Accuracy of neurological workhorse fMRI images thrown into doubt

Here's an interesting snippet courtesy of Forbes magazine. The possibility has been raised that functional magnetic resonance imaging (fMRI) images of the brain, a mainstay of neurological research for the past 15 years or so, may not be as reliable as previously thought.

The technique - which, to simplify horribly, involves measuring the tiny bursts of energy given off by the hydrogen atoms in the brain as they realign themselves after magnetic interference - is sound in itself. But the software used to generate those user-friendly colour-coded pictures of the brain may be suspect, according to some scientists. So, potentially, some 40,000 research studies on brain functionality (and the treatments based on those studies) over the last decade and a half might not be as accurate as thought. Which, if true, is pretty major.

The claims come from just one Swedish study, published in the Proceeding of the National Academy of Sciences of the United States, which purports to show that, in their studies of 499 healthy patients, the fMRI software does not always produce images of the correct parts of the brain lighting up (in some cases, false positives of 70% were recorded). It suggests that bugs or glitches in the software may be leading to faulty images, often to a dramatic extent.

I can get little feel from the article as to how extensive such errors might be, how many research reports might be affected, or even how many different software packages might involved (and how many are probably fine), so it is difficult to assess the possible impact of such a state of affairs. The Swedish study will need to be checked and repeated extensively (rattled fMRI companies are probably working on that right now) before we intemperately throw of our decades of valuable research.