I first volunteered to be a participant in an fMRI study as a wide-eyed college freshman ten years ago. I was so excited to get to see a picture of my brain, but once I was tightly packed into the scanner, a few worries entered my mind: would it turn out that I had a tumor, or was one of those people with half a brain? Would the experiment show that I’m not very smart, or vulnerable to mental illness? Would the graduate student administering the experiment know what I was thinking?
While my concerns were rather common, they were also rather unfounded. fMRI is rather good at predicting what you are thinking about in laboratory situations where you are given a very short list of things to think about. For example, an fMRI scan can predict whether you are thinking about a face or a place as the mental imagery for places and faces recruits different brain areas. And it is of note that this happens when your subjects are willing and able to think only about faces or houses for a 20 second run inside the scanner.
Earlier, I wrote a little about the analysis of fMRI data. The kind of inferences researchers make in these kinds of studies is in the form of “what area of the brain is more active for task 1 compared to task 2?” In prediction, the question becomes “given a pattern of brain activity, what was the participant seeing/hearing/doing?” Early prediction techniques relied on correlation: a voxel was predictive if its activity to a particular stimulus in one run was more highly correlated with activity to the same stimulus from another run than to activity from a different stimulus. More modern prediction studies make use of machine learning and statistical classifiers such as linear discriminant analysis, and support vector machines in particular. This approach to data analysis has been both popular and fruitful, particularly for vision research. I recommend this review for more on the state-of-the-art.
As impressed as we tend to be with both math-y and brain-y things, it is important to remember that we are still not able to predict arbitrary patterns of brain activity. When you are lying there in the scanner, it is possible to determine if you are daydreaming, but not the contents of your daydream. I have a particular concern over the over-selling of these studies in the titles of both lay and academic papers. Although the technology for fMRI decoding is advancing rapidly, I also do not see the scanner as a place where future civil liberties go to die. Getting readable data from the scanner requires your participant to be very co-operative, so its use as an interrogation device is limited. But for you paranoid types, perhaps you should consider some metal dental work… or even a hair tie!