Summary: Using fMRI to image larger portions of the brain at once—rather than focusing narrowly on small regions—captures additional, meaningful signals and provides a more complete view of how brain areas interact.
Source: Yale
Functional magnetic resonance imaging (fMRI) has revolutionized our ability to study human brain activity, but conventional approaches that concentrate on small, highly localized regions may miss important aspects of brain function.
A new study from Yale researchers compares a range of analysis strategies and shows that broadening the field of view—examining larger networks or whole-brain patterns—reveals effects that narrow, focal analyses often overlook. This wider perspective improves detection of neural signals and offers a clearer sense of how distributed brain systems work together.
The researchers also suggest that adopting broader-scale analyses could help address reproducibility challenges in neuroimaging, where some published results fail to be replicated by other teams. Their findings were published Aug. 4 in Proceedings of the National Academy of Sciences.
Traditional fMRI studies frequently search for small regions that show the strongest activity during a task and report those focal activations. However, complex cognitive, emotional, and social processes are rarely confined to single brain spots. Instead, these processes unfold across distributed networks that span many regions.
“The brain is a network. It’s complex,” said Dustin Scheinost, associate professor of radiology and biomedical imaging and the study’s senior author. He and his colleagues warn that oversimplifying brain activity by zeroing in on small areas can lead to incomplete or misleading conclusions.
“For more sophisticated cognitive processes, it’s unlikely that many brain areas are wholly uninvolved,” added Stephanie Noble, a postdoctoral associate in Scheinost’s lab and lead author. Focusing only on small, highly active regions can obscure other parts of the brain that contribute to the behavior or task under study.
To quantify these differences, the research team evaluated how well fMRI analyses at multiple scales could detect task-related effects—changes in signals that indicate engagement of particular brain systems. They used data from the Human Connectome Project, which includes scans from nearly a thousand individuals performing diverse tasks related to emotion, language, and social cognition.
Analyses ranged from the finest-grained level, examining single connections between two regions, to intermediate cluster-level approaches, and up to broad network-level and whole-brain methods. The results were consistent: larger-scale methods detected many more true effects than focal analyses.
This increased ability to identify genuine effects is known as statistical power. Noble noted, “We get better power with these broader-scale methods.” At the smallest scales, analyses identified roughly 10% of effects. At the network level, detection rates exceeded 80%.
The main trade-off is spatial specificity. Fine-grained analyses can localize effects very precisely within a small brain region. In contrast, network-level or whole-brain findings are less spatially specific: they indicate that effects are present across broad systems but do not pinpoint exact loci within those systems.
Noble frames the choice as one between two research priorities: being highly confident about a narrow piece of information, or obtaining a broader, somewhat less detailed picture that better reflects the complexity and distributed nature of many brain processes. The goal is to balance sensitivity and specificity so studies capture meaningful patterns without overstating localization.
Because broader-scale methods are straightforward to implement, the authors expect other teams to adopt them quickly. Increasing analytic scale and choosing appropriate statistical controls are practical steps that many labs can take to boost power and improve the robustness of results.
Low statistical power has been recognized as a contributor to the reproducibility problem across psychology and neuroscience. Studies with limited power tend to uncover only fragments of the underlying signal, which can appear inconsistent with other results rather than complementary pieces of a larger pattern. By raising power through broader-scale inference, apparent contradictions may be resolved as parts of a coherent whole.
“Moving up the scale—from low-level focal analyses to more complex networks—buys you a lot more power,” Scheinost said. “This is one of the tools we can use to address the reproducibility issues in the field.”
The authors emphasize that this perspective does not diminish the value of precise localization or the many methodological advances improving rigor in fMRI research. Rather, assessing power and choosing the appropriate scale of inference complements those efforts and helps ensure that studies better reflect the distributed nature of brain function.
Noble is developing a practical “power calculator” for fMRI to help researchers design studies that reach desired power levels across different inference scales.
About this neuroimaging research news
Author: Mallory Locklear
Source: Yale
Contact: Mallory Locklear – Yale
Image: The image is in the public domain
Original Research: Open access.
“Improving power in functional magnetic resonance imaging by moving beyond cluster-level inference” by Stephanie Noble et al. PNAS
Abstract
Improving power in functional magnetic resonance imaging by moving beyond cluster-level inference
Neuroimaging inference often targets focal brain areas or small circuits, but increasing evidence from well-powered studies points to broad-scale effects distributed across the brain. This suggests that many focal reports may represent only the most obvious portions of larger underlying patterns.
To evaluate how focal versus broad-scale perspectives influence scientific conclusions, the authors compared sensitivity and specificity across multiple inferential procedures. Using an empirical benchmarking framework, they resampled task-based connectomes from the Human Connectome Project (approximately 1,000 subjects, seven tasks, three resampling group sizes, and seven inferential procedures).
Broad-scale procedures—network-level and whole-brain approaches—achieved the conventional benchmark of 80% statistical power for detecting an average effect, outperforming focal methods (edge- and cluster-level) by more than 20 percentage points. Power also improved substantially when using false discovery rate (FDR) control rather than familywise error rate procedures.
The costs of broader-scale and FDR approaches were modest: the loss of spatial specificity and a slight increase in false positives were small compared with the large gains in power. The broad-scale methods introduced are simple, fast, and accessible, providing an immediate option for researchers to increase sensitivity.
Overall, shifting the scale of inference and adopting FDR control are feasible strategies that can help remedy the statistical power limitations affecting many fMRI studies and related fields, including task-based activation research.