US 12,223,650 B2
System for predicting disease with graph convolutional neural network based on multimodal magnetic resonance imaging
Yu Zhang, Hangzhou (CN); Chaoliang Sun, Hangzhou (CN); Zhichao Wang, Hangzhou (CN); Huan Zhang, Hangzhou (CN); Haotian Qian, Hangzhou (CN); and Tianzi Jiang, Hangzhou (CN)
Assigned to ZHEJIANG LAB, Hangzhou (CN)
Filed by ZHEJIANG LAB, Hangzhou (CN)
Filed on Aug. 6, 2024, as Appl. No. 18/796,239.
Application 18/796,239 is a continuation of application No. PCT/CN2023/124639, filed on Oct. 16, 2023.
Claims priority of application No. 202211276172.2 (CN), filed on Oct. 19, 2022.
Prior Publication US 2024/0394882 A1, Nov. 28, 2024
Int. Cl. G06T 7/00 (2017.01); G16H 30/40 (2018.01)
CPC G06T 7/0012 (2013.01) [G16H 30/40 (2018.01); G06T 2207/10088 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30016 (2013.01)] 8 Claims
OG exemplary drawing
 
1. A system for predicting disease with graph convolutional neural network based on multimodal magnetic resonance imaging, wherein the system comprises:
a multimodal magnetic resonance data acquisition module configured to extract information in multimodal magnetic resonance data according to brain atlases, wherein the information comprises structural images, resting-state magnetic resonance data and diffusion magnetic resonance data;
a data preprocessing module configured to preprocess the structural images, the resting-state magnetic resonance data and the diffusion magnetic resonance data;
a brain radiomics information extraction module configured to calculate information on volumes, thicknesses and surface areas of a cortex in different brain regions according to the structural images processed by the data preprocessing module; calculate information about amplitude of low frequency fluctuations and regional homogeneity of the different brain regions according to the resting-state magnetic resonance data processed by the data preprocessing module; and calculate information about fractional anisotropy, mean diffusivity, intracellular volume fraction and orientation dispersion index of the different brain regions according to the diffusion magnetic resonance data processed by the data preprocessing module;
a brain connectomic information extraction module configured to calculate a functional connection matrix for each subject according to the resting-state magnetic resonance data processed by the data preprocessing module; and calculate a structural connection matrix for each subject according to the diffusion magnetic resonance data processed by the data preprocessing module;
a brain graph structure construction module configured to construct a feature vector for the obtained multimodal information of each brain region by taking each brain region of the brain atlases as a node, wherein the obtained multimodal information comprises brain structure indexes of the volumes, the thicknesses and the surface areas of the cortex in the different brain regions extracted from the structural images, brain function indexes of the amplitudes of low frequency fluctuations (ALFF) and regional homogeneity (ReHo) values of the different brain regions calculated from the resting-state magnetic resonance data, and brain diffusion indexes of values of the fractional anisotropy (FA), the mean diffusivity (MD), the intracellular volume fraction (ICVF) and the orientation dispersion index (ODI) of each brain region calculated from the diffusion magnetic resonance data; multiply the normalized functional connection matrix and the normalized structural connection matrix as an adjacency matrix; and construct brain graph structured data G (V, E) based on the adjacency matrix, wherein a node set V comprises brain regions extracted from the brain atlases, and an edge set E comprises adjacency matrices obtained by multiplication; and
a graph convolutional neural network model construction module configured to construct a graph convolutional neural network model, train the graph convolutional neural network model by taking the brain graph structured data as a model input and a group label of the subject as a model output, and predict brain diseases by the trained graph convolutional neural network model.