A Unifying Model for Discordant and Concordant Results in Human Neuroimaging Studies of Facial Viewpoint Selectivity.

Recognizing faces regardless of their viewpoint is critical for social interactions. Traditional theories hold that view-selective early visual representations gradually become tolerant to viewpoint changes along the ventral visual hierarchy. Newer theories, based on single-neuron monkey electrophysiological recordings, suggest a three-stage architecture including an intermediate face-selective patch abruptly achieving invariance to mirror-symmetric face views. Human studies combining neuroimaging and multivariate pattern analysis (MVPA) have provided convergent evidence of view selectivity in early visual areas. However, contradictory conclusions have been reached concerning the existence in humans of a mirror-symmetric representation like that observed in macaques. We believe these contradictions arise from low-level stimulus confounds and data analysis choices. To probe for low-level confounds, we analyzed images from two face databases. Analyses of image luminance and contrast revealed biases across face views described by even polynomials-i.e., mirror-symmetric. To explain major trends across neuroimaging studies, we constructed a network model incorporating three constraints: cortical magnification, convergent feedforward projections, and interhemispheric connections. Given the identified low-level biases, we show that a gradual increase of interhemispheric connections across network-layers is sufficient to replicate view-tuning in early processing stages and mirror-symmetry in later stages. Data analysis decisions-pattern dissimilarity measure and data recentering-accounted for the inconsistent observation of mirror-symmetry across prior studies. Pattern analyses of human fMRI data (of either sex) revealed biases compatible with our model. The model provides a unifying explanation of MVPA studies of viewpoint selectivity and suggests observations of mirror-symmetry originate from ineffectively normalized signal imbalances across different face views.

Cross-species real time detection of trends in pupil size fluctuation.

Pupillometry is a popular method because pupil size is an easily measured and sensitive marker of neural activity and associated with behavior, cognition, emotion, and perception. Currently, there is no method for monitoring the phases of pupillary fluctuation in real time. We introduce rtPupilPhase - a software that automatically detects trends in pupil size in real time, enabling novel implementations of real time pupillometry towards achieving numerous research and translational goals.

Visual imagery vividness correlates with afterimage conscious perception.

Afterimages are illusory, visual conscious perceptions. A widely accepted theory is that afterimages are caused by retinal signaling that continues after the physical disappearance of a light stimulus. However, afterimages have been reported without preceding visual, sensory stimulation (e.g., conditioned afterimages and afterimages induced by illusory vision). These observations suggest the role of top-down, brain mechanisms in afterimage conscious perception. Therefore, some afterimages may share perceptual features with sensory-independent conscious perceptions (e.g., imagery, hallucinations, and dreams) that occur without bottom-up, sensory input. In the current investigation, we tested for a link between the vividness of visual imagery and afterimage conscious perception. Participants reported their vividness of visual imagery and perceived sharpness, contrast, and duration of negative afterimages. The afterimage perceptual features were acquired using perception matching paradigms that were validated on image stimuli. Relating these perceptual reports revealed that the vividness of visual imagery positively correlated with afterimage contrast and sharpness. These behavioral results support shared neural mechanisms between visual imagery and afterimages. This study encourages future research combining neurophysiology recording methods and afterimage paradigms to directly examine the neural mechanisms of afterimage conscious perception.

Whole-brain multivariate hemodynamic deconvolution for functional MRI with stability selection.

Conventionally, analysis of functional MRI (fMRI) data relies on available information about the experimental paradigm to establish hypothesized models of brain activity. However, this information can be inaccurate, incomplete or unavailable in multiple scenarios such as resting-state, naturalistic paradigms or clinical conditions. In these cases, blind estimates of neuronal-related activity can be obtained with paradigm-free analysis methods such as hemodynamic deconvolution. Yet, current formulations of the hemodynamic deconvolution problem have three important limitations: (1) their efficacy strongly depends on the appropriate selection of regularization parameters, (2) being univariate, they do not take advantage of the information present across the brain, and (3) they do not provide any measure of statistical certainty associated with each detected event. Here we propose a novel approach that addresses all these limitations. Specifically, we introduce multivariate sparse paradigm free mapping (Mv-SPFM), a novel hemodynamic deconvolution algorithm that operates at the whole brain level and adds spatial information via a mixed-norm regularization term over all voxels. Additionally, Mv-SPFM employs a stability selection procedure that removes the need to select regularization parameters and also lets us obtain an estimate of the true probability of having a neuronal-related BOLD event at each voxel and time-point based on the area under the curve (AUC) of the stability paths. Besides, we present a formulation tailored for multi-echo fMRI acquisitions (MvME-SPFM), which allows us to better isolate fluctuations of BOLD origin on the basis of their linear dependence with the echo time (TE) and to assign physiologically interpretable units (i.e., changes in the apparent transverse relaxation ΔR2∗) to the resulting deconvolved events. Remarkably, we demonstrate that Mv-SPFM achieves comparable performance even when using a single-echo formulation. We demonstrate that this algorithm outperforms existing state-of-the-art deconvolution approaches, and shows higher spatial and temporal agreement with the activation maps and BOLD signals obtained with a standard model-based linear regression approach, even at the level of individual neuronal events. Furthermore, we show that by employing stability selection, the performance of the algorithm depends less on the selection of temporal and spatial regularization parameters λ and ρ. Consequently, the proposed algorithm provides more reliable estimates of neuronal-related activity, here in terms of ΔR2∗, for the study of the dynamics of brain activity when no information about the timings of the BOLD events is available. This algorithm will be made publicly available as part of the splora Python package.

Manifold learning for fMRI time-varying functional connectivity.

Whole-brain functional connectivity (FC) measured with functional MRI (fMRI) evolves over time in meaningful ways at temporal scales going from years (e.g., development) to seconds [e.g., within-scan time-varying FC (tvFC)]. Yet, our ability to explore tvFC is severely constrained by its large dimensionality (several thousands). To overcome this difficulty, researchers often seek to generate low dimensional representations (e.g., 2D and 3D scatter plots) hoping those will retain important aspects of the data (e.g., relationships to behavior and disease progression). Limited prior empirical work suggests that manifold learning techniques (MLTs)-namely those seeking to infer a low dimensional non-linear surface (i.e., the manifold) where most of the data lies-are good candidates for accomplishing this task. Here we explore this possibility in detail. First, we discuss why one should expect tvFC data to lie on a low dimensional manifold. Second, we estimate what is the intrinsic dimension (ID; i.e., minimum number of latent dimensions) of tvFC data manifolds. Third, we describe the inner workings of three state-of-the-art MLTs: Laplacian Eigenmaps (LEs), T-distributed Stochastic Neighbor Embedding (T-SNE), and Uniform Manifold Approximation and Projection (UMAP). For each method, we empirically evaluate its ability to generate neuro-biologically meaningful representations of tvFC data, as well as their robustness against hyper-parameter selection. Our results show that tvFC data has an ID that ranges between 4 and 26, and that ID varies significantly between rest and task states. We also show how all three methods can effectively capture subject identity and task being performed: UMAP and T-SNE can capture these two levels of detail concurrently, but LE could only capture one at a time. We observed substantial variability in embedding quality across MLTs, and within-MLT as a function of hyper-parameter selection. To help alleviate this issue, we provide heuristics that can inform future studies. Finally, we also demonstrate the importance of feature normalization when combining data across subjects and the role that temporal autocorrelation plays in the application of MLTs to tvFC data. Overall, we conclude that while MLTs can be useful to generate summary views of labeled tvFC data, their application to unlabeled data such as resting-state remains challenging.

Cortex-wide neural dynamics predict behavioral states and provide a neural basis for resting-state dynamic functional connectivity.

Although resting-state functional magnetic resonance imaging (fMRI) studies have observed dynamically changing brain-wide networks of correlated activity, fMRI's dependence on hemodynamic signals makes results challenging to interpret. Meanwhile, emerging techniques for real-time recording of large populations of neurons have revealed compelling fluctuations in neuronal activity across the brain that are obscured by traditional trial averaging. To reconcile these observations, we use wide-field optical mapping to simultaneously record pan-cortical neuronal and hemodynamic activity in awake, spontaneously behaving mice. Some components of observed neuronal activity clearly represent sensory and motor function. However, particularly during quiet rest, strongly fluctuating patterns of activity across diverse brain regions contribute greatly to interregional correlations. Dynamic changes in these correlations coincide with changes in arousal state. Simultaneously acquired hemodynamics depict similar brain-state-dependent correlation shifts. These results support a neural basis for dynamic resting-state fMRI, while highlighting the importance of brain-wide neuronal fluctuations in the study of brain state.

The art and science of using quality control to understand and improve fMRI data.

Designing and executing a good quality control (QC) process is vital to robust and reproducible science and is often taught through hands on training. As FMRI research trends toward studies with larger sample sizes and highly automated processing pipelines, the people who analyze data are often distinct from those who collect and preprocess the data. While there are good reasons for this trend, it also means that important information about how data were acquired, and their quality, may be missed by those working at later stages of these workflows. Similarly, an abundance of publicly available datasets, where people (not always correctly) assume others already validated data quality, makes it easier for trainees to advance in the field without learning how to identify problematic data. This manuscript is designed as an introduction for researchers who are already familiar with fMRI, but who did not get hands on QC training or who want to think more deeply about QC. This could be someone who has analyzed fMRI data but is planning to personally acquire data for the first time, or someone who regularly uses openly shared data and wants to learn how to better assess data quality. We describe why good QC processes are important, explain key priorities and steps for fMRI QC, and as part of the FMRI Open QC Project, we demonstrate some of these steps by using AFNI software and AFNI's QC reports on an openly shared dataset. A good QC process is context dependent and should address whether data have the potential to answer a scientific question, whether any variation in the data has the potential to skew or hide key results, and whether any problems can potentially be addressed through changes in acquisition or data processing. Automated metrics are essential and can often highlight a possible problem, but human interpretation at every stage of a study is vital for understanding causes and potential solutions.

Highlight results, don't hide them: Enhance interpretation, reduce biases and improve reproducibility.

Most neuroimaging studies display results that represent only a tiny fraction of the collected data. While it is conventional to present "only the significant results" to the reader, here we suggest that this practice has several negative consequences for both reproducibility and understanding. This practice hides away most of the results of the dataset and leads to problems of selection bias and irreproducibility, both of which have been recognized as major issues in neuroimaging studies recently. Opaque, all-or-nothing thresholding, even if well-intentioned, places undue influence on arbitrary filter values, hinders clear communication of scientific results, wastes data, is antithetical to good scientific practice, and leads to conceptual inconsistencies. It is also inconsistent with the properties of the acquired data and the underlying biology being studied. Instead of presenting only a few statistically significant locations and hiding away the remaining results, studies should "highlight" the former while also showing as much as possible of the rest. This is distinct from but complementary to utilizing data sharing repositories: the initial presentation of results has an enormous impact on the interpretation of a study. We present practical examples and extensions of this approach for voxelwise, regionwise and cross-study analyses using publicly available data that was analyzed previously by 70 teams (NARPS; Botvinik-Nezer, et al., 2020), showing that it is possible to balance the goals of displaying a full set of results with providing the reader reasonably concise and "digestible" findings. In particular, the highlighting approach sheds useful light on the kind of variability present among the NARPS teams' results, which is primarily a varied strength of agreement rather than disagreement. Using a meta-analysis built on the informative "highlighting" approach shows this relative agreement, while one using the standard "hiding" approach does not. We describe how this simple but powerful change in practice-focusing on highlighting results, rather than hiding all but the strongest ones-can help address many large concerns within the field, or at least to provide more complete information about them. We include a list of practical suggestions for results reporting to improve reproducibility, cross-study comparisons and meta-analyses.

A unifying model for discordant and concordant results in human neuroimaging studies of facial viewpoint selectivity.

Our ability to recognize faces regardless of viewpoint is a key property of the primate visual system. Traditional theories hold that facial viewpoint is represented by view-selective mechanisms at early visual processing stages and that representations become increasingly tolerant to viewpoint changes in higher-level visual areas. Newer theories, based on single-neuron monkey electrophysiological recordings, suggest an additional intermediate processing stage invariant to mirror-symmetric face views. Consistent with traditional theories, human studies combining neuroimaging and multivariate pattern analysis (MVPA) methods have provided evidence of view-selectivity in early visual cortex. However, contradictory results have been reported in higher-level visual areas concerning the existence in humans of mirror-symmetrically tuned representations. We believe these results reflect low-level stimulus confounds and data analysis choices. To probe for low-level confounds, we analyzed images from two popular face databases. Analyses of mean image luminance and contrast revealed biases across face views described by even polynomials-i.e., mirror-symmetric. To explain major trends across human neuroimaging studies of viewpoint selectivity, we constructed a network model that incorporates three biological constraints: cortical magnification, convergent feedforward projections, and interhemispheric connections. Given the identified low-level biases, we show that a gradual increase of interhemispheric connections across network layers is sufficient to replicate findings of mirror-symmetry in high-level processing stages, as well as view-tuning in early processing stages. Data analysis decisions-pattern dissimilarity measure and data recentering-accounted for the variable observation of mirror-symmetry in late processing stages. The model provides a unifying explanation of MVPA studies of viewpoint selectivity. We also show how common analysis choices can lead to erroneous conclusions.

Manifold Learning for fMRI time-varying FC.

Whole-brain functional connectivity ( FC ) measured with functional MRI (fMRI) evolve over time in meaningful ways at temporal scales going from years (e.g., development) to seconds (e.g., within-scan time-varying FC ( tvFC )). Yet, our ability to explore tvFC is severely constrained by its large dimensionality (several thousands). To overcome this difficulty, researchers seek to generate low dimensional representations (e.g., 2D and 3D scatter plots) expected to retain its most informative aspects (e.g., relationships to behavior, disease progression). Limited prior empirical work suggests that manifold learning techniques ( MLTs )-namely those seeking to infer a low dimensional non-linear surface (i.e., the manifold) where most of the data lies-are good candidates for accomplishing this task. Here we explore this possibility in detail. First, we discuss why one should expect tv FC data to lie on a low dimensional manifold. Second, we estimate what is the intrinsic dimension (i.e., minimum number of latent dimensions; ID ) of tvFC data manifolds. Third, we describe the inner workings of three state-of-the-art MLTs : Laplacian Eigenmaps ( LE ), T-distributed Stochastic Neighbor Embedding ( T-SNE ), and Uniform Manifold Approximation and Projection ( UMAP ). For each method, we empirically evaluate its ability to generate neuro-biologically meaningful representations of tvFC data, as well as their robustness against hyper-parameter selection. Our results show that tvFC data has an ID that ranges between 4 and 26, and that ID varies significantly between rest and task states. We also show how all three methods can effectively capture subject identity and task being performed: UMAP and T-SNE can capture these two levels of detail concurrently, but L E could only capture one at a time. We observed substantial variability in embedding quality across MLTs , and within- MLT as a function of hyper-parameter selection. To help alleviate this issue, we provide heuristics that can inform future studies. Finally, we also demonstrate the importance of feature normalization when combining data across subjects and the role that temporal autocorrelation plays in the application of MLTs to tvFC data. Overall, we conclude that while MLTs can be useful to generate summary views of labeled tvFC data, their application to unlabeled data such as resting-state remains challenging.

The challenge of BWAs: Unknown unknowns in feature space and variance.

The recent paper by Marek et al.1 has shown that, to capture brain-wide associations using fMRI and MRI measures, thousands of individuals are required. These results can be potentially misunderstood to imply that MRI or fMRI lack sensitivity or specificity. This commentary discusses the demonstrated sensitivity of fMRI and focuses on methodology that may allow improvements in BWA studies. While individual variation may be an ultimate constraint, refinements in acquisition, population selection, and processing may bring about higher correlations.

Ultra-slow fMRI fluctuations in the fourth ventricle as a marker of drowsiness.

Wakefulness levels modulate estimates of functional connectivity (FC), and, if unaccounted for, can become a substantial confound in resting-state fMRI. Unfortunately, wakefulness is rarely monitored due to the need for additional concurrent recordings (e.g., eye tracking, EEG). Recent work has shown that strong fluctuations around 0.05Hz, hypothesized to be CSF inflow, appear in the fourth ventricle (FV) when subjects fall asleep, and that they correlate significantly with the global signal. The analysis of these fluctuations could provide an easy way to evaluate wakefulness in fMRI-only data and improve our understanding of FC during sleep. Here we evaluate this possibility using the 7T resting-state sample from the Human Connectome Project (HCP). Our results replicate the observation that fourth ventricle ultra-slow fluctuations (∼0.05Hz) with inflow-like characteristics (decreasing in intensity for successive slices) are present in scans during which subjects did not comply with instructions to keep their eyes open (i.e., drowsy scans). This is true despite the HCP data not being optimized for the detection of inflow-like effects. In addition, time-locked BOLD fluctuations of the same frequency could be detected in large portions of grey matter with a wide range of temporal delays and contribute in significant ways to our understanding of how FC changes during sleep. First, these ultra-slow fluctuations explain half of the increase in global signal that occurs during descent into sleep. Similarly, global shifts in FC between awake and sleep states are driven by changes in this slow frequency band. Second, they can influence estimates of inter-regional FC. For example, disconnection between frontal and posterior components of the Defulat Mode Network (DMN) typically reported during sleep were only detectable after regression of these ultra-slow fluctuations. Finally, we report that the temporal evolution of the power spectrum of these ultra-slow FV fluctuations can help us reproduce sample-level sleep patterns (e.g., a substantial number of subjects descending into sleep 3 minutes following scanning onset), partially rank scans according to overall drowsiness levels, and predict individual segments of elevated drowsiness (at 60 seconds resolution) with 71% accuracy.

How to Interpret Resting-State fMRI: Ask Your Participants.

Resting-state fMRI (rsfMRI) reveals brain dynamics in a task-unconstrained environment as subjects let their minds wander freely. Consequently, resting subjects navigate a rich space of cognitive and perceptual states (i.e., ongoing experience). How this ongoing experience shapes rsfMRI summary metrics (e.g., functional connectivity) is unknown, yet likely to contribute uniquely to within- and between-subject differences. Here we argue that understanding the role of ongoing experience in rsfMRI requires access to standardized, temporally resolved, scientifically validated first-person descriptions of those experiences. We suggest best practices for obtaining those descriptions via introspective methods appropriately adapted for use in fMRI research. We conclude with a set of guidelines for fusing these two data types to answer pressing questions about the etiology of rsfMRI.

Theta-burst TMS to the posterior superior temporal sulcus decreases resting-state fMRI connectivity across the face processing network.

Humans process faces by using a network of face-selective regions distributed across the brain. Neuropsychological patient studies demonstrate that focal damage to nodes in this network can impair face recognition, but such patients are rare. We approximated the effects of damage to the face network in neurologically normal human participants by using theta burst transcranial magnetic stimulation (TBS). Multi-echo functional magnetic resonance imaging (fMRI) resting-state data were collected pre- and post-TBS delivery over the face-selective right superior temporal sulcus (rpSTS), or a control site in the right motor cortex. Results showed that TBS delivered over the rpSTS reduced resting-state connectivity across the extended face processing network. This connectivity reduction was observed not only between the rpSTS and other face-selective areas, but also between nonstimulated face-selective areas across the ventral, medial, and lateral brain surfaces (e.g., between the right amygdala and bilateral fusiform face areas and occipital face areas). TBS delivered over the motor cortex did not produce significant changes in resting-state connectivity across the face processing network. These results demonstrate that, even without task-induced fMRI signal changes, disrupting a single node in a brain network can decrease the functional connectivity between nodes in that network that have not been directly stimulated.

Language lateralization from task-based and resting state functional MRI in patients with epilepsy.

We compared resting state (RS) functional connectivity and task-based fMRI to lateralize language dominance in 30 epilepsy patients (mean age = 33; SD = 11; 12 female), a measure used for presurgical planning. Language laterality index (LI) was calculated from task fMRI in frontal, temporal, and frontal + temporal regional masks using LI bootstrap method from SPM12. RS language LI was assessed using two novel methods of calculating RS language LI from bilateral Broca's area seed based connectivity maps across regional masks and multiple thresholds (p < .05, p < .01, p < .001, top 10% connections). We compared LI from task and RS fMRI continuous values and dominance classifications. We found significant positive correlations between task LI and RS LI when functional connectivity thresholds were set to the top 10% of connections. Concordance of dominance classifications ranged from 20% to 30% for the intrahemispheric resting state LI method and 50% to 63% for the resting state LI intra- minus interhemispheric difference method. Approximately 40% of patients left dominant on task showed RS bilateral dominance. There was no difference in LI concordance between patients with right-sided and left-sided resections. Early seizure onset (<6 years old) was not associated with atypical language dominance during task-based or RS fMRI. While a relationship between task LI and RS LI exists in patients with epilepsy, language dominance is less lateralized on RS than task fMRI. Concordance of language dominance classifications between task and resting state fMRI depends on brain regions surveyed and RS LI calculation method.

Imaging the spontaneous flow of thought: Distinct periods of cognition contribute to dynamic functional connectivity during rest.

Brain functional connectivity (FC) changes have been measured across seconds using fMRI. This is true for both rest and task scenarios. Moreover, it is well accepted that task engagement alters FC, and that dynamic estimates of FC during and before task events can help predict their nature and performance. Yet, when it comes to dynamic FC (dFC) during rest, there is no consensus about its origin or significance. Some argue that rest dFC reflects fluctuations in on-going cognition, or is a manifestation of intrinsic brain maintenance mechanisms, which could have predictive clinical value. Conversely, others have concluded that rest dFC is mostly the result of sampling variability, head motion or fluctuating sleep states. Here, we present novel analyses suggesting that rest dFC is influenced by short periods of spontaneous cognitive-task-like processes, and that the cognitive nature of such mental processes can be inferred blindly from the data. As such, several different behaviorally relevant whole-brain FC configurations may occur during a single rest scan even when subjects were continuously awake and displayed minimal motion. In addition, using low dimensional embeddings as visualization aids, we show how FC states-commonly used to summarize and interpret resting dFC-can accurately and robustly reveal periods of externally imposed tasks; however, they may be less effective in capturing periods of distinct cognition during rest.

A deconvolution algorithm for multi-echo functional MRI: Multi-echo Sparse Paradigm Free Mapping.

This work introduces a novel algorithm for deconvolution of the BOLD signal in multi-echo fMRI data: Multi-echo Sparse Paradigm Free Mapping (ME-SPFM). Assuming a linear dependence of the BOLD percent signal change on the echo time (TE) and using sparsity-promoting regularized least squares estimation, ME-SPFM yields voxelwise time-varying estimates of the changes in the apparent transverse relaxation (ΔR2⁎) without prior knowledge of the timings of individual BOLD events. Our results in multi-echo fMRI data collected during a multi-task event-related paradigm at 3 Tesla demonstrate that the maps of R2⁎ changes obtained with ME-SPFM at the times of the stimulus trials show high spatial and temporal concordance with the activation maps and BOLD signals obtained with standard model-based analysis. This method yields estimates of ΔR2⁎ having physiologically plausible values. Owing to its ability to blindly detect events, ME-SPFM also enables us to map ΔR2⁎ associated with spontaneous, transient BOLD responses occurring between trials. This framework is a step towards deciphering the dynamic nature of brain activity in naturalistic paradigms, resting-state or experimental paradigms with unknown timing of the BOLD events.