Individual functional brain region segmentation method and device based on improved density clustering

Abstract

The present invention discloses a method and device for individual functional brain region segmentation based on improved density clustering, which includes obtaining resting-state functional magnetic resonance imaging data of the individual subject to calculate the functional connection matrix and feature space distance matrix of the whole brain functional nodes; respectively calculating the local density and confidence of each functional node for each brain region, introducing the adjacency matrix to retain the labels of high-confidence functional nodes in the group map, combining the feature space distance and the geodesic distance to calculate the integrated distance, and assigning labels to the remaining functional nodes in the brain region according to the principle of being consistent with the higher local density point of the nearest integrated distance, thereby obtaining the preliminary individual functional brain region segmentation results of the brain region, performing correlation maximization processing until iterative convergence, and performing spatial continuity processing to obtain the final individual functional brain region segmentation results. The present invention aims to solve the problem of how to obtain effective and reliable individual functional brain region segmentation results using shorter resting-state functional magnetic resonance data.

Description

Technical Field

[0001] The present invention belongs to the field of medical image processing and neuroimaging, and in particular relates to a method and a device for segmenting individual functional brain regions based on improved density clustering. Background Art

[0002] The human brain is a very complex biological system with hundreds of billions of neurons. Neurons are connected by complex nerve fibers, forming neural networks through trillions of synapses, which constitute the complex cognitive and behavioral functions of the human brain. Human neuroscience research has long been trying to assign specific functions to clearly divided brain regions. Dividing the human brain into brain regions with different functions and constructing brain maps from them will help understand the connection and information transmission between different brain regions, and provide regions of interest (ROI) for other brain studies. Individual-level brain region dissection can reveal the differences in brain tissue between different individuals, thereby better understanding the neural basis of cognitive and behavioral differences between individuals, and better aligning brain regions between individuals during group analysis. In addition, individual-level dissection plays a key role in the study of brain diseases. By comparing the brain maps of patients and healthy people, it provides strong support for the diagnosis, etiology research and predicted progress of the disease. At the same time, the fine division of individual brain regions helps to accurately locate specific areas in the brain in medical treatment and assist clinical treatment. For example, transcranial magnetic stimulation (TMS) technology can effectively stimulate specific functional brain areas or brain networks to treat a variety of mental and neurological diseases. However, the therapeutic effect of TMS is difficult to guarantee in clinical practice, and only 30-40% of patients with refractory depression are relieved in conventional TMS treatment programs. This is often because conventional TMS treatment uses a standard universal cortical structure image template to locate the patient's brain area, ignoring individual differences. The expected functional brain area or brain network is not stimulated, thus affecting the clinical effect. Therefore, whether for scientific research or actual medical applications, the fine dissection of individualized brain areas is of great significance and value.

[0003] The information interaction of the human brain at multiple spatiotemporal scales is the basis of its complex cognitive and behavioral functions. The BOLD signals between specific brain regions have temporal correlation, namely functional connectivity, which is closely related to brain activity and function. Functional magnetic resonance imaging (fMRI), as a non-invasive in vivo brain functional imaging method, has high spatiotemporal resolution and provides an effective means for studying the functional connectivity pattern of the human brain. Among them, resting-state functional connectivity (RSFC) calculated based on resting-state fMRI data has become an effective means of brain region segmentation. At present, the technology of brain region segmentation at the group level is relatively mature. However, due to the influence of factors such as the low signal-to-noise ratio of individual fMRI data and the uncertainty of individual differences, the group-level segmentation method cannot be directly applied to individual data. In response to this problem, there are usually two solutions for individual-level dissection. One is to conduct long-term, multi-round scans on a single subject. This method can obtain reliable individual brain area divisions, but it is obviously not practical in actual clinical applications because it is time-consuming and labor-intensive. The second is to use the group-level brain map as the initial template for individual dissection, and obtain the final results of individual brain areas through iterative adjustments. This is also a more commonly used method for individual-level dissection. However, some previous methods have problems such as too strong group prior information constraints, overly complex models, and time-consuming training. In addition, most individual-level dissections are performed at the brain network level, and there are relatively few dissections at the brain region level. Summary of the invention

[0004] Technical problem to be solved by the present invention: In view of the above-mentioned problems in the prior art, a method and device for individual functional brain area segmentation based on improved density clustering are provided. The present invention aims to solve the problem of how to obtain effective and reliable individual functional brain area segmentation results using shorter resting-state functional magnetic resonance imaging data.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for individual functional brain region segmentation based on improved density clustering comprises the following steps:

[0007] S1, obtain the resting-state functional magnetic resonance imaging data of the individual subjects;

[0008] S2, perform data preprocessing to extract node time series of functional nodes, where the functional nodes are cortical vertices or subcutaneous nuclei;

[0009] S3, calculating the functional connection matrix of the whole brain functional nodes according to the node time series of the functional nodes;

[0010] S4, calculating the feature space distance between the functional nodes according to the functional connection matrix to obtain the feature space distance matrix;

[0011] S5, using the initial brain region template to divide the brain region, calculating the local density and confidence of each functional node in each brain region, and introducing the adjacency matrix to retain the labels of high-confidence functional nodes in the group map;

[0012] S6, respectively, calculate the integrated distance for each brain region by combining the feature space distance and the geodesic distance, and assign labels to the remaining functional nodes in the brain region according to the principle of being consistent with the higher local density point of the nearest integrated distance, so as to obtain the preliminary individual functional brain region dissection results;

[0013] S7, perform correlation maximization processing on each brain region until iterative convergence;

[0014] S8, perform spatial continuity processing on each brain region to obtain the final individual functional brain region dissection results.

[0015] Optionally, calculating the feature space distance matrix according to the functional connectivity matrix in step S4 includes:

[0016] S4.1, measuring the similarity between two functional nodes in the functional connection matrix by the correlation ratio, thereby obtaining a similarity matrix composed of the similarities between two functional nodes;

[0017] S4.2, reducing the dimension of the similarity matrix by Laplace eigenmap algorithm LE to obtain the normalized Laplace eigenvector;

[0018] S4.3, for the normalized Laplace eigenvector, the distance between each pair of functional nodes in the low-dimensional Laplace feature space is calculated as the feature space distance, and a feature space distance matrix consisting of the feature space distances between each pair of functional nodes is obtained.

[0019] Optionally, the processing for any brain region in step S5 includes:

[0020] S5.1, combined with the feature space distance matrix, calculate the local density of each functional node in the current brain region according to the following formula:

[0021] ,

[0022] In the above formula, For functional nodes The local density of For functional nodes To function node The feature space distance of is the cutoff distance, which is used to ensure that all points are within the cutoff distance The average number of functional nodes within the range accounts for a proportion of the total number of nodes that is not less than the preset threshold;

[0023] S5.2, combined with the feature space distance matrix, calculate the nearest neighbor distance of each functional node according to the following formula:

[0024] ,

[0025] In the above formula, For functional nodes The nearest neighbor distance, For functional nodes The local density of

[0026] S5.3, calculate the confidence of the function node based on the local density and nearest neighbor distance of the function node:

[0027] ,

[0028] In the above formula, For functional nodes confidence level;

[0029] S5.4, traverse the functional nodes in each sub-brain region of the current brain region in order of confidence from high to low, and retain the functional node labels of the group map in combination with the introduced adjacency matrix until the proportion of functional nodes assigned with labels reaches a preset proportion threshold; the adjacency matrix records the adjacent relationship between functional nodes in the spatial grid, and the element in the adjacency matrix is ​​"1" indicating that the corresponding two functional nodes are adjacent and share a triangle edge, and "0" indicating that the corresponding two functional nodes are not adjacent.

[0030] Optionally, step S5.4 includes:

[0031] S5.4.1, traverse each sub-brain region corresponding to the current brain region in the group atlas;

[0032] S5.4.2, find the functional node with the highest confidence in the current sub-brain region, and find its adjacent functional nodes according to the adjacency matrix; the adjacency matrix records the adjacent relationship between the functional nodes in the spatial grid, and an element in the adjacency matrix of "1" indicates that the corresponding two functional nodes are adjacent and share a triangle edge, and an element of "0" indicates that the corresponding two functional nodes are not adjacent;

[0033] S5.4.3, assigning the label of the functional node with the highest confidence as a group-level label, and also assigning the functional nodes whose group-level labels are consistent with the label of the functional node among its adjacent functional nodes as group-level labels; jumping to step S5.4.2 until the proportion of functional nodes assigned with labels in the current sub-brain region reaches a preset proportion threshold;

[0034] S5.4.4, determine whether the sub-brain regions of the current brain region in the group atlas have been traversed. If not, jump to step S5.4.1 to continue traversing the remaining sub-brain regions. Otherwise, determine that the operation of retaining the high-confidence functional node labels in the group atlas for the current brain region has been completed, and jump to step S6.

[0035] Optionally, the processing for any brain region in step S6 includes:

[0036] S6.1, the adjacency matrix is ​​regarded as an undirected graph between functional nodes, the paths between functional nodes with adjacency relationship are recorded as 1, and the shortest paths between all functional nodes are calculated as the geodesic distance between functional nodes;

[0037] S6.2, normalize the geodesic distance and feature space distance between functional nodes and average them to obtain the integrated distance;

[0038] S6.3, traverse all functional nodes in the brain region in order of local density from large to small. If the traversed functional node has been assigned a label, skip it; if the traversed functional node has not yet been assigned a label, find a set of functional nodes with a higher local density than the traversed functional node, and assign the label of the functional node closest to the traversed functional node in the set of functional nodes with higher local density to the traversed functional node, and finally obtain the preliminary individual functional brain region dissection results.

[0039] Optionally, the processing for any brain region in step S7 includes:

[0040] S7.1, traverse the function nodes in descending order of local density. If the traversal is not completed, jump to the next step; otherwise, jump to step S7.5;

[0041] S7.2, searching for adjacent functional nodes of the current functional node according to the adjacency matrix, and determining the functional nodes included in all categories corresponding to the adjacent functional nodes of the current functional node based on the preliminary individual functional brain region dissection results of the brain region;

[0042] S7.3, calculating the average distance between the current function node and the function nodes included in all categories in step S7.2 according to the feature space distance matrix, so as to measure the relevance of the current function node to each category, and the smaller the average distance, the stronger the relevance;

[0043] S7.4, if the label of the current function node is consistent with the category with the smallest average distance, then skip the current function node; otherwise, modify the label of the current function node to the category with the smallest average distance;

[0044] S7.5, determine whether the proportion of functional nodes whose labels do not change in this round exceeds the preset ratio. If it exceeds the preset ratio, it is determined that the correlation maximization processing of the individual functional brain area dissection results of the brain area is completed, and jump to step S8; otherwise, jump to step S7.1 to enter the next round of functional node traversal.

[0045] Optionally, the processing of any brain region in step S8 includes: judging whether the brain region contains a spatially discontinuous connected domain based on the individual functional brain region dissection results optimized for each brain region after the correlation maximization processing, retaining the connected domain containing the largest number of functional nodes for the brain region containing the spatially discontinuous connected domain, and reallocating labels to the remaining connected domains according to the category with the most labels in its peripheral functional nodes, and merging them with the peripheral brain regions, and obtaining the final individual functional brain region dissection results after all brain regions complete the above processing; the judging whether the brain region contains a spatially discontinuous connected domain includes using an adjacency matrix to find the connection relationship between functional nodes, and if two connected domains do not contain a pair of functional nodes with a connection relationship, then the two connected domains are spatially discontinuous connected domains.

[0046] In addition, the present invention also provides an individual functional brain region segmentation system based on improved density clustering, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering.

[0047] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering through a processor.

[0048] In addition, the present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering through a processor.

[0049] Compared with the prior art, the present invention mainly has the following advantages:

[0050] 1. The individual functional brain area segmentation method based on improved density clustering of the present invention can reliably and effectively segment short-term resting-state functional magnetic resonance imaging data to obtain more detailed individual functional brain area segmentation results, which has important application value in clinical applications.

[0051] 2. The present invention retains the group graph labels of high-confidence vertices in combination with the individual's own characteristics, thereby avoiding the situation where the individual graph is overly constrained by the group's prior information and taking into account the individual's specificity.

[0052] 3. The present invention comprehensively considers feature space distance and geodesic distance to perform label allocation, which greatly reduces the impact of low data signal-to-noise ratio.

[0053] 4. The individualized functional brain region dissection method proposed in the present invention does not require training. It only needs to input individual data to obtain the individual functional brain region dissection results. It can be directly applied to a single individual and has high efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of the basic flow of the method of the embodiment of the present invention.

[0055] Figure 2 A schematic diagram of a strategy for retaining high-confidence function node labels in a group graph in an embodiment of the present invention.

[0056] Figure 3 Schematic diagram of a correlation maximization processing method in an embodiment of the present invention.

[0057] Figure 4 It is a schematic diagram of calculating the individual recognition accuracy in an embodiment of the present invention.

[0058] Figure 5 This is the verification result of task activation heterogeneity within individual functional brain regions in an embodiment of the present invention (I).

[0059] Figure 6 This is the verification result of task activation heterogeneity within individual functional brain regions in the embodiment of the present invention (II). DETAILED DESCRIPTION

[0060] like Figure 1 As shown, the individual functional brain region segmentation method based on improved density clustering in this embodiment includes the following steps:

[0061] S1, obtain the resting-state functional magnetic resonance imaging data of the individual subjects;

[0062] S2, perform data preprocessing to extract node time series of functional nodes, where the functional nodes are cortical vertices or subcutaneous nuclei;

[0063] S3, calculating the functional connection matrix of the whole brain functional nodes according to the node time series of the functional nodes;

[0064] S4, calculating the feature space distance between the functional nodes according to the functional connection matrix to obtain the feature space distance matrix;

[0065] S5, using the initial brain region template to divide the brain region, calculating the local density and confidence of each functional node in each brain region, and introducing the adjacency matrix to retain the labels of high-confidence functional nodes in the group map;

[0066] S6, respectively, calculate the integrated distance for each brain region by combining the feature space distance and the geodesic distance, and assign labels to the remaining functional nodes in the brain region according to the principle of being consistent with the higher local density point of the nearest integrated distance, so as to obtain the preliminary individual functional brain region dissection results;

[0067] S7, perform correlation maximization processing on each brain region until iterative convergence;

[0068] S8, perform spatial continuity processing on each brain region to obtain the final individual functional brain region dissection results.

[0069] In step S1 of this embodiment, resting-state functional magnetic resonance imaging data of the individual subject is obtained. By using resting-state functional magnetic resonance imaging data of a relatively short time, an effective and reliable individual brain map can be obtained, which has great potential value in clinical practice.

[0070] In step S2 of this embodiment, the node time series of data preprocessing and function node extraction is implemented based on the statistical parameter mapping toolkit SPM, and the main process includes: removing the first 5 frames of scanning data to avoid magnetic saturation effect, inter-layer time correction, head motion correction (single frame rotation angle greater than 2.5° or translation greater than 2.5mm), spatial standardization to align to the standard MNI152 space, Gaussian space smoothing (Gaussian kernel half-width is 6mm), removing linear drift and time domain filtering (Chebyshev bandpass filter: 0.01~0.08Hz), and irrelevant variable regression. Finally, the cerebral cortex surface is reconstructed with the help of the FreeSurfer toolkit, and the cerebral cortex data is projected into the 32k_fs_LR space according to the cross-individual regional matching algorithm.

[0071] In step S2 of this embodiment, the node time series of the functional nodes is extracted for brain region segmentation. It should be noted that the method of this embodiment is applicable to cortical vertices and subcutaneous nuclei, and the cortical vertex data is used as an example in this embodiment.

[0072] In step S3 of this embodiment, when calculating the functional connection matrix of the whole brain functional nodes according to the node time series of the functional nodes, in order to improve the calculation efficiency, reduce the storage requirement, and avoid information loss as much as possible, 1483 cortical vertices evenly distributed throughout the brain are selected as the region of interest ROI, and the Pearson correlation coefficient from all vertices of the whole brain to the region of interest ROI is calculated, and Fisher-z transformation is performed to obtain the functional connection matrix of the whole brain functional nodes.

[0073] In step S4 of this embodiment, calculating the feature space distance matrix according to the functional connectivity matrix includes:

[0074] S4.1, by correlation ratio (eta 2) is used to measure the similarity between two functional nodes in the functional connection matrix, thereby obtaining a similarity matrix composed of the similarities between two functional nodes;

[0075] S4.2, the similarity matrix is ​​reduced in dimension by the Laplace eigenmap algorithm LE to obtain a normalized Laplace eigenvector; when the similarity matrix is ​​reduced in dimension by the Laplace eigenmap algorithm LE, the eigenvector dimension is the optimal dimensionality reduction dimension obtained when performing group level partitioning, which can be considered as a known condition. The dimensionality reduction dimensions of different brain regions are different. The present invention has also verified in experiments that the dimensionality reduction dimension can be randomly selected within the range of 3 to 15, and relatively reliable results can be obtained;

[0076] S4.3, for the normalized Laplace eigenvector, the distance between each pair of functional nodes in the low-dimensional Laplace feature space is calculated as the feature space distance, and a feature space distance matrix consisting of the feature space distances between each pair of functional nodes is obtained.

[0077] In this embodiment, when the initial brain region template is used to divide the brain region in step S5, the DK (Desikan-Killiany) template is specifically used as the initial brain region template. The DK template contains a total of 68 brain regions, each of which corresponds to an independent functional connection matrix of the brain region. The task of individual functional brain region segmentation is to finely divide these brain regions into sub-regions with different functions. It should be noted that the initial brain region template can also select other general templates. In addition, in order to make the segmentation result more accurate, the DK template of the subject itself can be used, or the group average template can be used. The difference with the individual template is very small and will not have a significant impact on the segmentation result. This example uses the group average DK template determined by the voting method.

[0078] Taking into account the lateralization of the left and right hemispheres, the brain regions across the left and right hemispheres are divided separately in this embodiment, so when calculating the feature space distance matrix, the left and right hemisphere brain regions are calculated separately.

[0079] In step S5 of this embodiment, the processing for any brain region includes:

[0080] S5.1, combined with the feature space distance matrix, calculate the local density of each functional node in the current brain region according to the following formula:

[0081] ,

[0082] In the above formula, For functional nodes The local density of For functional nodes To function node The feature space distance of is the cutoff distance, which is used to ensure that all points are within the cutoff distance The average number of functional nodes in the range accounts for a proportion of the total number of points not less than the preset threshold. For example, in this embodiment, the preset threshold is 2%. The cutoff distance is used to ensure that all points are within the cutoff distance. The average number of functional nodes within the scope accounts for no less than 2% of the total number of points;

[0083] S5.2, combined with the feature space distance matrix, calculate the nearest neighbor distance of each functional node according to the following formula:

[0084] ,

[0085] In the above formula, For functional nodes The nearest neighbor distance, For functional nodes The local density of ; for the point with the highest local density, it is defined as: ,in Indicates taking The largest , For functional nodes To function node The feature space distance of

[0086] S5.3, calculate the confidence of the function node based on the local density and nearest neighbor distance of the function node:

[0087] ,

[0088] In the above formula, For functional nodes The higher the confidence, the more likely it is to retain the group graph information;

[0089] S5.4, the functional nodes in each sub-brain region of the current brain region are traversed in order from high to low confidence, and the functional node labels of the group map are retained in combination with the introduced adjacency matrix until the proportion of functional nodes assigned with labels reaches a preset ratio threshold; in this embodiment, the group map is a group level segmentation template, which can be the current mainstream brain region level group map, or it can be a group map obtained by group level segmentation using existing data. The group map used in this embodiment is the group level segmentation result obtained by feature dimension reduction density clustering (EIC) using the average functional connection matrix calculated from all the subject data in step S1. It has been verified in the experiment that the reliability of the segmentation results obtained when the ratio threshold is within the range of 0.5~0.9 is relatively high, and the ratio threshold selected in this embodiment is 0.8. In addition, the adjacency matrix records the adjacent relationship between the functional nodes in the spatial grid, and the element in the adjacency matrix is ​​"1" indicating that the corresponding two functional nodes are adjacent and share a triangle edge, and "0" indicates that the corresponding two functional nodes are not adjacent. The adjacency matrix used in this embodiment represents the adjacent relationship between cortical vertices in the standard 32k_fs_LR space. The element "1" in the adjacency matrix indicates that the corresponding two vertices are adjacent. In the spatial grid, the two vertices share a triangle edge, and the element "0" indicates that the corresponding two vertices are not adjacent. The adjacency matrix can be obtained from FreeSurfer and is known information.

[0090] In this embodiment, step S5.4 includes:

[0091] S5.4.1, traverse each sub-brain region corresponding to the current brain region in the group atlas;

[0092] S5.4.2, find the functional node with the highest confidence in the current sub-brain region, and find its adjacent functional nodes according to the adjacency matrix, such as Figure 2 As shown in (a), vertex 1 is the vertex with the highest confidence, and the dotted circle around it represents its adjacent vertices;

[0093] S5.4.3, assign the label of the function node with the highest confidence as a group-level label, and also assign the label of the function node with the same group-level label as the function node among its adjacent function nodes; jump to step S5.4.2 until the proportion of function nodes assigned labels in the current sub-brain region reaches a preset proportion threshold (the proportion threshold selected in this embodiment is 0.8). The label of the function node (vertex) with the highest confidence is assigned as a group-level label, and the vertices with the same group-level label as the vertex among its adjacent vertices also retain the group-level label, such as Figure 2 As shown in (b) in ; then find the vertex with the second largest confidence and determine its adjacent vertices, such as Figure 2 In (c), vertex 2 and its adjacent vertices represented by the dashed circle around it retain the group horizontal labels according to the above method, such as Figure 2 As shown in (d) in the figure, and so on;

[0094] S5.4.4, determine whether the sub-brain regions of the current brain region in the group atlas have been traversed. If not, jump to step S5.4.1 to continue traversing the remaining sub-brain regions. Otherwise, determine that the operation of retaining the high-confidence functional node labels in the group atlas for the current brain region has been completed, and jump to step S6.

[0095] In step S6 of this embodiment, the processing for any brain region includes:

[0096] S6.1, the adjacency matrix is ​​regarded as an undirected graph between functional nodes, the paths between functional nodes with adjacency relationship are recorded as 1, and the shortest paths between all functional nodes are calculated as the geodesic distance between functional nodes;

[0097] S6.2, normalize the geodesic distance and feature space distance between functional nodes and average them to obtain the integrated distance;

[0098] S6.3, traverse all functional nodes in the brain region in order of local density from large to small. If the traversed functional node has been assigned a label, skip it; if the traversed functional node has not yet been assigned a label, find a set of functional nodes with a higher local density than the traversed functional node, and assign the label of the functional node closest to the traversed functional node in the set of functional nodes with higher local density to the traversed functional node, and finally obtain the preliminary individual functional brain region dissection results.

[0099] Step S7 performs correlation maximization processing on each brain region until iterative convergence in order to reduce the problem of too many scattered points caused by low data signal-to-noise ratio to a greater extent. The processing for any brain region in step S7 includes:

[0100] S7.1, traverse the function nodes in descending order of local density. If the traversal is not completed, jump to the next step; otherwise, jump to step S7.5;

[0101] S7.2, searching for adjacent functional nodes of the current functional node according to the adjacency matrix, and determining the functional nodes included in all categories corresponding to the adjacent functional nodes of the current functional node based on the preliminary individual functional brain region dissection results of the brain region;

[0102] S7.3, calculating the average distance between the current function node and the function nodes included in all categories in step S7.2 according to the feature space distance matrix, so as to measure the relevance of the current function node to each category, and the smaller the average distance, the stronger the relevance;

[0103] S7.4, if the label of the current function node is consistent with the category with the smallest average distance, then skip the current function node; otherwise, modify the label of the current function node to the category with the smallest average distance;

[0104] S7.5, determine whether the proportion of functional nodes whose labels do not change in this round exceeds the preset ratio. If it exceeds the preset ratio, it is determined that the correlation maximization of the individual functional brain region dissection results of the brain region has been completed, and jump to step S8; otherwise, jump to step S7.1 to enter the next round of functional node traversal. Among them, the preset ratio can be set according to actual needs. For example, as an optional implementation, the preset ratio in this embodiment is 98%. Figure 3 As shown in (a), the vertex 1 corresponding to the solid circle has adjacent vertices of categories including category 1: 1xM1, category 2: 1xM2 and category 3: 1xM3. Step S7.3 calculates the average distance from the current function node to the function nodes included in all categories in step S7.2 according to the feature space distance matrix, and finds the category m with the smallest average distance. Since the category m with the smallest average distance is inconsistent with the label "1" of the current vertex, the label "1" of the current vertex is modified to the category m with the smallest average distance, as shown in FIG. Figure 3 As shown in (b) in .

[0105] In this embodiment, the processing for any brain region in step S8 includes: judging whether the brain region contains a spatially discontinuous connected domain based on the optimized individual functional brain region dissection results of each brain region after the correlation maximization processing, retaining the connected domain containing the largest number of functional nodes for the brain region containing the spatially discontinuous connected domain, and reallocating labels to the remaining connected domains according to the category with the most labels in its peripheral functional nodes, and merging with the peripheral brain regions. After all brain regions complete the above processing, the final individual functional brain region dissection results are obtained; the judging whether the brain region contains a spatially discontinuous connected domain includes using an adjacency matrix to find the connection relationship between functional nodes. If two connected domains do not contain a pair of functional nodes with a connection relationship, the two connected domains are spatially discontinuous connected domains.

[0106] Steps S5 to S8 are performed on each brain region included in the DK template to obtain the dissection results of all brain regions of the whole brain, that is, the whole brain functional brain map. The number of individual functional brain regions obtained in this embodiment is theoretically consistent with the number of brain regions in the group map or reference map. If the number of individual brain regions finally obtained is less than the number of brain regions in the group map or reference map, it can be considered that it is caused by data quality problems or lesions in the brain regions of the individual subjects, and further analysis is required. Therefore, the method of this embodiment also has a certain effect on screening qualified data and discovering lesion individuals.

[0107] In order to verify the reliability and actual performance of the individual functional brain region segmentation method based on improved density clustering in this embodiment, the individual functional brain region segmentation method based on improved density clustering in this embodiment was implemented based on MATLAB, and segmented on a public data set containing 175 subjects (each subject contained 7T resting fMRI data of 4 scans in 2 sessions and 3T task-state fMRI data of 7 tasks), and implemented the following 4 aspects of verification: I, intra-individual repeatability and inter-individual variability; II, individual recognition accuracy; III, functional connectivity homogeneity within individual functional sub-regions; IV, task activation heterogeneity within individual functional sub-regions. Each subject was segmented into 485 functional brain regions.

[0108] I. Intra-individual repeatability and inter-individual variability.

[0109] The data of the two sessions of all subjects in the data set were separately subjected to individual-level whole-brain brain region dissection, and the repeatability between the dissection results of the two sessions of the same individual was calculated to obtain the intra-individual repeatability, and the average repeatability between the dissection results of any two different subjects was calculated to obtain the inter-individual repeatability. Each session contains two runs, and the average functional connection matrix obtained from the two runs is used for dissection to obtain the dissection results under the session. Repeatability is measured by calculating the Dice coefficient of the whole-brain dissection label. The larger the Dice coefficient, the higher the repeatability. In this embodiment, the average intra-individual repeatability of all subjects was 0.6854±0.0344, and the average inter-individual repeatability was 0.5974±0.0169. The intra-individual repeatability was significantly higher than the inter-individual repeatability, which verified the repeatability of the dissection method. The inter-individual repeatability not only shows that the functional brain regions of different individuals maintain a high degree of similarity, but also reflects the differences in functional brain regions between individuals.

[0110] II. Individual identification accuracy.

[0111] The individual recognition accuracy is calculated based on the segmentation results of the two individual sessions. For a subject, the Dice coefficient between the segmentation results of the individual in session 1 and all the individual segmentation results in session 2 is calculated, and the Dice coefficient of the corresponding result of the individual in session 2 is tested to see whether it is the largest. Figure 4As shown, the first line is the data of session session1, and the second line is the data of session session2. According to the calculation method, the Dice coefficient between the segmentation result under the individual session session2 and all the individual segmentation results under session session1 is also calculated to check whether the Dice coefficient of the corresponding result of the individual under session session1 is the largest. If the Dice coefficient corresponding to the same individual is the largest, it means that the individual can be successfully identified, and the individual recognition accuracy is obtained by calculating the proportion of the number of individuals that can be successfully identified. The average individual recognition accuracy obtained by the two calculation results is used as the final individual recognition accuracy. The individual recognition accuracy calculated on all subjects in this embodiment is 97.71%, which shows that the individualized functional brain area fine segmentation method of the present invention can distinguish different individuals and can characterize the differences in individual functional connection patterns.

[0112] III. Functional connectivity homogeneity within individual functional sub-regions.

[0113] Compared with the group atlas, the functional connectivity homogeneity is improved. According to the individual level segmentation results, the average correlation between the functional connectivity of all vertices in each segmented subregion and the average functional connectivity of the subregion is calculated, that is, the functional connectivity homogeneity. After weighted averaging, the functional connectivity homogeneity of the whole brain is obtained. The calculation formula is shown as follows:

[0114] ,

[0115] ,

[0116] In the above formula, is the functional connectivity homogeneity of the whole brain, For the Functional connectivity homogeneity of brain regions, For the The number of vertices in each brain region, is the average functional connectivity of all vertices in the brain region, This brain region The functional connection of each vertex. The group map is mapped to the individual and the functional connection homogeneity of the group map on the individual is calculated according to the above method. In this embodiment, the average individual map functional connection homogeneity calculated on all subjects is 0.8618±0.0132, the average group map functional connection homogeneity is 0.8392±0.0142, and the individual map functional connection homogeneity of all subjects is higher than the group map functional connection homogeneity, which verifies the effectiveness of the segmentation method. The individual map obtained is more representative of the individual functional brain area division than the group map.

[0117] IV. Task activation heterogeneity within individual functional sub-regions.

[0118] The task-state fMRI data of the subjects in the data set were used to verify the task activation heterogeneity within the individual functional sub-areas. Each subject had 3T task-state fMRI data of 7 tasks, namely motor task, social cognition task, game task, relationship task, language task, working memory task and emotional task. For each stimulus under each task, the standard deviation of the task activation value in each sub-area was calculated according to the individual segmentation results, which was the task activation heterogeneity of the sub-area. The weighted average of all sub-areas was used to obtain the task activation heterogeneity of the whole brain. The calculation formula is shown as follows:

[0119] ,

[0120] In the above formula, Activate heterogeneity for tasks across the whole brain, For the Heterogeneity of task activation in brain regions For the The number of vertices in each brain region. The task activation heterogeneity calculated for all stimuli under each task is averaged to obtain the task activation heterogeneity of the task. The lower the task activation heterogeneity, the more consistent the task activation within the sub-region, which means that the sub-region segmentation is more reasonable. Compare whether the task activation heterogeneity of the individual map is reduced relative to the group map to verify the effectiveness of the segmentation method. In this embodiment, for each individual subject, the task activation heterogeneity of the individual map and the group map on its task activation map is calculated separately. The task activation heterogeneity of the individual map and the group map under 7 tasks (sports, social cognition, game, relationship, language, working memory and emotion) are compared, and the proportion of individual maps with lower task activation heterogeneity in all subjects is counted, such as Figure 5 As shown, the average individual atlas task activation heterogeneity and group atlas task activation heterogeneity of all subjects were calculated, as Figure 6 As shown. Figure 5 and Figure 6 The experimental results show that the individual map task activation heterogeneity of most subjects is lower, and the average individual map task activation heterogeneity under each task is lower than that of the group map, which verifies the rationality of the sub-region segmentation. In this embodiment, the data of 2 sessions of each subject are used for segmentation when verifying the homogeneity of functional connectivity and the heterogeneity of task activation, that is, the average functional connectivity matrix calculated from 4 scanning data is used as the input of the segmentation algorithm. In this embodiment, the intra-individual repeatability of all subjects is significantly higher than the inter-individual repeatability, the individual recognition accuracy is high, the connection homogeneity of the individual map is significantly higher than that of the group map, the average individual map task activation heterogeneity under each task is lower than that of the group map, and the individual map task activation heterogeneity of most subjects is lower than that of the group map, which verifies the reliability and effectiveness of the functional sub-region segmentation method of the present invention.

[0121] In summary, in response to the current problem of individual level segmentation, the method of this embodiment uses resting-state functional magnetic resonance imaging to establish a method for individual functional brain region segmentation based on improved density clustering, including obtaining the resting-state functional magnetic resonance imaging data of the individual subjects to calculate the functional connection matrix and feature space distance matrix of the whole brain functional nodes; calculating the local density and confidence of each functional node for each brain region respectively, introducing the adjacency matrix to retain the labels of high-confidence functional nodes in the group map, combining the feature space distance and the geodesic distance to calculate the integrated distance, and assigning labels to the remaining functional nodes in the brain region according to the principle of being consistent with the higher local density point of the nearest integrated distance, so as to obtain the preliminary individual functional brain region segmentation results of the brain region, perform correlation maximization processing until iterative convergence, perform spatial continuity processing to obtain the final individual functional brain region segmentation results, and evaluate the effectiveness of the method from multiple angles. In order to solve the technical problem of establishing an effective and reliable individual-level functional brain area fine-division scheme using resting-state functional magnetic resonance data with a relatively short time on the basis of density clustering, the key technical means adopted by the method of this embodiment include: key technical means 1, retaining high-confidence vertex labels: introducing an adjacency matrix, retaining the labels of high-confidence vertices in the group map based on individual characteristics; key technical means 2, integrating distance to allocate vertex labels: combining feature space distance and geodesic distance to allocate remaining vertex labels, minimizing the impact of low signal-to-noise ratio caused by short scanning time of individual data; key technical means 3, correlation maximization processing: by checking whether each vertex belongs to the class with the largest correlation among the classes to which the adjacent vertices belong, further eliminating noise points until the iteration converges. In the related individualized segmentation technology, the group-level brain map or reference map is usually directly used as the initial template for individual segmentation, and then the final segmentation result of the individual brain area is obtained by iterative adjustment. However, this approach will make the individual segmentation results too constrained by the prior information of the group map or reference map, and the specificity of the individual partition is limited. At the same time, theoretically, the brain area boundaries of the group map or reference map vary the most between individuals, and directly using the entire map as the initialization label may interfere with the determination of the individual brain area boundaries. Therefore, in key technical means 1, the present invention determines the vertices (called high-confidence vertices) in the individual that are most likely to be consistent with the group map or reference map label based on individual characteristics, and introduces an adjacency matrix, starting from the vertex with the highest confidence, and retains the high-confidence vertex labels in turn until the number of vertices with retained labels reaches the specified number of vertices. This approach not only screens high-confidence vertices, but also takes into account individual characteristics. Although different individuals retain the same number of vertex labels of the group map or reference map, these vertices are not exactly the same. They are obtained by the characteristics of the individual itself, which not only ensures that the functional sub-area has good consistency with the group map or reference map, but also ensures the specificity of the individual. In the basic framework of correlation density clustering, the density is usually calculated according to the distance relationship of samples in the feature space, and then the cluster center is determined, and the vertex labels of non-cluster centers are assigned.For high-dimensional data such as the brain, the segmentation results of this method are easily affected by the low signal-to-noise ratio of the data, resulting in more noise points. Therefore, in key technical means 2, this embodiment comprehensively considers the feature space distance and the geodesic distance, and assigns labels to the remaining vertices after retaining the labels of some vertices in the group map or reference map, so as to minimize the impact of the low signal-to-noise ratio caused by the short scanning time of individual data. This embodiment is to perform individualized functional brain area division, which assumes that the vertices in each functional brain area have high correlation. Therefore, in key technical means 3, this embodiment performs correlation maximization processing, assigning each vertex to the class with the largest correlation among the classes to which its adjacent vertices belong, so as to further eliminate noise points until the iteration converges. The method of this embodiment can solve the problem of how to use relatively short resting state functional magnetic resonance data to obtain effective and reliable individual functional brain area segmentation results, and has the following effects: (1) The individual functional brain area segmentation method based on improved density clustering in this embodiment can reliably and effectively segment relatively short resting state functional magnetic resonance data to obtain reasonable individual maps, which has important potential value in clinical applications. (2) This embodiment retains the group map labels of high-confidence vertices in combination with the individual's own characteristics, avoiding the situation where the individual map is overly constrained by the group prior information and taking into account the individual specificity. (3) This embodiment comprehensively considers the feature space distance and the geodesic distance for label allocation, greatly reducing the impact of low data signal-to-noise ratio. (4) The individualized functional brain area segmentation method proposed in this embodiment does not require training. It only needs to input individual data to obtain the individual functional brain area segmentation results. It can be directly applied to a single individual and has high efficiency. The method of this embodiment uses resting-state functional magnetic resonance imaging data with a relatively short time to obtain an effective and reliable individual brain map, which has great potential value in clinical practice.

[0122] In addition, this embodiment also provides an individual functional brain region segmentation system based on improved density clustering, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering.

[0123] In addition, this embodiment also provides a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering through a processor.

[0124] In addition, this embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering through a processor.

[0125] Those skilled in the art should understand that the technical solutions provided by the embodiments of the present application may be in the form of methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0126] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for individual functional brain region segmentation based on improved density clustering, characterized in that: The steps include: S1, obtain the resting-state functional magnetic resonance imaging data of the individual subjects; S2, perform data preprocessing to extract node time series of functional nodes, where the functional nodes are cortical vertices or subcutaneous nuclei; S3, calculating the functional connection matrix of the whole brain functional nodes according to the node time series of the functional nodes; S4, calculating the feature space distance between the functional nodes according to the functional connection matrix to obtain the feature space distance matrix; S5, using the initial brain region template to divide the brain region, calculating the local density and confidence of each functional node in each brain region, and introducing the adjacency matrix to retain the labels of high-confidence functional nodes in the group map; S6, respectively, calculate the integrated distance for each brain region by combining the feature space distance and the geodesic distance, and assign labels to the remaining functional nodes in the brain region according to the principle of being consistent with the higher local density point of the nearest integrated distance, so as to obtain the preliminary individual functional brain region dissection results; S7, perform correlation maximization processing on each brain region until iterative convergence; S8, perform spatial continuity processing on each brain region to obtain the final individual functional brain region dissection results.

2. The method for individual functional brain region segmentation based on improved density clustering according to claim 1, characterized in that: Calculating the feature space distance matrix according to the functional connectivity matrix in step S4 includes: S4.1, measuring the similarity between two functional nodes in the functional connection matrix by the correlation ratio, thereby obtaining a similarity matrix composed of the similarities between two functional nodes; S4.2, reducing the dimension of the similarity matrix by Laplace eigenmap algorithm LE to obtain the normalized Laplace eigenvector; S4.3, for the normalized Laplace eigenvector, the distance between each pair of functional nodes in the low-dimensional Laplace feature space is calculated as the feature space distance, and a feature space distance matrix consisting of the feature space distances between each pair of functional nodes is obtained.

3. The method for individual functional brain region segmentation based on improved density clustering according to claim 1, characterized in that: The processing for any brain region in step S5 includes: S5.1, combined with the feature space distance matrix, calculate the local density of each functional node in the current brain region according to the following formula: ， In the above formula, For functional nodes The local density of For functional nodes To function node The feature space distance of is the cutoff distance, which is used to ensure that all The average number of functional nodes within the range accounts for a proportion of the total number of nodes that is not less than the preset threshold; S5.2, combined with the feature space distance matrix, calculate the nearest neighbor distance of each functional node according to the following formula: ， In the above formula, For functional nodes The nearest neighbor distance, For functional nodes The local density of S5.3, calculate the confidence of the function node based on the local density and nearest neighbor distance of the function node: ， In the above formula, For functional nodes Confidence level; S5.4, traverse the functional nodes in each sub-brain region of the current brain region in order from high to low confidence, and retain the functional node labels of the group map in combination with the introduced adjacency matrix until the proportion of functional nodes assigned with labels reaches a preset proportion threshold; the adjacency matrix records the adjacent relationship between functional nodes in the spatial grid, and the element in the adjacency matrix is ​​"1" indicating that the corresponding two functional nodes are adjacent and share a triangle edge, and "0" indicating that the corresponding two functional nodes are not adjacent.

4. The method for individual functional brain region segmentation based on improved density clustering according to claim 3, characterized in that: Step S5.4 includes: S5.4.1, traverse each sub-brain region corresponding to the current brain region in the group atlas; S5.4.2, find the functional node with the highest confidence in the current sub-brain region, and find its adjacent functional nodes according to the adjacency matrix; the adjacency matrix records the adjacent relationship between the functional nodes in the spatial grid, and an element in the adjacency matrix of "1" indicates that the corresponding two functional nodes are adjacent and share a triangle edge, and an element of "0" indicates that the corresponding two functional nodes are not adjacent; S5.4.3, assigning the label of the functional node with the highest confidence as a group-level label, and also assigning the functional nodes whose group-level labels are consistent with the label of the functional node among its adjacent functional nodes as group-level labels; jumping to step S5.4.2 until the proportion of functional nodes assigned with labels in the current sub-brain region reaches a preset proportion threshold; S5.4.4, determine whether the sub-brain regions of the current brain region in the group atlas have been traversed. If not, jump to step S5.4.1 to continue traversing the remaining sub-brain regions. Otherwise, determine that the operation of retaining the high-confidence functional node labels in the group atlas for the current brain region has been completed, and jump to step S6.

5. The method for individual functional brain region segmentation based on improved density clustering according to claim 1, characterized in that: The processing for any brain region in step S6 includes: S6.1, the adjacency matrix is ​​regarded as an undirected graph between functional nodes, the paths between functional nodes with adjacency relationship are recorded as 1, and the shortest paths between all functional nodes are calculated as the geodesic distance between functional nodes; S6.2, normalize the geodesic distance and feature space distance between functional nodes and average them to obtain the integrated distance; S6.3, traverse all functional nodes in the brain region in order of local density from large to small. If the traversed functional node has been assigned a label, skip it; if the traversed functional node has not yet been assigned a label, find a set of functional nodes with a higher local density than the traversed functional node, and assign the label of the functional node closest to the traversed functional node in the set of functional nodes with higher local density to the traversed functional node, and finally obtain the preliminary individual functional brain region dissection results.

6. The method for individual functional brain region segmentation based on improved density clustering according to claim 1, characterized in that: The processing for any brain region in step S7 includes: S7.1, traverse the function nodes in descending order of local density. If the traversal is not completed, jump to the next step; otherwise, jump to step S7.5; S7.2, searching for adjacent functional nodes of the current functional node according to the adjacency matrix, and determining the functional nodes included in all categories corresponding to the adjacent functional nodes of the current functional node based on the preliminary individual functional brain region dissection results of the brain region; S7.3, calculating the average distance between the current function node and the function nodes included in all categories in step S7.2 according to the feature space distance matrix, so as to measure the relevance of the current function node to each category, and the smaller the average distance, the stronger the relevance; S7.4, if the label of the current function node is consistent with the category with the smallest average distance, then skip the current function node; otherwise, modify the label of the current function node to the category with the smallest average distance; S7.5, determine whether the proportion of functional nodes whose labels do not change in this round exceeds the preset ratio. If it exceeds the preset ratio, it is determined that the correlation maximization processing of the individual functional brain area dissection results of the brain area is completed, and jump to step S8; otherwise, jump to step S7.1 to enter the next round of functional node traversal.

7. The method for individual functional brain region segmentation based on improved density clustering according to claim 1, characterized in that: The processing for any brain region in step S8 includes: judging whether the brain region contains a spatially discontinuous connected domain according to the individual functional brain region dissection results optimized for each brain region after the correlation maximization processing, retaining the connected domain containing the largest number of functional nodes for the brain region containing the spatially discontinuous connected domain, and reallocating labels to the remaining connected domains according to the category with the most labels in its peripheral functional nodes, and merging with the peripheral brain regions, and obtaining the final individual functional brain region dissection results after all brain regions complete the above processing; the judging whether the brain region contains a spatially discontinuous connected domain includes using an adjacency matrix to find the connection relationship between functional nodes, and if two connected domains do not contain a functional node pair with a connection relationship, then the two connected domains are spatially discontinuous connected domains.

8. An individual functional brain region segmentation system based on improved density clustering, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program or instruction stored therein, characterized in that: The computer program or instruction is programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering according to any one of claims 1 to 7 through a processor.

10. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is programmed or configured to execute the individual functional brain region segmentation method based on improved density clustering according to any one of claims 1 to 7 through a processor.

Images