Individual difference brain network construction method based on sparse kernel matrix optimization
The construction of individual differential brain networks through sparse nuclear matrix optimization method solves the problem of using the average functional connection matrix to mask individual differences in the existing technology, and achieves more accurate brain network construction and mental illness symptoms prediction.
Patent Information
- Application Number
- CN202411163526.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-08-23
AI Technical Summary
In the construction of individual differences brain networks, the use of the average functional connection matrix of the normal group may mask subtle differences between individuals and cannot fully reflect the unique pathological characteristics of each patient, resulting in limited accuracy of individualized diagnosis.
Using a method based on sparse kernel matrix optimization, the sparse kernel matrix is obtained by calculating the similarity matrix difference between the brain time series of abnormal subjects and all normal subjects, and multiplying it with the average value of other brain region difference matrix. Form an optimized brain region difference matrix, and then build an individual differentiated brain network.
It improves the accuracy of the brain network, can more accurately identify the symptoms and cognitive function damage of mental illness patients, improves the accuracy of predicting symptoms of mental illnesses, and distinguishes patients from normal people, which has good robustness.
Smart Images

Figure CN119066494A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to brain network construction, and in particular to an individual difference brain network construction method based on sparse kernel matrix optimization. Background Art
[0002] The study of brain functional networks is of great significance for understanding the complex operating mechanisms of the human brain. The human brain is a highly integrated network, not an isolated anatomical region, and forms a complex and precise connection network through synaptic interactions between neurons. Related studies have shown that various advanced cognitive functions of the brain rely on the coordinated cooperation of different brain regions, not just a single specific region. The pathogenesis of many mental illnesses can be attributed to the disordered connectivity between the brain regions involved to some extent. In recent years, the development of neuroimaging technology has made it possible to study abnormal brain connections in mental illness. Functional magnetic resonance imaging (fMRI) has been shown to be of great value in revealing the intrinsic functional connectivity of the brain. This technology reflects the neuronal activity of brain regions in the resting state or task state by measuring the blood oxygen level dependent (BOLD) signal in different brain regions. Functional connectivity can reflect the coordinated activity or synchronization between different brain regions in a specific time period, thereby revealing the relationship between the organizational structure and function of the brain network. Therefore, by using fMRI to analyze the differences in functional connectivity between patients and healthy controls, researchers can gain a deeper understanding of the pathological mechanisms of various brain diseases.
[0003] In clinical medicine, schizophrenia (SZ) is a chronic and highly disabling severe mental illness that usually develops in adolescence or early adulthood and manifests as multiple disorders of thinking, emotion and behavior. Schizophrenia has a variety of psychopathological characteristics, and its core features include positive symptoms (such as delusions, hallucinations, disordered thinking and chaotic behavior), negative symptoms (including impaired motivation, reduced spontaneous speech and social withdrawal) and cognitive impairment (including attention, memory, learning, thinking and executive function defects), affecting about 1% of the world's population. At present, there are as many as 26 million people suffering from schizophrenia worldwide, which not only brings great pain to patients, but also brings a heavy burden to families and society. Therefore, it is of great significance to use modern medical imaging technology to study the pathological mechanism of schizophrenia and achieve early warning and clinical intervention for schizophrenia. On the other hand, individual difference functional connectivity has important potential value in the field of neuroscience and brain imaging. First, it helps to understand the neural diversity of the human brain and reveal differences in cognitive ability, behavioral performance and emotional state. Secondly, by comparing the functional connectivity patterns of patients and healthy individuals, biomarkers for specific diseases can be identified, promoting symptom prediction and classification research. At the same time, the study of individual differences in functional connectivity can promote the development of personalized medicine, formulate personalized treatment plans based on the unique functional connectivity characteristics of patients, and improve the treatment effect. In addition, the study of individual differences in functional connectivity helps to reveal the basic mechanisms of brain network changes and promote the understanding of brain function. These studies can not only promote the development of neuroscience, but also show broad application prospects in personalized clinical diagnosis and medical treatment.
[0004] However, in current related studies, researchers mostly calculate the functional connectivity matrix based on the correlation of the time series of the subjects' brain regions, and compare the differences between the average functional connectivity matrix of each patient with mental illness and the healthy control group to construct an individual difference matrix. However, due to individual differences and the unique brain activity patterns of patients with mental illness, it is still challenging to accurately analyze and integrate these complex brain networks. Using the average functional connectivity matrix of the normal group may mask subtle differences between individuals and fail to fully reflect the unique pathological characteristics of each patient, resulting in the loss of individual characteristics, thereby limiting the accuracy of individualized diagnosis. Summary of the invention
[0005] Purpose of the invention: In view of the above shortcomings, the present invention provides a method for constructing individual difference brain networks based on sparse kernel matrix optimization to improve the accuracy of brain networks.
[0006] Technical solution: To solve the above problems, the present invention adopts a method for constructing an individual difference brain network based on sparse kernel matrix optimization, comprising the following steps:
[0007] (1) Using the brain network atlas, brain region time series are extracted from the normal brain image data of several normal subjects and the abnormal brain images of several abnormal subjects, respectively, to obtain the brain region time series of several normal subjects and the brain region time series of several abnormal subjects;
[0008] (2) Calculate the similarity matrix of the time series of the brain regions of several normal subjects and the similarity matrix of the time series of the brain regions of several abnormal subjects respectively through similarity measurement;
[0009] (3) Calculate the absolute value of the difference between the similarity matrix of each abnormal subject and the similarity matrix of all normal subjects to obtain the difference matrices of several brain regions of the abnormal subjects;
[0010] (4) For the vth brain region difference matrix of each abnormal subject, the similarities of each row of the brain region difference matrix are arranged from large to small, the values of the first N similarities are taken, and the values of other similarities are set to zero to obtain the vth sparse core matrix of the abnormal subject, and the vth sparse core matrix is multiplied by the average value of the remaining brain region difference matrices to obtain the vth optimized brain region difference matrix. Through this step, all optimized brain region difference matrices of all abnormal subjects are obtained;
[0011] (5) The average of the optimized brain region difference matrices of all brain regions of each abnormal subject is calculated to obtain the individual difference brain network of the abnormal subject.
[0012] Furthermore, in step (4), the vth sparse kernel matrix is multiplied by the average value of other normalized brain region difference matrices to obtain the optimized sparse kernel matrix of the vth brain region.
[0013] Furthermore, the vth optimized brain region difference matrix P (v) for:
[0014]
[0015] Among them, S (v) represents the vth sparse kernel matrix, D′ (k) represents the kth normalized brain region difference matrix, and m represents the total number of normal subjects.
[0016] Furthermore, the similarity calculation formula is:
[0017]
[0018] Among them, W(i,j) represents the brain area x of the subject. i and brain region x j Similarity value between time series, ρ 2 (x i ,x j ) represents the brain area x of the subjecti and brain region x j The square of the Euclidean distance between time series, μ represents the hyperparameter used for empirical settings, and ε i,j represents the hyperparameter used to eliminate the scaling problem.
[0019] Furthermore, the step (4) specifically includes:
[0020]
[0021] Among them, S(i,j) represents the value of the i-th row and j-th column in the sparse kernel matrix S; D(i,j) is the value of the i-th row and j-th column in the brain region difference matrix D, which represents the absolute value of the difference between W(i,j) of the abnormal subject and W(i,j) of the normal subject; D(i,k) is the value of the i-th row and k-th column in the brain region difference matrix D.
[0022] Furthermore, the brain region difference matrix is normalized:
[0023]
[0024] Wherein, D′ represents the matrix after normalizing the brain region difference matrix D, and D′(i,j) represents the value of the i-th row and j-th column in the normalized brain region difference matrix D′.
[0025] Furthermore, the normal brain image data and the abnormal brain image data are structural magnetic resonance image data from a public data center, which are preprocessed separately to obtain preprocessed images of normal brain region data and abnormal brain region data; the obtained preprocessed images are divided into brain regions, and the average time series of all voxels in each brain region in each data is calculated.
[0026] Furthermore, the preprocessed brain image data are divided into brain regions according to brain maps, and the brain maps include AAL maps and BA maps.
[0027] The present invention also adopts a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the above method when executing the computer program.
[0028] The present invention also adopts a computer-readable storage medium on which a computer program is stored, characterized in that the computer program implements the steps of the above method when executed by a processor.
[0029] Beneficial effects: Compared with the prior art, the significant advantage of the present invention is that it constructs a difference matrix between the brain regions of the abnormal and normal control groups, and fully explores and fuses the important features in the difference matrix of each brain region by multiplying the sparse kernel matrix of each brain region difference with the average value of the difference matrix of the remaining brain regions. The constructed individual difference matrix is used as a priori guidance for feature selection in related mental illness symptoms, cognitive prediction and classification, which improves the prediction accuracy of mental illness symptoms and cognition to a certain extent, and can also distinguish patients. In addition, the embodiments of the present invention have good robustness and can be used to extract important features in complex brain disease-related functional networks, and reveal physiological and pathological mechanisms related to symptoms and cognitive impairment. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 Schematic diagram of the process of constructing a midbrain network according to the present invention.
[0031] Figure 2 This is a schematic diagram of the accuracy of the brain network construction method based on the AAL atlas in the present invention in predicting the positive / negative symptoms, total symptom score and cognitive score of schizophrenia on the BSNIP1 dataset.
[0032] Figure 3 This is a schematic diagram of the accuracy and confusion matrix of the brain network construction method based on the AAL atlas in the present invention for distinguishing schizophrenia from normal people on the BSNIP1 dataset.
[0033] Figure 4 This is a schematic diagram of the accuracy of the brain network construction method based on BA atlas in the present invention in predicting positive / negative symptoms, total symptom scores and cognitive scores of schizophrenia on the BSNIP1 dataset.
[0034] Figure 5 This is a schematic diagram of the accuracy and confusion matrix of the brain network construction method based on BA atlas in the present invention for distinguishing schizophrenia from normal people on the BSNIP1 dataset. DETAILED DESCRIPTION
[0035] like Figure 1 As shown, a method for constructing an individual difference brain network based on sparse kernel matrix optimization in this embodiment includes the following steps:
[0036] Step S1: Using the brain network atlas, extract the brain region time series of the subject's functional magnetic resonance imaging.
[0037] In this step, the brain atlas used may be Brainnetome Atlas (BA), Automated Anatomy Labeling Atlas (AAL), etc. In the actual implementation process, any one of them or other brain atlases may be selected for implementation.
[0038] In this embodiment, only functional magnetic resonance imaging data of schizophrenia is used as an example for explanation, but the protection scope of the present invention is not limited to this. The data can be functional magnetic resonance imaging data of all mental disorders, including mental illnesses such as schizophrenia and depression, and functional magnetic resonance imaging data of non-mental diseases such as bipolar disorder.
[0039] In this embodiment, the preprocessing and feature extraction of functional magnetic resonance imaging data are specifically as follows: (1) head motion correction; (2) inter-layer time correction; (3) standardization to the Montreal standard space, preferably, resampling to 3×3×3mm; (4) regression of 6 head motion parameters and white matter (WM), gray matter (GM), cerebrospinal fluid (CSF); (5) smoothing with 8mm full width half max (FWHM) Gaussian filter. Then, the brain is divided into specific areas using BA or AAL brain atlas, and the average time series of all voxels in each area are calculated to obtain the time series features of each node.
[0040] Step S2: Use a similarity measurement method to calculate the similarity of brain region time series of patients with mental illness and normal people respectively.
[0041] The similarity measurement method includes but is not limited to Euclidean distance, cosine distance, and other correlation calculation methods.
[0042]
[0043] Specifically, the similarity calculation process can be as follows: According to formula (1), the time series feature matrix obtained after processing in step S1 is calculated for brain region x i and x j The square of the Euclidean distance between them is then used to determine the weights of brain region similarity using a scaled exponential similarity kernel, where μ represents an empirically set hyperparameter and ε i,j represents the hyperparameters used to eliminate scaling issues;
[0044]
[0045] Among them, mean(ρ(x i ,N)) represents x iThe average distance to each point in its neighborhood N, mean(ρ(x j ,N)) represents x j The average distance to each point in its neighborhood N, ρ(x i ,x j ) is x i and x j The Euclidean distance between the time series; by introducing the local scale parameter ε i,j , which can more accurately reflect x i and x j This method takes into account the local distribution of points when calculating the similarity weight, thereby improving the robustness and accuracy of the similarity metric. The smaller the distance, the larger the similarity weight W(i,j); the larger the distance, the smaller the similarity weight W(i,j). Through the form of an exponential function, it can be ensured that the weight is always between 0 and 1. The above calculations are performed on all schizophrenia patients and normal people, and the brain region similarity matrix of each subject is obtained.
[0046] Step S3: Calculate the absolute value of the difference between the brain region similarity matrix of each mental illness patient and all normal people as the brain region difference matrix D between the patient and the healthy control group.
[0047] Specifically, the brain region similarity matrix of each mental illness patient is subtracted from the brain region similarity matrix of m normal people in turn, and the absolute values of the obtained m brain region difference matrices are taken to obtain the m brain region difference matrices D between each mental illness patient and the m normal people.
[0048] Step S4: Calculate the top N values with higher similarity in each row of the vth brain region difference matrix, and set the values with lower similarity to zero as the sparse core matrix in the fusion process; then multiply the sparse core matrix with the average value of the remaining m-1 brain region difference matrices, and repeat this process until all brain region difference matrices of each patient with mental illness are traversed.
[0049] Specifically, based on the brain region difference matrix obtained in step S3, according to formula (2), the sparse kernel matrix with large differences between patients and healthy controls is first calculated, that is, the brain region x is calculated. i The values of the top N brain regions with higher similarity are set to zero, and the sparse kernel matrix is normalized.
[0050]
[0051] Then, the brain region difference matrix is normalized according to formula (3), the diagonal elements are set to 0.5, and x i The sum of similarities with all brain regions is 1. This normalization can eliminate the influence of unstable values.
[0052]
[0053] Based on the sparse kernel matrix and brain region difference matrix obtained by formulas (2) and (3), they are fused according to formula (4).
[0054]
[0055] The sparse kernel matrix contains the top N brain region connections with the highest similarity in each brain region difference matrix; Indicates that the average value of m-1 normalized brain region difference matrices is calculated; the obtained sparse kernel matrix is multiplied by the average value of the remaining m-1 brain region difference matrices to obtain the vth optimized brain region difference matrix P (v) The calculation process is repeated m times until all m brain region difference matrices of each patient with mental illness are traversed.
[0056] Step S5: Calculate the average value of the optimized brain region difference matrix obtained in step S4 as the individual difference matrix of patients with mental illness, thereby realizing the construction of individual difference functional brain network.
[0057] The present invention uses the Bipolar and Schizophrenia Network for Intermediate Phenotypes (BSNIP) dataset 1 (including fMRI data of 168 schizophrenia patients and 219 normal control subjects) and the normal person data of dataset 2 (including fMRI data of 397 normal control subjects) for testing. Using the constructed individual difference brain network as a priori guidance for feature selection, the symptoms and cognitive scores of schizophrenia are predicted on the BSNIP1 dataset, and patients and normal people are classified at the same time.
[0058] like Figure 2 , 4 The figure shows a comparison of the accuracy of individual difference functional brain network features constructed based on different brain maps in predicting the positive / negative symptoms, total symptom scores and cognitive scores of schizophrenia on the BSNIP1 dataset. The individual difference functional brain network features constructed by this method based on the AAL and BA brain maps can stably predict the multidimensional clinical manifestations of schizophrenia to a certain extent.
[0059] like Figure 3 , 5As shown in the figure, the accuracy and confusion matrix of individual difference functional brain network features constructed based on different brain maps in distinguishing schizophrenia from normal people on the BSNIP1 dataset are compared. Each point represents the accuracy of 10-fold cross validation, and the black dotted line represents the average accuracy of 10-fold cross validation; the colored cells on the diagonal in each confusion matrix (true positive rate) represent the matched true labels and predicted labels; the cells outside the diagonal represent the percentage of misclassification. Overall, the individual difference functional brain network features constructed by this method based on AAL and BA brain maps show similar classification accuracy in distinguishing schizophrenia from normal people.
Claims
1. A method for constructing individual difference brain networks based on sparse kernel matrix optimization, characterized in that: The following steps are involved: (1) Using the brain network atlas, brain region time series are extracted from the normal brain image data of several normal subjects and the abnormal brain images of several abnormal subjects, respectively, to obtain the brain region time series of several normal subjects and the brain region time series of several abnormal subjects; (2) Calculate the similarity matrix of the time series of the brain regions of several normal subjects and the similarity matrix of the time series of the brain regions of several abnormal subjects respectively through similarity measurement; (3) Calculate the absolute value of the difference between the similarity matrix of each abnormal subject and the similarity matrix of all normal subjects to obtain the difference matrices of several brain regions of the abnormal subjects; (4) For the vth brain region difference matrix of each abnormal subject, the similarities of each row of the brain region difference matrix are arranged from large to small, the values of the first N similarities are taken, and the values of other similarities are set to zero to obtain the vth sparse core matrix of the abnormal subject, and the vth sparse core matrix is multiplied by the average value of the remaining brain region difference matrices to obtain the vth optimized brain region difference matrix. Through this step, all optimized brain region difference matrices of all abnormal subjects are obtained; (5) Calculate the average of all optimized brain region difference matrices of each abnormal subject to obtain the individual difference brain network of the abnormal subject.
2. The method for constructing individual difference brain networks according to claim 1, characterized in that: In the step (4), the vth sparse kernel matrix is multiplied by the average value of other normalized brain region difference matrices to obtain the optimized brain region difference matrix of the vth brain region.
3. The method for constructing individual difference brain networks according to claim 2, characterized in that: The vth optimized brain region difference matrix P (v) for: Among them, S (v) represents the vth sparse kernel matrix, D′ (k) represents the kth normalized brain region difference matrix, and m represents the total number of normal subjects.
4. The method for constructing individual difference brain networks according to claim 3, characterized in that: The similarity calculation formula is: Where W(i, j) represents the brain region x of the subject. i and brain region x j Similarity value between time series, ρ 2 (x i , x j ) represents the brain area x of the subject i and brain region x j The square of the Euclidean distance between time series, μ represents the hyperparameter used for empirical settings, and ε i,j represents the hyperparameter used to eliminate the scaling problem.
5. The method for constructing individual difference brain networks according to claim 4, characterized in that: The step (4) specifically comprises: Among them, S(i, j) represents the value of the i-th row and j-th column in the sparse kernel matrix S; D(i, j) is the value of the i-th row and j-th column in the brain region difference matrix D, which represents the absolute value of the difference between W(i, j) of abnormal subjects and W(i, j) of normal subjects; D(i, k) is the value of the i-th row and k-th column in the brain region difference matrix D.
6. The method for constructing individual difference brain networks according to claim 5, characterized in that: Normalize the brain region difference matrix: Wherein, D′ represents the matrix obtained by normalizing the brain region difference matrix D, and D′(i, j) represents the value of the i-th row and j-th column in the normalized brain region difference matrix D′.
7. The method for constructing individual difference brain networks according to claim 1, characterized in that: The normal brain image data and the abnormal brain image data are structural magnetic resonance image data from a public data center, which are preprocessed separately to obtain preprocessed images of normal brain region data and abnormal brain region data; the obtained preprocessed images are divided into brain regions, and the average time series of all voxels in each brain region in each data is calculated.
8. The method for constructing a high-order individual difference brain network according to claim 1, characterized in that: The preprocessed brain image data are divided into brain regions according to the brain atlas, which includes the AAL atlas and the BA atlas.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.
Citation Information
Patent Citations
Fusion brain network construction method based on structural connection and functional connection
CN107658018A
Construction method of first episode schizophrenia individualized prediction model
CN111627553A
Brain functional connectivity correlation value adjustment method, brain functional connectivity correlation value adjustment system, brain activity classifier harmonization method, brain activity classifier harmonization system, and brain activity biomarker system
EP3865061A1
