Resting-State Functional Network Scale Effects and Statistical Significance-Based Feature Selection in Machine Learning Classification

Hao Guo,Junjie Chen,Yao Li,Jie Xiang,Dongwei Chen,Yong Xu,Godfred Kim Mensah

doi:10.1155/2019/9108108

Abstract

In recent years, functional brain network topological features have been widely used as classification features. Previous studies have found that network node scale differences caused by different network parcellation definitions significantly affect the structure of the constructed network and its topological properties. However, we still do not know how network scale differences affect the classification accuracy, performance of classification features, and effectiveness of the feature selection strategy using P values in terms of the machine learning method. This study used five scale parcellations, involving 90, 256, 497, 1003, and 1501 nodes. Three local properties of resting-state functional brain networks were selected (degree, betweenness centrality, and nodal efficiency), and the support vector machine method was used to construct classifiers to identify patients with major depressive disorder. We analyzed the impact of the five scales on classification accuracy. In addition, the effectiveness and redundancy of features obtained by the different scale parcellations were compared. Finally, traditional statistical significance (P value) was verified as a feature selection criterion. The results showed that the feature effectiveness of different scales was similar; in other words, parcellation with more regions did not provide more effective discriminative features. Nevertheless, parcellation with more regions did provide a greater quantity of discriminative features, which led to an improvement in the accuracy of the classification. However, due to the close distance between brain regions, the redundancy of parcellation with more regions was also greater. The traditional P value feature selection strategy is feasible with different scales, but our analysis showed that the traditional P < 0.05 threshold was too strict for feature selection. This study provides an important reference for the selection of network scales when applying topological properties of brain networks to machine learning methods.

Highlights

Machine learning and pattern recognition methods have been widely used in functional magnetic resonance data analysis in studies of brain thinking and cognitive state. e classification features selected for fMRI analysis are mostly direct features of the blood oxygen level-dependent (BOLD) signal, including peak, peak time, and slope, regardless of whether the analysis involves task-state fMRI [3, 4] or resting-state fMRI [5]
With the development of functional brain network research, more and more researchers have found that the rich topological information of functional networks can be used as biological markers for various neuropsychiatric diseases [6,7,8,9]. e extracted network topological features are widely used in the construction of classification models to assist in the diagnosis of brain diseases
The study had the following four objectives: (1) to define five scale parcellations, comprising 90, 256, 497, 1003, and 1501 nodes; (2) to analyze the effects of these five parcellations on classification accuracy; (3) to analyze and compare the feature effectiveness and redundancy among the different scales; and (4) to investigate the feasibility and reasonable threshold regarding the P value feature selection strategy. is study provides an important reference for selecting the network scale for future applications of brain network topological properties to machine learning methods

Summary

Introduction

Machine learning and pattern recognition methods have been widely used in functional magnetic resonance (fMRI) data analysis in studies of brain thinking and cognitive state (for review, see [1]). e classification features selected for fMRI analysis are mostly direct features of the blood oxygen level-dependent (BOLD) signal, including peak, peak time, and slope (for review, see [2]), regardless of whether the analysis involves task-state fMRI [3, 4] or resting-state (rs) fMRI [5]. We still do not know precisely how network scale affects the performance of classification features and classification accuracy, and effectiveness of the feature selection strategy using P values. In this context, this study applied five different node parcellations to construct and analyze the resting-state functional brain network of a control group and a disease group (involving patients with depression). The study had the following four objectives: (1) to define five scale parcellations, comprising 90, 256, 497, 1003, and 1501 nodes; (2) to analyze the effects of these five parcellations on classification accuracy; (3) to analyze and compare the feature effectiveness and redundancy among the different scales; and (4) to investigate the feasibility and reasonable threshold regarding the P value feature selection strategy. The study had the following four objectives: (1) to define five scale parcellations, comprising 90, 256, 497, 1003, and 1501 nodes; (2) to analyze the effects of these five parcellations on classification accuracy; (3) to analyze and compare the feature effectiveness and redundancy among the different scales; and (4) to investigate the feasibility and reasonable threshold regarding the P value feature selection strategy. is study provides an important reference for selecting the network scale for future applications of brain network topological properties to machine learning methods

Materials and Methods

Results

Conflicts of Interest