Biology subjects

Mumford, J. A.

Publications and source records attributed to Mumford, J. A..

9 recordsLinked to original sources

bioRxiv2025

Refining RDoC Using Individual-Level Task fMRI Factor Models Reveals Reproducible Brain-wide Motifs

The Research Domain Criteria (RDoC) framework was introduced to guide psychiatric research using biologically grounded, dimensional constructs of mental function. However, its hierarchical domain structure remains largely unvalidated against individual-level brain and behavioral data. Building on prior group-level work, we applied a multi-stage validation framework to Human Connectome Project (HCP) task-fMRI data to test whether individual-level, data-driven models more accurately capture the organization of brain activity and behavior than RDoC-based models. Using confirmatory factor analysis in two independent cohorts, we found that data-driven bifactor models consistently outperformed RDoC-based models across multiple fit indices. The general factor derived from these models revealed a reproducible, low-dimensional axis spanning visual-attentional to default mode-auditory systems, aligning with canonical macroscale cortical gradients. Community detection further identified reproducible spatial motifs whose centroids corresponded to interpretable functional systems and whose alignment predicted individual performance on working memory and relational reasoning tasks. To assess whether these findings extended beyond neural data, we analyzed behavioral measures in HCP and in an independent transdiagnostic dataset (LA5c). In both datasets, data-driven behavioral models outperformed RDoC-based models, although the relative support for bifactor versus specific factor structure differed by dataset. Extending the neural analyses to LA5c, which included healthy controls and individuals with ADHD, bipolar disorder, and schizophrenia, showed that data-driven bifactor models generalized across diagnostic groups and that alignment with data-driven community centroids related to symptom severity, whereas RDoC-based representations showed weaker or no associations. Finally, topological analysis of task-evoked brain activity revealed that data-driven representations better captured the global organization of brain states than RDoC domains. Together, these findings demonstrate that individual-level, empirically derived models provide a more accurate, generalizable, and behaviorally relevant account of brain organization than the current RDoC framework. By integrating neural, behavioral, and clinical validation, this work advances precision neuroscience and supports the empirical refinement of dimensional psychiatric frameworks.

neuroscience↗

bioRxiv2025

Unintended bias in the pursuit of collinearity solutions in fMRI analysis

In task functional magnetic resonance imaging (fMRI), collinearity between task regressors in time series models may impact power. When collinearity is identified after data collection, researchers often modify the model in an effort to reduce collinearity. However, some model adjustments are suboptimal and may introduce bias into parameter estimates. Although relevant to many task-fMRI studies, we highlight these issues using the Monetary Incentive Delay (MID) task data from the Adolescent Brain Cognitive Development (ABCD(R)) study. We introduce a procedure to more directly quantify the impact of collinearity on task-relevant measures: a contrast-based variance inflation factor (cVIF). We also show that collinearity reduction strategies--such as omitting regressors for specific task components, using impulse regressors for extended activations, and ignoring response time variability--can bias contrast estimates. Finally, we present a "Saturated" model that includes all task components, including response times, aiming to reduce these biases while maintaining comparable levels of collinearity, as assessed by cVIF.

neuroscience↗

bioRxiv2024

Predicting Task Activation Maps from Resting-State Functional Connectivity using Deep Learning

Deep learning models have demonstrated the potential to predict task-evoked brain activation from resting-state fMRI, offering a pathway toward individualized brain mapping without requiring task-based data. In this study, we systematically evaluate architectural strategies for improving the efficiency and scalability of such models. Using data from the Human Connectome Project, we replicate the BrainSurfCNN framework and introduce two extensions: BrainSERF, which incorporates channel-wise attention through squeeze-and-excitation modules, and BrainSurfGCN, a graph-based model that leverages cortical mesh topology for efficient message passing. Across multiple evaluation metrics, including spatial correlation, Dice score, Dice AUC, and subject identification accuracy, all models achieve comparable predictive performance. Despite similar accuracy, the proposed models offer distinct advantages. BrainSERF provides modest improvements in capturing individual-specific features, while BrainSurfGCN achieves substantial reductions in model size and training time, highlighting a favorable trade-off between performance and computational efficiency. Beyond architectural comparisons, we investigate factors driving variability in prediction accuracy. We find that behavioral task performance, resting-state data quality, and inter-subject variability in task activation jointly constrain prediction fidelity. In particular, contrasts with lower signal reliability and higher variability exhibit reduced predictability across all models. Together, these findings demonstrate that incorporating topological and functional structural priors can improve the efficiency of deep learning models without sacrificing accuracy, while also emphasizing that prediction performance is fundamentally limited by the reliability of the underlying neural signals.

neuroscience↗

bioRxiv2024

Impact of analytic decisions on test-retest reliability of individual and group estimates in functional magnetic resonance imaging: a multiverse analysis using the monetary incentive delay task

Empirical studies reporting low test-retest reliability of individual blood oxygen-level dependent (BOLD) signal estimates in functional magnetic resonance imaging (fMRI) data have resurrected interest among cognitive neuroscientists in methods that may improve reliability in fMRI. Over the last decade, several individual studies have reported that modeling decisions, such as smoothing, motion correction and contrast selection, may improve estimates of test-retest reliability of BOLD signal estimates. However, it remains an empirical question whether certain analytic decisions consistently improve individual and group level reliability estimates in an fMRI task across multiple large, independent samples. This study used three independent samples (Ns: 60, 81, 119) that collected the same task (Monetary Incentive Delay task) across two runs and two sessions to evaluate the effects of analytic decisions on the individual (intraclass correlation coefficient [ICC(3,1)]) and group (Jaccard/Spearman rho) reliability estimates of BOLD activity of task fMRI data. The analytic decisions in this study vary across four categories: smoothing kernel (five options), motion correction (four options), task parameterizing (three options) and task contrasts (four options), totaling 240 different pipeline permutations. Across all 240 pipelines, the median ICC estimates are consistently low, with a maximum median ICC estimate of .43 - .55 across the three samples. The analytic decisions with the greatest impact on the median ICC and group similarity estimates are the Implicit Baseline contrast, Cue Model parameterization and a larger smoothing kernel. Using an Implicit Baseline in a contrast condition meaningfully increased group similarity and ICC estimates as compared to using the Neutral cue. This effect was largest for the Cue Model parameterization; however, improvements in reliability came at the cost of interpretability. This study illustrates that estimates of reliability in the MID task are consistently low and variable at small samples, and a higher test-retest reliability may not always improve interpretability of the estimated BOLD signal.

neuroscience↗

bioRxiv2024

A Data-Driven Latent Variable Approach to Validating the Research Domain Criteria (RDoC) Framework

Despite the widespread use of the Research Domain Criteria (RDoC) framework in psychiatry and neuroscience, recent studies suggest that the RDoC is insufficiently specific or excessively broad relative to the underlying brain circuitry it seeks to elucidate. To address these concerns, we employed a latent variable approach using bifactor analysis. We examined 84 whole-brain task-based fMRI (tfMRI) activation maps from 19 studies with 6,192 participants. A curated subset of 37 maps with a balanced representation of RDoC domains constituted the training set, and the remaining held-out maps formed the internal validation set. External validation was conducted using 36 peak coordinate activation maps from Neurosynth, using terms of RDoC constructs as seeds for topic meta-analysis. Here, we show that a bifactor model incorporating a task-general domain and splitting the cognitive systems domain better fits the examined corpus of tfMRI data than the current RDoC framework. We also identify the domain of arousal and regulatory systems as underrepresented. Our data-driven validation supports revising the RDoC framework to reflect underlying brain circuitry more accurately.

neuroscience↗

bioRxiv2023

Cognitive tasks, anatomical MRI, and functional MRI data evaluating the construct of self-regulation

We describe the following shared data from N=103 healthy adults who completed a broad set cognitive tasks, surveys, and neuroimaging measurements to examine the construct of self-regulation. The neuroimaging acquisition involved task-based fMRI, resting fMRI, and structural MRI. Each subject completed the following ten tasks in the scanner across two 90- minute scanning sessions: attention network test (ANT), cued task switching, Columbia card task, dot pattern expectancy (DPX), delay discounting, simple and motor selective stop signal, Stroop, a towers task, and a set of survey questions. Subjects also completed resting state scans. The dataset is shared openly through the OpenNeuro project, and the dataset is formatted according to the Brain Imaging Data Structure (BIDS) standard.

neuroscience↗

bioRxiv2023

When multi-voxel pattern similarity and global activation are intertwined: Approaches to disentangling correlation from activation

Pattern similarity analysis, which uses correlation to examine similarities between neural activation patterns evoked by different trials or conditions, is often leveraged to test hypotheses not easily answerable with univariate comparisons, such as how events are represented or processed and the relationships between representations or processing of events. In principle, univariate analyses of global activation and multivariate analyses of pattern similarity can be used to answer substantively different questions about psychological and neural processing. For this to hold, it is necessary that pattern similarity estimates are not contaminated by differences in global activation across experimental events. Here, we report simulated data that demonstrate that global activation and pattern similarity (as assessed by correlation), although theoretically independent, are often intertwined. We present two plausible scenarios that illustrate how condition-specific changes in global activation can elicit condition-specific increases in pattern similarity by interacting with underlying across-voxel activation patterns. First, we consider a scenario in which a target region contains subpopulations of voxels such that only some voxels in a region are sensitive to a psychological variable and the remaining voxels are not modulated by this variable. In this scenario, this spatial pattern of responsive and unresponsive voxels adds new, shared across-voxel variability for events in the active condition, thereby increasing pattern similarity between these events. Second, we consider a scenario in which trials from all conditions elicit a shared across-voxel pattern of activation, but this shared across-voxel pattern is amplified for trials within one condition due to greater global activation. In this scenario, the change in activation for a given condition increases the ability to detect pre-existing, shared across-voxel variability across events in that condition, thereby increasing pattern similarity between these events. Given the observed influence of global activation on pattern similarity, we then assess whether it is possible to statistically separate the contributions of global activation and pattern similarity to observed activation patterns (using regression approaches, matching activation across conditions, and inclusion of control conditions). Additional simulations demonstrate that use of these techniques is not always effective in removing the influence of global activation on pattern similarity --the efficacy of these techniques depends on a variety of signal parameters that will likely vary across experiments and participants, highlighting the need for tailored control analyses that are targeted at addressing the particular hypotheses and potential global activation confounds of a given experiment. [Note: The reported simulations and this resulting white paper were generated in 2014. We share, without update, this paper given the continued relevance of understanding and controlling for global activation confounds when conducting multi-variate pattern analyses.]

neuroscience↗

bioRxiv2023

The response time paradox in functional magnetic resonance imaging analyses

The functional MRI (fMRI) signal is a proxy for an unobservable neuronal signal, and differences in fMRI signals on cognitive tasks are generally interpreted as reflecting differences in the intensity of local neuronal activity. However, changes in either intensity or duration of neuronal activity can yield identical differences in fMRI signals. When conditions differ in response times (RTs), it is thus impossible to determine whether condition differences in fMRI signals are due to differences in the intensity of neuronal activity or to potentially spurious differences in the duration of neuronal activity. The most common fMRI analysis approach ignores RTs, making it difficult to interpret condition differences that could be driven by RTs and/or intensity. Because differences in response time are one of the most important signals of interest for cognitive psychology, nearly every task of interest for fMRI exhibits RT differences across conditions of interest. This results in a paradox, wherein the signal of interest for the psychologist is a potential confound for the fMRI researcher. We review this longstanding problem, and demonstrate that the failure to address RTs in the fMRI time series model can also lead to spurious correlations at the group level related to RTs or other variables of interest, potentially impacting the interpretation of brain-behavior correlations. We propose a simple approach that remedies this problem by including RT in the fMRI time series model. This model separates condition differences from RT differences, retaining power for detection of unconfounded condition differences while also allowing the identification of RT-related activation. We conclude by highlighting the need for further theoretical development regarding the interpretation of fMRI signals and their relationship to response times.

neuroscience↗

bioRxiv2017

Behavioral and neural indices of affective coloring for neutral social stimuli

Emotional processing often continues beyond the presentation of emotionally evocative stimuli, which can result in affective biasing or coloring of subsequently encountered events. Here, we describe neural correlates of affective coloring and examine how individual differences in affective style impact the magnitude of affective coloring. We conducted functional magnetic resonance imaging in 117 adults who passively viewed negative, neutral, and positive pictures presented 2s prior to neutral faces. Brain responses to neutral faces were modulated by the valence of preceding pictures, with greater activation for faces following negative (vs. positive) pictures in the amygdala, dorsomedial and lateral prefrontal cortex, ventral visual cortices, posterior superior temporal sulcus, and angular gyrus. Three days after the MRI scan, participants rated their memory and liking of previously encountered neutral faces. Individuals higher in trait positive affect and emotional reappraisal rated faces as more likable when preceded by emotionally arousing (negative or positive) pictures. Additionally, greater amygdala responses to neutral faces preceded by positively valenced pictures were associated with greater memory for these faces three days later. Collectively, these results reveal individual differences in how emotions spill over onto the processing of unrelated social stimuli, resulting in persistent and affectively biased evaluations of such stimuli.

neuroscience↗