Stanford scientists identify fMRI inaccuracies

Scientists at Stanford University believe that results from some functional magnetic resonance imaging scans have been unintentionally distorted and may put into question the conclusions of hundreds of scholarly studies that have relied on the imaging technique.

In a study published in the journal NeuroImage, Brian Knutson, M.D., an associate professor in the Stanford University psychology department, and Matthew Sacchet, a Ph.D. student in the Neurosciences Program at the Stanford School of Medicine, wrote that many researchers have used a "one-size-fits-all" approach to processing fMRI findings that has result in skewing research results.

Functional MRI measures the changes in blood flow in the brain by detecting the change in magnetization between oxygen-rich and oxygen-deprived blood. In order to interpret that data, researchers must statistically process it by different means, one of which is spatial smoothing. This involves averaging out the activity seen in part of the brain.

The problem, Knutson and Sacchet found, is that when researchers process the data with a traditional "smoothing kernel" of 8mm, they can give too much weight to large brain structures, and too little weight to smaller structures.

"It might seem strange that a systematic bias like that could bias the whole field," Knutson said, according to an announcement. "But if half the people use 8mm and half use 4mm, you might end up with very different results, and it could add up."

Knutson and Sacchet particularly are interested in the area of the brain known as the nucleus accumbens, which plays a central role in the reward circuit. During the course of their research, using the smaller 4mm smoothing kernel, they determined forward and rear sections of the nucleus accumbens have different functions. But other studies that failed to confirm their findings used the 8mm smoothing kernel, which may have skewed the results by making it look like all the activity appeared in the rear of the nucleus accumbens.

Sacchet said they've already heard from a researcher in Germany who had her data re-analyzed and came to the same conclusion as the Stanford scientists.

"I honestly think most people want good data," Knutson said. "I'm excited that we can make this kind of research more rigorous."

To learn more:
- see the study abstract
- read the announcement

Suggested Articles

To build a successful network for health information exchange, policies must provide enough time and flexibility for the networks to develop and scale

Welcome to this week's Chutes & Ladders, our roundup of hirings, firings and retirings throughout the industry.

Federal lawmakers are taking a hard look at how the VA protects patient data shared with VA-approved health apps.