Zoom Meeting link: https://unimelb.zoom.us/j/94123635321?pwd=UGlIakRtTUN5NmFPUnZlY3F5NWM3dz09
Short video: https://youtu.be/Ew15Jcgatvc
Sina Mansour L.,
Vanessa Cropley,
Ye Tian,
Andrew ZaleskyIntroduction:No two brains are alike. Neuroimaging can be used to elucidate the neural basis of human identity and to map neural correlates of human behavior and cognition. While most neuroimaging features are commonly studied in the highest resolution provided by the scanner, brain connectivity studies are traditionally conducted in a lower resolution of a predefined atlas. The aim of this study was to contrast the identification (fingerprinting) and behavior prediction capabilities of the atlas-based brain networks with a novel ultra- high-resolution model of brain networks. This comparison will explore the potential gains of conducting brain connectivity analyses in a higher resolution.
Methods:We analyzed neuroimaging and behavioral data acquired from 1000 individuals participating in the Human Connectome Project (HCP). These individuals formed a "test" group. A repeated MRI scan on a different day from 42 of these individuals formed a "retest" group. Structural connectivity was mapped using probabilistic tractography of Diffusion MRI. Functional connectivity was mapped using resting-state functional MRI. Additionally, surface maps for cortical measures of Thickness, curvature, sulcal depth, and myelination were sourced from the HCP. Hence each scan was associated with 6 different neuroimaging characteristics of structure, morphology, and connectivity, each of which was mapped at the higher resolution of vertices (~32,000 nodes per hemisphere), as well as regions comprising an established atlas (180 regions per hemisphere) yielding 12 total measures.
For each measure, a similarity metric for all scan pairs was computed. This similarity information was used to quantify the extent of identifiable information captured by neuroimaging measures. We computed the effect size difference in intra- and inter-subject similarity distributions as an identifiability metric. Hence, higher effect size differences translated to higher precision in identification. To capture the extent of behavioral associations of every measure, Independent component analysis was used to decompose 109 behavioral measures sourced from HCP to five core continuous dimensions, characterizing cognitive performance, illicit substance use, tobacco use, personality-emotion traits, and mental health. Variance component modeling was used to evaluate the extent to which each neuroimaging measure could explain individual variation in each behavioral dimension.
Results:Comparing the correlates of neural identity and behavior in atlas-based models with the ultra-high-resolution alternative revealed the extent of information gain achieved by the increase in spatial resolution (Fig.1). Our findings show behavioral associations of all neuroimaging modalities significantly increased as a result of the increase in spatial resolution. In particular, behavior associations of structural connectivity, functional connectivity, and cortical thickness benefitted the most from ultra-high-resolution analyses. The neural correlates of individual identity were also better detected in higher resolution, especially for structural connectivity and all measures of morphology (cortical curvature, thickness, and sulcal depth). The identification improvements of functional connectivity and myelination were minimal. We proposed a novel model of high-resolution structural brain networks that surpasses the ability of atlas-based alternatives in both identification and behavior explanation of individuals.
