresponding reVarp. e reVarn e 、reKSn e and reKSp e Perform convolution to obtain a dimension-reduced matrix, and then sum all the dimension-reduced matrices to obtain the dimension-reduced functional connection matrix.

[0030] The beneficial effects of the above technical solution are as follows: Since the data of dynamic functional connectivity itself contains data from multiple time layers, it is not possible to simply take the average value of the time layers to obtain a two-dimensional functional connectivity matrix. This solution uses the gradient change of local similarity between time layers to reduce the dimension of the functional connectivity matrix, finds the optimal gradient change direction for dimension reduction, and can preserve the changes of the signals of its neighboring neurons over time as much as possible, thereby better describing the dynamic changes of functional connectivity.

[0031] Dynamic functional connectivity (DFM) can effectively describe the dynamic changes in neuronal signals, but subsequent fine-grained subdivision of DFM across all subjects is computationally intensive and complex. This dimensionality reduction method can better preserve the original dynamic trend while simplifying the computational burden of subsequent fine-grained subdivision (i.e., transforming multi-layer clustering into single-layer clustering).

[0032] Furthermore, KSn e and KSp e The calculation formula is the same, where KSn e The calculation formula is:

[0033]

[0034] Among them, KSn e Sn is the dynamic trend matrix of the inter-layer differences at the e-th level. g，e z is the g-th value in the gradient matrix of the e-th layer; e This is the time series of the e-th layer; For Sn of all layers g，e Mean; For z of all time series layers e Mean; WN is the total number of sliding window time series;

[0035] The formula for calculating the dimension reduction functional connectivity matrix is:

[0036]

[0037] Among them, FCS e For the e-th interlayer difference; reVarp e reVarn e 、reKSn e and reKSp e The Varp values ​​corresponding to the e-th interlayer differences are respectively e Varn e KSn e and KSp e The reconstructed matrix.

[0038] The beneficial effects of the above technical solution are as follows: based on the dynamic change trends between multiple layers, the original multi-layer functional connection matrix is ​​reduced to a two-dimensional functional connection matrix, which simplifies the computational complexity and retains the overall change trend information of the dynamic functional connection.

[0039] Furthermore, the positive gradient matrix and the negative gradient matrix, wherein the formula for calculating the positive gradient matrix is:

[0040] SNP eox =SAP eo(x+1，y) -SAP eo(x-1，y)

[0041] SNP eoy =SAP eo(x，y+1) -SAP eo(x，y-1)

[0042] SNP eo =|SNP eox(x，y) |+|SNP eoy(x，y) |

[0043] Among them, SAP eo The positive gradient matrix of the o-th mask in the e-th interlayer difference; SAP eo(x+1，y) and SAP eo(x-1，y) These represent the next and previous x-axis values ​​within the o-th mask representing the interlayer difference of the e-th layer; SAP eo(x，y+1) and SAP eo(x，y-1) These represent the difference between the e-th interlayer and the previous y-axis value within the o-th mask; SNP eox SNP represents the X-direction gradient component of the o-th mask in the e-th interlayer difference. eoy Let Y be the gradient component of the o-th mask in the e-th interlayer difference.

[0044] The beneficial effects of the above technical solution are: gradient calculation can better reflect the changing trend of interlayer images, and the absolute value is used to replace the square and square root operations to reduce the amount of computation.

[0045] Furthermore, methods for fusing the dimensionality-reduced functional connectivity matrices of all normal individuals based on weighted multi-source data fusion include:

[0046] A31. Merge the dimension-reduced functional connectivity matrices of all normal individuals to form a three-dimensional functional connectivity matrix Fm of [m, m, D], where m is the number of bodies in the target brain region and D is the total number of normal individuals.

[0047] A32. Remove outliers from the 3D functional connectivity matrix Fm, and then standardize each voxel of each normal person in the Fm matrix:

[0048]

[0049] Among them, FM rta Fm represents the functional connection between the r-th and t-th voxels of the a-th normal individual after standardization. rta Fm represents the functional connection between the r-th voxel and the t-th voxel of the a-th normal individual. rtD Functional connectivity between voxels r and t in all normal participants in the study;

[0050] A33. Divide the Fm matrix into m matrices FMx of size [m, D] along the X-axis.t t = 1 ~ m; calculate FMx t The proportion of functional connections (pFC) for all normal users at each site in the matrix:

[0051]

[0052] Among them, pFC uj For the Sth voxel of the uth voxel j The proportion of average functional connections at each site to the total functional connections of all normal users, denoted as [S, m]; FMx t(l) For the Sth j The functional connection value of the l-th normal person at each station on the X-axis; L is the function connection value of the S-th person. j Total number of regular staff at each site; For FMx t Sth j The average functional connectivity matrix of all voxels at each site is of size [1, m]; mFC represents a matrix of size [S, m].

[0053] A34. Calculate the uncertainty epFC of the average functional connection pFC for each station. Sj :

[0054]

[0055] Among them, epFC Sj The size is [1, S];

[0056] A35. Based on the functional connection uncertainty epFC of each station Sj Calculate the functional connectivity weight w of the current voxel. Sj :

[0057]

[0058] Among them, w Sj The size is [1, S];

[0059] A36. Based on the connection weight of all functions of each site, w Sj Calculate FMx t Voxel functional connectivity matrix composed of all normal personnel in the matrix

[0060]

[0061] Among them, FMx t vFMx is the functional connectivity matrix of the t-th voxel on the X-axis. St The size is [1, m];

[0062] A37. Divide the Fm matrix into m matrices of size [m, D] along the Y-axis: FMi t For t = 1 to m, calculate matrix FMY using steps A33 to A36. t All voxel functional connectivity matrices

[0063] A38, for all vFMx along the X-axis Sj Merge to form a pairwise matrix vFMx, and for all vFMy along the Y-axis Sj The components are merged to form a pairwise matrix vFMy;

[0064] A39. Calculate the final functional connectivity matrix vFM based on the pairwise matrices vFMy and vFMx:

[0065] vFM=(vFMx+vFMy T ) / 2

[0066] Among them, vFMy T This is the transpose of vFMy.

[0067] The beneficial effects of the above technical solution are as follows: This solution calculates the information weight of each site based on weighted multi-source data fusion. Compared with single-site data analysis, fusing multi-site data makes the results more robust. It combines the information content of each site to assign weights to each voxel of different sites. Sites with fewer normal personnel do not necessarily contain less information, and sites with more normal personnel have different data variability. By weighting the information content of all voxels at each site, the overall variability of the data can be described more comprehensively.

[0068] Furthermore, methods for preprocessing functional magnetic resonance imaging data include:

[0069] B1. Remove the first N time points of functional magnetic resonance imaging (fMRI) data from autistic patients / normal individuals;

[0070] B2. Register the functional magnetic resonance imaging after step B1 and remove data with head movement greater than 2 mm or 2 degrees in any direction;

[0071] B3. Use the EPI template to match the image processed in step B2 to the EPI template, and resample to a 3*3*3 voxel size, and then remove the linear trend;

[0072] B4. The data with linear trend removed were bandpass filtered at 0.01 to 0.1 Hz, and then the Friston's-24 head motion, white matter, cerebrospinal fluid, and whole brain average signals were regressed using DPARSFA software.

[0073] B5. Remove low-quality data with an average frame shift greater than 0.5 from the data processed in step B4, and then smooth the image using an 8mm half-height full-width Gaussian kernel.

[0074] The beneficial effects of the above technical solution are as follows: This solution uses DPARSFA software to preprocess the acquired images, which can reduce experimental errors, improve the accuracy of subsequent clustering of the reference clustering template, and thus ensure the accuracy of brain region differentiation in autistic patients.

[0075] Furthermore, the formula for calculating the functional connectivity matrix in step S4 is as follows:

[0076]

[0077] Among them, TS mi The time series of all voxel points in the i-th sliding window time series of the target brain region; TS n σTS represents the time series of all voxel points in the gray matter of the whole brain. m σTS represents the standard deviation of the time series of all voxel points in the i-th sliding window time series. n denoted as , where is the standard deviation of the time series of all voxels in the whole brain gray matter; Cov(·) is the covariance function; m is the number of target voxels, and n is the number of voxels in the whole brain gray matter.

[0078] The beneficial effects of the above technical solution are: using voxels of the target brain region and the functional connectivity of the whole brain can better describe the connection between the target brain region and the whole brain, thereby better locating the specific changes in the target brain region. Attached Figure Description

[0079] Figure 1 A flowchart illustrating the brain region target delineation method for neuromodulation in autism. Detailed Implementation

[0080] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0081] refer to Figure 1 , Figure 1 A flowchart illustrating the brain region target delineation method for neuromodulation in autism is shown; for example... Figure 1 As shown, the method S includes steps S1 to S6.

[0082] In step S1, resting-state functional magnetic resonance imaging (fMRI) data of autistic patients are acquired, and the fMRI data are preprocessed.

[0083] In implementation, the preferred methods for preprocessing functional magnetic resonance imaging (fMRI) data in this scheme include:

[0084] B1. Remove the first N time points of functional magnetic resonance imaging (fMRI) data from autistic patients / normal individuals. Since normal individuals may not remain stable during data acquisition, removing the image data from the first N time points can ensure that the image data does not deviate significantly.

[0085] B2. Register the functional magnetic resonance imaging after processing in step B1, and remove data in any direction where the head movement is greater than 2 mm or 2 degrees. This method can prevent the experimental results from being affected by excessive head movement of individual normal personnel. Identify and delete the data of normal personnel with excessive head movement.

[0086] B3. Match the image processed in step B2 to the EPI template and resample it to a 3*3*3 voxel size. Then remove the linear trend. Removing the linear trend can focus the analysis on the fluctuation of the trend data and better analyze the characteristics of the original data.

[0087] B4. Apply a bandpass filter of 0.01 to 0.1 Hz to the data after removing the linear trend, and then use the regression Friston's-24 head motion, white matter, cerebrospinal fluid, and whole brain average signals in the DPARSFA software; use a bandpass filter to remove excessively high or low noise signals generated by other organs of normal individuals.

[0088] B5. Remove low-quality data with an average frame shift greater than 0.5 from the data processed in step B4, and then smooth the image using an 8mm half-height full-width Gaussian kernel.

[0089] In step S2, the Brodmann template is used to extract the ba9 and ba46 of the dorsolateral prefrontal cortex from the preprocessed data as target brain regions; the voxels contained in the ba9 and ba46 of the dorsolateral prefrontal cortex are used as target brain regions, containing a total of 1231 voxels; the whole brain gray matter voxel mask contains a total of 67541 voxels.

[0090] In step S3, the preprocessed image data is expanded into a two-dimensional time series, that is, the original coordinate space [x, y, z, timepoint] is expanded into [x*y*z, timepoint], and the time series is processed by sliding window to obtain several sliding window time series;

[0091] When using a sliding window, WL time points are used as one time window. The length of each sliding window is WS, the overlap rate WO between windows is WS / WL, and the number of sliding windows WN is (TP-WL) / WS, resulting in WN time series TP of length WL. i (i = 1 ~ WN).

[0092] In step S4, the functional connectivity matrix between the voxel points of the target brain region and all voxel points in the whole brain gray matter in each sliding window time series is calculated:

[0093]

[0094] Among them, TS mi The time series of all voxel points in the i-th sliding window time series of the target brain region; TS n σTS represents the time series of all voxel points in the gray matter of the whole brain. m σTS represents the standard deviation of the time series of all voxel points in the i-th sliding window time series. n denoted as , where is the standard deviation of the time series of all voxels in the whole brain gray matter; Cov(·) is the covariance function; m is the number of target voxels, and n is the number of voxels in the whole brain gray matter.

[0095] In step S5, all functional connectivity matrices are dimensionality-reduced based on local similarity to obtain dimensionality-reduced functional connectivity matrices.

[0096] In one embodiment of the present invention, step S5 further includes:

[0097] S51. Calculate the Euclidean distance S between each row in each functional connectivity matrix. ij And all the Euclidean distances are arranged into a square matrix to form a functional connectivity similarity matrix of size m*m; where S ij For ρ mn Any two row vectors ρ i and ρ j Euclidean distance between them:

[0098] S52. Merge all functional connectivity similarity matrices of the same normal individual to obtain a dynamic functional connectivity similarity matrix FCS of size [m, m, WN]. a (a = 1 ~ WN);

[0099] S53. Subtracting adjacent layers from the dynamic functional similarity matrix yields the interlayer differences:

[0100] dFCS e =FCS (m，m，a+1) -FCS (m，m，a)

[0101] Among them, dFCS e For the (a+1)th functional connection matrix FCS (m，m，a+1) Connection matrix FCS with the a-th function (m，m，a) The interlayer difference, e = 1 ~ WN-1.

[0102] Then, a mask matrix of size [k, k] is defined, and the mask matrix is ​​used to traverse each layer with interlayer differences at a preset step size;

[0103] S54. Calculate all positive values ​​to form a positive gradient matrix in each mask region, and calculate all negative values ​​to form a negative gradient matrix.

[0104] Among them, there are positive gradient matrices and negative gradient matrices, where the formula for calculating the positive gradient matrix is:

[0105] SNP eox =SAP eo(x+1，y) -SAP eo(x-1，y)

[0106] SNP eoy =SAP eo(x，y+1) -SAP eo(x，y-1)

[0107] SNP eo =|SNP eox(x，y) |+|SNP eoy(x，y) |

[0108] Among them, SAP eo The positive gradient matrix of the o-th mask in the e-th interlayer difference; SAP eo(x+1，y) and SAP eo(x-1，y) These represent the next and previous x-axis values ​​within the o-th mask representing the interlayer difference of the e-th layer; SAP eo(x，y+1) and SAP eo(x，y-1) These represent the difference between the e-th interlayer and the previous y-axis value within the o-th mask; SNP eox SNP represents the X-direction gradient component of the o-th mask in the e-th interlayer difference. eoy Let Y be the gradient component of the o-th mask in the e-th interlayer difference.

[0109] S55. Extract the maximum and minimum values ​​from all positive and negative gradient matrices of the inter-layer difference, square them, and rearrange them into local gradient increase matrices Sp. e and the local gradient reduction matrix Sn e ;

[0110] S56. Calculate the variance of all positive gradient matrices and the variance of all negative gradient matrices for the inter-layer differences, constructing positive and negative variance matrices respectively, and normalize them to obtain the normalized positive variance matrix Varp. e and the normalized negative variance matrix Varn e ;

[0111] S57. Increase the matrix Sp for each local gradient. e and the local gradient reduction matrix Sn e A first-order linear fit is performed on the window time series of each element to obtain the dynamic trend matrix KSn. e and KSp e ;

[0112] KSn e and KSp e The calculation formula is the same, where KSn e The calculation formula is:

[0113]

[0114] Among them, KSn e Sn is the dynamic trend matrix of the inter-layer differences at the e-th level. g，e z is the g-th value in the gradient matrix of the e-th layer; e This is the time series of the e-th layer; For Sn of all layers g，e Mean; For z of all time series layers e Mean; WN is the total number of sliding window time series;

[0115] S58, Varp respectively e Varn e KSn e and KSp e Reconstruction is performed, and the reconstruction matrix reVarp is obtained respectively. e reVarn e 、reKSn e and reKSp e The reconstruction method is as follows: multiply each element in the above matrix by a matrix of size [k, k] consisting entirely of 1s (all values ​​are 1).

[0116]

[0117] Among them, KSn ef To traverse all points in the KSn matrix; mask eo It is the o-th mask for the e-th interlayer difference, with a value of 1, and also contains the mask's position information.

[0118] S59. Assign each interlayer difference to its corresponding reVarp. e reVarn e 、reKSn e and reKSp e Perform convolution to obtain a dimension-reduced matrix, and then sum all the dimension-reduced matrices to obtain the dimension-reduced functional connection matrix.

[0119] The formula for calculating the dimension reduction functional connectivity matrix is:

[0120]

[0121] Among them, FCS e For the e-th interlayer difference; reVarp e reVarn e 、reKSn e and reKSp e The Varp values ​​corresponding to the e-th interlayer differences are respectively e Varn e KSn e and KSp e The reconstructed matrix.

[0122] In step S6, a reference clustering template of normal individuals is obtained, and K-means clustering is performed on the functional connectivity matrix based on the number of clusters to obtain the subregion division of the target brain region.

[0123] In one embodiment of the present invention, a method for constructing a reference clustering template for normal individuals includes:

[0124] A1. Acquire resting-state functional magnetic resonance imaging data from normal individuals at different sites and perform preprocessing. The preprocessing method here is the same as that in steps B1 to B5.

[0125] This study used resting-state functional magnetic resonance imaging (fMRI) data from healthy individuals acquired using different MRI scanners at eight hospitals. The data included 224 healthy men and 48 healthy women, all with no history of significant physical or mental illnesses such as autism, schizophrenia, or depression. Acquisition parameters were as follows: repeat scan time (TR) of 1500ms or 2000ms; acquisition time points (Number of Measurements) of 240, 200, 180, 160, and 150°; and field of view (FOV) of 200*200, 220*220, 240*240, 224*224, and 230*230°.

[0126] A2. Obtain the dimensionality reduction functional connection matrix for each normal person using steps S1 to S5;

[0127] A3. Based on the weighted multi-source data fusion, the dimensionality-reduced functional connection matrix of all normal personnel is fused to obtain the final functional connection matrix;

[0128] A4. Calculate the Euclidean distance between voxel points in the final functional connectivity matrix as the similarity matrix, and perform group-level AP clustering on the similarity matrix to obtain the initial group-level clusters.

[0129] A5. Based on the silhouette coefficient and Calinski-Harabaz Index, find the largest cluster silhouette of the initial group-level cluster and obtain the brain regions of the normal group-level sub-regions as reference clustering templates.

[0130] In one embodiment of the present invention, a method for fusing the dimensionality-reduced functional connectivity matrix of all normal individuals based on weighted multi-source data fusion includes:

[0131] A31. Merge the dimension-reduced functional connectivity matrices of all normal individuals to form a three-dimensional functional connectivity matrix Fm of [m, m, D], where m is the number of bodies in the target brain region and D is the total number of normal individuals.

[0132] A32. Remove outliers from the 3D functional connectivity matrix Fm, and then standardize each voxel of each normal person in the Fm matrix:

[0133]

[0134] Among them, FM rta Fm represents the functional connection between the r-th and t-th voxels of the a-th normal individual after standardization. rta Fm represents the functional connection between the r-th voxel and the t-th voxel of the a-th normal individual. rtD Functional connectivity between voxels r and t in all normal participants in the study;

[0135] A33. Divide the Fm matrix into m matrices FMx of size [m, D] along the X-axis. t t = 1 ~ m; calculate FMx t The proportion of functional connections (pFC) for all normal users at each site in the matrix:

[0136]

[0137] Among them, pFC uj For the Sth voxel of the uth voxel j The proportion of average functional connections at each site to the total functional connections of all normal users, denoted as [S, m]; FMx t(l) For the Sth j The functional connection value of the l-th normal person at each station on the X-axis; L is the function connection value of the S-th person.j Total number of regular staff at each site; For FMx t Sth j The average functional connectivity matrix of all voxels at each site is of size [1, m]; mFC represents a matrix of size [S, m].

[0138] A34. Calculate the uncertainty epFC of the average functional connection pFC for each station. Sj :

[0139]

[0140] Among them, epFC Sj The size is [1, S];

[0141] A35. Based on the functional connection uncertainty epFC of each station Sj Calculate the functional connectivity weight w of the current voxel. Sj :

[0142]

[0143] Among them, w Sj The size is [1, S];

[0144] A36. Based on the connection weight of all functions of each site, w Sj Calculate FMx t Voxel functional connectivity matrix composed of all normal personnel in the matrix

[0145]

[0146] Among them, FMx t vFMx is the functional connectivity matrix of the t-th voxel on the X-axis. St The size is [1, m];

[0147] A37. Divide the Fm matrix into m matrices of size [m, D] along the Y-axis: FMi t For t = 1 to m, calculate matrix FMY using steps A33 to A36. t All voxel functional connectivity matrices

[0148] A38, for all vFMx along the X-axis Sj Merge to form a pairwise matrix vFMx, and for all vFMy along the Y-axis Sj The components are merged to form a pairwise matrix vFMy;

[0149] A39. Calculate the final functional connectivity matrix vFM based on the pairwise matrices vFMy and vFMx:

[0150] vFM=(vFMx+vFMy T ) / 2

[0151] Among them, vFMy T This is the transpose of vFMy.

[0152] In summary, this approach divides the DLPFC into subregions to identify subregions where brain neural networks exhibit significant differences. These subregions are then designated as key research areas for researchers, thereby ensuring the accuracy of their research focus.

Claims

1. A method for delineating brain regions as targets for neuromodulation in autism, characterized in that, Including the following steps: S1. Obtain resting-state functional magnetic resonance imaging (fMRI) data of autistic patients and preprocess the fMRI data. S2. The Brodmann template was used to extract ba9 and ba46 from the dorsolateral prefrontal cortex of the preprocessed data as target brain regions. S3. Expand the preprocessed image data into a two-dimensional time series, and perform sliding window processing on the time series to obtain several sliding window time series; S4. Calculate the functional connectivity matrix between the voxel points of the target brain region and all voxel points in the whole brain gray matter in each sliding window time series. S5. Based on local similarity, all functional connectivity matrices are dimensionality-reduced to obtain dimensionality-reduced functional connectivity matrices. S6. Obtain the constructed reference clustering template for normal individuals, and perform K-means clustering on the functional connectivity matrix of autistic patients based on the number of clusters to obtain the sub-region division of the target brain region. Methods for constructing reference clustering templates for normal individuals include: A1. Acquire resting-state functional magnetic resonance imaging data from normal individuals at different sites and perform preprocessing; A2. Obtain the dimensionality reduction functional connection matrix for each normal person using steps S1 to S5; A3. Based on the weighted multi-source data fusion, the dimensionality-reduced functional connection matrix of all normal personnel is fused to obtain the final functional connection matrix; A4. Calculate the Euclidean distance between voxel points in the final functional connectivity matrix as the similarity matrix, and perform group-level AP clustering on the similarity matrix to obtain the initial group-level clusters. A5. Based on the silhouette coefficient and Calinski Harabaz Index, find the largest cluster silhouette of the initial group-level cluster and obtain the brain regions of the normal group-level sub-regions as reference clustering templates.

2. The brain region target delineation method for autism neural modulation according to claim 1, characterized in that, Step S5 further includes: S51. Calculate the Euclidean distance between each row in each functional connectivity matrix, and arrange all the Euclidean distances into a square matrix to form a functional connectivity similarity matrix. S52. Merge all functional connectivity similarity matrices of the same normal person / autistic patient to obtain a dynamic functional connectivity similarity matrix; S53. Subtract the adjacent two layers in the dynamic function connection similarity matrix to obtain the inter-layer difference, and define a mask matrix. Use the mask matrix to traverse each layer of the inter-layer difference with a preset step size. S54. Calculate all positive values ​​to form a positive gradient matrix in each mask region, and calculate all negative values ​​to form a negative gradient matrix. S55. Extract the maximum and minimum values ​​from all positive and negative gradient matrices of the inter-layer differences, square them, and rearrange them into local gradient increase matrices. and local gradient reduction matrix ; S56. Calculate the variance of all positive gradient matrices and the variance of all negative gradient matrices for the inter-layer differences, construct the positive variance matrix and the negative variance matrix respectively, and normalize them to obtain the normalized positive variance matrix. and normalized negative variance matrix ; S57. Add matrices to the local gradients respectively. and local gradient reduction matrix A first-order linear fit is performed on the window time series of each element to obtain the dynamic change trend matrix. and ; S58, respectively for , , and Reconstruction is performed to obtain reconstruction matrices. , , and ; S59. Assign each interlayer difference to its corresponding... , , and Perform convolution to obtain a dimension-reduced matrix, and then sum all the dimension-reduced matrices to obtain the dimension-reduced functional connection matrix.

3. The brain region target delineation method for autism neuromodulation according to claim 2, characterized in that, and The calculation formula is the same, where The calculation formula is: in, For the first e Dynamic trend matrix of inter-layer differences; For the first e The first layer gradient matrix g One value; For the first e Layered time series; For all layers Mean; For all time series layers Mean; This represents the total number of sliding window time series; The formula for calculating the dimension reduction functional connectivity matrix is: in, For the first e Differences between layers; , , and The first e The inter-layer differences correspond to , , and The reconstructed matrix.

4. The method for brain region target delineation for autism neuromodulation according to claim 2, characterized in that, The positive gradient matrix and the negative gradient matrix, wherein the formula for calculating the positive gradient matrix is: in, For the first e The first of the inter-layer differences o The positive gradient matrix of each mask; and The first e The first interlayer difference o Inside the mask x The next value and the previous value on the axis; and The first e Inter-layer differences o Inside the mask y The next value and the previous value on the axis; For the first e The first of the inter-layer differences o A mask X Oriented gradient components; For the first e The first of the inter-layer differences o A mask Y Oriented gradient components.

5. The method for brain region target delineation for autism neuromodulation according to claim 1 or 2, characterized in that, Methods for fusing the dimensionality-reduced functional connectivity matrix of all normal individuals based on weighted multi-source data fusion include: A31. Merge the dimension-reduced functional connection matrices of all normal personnel to form a [ m , m , D ] three-dimensional functional connection matrix Fm , m The number of voxels in the target brain region. D This represents the total number of normal personnel. A32. Remove the 3D functional connection matrix Fm Outliers, then... Fm Each voxel of each normal person in the matrix is ​​standardized: in, For the standardized first a The first normal person r Individual elements and the first t Functional connectivity of individual elements; For the first a The first normal person r Individual elements and the first t Functional connectivity of individual elements; For all normal personnel participating in the study r Individual elements and the first t Functional connectivity of individual elements; A33, along Fm The matrix is ​​divided into m matrices of size [m, D] by the X-axis. ;calculate The proportion of functional connections of all normal personnel at each site in the matrix : ， in, For the first u The first individual element The average percentage of functional connections at each site relative to the total functional connections of all normal users is [ ]. S , m ]; For the first The first site l Functional connection values ​​for a normal person on the X-axis; L For the first S j Total number of regular staff at each site; for The Middle The average functional connectivity matrix of all voxels at each site is of size [1, ..., ...]. m ]; Indicates a size of [S, m A matrix of ], where ; A34. Calculate the average functional connections for each site. uncertainty : in, The size is [1, S]; A35. Based on the functional connection uncertainty of each station Calculate the functional connectivity weights of the current voxel. : in, The size is [1, S]; A36. Based on the connection weight of all functions of each site ,calculate Voxel functional connectivity matrix composed of all normal personnel in the matrix : in, For the functional connection matrix X-axis t Functional connectivity matrix of individual elements; Its size is [1, m]; A37. Divide the Fm matrix into m matrices of size [m, D] along the Y-axis. The matrix is ​​calculated using steps A33-A36. All voxel functional connectivity matrices ; A38, All X-axis Merge to form a pair matrix For all of the Y-axis Merge to form a pair matrix ; A39. Based on the complementary matrix and Calculate the final functional connectivity matrix. : in, for The transpose of .

6. The method for brain region target delineation for autism neuromodulation according to claim 2, characterized in that, Methods for preprocessing functional magnetic resonance imaging (fMRI) data include: B1. Remove the first N time points of functional magnetic resonance imaging (fMRI) data from autistic patients / normal individuals; B2. Register the functional magnetic resonance imaging after step B1 and remove data with head movement greater than 2 mm or 2 degrees in any direction; B3. Match the image processed in step B2 to the EPI template, resample to a 3*3*3 voxel size, and then remove the linear trend. B4. The data with linear trend removed were bandpass filtered at 0.01~0.1Hz, and then the Friston's-24 head motion, white matter, cerebrospinal fluid and whole brain average signals were regressed using DPARSFA software. B5. Remove low-quality data with an average frame shift greater than 0.5 from the data processed in step B4, and then smooth the image using an 8mm half-height full-width Gaussian kernel.

7. The method for delineating brain regions for neuromodulation of autism according to any one of claims 1-4 and 6, characterized in that, The formula for calculating the functional connectivity matrix in step S4 is: in, For the target brain region i A sliding window time series of all voxel points; The time series of all voxel points in the gray matter of the whole brain; For the first i The standard deviation of the time series of all voxel points in a sliding window time series; The standard deviation of the time series of all voxels in the whole brain gray matter; (•) represents the covariance function; m The number of target voxels. n The number of voxels in the gray matter of the whole brain.

Images