CN-117274163-B - Autism subtype detection method and system based on multi-hypergraph collaborative optimization

CN117274163BCN 117274163 BCN117274163 BCN 117274163BCN-117274163-B

Abstract

The invention discloses an autism subtype detection method and system based on multi-hypergraph collaborative optimization, wherein the method can acquire image data comprising a first mode and a second mode, and extract a low-frequency fluctuation amplitude graph of the image data of the first mode and a brain tissue density graph of the image data of the second mode. And dividing brain regions according to the automatic labeling atlas, and extracting the average characteristic sequence of the low-frequency fluctuation amplitude and the brain tissue density of each brain region. And generating a first data matrix and a second data matrix through the average characteristic sequence, and constructing a first super network and a second super network according to the data matrices. Aiming at a first super network and a second super network, a community detection method oriented to multi-super graph collaborative optimization is provided to generate a clustering result as a detection result of the autism subtype. According to the method, the related super network is built based on the multi-mode image data, then the multi-mode super network is fused and the collaborative optimization is performed to detect the autism subtype, so that the accuracy of the autism subtype detection can be improved.

Inventors

HU BIN
ZHENG WEIHAO
LI JIALONG
YAO ZHIJUN
TANG WANYOU

Assignees

兰州大学

Dates

Publication Date: 20260512
Application Date: 20230828

Claims (8)

1. An autism subtype detection method based on multi-hypergraph collaborative optimization is characterized by comprising the following steps of: Acquiring image data, wherein the image data comprises first-mode image data and second-mode image data, the first-mode image data is functional magnetic resonance imaging, and the second-mode image data is structural magnetic resonance imaging; extracting a low-frequency fluctuation amplitude feature map of the first-mode image data and extracting a brain tissue density feature map of the second-mode image data; Dividing the brain region of the low-frequency fluctuation amplitude characteristic map and the brain tissue density characteristic map according to an automatic labeling map; extracting an average characteristic sequence of the brain region; Obtaining a data matrix through the average characteristic sequence, wherein the data matrix comprises a first data matrix and a second data matrix, the first data matrix is obtained based on the average characteristic sequence of the low-frequency fluctuation amplitude characteristic graph, and the second data matrix is obtained based on the average characteristic sequence of the density characteristic graph; Constructing a first super network and a second super network according to the data matrix, wherein sub-samples are acquired, the sub-samples are elements of the first data matrix and the second data matrix, the sub-samples are taken as centroids, a sparse matrix is solved according to the centroids, the column dimension of the sparse matrix is equal to the number of the sub-samples, the sparseness of the sparse matrix is represented based on regularization parameters, the sparse matrix is spliced and binarization processing is carried out to obtain a hypergraph incidence matrix, the columns of the hypergraph incidence matrix are marked as superedges, and the first super network and the second super network are obtained based on the superedges; Performing multi-hypergraph collaborative optimized community detection on the first and second hypernetworks to generate a cluster tag, wherein clustering is performed on a non-zero value matrix before binarization of the hypergraph incidence matrix based on a Lun multi-layer community detection algorithm to generate an initial cluster tag, optimizing a modularized target of the first hypernetwork according to the initial cluster tag to generate a first cluster result, optimizing the modularized target of the second hypernetwork according to the first cluster result to generate a second cluster result, optimizing the modularized target of the first hypernetwork through the second cluster result, and optimizing the modularized target of the second hypernetwork through the first cluster result, and stopping multi-hypergraph collaborative optimized community detection of the first and second hypernetworks when the modularized targets of the first and second hypernetworks converge; and outputting the clustering label when the community detection of the multi-hypergraph collaborative optimization is finished.
2. The multiple hypergraph co-optimization based autism subtype detection method of claim 1, further comprising: removing time points of target quantity in the first mode image data; performing temporal layer correction and head motion correction on the image data; registering the first modality image data to a standard space, and performing temporal filtering and spatial smoothing on the first modality image data, the temporal filtering including a de-linearity trend and a band pass filtering.
3. The multiple hypergraph co-optimization based autism subtype detection method of claim 1, further comprising: performing a pre-fusion correction on the second modality image data; performing de-characterization and segmentation processing on the second modality image data to generate a gray matter density map and a white matter density map of the second modality image data; registering the gray matter density map and the white matter density map to a standard space.
4. The multiple hypergraph co-optimization based autism subtype detection method of claim 3, wherein extracting the brain tissue density feature graph of the second modality image data comprises: Acquiring the gray matter density map and the white matter density map; and splicing the gray matter density map and the white matter density map.
5. The multiple hypergraph co-optimization based autism subtype detection method of claim 1, further comprising: detecting head motion of the first modality image data; Removing the first mode image data with the head movement greater than 1 millimeter; acquiring a preprocessing grade of the second mode image data; removing the second mode image data with the preprocessing level lower than a target threshold value; finally, the consistency of the samples in the first mode image data and the second mode image data is ensured.
6. The multiple hypergraph co-optimization based autism subtype detection method according to claim 1, wherein the regularization parameters are determined by means of ten-fold cross validation.
7. The multiple hypergraph collaborative optimization-based autism subtype detection method according to claim 1, wherein the optimizing the modular objective of the first or second hypernetwork comprises: moving nodes in the first or second super network into clusters of neighboring nodes; calculating the variation of the modularity, wherein the variation comprises variation values of a segmentation term and a volume term; detecting the gain of the modularity through the variation; if the modularity exists a gain, moving the node into a cluster of adjacent nodes, and calculating the variation of the modularity; if the modularity does not have gain, the modular goal is marked as converging.
8. An autism subtype detection system based on multi-hypergraph collaborative optimization, comprising: The acquisition module is configured to acquire image data, wherein the image data comprises first-mode image data and second-mode image data, the first-mode image data is functional magnetic resonance imaging, and the second-mode image data is structural magnetic resonance imaging; The processing module is configured to extract a low-frequency fluctuation amplitude characteristic map of the first mode image data and extract a brain tissue density characteristic map of the second mode image data, divide the low-frequency fluctuation amplitude characteristic map and a brain region of the brain tissue density characteristic map according to an automatic labeling map, extract an average characteristic sequence of the brain region; The multi-task hypergraph construction module is configured to obtain a data matrix through the average characteristic sequence, wherein the data matrix comprises a first data matrix and a second data matrix, the first data matrix is obtained based on the average characteristic sequence of the low-frequency fluctuation amplitude characteristic graph, the second data matrix is obtained based on the average characteristic sequence of the brain tissue density characteristic graph, a first hypernetwork and a second hypernetwork are constructed according to the data matrix, the subsamples are elements of the first data matrix and the second data matrix, the subsamples are taken as mass centers, a sparse matrix is solved according to the mass centers, the column dimension of the sparse matrix is equal to the number of the subsamples, the sparseness of the sparse matrix is represented based on regularization parameters, the sparse matrix is spliced and binarization processing is carried out to obtain a hypergraph incidence matrix, the columns of the hypergraph incidence matrix are marked as superedges, and the first hypernetwork and the second hypernetwork are obtained based on the superedges; The multi-hypergraph collaborative community detection module is configured to perform multi-hypergraph collaborative optimization community detection on the first and second hypernetworks to generate a clustering label, wherein clustering is performed on a non-zero value matrix before binarization of the hypergraph incidence matrix based on a Luventuri multi-layer community detection algorithm to generate an initial clustering label, the modularized target of the first hypernetwork is optimized according to the initial clustering label to generate a first clustering result, the modularized target of the second hypernetwork is optimized according to the first clustering result to generate a second clustering result, the modularized target of the first hypernetwork is optimized through the second clustering result, and the modularized target of the second hypernetwork is optimized through the first clustering result, and the multi-hypergraph collaborative optimized community detection of the first and second hypernetworks is stopped when the modularized targets of the first and second hypernetworks are converged; and the output module is configured to output the cluster label when the community detection of the multi-hypergraph collaborative optimization is finished.

Description

Autism subtype detection method and system based on multi-hypergraph collaborative optimization Technical Field The application relates to the technical field of artificial intelligence, in particular to an autism subtype detection method and system based on multi-hypergraph collaborative optimization. Background Functional magnetic resonance imaging (functional magnetic resonance imaging, fMRI) and structural magnetic resonance imaging (structural magnetic resonance imaging, sMRI) techniques are very widely used in the medical imaging field. Functional magnetic resonance imaging is the measurement of hemodynamic changes induced by neuronal activity by stimulating specific senses to cause neural activity (functional area activation) in the corresponding parts of the cerebral cortex. In the analysis angle of the functional image, a researcher establishes a sample similarity matrix through a random forest, 3 biological subtypes of autism are identified by adopting a spectral clustering algorithm, and the structural magnetic resonance imaging is a non-invasive medical imaging technology and is used for scanning a human body. From the analysis point of view of structural images, researchers classified autism into 3 biological subtypes by performing unsupervised clustering on the grey brain anatomy. The traditional graph can be used for describing the association relationship between samples, wherein the vertex represents the sample and the side represents the association relationship, but the traditional network constructed based on linear correlation can only capture the second-order relationship between every two samples, and cannot represent the cooperative change relationship between a plurality of samples. Because the hyperedges in the hypergraph can connect any number of samples, not just two samples, the higher order relationship between the samples can be well characterized. In the past, only the traditional unsupervised algorithm is adopted to cluster the data of a single mode, so that complex high-order association characteristics between samples cannot be modeled, and the changes of brain functions and structures caused by autism are ignored, so that the detection effect of biological subtypes is poor, and the subtypes cannot comprehensively reflect the heterogeneity of the pathological change mode of the autism brain. The hypergraph is used for describing the high-order similarity relationship between samples, and a hypernetwork constructed by the multi-mode brain image data is fused, so that more objective and reliable autism subtype can be detected. Disclosure of Invention The application provides an autism subtype detection method and system based on multi-hypergraph collaborative optimization, which are used for solving the problem of difficult analysis of biological heterogeneity of autism. In a first aspect, the invention discloses an autism subtype detection method based on multi-hypergraph collaborative optimization, which comprises the following steps: Acquiring image data, wherein the image data comprises first-mode image data and second-mode image data, the first-mode image data is functional magnetic resonance imaging, and the second-mode image data is structural magnetic resonance imaging; extracting a low-frequency fluctuation amplitude feature map of the first-mode image data and extracting a brain tissue density feature map of the second-mode image data; Dividing the brain region of the low-frequency fluctuation amplitude characteristic map and the brain tissue density characteristic map according to an automatic labeling map; extracting an average characteristic sequence of the brain region; Obtaining a data matrix through the average characteristic sequence, wherein the data matrix comprises a first data matrix and a second data matrix, the first data matrix is obtained based on the average characteristic sequence of the low-frequency fluctuation amplitude characteristic diagram, and the second data matrix is obtained based on the average characteristic sequence of the brain tissue density characteristic diagram; constructing a first super network and a second super network according to the data matrix; Performing community detection of multi-hypergraph collaborative optimization on the first and second hypernetworks to generate cluster labels; and outputting the clustering label when the community detection of the multi-hypergraph collaborative optimization is finished. Optionally, the method further comprises removing a target number of time points in the first modality image data, performing temporal layer correction and head movement correction on the image data, registering the first modality image data to a standard space, and performing temporal filtering and spatial smoothing on the first modality image data, the temporal filtering comprising a nonlinear trend and band pass filtering. Optionally, the method further comprises performing pre-commissure correction on the secon