Brain network analysis method based on diffusion MRI image

Abstract

According to the brain network analysis method based on the diffusion MRI image, a double-task model training method is adopted at the same time, and comprehensive analysis of the brain network is achieved. The problem that a brain network carries out representation learning in a non-Euclidean space is solved. Meanwhile, a double-task model training method is adopted, clinical data are fully utilized, and qualitative and quantitative analysis results of the brain network are obtained. According to the method, a graph neural network (GNNs) model capable of sensing irregular structure data features in a non-Euclidean space is adopted to carry out representation learning on a brain network, so that the brain network representation accuracy of the model is improved. In addition, network parameters are adjusted by adopting a double-task training strategy, so that the practicability of an analysis result is improved.

Description

technical field [0001] The invention belongs to the technical fields of computer vision and medical image processing, and in particular relates to a brain network analysis method based on diffusion MRI images. Background technique [0002] Brain network analysis based on medical imaging is a research hotspot in the field of computer vision in recent years. Image feature extraction and brain network representation are two key steps in brain image analysis. The method of brain network representation guides the results of image feature extraction. The quality of extracted features is related to the quality of brain network representation learning, and ultimately affects the performance of the trained network model. Therefore, the brain network representation method is the core of brain image analysis technology. part. According to the analysis method of extracting brain network representation, the research of brain network is mainly divided into two directions: analysis based...

AI Technical Summary

Problems solved by technology

Due to the characteristics of the matrix grid, the similarity of features can only be matched in the Euclidean space, ignoring the rich relationship information of the brain network in the non-Euclidean space

[0004] In addition, most deep learning-based brain network representation methods focus on the classification task of brain networks and extract group-level features, while ignoring other tasks that combine structured image data with unstructured data (i.e., graph-structured data). (such as score regression tasks), cannot make full use of effective data in clinical applications, and improve the practicability of brain network analysis

[0005] As mentioned above, the currently commonly used deep learning methods are all aimed at data with regular grid structures, and representing the brain network as structured data for feature learning in Euclidean space will lose its rich relational information in non-Euclidean space. Therefore, the results are limited and cannot fully utilize the advantages of brain network analysis in clinical applications

[0006] In addition, most deep learning-based brain network representation methods focus on the classification tasks of brain networks, which is not conducive to the comprehensive application of structured image data and unstructured data (such as clinical scores, electronic medical records, etc.) (such as score regression tasks) , can not make full use of clinical data for more comprehensive and effective brain network analysis

Images