Functional magnetic resonance imaging, or fMRI for short, is a non-invasive method with which changes in local cerebral blood flow can be viewed superimposed on anatomical reconstructions of the brain in humans. The spin axes of the nuclei of hydrogen atoms are aligned upright in a strong magnetic field. Short radio pulses deflect the spin axes. Small discrepancies in the time that the nuclei take to newly upright their axes are dependent on the blood oxygenation. fMRI uses blood oxygen level-dependent (BOLD) signals. Brain tissue cannot store the oxygen and sugar brain cells need to function. Therefore, local blood flow increases when nerve cells start processing information and the blood oxygen level rises. Conversely, blood flow diminishes when nerve cell activity is inhibited. With fMRI, scientists exploit the relationship between nerve cell activity and blood flow to reconstruct functional cerebral activation maps. It is important to note that, while nerve cells respond to sensory stimulation within fractions of a second, the blood flow response is slower, taking about two seconds to change. Moreover, more than 10,000 nerve cells must change activity in order to measurably affect blood flow. This is the estimated number of cells (Rakic, 2008) in the smallest volume of cortical tissue in which my colleagues and I could detect a stimulus-related change in BOLD signal at 4.7T (Sachdev and others, 2003). Because the detectable blood flow changes result from the energy requirements of such large populations of nerve cells, we do not know whether the observed changes stem from a high activation of a select subpopulation of particularly responsive nerve cells or from a low activation of a large population of moderately-activated cells. A small number of highly activated nerve cells may introduce a significant partial volume effect into our measurements.
Since Ogawa and others (1993) at Bell Laboratories and Kwong and others (1992) at the National Institutes of Health independently discovered fMRI in the early 1990's, the method has become widely used in the neurosciences. Psychiatrists and psychologists have developed a large body of knowledge on the location of stimulus- and task-related activation in cerebral cortex.
The fMRI signals detected during a scanning session are very dynamic. Signal strength may vary in a wide range. Many repetitions of data collection are needed. As a consequence, brain activation maps obtained with fMRI are rendered from complex statistical analyses and, therefore, are probabilistic. Taking signal strength and variability into consideration, thresholds need to be defined, below which a change in blood flow is no longer considered relevant. Conversely, regions with blood flow above threshold, are often heralded as "lit up". This division in all or nothing imposes limitations on the interpretations of the findings of such studies which I illustrate with an analogy below.
I used to live in the Lemanic region of Switzerland. The airport to fly in is Cointrin near Geneva. Peering through the window on the approach to Cointrin in November, the observer may conclude more often than not that the Lemanic region consists of an arctic plane stretching from a low mountain range on one side to the high peaks of the Mont Blanc massif on the other. Then, the airplane descends onto the plane, which turns out to consist of thick clouds. Once the layer is penetrated, the observer is surprised to detect a diverse landscape with a large crescent-shaped lake surrounded by ridges of many more mountains. There are cities and towns. There are forests, vineyards, fields and ponds. One glance cannot comprehend the multitude. The observer has to pay attention to one cue at a time.
Analyzing fMRI data consists of a similar experience. Potentially pertinent information remains undiscovered, hidden under the clouds of statistical thresholding. We are pressed to ignore the landscape under the clouds, because of its transience. The profiles change from glance to glance, escaping our scrutiny. We lack the tools of comprehension. Striving for simplicity, we resort to thresholding.
However, the brain areas with the strongest and most statistically significant activation may not be the most instrumental for the mental processes under investigation. In fact, the regions of cerebral cortex that may be involved in making decisions, taking risks, and making plans commonly receive input from multiple senses. The input has been preprocessed in the primary areas, that is the cortical regions first to receive the information from sensory organs. Moreover, the nerve cells in the higher order areas are subject to feedback and re-entrant input, activating the nerve cells with delay. The comparatively small contribution of the stimulus and the delay in nerve cell responses may result in activation too feeble and incoherent for pushing the region above threshold. This complication may pose the greatest impediment in fMRI of higher brain function.
Apart from the statistical variability inherent in the method, the selection of the participants and sample size decisively influence the outcome of a fMRI study. Drugs are commonly tested on more than 1000 patients in phase III clinical trials before the FDA considers approval [Zeke Ashton (2000) The FDA and Clinical Trials: A Short History, THE BODY]. In a widely publicised study (Maggie Fox's and Xavier Briand's post entitled "Brain differences mark those with depression risk" on Reuters, Mar. 23, 2009; Roni Caryn Rabin's post entitled "Study Links Depression to Thinning of Brain’s Cortex" in The New York Times, March 24, 2009), more than 100 participants needed to be recruited to demonstrate a thinning of the cerebral cortex in people predisposed for depression. Social scientists commonly incorporate the responses of 2000 and more participants in their analyses to be able to provide answers of significance. By contrast, fMRI studies on fundamental questions will hardly ever reach the enrollment necessary for assessments on such large-scale because of forbidding cost. Careful consideration is necessary to determine whether a minute difference is located in an area important enough to warrant continued enrollment. In addition, the investigators have to ensure that the participants constitute a representative sample for the question of study, differing only in the feature to be examined. Sampling bias and co-linked differences may introduce systematic errors, leading to observations the underlying mechanisms of which can not be disambiguated.
In the next two essays of this trilogy, I discuss two recent studies illustrating the achievements accomplished with and the limits of fMRI. The underlying neural mechanisms for the findings of the first study could be explored in animal models. By contrast, no animal models are readily available for the findings of the second study. The installments were published Mar. 27 and Mar. 31, 2009.
- Yesterday, National Public Radio's All Things Considered broadcast the third installment of Barbara Bradley Hagerty's journalistic journey into spirituality and the brain entitled "Prayer May Reshape Your Brain...And Your Reality". The description of the scientific methods, on which the findings portrayed in this segment were based, was superficial by any standards. The brain does not light up. The compounds used to image cerebral activation with single photon computed tomography (SPECT) are not dyes. They are radioactively-labeled tracers that accumulate in the brain according to local blood flow. Local blood flow is coupled to energy metabolism. The energy metabolism fluctuates with nerve cell activation. The inaccuracies of explanation in this report degrade the scientific validity of methods carefully developed for the use in diagnostic medicine, discrediting the scientific merit of the studies involved (05/21/09).
- On Jun. 30, 2009, Public Broadcasting Service's Nova premiered a show on musicophilia with Oliver Sacks entitled "Musical Minds". I have written about Professor Sacks' work on musicophilia in my post dated Jan. 30, 2008. In the video extra accompanying the show, he is undergoing fMRI while listening to pieces of music composed by Bach and Beethoven. Although Professor Sacks confessed his confusion about the provenance of the pieces, fMRI rendered distinctly more activated brain regions in response to the piece by Bach - his favorite - than to the piece by Beethoven. The lead imaging scientist concluded that "your brain can distinguish them, even when you don't!" Such discrepancies are difficult to reconcile. The part was excluded from the show (07/02/09).
- Today, Jon Hamilton reported in a segment entitled "False Signals Cause Misleading Brain Scans" broadcast on National Public Radio's All Things Considered on the findings of two recent scientific studies published in the journals Nature Neuroscience (Kriegeskorte and others, 2009) and Perspectives on Psychological Science (Vul and others, 2009). The studies demonstrate that recurrent analyses of the same fMRI data, known as double dipping, can produce statistically significant, false positive brain activation (07/07/09).
- Kriegeskorte N, Simmons WK, Bellgowan PS, Baker CI (2009) Circular analysis in systems neuroscience: the dangers of double dipping. Nat Neurosci 12: 535-40.
- Kwong KK, Belliveau JW, Chesler DA, Goldberg IE, Weisskoff RM, Poncelet BP, Kennedy DN, Hoppel BE, Cohen MS, Turner R, et al. (1992) Dynamic magnetic resonance imaging of human brain activity during primary sensory stimulation. Proc Natl Acad Sci USA 89: 5675-5679.
- Ogawa S, Menon RS, Tank DW, Kim SG, Merkle H, Ellermann JM, Ugurbil K (1993) Functional brain mapping by blood oxygenation level-dependent contrast magnetic resonance imaging. A comparison of signal characteristics with a biophysical model. Biophys J 64: 803-812.
- Racic P (2008) Confusing cortical columns. Proc Natl Acad Sci USA 105: 12099-12100.
- Sachdev RN, Champney GC, Lee H, Price RR, Pickens DR 3rd, Morgan VL, Stefansic JD, Melzer P, Ebner FF (2002) Experimental model for functional magnetic resonance imaging of somatic sensory cortex in the unanesthetized rat. Neuroimage 19: 742-750.
- Vul E, Harris C, Winkielman P, Pashler H (2009) Reply to Comments on "Puzzlingly high correlations in fMRI studies of emotion, personality, and social cognition". Perspectives Psychol Sci 4: 319-324.
- fMRI III: Religiosity & Brain Activation
- fMRI II: Memory of Simple Stimuli & Brain Activation
- About Cochlear Implants & The Quality of Music