Forget the forest, it's all in the trees
fMRI studies of language are typically done by group analysis, meaning that researchers test 10, 20 or even 50 subjects, then average data together onto a common brain space to search for regions that are active across brains.
But Fedorenko says this is not an ideal way to do things, mainly because the fine-grained anatomical differences between brains can cause data "smearing," making it look as if one region is active in two different tasks when in reality, the tasks activate two neighboring but not overlapping regions in each individual subject.
By way of analogy, she says, imagine taking pictures of 10 people's faces and overlaying them, one on top of another, to achieve some sort of average face. While the resulting image would certainly look like a face, when you compared it back to the original pictures, it would not line up perfectly with any of them. That's because there is natural variation in our features the size of our foreheads, the width of our noses, the distance between our eyes.
It's the same way for brains. "Brains are different in their folding patterns, and where exactly the different functional areas fall relative to these patterns," Fedorenko says. "The general layout is similar, but there isn't fine-grained matching." So, she says, analyzing data by "aligning brains in some common space is just never going to be quite right."
Ideally, then, data would be analyzed for each subject individually; that is, patterns of activity in one brain would only ever be compared to patterns of activity from that same brain. To do this, the researchers spend the first 10 to 15 minutes of each fMRI scan having
|Contact: Caroline McCall|
Massachusetts Institute of Technology