Traditional functional brain imaging studies rely on paradigms with tightly controlled experimental event timing. This allows researchers to build models of how the signal from regions responsive to those events should plausibly fluctuate over time. These models then can be used to perform statistical hypothesis testing on the data (e.g., GLM and ANOVA). But, unfortunately, many cognitively and clinically relevant phenomena such as mind wandering, unprompted memory recall, self-initiated motor behaviors, induction of emotions and mood, responses to naturalistic stimuli, and epileptic seizures are not amenable to scientific scrutiny using such tightly controlled experiments because events of interest in these research scenarios happen at moments and for durations that cannot be easily or fully pre-defined, hence the term paradigm-free functional brain imaging.
Given the importance of being able to study time-uncontrollable phenomena for gaining a complete understanding of healthy and diseased brain function, neuroimagers have devised many ways to deal with timing uncertainty. For example, in resting-state studies —namely those in which subjects are instructed to freely let their mind wander—it is common to rely on seed-based correlational analysis, model-free multivariate decomposition algorithms (e.g., independent-component analysis) or clustering approaches (e.g. k-means, hierarchical clustering) to uncover spatially distributed patterns of temporally synchronous activity from hemodynamic (i.e. functional MRI, functional near infrared spectroscopy) or electrophysiological (EEG, MEG, ECOG) data. Those techniques can be used to build network models of the brain (i.e., functional connectomes) and explore how the topological properties of those models change across populations or experimental conditions (e.g. sleep vs. awake). In parallel, point-process and deconvolution methods can be used to detect temporally sparse instances of activity-inducing events that can be localized to individual regions. Finally, inter-subject correlational analysis for naturalistic stimuli, as well as advanced machine learning algorithms (e.g., non-linear dimensionality reduction techniques) are now also being applied to synchronous paradigm-free data in order to study how brain activity evolves freely over time.
Despite tremendous progress, the analysis and interpretation of paradigm-free data to explore brain function remains challenging. For example, fMRI data is noisy and the temporal profile of hemodynamic responses (e.g., those linking neuronal events to fMRI recordings) is varying both across subjects and also across regions. These properties of the data create significant challenges for methods that must make some assumptions (e.g., a given canonical hemodynamic response function) to separate neuronal-induced signal from noise. Similar arguments regarding limited signal-to-noise and unmodeled inter-subject response variability apply universally across all functional neuroimaging modalities. Consequently, results from paradigm-free analyses are cumbersome to validate and interpret, often relying on concurrent recordings or multimodal data (e.g., EEG or skin conductance).
The goal of this research topic is to bring together experts that work with paradigm-free experiments and data in order to advance our understanding of brain function. We welcome original research describing novel methodologies and algorithms, interpretational frameworks and applications to both basic and clinical science across the wide range of functional neuroimaging techniques (e.g. fMRI, fNIRS, EEG, MEG, ECOG, optical imaging, neural recordings, etc.). Opinion and review style manuscripts that will foster scientifically constructive discussion regarding the use and implementation of paradigm-free approaches are also welcome. Finally, we will also consider negative studies that can help highlight misconceptions and pitfalls on the utility of paradigm-free approaches in functional neuroimaging.
Traditional functional brain imaging studies rely on paradigms with tightly controlled experimental event timing. This allows researchers to build models of how the signal from regions responsive to those events should plausibly fluctuate over time. These models then can be used to perform statistical hypothesis testing on the data (e.g., GLM and ANOVA). But, unfortunately, many cognitively and clinically relevant phenomena such as mind wandering, unprompted memory recall, self-initiated motor behaviors, induction of emotions and mood, responses to naturalistic stimuli, and epileptic seizures are not amenable to scientific scrutiny using such tightly controlled experiments because events of interest in these research scenarios happen at moments and for durations that cannot be easily or fully pre-defined, hence the term paradigm-free functional brain imaging.
Given the importance of being able to study time-uncontrollable phenomena for gaining a complete understanding of healthy and diseased brain function, neuroimagers have devised many ways to deal with timing uncertainty. For example, in resting-state studies —namely those in which subjects are instructed to freely let their mind wander—it is common to rely on seed-based correlational analysis, model-free multivariate decomposition algorithms (e.g., independent-component analysis) or clustering approaches (e.g. k-means, hierarchical clustering) to uncover spatially distributed patterns of temporally synchronous activity from hemodynamic (i.e. functional MRI, functional near infrared spectroscopy) or electrophysiological (EEG, MEG, ECOG) data. Those techniques can be used to build network models of the brain (i.e., functional connectomes) and explore how the topological properties of those models change across populations or experimental conditions (e.g. sleep vs. awake). In parallel, point-process and deconvolution methods can be used to detect temporally sparse instances of activity-inducing events that can be localized to individual regions. Finally, inter-subject correlational analysis for naturalistic stimuli, as well as advanced machine learning algorithms (e.g., non-linear dimensionality reduction techniques) are now also being applied to synchronous paradigm-free data in order to study how brain activity evolves freely over time.
Despite tremendous progress, the analysis and interpretation of paradigm-free data to explore brain function remains challenging. For example, fMRI data is noisy and the temporal profile of hemodynamic responses (e.g., those linking neuronal events to fMRI recordings) is varying both across subjects and also across regions. These properties of the data create significant challenges for methods that must make some assumptions (e.g., a given canonical hemodynamic response function) to separate neuronal-induced signal from noise. Similar arguments regarding limited signal-to-noise and unmodeled inter-subject response variability apply universally across all functional neuroimaging modalities. Consequently, results from paradigm-free analyses are cumbersome to validate and interpret, often relying on concurrent recordings or multimodal data (e.g., EEG or skin conductance).
The goal of this research topic is to bring together experts that work with paradigm-free experiments and data in order to advance our understanding of brain function. We welcome original research describing novel methodologies and algorithms, interpretational frameworks and applications to both basic and clinical science across the wide range of functional neuroimaging techniques (e.g. fMRI, fNIRS, EEG, MEG, ECOG, optical imaging, neural recordings, etc.). Opinion and review style manuscripts that will foster scientifically constructive discussion regarding the use and implementation of paradigm-free approaches are also welcome. Finally, we will also consider negative studies that can help highlight misconceptions and pitfalls on the utility of paradigm-free approaches in functional neuroimaging.
