Truth in the Brain?

I've been thinking quite a bit about an article I saw last week, over at the New Scientist website, on the beginnings of the use of fMRI scans as a possible means of determining the truth of a statement. An extract:

In what amounted to the world’s first commercial lie-detection test using function magnetic resonance imaging (fMRI), technicians at No Lie mapped blood flow within Nathan’s brain while he answered a battery of questions about the deli fire and compared the results to control tests during which Nathan was asked to lie. The differences in the way his brain responded to these tasks appear to confirm his innocence.

It seems to me that a perfect truth or lie detector would be an immensely powerful tool, although not without the potential for misuse. The difficulty with any truth or lie detection is in the ability to determine those people who believe that they are guilty of a crime, when they are not, or those that are guilty, but do not show the same kind of emotional response as other people. For example if a psychopathic murderer can pass detection, then it is not one hundred percent effective, which must be the requirement if such a thing is going to be useful within a law and order setting. Nobody will be happy if innocent people are locked up and the guilty set free, because this technique isn't reliable.

In the end though, this technique may end up being no more reliable than the polygraph (which seems more gimmick than actual science). From Skepdic:

Is there any evidence that the polygraph is really able to detect lies? The machine measures changes in blood pressure, breath rate, and respiration rate. When a person lies it is assumed that these physiological changes occur in such a way that a trained expert can detect whether the person is lying. Is there a scientific formula or law which establishes a regular correlation between such physiological changes and lying? No. Is there any scientific evidence that polygraph experts can detect lies using their machine at a significantly better rate than non-experts using other methods? No. There are no machines and no experts that can detect with a high degree of accuracy when people, selected randomly, are lying and when they are telling the truth.

To accurately and reliably detect lies must be the holy grail of many researchers (not to mention a few chat show hosts). But as to whether such a thing is actually possible, is for the time-being still an unanswered question.