Functional Brain Network Estimation Based on Weighted BOLD Signals for MCI Identification

• Published in: Pattern Recognition and Computer Vision Vol. 11858; pp. 29 - 40
• Main Author: Chen, Huihui
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Functional brain network; Mild cognitive impairment (MCI); Pearson’s correlation; Resting-state functional magnetic resonance imaging (rs-fMRI); Scrubbing; Sparse representation
• ISBN: 9783030317225; 3030317226
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-31723-2_3

Functional brain network (FBN) provides an important way to reveal the inherent organization of the brain and explore informative biomarkers of neurological disorders. Due to its increasing potential in clinical applications, many methods, such as Pearson’s correlation and sparse representation, have been proposed in the recent years for FBN estimation. In practice, before the FBN estimation, a complex data preprocessing pipeline is involved to improve the quality of the data (i.e., fMRI signals in this paper), in which the scrubbing is an optional scheme for removing some “bad” time points (or volumes) from the fMRI signals according to a hard threshold related to, for example, the frame-wise displacement (FD). However, on one hand, the direct removal of time points may cause the loss of some useful information in data, and, on the other hand, the remaining time points may be not clean enough. In addition, with a fixed threshold, different numbers of volumes are generally scrubbed for different subjects, resulting in a bias or inconsistency in the estimated FBNs. To address these issues, in this paper, we develop a motion-dependent FBN estimation scheme by weighting the fMRI signals according to the values of FD. As a result, the proposed method can not only reduce the difficulty of threshold selection involved in the traditional scrubbing scheme, but also provide a more flexible framework that scrubs the data in the subsequent FBN estimation model. To verify the effectiveness of the proposed approach, we conduct experiments to identify subjects with mild cognitive impairment (MCI) from normal controls on a publicly available dataset. The experimental results show that our newly estimated FBNs can significantly improve the final classification accuracy.