Search

Brain decoding aims to infer cognitive states from patterns of brain activity. Substantial inter-individual variations in functional brain organization challenge accurate decoding performed at the group level. In this paper, we tested whether accurate brain decoding models can be trained entirely at the individual level. We trained several classifiers on a dense individual functional magnetic resonance imaging (fMRI) dataset for which six participants completed the entire Human Connectome Project (HCP) task battery >13 times over ten separate fMRI sessions. We evaluated nine decoding methods, from Support Vector Machines (SVM) and Multi-Layer Perceptron (MLP) to Graph Convolutional Neural Networks (GCN). All decoders were trained to classify single fMRI volumes into 21 experimental conditions simultaneously, using ∼7 h of fMRI data per participant. The best prediction accuracies were achieved with GCN and MLP models, whose performance (57-67 % accuracy) approached state-of-the-art accuracy (76 %) with models trained at the group level on >1 K hours of data from the original HCP sample. Our SVM model also performed very well (54-62 % accuracy). Feature importance maps derived from MLP -our best-performing model- revealed informative features in regions relevant to particular cognitive domains, notably in the motor cortex. We also observed that inter-subject classification achieved substantially lower accuracy than subject-specific models, indicating that our decoders learned individual-specific features. This work demonstrates that densely-sampled neuroimaging datasets can be used to train accurate brain decoding models at the individual level. We expect this work to become a useful benchmark for techniques that improve model generalization across multiple subjects and acquisition conditions., Competing Interests: Declaration of Competing Interest The authors declare they have no conflict of interest related to this study., (Copyright © 2023. Published by Elsevier Inc.)

A key goal in neuroscience is to understand brain mechanisms of cognitive functions. An emerging approach is "brain decoding", which consists of inferring a set of experimental conditions performed by a participant, using pattern classification of brain activity. Few works so far have attempted to train a brain decoding model that would generalize across many different cognitive tasks drawn from multiple cognitive domains. To tackle this problem, we proposed a multidomain brain decoder that automatically learns the spatiotemporal dynamics of brain response within a short time window using a deep learning approach. We evaluated the decoding model on a large population of 1200 participants, under 21 different experimental conditions spanning six different cognitive domains, acquired from the Human Connectome Project task-fMRI database. Using a 10s window of fMRI response, the 21 cognitive states were identified with a test accuracy of 90% (chance level 4.8%). Performance remained good when using a 6s window (82%). It was even feasible to decode cognitive states from a single fMRI volume (720ms), with the performance following the shape of the hemodynamic response. Moreover, a saliency map analysis demonstrated that the high decoding performance was driven by the response of biologically meaningful brain regions. Together, we provide an automated tool to annotate human brain activity with fine temporal resolution and fine cognitive granularity. Our model shows potential applications as a reference model for domain adaptation, possibly making contributions in a variety of domains, including neurological and psychiatric disorders., Competing Interests: Declaration of Competing Interest The authors declare no competing financial interests., (Copyright © 2021. Published by Elsevier Inc.)

Increasing the reproducibility of neuroimaging measurement addresses a central impediment to the advancement of human neuroscience and its clinical applications. Recent efforts demonstrating variance in functional brain organization within and between individuals shows a need for improving reproducibility of functional parcellations without long scan times. We apply bootstrap aggregation, or bagging, to the problem of improving reproducibility in functional parcellation. We use two large datasets to demonstrate that compared to a standard clustering framework, bagging improves the reproducibility and test-retest reliability of both cortical and subcortical functional parcellations across a range of sites, scanners, samples, scan lengths, clustering algorithms, and clustering parameters (e.g., number of clusters, spatial constraints). With as little as 6 ​min of scan time, bagging creates more reproducible group and individual level parcellations than standard approaches with twice as much data. This suggests that regardless of the specific parcellation strategy employed, bagging may be a key method for improving functional parcellation and bringing functional neuroimaging-based measurement closer to clinical impact., (Copyright © 2020 The Authors. Published by Elsevier Inc. All rights reserved.)

Studies using resting-state functional magnetic resonance imaging (rsfMRI) are increasingly collecting data at multiple sites in order to speed up recruitment or increase sample size. The main objective of this study was to assess the long-term consistency of rsfMRI connectivity maps derived at multiple sites and vendors using the Canadian Dementia Imaging Protocol (CDIP, www.cdip-pcid.ca). Nine to 10 min of functional BOLD images were acquired from an adult cognitively healthy volunteer scanned repeatedly at 13 Canadian sites on three scanner makes (General Electric, Philips and Siemens) over the course of 2.5 years. The consistency (spatial Pearson's correlation) of rsfMRI connectivity maps for seven canonical networks ranged from 0.3 to 0.8, with a negligible effect of time, but significant site and vendor effects. We noted systematic differences in data quality (i.e. head motion, number of useable time frames, temporal signal-to-noise ratio) across vendors, which may also confound some of these results, and could not be disentangled in this sample. We also pooled the long-term longitudinal data with a single-site, short-term (1 month) data sample acquired on 26 subjects (10 scans per subject), called HNU1. Using randomly selected pairs of scans from each subject, we quantified the ability of a data-driven unsupervised cluster analysis to match two scans of the same subjects. In this "fingerprinting" experiment, we found that scans from the Canadian subject (Csub) could be matched with high accuracy intra-site (>95% for some networks), but that the accuracy decreased substantially for scans drawn from different sites and vendors, even falling outside of the range of accuracies observed in HNU1. Overall, our results demonstrate good multivariate stability of rsfMRI measures over several years, but substantial impact of scanning site and vendors. How detrimental these effects are will depend on the application, yet our results demonstrate that new methods for harmonizing multisite analysis represent an important area for future work., (Copyright © 2019. Published by Elsevier Inc.)

Moving from group level to individual level functional parcellation maps is a critical step for developing a rich understanding of the links between individual variation in functional network architecture and cognitive and clinical phenotypes. Still, the identification of functional units in the brain based on intrinsic functional connectivity and its dynamic variations between and within subjects remains challenging. Recently, the bootstrap analysis of stable clusters (BASC) framework was developed to quantify the stability of functional brain networks both across and within subjects. This multi-level approach utilizes bootstrap resampling for both individual and group-level clustering to delineate functional units based on their consistency across and within subjects, while providing a measure of their stability. Here, we optimized the BASC framework for functional parcellation of the basal ganglia by investigating a variety of clustering algorithms and similarity measures. Reproducibility and test-retest reliability were computed to validate this analytic framework as a tool to describe inter-individual differences in the stability of functional networks. The functional parcellation revealed by stable clusters replicated previous divisions found in the basal ganglia based on intrinsic functional connectivity. While we found moderate to high reproducibility, test-retest reliability was high at the boundaries of the functional units as well as within their cores. This is interesting because the boundaries between functional networks have been shown to explain most individual phenotypic variability. The current study provides evidence for the consistency of the parcellation of the basal ganglia, and provides the first group level parcellation built from individual-level cluster solutions. These novel results demonstrate the utility of BASC for quantifying inter-individual differences in the functional organization of brain regions, and encourage usage in future studies., (Copyright © 2018. Published by Elsevier Inc.)

Connectivity studies using resting-state functional magnetic resonance imaging are increasingly pooling data acquired at multiple sites. While this may allow investigators to speed up recruitment or increase sample size, multisite studies also potentially introduce systematic biases in connectivity measures across sites. In this work, we measure the inter-site effect in connectivity and its impact on our ability to detect individual and group differences. Our study was based on real, as opposed to simulated, multisite fMRI datasets collected in N=345 young, healthy subjects across 8 scanning sites with 3T scanners and heterogeneous scanning protocols, drawn from the 1000 functional connectome project. We first empirically show that typical functional networks were reliably found at the group level in all sites, and that the amplitude of the inter-site effects was small to moderate, with a Cohen's effect size below 0.5 on average across brain connections. We then implemented a series of Monte-Carlo simulations, based on real data, to evaluate the impact of the multisite effects on detection power in statistical tests comparing two groups (with and without the effect) using a general linear model, as well as on the prediction of group labels with a support-vector machine. As a reference, we also implemented the same simulations with fMRI data collected at a single site using an identical sample size. Simulations revealed that using data from heterogeneous sites only slightly decreased our ability to detect changes compared to a monosite study with the GLM, and had a greater impact on prediction accuracy. However, the deleterious effect of multisite data pooling tended to decrease as the total sample size increased, to a point where differences between monosite and multisite simulations were small with N=120 subjects. Taken together, our results support the feasibility of multisite studies in rs-fMRI provided the sample size is large enough., (Copyright © 2017. Published by Elsevier Inc.)

Resting-state functional connectivity (RSFC) studies have provided strong evidences that visual deprivation influences the brain's functional architecture. In particular, reduced RSFC coupling between occipital (visual) and temporal (auditory) regions has been reliably observed in early blind individuals (EB) at rest. In contrast, task-dependent activation studies have repeatedly demonstrated enhanced co-activation and connectivity of occipital and temporal regions during auditory processing in EB. To investigate this apparent discrepancy, the functional coupling between temporal and occipital networks at rest was directly compared to that of an auditory task in both EB and sighted controls (SC). Functional brain clusters shared across groups and cognitive states (rest and auditory task) were defined. In EBs, we observed higher occipito-temporal correlations in activity during the task than at rest. The reverse pattern was observed in SC. We also observed higher temporal variability of occipito-temporal RSFC in EB suggesting that occipital regions in this population may play the role of a multiple demand system. Our study reveals how the connectivity profile of sighted and early blind people is differentially influenced by their cognitive state, bridging the gap between previous task-dependent and RSFC studies. Our results also highlight how inferring group-differences in functional brain architecture solely based on resting-state acquisition has to be considered with caution., (Copyright © 2016 Elsevier Inc. All rights reserved.)

In 2011, the "ADHD-200 Global Competition" was held with the aim of identifying biomarkers of attention-deficit/hyperactivity disorder from resting-state functional magnetic resonance imaging (rs-fMRI) and structural MRI (s-MRI) data collected on 973 individuals. Statisticians and computer scientists were potentially the most qualified for the machine learning aspect of the competition, but generally lacked the specialized skills to implement the necessary steps of data preparation for rs-fMRI. Realizing this barrier to entry, the Neuro Bureau prospectively collaborated with all competitors by preprocessing the data and sharing these results at the Neuroimaging Informatics Tools and Resources Clearinghouse (NITRC) (http://www.nitrc.org/frs/?group_id=383). This "ADHD-200 Preprocessed" release included multiple analytical pipelines to cater to different philosophies of data analysis. The processed derivatives included denoised and registered 4D fMRI volumes, regional time series extracted from brain parcellations, maps of 10 intrinsic connectivity networks, fractional amplitude of low frequency fluctuation, and regional homogeneity, along with grey matter density maps. The data was used by several teams who competed in the ADHD-200 Global Competition, including the winning entry by a group of biostaticians. To the best of our knowledge, the ADHD-200 Preprocessed release was the first large public resource of preprocessed resting-state fMRI and structural MRI data, and remains to this day the only resource featuring a battery of alternative processing paths., (Copyright © 2016 Elsevier Inc. All rights reserved.)

In systems neuroscience, the term "connectivity" has been defined in numerous ways, according to the particular empirical modality from which it is derived. Due to large differences in the phenomena measured by these modalities, the assumptions necessary to make inferences about axonal connections, and the limitations accompanying each, brain connectivity remains an elusive concept. Despite this, only a handful of studies have directly compared connectivity as inferred from multiple modalities, and there remains much ambiguity over what the term is actually referring to as a biological construct. Here, we perform a direct comparison based on the high-resolution and high-contrast Enhanced Nathan Klein Institute (NKI) Rockland Sample neuroimaging data set, and the CoCoMac database of tract tracing studies. We compare four types of commonly-used primate connectivity analyses: tract tracing experiments, compiled in CoCoMac; group-wise correlation of cortical thickness; tractographic networks computed from diffusion-weighted MRI (DWI); and correlational networks obtained from resting-state BOLD (fMRI). We find generally poor correspondence between all four modalities, in terms of correlated edge weights, binarized comparisons of thresholded networks, and clustering patterns. fMRI and DWI had the best agreement, followed by DWI and CoCoMac, while other comparisons showed striking divergence. Networks had the best correspondence for local ipsilateral and homotopic contralateral connections, and the worst correspondence for long-range and heterotopic contralateral connections. k-Means clustering highlighted the lowest cross-modal and cross-species consensus in lateral and medial temporal lobes, anterior cingulate, and the temporoparietal junction. Comparing the NKI results to those of the lower resolution/contrast International Consortium for Brain Imaging (ICBM) dataset, we find that the relative pattern of intermodal relationships is preserved, but the correspondence between human imaging connectomes is substantially better for NKI. These findings caution against using "connectivity" as an umbrella term for results derived from single empirical modalities, and suggest that any interpretation of these results should account for (and ideally help explain) the lack of multimodal correspondence., (Copyright © 2015 Elsevier Inc. All rights reserved.)

A recent trend in functional magnetic resonance imaging is to test for association of clinical disorders with every possible connection between selected brain parcels. We investigated the impact of the resolution of functional brain parcels, ranging from large-scale networks to local regions, on a mass univariate general linear model (GLM) of connectomes. For each resolution taken independently, the Benjamini-Hochberg procedure controlled the false-discovery rate (FDR) at nominal level on realistic simulations. However, the FDR for tests pooled across all resolutions could be inflated compared to the FDR within resolution. This inflation was severe in the presence of no or weak effects, but became negligible for strong effects. We thus developed an omnibus test to establish the overall presence of true discoveries across all resolutions. Although not a guarantee to control the FDR across resolutions, the omnibus test may be used for descriptive analysis of the impact of resolution on a GLM analysis, in complement to a primary analysis at a predefined single resolution. On three real datasets with significant omnibus test (schizophrenia, congenital blindness, motor practice), markedly higher rate of discovery were obtained at low resolutions, below 50, in line with simulations showing increase in sensitivity at such resolutions. This increase in discovery rate came at the cost of a lower ability to localize effects, as low resolution parcels merged many different brain regions together. However, with 30 or more parcels, the statistical effect maps were biologically plausible and very consistent across resolutions. These results show that resolution is a key parameter for GLM-connectome analysis with FDR control, and that a functional brain parcellation with 30 to 50 parcels may lead to an accurate summary of full connectome effects with good sensitivity in many situations., (Copyright © 2015 Elsevier Inc. All rights reserved.)

The consolidation of motor sequence learning is known to depend on sleep. Work in our laboratory and others have shown that the striatum is associated with this off-line consolidation process. In this study, we aimed to quantify the sleep-dependent dynamic changes occurring at the network level using a measure of functional integration. We directly compared changes in connectivity before and after sleep or the simple passage of daytime. As predicted, the results revealed greater integration within the cortico-striatal network after sleep, but not an equivalent daytime period. Importantly, a similar pattern of results was also observed using a data-driven approach; the increase in integration being specific to a cortico-striatal network, but not to other known functional networks. These findings reveal, for the first time, a new signature of motor sequence consolidation: a greater between-regions interaction within the cortico-striatal system., (Copyright © 2014 Elsevier Inc. All rights reserved.)

The effect of regressing out the global average signal (GAS) in resting state fMRI data has become a concern for interpreting functional connectivity analyses. It is not clear whether the reported anti-correlations between the Default Mode and the Dorsal Attention Networks are intrinsic to the brain, or are artificially created by regressing out the GAS. Here we introduce a concept, Impact of the Global Average on Functional Connectivity (IGAFC), for quantifying the sensitivity of seed-based correlation analyses to the regression of the GAS. This voxel-wise IGAFC index is defined as the product of two correlation coefficients: the correlation between the GAS and the fMRI time course of a voxel, times the correlation between the GAS and the seed time course. This definition enables the calculation of a threshold at which the impact of regressing-out the GAS would be large enough to introduce spurious negative correlations. It also yields a post-hoc impact correction procedure via thresholding, which eliminates spurious correlations introduced by regressing out the GAS. In addition, we introduce an Artificial Negative Correlation Index (ANCI), defined as the absolute difference between the IGAFC index and the impact threshold. The ANCI allows a graded confidence scale for ranking voxels according to their likelihood of showing artificial correlations. By applying this method, we observed regions in the Default Mode and Dorsal Attention Networks that were anti-correlated. These findings confirm that the previously reported negative correlations between the Dorsal Attention and Default Mode Networks are intrinsic to the brain and not the result of statistical manipulations. Our proposed quantification of the impact that a confound may have on functional connectivity can be generalized to global effect estimators other than the GAS. It can be readily applied to other confounds, such as systemic physiological or head movement interferences, in order to quantify their impact on functional connectivity in the resting state., (© 2013.)

Imaging cerebral glucose metabolism with positron emission tomography (PET) in Alzheimer's disease (AD) has allowed for improved characterisation of this pathology. Such patterns are typically analysed using either univariate or multivariate statistical techniques. In this work we combined voxel-based group analysis and independent component analysis to extract differential characteristic patterns from PET data of glucose metabolism in a large cohort of normal elderly controls and patients with AD. The patterns were used in conjunction with a support vector machine to discriminate between subjects with mild cognitive impairment (MCI) at risk or not of converting to AD. The method was applied to baseline fluoro-deoxyglucose (FDG)-PET images of subjects from the ADNI database. Our approach achieved improved early detection and differentiation of typical versus pathological metabolic patterns in the MCI population, reaching 80% accuracy (85% sensitivity and 75% specificity) when using selected regions. The method has the potential to assist in the advance diagnosis of Alzheimer's disease, and to identify early in the development of the disease those individuals at high risk of rapid cognitive decline who could be candidates for new therapeutic approaches., (Copyright © 2012 Elsevier Inc. All rights reserved.)

Empirical evidence increasingly supports the hypothesis that patterns of intrinsic functional connectivity (iFC) are sculpted by a history of evoked coactivation within distinct neuronal networks. This, together with evidence of strong correspondence among the networks defined by iFC and those delineated using a variety of other neuroimaging techniques, suggests a fundamental brain architecture detectable across multiple functional and structural imaging modalities. Here, we leverage this insight to examine the functional organization of the human insula. We parcellated the insula on the basis of three distinct neuroimaging modalities - task-evoked coactivation, intrinsic (i.e., task-independent) functional connectivity, and gray matter structural covariance. Clustering of these three different covariance-based measures revealed a convergent elemental organization of the insula that likely reflects a fundamental brain architecture governing both brain structure and function at multiple spatial scales. While not constrained to be hierarchical, our parcellation revealed a pseudo-hierarchical, multiscale organization that was consistent with previous clustering and meta-analytic studies of the insula. Finally, meta-analytic examination of the cognitive and behavioral domains associated with each of the insular clusters obtained elucidated the broad functional dissociations likely underlying the topography observed. To facilitate future investigations of insula function across healthy and pathological states, the insular parcels have been made freely available for download via http://fcon_1000.projects.nitrc.org, along with the analytic scripts used to perform the parcellations., (Copyright © 2012 Elsevier Inc. All rights reserved.)

Carbon dioxide (CO(2)), a potent vasodilator, is known to have a significant impact on the blood-oxygen level dependent (BOLD) signal. With the growing interest in studying synchronized BOLD fluctuations during the resting state, the extent to which the apparent synchrony is due to variations in the end-tidal pressure of CO(2) (PETCO(2)) is an important consideration. CO(2)-related fluctuations in BOLD signal may also represent a potential confound when studying task-related responses, especially if breathing depth and rate are affected by the task. While previous studies of the above issues have explored retrospective correction of BOLD fluctuations related to arterial PCO(2), here we demonstrate an alternative approach based on physiological clamping of the arterial CO(2) level to a near-constant value. We present data comparing resting-state functional connectivity within the default-mode-network (DMN), as well as task-related BOLD responses, acquired in two conditions in each subject: 1) while subject's PETCO(2) was allowed to vary spontaneously; and 2) while controlling subject's PETCO(2) within a narrow range. Strong task-related responses and areas of maximal signal correlation in the DMN were not significantly altered by suppressing fluctuations in PETCO(2). Controlling PETCO(2) did, however, improve the performance of retrospective physiological noise correction techniques, allowing detection of additional regions of task-related response and resting-state connectivity in highly vascularized regions such as occipital cortex. While these results serve to further rule out systemic physiological fluctuations as a significant source of apparent resting-state network connectivity, they also demonstrate that fluctuations in arterial CO(2) are one of the factors limiting sensitivity in task-based and resting-state fMRI, particularly in regions of high vascular density. This must be considered when comparing subject groups who might exhibit differences in respiratory physiology or breathing patterns., (Copyright © 2012 Elsevier Inc. All rights reserved.)

A variety of methods have been developed to identify brain networks with spontaneous, coherent activity in resting-state functional magnetic resonance imaging (fMRI). We propose here a generic statistical framework to quantify the stability of such resting-state networks (RSNs), which was implemented with k-means clustering. The core of the method consists in bootstrapping the available datasets to replicate the clustering process a large number of times and quantify the stable features across all replications. This bootstrap analysis of stable clusters (BASC) has several benefits: (1) it can be implemented in a multi-level fashion to investigate stable RSNs at the level of individual subjects and at the level of a group; (2) it provides a principled measure of RSN stability; and (3) the maximization of the stability measure can be used as a natural criterion to select the number of RSNs. A simulation study validated the good performance of the multi-level BASC on purely synthetic data. Stable networks were also derived from a real resting-state study for 43 subjects. At the group level, seven RSNs were identified which exhibited a good agreement with the previous findings from the literature. The comparison between the individual and group-level stability maps demonstrated the capacity of BASC to establish successful correspondences between these two levels of analysis and at the same time retain some interesting subject-specific characteristics, e.g. the specific involvement of subcortical regions in the visual and fronto-parietal networks for some subjects., (Copyright (c) 2010 Elsevier Inc. All rights reserved.)

Cognition is thought to result from interactions within large-scale networks of brain regions. Here, we propose a method to identify these large-scale networks using functional magnetic resonance imaging (fMRI). Regions belonging to such networks are defined as sets of strongly interacting regions, each of which showing a homogeneous temporal activity. Our method of large-scale network identification (LSNI) proceeds by first detecting functionally homogeneous regions. The networks of functional interconnections are then found by comparing the correlations among these regions against a model of the correlations in the noise. To test the LSNI method, we first evaluated its specificity and sensitivity on synthetic data sets. Then, the method was applied to four real data sets with a block-designed motor task. The LSNI method correctly recovered the regions whose temporal activity was locked to the stimulus. In addition, it detected two other main networks highly reproducible across subjects, whose activity was dominated by slow fluctuations (0-0.1 Hz). One was located in medial and dorsal regions, and mostly overlapped the "default" network of the brain at rest [Greicius, M.D., Krasnow, B., Reiss, A.L., Menon, V., 2003. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis. Proceedings of the National Academy of Sciences of the U.S.A. 100, 253-258]; the other was composed of lateral frontal and posterior parietal regions. The LSNI method we propose allows to detect in an exploratory and systematic way all the regions and large-scale networks activated in the working brain.