Search

CN-122020040-A - Brain state analysis method and device based on fiber bundle guide map self-encoder

CN122020040ACN 122020040 ACN122020040 ACN 122020040ACN-122020040-A

Abstract

The invention provides a brain state analysis method and a brain state analysis device based on a fiber bundle guide map self-encoder, which take anatomical connection of white matter fiber bundles as prior constraint, limit an analysis range to brain areas connected by specific fiber bundles in a functional connection dynamic map construction stage, and utilize a map self-encoder model to characterize a propagation path of binary matrix constraint functional information of the anatomical connection in a neural network, thereby realizing deep fusion of structural functional information in nonlinear dimension reduction, and finally identifying fiber bundle related brain states with definite anatomical attribution and high specificity and dynamic indexes thereof. The invention breaks through the homogenization assumption of the traditional whole brain state analysis, dynamically and accurately anchors the macroscopic function on the micro-dissection path, and remarkably improves the dissection specificity and biological interpretability of the identified brain state.

Inventors

  • ZONG FANGRONG
  • Shi Qiji
  • ZHU ZAIMIN
  • ZHOU LEQING

Assignees

  • 北京邮电大学

Dates

Publication Date
20260512
Application Date
20260126

Claims (10)

  1. 1. A method for analyzing brain states based on a fiber bundle guide map self-encoder, the method comprising the steps of: preprocessing functional magnetic resonance imaging data of a brain of a subject to be analyzed, and performing spatial registration with brain structures of individuals to extract blood oxygen dependent level data time sequences of a plurality of interest areas, wherein the interest areas are gray brain areas connected by appointed white matter fiber bundles; Extracting target low-frequency oscillation components from the blood oxygen dependent level data time sequences of the interest areas through a band-pass filter, performing Hilbert transformation, and calculating the instantaneous phase of a signal at each time point to obtain an instantaneous phase time sequence of each interest area; Calculating instantaneous phase differences among all the interest areas at each time point, calculating corresponding cosine values as phase coherence values, and establishing connection relations among the interest areas corresponding to the phase coherence values with higher set proportion to form a brain connection diagram, wherein the brain connection diagram comprises a binary adjacent matrix representing the connection relations among the interest areas and a node characteristic matrix formed by the phase coherence values; based on a pre-trained graph self-encoder model, extracting a low-dimensional potential representation corresponding to each region of interest by taking the node characteristic matrix of each time point as input and the binary adjacency matrix as constraint; Clustering the low-dimensional potential representations of the interest areas at all time points, wherein each cluster represents a brain state, assigning a unique state label for each time point to obtain a discrete sequence of brain state switching along with time, and calculating the occurrence probability, average duration and state transition probability of each brain state.
  2. 2. The method of brain state analysis based on a fiber bundle guide map self-encoder according to claim 1, wherein before preprocessing functional magnetic resonance imaging data of a brain of a subject to be analyzed, the method further comprises: Superposing a plurality of target white matter fiber bundles of the human brain and a preset anatomical map, and determining a gray matter brain area connected with the target white matter fiber bundles as an interest area, wherein the target white matter fiber bundles comprise bilateral anterior thalamus radiation, lower frontal occipital bundles, lower longitudinal bundles, upper longitudinal bundles, temporal upper longitudinal bundles, hook bundles, cingulate hippocampus, cingulate gyrus and corticospinal bundles; Preprocessing functional magnetic resonance imaging data of a brain of a subject to be analyzed, including performing temporal layer correction, head motion correction, spatial normalization and smoothing, and filtering denoising on the functional magnetic resonance imaging data.
  3. 3. The method for analyzing brain states based on a fiber bundle guide map self-encoder according to claim 1, wherein the calculation formula of the phase coherence value is: ; Wherein, the Is shown at the time point Time interest zone With the region of interest The phase coherence value in between, Indicating a point in time Time interest zone Corresponding to the temporal phase of the time series of blood oxygen dependent level data, Indicating a point in time Time interest zone Corresponding to the temporal phase of the time series of blood oxygen dependent level data.
  4. 4. The brain state analysis method based on the fiber bundle guide map self-encoder according to claim 1, wherein the frequency range of the band-pass filter is 0.01-0.1 hz; establishing a connection relation between the interest areas corresponding to the phase wanted dry values with higher set proportion comprises the steps of setting the set proportion to 40-45% based on a proportion threshold method.
  5. 5. The method for analyzing brain states based on a fiber bundle guided graph self-encoder according to claim 1, wherein the training step of the graph self-encoder comprises: Acquiring a training sample set comprising a sample brain connection graph of a plurality of time points, wherein the sample brain connection graph comprises a sample node feature matrix formed by a sample binary adjacency matrix representing connection relations between the interest areas and the phase coherence values; acquiring an initial encoder and an initial decoder, wherein the encoder is obtained by stacking a multi-layer graph convolution network or a graph annotation network, and the initial decoder comprises a full-connection layer, an inner product operation layer and an activation function layer; The initial encoder takes the sample brain connection graph as input, transmits and aggregates information carried by the sample node feature matrix along the edges with connection relation of the sample binary adjacent matrix marks, suppresses the edges without connection relation, outputs sample low-dimensional potential representation, takes the sample low-dimensional potential representation and the sample binary adjacent matrix as input, and outputs a reconstruction matrix of the sample node feature matrix, and carries out parameter updating on the initial encoder and the initial decoder based on binary cross entropy loss of the reconstruction matrix and the sample node feature matrix to obtain the graph self-encoder and the graph self-decoder.
  6. 6. The method of claim 1, wherein clustering the low-dimensional potential representations of each of the regions of interest at all points in time, further comprises: clustering is performed by adopting a K-means algorithm, a Gaussian mixture model or a spectral clustering algorithm, and the optimal clustering number is determined by adopting an elbow rule or a contour coefficient.
  7. 7. The method of brain state analysis based on a fiber bundle guide map self-encoder according to claim 1, further comprising: and adopting a hidden Markov model to perform brain state clustering division and modeling state transition probability on the low-dimensional potential representation.
  8. 8. A fiber bundle guide map self-encoder based brain state analysis device comprising a processor, a memory and a computer program or instructions stored on the memory, characterized in that the processor is adapted to execute the computer program or instructions, which when executed, implement the steps of the method according to any one of claims 1 to 7.
  9. 9. A computer-readable storage medium, on which a computer program or instructions is stored, which, when executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.
  10. 10. A computer program product comprising a computer program or instructions which, when executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.

Description

Brain state analysis method and device based on fiber bundle guide map self-encoder Technical Field The invention relates to the technical field of neuroscience, in particular to a brain state analysis method and device based on a fiber bundle guide map self-encoder. Background The brain acts as a highly complex dynamic system, and its cognitive function is realized by the cooperation and recombination of large-scale neural clusters in the space-time dimension. With the development of neural imaging techniques such as functional magnetic resonance imaging, researchers can non-invasively observe the activity of the whole brain nerve, and further propose a concept of brain state for describing a dynamic framework of the evolution of the overall activity pattern of the nervous system over time. The research on brain states and dynamic transition rules thereof is a key for understanding the mechanisms of cognitive functional nerve basis and nervous system diseases. At present, a plurality of technical schemes for analyzing brain states based on functional magnetic resonance imaging data have been developed in the field of computational neuroscience. However, the brain state identified by the existing scheme is completely dependent on the transient statistical characteristics derived from the functional magnetic resonance imaging signals in the extraction process, and cannot be fused with the inherent anatomical structure information of the brain, in particular the white matter fiber bundle connection network which is the physical basis of nerve information transmission. The assumption of the homogeneity of the whole brain is implied in the actual operation, i.e. all possible functional connections of the whole brain are given equal weights in the state composition. This results in the identified state often being a global, ambiguous macroscopic pattern, possibly confounded with functional signals from different anatomic passageways, and thus lacking explicit anatomic attribution and specificity. This limitation weakens the potential of this solution in revealing the functional dynamics of specific nerve loops, locating abnormal loops associated with neurological diseases, and providing clinical biomarkers with clear biological interpretation. Therefore, a new technical solution is needed that can break through the above limitations in order to identify brain functional states with greater anatomical specificity, biological interpretability and clinical transformation potential by a priori knowledge of the anatomy of the mass fiber bundles when analyzing brain functional state dynamics. Disclosure of Invention In view of this, embodiments of the present invention provide a brain state analysis method and apparatus based on a fiber bundle guide map self-encoder, so as to eliminate or improve one or more drawbacks existing in the prior art, and overcome the problems of state characterization confounding, insufficient anatomical specificity and poor biological interpretability caused by lack of constraints of white matter fiber bundles in the existing method. One aspect of the present invention provides a brain state analysis method based on a fiber bundle guide map self-encoder, the method comprising the steps of: preprocessing functional magnetic resonance imaging data of a brain of a subject to be analyzed, and performing spatial registration with brain structures of individuals to extract blood oxygen dependent level data time sequences of a plurality of interest areas, wherein the interest areas are gray brain areas connected by appointed white matter fiber bundles; Extracting target low-frequency oscillation components from the blood oxygen dependent level data time sequences of the interest areas through a band-pass filter, performing Hilbert transformation, and calculating the instantaneous phase of a signal at each time point to obtain an instantaneous phase time sequence of each interest area; Calculating instantaneous phase differences among all the interest areas at each time point, calculating corresponding cosine values as phase coherence values, and establishing connection relations among the interest areas corresponding to the phase coherence values with higher set proportion to form a brain connection diagram, wherein the brain connection diagram comprises a binary adjacent matrix representing the connection relations among the interest areas and a node characteristic matrix formed by the phase coherence values; based on a pre-trained graph self-encoder model, extracting a low-dimensional potential representation corresponding to each region of interest by taking the node characteristic matrix of each time point as input and the binary adjacency matrix as constraint; Clustering the low-dimensional potential representations of the interest areas at all time points, wherein each cluster represents a brain state, assigning a unique state label for each time point to obtain a discrete sequence of brain