The Evaluation of Brain Age Prediction by Different Functional Brain Network Construction Methods

Abstract

Brain functional network (BFN) analysis based on functional magnetic resonance imaging (fMRI) has proven to be a value method for revealing organization architectures in normal aging brains. However, a comprehensive comparison of different BFN methods for predicting brain age remains lacking. In this paper, we introduce a novel method to establish the BFN by using the Schatten-0 (\( S_0 \)) and \( \ell _0 \)-regularized low rank sparse representation (\({S_0}{{/}}{\ell _{{0}}}\) LSR) method. Moreover, the performance of different BFN methods in the brain age prediction with different feature extraction methods is evaluated. A support vector regression (SVR) is applied to the BFN data to predict brain age. Experimental results for resting state fMRI data sets show that compared with the Pearson correlation (PC), sparse representation (SR), low rank representation (LR), and low rank sparse representation (LSR) methods, the LSR method can achieve better modularity and predict brain age more accurately. The novel approach can enhance our understanding of the functional network of the aging brain.