MRIs That Read Your Thoughts 28
Nicholas Roussos writes "Functional MRIs have been used in several studies to accurately predict what volunteers were looking at even when they themselves were unsure. According to the BBC, 'When two images were flashed in quick succession, the volunteers only consciously saw the second one and were unable to make out the first. But the brain scans clearly distinguished the patterns of brain activity created by the "invisible" images.'"
Some info on fMRI (Score:3, Informative)
I work in the analysis of fMRI data, so it might be good to disspell some myth and provide some key info.
fMRI is derived from "normal" MRI, and in fact, any fMRI scanner can also make a normal MRI. MRI, or actually, NMRI, stands for Nuclear Magnetic Resonance Imaging. The "N" for Nuclear is usually not included in the abbreviation because it seems to make the public nervous. An MRI makes an image by using a radio-frequency magnetic field to excite a magnetic spin-state in the nuclei of hydrogen atoms (i.e., in single protons); the more an area resonates, the more the measured signal is modulated, and the higher the hydrogen concentration assumed to be (typically that means: the higher the water-concentration is). Imaging can be done on a three dimensional grid, because the exact frequency at which the spin-state is excited can be modulated by static (compared to the RF field) magnetic fields. To this end, a large static field (1-7 Tesla) and 2 auxiliary fields are used to determine at which point in the grid the resonance takes place. This, then, is done for each point in the grid. The result is a 3D proton (or water) density map of the brain (usually refered to as the "anatomy (map)" by people working in the field).
In functional MRI (fMRI), one takes the MRI process one step further by exploting the paramagnetic properties of oxygen; the amount of oxygen (in the blood, attached to hemoglobine) present in a certain area will modulate the proton spin resonance frequency as well, and this extra modulation can be measured. In this way, one can also make an oxygen-density map (one has to correct and compensate for a lot of things here, but that's a long story).
The oxygen density map can be used, because areas in the brain that are active "draw blood" towards themselves. This is called the BOLD (blood oxygen level dependent) response. Typically, this response lags one to a couple of seconds behind the actual activation of the brain area; in the fMRI data, one sees the signal in that area become higher. One can thus detect which areas of the brain are active.
The trick that the article describes is sometimes called brain state prediction, and that's a more difficult problem. Typically, one measures the fMRI signal while first supplying stimulus "1" N times, and them stimulus "2" N times. Determining which areas are active for stimulus "1" OR "2" (or both), usually has a reasonable SNR; based on the measured example, something simple like an F-test will suffice. However, given which part of the brain is active for either, the task of determining which of these active areas correspond more to simimulus "1" or "2", is more diffucult: usually the entire active area responds to both responses, and the difference is in the magnitude (or delay, or width) of the BOLD response in that area w.r.t. stimulus "1" or "2". These difference (sometimes expressed in a Contrast-to-Noise Ratio), is much smaller than the SNR for the activation. Consequently, this is an active research area.
To disspell some myths: the fMRI data can only be obtained using a HUGE scanner, and your head has to be completely inside. Furthermore, the sampling frequency is rather low (1.5 seconds between scans, usually), and the spatial resolution isn't that high either (64x64x64 voxels, in that order of magnitude).