Data-driven Models to Identify Functional Brain Networks
Author | : Wei Zhang |
Publisher | : |
Total Pages | : 250 |
Release | : 2019 |
ISBN-10 | : OCLC:1122566010 |
ISBN-13 | : |
Rating | : 4/5 (10 Downloads) |
Book excerpt: There have been extensive studies of brain functional networks in both human and primate's brains using fMRI signals. With the dramatic development of fMRI techniques, to interpret the functional networks at connectome-scale is necessary; so far, despite growing interests in the field, the techniques are employed to identify the functional networks are limited to the traditional methods, such as Principle Component Analysis (PCA) and Independent Component Analysis (ICA); these methods may have some potential redundant constraints and cannot work on the connectome-scale; thus, I propose a series of novel computational framework in order to identify connectome-scale functional networks using whole-brain fMRI data. At first, I apply sparse dictionary learning (SDL) on resting-state fMRI data of macaque brain. This new technique can successfully identify the 70 consistent intrinsic functional networks of macaque brain; and experimental results demonstrate that 70 consistent intrinsic functional networks expand currently known macaque ICNs already reported in the literature. Moreover, to comprehensively compare the performance of ICAs and SDLs, I propose a series of simulated experiments and provide the guidelines to use ICAs and SDLs based on different scenarios; and the experimental result offers empirical better understanding of ICA and SDL algorithms in inferring functional networks from fMRI data; finally, since hierarchical organization of brain function has been an established concept in the neuroscience field for a long time, however, it has been rarely demonstrated how such hierarchical macroscale functional networks are actually organized in the human brain. To answer this question, this dissertation proposes a new method of Hybrid Spatiotemporal Deep Learning (HSDL), by jointly using Deep Belief Networks (DBN) and Deep LASSO to reveal the temporal hierarchical features and spatial hierarchical features based on the Human Connectome Project (HCP) 900 fMRI datasets. Briefly, the key idea of HSDL is to extract the weights between two adjacent layers of DBN, which are then treated as the hierarchical dictionaries for Deep LASSO to identify the corresponding hierarchical spatial maps. Our results demonstrate that both spatial and temporal aspects of those dozens of functional networks exhibit multiscale properties that can be well characterized and interpreted based on existing computational tools and neuroscience knowledge.