Abnormal region identification method for brain map data

Abstract

The invention provides an abnormal region recognition method for brain map data, which comprises the following steps: acquiring the brain map data of a target object, the brain map data comprising time sequence data and a connection strength matrix of M brain regions in the brain of the target object; performing sliding window analysis on time sequence data of M brain regions in the brain map data, and determining K groups of time-varying connection matrixes; performing statistical feature extraction and dynamic feature extraction based on the K groups of time-varying connection matrixes, and determining statistical features and dynamic fluctuation features of each brain region; based on the statistical characteristics and the dynamic fluctuation characteristics of each brain region, determining input data of each brain region; and inputting the input data of each brain region into the abnormal region recognition model to obtain an abnormal region recognition result output by the abnormal region recognition model. According to the scheme, whether the abnormal brain region exists or not can be detected by analyzing the connection function of each brain region in the brain map data of the target object.

Description

Technical Field

[0001] The present application relates to the technical field of brain connection data processing, and in particular to a method for identifying abnormal regions in brain map data. Background Art

[0002] In the field of brain science, brain region abnormality identification technology, a cutting-edge research direction in neuroimaging, is evolving towards multimodal fusion. Currently, abnormality detection technology based on structural neuroimaging (such as sMRI and CT) has developed a mature technical approach: voxel-level feature extraction from high-resolution brain structural images using deep learning algorithms (such as 3D-CNN and U-Net architectures), combined with brain map registration technology to quantify morphological parameters such as gray matter volume and cortical thickness, effectively identifying structural lesions such as brain tumors and localized atrophy.

[0003] However, this type of method has fundamental limitations in its sensitivity to functional network abnormalities. Studies have shown that more than 60% of neuropsychiatric diseases (such as depression, bipolar disorder, etc.) are not accompanied by significant structural abnormalities, but will manifest as changes in the topological characteristics of the functional connectivity group, that is, abnormal connectivity functions or abnormal connectivity patterns in some brain regions. The abnormal area identification schemes in the existing technology focus more on the identification of abnormalities at the structural level, and are difficult to apply to the analysis of brain regions with non-structural abnormalities. At present, the problem that needs to be solved urgently in this field is to develop a technology that can analyze brain area abnormalities from the perspective of functional connectivity. Summary of the Invention

[0004] The purpose of the embodiments of the present application is to provide a method for identifying abnormal regions in brain map data, so as to detect whether there are abnormal brain regions by analyzing the connectivity functions of various brain regions in the brain map data of a target object.

[0005] In order to achieve the above objectives, the embodiments of the present application are implemented in the following manner:

[0006] An embodiment of the present application provides a method for identifying abnormal regions in brain map data, comprising: acquiring brain map data of a target object, wherein the brain map data includes time series data and a connection strength matrix of M brain regions in the brain of the target object; performing a sliding window analysis on the time series data of the M brain regions in the brain map data to determine K groups of time-varying connection matrices; performing statistical feature extraction and dynamic feature extraction based on the K groups of time-varying connection matrices to determine the statistical features and dynamic fluctuation features of each brain region; determining input data of each brain region based on the statistical features and dynamic fluctuation features of each brain region; and inputting the input data of each brain region into an abnormal region recognition model to obtain an abnormal region recognition result output by the abnormal region recognition model.

[0007] Furthermore, the time series data of M brain regions in the brain map data contain data of T time nodes. A sliding window analysis is performed on the time series data of the M brain regions in the brain map data to determine K groups of time-varying connection matrices, including: setting the length of the sliding window to 30 seconds and the step length to 10 seconds, and determining a total of K window data; for the kth window data: calculating the correlation coefficient between the i-th brain region and the j-th brain region in the window data, and determining a set of correlation coefficients corresponding to the i-th brain region and the j-th brain region in the k-th window data; based on a set of correlation coefficients between every two brain regions in the window data, forming a set of time-varying connection matrices corresponding to the k-th window data.

[0008] Furthermore, the correlation coefficients between the i-th brain region and the j-th brain region in the window data are calculated, and a set of correlation coefficients corresponding to the i-th brain region and the j-th brain region in the k-th window data are determined, including:

[0009] For the i-th brain region in the k-th window data:

[0010] The following formula is used to calculate a set of correlation coefficients between the i-th brain region and the j-th brain region:

[0011]

[0012] in, It represents the correlation coefficient between the i-th brain region and the j-th brain region in the k-th window data, and contains four attribute data, namely, the 0-node lag correlation coefficient 1-node lagged correlation coefficient 2-node lagged correlation coefficient and the 3-node lagged correlation coefficient T k is the number of time nodes of the k-th window data, is the signal value of the i-th brain region in the window data at time node t, is the average signal value of the i-th brain region at each time node in the window data, is the signal value of the jth brain region in the window data at time node t, is the average signal value of the i-th brain region at each time node in the window data, is the signal value of the jth brain region in the window data at time node t+1, is the signal value of the jth brain region in the window data at time node t+2, is the signal value of the jth brain region in the window data at time node t+3.

[0013] Furthermore, based on K groups of time-varying connection matrices, statistical feature extraction and dynamic feature extraction are performed to determine the statistical features and dynamic fluctuation features of each brain region, including: for the i-th brain region: based on the (M-1) group correlation coefficients between brain region i and other brain regions in the k-th group of time-varying connection matrices, the mean connection strength within the time window of brain region i in the k-th group of time-varying connection matrices is calculated; based on the mean connection strength within the time window of brain region i in the k-th group of time-varying connection matrices, the variance of the connection strength corresponding to brain region i in the k-th group of time-varying connection matrices is calculated; based on the mean connection strength within the time window of brain region i in each group of time-varying connection matrices, the mean connection strength across time windows of brain region i is calculated; based on the mean connection strength within the time window of brain region i and the mean connection strength across time windows, the dynamic fluctuation features of brain region i are calculated.

[0014] Furthermore, based on the (M-1) group correlation coefficients between brain region i and other brain regions in the kth group of time-varying connection matrix, the mean connection strength within the time window of brain region i in the kth group of time-varying connection matrix is calculated, including: for the kth group of time-varying connection matrix: using the following formula to calculate the mean connection strength within the time window of brain region i:

[0015]

[0016] in, is the mean value of the connection strength within the time window of brain region i in the k-th window data, which contains four mean data, namely the mean value of the connection strength within the 0-node lag time window Mean connection strength within the 1-node lag time window Mean connection strength within the 2-node lag time window Mean connection strength within the 3-node lag time window Based on the 0-node lagged correlation coefficient 1-node lagged correlation coefficient 2-node lagged correlation coefficient and the 3-node lagged correlation coefficient Calculated.

[0017] Furthermore, based on the mean value of the connection strength within the time window of the brain region i in the kth group of time-varying connection matrices, the variance of the connection strength corresponding to the brain region i in the kth group of time-varying connection matrices is calculated, including: for the kth group of time-varying connection matrices: using the following formula to calculate the variance of the connection strength of the brain region i:

[0018]

[0019] in, is the variance of the connection strength corresponding to brain region i in the kth group of time-varying connection matrix, which contains four variance data, namely, the variance of the 0-node lagged connection strength 1-node lagged connection strength variance 2-node lagged connection strength variance 3-node lagged connection strength variance Based on the 0-node lagged correlation coefficient The average connection strength within the lag time window with the 0 node 1-node lagged correlation coefficient The average connection strength within the lag time window with 1 node 2-node lagged correlation coefficient The average connection strength within the 2-node lag time window 3-node lagged correlation coefficient The average connection strength within the 3-node lag time window Calculated.

[0020] Furthermore, based on the mean value of the connection strength within the time window of brain region i in each set of time-varying connection matrices, the mean value of the connection strength across time windows of brain region i is calculated, including: using the following formula to calculate the mean value of the connection strength across time windows of brain region i:

[0021]

[0022] in, is the mean value of the cross-time window connection strength of brain region i, which contains four mean data, namely, the mean value of the cross-time window connection strength of 0 node lag Mean connection strength across time windows with 1 node lag Mean of the connection strength across the time window with 2 nodes lag Mean of the connection strength across the time window with 3 nodes lag Based on the mean connection strength within the 0 node lag time window Mean connection strength within the 1-node lag time window Mean connection strength within the 2-node lag time window Mean connection strength within the 3-node lag time window Calculated, K is the total number of window data.

[0023] Furthermore, based on the mean connection strength within the time window and the mean connection strength across time windows of brain region i, the dynamic fluctuation characteristics of brain region i are calculated, including:

[0024] The local dynamic fluctuation of brain area i is calculated using the following formula:

[0025]

[0026] Among them, V i is the local dynamic fluctuation of brain area i, which includes four sets of data: 0 node lag local dynamic fluctuation V i(0), 1 node lag local dynamic volatility V i (1) 2-node lagged local dynamic volatility V i (2) 3-node lagged local dynamic volatility V i (3);

[0027] The overall dynamic fluctuation of brain area i was calculated using the following formula:

[0028]

[0029] Among them, V' i is the overall dynamic volatility of brain area i, which also contains four sets of data, namely, the overall dynamic volatility V' after 0 node lag i (0), 1 node lag overall dynamic volatility V' i (1) 2-node lagged overall dynamic volatility V' i (2) 3-node lagged overall dynamic volatility V' i (3), μ i is the mean overall connection strength of brain region i, Represents the correlation coefficient between brain region i and brain region j in the connection strength matrix of brain map data.

[0030] Furthermore, based on the statistical characteristics and dynamic fluctuation characteristics of each brain region, the input data of each brain region is determined, including: for brain region i: the average connection strength of brain region i in the time window of each window data is calculated. Connection strength variance and the local dynamic fluctuation V of brain area i i and overall dynamic volatility V' i , forming a 4×(2K+2) feature matrix; flattening the feature matrix to form a feature vector with a length of (8K+8) as the input data of brain area i.

[0031] Furthermore, the abnormal region recognition model is constructed based on the multi-layer perceptron model.

[0032] Beneficial effects:

[0033] This approach utilizes brain map data (consisting of time series data and connection strength matrices for M brain regions) for sliding window analysis, identifying K time-varying connectivity matrices. This approach then extracts statistical features (window-level features) and dynamic features (including local and global fluctuations), and performs multi-scale feature fusion to form input data for each brain region. This data is then used in an abnormal region identification model (based on a multi-layered multi-processor logic layer) to identify abnormal regions. This approach overcomes the limitations of traditional static functional connectivity analysis by constructing a time-varying connectivity matrix using multi-time lag cross-correlation calculations (0-3 lags). This approach effectively reflects the time-varying nature of functional connectivity and captures the directional nature of information transfer between brain regions, complementing the principles of causal analysis while maintaining low computational complexity. Furthermore, by calculating the mean cross-window connection strength and local dynamic fluctuations, it simultaneously reflects the temporal stability and instantaneous fluctuations of functional connectivity, making it highly useful for identifying psychiatric disorders such as schizophrenia, which are associated with disturbed dynamic connectivity. The extracted window-level features can reflect the strength distribution of brain region connections within a single sliding window, thereby revealing the instantaneous state of the functional network. Dynamic features, on the other hand, can quantify the temporal variability of brain region connections by calculating cross-window volatility indicators (including local volatility and overall volatility). This indicator is beneficial for identifying default network abnormalities in depression. Finally, the features are fused and flattened to form input data. The abnormal region recognition model constructed based on MLP (because the sample size currently available for training in this solution is not too large, but the feature processing is relatively good, including sufficient multi-level features, does not require further spatial feature extraction through convolution kernels. On a relatively smaller sample size, the parameter efficiency of MLP is higher) is used to identify abnormal regions, which can maximize the recognition accuracy. Based on this, this solution can detect the presence of abnormal brain regions by analyzing the connectivity functions of each brain region in the brain map data of the target subject.

[0034] The innovative design of extracting multi-order lagged correlation features effectively reduces the interference of high-frequency noise in the signal and improves the signal-to-noise ratio. Furthermore, the extraction of multi-order lagged correlation features can reflect the direction of information transmission between brain regions and the causal patterns between them, thus enhancing the information-reflecting power of the features. The window-level features extracted based on this (i.e., statistical features of window data) can reflect the strength distribution of brain region connections within a single time window and the instantaneous state of the functional network. Furthermore, the dynamic features extracted complement each other: global dynamic fluctuation and local dynamic fluctuation. Local dynamic fluctuation reflects the temporal variability of brain region connection strength, while global dynamic fluctuation reflects the degree to which brain region connectivity patterns deviate from the population norm. Therefore, it is more sensitive to diseases associated with decreased functional connectivity stability (such as Alzheimer's disease) and disorders with abnormal connectivity patterns (such as autism), which helps improve the effectiveness of abnormal region identification. The input data constructed based on this (feature vector of length (8K+8)) has a low overall computational complexity, ensuring high system efficiency.

[0035] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 This is a flowchart of a method for identifying abnormal areas in brain map data provided in an embodiment of the present application.

[0038] Figure 2 This is a schematic diagram showing the conditions for introducing visual abnormal areas into our unit's brain mapping software. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0040] Since the abnormal area recognition model used in this embodiment relies on MLP (Multilayer Perceptron) construction, and the model construction process is relatively simple, it is sufficient to use TensorFlow to define the model and set relevant parameters. The focus is on the processing of feature data. The feature processing process of training data is also similar. Therefore, this embodiment does not introduce the processing process of training data separately, but instead introduces the process of abnormal area recognition of brain map data of a target object.

[0041] See also Figure 1 , Figure 1 Flowchart of the abnormal region identification method of brain map data provided in an embodiment of the present application. In this embodiment, the abnormal region identification method of brain map data may include steps S10, S20, S30, S40, and S50.

[0042] In this embodiment, step S10 may be executed first.

[0043] Step S10: Acquire brain map data of the target object, wherein the brain map data includes time series data and connection strength matrix of M brain regions in the brain of the target object.

[0044] In this embodiment, brain map data of the target subject can be obtained. The brain map data includes time series data and a connection strength matrix for M brain regions in the target subject's brain. Currently, there are many software programs that can use the target subject's fMRI data (combined with sMRI data or T1 structural imaging, etc.) to generate a brain network map (including the M brain regions and the connectivity relationships between brain regions) of the target subject, thereby obtaining the time series data and connection strength matrix for the M brain regions of the target subject. This will not be described in detail here.

[0045] After obtaining the brain map data of the target object, step S20 may be further executed.

[0046] Step S20: performing sliding window analysis on the time series data of the M brain regions in the brain map data to determine K groups of time-varying connection matrices.

[0047] In this embodiment, in order to meet the minimum time unit requirement for dynamic analysis of functional connectivity, the length of the sliding window is set to 30 seconds, the step length is 10 seconds, and a total of K window data are determined.

[0048] For example, if the time series data of a brain region is 10 minutes (the signal is collected every 2 seconds), then it contains 300 nodes, and each sliding window contains 15 nodes (corresponding to a step size of 5 nodes), corresponding to 58 window data.

[0049] For each window data, taking the kth window data as an example, the following processing is required:

[0050] The correlation coefficients between the i-th brain region and the j-th brain region in the window data are calculated, and a set of correlation coefficients corresponding to the i-th brain region and the j-th brain region in the k-th window data are determined.

[0051] For example, for the i-th brain region in the k-th window data, the following formula is used to calculate a set of correlation coefficients between the i-th brain region and the j-th brain region:

[0052]

[0053]

[0054] in, It represents the correlation coefficient between the i-th brain region and the j-th brain region in the k-th window data, and contains four attribute data, namely, the 0-node lag correlation coefficient 1-node lagged correlation coefficient 2-node lagged correlation coefficient and the 3-node lagged correlation coefficient T k is the number of time nodes of the k-th window data, is the signal value of the i-th brain region in the window data at time node t, is the average signal value of the i-th brain region at each time node in the window data, is the signal value of the jth brain region in the window data at time node t, is the average signal value of the i-th brain region at each time node in the window data, is the signal value of the jth brain region in the window data at time node t+1, is the signal value of the jth brain region in the window data at time node t+2, is the signal value of the jth brain region in the window data at time node t+3.

[0055] Different from the traditional correlation coefficient calculation scheme, this method introduces multi-order lag for subsequent processing to extract features that can reflect the causal pattern (i.e., the connection direction) to improve the detection effect of abnormal areas.

[0056] Based on this, a set of correlation coefficients between every two brain regions in each window data is calculated (the correlation coefficient between brain region i and itself can be recorded as 1), so that a set of time-varying connection matrices corresponding to the k-th window data can be formed (in fact, there are 4 time-varying connection matrices, corresponding to the time-varying connection matrices formed by different numbers of lagging nodes).

[0057] After obtaining a set of time-varying connection matrices corresponding to each window data (a total of K sets of time-varying connection matrices are obtained), step S30 may be further performed.

[0058] Step S30: Based on the K groups of time-varying connection matrices, statistical feature extraction and dynamic feature extraction are performed to determine the statistical features and dynamic fluctuation features of each brain region.

[0059] In this embodiment, it is necessary to perform feature extraction on each brain region in each set of time-varying connection matrices (i.e., corresponding to each window data). Here, the i-th brain region in the k-th set of time-varying connection matrices is used as an example for description:

[0060] Specifically, the mean connection strength within the time window of brain region i in the kth group of time-varying connection matrix can be calculated based on the (M-1) group correlation coefficients between brain region i and other brain regions in the kth group of time-varying connection matrix.

[0061] For example, for the kth group of time-varying connection matrices, the following formula is used to calculate the mean connection strength within the time window of brain region i:

[0062]

[0063] in, is the mean value of the connection strength within the time window of brain region i in the k-th window data, which contains four mean data, namely the mean value of the connection strength within the 0-node lag time window Mean connection strength within the 1-node lag time window Mean connection strength within the 2-node lag time window Mean connection strength within the 3-node lag time window Based on the 0-node lagged correlation coefficient 1-node lagged correlation coefficient 2-node lagged correlation coefficient and the 3-node lagged correlation coefficient The calculation is obtained, and the formula (6) can be entered during the calculation, which is not shown here.

[0064] Afterwards, the variance of the connection strength corresponding to the brain region i in the kth group of time-varying connection matrices can be calculated based on the mean value of the connection strength within the time window of the brain region i in the kth group of time-varying connection matrices.

[0065] For the kth group of time-varying connection matrices, the following formula is used to calculate the connection strength variance of brain region i:

[0066]

[0067] in, is the variance of the connection strength corresponding to brain region i in the kth group of time-varying connection matrix, which contains four variance data, namely, the variance of the 0-node lagged connection strength 1-node lagged connection strength variance 2-node lagged connection strength variance 3-node lagged connection strength variance Based on the 0-node lagged correlation coefficient The average connection strength within the lag time window with the 0 node 1-node lagged correlation coefficient The average connection strength within the lag time window with 1 node 2-node lagged correlation coefficient The average connection strength within the 2-node lag time window 3-node lagged correlation coefficient The average connection strength within the 3-node lag time window The calculation is obtained, and the formula (7) can be entered during the calculation, which is not shown here.

[0068] Furthermore, the mean value of the connection strength across time windows of brain region i can be calculated based on the mean value of the connection strength within the time window of brain region i in each set of time-varying connection matrices.

[0069] For brain region i, the following formula is used to calculate the mean connection strength of brain region i across time windows:

[0070]

[0071] in, is the mean value of the cross-time window connection strength of brain region i, which contains four mean data, namely, the mean value of the cross-time window connection strength of 0 node lag Mean connection strength across time windows with 1 node lag Mean of the connection strength across the time window with 2 nodes lag Mean of the connection strength across the time window with 3 nodes lag Based on the mean connection strength within the 0 node lag time window Mean connection strength within the 1-node lag time window Mean connection strength within the 2-node lag time window Mean connection strength within the 3-node lag time window Calculate (enter formula (8) during calculation, not shown here), K is the total number of window data.

[0072] Furthermore, the dynamic fluctuation characteristics of brain region i can be calculated based on the mean connection strength within the time window and the mean connection strength across time windows of brain region i. In this embodiment, the dynamic fluctuation characteristics include two types of fluctuation calculations: local dynamic fluctuation and global dynamic fluctuation.

[0073] First, the local dynamic fluctuation of brain area i is calculated using the following formula:

[0074]

[0075] Among them, V i is the local dynamic fluctuation of brain area i, which includes four sets of data: 0 node lag local dynamic fluctuation V i (0), 1 node lag local dynamic volatility V i (1) 2-node lagged local dynamic volatility V i (2) 3-node lagged local dynamic volatility V i (3).

[0076] Secondly, the overall dynamic fluctuation of brain area i was calculated using the following formula:

[0077]

[0078] Among them, V' i is the overall dynamic volatility of brain area i, which also contains four sets of data, namely, the overall dynamic volatility V' after 0 node lag i (0), 1 node lag overall dynamic volatility V' i (1) 2-node lagged overall dynamic volatility V' i (2) 3-node lagged overall dynamic volatility V' i (3), μ i is the mean overall connection strength of brain region i, Represents the correlation coefficient between brain region i and brain region j in the connection strength matrix of brain map data.

[0079] After the statistical features of each window data in each brain region and the dynamic fluctuation features of each brain region are calculated accordingly, step S40 may be executed.

[0080] Step S40: Based on the statistical characteristics and dynamic fluctuation characteristics of each brain region, the input data of each brain region is determined.

[0081] In this embodiment, each brain region is also processed as a unit. For brain region i:

[0082] The mean connection strength of brain region i in the time window of each window data can be Connection strength variance and the local dynamic fluctuation V of brain area i i and overall dynamic volatility V' i , forming a 4×(2K+2) feature matrix:

[0083]

[0084] in, and is K elements (taking 58 window data as an example, then, and are 58 elements), and V i and V' i They are all one element, so the number of columns of the matrix is (2K+2).

[0085] Afterwards, the feature matrix can be flattened to form a feature vector of length (8K+8) as the input data of brain region i. For example, Of course, the actual processing process can also be other forms of feature fusion. This embodiment considers the abnormal area recognition model built on the basis of MLP. Therefore, the features of each level of each brain region are fused into a feature vector with a length of (8K+8), which is conducive to model processing.

[0086] After obtaining the input data corresponding to each brain region, step S50 may be executed.

[0087] Step S50: inputting the input data of each brain region into the abnormal region recognition model to obtain the abnormal region recognition result output by the abnormal region recognition model.

[0088] In this embodiment, the input data for each brain region can be fed into the abnormal region identification model, which then outputs the abnormal region identification results. Considering that combining the input data from all brain regions to construct a whole-brain model would be difficult to achieve with the current amount of available training data, this embodiment temporarily implements a multi-level, parallel, single-brain region model.

[0089] That is, we need to build M sub-models (each sub-model is built based on MLP, which can be done with TensorFlow. Take 3 hidden layers as an example, the number of neurons in each layer decreases, the hidden layer activation function is ReLU, the output layer activation function is Sigmoid, Dropout is set to 0.5, and the L2 regularization weight decay is set to e -4 , Adam is selected as the optimizer, and the learning rate is set to 0.001), and each sub-model is numbered according to the brain region, and brain region-level training data is used (the same feature processing method is also used, the difference is that it has labels). At present, the labels of data in this field are usually at the object level. Therefore, the labeling work of this part of the brain region-level labeled data (as training data) needs to be completed manually. Of course, a pseudo-label strategy can also be used to determine the abnormal areas through object-level labels for labeling, and the labels of the remaining areas are designed to be credible. For normal subjects, the labels of all areas are marked as normal, and different weights are assigned to the three types of labels. For example, the abnormal labels are weighted 1.5, the normal labels are weighted 1.0, and the credible labels are weighted 0.8. In this way, the processing of the training data is completed and a training set is formed to train the corresponding sub-model.

[0090] Finally, the abnormal region recognition model can be used to determine the abnormal region recognition results at the brain region level to achieve abnormal region recognition.

[0091] The abnormal area recognition results can be fed back into the brain map data and visualized by coloring (different colors), such as the brain map software of our unit (but it has not yet been connected to this software function, and may be introduced in subsequent versions), such as Figure 2 shown.

[0092] In summary, the embodiment of the present application provides a method for identifying abnormal areas of brain map data, which uses brain map data (including time series data and connection strength matrices of M brain regions in the brain) for sliding window analysis to determine K groups of time-varying connection matrices, further perform statistical feature extraction (window-level features) and dynamic feature extraction (including local volatility and overall volatility), perform multi-scale feature fusion, form input data for each brain region, and use the abnormal region identification model (based on MLP construction) to obtain abnormal region identification results. This solution breaks through the limitations of traditional static functional connectivity analysis and constructs a time-varying connection matrix through multi-time lag cross-correlation calculation (0-3 order lag). It can effectively reflect the time-varying characteristics of functional connectivity and effectively capture the directional characteristics of information transmission between brain regions, which complements the principle of causal analysis and has low computational complexity. By calculating the mean value of cross-window connection strength and local dynamic volatility, the temporal stability and instantaneous fluctuation characteristics of functional connectivity can be reflected simultaneously, which is very beneficial for identifying mental illnesses such as schizophrenia accompanied by dynamic connectivity disorders. The extracted window-level features can reflect the intensity distribution characteristics of brain area connections within a single sliding window, thereby revealing the instantaneous state of the functional network. The dynamic features can quantify the temporal variability of brain area connections by calculating cross-window volatility indicators (including local volatility and global volatility). This indicator is conducive to the identification of default network abnormalities in depression. Finally, the features are fused and flattened to form input data. The abnormal area recognition model constructed based on MLP (because the sample size currently available for training in this scheme is not too large, but the feature processing is relatively good, containing sufficient multi-level features, does not require further spatial feature extraction through convolution kernels. On a relatively smaller sample size, the parameter efficiency of MLP is higher) is used to identify abnormal areas, which can maximize the recognition accuracy. Based on this, this scheme can detect the presence of abnormal brain areas by analyzing the connectivity functions of each brain area in the brain map data of the target subject.

[0093] The innovative design of extracting multi-order lagged correlation features effectively reduces the interference of high-frequency noise in the signal and improves the signal-to-noise ratio. Furthermore, the extraction of multi-order lagged correlation features can reflect the direction of information transmission between brain regions and the causal patterns between them, thus enhancing the information-reflecting power of the features. The window-level features extracted based on this (i.e., statistical features of window data) can reflect the strength distribution of brain region connections within a single time window and the instantaneous state of the functional network. Furthermore, the dynamic features extracted complement each other: global dynamic fluctuation and local dynamic fluctuation. Local dynamic fluctuation reflects the temporal variability of brain region connection strength, while global dynamic fluctuation reflects the degree to which brain region connectivity patterns deviate from the population norm. Therefore, it is more sensitive to diseases associated with decreased functional connectivity stability (such as Alzheimer's disease) and disorders with abnormal connectivity patterns (such as autism), which helps improve the effectiveness of abnormal region identification. The input data constructed based on this (feature vector of length (8K+8)) has a low overall computational complexity, ensuring high system efficiency.

[0094] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for identifying abnormal regions in brain map data, characterized in that: include: Acquiring brain map data of a target subject, wherein the brain map data includes time series data and a connection strength matrix of M brain regions in the target subject's brain; Perform sliding window analysis on the time series data of M brain regions in the brain map data to determine K groups of time-varying connection matrices; Based on K groups of time-varying connection matrices, statistical feature extraction and dynamic feature extraction are performed to determine the statistical features and dynamic fluctuation features of each brain region; Based on the statistical characteristics and dynamic fluctuation characteristics of each brain region, the input data of each brain region is determined; The input data of each brain region is input into the abnormal region recognition model to obtain the abnormal region recognition result output by the abnormal region recognition model.

2. The abnormal area identification method of brain map data according to claim 1, characterized in that: The time series data of M brain regions in the brain map data contain data of T time nodes. Sliding window analysis is performed on the time series data of M brain regions in the brain map data to determine K groups of time-varying connection matrices, including: The length of the sliding window is set to 30 seconds, the step length is 10 seconds, and a total of K window data are determined; For the k-th window data: Calculate the correlation coefficient between the i-th brain region and the j-th brain region in the window data, and determine a set of correlation coefficients corresponding to the i-th brain region and the j-th brain region in the k-th window data; Based on a set of correlation coefficients between every two brain regions in the window data, a set of time-varying connection matrices corresponding to the k-th window data is formed.

3. The abnormal region identification method of brain map data according to claim 2, characterized in that: Calculate the correlation coefficient between the i-th brain region and the j-th brain region in the window data, and determine a set of correlation coefficients corresponding to the i-th brain region and the j-th brain region in the k-th window data, including: For the i-th brain region in the k-th window data: The following formula is used to calculate a set of correlation coefficients between the i-th brain region and the j-th brain region: in, It represents the correlation coefficient between the i-th brain region and the j-th brain region in the k-th window data, and contains four attribute data, namely, the 0-node lag correlation coefficient 1-node lagged correlation coefficient 2-node lagged correlation coefficient and the 3-node lagged correlation coefficient T k is the number of time nodes of the k-th window data, is the signal value of the i-th brain region in the window data at time node t, is the average signal value of the i-th brain region at each time node in the window data, is the signal value of the jth brain region in the window data at time node t, is the average signal value of the i-th brain region at each time node in the window data, is the signal value of the jth brain region in the window data at time node t+1, is the signal value of the jth brain region in the window data at time node t+2, is the signal value of the jth brain region in the window data at time node t+3.

4. The abnormal region identification method of brain map data according to claim 3, characterized in that: Based on K groups of time-varying connection matrices, statistical feature extraction and dynamic feature extraction are performed to determine the statistical features and dynamic fluctuation features of each brain region, including: For the i-th brain region: Based on the (M-1) correlation coefficients between brain region i and other brain regions in the kth group of time-varying connectivity matrix, the mean connection strength within the time window of brain region i in the kth group of time-varying connectivity matrix was calculated; Based on the mean connection strength within the time window of brain region i in the kth group of time-varying connection matrix, the variance of the connection strength corresponding to brain region i in the kth group of time-varying connection matrix is calculated; Based on the mean connection strength within the time window of brain region i in each set of time-varying connection matrices, the mean connection strength across time windows of brain region i is calculated; Based on the mean connection strength within the time window and the mean connection strength across time windows of brain region i, the dynamic fluctuation characteristics of brain region i are calculated.

5. The abnormal region identification method of brain map data according to claim 4, characterized in that: Based on the (M-1) correlation coefficients between brain region i and other brain regions in the kth group of time-varying connectivity matrix, the mean connection strength within the time window of brain region i in the kth group of time-varying connectivity matrix is calculated, including: For the kth group of time-varying connection matrix: The mean connection strength within the time window of brain region i is calculated using the following formula: in, is the mean value of the connection strength within the time window of brain region i in the k-th window data, which contains four mean data, namely the mean value of the connection strength within the 0-node lag time window Mean connection strength within the 1-node lag time window Mean connection strength within the 2-node lag time window Mean connection strength within the 3-node lag time window Based on the 0-node lagged correlation coefficient 1-node lagged correlation coefficient 2-node lagged correlation coefficient and the 3-node lagged correlation coefficient Calculated.

6. The abnormal region identification method of brain map data according to claim 5, characterized in that: Based on the mean connection strength within the time window of brain region i in the kth group of time-varying connection matrices, the variance of the connection strength corresponding to brain region i in the kth group of time-varying connection matrices is calculated, including: For the kth group of time-varying connection matrix: The following formula is used to calculate the variance of the connection strength of brain region i: in, is the variance of the connection strength corresponding to brain region i in the kth group of time-varying connection matrix, which contains four variance data, namely, the variance of the 0-node lagged connection strength 1-node lagged connection strength variance 2-node lagged connection strength variance 3-node lagged connection strength variance Based on the 0-node lagged correlation coefficient The average connection strength within the lag time window with the 0 node 1-node lagged correlation coefficient The average connection strength within the lag time window with 1 node 2-node lagged correlation coefficient The average connection strength within the 2-node lag time window 3-node lagged correlation coefficient The average connection strength within the 3-node lag time window Calculated.

7. The abnormal region identification method of brain map data according to claim 5, characterized in that: Based on the mean connection strength within the time window of brain region i in each set of time-varying connection matrices, the mean connection strength across time windows of brain region i is calculated, including: The following formula is used to calculate the mean connectivity strength of brain region i across time windows: in, is the mean value of the cross-time window connection strength of brain region i, which contains four mean data, namely, the mean value of the cross-time window connection strength of 0 node lag Mean connection strength across time windows with 1 node lag Mean of the connection strength across the time window with 2 nodes lag Mean of the connection strength across the time window with 3 nodes lag Based on the mean connection strength within the 0 node lag time window Mean connection strength within the 1-node lag time window Mean connection strength within the 2-node lag time window Mean connection strength within the 3-node lag time window Calculated, K is the total number of window data.

8. The abnormal region identification method of brain map data according to claim 7, characterized in that: Based on the mean connection strength within the time window and the mean connection strength across time windows of brain region i, the dynamic fluctuation characteristics of brain region i are calculated, including: The local dynamic fluctuation of brain area i is calculated using the following formula: Among them, V i is the local dynamic fluctuation of brain area i, which includes four sets of data: 0 node lag local dynamic fluctuation V i (0), 1 node lag local dynamic volatility V i (1) 2-node lagged local dynamic volatility V i (2) 3-node lagged local dynamic volatility V i (3); The overall dynamic fluctuation of brain area i was calculated using the following formula: Among them, V' i is the overall dynamic volatility of brain area i, which also contains four sets of data, namely, the overall dynamic volatility V' after 0 node lag i (0), 1 node lag overall dynamic volatility V' i (1) 2-node lagged overall dynamic volatility V' i (2) 3-node lagged overall dynamic volatility V' i (3), μ i is the mean overall connection strength of brain region i, Represents the correlation coefficient between brain area i and brain area j in the connection strength matrix of brain map data.

9. The abnormal region identification method of brain map data according to claim 5, characterized in that: Based on the statistical characteristics and dynamic fluctuation characteristics of each brain region, the input data of each brain region is determined, including: Targeting brain area i: The average connection strength of brain region i in the time window of each window data Connection strength variance and the local dynamic fluctuation V of brain area i i and overall dynamic volatility V' i , forming a 4×(2K+2) feature matrix; The feature matrix is flattened to form a feature vector of length (8K+8) as the input data of brain region i.

10. The abnormal region identification method of brain map data according to claim 1, characterized in that: The abnormal region recognition model is built based on the multi-layer perceptron model.

Images