Summary: Zooming out to image larger areas of the brain while using fMRI technology allows researchers to pick up additional, relevant information, providing a better understanding of neural interaction.
Researchers have learned a lot about the human brain through functional magnetic resonance imaging (fMRI), a technology that can give insight into brain function. Typical fMRI methods may miss key information and provide only part of the picture, the Yale researchers say.
In a new study, they evaluated different approaches and found that zooming out and taking a wider field of view captures additional relevant information left by a narrow focus, providing a greater understanding of neural interaction.
Moreover, these more comprehensive findings may help to address the problem of neuroimaging reproducibility, as some of the results presented in studies cannot be reproduced by other researchers.
The results were published on August 4 in Proceedings of the National Academy of Sciences.
Studies using fMRI usually focus on small areas of the brain. As one example of this approach, researchers are looking for areas of the brain that become more “active” when a particular activity is performed, and head to small areas with the strongest activation. But a growing body of evidence shows that brain processes, and complex processes in particular, are not limited to small parts of the brain.
“The brain is a network. It’s complex,” said Dustin Schinoust, assistant professor of radiology and biomedical imaging and senior author of the study. Oversimplification leads to inaccurate conclusions, he said.
“For more complex cognitive processes, it is unlikely that many brain regions are completely uninvolved,” added Stephanie Noble, a postdoctoral fellow in the Scheinost Laboratory at Yale University School of Medicine and the study’s lead author.
Focusing on small areas leads to the exclusion of other areas that may be involved in the behavior or process under study, which can influence the direction of future research as well.
“You’re developing this wrong picture of what’s really going on in the brain,” she said.
For the study, the researchers evaluated how well fMRI analyzes across a range of scales were at detecting effects or changes in fMRI signals as participants performed different activities, revealing which parts of the brain were working.
They used data from the Human Connectome Project, which collected scans of individuals’ brains as they performed various tasks related to complex processes such as emotion, language and social interactions.
The research team looked for effects in very small parts of the brain’s network – such as connections between only two regions – as well as in combinations of connections, diffuse networks and whole brains.
They found that the larger the scale, the better they could detect effects. This ability to detect influences is known as “strength”.
“We get better strength with these large-scale methods,” Noble said.
At smaller scales, the researchers were only able to detect about 10% of the effects. But at the network level, they can detect more than 80% of them.
The trade-off for the additional strength was that the broader views did not convey information as spatially accurate as those of the smaller analyses. For example, at the smaller scale, researchers could say with confidence that the effects they observed were occurring throughout the small area.
However, at the network level, they could only say that the effects were occurring across a large portion of the network, not exactly where the network was.
The goal, says Noble, is to balance the advantages and disadvantages of different methods.
“Would you rather be very confident with a small piece of relevant information—in other words, have a very clear picture of just the tip of the iceberg?” She said.
“Or would you rather have a really big picture of the whole iceberg that might be a little blurry but give you a sense of the complexity and the wide spatial scope of where things happen in the brain?”
For other researchers, this approach is easy to implement, and Noble said she is looking forward to seeing how other scientists use it.
She notes that fields of psychology and neuroscience, including neuroimaging, have had trouble reproducing. The low power of the fMRI analyzes contributes to this: low power studies reveal only small parts of the story, which can be considered contradictory rather than parts of the whole.
Increasing the power of the fMRI, as she and her colleagues have done here by increasing the volume of their analyzes, may be one way to address reproducibility challenges by revealing how consistent the seemingly contradictory results are in reality.
“Going up the food chain, so to speak, going from a very low level to more complex networks, gives you more power,” Schinoust said. “This is one of the tools we can use to help with the reproductive problem.”
Noble said scientists should not dispose of the baby with the bath water. A lot of good work is being done to improve methods and enhance rigor, she said, and fMRI remains a valuable tool: “I believe that assessing strength, rigor, and reproducibility is healthy for any field. Especially those dealing with the complexity of organisms and mental processes.”
Noble is now developing an “energy calculator” for fMRI, to help others design studies in a way that achieves the desired level of power.
About this neuroimaging research news
original search: open access.
“Improving ability in functional magnetic resonance imaging by bypassing mass-level inferenceWritten by Stephanie Noble et al. PNAS
Improving ability in functional magnetic resonance imaging by bypassing mass-level inference
Inference in neuroimaging usually occurs at the level of the brain’s focal regions or circuits. However, increasingly robust studies paint a richer picture of large-scale effects distributed throughout the brain, suggesting that many focal reports may only reflect the tip of the iceberg of the core effects.
How focal versus broad perspectives affect the conclusions we reach has not been comprehensively evaluated using real data.
Here, we compare sensitivity and specificity across procedures representing multiple levels of inference using an empirical benchmark procedure that replicates task-based neural network models from the Human Connectome Project dataset (∼1,000 subjects, 7 tasks, 3 resampling group sizes, 7 inferential procedures).
Only large-scale (network and whole-brain) procedures had a conventional statistical power level of 80% for average effect detection, reflecting a 20% greater statistical power over focal (edge and group) procedures. The power also significantly increased the false discovery rate – compared to the familial error rate – control procedures.
The downsides are somewhat limited. The loss in specificity for large-scale and FDR procedures was relatively modest compared to the gain in power. Furthermore, the large-scale methods we offer are simple, quick and easy to use, and provide a direct starting point for researchers.
This also points to the promise of more sophisticated broad-based methods for not only functional connectivity but also related areas, including task-based activation.
Altogether, this work shows that changing the inference scale and selecting the FDR control are immediately achievable and can help address problems using the statistical power that has plagued model studies in this field.