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

Abstract

Resting-state functional MRI (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, i.e., should not span multiple lobes. There is disagreement about whether areal-level parcels should be strictly contiguous or comprise multiple non-contiguous components, therefore we considered three areal-level MS-HBM variants spanning these range of possibilities. Individual-specific MS-HBM parcellations estimated using 10min of data generalized better than other approaches using 150min of data to out-of-sample rs-fMRI and task-fMRI from the same individuals. Resting-state functional connectivity (RSFC) derived from MS-HBM parcellations also achieved the best behavioral prediction performance. Among the three MS-HBM variants, the strictly contiguous MS-HBM (cMS-HBM) exhibited the best resting-state homogeneity and most uniform within-parcel task activation. In terms of behavioral prediction, the gradient-infused MS-HBM (gMS-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 (GITHUB_LINK).

Predicting intelligence from fMRI data of the human brain in a few minutes of scan time

“…There is a growing hope that functional MRI (fMRI) may help to fill this gap. In this context, the suitability of fMRI for predicting individual behaviour has been investigated in a number of recent studies [1][2][3][4][5][6][7][8][9].…”

BAI-Net: Individualized Anatomical Cerebral Cartography using Graph Neural Network

“…To test this, we applied a simple linear model to predict individual cognition behaviors using the functional connectome of parcellations derived from the group-registered, IC or BAI-Net methods, respectively. We estimated the performance on 58 cognition behaviors (the same with [33,37]), using Pearson correlation (r score) to quantify the association between predicted values and observed behavioral scores in the leave-one-out cross validation. Among these scores, 34 behaves are significantly predicted (p < .05) by at least one of functional connectomes extracted from these parcellations.…”

“…Many studies adopted cortical fingerprint-based (e.g., anatomical and functional connectivity) mappings for individualizing the population atlas to reveal individualspecific variability of cortical areas [13,[25][26][27][28][29][30][31][32][33]. Population atlas priors provide a blueprint of functional and anatomical architectures of the human brain in these mappings to guide individual-specific topography presented by connection fingerprints [13,23,26,[32][33][34][35][36][37]. However, there still exists a great challenge for achieving a robust, individual-specific cortical atlas that can reveal individual areal differences, which are obscured by population averaging, and can highly replicate such differences across repeated sessions [38].…”

“…Population priors should be integrated reliably into the mapping to handle the variations of the connectivity fingerprint as well as to avoid obvious outliers in the individual parcellations. Previous methods have made lots of efforts on the utilization of population priors, such as providing rough spatial locations of cortical areas or the distribution of cortical fingerprints on these locations, or iterating group connection fingerprints [13,26,27,30,37]. For example, Han et al presented well delineation of individualized cerebral cartography by iteratively updating reference anatomical connectivities for each area, and matching them with vertex-level anatomical connectivities to identify individual cortical areas (denoted as iterative clustering (IC) methods) [26,30], but it might present areal outliers far from its group locations when faced with fingerprint noises or population biases on certain small areas.…”

“…For example, Han et al presented well delineation of individualized cerebral cartography by iteratively updating reference anatomical connectivities for each area, and matching them with vertex-level anatomical connectivities to identify individual cortical areas (denoted as iterative clustering (IC) methods) [26,30], but it might present areal outliers far from its group locations when faced with fingerprint noises or population biases on certain small areas. From functional aspects, the functional connectome reflected by an individual parcellation should have more associations with related individual cognition behaviors or the illness symptoms, and provides a more precise individual measurement for the clinic practices [37,40].…”