Graph convolution method and device for enhancing rs-fmri feature analysis specific roi

Abstract

The application provides a kind of graph convolution method and device for enhancing rs-fMRI feature analysis specific ROI, it is related to artificial intelligence and medical imaging technical field, the method comprises: obtaining resting-state magnetic resonance imaging data;Extract the blood oxygen level dependent signal of each region of interest in resting-state magnetic resonance imaging data;Blood oxygen level dependent signal is carried out chaos processing, and blood oxygen level dependent chaotic signal is obtained;Blood oxygen level dependent signal and blood oxygen level dependent chaotic signal are dynamically fused, and a fusion signal matrix is obtained;Sample sequence of the fusion signal matrix is obtained based on the preset time window, and an adjacency matrix is constructed based on the sample sequence;The data corresponding to the adjacency matrix is input into the neural network model, and the calculation result is output;Specific region of interest pair is obtained based on the calculation result.It makes up for the deficiency of the existing method for extracting potential features of the brain, thereby improving the classification accuracy, thereby positioning the specific ROI of the nervous system disease.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and medical imaging technology, and particularly relates to a graph convolution method for enhancing rs-fMRI feature analysis of specific ROIs, a device, an electronic device and a computer readable storage medium. BACKGROUND

[0002] Neurological diseases cover a wide range, such as autism spectrum disorders and cognitive disorders, caused by a combination of environmental, genetic and psychological factors. The identification of special biomarkers for such mental diseases is of great help to subsequent targeted research. The lack of accurate pathological identification as a diagnostic reference will affect targeted diagnosis.

[0003] Resting state functional magnetic resonance imaging (rs-fMRI) is a non-invasive technique for measuring spontaneous brain activity through blood-oxygen-level-dependent (BOLD) signals. It is now increasingly used to study neurological system diseases. However, current research still needs to be improved for accurate identification of specific regions of interest (ROIs) for each mental disease.

[0004] The ability of graph neural networks to construct graph nodes through each ROI for the diagnosis of neurological system diseases is gradually improving, but existing research has ignored the interference of surrounding factors with the BOLD signals of each ROI during the acquisition process, resulting in deviations in the acquired BOLD signals and subsequent inaccurate ROI analysis.

[0005] The current deep learning technology used for feature extraction is not sufficient, although many existing methods have already achieved high diagnostic accuracy, the preservation and enhancement of original signal features still need to be improved. SUMMARY

[0006] Therefore, the present application provides a graph convolution method for enhancing rs-fMRI feature analysis of specific ROIs, a device, an electronic device and a computer readable storage medium, which can solve the problem that the current deep learning technology used for feature extraction is not sufficient and the preservation and enhancement of original signal features are not perfect.

[0007] Some embodiments of the present application provide a graph convolution method for enhancing rs-fMRI feature analysis of specific ROIs. The following aspects of the present application are introduced from multiple aspects, and the embodiments and advantages of the following aspects can be mutually referenced.

[0008] In a first aspect, the present application provides a graph convolution method for enhancing rs-fMRI feature analysis of specific ROIs, comprising:

[0009] obtaining resting-state magnetic resonance imaging data;

[0010] extracting blood oxygen level dependent signals of each region of interest in the resting-state magnetic resonance imaging data;

[0011] chaotically processing the blood oxygen level dependent signals based on a chaotic model to obtain blood oxygen level dependent chaotic signals;

[0012] dynamically fusing the blood oxygen level dependent signals and the blood oxygen level dependent chaotic signals to obtain a fusion signal matrix;

[0013] obtaining a sample sequence of the fusion signal matrix based on a preset time window, and constructing an adjacency matrix based on the sample sequence, the adjacency matrix being used to represent a time transformation relationship of the blood oxygen level dependent signals in the region of interest;

[0014] inputting data corresponding to the adjacency matrix into a neural network model to output a calculation result;

[0015] obtaining a specific region of interest pair based on the calculation result.

[0016] In a possible implementation of the first aspect, the method further comprises:

[0017] performing data correction based on the resting-state magnetic resonance imaging data to obtain corrected data;

[0018] filling the corrected data into an automatic anatomical labeling template to obtain filled data;

[0019] extracting the blood oxygen level dependent signals of each region of interest based on the automatic anatomical labeling template and the filled data.

[0020] In a possible implementation of the first aspect, the method further comprises:

[0021] selecting a blood oxygen level dependent signal of a target region with the highest functional connectivity in the region of interest as data input of an initial chaotic model;

[0022] dynamically modulating a period control parameter of the initial chaotic model based on the blood oxygen level dependent signal of the target region to obtain a chaotic model;

[0023] solving a differential equation of the chaotic model to obtain the blood oxygen level dependent chaotic signals.

[0024] In a possible implementation of the first aspect, the blood oxygen level dependent signals and the blood oxygen level dependent chaotic signals are dynamically fused to obtain a fused signal matrix, including:

[0025] The blood oxygen level dependent chaotic signals are standardized.

[0026] One component in the blood oxygen level dependent chaotic signals is selected.

[0027] The component is dynamically fused with the blood oxygen level dependent signals to obtain a fused signal matrix.

[0028] In a possible implementation of the first aspect, the data corresponding to the adjacency matrix is input into a neural network model, and a calculation result is output, including:

[0029] Local features are aggregated based on the data corresponding to the adjacency matrix to obtain the local features.

[0030] Global feature enhancement is performed based on the local features to obtain enhanced features.

[0031] Feature extraction is performed based on the enhanced features to obtain multi-scale features.

[0032] The calculation result is obtained based on the multi-scale features.

[0033] In a possible implementation of the first aspect, the specific ROI pair is obtained based on the calculation result, including:

[0034] The adjacency matrix of the abnormal resting-state magnetic resonance imaging data is output based on the calculation result.

[0035] The adjacency matrix of the abnormal resting-state magnetic resonance imaging data is compared with the adjacency matrix of the control group to obtain the specific ROI pair.

[0036] In a possible implementation of the first aspect, after the blood oxygen level dependent signals of each ROI in the resting-state magnetic resonance imaging data are extracted, the method further includes:

[0037] The blood oxygen level dependent signals of each ROI are standardized.

[0038] In a second aspect, the present application provides a graph convolution device for enhancing rs-fMRI feature analysis specificity ROI, including:

[0039] An acquisition module is configured to acquire resting-state magnetic resonance imaging data.

[0040] An extraction module is configured to extract blood oxygen level dependent signals of each ROI in the resting-state magnetic resonance imaging data.

[0041] A chaos processing module is configured to perform chaos processing on the blood oxygen level dependent signal based on a chaos model to obtain a blood oxygen level dependent chaos signal.

[0042] A fusion module is configured to dynamically fuse the blood oxygen level dependent signal and the blood oxygen level dependent chaos signal to obtain a fusion signal matrix.

[0043] A construction module is configured to obtain a sample sequence of the fusion signal matrix based on a preset time window, and construct an adjacency matrix based on the sample sequence, the adjacency matrix being configured to represent a time transformation relationship of the blood oxygen level dependent signal in the region of interest.

[0044] A classification module is configured to input data corresponding to the adjacency matrix into a neural network model to output a calculation result.

[0045] An analysis module is configured to obtain a specific region of interest pair based on the calculation result.

[0046] In a third aspect, the present application provides an electronic device, comprising: a processor; and a memory having computer program instructions stored therein.

[0047] When the computer program instructions are run by the processor, the processor executes the graph convolution method for enhancing rs-fMRI feature analysis specificity ROI based on a chaos system.

[0048] In a fourth aspect, the present application provides a computer readable storage medium storing a computer program, which, when run by a processor, causes the processor to execute the graph convolution method for enhancing rs-fMRI feature analysis specificity ROI based on a chaos system.

[0049] In a fifth aspect, the present application discloses a device, comprising:

[0050] A memory is configured to store instructions executed by one or more processors of the device, and the processor is one of the processors of the device and is configured to execute the method disclosed in any one of the first to fourth aspects.

[0051] The present application can effectively fuse the blood oxygen level dependent chaos signal with the blood oxygen level dependent signal of each region of interest, thereby reducing the influence of external factors on the blood oxygen level dependent signal, improving the accuracy of neural network calculation, and making the found difference-specific region of interest pair more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 A flowchart of the graph convolution method for enhancing rs-fMRI feature analysis specificity ROI of the embodiments of the present application;

[0053] Figure 2A model graph for a graph convolution method of the embodiment of the application for enhancing rs-fMRI feature analysis specific ROIs;

[0054] Figure 3 A flowchart of step S120 of the embodiment of the application;

[0055] Figure 4 A flowchart of step S130 of the embodiment of the application;

[0056] Figure 5 A flowchart of step S140 of the embodiment of the application;

[0057] Figure 6 A visualized experimental result graph of the embodiment of the application;

[0058] Figure 7 A flowchart of step S160 of the embodiment of the application;

[0059] Figure 8 A flowchart of step S170 of the embodiment of the application;

[0060] Figure 9 A block diagram of the device of the embodiment of the application;

[0061] Figure 10 A block diagram of the SoC (System on Chip) of the embodiment of the application. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are some but not all of the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the application.

[0063] In order to facilitate the understanding of the technical solutions of the application, the technical problems to be solved by the application are first described.

[0064] Resting-state magnetic resonance imaging is a non-invasive technique for measuring spontaneous brain activity through blood oxygen level dependent signals, and is now increasingly used to study neurological system diseases. However, the current research on accurately identifying specific ROIs of various mental diseases still needs to be improved.

[0065] With the increasing ability of graph neural networks to construct graph nodes for each ROI for the diagnosis of neurological system diseases, it is convenient to identify specific ROIs of various mental diseases, but at the same time, there are some technical problems, such as:

[0066] (1) In the process of BOLD signal acquisition, it will be interfered by environmental noise (such as scanner artifacts, physiological noise, head motion) and neurovascular coupling effect, which will cause signal distortion and specific ROI identification deviation.

[0067] (2) Traditional graph neural network relies on static adjacency matrix, and does not fully consider the time-varying characteristics of BOLD signal, which leads to insufficient spatiotemporal dynamic modeling of functional connection network between ROIs.

[0068] (3) The original features are not preserved enough, and single modality (such as only BOLD signal) is difficult to fully represent the functional characteristics of ROI.

[0069] In order to solve the above technical problems, the present application provides a graph convolution method for enhancing rs-fMRI feature analysis of specific ROI, which modulates the BOLD signal by a chaotic model to enhance the signal noise immunity and feature robustness; dynamically fuses the BOLD signal and the chaotic signal, and realizes signal complementation by using standardized components; extracts the sample sequence of the fused signal by using a preset time window, constructs an adjacency matrix reflecting the dynamic evolution of the signal, captures the time dependence between ROIs, and improves the adaptability of the graph structure to the dynamic brain network representation; compares the adjacency matrix difference between the abnormal group and the control group, combines the multi-scale feature classification of the neural network, and selects the abnormal specific ROI pair, so as to improve the reliability of biomarker identification.

[0070] The graph convolution method for enhancing rs-fMRI feature analysis of specific ROI of the embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0071] Reference Figure 1 and Figure 2 , Figure 1 The flowchart of the graph convolution method for enhancing rs-fMRI feature analysis of specific ROI of the embodiment of the present application is shown. Figure 2 The model diagram of the graph convolution method for enhancing rs-fMRI feature analysis of specific ROI of the embodiment of the present application is shown.

[0072] The method comprises steps S110-S170.

[0073] S110, acquiring resting state magnetic resonance imaging data.

[0074] In the embodiment of the present application, the resting state magnetic resonance imaging data of the subject in the closed eye or open eye state without external task or stimulation, keeping awake but avoiding active thinking activity, can be collected by the MRI scanner with an acquisition time length of 5-15 minutes.

[0075] As Figure 2As shown in the implementation of the present application, step S110 is equivalent to step A inputting the rs-fMRI file, which is used to obtain the resting-state magnetic resonance imaging data.

[0076] S120, extracting the blood oxygen level dependent signals of each region of interest in the resting-state magnetic resonance imaging data.

[0077] As shown in the implementation of the present application, step S120 is equivalent to step B registering the template to divide the ROI and step C extracting the time series, which is used to register the corrected resting-state magnetic resonance imaging data to the automatic anatomical marker template and extract the time series of each ROI. Figure 2 It should be noted that by performing data correction based on the resting-state magnetic resonance imaging data, the noise, artifacts and other interference factors in the original data can be effectively removed, so that more accurate and reliable corrected data can be obtained.

[0078] By filling the corrected data into the automatic anatomical marker template, the standardized structure and positioning information of the template can be used to accurately identify and locate each region of interest in the brain.

[0079] Based on the automatic anatomical marker template, the filled data can be extracted to obtain the blood oxygen level dependent signals of each region of interest, improve the efficiency and accuracy of signal extraction, avoid errors and subjectivity caused by manual extraction, and provide key data support for in-depth study of brain function activity.

[0080] S130, performing chaotic processing on the blood oxygen level dependent signals based on a chaotic model to obtain blood oxygen level dependent chaotic signals.

[0081] As shown in the implementation of the present application, step S130 is equivalent to step D generating chaotic signals, which is used to obtain the blood oxygen level dependent chaotic signals.

[0082] Figure 2 It should be noted that the blood oxygen level dependent chaotic signals are high-dimensional chaotic signals reconstructed by a chaotic model based on the nonlinear dynamic characteristics of the blood oxygen level dependent signals in the resting-state magnetic resonance imaging data.

[0083] By processing the blood oxygen level dependent signals based on the chaotic model, the small dynamic changes and complex characteristics in the signals can be more sensitively captured, which helps to discover potential information that is difficult to detect by traditional linear analysis methods, so as to more deeply understand the complex mechanism of brain activity.

[0084]

[0085] ​​By dynamically modulating the periodic control parameters of the initial chaotic model based on the blood oxygen level-dependent signal of the target region, the chaotic model can better adapt to the characteristics and changing patterns of the actual signal, thereby improving the model's fitting accuracy and predictive ability for the blood oxygen level-dependent signal.

[0086] S140 dynamically fuses the blood oxygen level-dependent signal and the blood oxygen level-dependent chaotic signal to obtain a fused signal matrix.

[0087] like Figure 2 As shown, in the implementation of this invention, step S140 is equivalent to step D generating a chaotic signal and spreading it to the whole brain, and step E dynamically fusing it with the standardized chaotic signal. This is used to map the nonlinear threshold component in the chaotic signal that the blood oxygen level depends on the whole brain to obtain a mapping signal, and then dynamically fusing the mapping signal with the standardized chaotic signal.

[0088] It should be noted that by dynamically fusing blood oxygen level-dependent signals and blood oxygen level-dependent chaotic signals, signal features from different dimensions can be integrated more accurately. This allows the fused signal matrix to retain the physiological information of the original signals and the dynamic characteristics of the chaotic signals, providing a more comprehensive and richer information foundation for subsequent analysis. The dynamic fusion approach can reflect the interaction and dynamic changes between blood oxygen level-dependent signals and chaotic signals in real time, capturing the dynamic characteristics of brain activity under different physiological states.

[0089] S150: Obtain sample sequences of the fused signal matrix based on a preset time window, and construct an adjacency matrix based on the sample sequences.

[0090] Specifically, the size of the preset time window is equal to the size of the blood oxygen level-dependent signal, and the adjacency matrix expression is:

[0091] A = corrcoef(Q)

[0092] Where A is the adjacency matrix and Q is the sample sequence.

[0093] like Figure 2 As shown, in the implementation of this invention, step S150 is equivalent to step F, where each ROI is processed sequentially within the time window to obtain a sample sequence of the fused signal matrix and to construct an adjacency matrix based on the sample sequence.

[0094] It should be noted that the adjacency matrix is ​​a square matrix, that is, the number of rows equals the number of columns, and the element A(i,j) in the matrix represents the correlation between the i-th sample and the j-th sample in the sample sequence.

[0095] By acquiring sample sequences of the fused signal matrix based on a preset time window, it is possible to capture the dynamic changes of the fused signal within a specific time range, which helps to analyze the time-varying characteristics and short-term behavior of the signal.

[0096] Based on the sample sequence, the adjacency matrix is constructed, the similarity or correlation between signals in different time windows can be quantified, and the adjacency matrix of the signal is constructed, which provides a basis for studying the complex interaction of the signal.

[0097] Since the size of the preset time window is equal to the size of the blood oxygen level dependent signal, it can ensure that the key information of the original signal is not lost during the analysis process, and at the same time avoid signal distortion or information omission caused by too large or too small time window.

[0098] S160, input the data corresponding to the adjacency matrix into the neural network model, and output the calculation result.

[0099] As shown in Figure 2 In the implementation process of the present application, step S160 is equivalent to step G to generate features F, which is used to judge whether the input rs-fMRI image is an abnormal image based on the adjacency matrix through the neural network.

[0100] It should be noted that the neural network model in the above steps includes but is not limited to a graph convolutional neural network model, a graph attention network model, and a graph sampling and aggregation model.

[0101] Preferably, the data corresponding to the adjacency matrix is input into the graph convolutional neural network model, and the calculation result is output.

[0102] It should be noted that the graph convolutional neural network model includes local feature aggregation, global feature enhancement, and multi-scale feature extraction steps. Through the step-by-step processing of local feature aggregation, global feature enhancement, and multi-scale feature extraction, the blood oxygen level dependent signal features can be refined and enhanced layer by layer, making the final multi-scale features more rich and representative.

[0103] On the basis of local feature aggregation, global feature enhancement is performed, which not only retains the local detail information of the blood oxygen level dependent signal, but also integrates the global context information, making the feature more complete and accurate, which is helpful to fully characterize the complex characteristics of the signal.

[0104] Through multi-scale feature extraction, the features of the signal can be fully captured from different scales, avoiding information loss caused by single-scale analysis, improving the recognition ability of the complex pattern of the blood oxygen level dependent signal, and enhancing the reliability of the calculation result.

[0105] Through the deep processing of the neural network on the adjacency matrix data, the hidden complex relationships and deep features in the blood oxygen level dependent signal can be mined, the depth and accuracy of the blood oxygen level dependent signal analysis can be improved, and more valuable information can be provided for subsequent research and application.

[0106] By inputting adjacency matrix data into a neural network model for automated feature extraction and analysis, manual intervention can be reduced, processing efficiency can be improved, and abnormal rs-fMRI image data can be quickly found from a large amount of rs-fMRI image data.

[0107] S170, based on the calculation results, specific regions of interest pairs are obtained.

[0108] like Figure 1 As shown, in the implementation of this invention, step S170 is equivalent to step H. If the disease is present, a specific ROI is analyzed to find the difference ROI pair between the abnormal rs-fMRI image data and the control group image.

[0109] It should be noted that by comparing the adjacency matrix of abnormal resting-state magnetic resonance imaging data with that of the control group, specific regions of interest pairs in the abnormal data can be accurately identified.

[0110] Below, on Figure 3 Steps S120-S140 and S160-S170 are further explained in detail.

[0111] refer to Figure 3 , Figure 4 A flowchart of step S120 of an embodiment of the present invention is shown. Step S120 includes:

[0112] S121, Data correction is performed based on resting-state magnetic resonance imaging data to obtain corrected data.

[0113] It should be noted that the calibration data includes image data and header information. The image data is a three-dimensional or four-dimensional array used to represent the signal intensity of each voxel. The header information includes the affine transformation matrix, voxel size, slice order, and spatial orientation. The file format is as follows:

[0114] The commonly used data correction methods include rigid body transformation, similarity transformation, nonlinear deformation, and affine transformation.

[0115] Preferably, the resting-state magnetic resonance imaging data is corrected by affine transformation to obtain corrected data.

[0116] It should be noted that the advantages of choosing affine transformation for data correction are:

[0117] (1) Affine transformation (12 degrees of freedom) can model translation, rotation, scaling and shearing simultaneously. Therefore, compared with 6-degree-of-freedom rigid body transformation or 7-degree-of-freedom similarity transformation, it has greater flexibility and can more accurately align brain structures with proportional differences or slight deformations between different subjects.

[0118] (2) Compared with the nonlinear transformation (such as the high-dimensional deformation model provided by FNIRT), the affine transformation is more efficient in calculation, has moderate parameters, and reduces the risk of overfitting.

[0119] (3) The affine transformation can ensure the continuity of the local structure during spatial changes and will not introduce large non-physiological distortion, so it is more robust and reliable for subsequent ROI time series extraction, brain network analysis and other macro-region-based processing tasks.

[0120] S122, fill the correction data to the automated anatomical labeling template to obtain filled data.

[0121] It should be noted that the automated anatomical labeling (Automated Anatomical Labeling, AAL for short) is a segmentation tool based on anatomical standards, which is used to divide the brain into multiple predefined regions of interest (ROI). The core purpose is to standardize the division of brain regions, facilitate cross-study functional connectivity analysis and data comparability. Subsequent versions include Automated Anatomical Labeling 2.0 template (AAL2 for short) and Automated Anatomical Labeling 3.0 template (AAL3 for short).

[0122] Preferably, the correction data is aligned with AAL3 by the FLIRT tool to obtain aligned data, and the aligned data is resampled to AAL3 by linear interpolation method to obtain filled data.

[0123] It should be noted that the advantage of selecting AAL3 lies in that the high-precision ROI division of AAL3 reduces signal aliasing, ensures the purity of BOLD signal in the target region with the highest functional connectivity, and provides reliable input for dynamic modulation of the mixed model.

[0124] S123, extracting the filled data based on the automated anatomical labeling template to obtain blood oxygen level dependent signals of each region of interest.

[0125] Preferably, the filled data is extracted based on AAL3 by the NiftiLabelsMasker tool to obtain blood oxygen level dependent signals of each region of interest.

[0126] It should be noted that the NiftiLabelsMasker tool in the Nilearn library is used to extract the signal of the functional image based on the AAL3 brain region division. Compared with other common signal extraction methods (such as NiftiMasker, NiftiMapsMasker or directly using ROI average), NiftiLabelsMasker is particularly suitable for region division based on discrete label maps (such as AAL3), and can efficiently average or aggregate the voxel signals in each brain region, ensuring that the extraction result strictly corresponds to the anatomical structure. At the same time, the tool has built-in preprocessing functions such as standardization (such as z-score), detrending, smoothing, etc., which can facilitate the simultaneous completion of data normalization processing during extraction, and improve the accuracy and consistency of downstream analysis.

[0127] Reference Figure 4 , Figure 5 A flow chart of step S130 of the embodiment of the application is shown, and step S130 comprises:

[0128] S131, selecting the blood oxygen level dependent signal of the target region with the highest functional connectivity in the region of interest as the data input of the initial chaotic model.

[0129] It should be noted that the target region with the highest functional connectivity in the region of interest includes: precentral gyrus, insular cortex, dorsolateral prefrontal cortex, etc., and the chaotic model includes: Chaotic model, Lorenz model, Duffing model, etc.

[0130] Preferably, the blood oxygen level dependent signal of the precentral gyrus is selected as the data input of the initial Chaotic model, wherein the initial Chaotic model is represented by the following differential equation:

[0131]

[0132] Wherein, x represents the fast oscillation component of the blood oxygen level dependent chaotic signal, y represents the slow phase component of the blood oxygen level dependent chaotic signal, z represents the nonlinear threshold component of the blood oxygen level dependent chaotic signal, a represents the nonlinear control parameter, b represents the dissipation parameter, c represents the periodic control parameter, and dt represents the time step.

[0133] It should be noted that the advantages of selecting the blood oxygen level dependent signal of the precentral gyrus as the input are:

[0134] (1) The precentral gyrus ranks in the top 5% of the whole brain in the resting state functional network node degree, indicating that the precentral gyrus has extensive connections with multiple functional blocks, and is suitable for global dynamic characteristics input of the chaotic model.

[0135] (2) The betweenness centrality value of the precentral gyrus ranks first in the sensorimotor network, indicating that the precentral gyrus plays a key relay role in cross-network transmission and can reflect the dynamic integration characteristics of the whole brain.

[0136] (3) The entropy range of the BOLD signal sample in the precentral gyrus is 0.25-0.35, which is within the optimal input range of the chaotic model (0.2-0.5), thus avoiding both excessive randomness and excessive regularization.

[0137] (4) The maximum Lyapunov exponent (λ_max = 0.05 ± 0.01) of the BOLD signal in the precentral gyrus is significantly positive (P < 0.01), indicating that it has deterministic chaotic characteristics, while the chaotic characteristics of the peripheral system (such as the amygdala λ_max = 0.02 ± 0.01) are weaker.

[0138] choose The advantages of chaos models are:

[0139] (1) The parameters of the chaotic model can be dynamically adjusted to adapt to the dynamic differences in BOLD signals of different individuals.

[0140] (2) Chaotic model topologies are sensitive to initial conditions but possess structural stability, making them suitable for long-term signal analysis.

[0141] (3) Embed the BOLD signal The phase space of the chaotic model can preserve the key dynamic features of the BOLD signal and avoid information loss.

[0142] (4) Through The chaotic model performs a linear transformation on the BOLD signal, which can amplify the pathological specific characteristics.

[0143] S132, the periodic control parameters of the initial chaotic model are dynamically modulated based on the blood oxygen level dependent signal of the target region to obtain the chaotic model.

[0144] Specifically, based on the blood oxygenation level-dependent signal in the precentral gyrus, the initial... The periodic control parameters of the chaotic model are dynamically modulated. The expression for these periodic control parameters is as follows:

[0145] c = C base +k*S BOLD

[0146] Among them, c base Let S be the center value of the chaotic interval, k be the scaling factor, and S be the center value of the interval. BOLD This is a blood oxygenation level-dependent signal in the precentral gyrus region.

[0147] S133, solve the differential equation of the chaotic model to obtain the blood oxygen level-dependent chaotic signal.

[0148] Specifically, the differential equation based on the chaotic model is discretized using Euler to obtain a blood oxygen level-dependent chaotic signal. The expression for this blood oxygen level-dependent chaotic signal is as follows:

[0149]

[0150] Where n is a variable parameter.

[0151] In the above steps, The nonlinear control parameter a of the chaotic model is set to 0.2, the dissipation parameter b is set to 0.4, the periodic control parameter c is set to 1.0, and the time step dt is set to 0.07. The initial state x0 of the fast oscillation component of the blood oxygen level dependent on the chaotic signal is set to 0.1, the initial state y0 of the slow phase component of the blood oxygen level dependent on the chaotic signal is set to 0.1, and the initial state z0 of the nonlinear threshold component of the blood oxygen level dependent on the chaotic signal is set to 0.1.

[0152] It should be noted that, through experiments, setting the initial states of the fast oscillation component, slow phase component, and nonlinear threshold component to 0.1 allows the nonlinear threshold component to accurately reflect changes in the blood oxygen level dependent signal. However, the chaotic model is sensitive to the initial states of these components. Setting them to 0.2 results in significant signal fluctuations. Setting them to 0.01 produces little change. After adjustments, 0.1 was determined to be the most suitable setting.

[0153] Setting the nonlinear control parameter 'a' to 0.2 maintains the system in a chaotic state while ensuring numerical stability. Setting the dissipation parameter 'b' to 0.4 ensures energy dissipation, which helps form a stable chaotic attractor and prevents divergence. Setting the periodic control parameter 'c' to 1.0 ensures the system exhibits complex dynamic behavior and can maintain chaotic oscillations for extended periods. A lower 'c' may lead to periodic oscillations, while a higher 'c' can cause excessive divergence. Setting the time step 'dt' to 0.07 strikes a balance between numerical stability and computational efficiency. An excessively large step may lead to numerical instability or loss of chaotic characteristics, while an excessively small step increases computational overhead. This value has been experimentally verified to effectively capture [the chaotic state]. The complex dynamic behavior of the system.

[0154] refer to Figure 5 , Figure 6A flowchart of step S140 of an embodiment of the present invention is shown. Step S140 includes:

[0155] S141 standardizes blood oxygen levels based on chaotic signals.

[0156] It should be noted that standardization of blood oxygen level-dependent signals is used to eliminate dimensional differences, enhance data comparability, and optimize model convergence speed. Standardization methods include, but are not limited to, Min-Max standardization, decimal scaling standardization, Robus standardization, vector normalization, and zero-mean unit variance standardization.

[0157] Preferably, the blood oxygen level-dependent chaotic signal is standardized using a zero-mean unit variance standardization algorithm.

[0158] It should be noted that the advantages of zero-mean unit variance standardization are:

[0159] (1) Zero mean variance standardization only shifts and scales the data, without changing the distribution pattern of blood oxygen level dependent chaotic signal, and can preserve the nonlinear dynamic characteristics of blood oxygen level dependent chaotic signal.

[0160] (2) Zero mean variance standardization is sensitive to extreme outliers and can help detect residual noise.

[0161] In the above steps, the standardized blood oxygen level, dependent on the chaotic signal, is represented as follows:

[0162]

[0163] Among them, s' ij For standardized blood oxygen levels to depend on chaotic signals, s ij For blood oxygenation levels to depend on chaotic signals, μ j σ is the mean of the nonlinear threshold components. j denoted as the standard deviation of the nonlinear threshold component, ∈ = 10⁻⁸, used to prevent the denominator from being 0.

[0164] S142, select one component of the chaotic signal that the blood oxygen level depends on.

[0165] It should be noted that blood oxygen levels depend on chaotic signals, including fast oscillation components, slow phase vectors, and nonlinear threshold components.

[0166] Preferably, a nonlinear threshold component of the chaotic signal that depends on blood oxygen levels is selected.

[0167] It should be noted that the current reference is... Figure 7, as shown in the visualization of experimental results, where x is the fast oscillation component, y is the slow phase component, and z is the nonlinear threshold component. The nonlinear threshold component of the blood oxygen level-dependent chaotic signal is selected for mapping to all regions of interest because it changes more smoothly and contains slow drift and sudden jumps, which are more similar to the blood oxygen level-dependent signal.

[0168] The advantage of selecting the nonlinear threshold component is that:

[0169] (1) The nonlinear threshold component can capture the nonlinear changes in the BOLD signal, which are closely related to the complex physiological mechanisms of brain activity.

[0170] (2) By fusing with the BOLD signal, we can better understand the complex dynamic behavior of the brain under different physiological conditions.

[0171] (3) The nonlinear threshold component can help remove noise in the signal and enhance the stability of the BOLD signal.

[0172] (4) By fusing the nonlinear threshold component, we can improve the overall quality of the BOLD signal, making it more reliable in analysis.

[0173] S143, map the nonlinear threshold component to all regions of interest to obtain a mapping signal.

[0174] It should be noted that the mapping signal is a nonlinear threshold feature data containing each ROI, usually a three-dimensional or four-dimensional array, and each element in the mapping signal represents the nonlinear threshold feature value of the corresponding region of interest.

[0175] In the above steps, the mapping signal expression is:

[0176] N roi =tile(N z ,(1,M))

[0177] where N roi is the mapping signal, N z is the nonlinear threshold component, and M is the number of regions of interest.

[0178] S144, based on the mapping signal, adjust the signal intensity through a scaling model to obtain an adjusted signal.

[0179] It should be noted that the adjusted signal is data with the same structure as the mapping signal, but the adjusted signal intensity has been scaled according to certain rules.

[0180] The adjusted signal expression is:

[0181] N'=λ*N roi

[0182] Wherein, N' is the adjustment signal, and λ is the intensity control value.

[0183] S145, the optimal weight is obtained, the adjustment signal and the blood oxygen level dependent signal are dynamically fused based on the optimal weight, and a fusion signal matrix is obtained.

[0184] It should be noted that the expression of the fusion signal matrix is:

[0185] fusion=ω * ·N′+(1-ω * )·real_signal

[0186] Wherein, fusion is the fusion signal matrix, ω* is the optimal weight, and real_signal is the blood oxygen level dependent signal.

[0187] It should be noted that the optimal weight needs to be calculated by a weight optimization objective function, and the expression of the weight optimization objective function is:

[0188]

[0189] Wherein, ω is the test weight, and temp_fusion is the temporary fusion signal.

[0190] Reference Figure 7 , Figure 8 The flow chart of step S160 of the embodiment of the application is shown, and step S160 comprises:

[0191] S161, local features are obtained by performing local feature aggregation based on the data corresponding to the adjacency matrix. It should be noted that the expression of the local feature is:

[0192] Z (1) =ReLU(A·(Q·W1+b1))

[0193] Wherein, Z (1) is the local feature, W1∈R T*H is the learning weight, b1 is the bias term, T is the length of the blood oxygen level dependent signal, and H is the hidden layer dimension.

[0194] S162, enhanced features are obtained by performing global feature enhancement based on the local features.

[0195] It should be noted that the expression of the enhanced features is:

[0196] Z (2) =LayerNorm((ReLU(W2Z (1) +b2)))+Z (1)

[0197] wherein, Z (2) is an enhanced feature, W2∈R H*H is a global enhancement weight, and b2 is a bias term.

[0198] S163, performing feature extraction based on the enhanced feature to obtain a multi-scale feature.

[0199] It should be noted that the multi-scale feature expression is:

[0200] F=MaxPool(Z (2) )

[0201] wherein, F is the multi-scale feature.

[0202] S164, flattening the multi-scale feature to obtain a flattened feature.

[0203] It should be noted that the flattened feature expression is:

[0204] F flat =Flatten(F)

[0205] wherein, F flat is the flattened feature.

[0206] S165, calculating a calculation result based on the flattened feature through a fully connected layer.

[0207] It should be noted that the calculation result expression is:

[0208] Y=softmax(W f ·F flat +b f )

[0209] wherein, is a learnable weight, b f ∈R C is a bias term, C is the number of calculation result categories, and N is the number of samples in the sample sequence.

[0210] Referring to Figure 8 , Figure 9 a flowchart of step S170 of the embodiment of the present application is shown, and step S170 includes:

[0211] S171, outputting an adjacency matrix of abnormal resting state magnetic resonance imaging data based on the calculation result.

[0212] Specifically, a comparison threshold is preset, and it is determined whether the calculation result is greater than the comparison threshold. If yes, it is determined that the resting state magnetic resonance imaging data is abnormal resting state magnetic resonance imaging data, and the adjacency matrix of the abnormal resting state magnetic resonance imaging data is output. If no, no processing is performed.

[0213] S172, comparing the adjacency matrix of the abnormal resting state magnetic resonance imaging data with the adjacency matrix of the control group to obtain a comparison result.

[0214] It should be noted that the comparison result expression is:

[0215] FC Difference = |FC condition1 -FC condition2 |

[0216] Wherein, FC Difference is the comparison result, FC condition1 indicates the condition1th adjacency matrix of the abnormal resting state magnetic resonance imaging data, FC condition2 indicates the condition2th adjacency matrix of the control group, condition1 is a variable parameter, and condition2 is a variable parameter.

[0217] S173, a second comparison threshold is preset, and it is judged whether the comparison result is greater than the second comparison threshold, if yes, the adjacency matrix of the abnormal resting state magnetic resonance imaging data is outputted with the adjacency matrix of the control group to obtain a specific region of interest pair; otherwise, no processing is performed.

[0218] The application provides a kind of graph convolutional neural network device based on chaotic system enhancement rs-fMRI feature analysis specific ROI, comprising:

[0219] The acquisition module is used to acquire resting state magnetic resonance imaging data.

[0220] The extraction module is used to extract blood oxygen level dependent signals of each region of interest in the resting state magnetic resonance imaging data.

[0221] The chaotic processing module is used to perform chaotic processing on the blood oxygen level dependent signals based on a chaotic model to obtain blood oxygen level dependent chaotic signals.

[0222] The fusion module is used to dynamically fuse the blood oxygen level dependent signals and the blood oxygen level dependent chaotic signals to obtain a fusion signal matrix.

[0223] The construction module is used to obtain a sample sequence of the fusion signal matrix based on a preset time window, and construct an adjacency matrix based on the sample sequence, wherein the adjacency matrix is used to represent the time transformation relationship of the blood oxygen level dependent signals in the region of interest.

[0224] The classification module is used to input the data corresponding to the adjacency matrix into a neural network model to output a calculation result.

[0225] The analysis module is used to obtain a specific region of interest pair based on the calculation result.

[0226] The modules in the device of the present application perform the method of the above-mentioned embodiments, and the specific functions can refer to the description of the above-mentioned embodiments, which will not be repeated here.

[0227] The present application provides an electronic device, comprising: a processor; and a memory, wherein computer program instructions are stored in the memory.

[0228] The computer program instructions are run by the processor, so that the processor executes the graph convolution method for enhancing rs-fMRI feature analysis specific ROI based on a chaotic system.

[0229] The present application provides a computer readable storage medium, which stores a computer program, and the computer program is run by the processor, so that the processor executes the graph convolution method for enhancing rs-fMRI feature analysis specific ROI based on a chaotic system.

[0230] The present application discloses a device, comprising:

[0231] The memory is used for storing instructions executed by one or more processors of the device, and the processor is one of the processors of the device, and is used for executing the method disclosed in any one of the first to fourth aspects.

[0232] Now referring to Figure 8 , a block diagram of a device 1200 according to one embodiment of the present application is shown. The device 1200 can include one or more processors 1201 coupled to a controller hub 1203. For at least one embodiment, the controller hub 1203 communicates with the processor(s) 1201 via a multi-drop bus, such as a Front Side Bus (FSB), point-to-point interface, such as a Quick Path Interconnect (QPI), or similar communication link 1206. The processor(s) 1201 execute instructions to perform a general type of processing operation. In an embodiment, the controller hub 1203 includes, without limitation, a Graphics Memory Controller Hub (GMCH) (not shown), which includes memory and graphics controllers, and an Input / Output Hub (IOH) (which can be on a separate chip) (not shown), that couples with the GMCH and with the processor(s) 1201.

[0233] The device 1200 can also include a coprocessor 1202 coupled to the controller hub 1203. Alternatively, one or both of the memory and the GMCH can be integrated into the processor (as described in the present disclosure), the memory 1204 and the coprocessor 1202 are directly coupled to the processor 1201 and the controller hub 1203 is integrated into the IOH. The memory 1204 can be, for example, a dynamic random access memory (DRAM) such as a synchronous dynamic random access memory (SDRAM), a PCMA, or a combination of such memories. In one embodiment, the coprocessor 1202 is a special-purpose processor, such as, for example, a high-throughput MIC processor (MIC), a network or communication processor, compression engine, graphics processor, GPGPU, embedded processor, etc. The coprocessor 1202 can be a shared coprocessor, e.g., a coprocessor 1202 shared by the processor 1201 and the controller hub 1203. In one embodiment, the coprocessor 1202 is a separate processor that is part of the device 1200. In various embodiments, the coprocessor 1202 is integrated on the same die with the processor 1201 and / or controller hub 1203. In various embodiments, the memory 1204 is on-board the coprocessor 1202. In various embodiments, the memory 1204 is off-board of the coprocessor 1202. The optional nature of the coprocessor 1202 is denoted in parentheses in FIG. 12. Figure 9

[0234] The memory 1204, as a computer-readable storage medium, can include one or more tangible, non-transitory, computer-readable media used to store data and / or instructions for use by or in connection with the computer system 1200. In the illustrated example, the memory 1204 includes a volatile memory 1205 and a non-volatile memory 1207, but can include other types of memories as well, such as one or more of a suitably configured flash memory, one or more Hard-Disk Drives (HDDs), one or more Compact Disc (CD) drives, and / or one or more Digital Versatile Disc (DVD) drives.

[0235] In one embodiment, the device 1200 can further include a network interface controller (NIC) 1206. The network interface 1206 can include a transceiver to provide a radio interface to the device 1200 to enable communications with any other suitable device, such as a front end module, an antenna, etc. In various embodiments, the network interface 1206 can be integrated with other components of the device 1200. The network interface 1206 can implement the functionality of the communication unit in the above-described embodiments.

[0236] ​The device 1200 can further include an input / output (I / O) device 1205. The I / O 1205 can include a user interface designed to enable a user to interact with the device 1200, a peripheral component interface designed to enable peripheral components to interact with the device 1200, and / or a sensor designed to determine environmental conditions and / or location information related to the device 1200.

[0237] Notably, Figure 9 are merely exemplary. That is, although Figure 9 In the embodiment shown in FIG. 12, the device 1200 includes a processor 1201, a controller hub 1203, a memory 1204, and the like, but in actual applications, a device using the methods of the present application can include only a part of the devices of the device 1200, for example, can include only the processor 1201 and the NIC 1206. Figure 10 The nature of optional devices in FIG. 12 is shown in dashed lines. According to some embodiments of the present application, the memory 1204 as a computer readable storage medium stores instructions which, when executed on a computer, cause the system 1200 to perform the graph convolution method for enhancing specificity of rs-fMRI feature analysis of ROI based on a chaotic system according to the above embodiments, which can be specifically referred to the method of the above embodiments, and will not be repeated here.

[0238] Reference is now made to Figure 10 FIG. 13, which shows a block diagram of a SoC (System on Chip) 1300 according to an embodiment of the present application. In Figure 10 Similar elements in FIG. 13 bear like reference numerals. In addition, dashed lined boxes are optional features of more advanced SoCs. In ​ In FIG. 13, the SoC 1300 includes an interconnect unit 1350 coupled to an application processor 1310; a system agent unit 1380; a bus controller unit 1390; an integrated memory controller unit 1340; a set or one or more coprocessors 1320, which can include integrated graphics logic, an image processor, an audio processor, and a video processor; a static random access memory (SRAM) unit 1330; and a direct memory access (DMA) unit 1360. In one embodiment, the coprocessors 1320 include a special-purpose processor, such as for example a network or communication processor, compression engine, GPGPU, a high-throughput MIC processor, or embedded processor, among others.

[0239] The static random access memory (SRAM) unit 1330 can include one or more computer-readable media for storing data and / or instructions. The computer-readable storage media can store instructions, in particular, a transient or a permanent copy of the instructions. The instructions can include causing the Soc 1300 to perform the graph convolution method for enhancing specificity of ROIs in rs-fMRI feature analysis based on chaotic systems according to the above-described embodiments, in particular, the method of the above-described embodiments, which will not be described herein again.

[0240] Embodiments of the mechanisms disclosed herein can be implemented in hardware, software, firmware, or any combination thereof. Embodiments of the application can be implemented as computer programs or program code executing on programmable systems comprising at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device.

[0241] Program code can be applied to input instructions to perform the functions described herein and generate output information. The output information can be applied to one or more output devices, in known fashion. For purposes of this application, a processing system includes any system that has a processor, such as for example a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC), or a microprocessor.

[0242] The program code can be implemented in a high level procedural or object oriented programming language to be executed by a processing system. As necessary, the program code can be implemented in assembly or machine language, if desired. In fact, the mechanisms described herein are not limited in scope to any particular programming language. In any case, the language can be a compiled or interpreted language.

[0243] In some cases, the disclosed embodiments can be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments can also be implemented as instructions carried by or stored on one or more transitory or non-transitory machine- readable (e.g., computer-readable) media, which can be read and executed by one or more processors. For example, the instructions can be distributed over the network or by other computer readable media. Thus, a machine-readable medium can include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computer), including without limitation, a floppy disk, an optical disc (e.g., a CD or DVD), a magnetic disk or tape, a read only memory (ROM), a random access memory (RAM), an erasable programmable read only memory (EPROM), an electrically erasable programmable read only memory (EEPROM), a magnetic or optical card, a flash memory, or a tangible, machine-readable storage medium. Accordingly, a machine-readable medium includes any medium that is suitable for storing or transmitting instructions for controlling a machine (e.g., a computer) to perform any of the operations disclosed herein.

[0244] In the drawings, some of the structural or methodological features can be shown in particular arrangements and / or orders. However, it should be understood that such specific arrangements and / or orders can not be required. Instead, in some embodiments, the features can be arranged in a different manner and / or order than shown in the figures of the specification. Additionally, inclusion of a structural or methodological feature in a particular figure is not meant to imply that such feature is required in all embodiments, and in some embodiments, the features can not be included or can be combined with other features.

[0245] It should be noted that each unit / module mentioned in each device embodiment of the present application is a logical unit / module, and in the physical world, one logical unit / module can be a physical unit / module, or a part of a physical unit / module, or be realized in a combination of multiple physical unit / modules, and the physical realization of these logical units / modules is not the most important, and the combination of the functions implemented by these logical units / modules is the key to solving the technical problems proposed by the present application. In addition, in order to highlight the innovative part of the present application, the above-mentioned device embodiments of the present application do not introduce the units / modules that are not closely related to solving the technical problems proposed by the present application, which does not mean that the above-mentioned device embodiments do not have other units / modules.

[0246] It should be noted that in the examples and descriptions of the present patent, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including one" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0247] Although the present application has been illustrated and described with reference to certain preferred embodiments thereof, it should be understood that various changes in form and detail can be made therein without departing from the spirit and scope of the application.

Claims

1. A graph convolution method for enhancing rs-fMRI feature analysis specific ROI, characterized in that, The method comprises the following steps: acquiring resting state magnetic resonance imaging data; extracting blood oxygen level dependent signals of each region of interest in the resting state magnetic resonance imaging data; chaotic processing the blood oxygen level dependent signals based on a chaotic model to obtain blood oxygen level dependent chaotic signals, comprising: selecting the blood oxygen level dependent signal of a target region with the highest functional connectivity in the region of interest as the data input of an initial chaotic model; dynamically modulating the period control parameter of the initial chaotic model based on the blood oxygen level dependent signal of the target region to obtain the chaotic model; solving the differential equation of the chaotic model to obtain the blood oxygen level dependent chaotic signal; dynamically fusing the blood oxygen level dependent signal and the blood oxygen level dependent chaotic signal to obtain a fusion signal matrix; acquiring a sample sequence of the fusion signal matrix based on a preset time window, and constructing an adjacency matrix based on the sample sequence, wherein the adjacency matrix is used to represent the time transformation relationship of the blood oxygen level dependent signals in the region of interest; inputting the data corresponding to the adjacency matrix into a neural network model to output a calculation result; obtaining a specific region of interest pair based on the calculation result.

2. The method of claim 1, wherein, The method comprises the following steps: based on the resting state magnetic resonance imaging data, data correction is performed to obtain corrected data; the corrected data is filled into an automatic anatomical marker template to obtain filled data; based on the automatic anatomical marker template, the filled data is extracted to obtain the blood oxygen level dependent signals of each region of interest.

3. The method of claim 1, wherein, The method comprises the following steps: standardizing the blood oxygen level dependent chaotic signal; selecting a component in the blood oxygen level dependent chaotic signal; dynamically fusing the component and the blood oxygen level dependent signal to obtain the fusion signal matrix.

4. The method of claim 1, wherein, The method comprises the following steps: based on the data corresponding to the adjacency matrix, local features are aggregated to obtain local features; based on the local features, global feature enhancement is performed to obtain enhanced features; based on the enhanced features, feature extraction is performed to obtain multi-scale features; based on the multi-scale features, the calculation result is obtained.

5. The method of claim 1, wherein, The method comprises the following steps: based on the calculation result, an adjacency matrix of abnormal resting state magnetic resonance imaging data is output; the adjacency matrix of the abnormal resting state magnetic resonance imaging data is compared with the adjacency matrix of a control group to obtain the specific region of interest pair.

6. The method of claim 1, wherein, After extracting the blood oxygen level dependent signals of each region of interest in the resting state magnetic resonance imaging data, the method further comprises the following steps: standardizing the blood oxygen level dependent signals of each region of interest.

7. A graph convolutional apparatus for enhancing rs-fMRI feature analysis specific ROI, characterized in that, The method comprises the following steps: an acquisition module is configured to acquire resting state magnetic resonance imaging data; An extraction module is configured to extract blood oxygen level dependent signals of each region of interest in the resting state magnetic resonance imaging data; A chaos processing module is configured to perform chaos processing on the blood oxygen level dependent signals based on a chaos model to obtain blood oxygen level dependent chaos signals, including: selecting a blood oxygen level dependent signal of a target region with the highest functional connectivity in the region of interest as data input of an initial chaos model; dynamically modulating a period control parameter of the initial chaos model based on the blood oxygen level dependent signal of the target region to obtain the chaos model; solving a differential equation of the chaos model to obtain the blood oxygen level dependent chaos signals; a fusion module is configured to dynamically fuse the blood oxygen level dependent signals and the blood oxygen level dependent chaos signals to obtain a fusion signal matrix; a construction module is configured to obtain a sample sequence of the fusion signal matrix based on a preset time window, and construct an adjacency matrix based on the sample sequence, the adjacency matrix being configured to represent a time transformation relationship of the blood oxygen level dependent signals in the region of interest; a classification module is configured to input data corresponding to the adjacency matrix into a neural network model to output a calculation result; an analysis module is configured to obtain specific regions of interest based on the calculation result.

8. An electronic device, comprising: including: a processor; and a memory, in which computer program instructions are stored, wherein the computer program instructions, when executed by the processor, cause the processor to perform the method of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program, when executed by a processor, causes the processor to perform the method of claims 1-6.

Images