Individual Variability and Test-Retest Reliability Revealed by Ten Repeated Resting-State Brain Scans over One Month

Abstract

Individual differences in mind and behavior are believed to reflect the functional variability of the human brain. Due to the lack of a large-scale longitudinal dataset, the full landscape of variability within and between individual functional connectomes is largely unknown. We collected 300 resting-state functional magnetic resonance imaging (rfMRI) datasets from 30 healthy participants who were scanned every three days for one month. With these data, both intra- and inter-individual variability of six common rfMRI metrics, as well as their test-retest reliability, were estimated across multiple spatial scales. Global metrics were more dynamic than local regional metrics. Cognitive components involving working memory, inhibition, attention, language and related neural networks exhibited high intra-individual variability. In contrast, inter-individual variability demonstrated a more complex picture across the multiple scales of metrics. Limbic, default, frontoparietal and visual networks and their related cognitive components were more differentiable than somatomotor and attention networks across the participants. Analyzing both intra- and inter-individual variability revealed a set of high-resolution maps on test-retest reliability of the multi-scale connectomic metrics. These findings represent the first collection of individual differences in multi-scale and multi-metric characterization of the human functional connectomes in-vivo, serving as normal references for the field to guide the use of common functional metrics in rfMRI-based applications.

Fig 1. The overall analytic strategy implemented in the Connectome Computation System (CCS).

All individual rfMRI images (A) are first preprocessed through the CCS pipeline and then CCS projected individual rfMRI time series onto a uniform cortical surface grid based upon the brain tissue classification (B/C). As a proof of concept, an individual-level functional connectome can be modelled as a graph or network with vertices as nodes and pair-nodes dependency as edges (D). To achieve a systematic and comprehensive characterization of the brain graph, we proposed a set of metrics at three different scales from a single vertex to the whole cortex for measuring the amplitude and homogeneity at local scales, subnetworks or modules at meso-scales, and connectome centrality at the global scale (E). The connectome graph is reproduced and modified from [15].

Fig 2. Quantification of variability of seven common rfMRI metrics across the whole cortical mantle.

A) Auantification measures of intra-individual variability strength are plotted in polar form for amplitude of low-frequency fluctuations (ALFF), fractional ALFF (fALFF), regional homogeneity with length-one (ReHo1) and length-two (ReHo2) neighbors, seed-based connectivity analysis (SCA), weighted degree centrality (DCw) and eigenvector centrality (ECw). B) Auantification measures of inter-individual variability strength are plotted in polar form for the rfMRI metrics. C) Auantification measures of test-retest reliability strength are plotted in polar form for the rfMRI metrics. Note that for the purpose of visualization, all the strengths are normalized into a standard value between 0 and 1 by dividing the specific variance with the overall variance.

Fig 3. Renders of canonical large-scale networks and cognitive components.

A) The seven large-scale networks derived from a large sample (N = 1,000) of healthy resting-state brains [60] including Visual (Purple), Somatomotor (Blue), Dorsal Attention (Green), Ventral Attention (Violet), Limbic (Cream), Frontoparietal Control (Orange), Default (Red). This render projects these networks onto the fsaverage surface grid with its dorsal, ventral, lateral, medial, anterior and posterior views of the left hemisphere (LH) and the right hemisphere (RH). Dark gray curves indicate the boundaries between the seven networks. B) A surface render of the cognitive ontology of the brain derived from a large data set of neuroimaging experiments (N = 10,449) that contains twelve (C1-Hand, C2-Mouth, C3-Auditory, C4-Visual, C5-Language, C6-Attention, C7-Autonomic, C8-Inhibition, C9-Working Memory, C10-Default, C11-Basal and C12-Reward) components of cognition [66]. This image is reproduced and modified from [66].

Fig 4. Vertex-wise statistical maps of four common rfMRI metrics at local scales.

Maps of group-level statistical significance strength (the first row), test-retest reliability (the second row), intra-individual variability (the third row) and inter-individual variability (the forth row) are rendered onto the fsaverage surface grid with its lateral and medial views for amplitude and homogeneity metrics including amplitude of low-frequency fluctuations (ALFF, the first column), fractional ALFF (fALFF, the second column), regional homogeneity with length-one neighbors (ReHo1, the third column) and regional homogeneity with length-two neighbors (ReHo2, the forth column). Dark gray curves indicate the boundaries between the seven canonical neural networks.

Fig 5. Vertex-wise statistical maps of the seed-based rfMRI connectivity.

Maps of group-level statistical significance strength (A), test-retest reliability (B), inter-individual variability (C) and inter-individual variability (D) are rendered onto the fsaverage surface grid with its lateral and medial views. Dark gray curves indicate the boundaries between the seven canonical neural networks.

Fig 6. Vertex-wise statistical maps of large-scale common modular metrics at meso scales of subnetworks.

Maps of group-level significance strength (A), test-retest reliability (B), intra-individual variability (C) and inter-individual variability (D) are rendered onto the fsaverage surface grid with its lateral and medial views for within-network functional specialization (spatial patterns) and between-network functional integration (temporal interactions or interplays) of the seven common large-scale neural networks: Visual, SomMot, DorsAttn, VentAttn, Limbic, Control, Default. Dark gray curves indicate the boundaries between the seven canonical neural networks. Each connection line is plotted with mixed colors of the two networks it connects.

Fig 7. Vertex-wise statistical maps of two common rfMRI metrics at connectome scales.

Maps of group-level statistical significance strength (the first row), test-retest reliability (the second row), intra-individual variability (the third row) and inter-individual variability (the forth row) are rendered onto the fsaverage surface grid with its lateral and medial views for weighted degree centrality (DCw) and weighted eigenvector centrality (ECw). Dark gray curves indicate the boundaries between the seven canonical neural networks.