RESEARCH LINES
Inferential and Predictive Models in Functional Connectivity
Dynamic Connectivity Models
Robust Denoising and Quality Control Strategies in fMRI
Intersubject Heterogeneity in Anatomy and Function
Embedded Files
Inferential and Predictive Models in Functional Connectivity
Functional MRI (fMRI) allows researchers to estimate patterns of synchronization between brain areas, known as functional connectivity, producing a detailed picture of the brain’s connectome. However, this richness comes at a cost. A typical study may involve millions or even billions of connections, while only collecting data from a relatively small number of participants. Traditional statistical approaches struggle in this setting and often fail to detect anything but the strongest effects.
This research line focuses on the development and application of whole-connectome inferential and predictive models that allow researchers to study brain-behavior relationships in a way that respects the fundamentally distributed nature of the brain's connectome.
Functional Connectivity Multivariate Pattern Analysis (fc-MVPA) is a novel method developed by the lab to support population-level inferences, individual-difference analyses, and decoding, that overcomes the limitations of traditional statistical approaches by shifting the focus from individual connections to patterns of connectivity.
Rather than testing each connection separately, the method examines how each location in the brain connects to the rest of the brain as a whole. For every brain location, fc-MVPA treats its full connectivity profile as a meaningful pattern and asks whether the shape of that pattern differs across individuals, groups, or experimental conditions. This perspective allows researchers to detect subtle but biologically meaningful differences that are distributed across many connections, even when no single connection stands out on its own.
By combining multivariate pattern analysis with classical statistical inference, fc-MVPA enables researchers to study the human connectome at full brain scale without sacrificing interpretability or statistical validity. fc-MVPA also provides a model-free low-dimensional description of how connectivity patterns vary across individuals, offering new insights into the diversity of brain organization and providing compact representations that can be used in predictive models for decoding and individual-level prediction based on whole-brain connectivity patterns.
More generally, fc-MVPA reflects a shift toward understanding brain function not from isolated functional connections, but from coordinated patterns of connectivity spanning the entire brain.
For more information see:
Nieto-Castanon, A. (2022). Brain-wide connectome inferences using functional connectivity MultiVariate Pattern Analyses (fc-MVPA). PLoS computational biology, 18(11), e1010634. https://doi.org/10.1371/journal.pcbi.1010634
Arnold Anteraper, S., Guell, X., D'Mello, A., Joshi, N., Whitfield-Gabrieli, S., & Joshi, G. (2019). Disrupted cerebrocerebellar intrinsic functional connectivity in young adults with high-functioning autism spectrum disorder: a data-driven, whole-brain, high-temporal resolution functional magnetic resonance imaging study. Brain connectivity, 9(1), 48-59. https://doi.org/10.1089/brain.2018.0581
Westfall, D. R., Anteraper, S. A., Chaddock-Heyman, L., Drollette, E. S., Raine, L. B., Whitfield-Gabrieli, S., ... & Hillman, C. H. (2020). Resting-state functional connectivity and scholastic performance in preadolescent children: a data-driven multivoxel pattern analysis (mvpa). Journal of Clinical Medicine, 9(10), 3198. https://doi.org/10.3390/jcm9103198
Morris, T. P., Chaddock-Heyman, L., Ai, M., Anteraper, S. A., Nieto-Castañon, A. N., Whitfield-Gabrieli, S., ... & Kramer, A. F. (2021). Enriching activities during childhood are associated with variations in functional connectivity patterns later in life. Neurobiology of aging, 104, 92-101. https://doi.org/10.1016/j.neurobiolaging.2021.04.002
Dynamic Connectivity Models
Dynamic connectivity research in our lab asks how the brain’s large-scale communication patterns reconfigure over time as people adapt to external or internal factors, shift attention, learn and develop. The goal is to characterize not only whether the functional connectome changes, but how those changes are organized: are they best understood as a sequence of discrete whole-brain states, or as the superposition of independent changes unfolding concurrently across different circuits and networks? This distinction matters because many popular dFC methods enforce a “one-state-at-a-time” representation, which can create the appearance of global state switching even when brain dynamics may be more mixed and multi-process in nature.
Dynamic Independent Component Analysis (dynamic ICA) is a superposition-capable method developed by the lab for characterizing time-varying connectivity. Rather than assigning each moment to a single global state, dynamic ICA represents dFC as the overlapping expression of multiple latent circuits, each circuit formed by a group of connections whose strengths fluctuate jointly over time. Applied to fMRI data, dynamic ICA yields a circuit-based description of brain dynamics: the functional connectome is decomposed into interrelated circuits, and changes in the functional organization are described by the temporal dynamics of each circuit and their coordination.
Together, this research line motivates a shift from “global state sequences” toward a circuit-based view of dynamic connectivity, in which the key questions become how multiple latent processes coexist, when they become coordinated, and how health, cognition, and clinical conditions reshape that coordination architecture.
For more information see:
Nieto-Castanon, A. (in prep). The Illusion of Brain States: Chimera fMRI Challenges Global Accounts of Brain Dynamics.
Anteraper, S. A., Nieto-Castanon, A., & Whitfield-Gabrieli, S. (2020). Functional MRI methods. In Neuroimaging in Schizophrenia (pp. 93-112). Cham: Springer International Publishing. https://doi.org/10.1007/978-3-030-35206-6_5
Nieto-Castanon, A. (2020). Handbook of functional connectivity magnetic resonance imaging methods in CONN. Hilbert Press. https://doi.org/10.56441/hilbertpress.2207.6598
Denoising and Quality Control Strategies in fMRI
This line of work starts from a simple (but often underappreciated) observation: in functional connectivity fMRI, many “noise” sources, head motion, respiration, cardiac effects, scanner instabilities, produce correlated fluctuations across large portions of the brain, which can directly bias connectivity estimates and threaten interpretability and replicability. Because these confounds can masquerade as brain-wide coupling, the lab’s strategy emphasizes denoising plus explicit quality control (QC) as a unified methodological program, rather than treating preprocessing as a box to check.
A central methodological contribution of the lab in the context of fcMRI denoising was the proposal of an anatomical component-based correction (aCompCor) as a principled alternative to global signal regression. aCompCor combined with motion modeling and band-pass filtering was shown to improve sensitivity/specificity for positive correlations while avoiding some of the interpretational pitfalls of global signal regression (which can “overshoot” and induce artifactual negatives).
The broader denoising framework developed by the lab generalizes this idea into a modular set of choices that can be tuned to each dataset and each scientific question. This framework combines component-based noise correction, motion regression, and scrubbing, in which individual time points identified as outliers due to excessive motion or signal intensity changes are modeled as nuisance regressors rather than outright removed. These steps are integrated in a single linear regression and combined with temporal filtering, typically band-pass filtering, to focus analyses on frequency ranges most relevant to neural fluctuations while suppressing slow drifts and high-frequency noise.
In this program, QC is not an afterthought; it is how researchers can optimize their denoising strategy to their own datasets, and ultimately verify that denoising achieved its intended goal and that any remaining data limitations are understood and reported. Some of the QC procedures developed by the lab include the computation of a Data Validity score, providing quantitative assessments of global distributional biases in functional connectivity, a Data Quality score, measuring the extent of motion-related and other confounds in functional connectivity measures, and a Data Sensitivity score, characterizing the precision of subject- and group-level estimates of functional connectivity strength. These scores can be used to evaluate the quality and reliability of functional connectivity estimates derived from the data, prompting revised denoising strategies or stricter data exclusion when necessary.
Denoising and quality control are essential safeguards that determine the validity of downstream connectivity analyses. This research program provides practical tools and conceptual guidance for identifying and mitigating confounds, thereby continuing to strengthen confidence in functional connectivity as a scientific measure.
For more information see:
Nieto-Castanon, A. (2025). Preparing fMRI data for statistical analysis. In fMRI techniques and protocols (pp. 163-191). New York, NY: Springer US. https://doi.org/10.1007/978-1-0716-4438-6_6
Morfini, F., Whitfield-Gabrieli, S., & Nieto-Castañón, A. (2023). Functional connectivity MRI quality control procedures in CONN. Frontiers in Neuroscience, 17, 1092125. https://doi.org/10.3389/fnins.2023.1092125
Nieto-Castanon, A. (2020). Handbook of functional connectivity magnetic resonance imaging methods in CONN. Hilbert Press. https://doi.org/10.56441/hilbertpress.2207.6598
Chai, X. J., Nieto-Castañón, A, Öngür, D., & Whitfield-Gabrieli, S. (2012). Anticorrelations in resting state networks without global signal regression. Neuroimage, 59(2), 1420-1428. https://doi.org/10.1016/j.neuroimage.2011.08.048
Intersubject Heterogeneity in Anatomy and Function
How can we meaningfully compare brain function across individuals in the presence of substantial intersubject heterogeneity? Intersubject heterogeneity arises both in the anatomical domain, where the size, shape, and precise location of cortical regions vary markedly across people, and in the functional domain, where the mapping between brain regions and cognitive operations is not perfectly aligned with anatomy alone. Subject-specific ROIs (Regions Of Interest) provide a natural framework for addressing this challenge.
One strategy focuses on addressing anatomical heterogeneity by defining regions of interest separately in each subject based on identifiable anatomical landmarks such as sulci and gyri. This approach starts from the observation that standard whole-brain normalization procedures leave substantial residual anatomical variability, even after nonlinear inter-subject alignment. To address this, the proposed framework defines subject-specific anatomical ROIs and performs statistical analyses directly at the level of each region, treating the ROI, not individual voxels, as the basic unit of inference. Functional responses within each ROI are analyzed using multivariate techniques, preserving within-region structure while avoiding spatial smoothing across potentially non-homologous areas.
This anatomically grounded ROI strategy reframes the problem of intersubject alignment: instead of asking whether individual voxels correspond across brains, it asks whether entire regions defined by shared anatomical constraints exhibit consistent functional properties across individuals.
A complementary strategy addresses functional heterogeneity by defining subject-specific ROIs not anatomically but functionally, using functional localizer tasks. In this approach, regions are identified independently in each subject using tasks designed to selectively engage specific cognitive processes. These functionally defined ROIs are then used as the basis for subsequent analyses across conditions or subjects. This strategy recognizes that functionally equivalent regions may not occupy identical anatomical locations across individuals, particularly in higher-order cortical areas. By localizing regions based on functional response properties rather than spatial coordinates, this approach aligns brain data across subjects at the level of functional identity, rather than anatomy.
Whole-brain voxel-based analyses
Another related method developed by the lab focuses on the use of functional localizers in the context of whole-brain voxel-based analyses, by explicitly separating voxel selection from effect estimation within each subject. In this framework, each subject’s data are divided into two independent sets. One subset of the data is used solely to identify a functional mask, defining the voxels that show a response consistent with a predefined contrast of interest in that individual. The second, independent subset of the data is then used to estimate the magnitude and direction of the effect within the voxels selected by that mask. This separation ensures that voxel selection does not bias the estimation of task-related effects.
Critically, this approach allows voxel-level analyses to accommodate extreme intersubject variability in the precise spatial location of functional responses. Within a larger anatomical parcel or functional territory, different subjects may express task-related activations in non-overlapping voxel sets, and some voxels within the same parcel may even show responses with opposite signs across subjects. Standard strategies such as spatial smoothing or averaging signals across an entire ROI would fail in this extreme context as they implicitly assume a consistent spatial layout of effects across voxels and would cause true effects to cancel out when aggregated within a subject.
By contrast, the localizer-based split-data approach ensures that, for each subject, only voxels that show the appropriate functional response are carried forward to the estimation stage. As a result, even when the spatial pattern of activation differs dramatically across individuals, the analysis can still detect that there exist voxels within each subject that respond positively and consistently to the task. Population-level inference is then performed by combining these subject-level estimates across participants, rather than by enforcing voxelwise correspondence at fixed spatial coordinates. This strategy provides a principled alternative for studying functional specialization in brain regions where intersubject variability is the rule rather than the exception.
Together, these ROI-level frameworks articulate a coherent research program centered on understanding and leveraging interindividual variability rather than treating it as noise. Anatomically defined ROIs provide a principled way to control for structural variability while preserving regional specificity, whereas functionally defined ROIs allow analyses to adapt to individual differences in functional organization, even in extreme cases where fine-scale representations (such as columnar or patchy cortical organization) are not expected to spatially align across subjects. Both approaches share a common goal: enabling robust, interpretable, and sensitive inference at the level of brain regions while explicitly acknowledging that human brain organization is heterogeneous across individuals.
For more information see:
Nieto-Castanon, A., Ghosh, S. S., Tourville, J. A., & Guenther, F. H. (2003). Region of interest based analysis of functional imaging data. Neuroimage, 19(4), 1303-1316. https://doi.org/10.1016/S1053-8119(03)00188-5
Nieto-Castañón, A., & Fedorenko, E. (2012). Subject-specific functional localizers increase sensitivity and functional resolution of multi-subject analyses. Neuroimage, 63(3), 1646-1669. https://doi.org/10.1016/j.neuroimage.2012.06.065
Fedorenko, E., Hsieh, P. J., Nieto-Castañón, A., Whitfield-Gabrieli, S., & Kanwisher, N. (2010). New method for fMRI investigations of language: defining ROIs functionally in individual subjects. Journal of neurophysiology, 104(2), 1177-1194. https://doi.org/10.1152/jn.00032.2010
Google Sites
Report abuse