Individual-Specific Areal-Level Parcellations Improve Functional Connectivity Prediction of Behavior

Abstract

Resting-state functional magnetic resonance imaging (rs-fMRI) allows estimation of individual-specific cortical parcellations. We have previously developed a multi-session hierarchical Bayesian model (MS-HBM) for estimating high-quality individual-specific network-level parcellations. Here, we extend the model to estimate individual-specific areal-level parcellations. While network-level parcellations comprise spatially distributed networks spanning the cortex, the consensus is that areal-level parcels should be spatially localized, that is, should not span multiple lobes. There is disagreement about whether areal-level parcels should be strictly contiguous or comprise multiple noncontiguous components; therefore, we considered three areal-level MS-HBM variants spanning these range of possibilities. Individual-specific MS-HBM parcellations estimated using 10 min of data generalized better than other approaches using 150 min of data to out-of-sample rs-fMRI and task-fMRI from the same individuals. Resting-state functional connectivity derived from MS-HBM parcellations also achieved the best behavioral prediction performance. Among the three MS-HBM variants, the strictly contiguous MS-HBM exhibited the best resting-state homogeneity and most uniform within-parcel task activation. In terms of behavioral prediction, the gradient-infused MS-HBM was numerically the best, but differences among MS-HBM variants were not statistically significant. Overall, these results suggest that areal-level MS-HBMs can capture behaviorally meaningful individual-specific parcellation features beyond group-level parcellations. Multi-resolution trained models and parcellations are publicly available (https://github.com/ThomasYeoLab/CBIG/tree/master/stable_projects/brain_parcellation/Kong2022_ArealMSHBM).

An integrated resource for functional and structural connectivity of the marmoset brain

“…It is also worth noting that our functional parcels do not exactly correspond to the traditional cytoarchitectonic definition of the cortical areas (Amunts and Zilles, 2015;Kaas, 1987). Consistent with many brain parcellations by non-invasive neuroimaging Kong et al, 2021;Schaefer et al, 2018;Van Essen and Glasser, 2018;Van Essen et al, 2012), our defined area-level functional parcels most likely reflect a different type of computational sub-units, in agreement with the idea that the brain is organized in multiple scales (Churchland and Sejnowski, 1988;van den Heuvel and Yeo, 2017). Therefore, compared with available structural atlases, MBMv4 captures the organization of functional connectivity accurately.…”

“…Since the neuronal tracing data is directional, its intactness is helpful for the mapping of the structural connectome, and the accurate linkage between the structural and functional connectome in our next stage, which is the unidirectional diffusion tractography not capture. Finally, our parcellation only used the resting-state functional connectivity information as in many human studies Kong et al, 2021;Schaefer et al, 2018). Thus, combining multiple task-fMRI data to improve functional parcellation becomes important in the future.…”

Two common and distinct forms of variation in human functional brain networks

“…At present, many resting-state and task-based fMRI studies aggregate or compare data across individuals based on spatial normalization, thus assuming that the same spatial layout of brain systems is conserved across individuals (Fedorenko, 2021). However, widespread individual differences in the localization of brain regions and systems may lead to detrimental effects when performing group-level analyses, including loss of sensitivity and functional resolution (Nieto-Castanon and Fedorenko, 2012), and prediction accuracy of task functional connectivity (Porter et al, 2021), task-evoked signals (Guntupalli et al, 2018; Haxby et al, 2020), and prediction of behavior from resting networks (Brennan et al, 2019; Fan et al, 2020; Kong et al, 2019; Kong et al, 2021). Together, these limitations lead group studies to fall short of providing comprehensive explanations of brain function and cognition as a whole.…”

“…This includes individualized approaches such as collecting functional localizer task data from each individual subject (Fedorenko, 2021; Fedorenko et al, 2013; Kanwisher et al, 1997; Saxe et al, 2006), adopting fcMRI methods to define brain systems and areas from subjects with large quantities of data (Braga and Buckner, 2017; Gordon et al, 2017c), or hyperalignment-based techniques to increase functional correspondence (Guntupalli et al, 2018; Haxby et al, 2011; Haxby et al, 2020). Other methods developed to address functional alignment include template-matching techniques (e.g., (Gordon et al, 2017a; Gordon et al, 2017b)), multi-modal functional/anatomical registration (Glasser et al, 2016), and hierarchical functional parcellation approaches (e.g., (Kong et al, 2019; Kong et al, 2021)) to identify brain systems and regions in individuals even with more modest amounts of data.…”

“…Much work treats all forms of cortical variation uniformly, assuming that they are primarily driven by boundary shifts in the borders between systems: a functional region may expand, contract, or be slightly offset relative to the typical pattern (e.g., as has been documented for V1, which can differ more than 2-fold in size across individuals (Dougherty et al, 2003; Frost and Goebel, 2012)). Border shifts are likely important for understanding individual differences in behavior and brain function (Cui et al, 2020; Kong et al, 2019; Kong et al, 2021), and could potentially be addressed through functional alignment methods that allow for individually-specific regions to be locally displaced (e.g., (Haxby et al, 2011; Kong et al, 2021)).…”