Ever seen those pictures of a brain with glowing red or yellow parts?

Maybe it was on the news during a science segment, or in some article you found on Reddit. These pictures seem to show up everywhere in recent psychology research.

And they're examples of functional magnetic resonance imaging, or fMRI. Magnetic resonance imaging scanners, or MRIs, are kind of amazing tools: they create a 3D image of the inside of living things without exposing them to radioactivity, like an x-ray would. And functional imaging can show changes inside of living things.

Like, which parts of the brain are active while someone is reading words or solving a puzzle. It's a super useful technique, but there's a lot that can go wrong, too. And we've already uncovered some major accidents in fMRI research.

Scientists know that active brain cells need oxygenated blood. So fMRI machines detect activity by measuring blood rushing to different parts of the brain, using something called the BOLD or blood oxygen-level dependent signal. The main ingredient in an fMRI machine is a powerful electromagnet— so don't go near one if there's any metal on you, or in you.

When that electromagnet is turned on, it creates a magnetic field. And it can be used to generate pulses of energy, which get absorbed by the hydrogen atoms in water molecules in your body. The atoms release that energy again at a specific frequency, which depends on what's affecting the magnetic field nearby— like whether the hemoglobin in your blood is carrying oxygen.

Then, sensors in the fMRI pick up these electromagnetic signals from the whole brain and generate an image! That entire process takes around 2 to 3 seconds, and then repeats to create lots of brain pictures. Now, there's still lots of work to do before you can learn anything from those images.

And most of the time, researchers use computer algorithms and many careful, small choices to do this work. First of all, it's hard to lie perfectly still through a whole fMRI scan, so researchers try to correct for head motion by lining up all the images from each person. But they don't trust computers to accurately adjust images any more than a couple millimeters.

So if someone wiggled around too much, researchers usually throw out that person's data. The next problem is that images of the brain aren't generated all at once. The picture gets created in horizontal slices.

So the last slice of an image, for example, might be showing 2 seconds later than the first. And it needs to be adjusted so it looks more like the last slice from the previous picture. That way, researchers can create brain images as if they were taken all at once.

A tiny time delay can mean a big change in blood flow and BOLD signal, because cells change so ridiculously fast! Third of all, people's heads aren't the same size. So to compare different brains, you need to fit everyone's images onto a standard template.

Software can do this automatically, kind of like Photoshop. But if you've ever stretched or squashed a photo, you know how even slightly distorted pictures can look a little funky. After scientists have adjusted all their images, they still have to compare them to draw any conclusions about what parts of your brain are involved in face recognition, instead of naming fruits.

To look for differences in brain activity, researchers use software to compare the activation in every single voxel—a 3-D pixel—in two images. That's a whole lot of comparisons, so there's a high chance of mistakes like false positives— when there seems to be a meaningful change in blood flow, even though there actually isn't. So researchers can adjust their software with careful statistical analysis to be more picky about what's a significant difference in brain activity.

In 2009, one team of psychologists showed that without doing this last step, they found brain activity in a dead salmon. I mean, there's a false positive for you! Thankfully, psychologists are scientists who know how their analyses can get messed up, and want to make sure they're not blowing bogus results out of proportion!

But things can still go wrong. A study from last year tested three popular fMRI software packages, which perform this last statistical analysis step slightly differently, on resting-state data from nearly 500 people. And what they found was a little worrying: because of errors, these programs may have seriously inflated the rates of false positives in some research, possibly even up to 70%.

Basically, researchers were clicking a box to make their analysis pickier, but the software wasn't picky enough. And now, there might be some dead salmon-level screw-ups lurking in some fMRI studies. But, this doesn't necessarily mean whole studies are completely wrong— they probably just suggested that bigger parts of the brain were active than actually were.

The good news is that these software bugs are fixed, and researchers are trying to share more data with each other to be sure they're analyzing it properly. But it still might be a while before all the past mistakes are uncovered and studies are replicated. Overall, fMRI data is most useful when you combine it with other methods and knowledge about the brain.

And, like any scientific research, you should never take any one fMRI study as the final truth. You have to take a step back, look at lots of studies, and squint! Thanks for watching this episode of SciShow Psychology, and especially to our patrons on Patreon for making it happen!

If you'd like to help us keep bringing you new content like this, go to patreon.com/scishow. And for more psychology research, go to youtube.com/scishowpsych and subscribe.