Brain network oriented structured characteristic selection method

Abstract

The invention provides a brain network oriented structured characteristic selection method. The graph kernel based gk-SFS structured characteristic selection method is provided by considering that network analysis performance is influenced due to the fact that complex data of the brain network uses network local measurement as a characteristic vector for following characteristic selection and classification and inherent topological structure information of the network is neglected. The method includes two regularization items; one is a sparse regularization item that includes an L1 norm-form regularization item and ensures that only the network characteristics with a small amount of discrimination power can be selected; and other is a Laplace regularization item which is used to reserve integral distribution information of the brain network data, the similarity of network data is calculated via the graph kernel, and topological structure information of the brain network data is reserved. According to two real brain-disease data sets, the method has a higher performance for brain diseases compared with existing methods.

Description

technical field [0001] The invention belongs to the fields of machine learning and medical image analysis, and in particular relates to a brain network-oriented structured feature selection method. Background technique [0002] Modern magnetic resonance imaging (MRI) techniques, including functional MRI (fMRI), provide a non-invasive way to explore the human brain, revealing previously unrevealed insights into brain structure and function. mechanism. Brain network analysis can describe the interaction between brain regions at the connection level, and has become a new research hotspot in medical image analysis and neuroimaging. [0003] Recently, machine learning methods have been used in the analysis and classification of brain networks. For example, researchers have used brain networks for the diagnosis and classification of early brain diseases and achieved good performance. In these studies, it is typical to extract brain local measures (such as clustering coefficient...

AI Technical Summary

Problems solved by technology

Existing methods usually extract local measures (such as edge weights or clustering coefficients) as features from network data and combine them into a long feature vector for subsequent feature selection and classification, while some useful network structures Information such as the overall topology of the network is lost, which may degrade the final classification performance

Images