Exploring Brain Networks Using Functional Magnetic Resonance Imaging

There are around 100 billion neurons and 100 trillion synapses in the human brain, which is likely the most complex system in the universe as we known. The coordination of information across multiple spatiotemporal scales is a prerequisite of brain functions spanning all cognitive and behavioral domains. To explore the human brain more systemically and accurately, the brain science and cognitive neuroscience should be done from a viewpoint of brain connectivity and networks. As a noninvasive brain imaging technique, functional magnetic resonance imaging(f MRI) has a promising spatiotemporal resolution, providing a powerful tool in the investigation of large-scale functional brain networks. Based on resting-state f MRI, the dissertation explores a neurobiological basis of head motion in brain imaging and spatiotemporal dynamics of the spontaneous brain activity of the human beings, and makes new progress in the supervised and unsupervised classification frameworks of brain states with functional connectivity MRI(fc MRI) and their clinical applications. The main contents and contributions of the dissertation are as follows:Neurobiological basis of head motion in brain imaging. Individual differences in neuroimaging-based brain metrics, especially f MRI-based functional connectivity, can correlate with individual differences in spontaneous head motion during data collection. The assumption has been that motion causes artifactual differences in brain connectivity that must and can be corrected. In the second chapter, we propose that brain connectivity differences can also represent a neurobiological trait that predisposes to head motion differences. We support this assertion with an analysis of intra- versus inter-subject differences in connectivity comparing high to low motion subgroups. Inter-subject analysis identified a biomarker for head motion consisting of reduced distant functional connectivity primarily in the default network in individuals with high head motion. Similar connectivity differences were not found in analysis of intra-subject data. Instead, this biomarker was a stable property in individuals across time. The current findings suggest that the connectivity differences cannot be simply attributed to motion artifacts but also reflect individual variability in functional organization. Differentiating the true disease biomarkers from that of motion tendency as reported here is critical for identifying the treatment targets and development of treatment strategy for patients with brain diseases.Supervised classification of major depression using whole-brain fc MRI. In the third chapter, we investigate the whole-brain fc MRI patterns of individuals with major depression, and attempt to identify depressed patients from controls with such functional connectivity patterns. In the multivariate pattern analysis, 94.3% of subjects including 100% patients and 89.7% healthy controls were correctly classified via cross-validation. The most functional connections with highest discriminative power were primarily located within or between the default, affective and visual networks, as well as the cerebellum, thereby indicating that the alterations of the relavent intrinsic networks may give rise to a portion of the complex of emotional and cognitive disturbances in depression. Moreover, the amygdala, anterior cingulate cortex, parahippocampal gyrus, and hippocampus, which exhibit high discriminative power in classification, may play important roles in the pathophysiology of this disorder. This study suggests that whole-brain fc MRI may provide potential effective biomarkers for the clinical diagnosis of major depression.Unsupervised classification of brain network patterns using fc MRI. The clinical diagnosis of neuropsychiatric disorders including depression are based on self-reported symptoms and clinical signs, which may be prone to patients’ behaviors and clinicians’ bias. In the fourth chapter, we develop an unsupervised machine learning approach for accurately identifying individuals with major depressive disorder based on single resting-state fc MRI scans in the absence of clinical information. First of all, we divide the perigenual cingulate cortex into two subregions by using clustering with the distinct functional connectivity patterns: subgenual and pregenual regions, and show that a maximum margin clustering-based unsupervised classification approach extractes sufficient information from the subgenual cingulate connectivity maps to distinguish patients from controls with a group-level clustering consistency and an individual-level classification consistency of 92.5%. Furthermore, the most discriminating subgenual cingulate connectivity network primarily includes the ventral prefrontal cortices, superior temporal gyrus and limbic system areas, suggesting a critical role of the connectivity in the pathophysiology of major depressive disorder. The current study implies that subgenual cingulate connectivity network signatures provide potential neurobiological markers for the clinical diagnosis of depression and that maximum margin clustering-based unsupervised classification approaches may have the potential to inform clinical practice and aid in research on neuropsychiatric diseases.Spatiotemporal dynamics of the spontaneous activity in the human brain. The evidence of the default and dorsal attention networks being diametrically opposed has suggested a possible feature of brain organization in which cognitive processes subserving competing goals are spatially and temporally segregated. In the fifth chapter, transient formation and dissolution of brain states revealed through clustering of single BOLD(blood oxygenation level dependent)-f MRI coactivation patterns shows that the brain is temporally organized into multiple competing and spatially inverted representations. The configurations departed from canonical network relationships and within regions, could dynamically organize to distinctly follow histological or functional boundaries across time. The transition probability between states is highly symmetric with a low probability of switching between opposing states. The temporal organization of spatial representations is restructured during performing a language task in which task-relevant states observed at rest are selectively recruited. Collectively, the current results imply that the resting human brain dynamically alternates among a fixed repertoire of patterns dominated by pairs of opposing representations. With an improved understanding of the properties of the brain’s apparently fixed functional repertoire(including determination of their underlying electrophysiological correlates) the transient states may provide insight into ongoing cognitive process, provide an index of behavioral variability, and serve as potential neurobiological markers of brain diseases.